Description:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:

METHOD AND SYSTEM FOR FORECASTING INTRADAY MARKET PRICE DYNAMICS OF ENERGY USING A PARTICLE-FILTERING APPROACH

Applicant

Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th floor,
Nariman point, Mumbai 400021,
Maharashtra, India

Preamble to the description:
The following specification particularly describes the invention and the manner in which it is to be performed.
TECHNICAL FIELD
The disclosure herein generally relates to forecasting market price of electricity, and, more particularly, to method and system for forecasting intraday market price dynamics of energy using a particle-filtering approach.
BACKGROUND
Electricity price forecasting is one of the special commodities that forecasts forward energy prices in electricity markets. Unlike other commodities, electricity ‘spot market’ is typically a day-ahead market that allows continuous trading, where agents submit their bids and offers delivery of electricity during each bid hour. Many countries are developing new energy policies to secure their energy systems, sustaining in economy development, and reducing negative environmental impact. Intraday power markets have become increasingly prominent in the context of balancing day-ahead markets in the recent past. Intraday power trading services balance the day-ahead markets by trading sell off surplus production or replenish shortfall. The inclusion of renewable energy sources (RES) in the supply side along with the existing challenge of elastic demand further adds market volatility.
Given the increasing prominence of intraday markets, forecasting intraday prices at short-term or at critical hours before the actual execution of delivery have become challenging in the recent past. Price forecasting in such scenarios helps in planning and trading power products, which in turn are expected to reduce market volatility. Electricity requirements differ with market regions, due to demand-side behaviors and climatic conditions, which reflects on trading behaviors. It is hence imperative that forecasting intraday prices is specific to the region under consideration. Further, meticulous data analysis concerning the historical intraday price data is necessary to provide reliable forecasts. Forecasting intraday power prices at critical hours before the actual hour of delivery allows traders to bid appropriately and this in turn is expected to reduce market volatility.
In conventional methods, price forecasting is achieved through extensive use of metadata information with adopted sophisticated formulation methods. Neural network-based price forecasts have been quite popular in energy price forecasting (EPF) committee which uses machine learning models related to ensemble averaging. The day-ahead price forecasting modelling techniques for power market price values, solar energy generation and wind information are often sophisticated with extensive formulation requirements. However, conventional methods lack in forecasting market price at critical hour of interest due to complex and data driven features.
SUMMARY
Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a system for forecasting intraday market price dynamics of energy using a particle-filtering approach is provided. The system includes receiving one or more product price data of electricity. Simultaneously, a plurality of price bidding parameters comprising of an average price, a maximum price, a minimum price for each market electrical energy data at every bidding hour are obtained for product data. Further, a stochastic constant drift model is created for the one or more product price data, wherein the stochastic constant drift model is updated at one or more non-zero bidding instances close to every bidding hour of interest based on a price measurement. Furthermore, the stochastic constant drift model for the plurality of price bidding parameters are analyzed based on one or more value change observed in each price bidding parameter at every bidding hour based on a market price velocity, wherein the stochastic constant drift model defines a next price bidding parameter value by incrementing a previous price bidding parameter value. Then, the price measurement is recomputed to update the stochastic constant drift model close to every bidding hour of interest and determining if each price bidding parameter lies within a predefined interval bounds based on a one or more Bollinger bands. Furthermore, by using a particle filtering approach, an actual market price estimate of each price bidding parameter is forecasted at every bidding hour of interest based on the updated stochastic constant drift model.
In accordance with an embodiment, the market price velocity is a ratio of a current bidding price of each product data with a previous bidding price of each product price data.
In accordance with an embodiment, the market price velocity determines a non-linear dynamics of each current price bidding parameter as maximum price at each bidding hour.
In accordance with an embodiment, the stochastic constant drift model increments each price bidding parameter by performing trial and error analysis on each product price data.
In accordance with an embodiment, updating the price direction of each product price close to each bidding hour by performing an arithmetic operation and updating the bidding factor when nonzero bidding instances is close to the forecasting bidding hour.
In accordance with an embodiment, updating the product price measurement variation close to each product price state at the forecast hour.
In another aspect, a method for forecasting intraday market price dynamics of energy using a particle-filtering approach is provided. The method includes receiving one or more product price data of electricity. Simultaneously, a plurality of price bidding parameters comprising of an average price, a maximum price, a minimum price for each market electrical energy data at every bidding hour are obtained for product data. Further, a stochastic constant drift model is created for the one or more product price data, wherein the stochastic constant drift model is updated at one or more non-zero bidding instances close to every bidding hour of interest based on a price measurement. Furthermore, the stochastic constant drift model for the plurality of price bidding parameters is analyzed based on one or more value change observed in each price bidding parameter at every bidding hour based on a market price velocity, wherein the stochastic constant drift model defines a next price bidding parameter value by incrementing a previous price bidding parameter value. Then, the price measurement is recomputed to update the stochastic constant drift model close to every bidding hour of interest and determining if each price bidding parameter lies within a predefined interval bounds based on a one or more Bollinger bands. Furthermore, by using a particle filtering approach, an actual market price estimate of each price bidding parameter is forecasted at every bidding hour of interest based on the updated stochastic constant drift model.
In accordance with an embodiment, the market price velocity is a ratio of a current bidding price of each product data with a previous bidding price of each product price data.
In accordance with an embodiment, the market price velocity determines a non-linear dynamics of each current price bidding parameter as maximum price at each bidding hour.
In accordance with an embodiment, the stochastic constant drift model increments each price bidding parameter by performing trial and error analysis on each product price data.
In accordance with an embodiment, updating the price direction of each product price close to each bidding hour by performing an arithmetic operation and updating the bidding factor when nonzero bidding instances is close to the forecasting bidding hour.
In accordance with an embodiment, updating the product price measurement variation close to each product price state at the forecast hour.
