DESC:FORM 2
THE PATENTS ACT, 1970
(39 of 1970)
&
The Patent Rules, 2003
COMPLETE SPECIFICATION
(See sections 10 & rule 13)
1. TITLE OF THE INVENTION
A SYSTEM AND METHOD FORECASTING FOR SOLAR PHOTOVOLTAIC POWER USING MACHINE-LEARNING

2. APPLICANT (S)
NAME NATIONALITY ADDRESS
DIVYASAMPARK IHUB ROORKEE FOR DEVICES MATERIALS AND TECHNOLOGY FOUNDATION IN Indian Institute of Technology Roorkee, Roorkee-247667, Uttarakhand, India.
3. PREAMBLE TO THE DESCRIPTION
COMPLETE SPECIFICATION

The following specification particularly describes the invention and the manner in which it is to be performed.

FIELD OF INVENTION:
[001] The present invention relates to the field of solar photovoltaic panel. The present invention in particular relates to a system and method forecasting for solar photovoltaic power using machine-learning.
DESCRIPTION OF THE RELATED ART:
[002] Gauging of force age from sustainable sources is critical for power framework combination and furthermore accommodating in age, transmission and dissemination for energy providers and power markets. The essential necessity of the activity and arranging of a service organization is to have exact model for age gauging.
[003] Reference may be made to the following:
[004] IN Publication No. 202341031911 relates to the short-term forecasts of renewable power generation are essential for effectively integrating renewable energy sources. With the waning and overrated petroleum product assets, the globe has at long last moved its concentration towards the utilization of environmentally friendly power Assets, chiefly Sun based Energy. In this time span, the world has likewise seen a flood in specialized developments in the field of information science and AI. Additionally, it turned out to be exceptionally fundamental for the energy business to anticipate the result of the sun based power and subsequently needed to utilize different AI procedures among different strategies. This work includes 24-hour ahead sun oriented and wind power anticipating utilizing AI calculations. Two AI calculations, to be specific Back spread brain organization and Irregular woods are tried with same dataset. As inexhaustible power age is profoundly reliant upon weather patterns thus, for this work meteorological information of specific area is taken as info information for preparing the organization. For assessment of determining model, a legitimate assessment measure has been utilized for both gauging model individually. Exhibitions of back spread and arbitrary woods calculations are thought about for summer, winter and blustery seasons for sun based power determining. As wind power doesn't rely upon seasons, complete 5 years information is taken for gauging. The model is likewise tried for the remarkable situations where sun oriented irradiance esteem changes radically to arbitrary qualities because of overcast cover.
[005] IN Publication No. 202241067929 relates to a hybrid model that integrates machine learning and statistical approaches is suggested for predicting future solar energy generation. In order to improve the accuracy of the suggested model, an ensemble of machine learning models was used in this study. The results of the simulation show that the proposed method has reduced placement cost, when compared with existing methods. When comparing the performance of an ensemble model that integrates all of the combination strategies to standard individual models, the suggested ensemble model outperformed the conventional individual models. According to the findings, a hybrid model that made use of both machine learning and statistics outperformed a model that made sole use of machine learning in its performance.
[006] IN Publication No. 202311019272 discloses a smart solar energy monitoring. The system comprises of various electrical components such as solar panel, LCD display, potentiometer, DC motor, current sensor and Arduino Uno. The working of the system comprises of 24V solar panel supplies current through Potentiometer to ACS712 current sensor, Current sensor capture the current value at a range of 0-1024 (analog value). Current Sensor ACS712 is connected with additional DC 5V. ACS712’s IP- pin then pass that current to DC Motor and the DC motor another terminal is connected with Solar Panel’s negative end. By utilizing this system a user can track temperature, humidity and solar power generation. The controlling system is very easy to handle and results in efficient output.
[007] IN Publication No. 202241067474 relates to large-scale renewable energy plants, the future of solar power forecasting is vital to their success. For reliable predictions of solar electricity generation one must take into consideration changes in weather patterns over time. A hybrid model that integrates machine learning and statistical approaches is suggested for predicting future solar energy generation. In order to improve the accuracy of the suggested model, an ensemble of machine learning models was used in this study. The results of the simulation show that the proposed method has reduced placement cost, when compared with existing methods. When comparing the performance of an ensemble model that integrates all of the combination strategies to standard individual models, the suggested ensemble model outperformed the conventional individual models. According to the findings, a hybrid model that made use of both machine learning and statistics outperformed a model that made sole use of machine learning in its performance.
[008] IN Publication No. 202211063868 relates to sunlight-based energy is unreservedly accessible and can be disengaged by introducing an enormous sun-oriented power plant. This way, such PV sunlight-based plants are critical supporters of cutting the energy shortage in distant regions. For a reason for expectation, the Neural model was created; we anticipated the upsides of the exhibition proportion (PR), creation sum (MWh), and plan of the exhibit (POA) of the sunlight-based plant for years. This AI expectation procedure is exceptionally compelling and productive, contrasted with other conventional expectation and determining strategies, for assessing the exhibition of the sun based power plant and the future situation with the sun oriented power plant. Because of these outcomes, the model for anticipating sun-oriented power age is compelling and can be used for any sun-based power plant.