In yet another aspect, a non-transitory computer readable medium for receiving one or more product price data of electricity. Simultaneously, a plurality of price bidding parameters comprising of an average price, a maximum price, a minimum price for each market electrical energy data at every bidding hour are obtained for product data. Further, a stochastic constant drift model is created for the one or more product price data, wherein the stochastic constant drift model is updated at one or more non-zero bidding instances close to every bidding hour of interest based on a price measurement. Furthermore, the stochastic constant drift model for the plurality of price bidding parameters is analyzed based on one or more value change observed in each price bidding parameter at every bidding hour based on a market price velocity, wherein the stochastic constant drift model defines a next price bidding parameter value by incrementing a previous price bidding parameter value. Then, the price measurement is recomputed to update the stochastic constant drift model close to every bidding hour of interest and determining if each price bidding parameter lies within a predefined interval bounds based on a one or more Bollinger bands. Furthermore, by using a particle filtering approach, an actual market price estimate of each price bidding parameter is forecasted at every bidding hour of interest based on the updated stochastic constant drift model.
In accordance with an embodiment, the market price velocity is a ratio of a current bidding price of each product data with a previous bidding price of each product price data.
In accordance with an embodiment, the market price velocity determines a non-linear dynamics of each current price bidding parameter as maximum price at each bidding hour.
In accordance with an embodiment, the stochastic constant drift model increments each price bidding parameter by performing trial and error analysis on each product price data.
In accordance with an embodiment, updating the price direction of each product price close to each bidding hour by performing an arithmetic operation and updating the bidding factor when nonzero bidding instances is close to the forecasting bidding hour.
In accordance with an embodiment, updating the product price measurement variation close to each product price state at the forecast hour.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS
The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
FIG. 1 illustrates an exemplary block diagram of a system (alternatively referred to as market price forecasting system) for forecasting an intraday market price, in accordance with some embodiments of the present disclosure.
FIG. 2 is a schematic overview of the market price forecasting system to forecast intraday market price dynamics of energy pricing, in accordance with some embodiments of the present disclosure.
FIG. 3A and FIG.3B depicts a flow diagram illustrating a method to forecast an actual market price of electricity by utilizing a particle filtering approach using the system of FIG.1, in accordance with some embodiments of the present disclosure.
FIG. 4 illustrates a price bidding pattern captured at a predefined duration to analyze one or more price variations observed in each price bidding parameter using the system of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.5 shows a graphical analysis chart of an actual market price, and a price measurement using the particle filtering approach using the system of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.6 shows Bollinger bands of a current bidding price of each product data of the energy and rolling average of averaged prices and the maximum price values for different bidding hours using the system of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.7A shows the actual market price estimated for 3rd bidding hour product data at a predefined duration (14:00 hours) using the particle filtering approach using the system of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.7B shows the actual market price estimated for 3rd bidding hour product data at the predefined duration (14:00 hours) with a maximum price range using the particle filtering approach using the system of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.7C shows the actual market price estimated for 3rd bidding hour product data at the predefined duration (14:00 hours) with the maximum price range and a confidence interval using the particle filtering approach using the system of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.8A and FIG.8B illustrates forecasting intraday market price based on a long-term short memory (LSTM) architecture using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.9A shows one or more price variations based on a plurality of auto-regressive features using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.9B shows raw data correlated with zero aggregation at every bidding hour using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.9C shows negative correlations between a price bidding time and a one or more volume bids using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.9D shows positive correlations between the price bidding time and the one or more volume bids using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.10 shows partial auto correlations with averaged intraday market price using the system of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.11 shows mean absolute percentage error (MAPE) for intraday electricity price forecast using the LSTM for every bidding hour product data of electricity using the system of FIG.1, in accordance with some embodiments of the present disclosure.
FIG.12 shows price forecasts for every bidding hour product data having the actual price and the predicted price using the system of FIG.1, in accordance with some embodiments of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS
Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments.
Embodiments herein provide a method and system for forecasting intraday market price dynamics of energy using a particle-filtering approach. The method disclosed, enables to forecast accurately market price of energy or electricity. Intraday price dynamics are extremely complex, using data-driven features such as auto-regressive features and price modelling for market price forecasting. Market price dynamics are forecasted using either one of a particle filtering approach and a long-term short memory (LSTM) framework. The method of the present disclosure initially processes one or more product price data of electricity received as input by using a particle filtering approach to determine a price measurement which is referred as an economic indicator to model one or more product data actual market price. The market price is modeled for every product data by creating a stochastic constant drift model to forecast an actual market price of each product price of electricity at critical hours by using the particle filtering approach. The stochastic constant drift model determines a next price bidding parameter value by incrementing a previous price bidding parameter value. The disclosed system is further explained with the method as described in conjunction with FIG.1 to FIG.12 below.
Referring now to the drawings, and more particularly to FIG. 1 through FIG.12, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
FIG. 1 illustrates an exemplary block diagram of a system (alternatively referred to as market price forecasting system), in accordance with some embodiments of the present disclosure. In an embodiment, the system 100 includes one or more hardware processors 104, communication interface device(s) or input/output (I/O) interface(s) 106 (also referred as interface(s)), and one or more data storage devices or memory 102 operatively coupled to the one or more hardware processors 104. The one or more processors 104 may be one or more software processing components and/or hardware processors. In an embodiment, the hardware processors can be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the processor(s) is/are configured to fetch and execute computer-readable instructions stored in the memory. In an embodiment, the system 100 can be implemented in a variety of computing systems, such as laptop computers, notebooks, hand-held devices, workstations, mainframe computers, servers, a network cloud, and the like.
The I/O interface device(s) 106 can include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like and can facilitate multiple communications within a wide variety of networks N/W and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. In an embodiment, the I/O interface device(s) can include one or more ports for connecting a number of devices to one another or to another server.
The memory 102 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random-access memory (SRAM) and dynamic-random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. The memory 102 further comprises (or may further comprise) information pertaining to input(s)/output(s) of each step performed by the systems and methods of the present disclosure. In other words, input(s) fed at each step and output(s) generated at each step are comprised in the memory 102 and can be utilized in further processing and analysis.