[009] IN Publication No. 202221051843 relates to large-scale renewable energy plants, the future of solar power forecasting is vital to their success. For reliable predictions of solar electricity generation, one must take into consideration changes in weather patterns over time. In this paper, a hybrid model that integrates machine learning and statistical approaches is suggested for predicting future solar energy generation. In order to improve the accuracy of the suggested model, an ensemble of machine learning models was used in this study. The results of the simulation show that the proposed method has reduced placement cost, when compared with existing methods. When comparing the performance of an ensemble model that integrates all of the combination strategies to standard individual models, the suggested ensemble model outperformed the conventional individual models. According to the findings, a hybrid model that made use of both machine learning and statistics outperformed a model that made sole use of machine learning in its performance.
[010] IN Publication No. 202211053233 relates to photovoltaic (PV) technology will be crucial in future energy generation. Advances in computing technology, data gathering and storage, and data-driven algorithms have boosted machine learning. We examine machine learning approaches for PV systems. First, electrical and thermal models for PV systems are presented. Machine learning is then used to analyse PV systems. We examine how machine learning might help mankind reach a cleaner environment in the global march towards carbon neutrality. This article examines the problems and potential of machine learning for PV system analysis. Many small-scale PV technologies, such as building integrated photovoltaic thermal systems (BIPV/T), have not yet benefited completely from machine learning in terms of system efficiency and economic sustainability. Machine learning might help PV systems produce more sustainable electricity.
[011] IN Publication No. 202241054828 relates to improve energy management services and distribution of renewable energy sources, new innovations in ML technology are needed to produce accurate learning models that can be used in the energy analysis process, such as monitoring, prediction, forecasting, scheduling and decision-making. The complexity of the problems in the smart grid system, which includes uncertainty and non-linearity, affects the more complex the energy data structure generated. Therefore, the simple ML method will not be able to perform the Learning process because it is limited to simple raw data processing. In this paper, Deep Neural Network (DNN) method will be developed using Long Short Term Memory (LSTM) as a Learning model to provide Future Accurate Prediction (FAP) on electricity use and on renewable energy plants. Prediction test using Confusion Matrix accuracy value and RMSE error value.
[012] IN Publication No. 202241058112 relates to solar power forecasting will have a significant impact on the future of large-scale renewable energy plants. Predicting photovoltaic power generation depends heavily on climate conditions, which fluctuate over time. In this research, we propose a hybrid model that combines machine-learning methods with Theta statistical method for more accurate prediction of future solar power generation from renewable energy plants. The machine learning models include long short-term memory (LSTM), gate recurrent unit (GRU), Auto Encoder LSTM (Auto-LSTM) and a newly proposed Auto-GRU. The experiments show that the proposed MLSHM, using all the combination methods, achieved higher accuracy compared to the prediction of the traditional individual models. Results demonstrate that a hybrid model combining machine-learning methods with statistical method outperformed a hybrid model that only combines machine-learning models without statistical method.
[013] IN Publication No. 202241063799 relates to an output power prediction of building integrated semi-transparent photovoltaic system using machine learning algorithm, comprises of, a predicting how well a solar photovoltaic system will function requires knowledge of factors such as location, climate, and data quality, it is a 5 crucial part of the most accurate method of forecasting. When making projections on a sub hourly time scale, satellite images can be quite helpful and therein, satellite data, on the other hand, can potentially anticipate PV output on a climatological time scale or provide projections up to six hours in advance. The output from PV systems can be converted using a number of different methods, such as deterministic models with three or five parameters, 10 parametric models, or other machine learning strategies.
[014] IN Publication No. 202241047924 relates to a method of forecasting solar irradiance on a photovoltaic (PV) panel comprising: collecting primary dataset from a region of interest, normalising data in the collected dataset, segregating plurality of variables from the normalised data into an input and output categories, employing a primary regressor for predicting a first set of values, initialising a secondary regressor with a relative error dataset obtained from the first set of values, deploying a reinforcement learning based RNN for kernel selection of the secondary regressor used to check for error mapping, and predicting overall outcome in real-time using combination of the primary and the secondary regressors output, wherein the method reduces overall memory requirements by minimizing number of feature variables in the input dataset and provides flexibility in prediction interval with increased accuracy of the forecasted values.
[015] IN Publication No. 202231045057 discloses a deep learning based enhanced solar energy forecasting with AI-driven IoT system. The present invention is comprised of, but not limited to, an Artificial Intelligence based receiving means for providing weather forecast data, the weather forecast data including a plurality of weather features; a processing unit in an IoT environment for processing the weather forecast data using a chain of a plurality of processing blocks of an artificial neural network to derive one or more of the plurality of weather features, wherein each of the plurality of processing blocks is having a neural network layer, an activation unit, and a pooling unit, wherein the neural network layer associates a filter to a region of the weather forecast data across a plurality of neural network layers in the weather forecast data; and an output unit for determining a solar power forecast for enhancing solar energy with AI-driven IoT connected modules through the derived weather features.