FIG. 2 is a schematic overview of the market price forecasting system 100 for electricity pricing, in accordance with some embodiments of the present disclosure. The FIG.2 includes a plurality of components comprising a stochastic constant drift model 202, a price measurement 204, and a particle filtering 206.
The market price forecasting system 100 receives a one or more product price data of electricity or energy in real time. Each product price data of electricity is processed using the stochastic constant drift model 202.
The stochastic constant drift model 202 of the market price forecasting system 100 determines one or more price variations for the received inputs based on a plurality of price bidding parameters.
The price measurement 204 of the market price forecasting system 100 determines values obtained from immediate previous price values of each product data of the electricity and a one or more deviation observed.
The particle filtering 206 of the market price forecasting system 100 forecasts an actual market price estimate of each price bidding parameter at every bidding hour of interest. The particle filtering approach is enabled with a generalized Kalman filtering approach which provides advantage of forecasting next bidding time using a simulation process.
The present disclosure is further explained considering an example, where the system 100 forecasts the market price for electricity using the system of FIG.1 and FIG.12.
FIG.3A and FIG.3B depicts a flow diagram illustrating a method to forecast an actual market price by utilizing a particle filtering approach using the system of FIG.1, in accordance with some embodiments of the present disclosure. In an embodiment, the system 100 comprises one or more data storage devices or the memory 102 operatively coupled to the processor(s) 104 and is configured to store instructions for execution of steps of the method 300 by the processor(s) or one or more hardware processors 104. The steps of the method 300 of the present disclosure will now be explained with reference to the components or blocks of the system 100 as depicted in FIG.1 and FIG.12 and the steps of flow diagram as depicted in FIG.3A and FIG.3B. Although process steps, method steps, techniques or the like may be described in a sequential order, such processes, methods, and techniques may be configured to work in alternate orders. In other words, any sequence or order of steps that may be described does not necessarily indicate a requirement that the steps to be performed in that order. The steps of processes described herein may be performed in any order practical. Further, some steps may be performed simultaneously.
Referring now to the steps of the method 300, at step 302, the one or more hardware processors 104 receive one or more product price data of electricity as input. Considering an example where the system 100 receives the one or more product price data of electricity for forecasting intraday power market price estimate for renewable energy source (RES) generation. Initially, the method determines a bidding pattern across every bidding hour for all volumes of each product data.
Referring now to FIG. 4, where FIG.4 illustrates a bidding pattern for a predefined duration using the system of FIG.1, in accordance with some embodiments of the present disclosure. Majority of electricity bid occurs between either 6th bidding hour of a current bidding day or a 3rd bidding hour of the current bidding day, that is between six and three. It is observed that the relation between each bidding price and the volume of electricity product data is varied for different bidding instances using different sample populations. Here, the volume of electricity includes a 1 megawatt (MW), 10 kilowatts (KW), 100 megawatts (MW) and thereof depending upon user requirements.
Referring now to the steps of the method 300, at step 304 the one or more hardware processors 104 obtain a plurality of price bidding parameters comprising of an average price, a maximum price, a minimum price for each market electrical energy data at every bidding hour. Referring to the above example, where for each price data of electricity, the plurality of price bidding parameters is fetched. The price bidding parameters comprises of the average price, the maximum price, the minimum price for each market electrical energy data at every bidding hour.
Referring now to the steps of the method 300, at step 306 the one or more hardware processors 104 create a stochastic constant drift model for the one or more product price data. The stochastic constant drift model is updated at one or more non-zero bidding instances close to every bidding hour of interest based on a price measurement. Here, for each product data of electricity the stochastic constant drift model is created based on non-linear modeling for maximum price forecast at every bidding hour.
The stochastic constant drift model stabilizes variance for linear estimation to learn electricity price formation. However, non-linear modeling of bidding price at each bidding hour uses a metaphorical long-tailed distribution enabling estimation of next bidding price value through the simulation methods. Here, Kalman filtering approach make use of the price measurement equation as represented in Equation 1 and Equation 2 to model stochasticity inherent in price transition. Here Z_k refers to the price measurements,
X_k=F_k X_k-1+w_k ------------ Equation 1
Z_k=H_k X_k+v_k ---------------- Equation 2
In one embodiment, the price measurement (referring now to FIG.5) is a ratio of a current bidding price of each product data with a previous bidding price of each product price data. The measurement set for the maximum price forecasting are defined as values obtained from immediate previous bidding price values using a ? deviation. where, ? is a constant noise factor. The market price dynamics in intraday market are Markovian in nature and bidding price at each bidding hour can be drifted through few steps using a constant drift factor to arrive at the price measurement at each price bidding hour of forecast. However, while in actual scenarios, the value of ? is not predictable and is a non-constant factor, the use of reliable economic indicators helps in obtaining price measurements using the constant ? factor. Such modeling is of noisy, and in modeling the jumps at each bidding price. It is noted that constructing a trajectory helps to arrive at the price value for each bidding hour of forecast, starting at any previous bidding price value.
In another embodiment, price state distribution for the price measurement is used to obtain price transitions across each bidding hour using state transition Equation 1 and Equation 2. The stochastic constant drift model forecasts maximum price at each bidding hour of interest. Therefore, price transition at certain bidding hour is due to the metaphorical long-tailed distribution. Accordingly, for the state X_k Equation 1 and Equation 2 are not single valued but are density distributions. Likewise, the forecasted state X_k+1 is also the long-tailed distribution. The forecasted state X_k+1 is the maximum price range at each bidding hour of interest. Implementation of state transition is done using the generalized Kalman filtering approach called the particle filtering which is mentioned in other embodiments of the present disclosure.
Referring now to the steps of the method 300, at step 308 the one or more hardware processors 104 analyze, the stochastic constant drift model for the plurality of price bidding parameters based on one or more variations observed in each price bidding parameter at every bidding hour based on a market price velocity, wherein the stochastic constant drift model defines a next price bidding parameter value by incrementing a previous price bidding parameter value. The market price velocity is a ratio of a current bidding price of each product data with a previous bidding price of each product price data.
In one embodiment, the market price velocity is one of a key economic indicators typically used in stock market trading to explore non-linear dynamics of maximum price of hourly products at every bidding hour. As differed with the global correlations exploited through correlation and regression analysis, non-linear indicators are expected to observe the price movement for each hourly product data only along different bidding hours. The market price velocity may also be referred as a price momentum or a region of curve (ROC) is given in Equation 3,
ROC=((P_present/P_previous )-1)*100 ------------Equation 3
The market price velocity indicates better understanding of a maximum price velocity for the current bidding price of each product data under consideration across the bidding hours. Here, the maximum price forecast for every bidding hourly product may be considered as analogous to the closing price in stock markets along with average price estimation. This helps in understanding a skewness of the market price as one moves closer to the actual delivery time. A typical example of the maximum price for each bidding hourly product data of the electricity is given in below Table 1.
Table 1 – Maximum prices for hourly product at 12:00 hours on 17th October 2019 date at 6th bidding hour
52.77 53.02 52.0 50.9 51.8 51.07 51.8 54.19 50.0 59.2
It is considered that the hourly product 12.00 hours on 17th October 2019 as a sample instance. The market price velocity is calculated according to the Equation and is given in Table 2,
Table 2 – Momentum of maximum prices for hourly product data at 12:00 hours on 17th October 2019 date until at 6th bidding hour
-15.54 8.37 -4.41 -1.40 1.42 -1.73 2.16 1.96 -0.47
Here the current bidding prices are considered from the 15th bidding hour until to the 6th bidding hour. The first value in the Table 1 corresponds to the maximum price for the 6th bidding hour.
In one embodiment, the market price velocity determines the non-linear dynamics of each current price bidding parameter as maximum price at each bidding hour.
Referring now to the steps of the method 300, at step 310 the one or more hardware processors 104 recompute the price measurement to update the stochastic constant drift model close to every bidding hour of interest and determining if each price bidding parameter lies within a predefined interval bounds based on a one or more Bollinger bands. It is noted that the Bollinger bands (referring now to FIG.6) are quite popular in financial market analysis and help to understand the standard deviation of the prices for the duration under consideration.
FIG.6 shows Bollinger bands of a current bidding price of each product data of the energy and rolling average of averaged prices and the maximum price values for different bidding hours using the system of FIG.1, in accordance with some embodiments of the present disclosure. These are computed as rolling standard deviation from the average prices for the duration. Typically, in stock market analysis a 20-day window is considered to analyze the average prices and the Bollinger bands are computed accordingly. Here, the bidding duration is modified, and each bidding length is considered to be the number of bidding hour for which the price dynamics of hourly product needs to be observed.
Bollinger bands are two-band values for each hourly product. Bollinger up (Equation 4) is the upper standard deviation from the average price and Bollinger down (Equation 5) is the lower standard deviation from the average prices. These values are utilized to understand the evolution of maximum bidding price values with respect to rolling average mean and rolling the standard deviation.
Bollinger bands are computed as,
?Bollinger Bands ?_up=MA(IDprices,n)+m*s ------------- Equation 4
?Bollinger Bands ?_down=MA(IDprices,n)-m*s ------------- Equation 5
where,
MA refers to a moving average of the averaged bidding prices,
n is the number of bidding hours considered,
m is number of standard deviations, and
s refers to the standard deviation.
In intraday market data it is identified that not all bidding hours have bidding instances. In order to explore the non-linear dynamics of any sample instance, the absence of bidding instances affects the price modeling as a Markovian process. Design of noisy estimates for such bidding hour demonstrates efficiency of the method of the present disclosure in modeling the stochasticity inherent with price variation. The stochastic constant drift model for the price movement across every bidding hour is the equivalence of such design and forms the base for the method of the proposed disclosure.
The market price velocity indicator improves estimation of forecasts at critical bidding hour using the particle filtering approach (referring to Table 3). The Bollinger bands estimates are computed for these bidding hours, to understand the one or more variations of maximum price with respect to the Bollinger bands derived from the average hourly prices. Here, the maximum price is one of the price bidding parameter.
Referring now to the steps of the method 300, at step 312 the one or more hardware processors 104 forecast by using a particle filtering approach, an actual market price estimate of each price bidding parameter at every bidding hour of interest based on the updated stochastic constant drift model. Referring now to the above example, where the particle filtering approach (referring Table 3) forecasts actual market price estimate by performing the below steps,
Step 1 - A total number of bidding hours is selected to forecast the market price dynamics of each product price data.
Step 2 – The price measurement for the product price data is computed using the stochastic constant drift model based on the previous maximum price bidding parameter value at each bidding hour with an incremental factor.
In another embodiment, the maximum price forecasting (FIG.7A through FIG.7C) shows for each bidding hour product complements the average price spread prediction for better understanding of the market behavior. FIG.7C shows the actual market price estimated for 3rd bidding hour product data at the predefined duration (14:00 hours) with the maximum price range and a confidence interval using the particle filtering approach using the system of FIG.1, in accordance with some embodiments of the present disclosure. Considering the maximum price per bidding hour, for the given hourly product data. Here, the trajectory until 3rd bidding hour for every bidding hour product data is volatile and yet slow varying. This implies that the maximum price at each bidding hour can be constructed as noisy instance for the previous bidding hour’s maximum price represented in Equation 6,
P_(max,current )=P_(max,prev)±? ----------- Equation 6
Where, ? is empirically chosen using trial-and-error method and is a small increment value. Here, short bidding hour duration are set close to each bidding hour of forecast for such modeling.
Equation 6 indicates the price measurement as the economic analysis. As mentioned above, the price measurement is generally a constantly changing factor as observed in Table 2. However, the slow-varying momentum of the bidding price can be constructed using a constant drift factor, between the previous bidding hour and the current bidding hour of forecast. If the constant drift factor is chosen such that at the bidding hour of forecast is the price estimate through the slow-varying trajectory is the approximate of the actual price just before the hour of forecast.
The stochastic constant drift model increments each price bidding parameter by performing trial and error analysis on each product price data.
Step 3 - Price direction of each product price data is updated with an incremental value in a constant drift factor at each bidding hour to approximate the price measurement.
In one embodiment, the price direction of each product price is updated close to each bidding hour by performing an arithmetic operation and updating the bidding factor when nonzero bidding instances is close to the forecasting bidding hour.
Price measurement criterion updates the correct price direction made by adding or subtracting the ? factor when a non-zero bidding instance is close to the hour of prediction. However, the correction is such that the updated price value is still noisy version of the actual price value at the bidding hour. The change is with respect to the upward motion, or the downward motion of the bidding price values for each product date. Noise in the price measurement ensures a level of stochasticity necessary to forecast through simulations while change in trajectory allows simulations to learn the change in direction of the price movement. After obtaining the price measurements, the particle filtering approach drifts each bidding price state to the next one, using the price measurements as correction steps.