[016] IN Publication No. 202111060836 discloses a system for solar PV forecasting using machine learning interface and method thereof. The system includes, but not limited to, one or more processing unit connected with a memory unit configured to execute: a machine learning based weather module for collecting forecast meteorological data for a predetermined geography; a plurality of sensors measuring solar irradiance parameters for the predetermined geography over predetermined time intervals to form a data set; a solar irradiance unit for estimating irradiance levels using parameters collected from the machine learning based weather module; a physical characteristics module associated with the solar PV cell of a solar energy generating system in the predetermined geography; and a user interface provided on a computing device for viewing and analysing solar energy production using the collected forecast meteorological data, estimated irradiance levels, and physical characteristics of the solar energy generating system.
[017] IN Publication No. 202011047662 discloses a processing system having machine learning for solar radiation estimation, in a pre-determined territory over predefined time intervals to form an optimized input data values. The system includes a processing unit; multiple transducers to measure current solar radiation parameters; a display device to show output data values; and a memory disposed in communication with the processing unit and storing processing unit executable instructions. Further, the instructions comprising instructions to: a data acquisition unit, which consists of the plurality of transducers and record the desired data and convert it in a desired format and presented to the processor; further, the processing unit is connected with the display device and the set of interfaces with the hardware.
[018] IN Publication No. 202011044299 discloses a system for forecasting solar irradiance for a solar-based photovoltaic system and method thereof, in a defined geographical region over predetermined time intervals to form a data set. The system includes a data processing unit; an array of sensors to measure current solar irradiance parameters; a display unit to show output data values; a set of interfaces with a hardware compatible to the processor; and a memory disposed in communication with the data processing unit and storing processor executable instructions. Further, the instructions comprising instructions to: a data acquisition unit, which consists of the several sensors and capture the desired data and convert it in a desired format and presented to a data processing unit; the data processing unit is connected with display unit and the set of interfaces with the hardware.
[019] IN Publication No. 202011045281 discloses a system for solar radiation estimation for a solar-based power system and method thereof, in a defined geographical region over predetermined time intervals to form a data set. The system includes a processor; a plurality of sensors to measure current solar radiation parameters; a display unit to show output data values; and a memory disposed in communication with the processor and storing processor executable instructions. Further, the instructions comprising instructions to: a data acquisition unit, which consists of the plurality of sensors and record the desired data and convert it in a desired format and presented to the processor; further, the processor is connected with display unit and the set of interfaces with the hardware.
[020] IN Publication No. 201847014056 relates to a power network (SN) having a photovoltaic system (PV), a time curve (L(T)) of a light radiation (L) of the photovoltaic system (PV) is determined, wherein an increase (A) of the time curve (L(T)) in relation to a reference curve (R(T)) of the light radiation (L) is detected. As a result of the detection of the increase (A), a preparatory measure (VS) is then introduced in order to prepare the power network (SN) for an upcoming power drop of the photovoltaic system (PV).
[021] IN Publication No. 9662/CHENP/2012 relates to a solar power forecasting system can provide forecasts of solar power output by photovoltaic plants over multiple time frames. A first time frame may be several hours from the time of the forecast which can allow utility personnel sufficient time to make decisions to counteract a forecasted shortfall in solar power output. For example the utility personnel can decide to increase power production and/or to purchase additional power to make up for any forecasted shortfall in solar power output. A second time frame can be several minutes from the time of the forecast which can allow for operations to mitigate effects of a forecasted shortfall in solar power output. Such mitigation operations can include directing an energy management system to shed noncritical loads and/or ramping down the power produced by the photovoltaic plants at a rate that is acceptable to the utility to which the photovoltaic plants provide power.
[022] IN Publication No. 10076/CHENP/2012 relates to a solar power forecasting system can provide forecasts of solar power output by photovoltaic plants over multiple time frames. A first time frame may be several hours from the time of the forecast which can allow utility personnel sufficient time to make decisions to counteract a forecasted shortfall in solar power output. For example the utility personnel can decide to increase power production and/or to purchase additional power to make up for any forecasted shortfall in solar power output. A second time frame can be several minutes from the time of the forecast which can allow for operations to mitigate effects of a forecasted shortfall in solar power output. Such mitigation operations can include directing an energy management system to shed noncritical loads and/or ramping down the power produced by the photovoltaic plants at a rate that is acceptable to the utility to which the photovoltaic plants provide power.
[023] IN Publication No. 42/DEL/2010 relates to a method of forecasting the electrical production of a photovoltaic device comprising photovoltaic modules comprising a first part of estimating the lighting that will be received in the plane of the photovoltaic modules and a second part of estimating the electrical production of the photovoltaic device, characterized in that it comprises the following first step: (El) determination of whether a period concerned is sunny or cloudy, and characterized in that it comprises the following second step (E2) of implementing at least one of the following two steps: (E2) correction of the second part of the method of forecasting the electrical production based on the measurement of the true electrical production of the photovoltaic modules if the period concerned is sunny; and/or - correction of the first part of the method of forecasting the electrical production based on the measurement of the true electrical production of the photovoltaic modules if the period concerned is cloudy.