Step 4 - The one or more product data lagging at each bidding hour of interest is modeled by estimating next product price data based on one or sampling methods for the price measurement.
This modeling technique is relevant for the product data under consideration as many non-zero bidding instances across different bidding hours. In such case to model the price evolution the current day biding price is not possible. Moreover, the market price dynamics across bidding hour for each hourly product is an advantage of non-linear modeling.
Step 5 – Further, a likelihood is obtained by slicing a present state of the product price data and consecutive measurement state posteriors into multiple data and weighing each present state product data against corresponding consecutive measurement product data based on the density of the samples in the price measurement.
Step 6 - Weights of the product price are updated for resampling and obtaining a next state posterior state distribution.
Step 7 – The steps are iteratively performed until the actual bidding hour is forecasted using the mean of the forecasted posterior state is the actual product price and the Bollinger bands at each step as critical intervals.
Table 3 – Particle Filtering Approach
Step 1 : Select number of bidding hours steps to forecast hour
Step 2 : Obtain price measurements using stochastic constant drift,
P_(max,current )=P_(max,prev)±?
Step 3: Update price direction close to forecast hour
Step 4: Model price at bidding hour i: mean of metaphorical long-tailed distribution P_ID1
Step 5: Importance Sampling : n Draw n samples P_n^i from P_ID1
Step 6: Use probability p(.) measure to weigh next price measurement y_n wrt P_IDi
Step 7: update weights : w_n^i=P(y_n |P_n^i)
Step 8: Resample P_n^i according to w_n^i
Step 9: Assign from step 8 P_n^(i+1 ) ? P_n^i
Step 10 : Repeat Step 5 ? Step 8 until forecast hour
Further, the particle filtering approach is implemented for the non-linear stochastic modeling of maximum bidding prices at every bidding hour. While there are many computational methods to perform particle filtering, utilizing sampling importance resampling filter (SIR) filter, is equivalent of Bayesian bootstrap filter. The maximum bidding price at every bidding hour can be represented as the mean of the metaphorical price distribution. The price measurements obtained through the non-linear modeling methods are used to transit from one price state at each bidding hour to the next. SIR filter updates the maximum price state distributions from the previous state. This approach is executed for the required number of bidding hours to make the forecast at the bidding hour of interest.
Further, selected short range of bidding hours are used to run the particle filtering approach. This is because stochastic updating for state changes or non-linear for price measurements possibly degenerate or go off track for large steps. Accordingly, for every three bidding hour steps, prior to the bidding hour of interest the system 100 obtains price measurements and therefore the maximum price forecasts at the bidding hour of forecast. Here, the 6th bidding hour and the 3rd bidding hour are observed to be critical bidding hours with maximum bids occurring six hours before the delivery time. The results of the particle filtering approach on sample hourly products are compared with the LSTM architecture for every hourly product from 17th October 2019 for ease representation. Results are presented by the present disclosure with respect to the point forecasts of maximum price for the 6th bidding hour and the 3rd bidding hour for 12.00 hours product in FIG.7B and 14.00 hours product in FIG.7C. The maximum price estimated is shown against the Bollinger bands and the actual price spread for both 12.00 hours and 14.00 hours products in FIG.7A through FIG.7C.
It is observed that the price measurement used along with the particle filtering approach demonstrates reliable point estimates for maximum prices at the 3rd bidding hour and the 6th bidding hour. The non-linear price measurement method along with the stochastic state transition using the particle filtering approach results are reliable to understand price dynamics. While both the LSTM and the particle filtering approaches have been tested on the entire span of data, few sample instances are represented as experimental results.
FIG.8A and FIG.8B illustrates forecasting intraday market price based on a long-term short memory (LSTM) architecture using the system of FIG.1, in accordance with some embodiments of the present disclosure.
Further, forecasting the intraday market price utilizing long short-term memory (LSTM) modeling is another efficient framework for average price spread for hourly products. The LSTM architectures are popular choice in the field of deep learning, for time-series data analysis, classification, and prediction. LSTM are known to track long-term dependencies in the input time-series data. The LSTM process is shown in FIG.8A. Given the volatility of the intraday power market, learning from stationary features implies the need for reliable long-term memory mechanism and this is readily provided by the LSTM architecture.
FIG.8A uses only the intraday market variables to predict average price spread for hourly products. The bidding price of every hourly product is in relation with volume, the number of bids and the day of the week. It is observed from the regression analysis that the bidding prices are correlating with prices from 24, 48 and 72 lag hours. Accordingly, the market price data were sliced according to the volume of the bid, the day of the week, the hour of actual execution and the bidding time. The average price from previous day (for the same hour) and the present-day price information until 6th bidding hour provides prior knowledge for the current bidding time price forecasting.
FIG.8B depicts the LSTM architecture forecasting with implementation and evaluation results. However, it is to be noted that the use of LSTMs exceeds the need to learn the dependencies between these factors. The use of LSTMs in learning temporal dependencies is well known in the literature. As intraday power markets are quite volatile, it is important to learn the dynamic relation between averaged prices between consecutive bidding hours. Therefore, while average price spread is conditioned on features such as hour, day of the week, volume, number of bids through correlational analyses, the temporal dependencies between different bidding hours also impact the average price spread prediction at the bidding hour of forecast. The LSTM architecture uses the input data along with its previous price predictions for the hourly product to provide the next output. This ensures temporal dependencies are exploited along with features correlating with price prediction using the LSTMs. For the LSTM architecture, two LSTM layers and two dense layers are used. The features were fed to one LSTM layer and the output is obtained at the dense layer as the average price spread.
In one embodiment, impact of one or more latent factors for forecasting the market price dynamics using the LSTM architecture are determined by performing the steps mentioned below. Here, the one or more latent factors refer to the variables influencing price dynamics such as a temperature, a day of the week, a time of day, a volume bid, a number of bids and thereof.
Step 1 - extracting for the one or more product price data of electricity, a set of auto-regressive features comprising an average price data, a minimum price bidding time, a maximum price bidding time, a standard deviation of price for the bidding time, the day of the week, and climatic information comprising the temperature and a wind speed.
Step 2 - The set of auto-regressive features are fed as input to the LSTM architecture, via the one or more hardware processors, to compute an average product price for determining a bidding pattern of the product for consecutive days.
Step 3 - A correlation matrix is constructed using the set of auto-regressing features between the product price and one or more bidding instances.
Step 4 – Generating an auto correlation function plot by using the set of auto-regressive features to determine a positive correlation and a negative correlation between the current bidding product price and the previous bidding product price to determine the impact of different latent factors on each product price variations.
FIG.9A shows a one or more price variations based on a plurality of auto-regressive features using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure. Correlation matrices are determined between each bidding hour product prices and different factors. The correlation between the product prices at different bidding hours and the latent relations with respect to volume and the bidding instances is determined using the correlation matrices. Correlation matrices help to understand and gain insights in the global correlations between hourly product prices and different factors. The features drawn from each hourly product data are average price data, minimum and maximum prices at the different bidding hour, the standard deviation of the price for the bidding hour, day of the week and climatic information such as a temperature and a wind speed.
FIG.9B shows raw data correlated with zero aggregation at every bidding hour using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure. It is noted that the average price, the maximum price, and the minimum price share week correlations for a current day of the week, the bidding time, and the climatic information (temperature).
FIG.9C shows negative correlations between a price bidding time and a one or more volume bids using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure. The correlation matrix as in FIG.9C shows relation such as the bidding hourly products with the price bid, the volume bid, and the bidding hour (ID hour).
FIG.9D shows positive correlations between the price bidding time and the one or more volume bids using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure. FIG.9D further shows hourly products in which positive correlations exist between the bidding price, the biding volume, and every bidding hour.
FIG.10 shows partial auto correlations function on averaged intraday prices using the system of FIG.1, in accordance with some embodiments of the present disclosure. The auto-regressive features for averaged price data for hourly products are plotted using one or more partial auto correlation function (PACF) plots. These plots help in understanding the positive correlation and the negative correlation between different hourly products and products at different time lags. The average price data is the average of the prices for an hourly product and the entire duration of the data is considered to understand the correlational behavior across all bidding hourly products. The plots are expected to depict the relation between each hourly product bid at say, ith hour and an earlier bid at say i- k hour for the entire duration of consideration. For the year 2019, for 365 days, each day has 24 hourly products. The PACF plot is plotted for this entire data set. The corresponding plot is shown in FIG.10, it is observed that the average price for hourly products share maximal correlation with corresponding products bid at 24 hours, 42 hours, and 72 hours lag. From the above analysis it is to be noted that the day of the week, the bidding hour and the PACF features can be used as features, along with the volume bid and the number of bidding instances, previous day’s averaged hourly product price to determine the average price for the critical bidding hours. The extracted features are used as input to the LSTM architecture that would condition the spread of average prices for the hour of forecast on the volume bid and the number of bids, apart from exploiting the temporal dependencies on the market price data.
Evaluation of the LSTM architecture and the particle filtering approach used publicly available EPEX intraday market price dataset for the duration 2019-2020. These price data were about 365 single files for the year 2019 and six months of data for the year 2020 and each file has no missing values. The minimum volume bid placed is 0.1 MW and the maximum volume bid is 100 MW. It is observed that majority of the price bid are for the 0-1MW volume range.
In one embodiment, data processing for the LSTM architecture slices the data according to hour of the day, the day of the week and the volume bid. To forecast average price for each bidding hour of an hourly product the price information from the 6th bidding hour before the bidding instance was considered. Prices for the 12 hours before the bidding instance were clubbed and considered as one bidding hour. This was done as the number of bids are for twelve hours before the bidding hour are usually few in number. The previous days averaged price for the hourly product from the previous day is considered along with the volume of bid.
FIG.11 shows mean absolute percentage error (MAPE) for intraday electricity price forecast using the LSTM for every bidding hour product data of electricity using the system of FIG.1, in accordance with some embodiments of the present disclosure. The MAPE distribution for the 3rd bidding hour forecasts the particle filtering approach is reliable to a large extent. However, the stochastic model is robust and volatile in nature for the intraday power markets.
FIG.12 shows price forecasts for every bidding hour product data having the actual price and the predicted price using the system 100 of FIG.1, in accordance with some embodiments of the present disclosure.
The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
The embodiments of present disclosure herein addresses unresolved problem of forecasting market price of electricity. The embodiment thus provides method and system for forecasting intraday market price dynamics of energy using a particle-filtering approach. Moreover, the embodiments herein further provide, stochastic modeling to forecast actual market price at critical bidding hours using the particle filtering approach. Intraday price dynamics are complex, the data-driven features such as the auto-regressive features and data-appropriate metrics benefits price modelling and price forecasting.
It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g., any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g., hardware means like e.g., an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g., an ASIC and an FPGA, or at least one microprocessor and at least one memory with software processing components located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g., using a plurality of CPUs.
The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various components described herein may be implemented in other components or combinations of other components. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise.
Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., be non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
It is intended that the disclosure and examples be considered as exemplary only, with a true scope of disclosed embodiments being indicated by the following claims.
, Claims:We Claim:
1. A processor implemented method for forecasting an intraday market price dynamics of energy using particle-filtering approach, the method comprising:
receiving (302), via one or more hardware processors, one or more product price data of electricity;
obtaining (304), via the one or more hardware processors, a plurality of price bidding parameters comprising of an average price, a maximum price, a minimum price for each market electrical energy data at every bidding hour;
creating (306), via the one or more hardware processors, a stochastic constant drift model for the one or more product price data, wherein the stochastic constant drift model is updated at one or more non-zero bidding instances close to every bidding hour of interest based on a price measurement;
analyzing (308), via the one or more hardware processors, the stochastic constant drift model for the plurality of price bidding parameters based on one or more value change observed in each price bidding parameter at every bidding hour based on a market price velocity, wherein the stochastic constant drift model defines a next price bidding parameter value by incrementing a previous price bidding parameter value;
recomputing (310), via the one or more hardware processors, the price measurement to update the stochastic constant drift model close to every bidding hour of interest and determining if each price bidding parameter lies within a predefined interval bounds based on a one or more Bollinger bands; and
forecasting (312), by using a particle filtering approach, via the one or more hardware processors, an actual market price estimate of each price bidding parameter at every bidding hour of interest based on the updated stochastic constant drift model.
2. The processor implemented method as claimed in claim 1, wherein the market price velocity is a ratio of a current bidding price of each product data with a previous bidding price of each product price data.