[024] IN Publication No. 9408/DELNP/2012 relates to methods and apparatuses to generate a forecast based on generalized differentiation or integration including but not limited to non-integer or variable order differentiation or integration.
[025] Publication No. WO2023054476 relates to a method for predicting solar-generated power, a device for predicting solar-generated power, and a solar power generation system with which it is possible to predict the power generated at a desired date and time. This method for predicting solar power generation includes executing, in order: a first step for saving past data in which weather data and solar cell power generation data outputted via a PCS at least one year before a date subject to prediction are associated with one another; a second step for calculating a clear-weather power generation curve from the past data; a third step for obtaining a yearly-transition curve indicating the yearly transition of the total amount of power generated; a fourth step for obtaining a clear-weather power generation model from the clear-weather power generation curve in conformance with the yearly-transition curve; a fifth step for obtaining a clear-weather power generation model for the entirety of one year; a sixth step for obtaining a weather coefficient; and a seventh step for acquiring a weather forecast and obtaining predicted power generation from the clear-weather power generation model and the weather coefficient.
[026] Publication No. US2023103959 relates to a method for generating a solar power output forecast for a solar power plant, comprising: using a processor, in a training mode, generating a trained artificial intelligence model using historical output data and historical input data including historical physical subsystem input data and historical physical subsystem forecasts for the solar power plant; in a runtime mode, for a predetermined forecast horizon, applying the trained artificial intelligence model to current input data including current physical subsystem input data and current physical subsystem forecasts for the solar power plant to produce the solar power output forecast; and, presenting the solar power output forecast on a display.
[027] Publication No. KR20220162360 relates to a method and apparatus for predicting the amount of solar power generation. The solar power generation prediction method includes determining the sun's trajectory using an image of a sky area in an area where a solar power generation system is installed; Tracking factors that inhibit solar power generation that exist within a predetermined radius on the determined sun's orbit; and estimating the amount of power generated by the photovoltaic power generation system by predicting a state change of the tracked solar power generation inhibitor.
[028] Publication No. KR20210147366 relates to a system for predicting a solar power generation amount to predict the solar power generation amount of a prediction target day comprises: a similar day selection module selecting at least one similar day for predicting the solar power generation amount of the prediction target day among candidate similar dates in a similar day selection range set on the basis of the prediction target day based on weather forecast data of the prediction target day; and a solar power generation amount prediction module predicting the solar power generation amount of the prediction target day based on past solar power generation amount data of the similar day selected by the similar day selection module and the weather forecast data of the prediction target day.
[029] Publication No. JP2021189154 relates to a photovoltaic power generation panel for more than a few days, the accuracy of forecasting an amount of photovoltaic power generation during snow cover is poor just by using forecast information of an amount of snowfall. A photovoltaic power generation amount prediction device comprises: a data receiving unit that receives an actual power generation amount measurement value, a snowfall amount forecast value, and a solar radiation amount forecast value of a photovoltaic power generation panel; a solar radiation amount conversion unit that converts the solar radiation amount forecast value into a predicted power generation amount according to the photovoltaic power generation panel; a snow cover estimation unit that estimates a current snow depth calculated by comparing the predicted power generation amount and the actual measured value of power generation, and a snow depth after the present based on the snowfall amount forecast value; and a power generation amount prediction correction unit that corrects the predicted power generation amount based on the snow depth estimated by the snow cover estimation unit.
[030] Publication No. TW202123603 relates to a method for solar power forecasting, wherein the method comprising steps of collecting solar radiation data for a period of time; storing the collected solar radiation data into a preliminary matrix, Y which further comprising a first matrix, M[alpha]; and processing said stored first matrix, M[alpha] data to obtain forecast of solar power.
[031] Publication No. KR20210077474 relates to a solar power generation amount prediction method is performed by a solar power generation amount prediction device according to an embodiment of the present invention. The solar power generation amount prediction method comprises the following steps of: receiving past solar power generation amount data, weather observation data, and weather forecast data; and predicting a solar power generation amount by using a machine learning model learned based on the past solar power generation amount data, the weather observation data, and the weather forecast data. The machine learning model includes a plurality of long short term memory (LSTM) layers and a repeat vector disposed there between.