3. The processor implemented method as claimed in claim 1, wherein the market price velocity determines a non-linear dynamics of each current price bidding parameter as maximum price at each bidding hour.

4. The processor implemented method as claimed in claim 1, wherein the stochastic constant drift model increments each price bidding parameter by performing trial and error analysis on each product price data.

5. The processor implemented method as claimed in claim 1, wherein forecasting the actual market price estimate of each price bidding parameter at every bidding hour of interest by using the particle filtering approach comprises:
selecting, a total number of bidding hours to forecast the market price dynamics of each product price data;
computing, the price measurement for the product price data using the stochastic constant drift model based on the previous maximum price bidding parameter value at each bidding hour with an incremental factor;
updating, a price direction of each product price data with an incremental value in a constant drift factor at each bidding hour to approximate the price measurement;
modeling, the one or more product data lagging at each bidding hour of interest by estimating next product price data based on one or sampling methods for the price measurement;
obtaining, a likelihood by slicing a present state of the product price data and consecutive measurement state posteriors into multiple data, and weighing each present state product data against corresponding consecutive measurement product data based on the density of the samples in the price measurement;
updating, weights of the product price for resampling and obtaining a next state posterior state distribution; and
iteratively performing, until the actual bidding hour is forecasted using the mean of the forecasted posterior state is the actual product price and the one or more Bollinger bands at each step as critical intervals.
6. The processor implemented method as claimed in claim 1, wherein updating the price direction of each product price close to each bidding hour by performing an arithmetic operation and updating the bidding factor when nonzero bidding instances is close to the forecasting bidding hour.