[032] Publication No. KR20200057821 relates to an apparatus for predicting a generation amount of solar photovoltaic power based on big data analysis which may collect various data on solar photovoltaic power and systematically apply various analysis schemes including multiplex artificial neural network analysis based on the collected data to forecast power generation, and confirm factors affecting the prediction of the generation amount of the solar photovoltaic power for every analysis to increase prediction accuracy with respect to various environmental changes, and a method thereof. In consideration of various climate factors associated with outside temperature, quantity of solar radiation, humidity and the like affecting the solar photovoltaic power generation and various factors including a physical change of a machine such as pollution and temperature of a module, by providing volatility prediction of power yield according to a climate state and physical elements of a machine through accurate power generation prediction and influence factor analysis and fundament information for efficient load operation and management of a power generation system, the fundament information may be widely used as fundament data to increase economics of an existing installation area, to review business of a solar power station, to determine an intention.
[033] Publication No. US2020257019 relates to measurements are simulated of direct normal irradiance, diffuse horizontal and global horizontal irradiance from groups of two or more photovoltaic arrays and/or irradiance sensors which are located in close proximity to each other and which have different tilt and azimuth angles. Irradiance measurements derived from solar power system power measurements are combined with measurements made by irradiance sensors to synthesize an image of ground level global horizontal irradiance which can be used to create a vector describing motion of that image of irradiance in an area of interest. A sequence of these irradiance images can be transformed into a time series from which a motion vector can be derived. The motion vector can be applied to a current image of ground level irradiance and that image can be projected to a future point in time to provide a solar radiation forecast. These forecasts can be converted into forecasts of solar power system power in the area of interest.
[034] Publication No. KR102092860 relates to a machine learning based photovoltaic power generation value prediction device which does not use future weather forecast data and a method thereof. The present invention suggests a technique of predicting not a photovoltaic power generation amount in a peak time zone by using the future weather forecast data with uncertainty, but a photovoltaic power generation amount in the peak time zone of the day based on weather data measured in a time zone before the peak time zone of the day. Therefore, the present invention can support a manager of a photovoltaic power generation plant to more accurately predict the photovoltaic power generation amount in the peak time zone.
[035] Publication No. WO2019021438 relates to a power generation amount record acquisition unit acquires, from a power amount measurement device, a power generation amount record for the amount of power generated by a solar power generation system. An extraterrestrial solar radiation amount calculation unit calculates the extraterrestrial solar radiation amount on a designated day and in a designated time period at the location at which the solar power generation system is installed. A weather information acquisition unit acquires weather forecast information for each time period of a day for which a prediction is to be made. A power generation amount record classification unit classifies power generation amount coefficients, obtained by dividing a power generation amount by an extraterrestrial solar radiation amount, into a given number of power generation amount record groups associated with weather attributes. A predicted power generation amount calculation unit calculates a predicted power generation amount on the basis of the extraterrestrial solar radiation amount in each time period for which a prediction is to be made, and the power generation amount record group associated with a weather attribute corresponding to the weather forecast information.
[036] Publication No. CN116093932 relates to a photovoltaic power generation prediction method and system based on fusion information, and relates to the field of digital energy; the steps of the method are: S1: using a time convolutional neural network TCN to obtain corresponding future time series features according to historical time series photovoltaic power generation data; S2: Use convolutional neural network ResNet-50 to extract environmental features according to the same historical time based on environmental data and calculate self-attention weights according to the environmental features.
[037] Patent No. US11545830 relates to a photovoltaic system can include multiple photovoltaic power inverters that convert sunlight to power. An amount of power for each of the inverters can be measured over a period of time. These measurements, along with other data, can be collected. The collected measurements can be used to generate artificial neural networks that predict the output of each inverter based on input parameters. Using these neural networks, the total solar power generation forecast for the photovoltaic system can be predicted.
[038] Publication No. CN107766990 relates to a photovoltaic power station generation power prediction method comprising the steps that six meteorological characteristics are daily extracted by using the historical meteorological data of a photovoltaic power station so that a meteorological characteristic library is established; the daily characteristic data in the meteorological characteristic library are clustered through a KFCM algorithm so as to realize weather type classification, and class marking is performed on the daily power data and the meteorological data; an SVR sub-model is established for each class of power data and meteorological data according to the class mark; the weather type of the target day is identified by using the SVM through the target day weather characteristics provided by numerical weather prediction and the corresponding SVR sub-model is selected; an ARIMA model is established by using the real-time monitoring data of the target day, and real-time prediction of the irradiation intensity and the temperature can be realized by using the rolling prediction model; and the prediction values of the irradiation intensity and the temperature are inputted to the selected SVR sub-model so that the photovoltaic power station power prediction result can be obtained. The photovoltaic power station generation power prediction accuracy can be enhanced.
[039] Publication No. CN113159102 relates to a multi-time-scale photovoltaic power prediction method and system. The method comprises the steps of obtaining historical power data; inputting historical power data into the trained power prediction model, and outputting a power prediction result at a to-be-predicted moment; and taking the power prediction result at the to-be-predicted moment as the input of the trained power prediction model to carry out rolling prediction, and outputting the predicted power of the to-be-predicted time period. The ultra-short-term prediction result and the short-term prediction result of the photovoltaic power are output at the same time.
[040] Publication No. CN113837434 relates to a solar photovoltaic power generation prediction method and device, electronic equipment and a storage medium, and belongs to the field of solar photovoltaic power generation. The method comprises the steps that: environmental data and PV power are collected to form a data set, wherein the environmental data and the PV power have time sequence characteristics; data preprocessing is performed on the data set; and the preprocessed data are input into a pre-trained mixed deep learning model, and a prediction result is obtained, wherein the mixed deep learning model comprises a variational mode decomposition model, a nested long and short term memory artificial neural network and a full connection layer. According to the embodiment of the invention, a plurality of models are integrated, the complexity between the photovoltaic data and the environmental data can be well processed, and the advantages of the time sequence characteristic and the hybrid deep learning model are combined, so that the prediction precision is improved, and the photovoltaic prediction result is more timely and accurate.
[041] Publication No. CN115545257 relates to a distributed photovoltaic power prediction model method based on a generative adversarial framework, and the method comprises the steps: firstly forming a historical power matrix (HPM) through employing continuous historical power data before a prediction day; secondly, in order to better utilize structural data and random information hidden in HPM, a distributed low-rank decomposition algorithm based on matrix low-rank attributes is provided. Thirdly, two pieces of gradient information of the HPM are calculated, and the purpose is to provide the fluctuation trend of power output for the prediction model; and finally, combining all the contents into a tensor as model driving data. In addition, in view of the fact that the PV system is highly sensitive to weather, a Meteorological Guidance Vector (MGV) is constructed to assist in training and execution of the prediction model.
[042] Publication No. KR102054163 relates to a photovoltaic power generation prediction system. The photovoltaic power generation prediction system comprises: a sensor unit for collecting surrounding information of a solar module; a first data collection unit for collecting first data information on weather information of the meteorological agency; a second data collection unit for collecting second data information on fine dust, lightning, and clouds from the sensor unit; a third data collection unit for collecting third data information on an surrounding object of the solar module and the information collected from the surrounding object from the sensor unit; and a prediction model generation unit for receiving the first data information, the second data information, and the third data information to predict a power generation amount by using a machine running technique. The present invention can effectively measure the power generation amount of sunlight by reflecting data information around the solar module in addition to climate information.
[043] In contrast, one patent (CN113837434) focused on the importance of data pre-processing in solar forecasting. They used a variational mode decomposition (VMD) to decompose the complex time series. However, this requires the need to preset the penalty term and the number of modes or IMFs, which is one of the main limitations of this technique.
[044] Additionally, breaking the complex series into several subseries increases the complexity of the forecaster due to the need to predict several IMFs. However, no solution has been found in the prior art for dealing with this shortcoming of the decomposition-based models. The invention proposes a noise-assisted decomposition approach that does not require presetting penalty term and IMFs. Additionally, the presents an approach to dealing with the complexity of decomposition-based models by reducing the large pool of IMFs. This subseries reduction approach fills the major research gap of previous work utilizing decomposition techniques. Moreover, long-short term memory (LSTM) is a generally used DL technique in the prior art. However, present invention applies an updated attention mechanism-based LSTM (amLSTM) for accurate day-ahead solar forecasting. The end-product is a web-based platform for day-ahead PV power prediction in a 15-minute time resolution. It comprises a high performance forecasting model (NA-MEMD-FFT-amLSTM) and a system for data collection. The data collection stage is crucial for better fitting of machine learning algorithms.
[045] In order to overcome above listed prior art, the present invention aims to provide system and method forecasting for solar photovoltaic power using machine-learning.
OBJECTS OF THE INVENTION:
[046] The principal object of the present invention is to provide a system and method forecasting for solar photovoltaic power using machine-learning.
[047] Another object of the present invention is to provide a hand held system for day-ahead PV power prediction in a short time resolution.
[048] Yet another object of the present invention is to reduce the complexities involved with the decomposition method using fast Fourier transform in the field of solar power forecasting.
[049] Yet another object of the present invention is to combine NA-MEMD, FFT and attention mechanism-based LSTM for application in solar PV forecasting.
SUMMARY OF THE INVENTION:
[050] The present invention relates to the system and method forecasting for solar photovoltaic power using machine-learning. The invention is focused on addressing the challenge of accurately forecasting day-ahead PV power in 15-minutes time resolution. It comprises a setup for data collection and a novel high performance forecasting method.
[051] Weather stations are installed at different locations. The data is gathered using sensors installed at the weather station, serving as the foundation for training our model. The invention involves decomposition-based processing, where the data is broken down into multiple signals known as IMFs (Intrinsic Mode Functions). These IMFs, remaining in the time domain, preserve extensive information across various frequencies, ensuring the conservation of critical data for future power prediction.
[052] The system integrates the AM-LSTM model into decomposition-based processing, it synergistically harness the strengths of both methodologies. The attention mechanisms in the AM-LSTM model enable it to attend to significant patterns within the IMFs, enhancing the predictive capabilities of the entire system.
BREIF DESCRIPTION OF THE INVENTION
[053] It is to be noted, however, that the appended drawings illustrate only typical embodiments of this invention and are therefore not to be considered for limiting its scope, for the invention may admit to other equally effective embodiments.