7. The processor implemented method as claimed in claim 1, updating the product price measurement variation close to each product price state at the forecast hour.

8. The processor implemented method as claimed in claim 1, further comprising determining an impact of one or more latent factors for forecasting the market price dynamics using an LSTM architecture by performing the steps of:
extracting for the one or more product price data of electricity, a plurality of auto-regressive features comprising an average price data, a minimum price bidding time, a maximum price bidding time, a standard deviation of price for the bidding time, a day of the week, and climatic information comprising a temperature and a wind speed;
feeding, the set of auto-regressive features to the LSTM architecture, via the one or more hardware processors, to compute an average product price for determining a bidding pattern of the product for consecutive days;
constructing, a correlation matrix using the set of auto-regressing features between the product price and one or more bidding instances; and
generating, an auto correlation function plot by using the set of auto-regressive features to determine a positive correlation and a negative correlation between the current bidding product price and the previous bidding product price to determine the impact of different latent factors on each product price variations.

9. A system (100), for forecasting an intraday market price dynamics of energy using particle-filtering approach comprising:
a memory (102) storing instructions;
one or more communication interfaces (106); and
one or more hardware processors (104) coupled to the memory (102) via the one or more communication interfaces (106), wherein the one or more hardware processors (104) are configured by the instructions to:
receive, one or more product price data of electricity;
obtain, a plurality of price bidding parameters comprising of an average price, a maximum price, a minimum price for each market electrical energy data at every bidding hour;
create, a stochastic constant drift model for the one or more product price data, wherein the stochastic constant drift model is updated at one or more non-zero bidding instances close to every bidding hour of interest based on a price measurement;
analyze, the stochastic constant drift model for the plurality of price bidding parameters based on one or more value change observed in each price bidding parameter at every bidding hour based on a market price velocity, wherein the stochastic constant drift model defines a next price bidding parameter value by incrementing a previous price bidding parameter value;
recompute, the price measurement to update the stochastic constant drift model close to every bidding hour of interest and determining if each price bidding parameter lies within a predefined interval bounds based on a one or more Bollinger bands; and
forecast by using a particle filtering approach, an actual market price estimate of each price bidding parameter at every bidding hour of interest based on the updated stochastic constant drift model.

10. The system as claimed in claim 9, wherein the market price velocity is a ratio of a current bidding price of each product data with a previous bidding price of each product price data.

11. The system as claimed in claim 9, wherein the market price velocity determines a non-linear dynamics of each current price bidding parameter as maximum price at each bidding hour.

12. The system as claimed in claim 9, wherein the stochastic constant drift model increments each price bidding parameter by performing trial and error analysis on each product price data.

13. The system as claimed in claim 9, wherein forecasting the actual market price estimate of each price bidding parameter at every bidding hour of interest by using the particle filtering approach comprises:
selecting, a total number of bidding hours to forecast the market price dynamics of each product price data;
computing, the price measurement for the product price data using the stochastic constant drift model based on the previous maximum price bidding parameter value at each bidding hour with an incremental factor;
updating, a price direction of each product price data with an incremental value in a constant drift factor at each bidding hour to approximate the price measurement;
modeling, the one or more product data lagging at each bidding hour of interest by estimating next product price data based on one or sampling methods for the price measurement;
obtaining, a likelihood by slicing a present state of the product price data and consecutive measurement state posteriors into multiple data, and weighing each present state product data against corresponding consecutive measurement product data based on the density of the samples in the price measurement;
updating, weights of the product price for resampling and obtaining a next state posterior state distribution; and
iteratively performing, until the actual bidding hour is forecasted using the mean of the forecasted posterior state is the actual product price and the one or more Bollinger bands at each step as critical intervals.
14. The system as claimed in claim 9, wherein updating the price direction of each product price close to each bidding hour by performing an arithmetic operation and updating the bidding factor when nonzero bidding instances is close to the forecasting bidding hour.

15. The system as claimed in claim 9, updating the product price measurement variation close to each product price state at the forecast hour.

16. The system as claimed in claim 9, further comprising determining an impact of one or more latent factors for forecasting the market price dynamics using an LSTM architecture by performing the steps of:
extracting for the one or more product price data of electricity, a plurality of auto-regressive features comprising an average price data, a minimum price bidding time, a maximum price bidding time, a standard deviation of price for the bidding time, a day of the week, and climatic information comprising a temperature and a wind speed;
feeding, the set of auto-regressive features to the LSTM architecture, via the one or more hardware processors, to compute an average product price for determining a bidding pattern of the product for consecutive days;
constructing, a correlation matrix using the set of auto-regressing features between the product price and one or more bidding instances; and
generating, an auto correlation function plot by using the set of auto-regressive features to determine a positive correlation and a negative correlation between the current bidding product price and the previous bidding product price to determine the impact of different latent factors on each product price variations.

Dated this 22nd Day of September 2022

Tata Consultancy Services Limited
By their Agent & Attorney

(Adheesh Nargolkar)
of Khaitan & Co
Reg No IN-PA-1086