[054] Figure 1 shows forecasting system, according to the present invention;
[055] Figure 2 shows flowchart according to the present invention.
DETAILED DESCRIPTION OF THE INVENTION:
[056] The present invention provides a system and method forecasting for solar photovoltaic power using machine-learning. The invention is focused on addressing the challenge of accurately forecasting day-ahead PV power in 15-minutes time resolution. It comprises a setup for data collection and a novel high performance forecasting method.
[057] Weather stations are installed at five different locations. Figure 1 shows the installed weather station mounted on cemented blocks. The solar and weather data are collected using six different sensors namely a pyranometer, passive cloud cover detector, rainfall sensor, air temperature and relative humidity sensor, and wind speed sensor. The weather station comprises a solar panel and a battery that provides power to all the sensors. The whole structure is mounted on cemented blocks through steel pipes. The system comes with a waterproof rating and an operating temperature range of -40oC to 75oC. The air temperature sensor has a range of - 40oC to 123.8oC with a resolution of 0.01oC. The relative humidity sensor has a range of 0 to 100% with a resolution of 0.5% RH typical. The wind speed sensor is a 3-cup anemometer type with a range of 0 to 67m/s and a resolution of 0.1 m/s. The solar irradiation sensor is having a spectral range of 305 to 2800nm and a measuring range of 0 to 2000 W/m2. The passive cloud cover sensor is having an output of 0 to 1V for CCF 0 to 100%, respectively. The access panel is to get the data manually through a data shuttle. The data is collected in 5 minutes resolution and uploaded to cloud storage, such that it can be accessed remotely from anywhere.
[058] The system consists of input unit for collecting dedicated data for solar photovoltaic (PV) forecasting, and a forecasting process as shown in Figure 2. The process comprises multivariate decomposition techniques, fast Fourier transform and a state-of-the-art machine learning technique. In contrast to the prior art data collection approach from different satellite sources, this demonstrates data collection through ground-based monitoring stations. It collects data on parameters such as solar data (GHI), temperature, wind speed, rainfall, humidity, and cloud cover. In addition, this invention is presenting a hybrid system of three diverse techniques that are, data decomposition, transformation, and machine learning. The overall hybrid system can be summarized as NA-MEMD-FFT-amLSTM.
[059] In this system, NA-MEMD-FFT-amLSTM deals with the non-linear and non-stationary issues of complex multivariate time series data using a noise-assisted multivariate empirical mode decomposition technique (NA-MEMD). Next, fast fourier transform (FFT) is applied to reduce the multi-dimensional data into only three sets of subseries. This subseries reduction approach fills the major research gap of prior art utilizing decomposition techniques. Additionally, in place of simple long-short term memory (LSTM), this invention applies an attention mechanism-based LSTM (amLSTM) for accurate day-ahead solar forecasting. The product is a hand held system for day-ahead PV power prediction in a 15-minute time resolution.
[060] The data is gathered using sensors installed at the weather station, serving as the foundation for training our model. The invention involves decomposition-based processing, where the data is broken down into multiple signals known as IMFs (Intrinsic Mode Functions). These IMFs, remaining in the time domain, preserve extensive information across various frequencies, ensuring the conservation of critical data for future power prediction.
[061] The decomposition of IMFs allows us to maintain the relationship between consecutive time series data, resulting in improved prediction accuracy. While training each individual IMF adds complexity, FFT (fast fourier transform) is employed to identify the principal frequency of the signal. By strategically combining IMFs based on their range beforehand, significantly reduces the number of times the ML model needs training.
[062] For instance, if initially there are 15 IMFs, training the model 15 times may prove intricate. However, by reducing the IMFs to just 5 through combination, the model requires only 5 training sessions, streamlining and simplifying the entire process.
[063] Thus, the method of decomposition-based processing, combined with FFT-based IMF reduction, optimizes the prediction process, ensuring both accuracy and efficiency in forecasting future power outcomes.
[064] Additionally, the approach incorporates the advanced AM-LSTM (Attention Mechanism with Long Short-Term Memory) model, which further enhances the forecasting accuracy. The AM-LSTM model combines the power of attention mechanisms and LSTM networks, allowing it to focus on crucial features while disregarding irrelevant ones. This adaptability ensures that the model allocates resources effectively and accurately captures temporal dependencies in the data.
[065] By integrating the AM-LSTM model into decomposition-based processing, it synergistically harness the strengths of both methodologies. The attention mechanisms in the AM-LSTM model enable it to attend to significant patterns within the IMFs, enhancing the predictive capabilities of the entire system.
[066] Through the strategic fusion of decomposition, FFT-based IMF reduction, and the AM-LSTM model, remarkable success is achieved in precise and efficient power forecasting. The resulting synergy empowers system to make accurate predictions, surpassing traditional methods, and opening doors for further advancements in the realm of energy forecasting.
[067] The forecasting algorithm proposed in this work consists of three stages addressing a different challenge in each stage (refer to Figure 2).
[068] The intermittent and stochastic nature of solar irradiance time series make the forecasting complicated and difficult due to its dependence on several meteorological variables, environmental conditions, latitude, time, etc. Data decomposition techniques are very suitable to get subseries of the complicated time series data. The first stage of forecasting method implements a novel Noise Assisted - Multivariate Empirical Mode Decomposition (NA-MEMD) technique for breaking the complex multivariate data of solar irradiance and weather variables into several subseries termed Intrinsic Mode Functions (IMFs). NA-MEMD can break the complex time series more effectively by reducing the mode mixing issue of traditional MEMD. Mode mixing occurs due to signal intermittency and causes a change in the physical meaning of IMFs. Adding only two or three noise channels with the original multivariate signal offers a uniformly distributed reference scale that enables MEMD to restore its filter bank property and helps to reduce the mixing of features in decomposed subseries.
[069] The fundamental problem with the decomposition techniques is the large number of created IMFs, which makes the model more difficult. IMFs are simple sub-signals of complex time series. The initial IMFs include a signal with a high frequency, which lowers as the number of IMFs increases. So, to reduce the complexity of decomposition-based models, the invention provides an approach based on the Fast Fourier transform (FFT).
[070] According to this approach, the complete large set of IMFs is divided into three broad categories, namely, high, medium, and low based on the range of frequency. So, in simple terms, a large pool of IMFs is converted into only three sets of IMFs. This technique reduces the complexity of the model with the need of predicting only three sets of IMFs, as opposed to all IMFs individually.
[071] Advancement in Machine Learning (ML) techniques has given sequence processing models, such as Long Short-term Memory (LSTM) networks. LSTM is capable to recall long-term dependencies in the data sequence and perform better non-linear mapping of input variable to output variable compared to the other ML models. The aim of this research is to predict the PV power for a complete day with a temporal resolution of 15 minutes. The third stage of the algorithm implements an attention-based LSTM model for predicting the IMFs obtained after the operation of NA-MEMD and FFT. The addition of the attention layer to conventional LSTM increased the capability of the model to process long-term dependencies among the data points, hence enhancing the prediction output's accuracy.
[072] Thus the invention reduces the complexities involved with the decomposition algorithm using fast Fourier transform in the field of solar power forecasting. It combines noise assisted - multivariate empirical mode decomposition (NA-MEMD), Fast fourier transform (FFT) and attention mechanism-based LSTM for application in solar PV forecasting. A seamless integration of decomposition based processing method, FFT based IMF reduction technique, and state of the art attention-based LSTM model which leverages the power of attention mechanism and LSTM networks is provided. This provides essential features and capturing of temporal dependencies in the dataset accurately.
[073] Numerous modifications and adaptations of the system of the present invention will be apparent to those skilled in the art, and thus it is intended by the appended claims to cover all such modifications and adaptations which fall within the true spirit and scope of this invention.
,CLAIMS:
WE CLAIM:
1. A forecasting system for solar photovoltaic power using machine-learning comprises-
a) six sensors characterized in that a pyranometer, passive cloud cover detector, rainfall sensor, air temperature and relative humidity sensor, and wind speed sensor wherein air temperature sensor has a range of - 40oC to 123.8oC with a resolution of 0.01oC, relative humidity sensor has a range of 0 to 100% with a resolution of 0.5% RH typical, wind speed sensor is a 3-cup anemometer type with a range of 0 to 67m/s and a resolution of 0.1 m/s, solar irradiation sensor is having a spectral range of 305 to 2800nm and a measuring range of 0 to 2000 W/m2 and passive cloud cover sensor is having an output of 0 to 1V for CCF 0 to 100%, respectively.
b) weather station characterized in that a solar panel and a battery that provides power to all the sensors. The whole structure is mounted on cemented blocks through steel pipes. The system comes with a waterproof rating and an operating temperature range of -40oC to 75oC and access panel is to get the data manually through a data shuttle which is collected in 5 minutes resolution and uploaded to cloud storage, such that it can be accessed remotely from anywhere.
c) input unit for collecting dedicated data for solar photovoltaic (PV) forecasting, and a forecasting process.
d) Control unit for storing and analyzing the data where the data collected from the input unit is decomposed using NA-MEMD technique.
2. The forecasting system, as claimed in claim 1, wherein the weather stations are installed at different locations.
3. The forecasting system, as claimed in claim 1, wherein data collection through ground-based monitoring stations on parameters such as solar data (GHI), temperature, wind speed, rainfall, humidity, and cloud cover.
4. The forecasting system, as claimed in claim 1, wherein decomposition process produces several subseries termed intrinsic mode functions (IMFs).
5. The forecasting system, as claimed in claim 1, wherein large numbers of IMFs are reduced into only three classes (High, medium, and low) using the fast Fourier Transform and the resultant functions are then passed through a modern attention-based LSTM neural network, which learns the patterns from three obtained classes and makes predictions of the day ahead GHI accurately.