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 REGIME BASED ANOMALY DETECTION IN VARIABLE LOAD OPERATED POWER PLANT

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 anomaly detection in power plant, and, more particularly, to a method and a system for regime based anomaly detection in variable operated power plant especially in coal power plant.

BACKGROUND
For the past few years, the industry has been experiencing challenges due to high competition across peers and shortage of skilled staff. Industrial assets are the machines which are responsible for performing certain processes which help in generating industrial output and contribute to revenue generation by meeting the consumer’s demands. The condition monitoring of the asset is a very critical aspect in the industry due to two reasons, firstly plant availability which directly impacts revenue generation and secondly the cost of asset maintenance i.e., spare parts cost and service cost. Maintenance cost increases significantly if the problem identification happens at a later stage. Unplanned failures reduce the reliability of the machine and return on investment.
Before the inclusion of renewable power generators in the grid, coal-based thermal power plants used to run primarily in full-load conditions and engineers were trained and skilled to operate the plant in that operating regime. With increased penetration of renewable power generation in the recent years to shift toward sustainable energy sources, there is a need to run coal-based power plant on variable load regime as a stable source of energy to overcome the uncertain or variation of the renewable sources. The coal based power plant business is facing problems due to emerging risk with rapid penetration of renewable energy and need right skills and tools to improve the availability of asset to meet the demand in the new scenario.
Existing coal-based power plants used to run more in steady state full load conditions for a majority of the time.. The dynamic changing condition in the grid due to adaptation of variable renewable energy is bringing new challenges in the industry which cannot be managed by existing technology. Existing technologies use different types of solution for conditional monitoring of tools. Some frameworks primarily depend on statistical analysis along with threshold, rule-based approach which has been built based on analytics and existing operation knowledge. Few frameworks are using fuzzy logic based asset health score calculation logic where lot of membership function need to be managed after statistical analysis of each sensor which works well for one regime e.g., full load running condition but will not work well for power plants which may operate on 40-60 capacity and in cyclic behavior. Causal analysis is performed primarily by rule-based techniques and correlation, but it does not consider variable load condition, trajectory, and different type of causality related to it. Often models are made to consider all different operational conditions, which makes the solution less scalable and eventually is like a black box and have not adopted explainable AI based on different regime. The existing frameworks also have limited capability to receive feedback and learn on continuous basis from external knowledge source, large language model with personalized language model or ChatGPT, Bard etc.

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 method for regime based anomaly detection in variable operated power plant is provided. The method includes receiving (i) a plurality of real-time operating conditions, (ii) a plurality of asset running conditions, and (iii) a plurality of environment conditions as input data associated with a power plant operating in real time. Further the method includes, extracting a set of features from the input data based on linear regression analysis of the plurality of real-time operating conditions. Furthermore, the method includes, classifying the set of features using a set of decision tree models to obtain a set of classified features and then identifying a set of real time regimes corresponding to the set of classified features. Further the method includes, identifying a set of relevant regimes using a pareto analysis technique performed on the set of regimes based on a domain knowledge and then detecting a set of anomalies. Finally, the method includes, performing a causality analysis for each relevant regime using the set of anomalies based on structure model learning on a pre-trained second Bayesian network model.
In another aspect, a system for regime based anomaly detection in variable operated power plant is provided. The system comprises memory storing instructions; one or more communication interfaces; and one or more hardware processors coupled to the memory via the one or more communication interfaces, wherein the one or more hardware processors are configured by the instructions to receive (i) a plurality of real-time operating conditions, (ii) a plurality of asset running conditions, and (iii) a plurality of environment conditions as input data associated with a power plant operating in real time. Further the system includes, extracting a set of features from the input data based on linear regression analysis of the plurality of real-time operating conditions. Furthermore, the system includes, classifying the set of features using a set of decision tree models to obtain a set of classified features and then identifying a set of real time regimes corresponding to the set of classified features. Further the system includes, identifying a set of relevant regimes using a pareto analysis technique performed on the set of regimes based on a domain knowledge and then detecting a set of anomalies. Finally, the system includes, performing a causality analysis for each relevant regime using the set of anomalies based on structure model learning on a pre-trained second Bayesian network model.
The power plant operates in a variable load condition and the set of features include (i) a set of trajectory parameters of a data distribution associated with the plurality of real-time operating conditions, (ii) a set of weather conditions from the plurality of environment conditions, and (iii) a set of seasonal demand features from the plurality of real-time operating conditions. The set of trajectory parameters comprises a slope and a coefficient of determination of the data distribution associated with the plurality of real-time operating conditions.
The training of the first Bayesian network model includes initially, receiving a historical input data corresponding to the power plant. Further it extracts a set of historical features from the historical input data and then obtains the set of decision tree models by a learning process. Further the training process combines the set of decision tree models to generate a hierarchical grid structure corresponding to the set of historical features and then performs regime segmentation on the hierarchical grid structure using a statistical information grid-based (STING) clustering technique to obtain a set of regimes. Finally, the first Bayesian network model is trained using the set of regimes and the set of historical features. It further generates the directed acyclic graph for the set of historical features from the first Bayesian network model.
The set of anomalies are detected by initially, identifying (i) a set of actual process variables, (ii) a set of manipulated variables, and (iii) a set of disturbance variables associated with the set of relevant regimes. Further a set of predicted process variables is predicted using a trained regression model associated with the set of relevant regimes. Then a residual error associated with each relevant regime is calculated based on a comparison between the set of predicted process variables and the set of actual process variables. Finally, an anomaly associated with each relevant regime is detected by comparing the residual error with a predefined regime threshold to obtain a set of anomalies.
In yet another aspect, there is provided a computer program product comprising a non-transitory computer readable medium having a computer readable program embodied therein, wherein the computer readable program, when executed on a computing device causes the computing device for regime based anomaly detection in variable operated power plant by receiving (i) a plurality of real-time operating conditions, (ii) a plurality of asset running conditions, and (iii) a plurality of environment conditions as input data associated with a power plant operating in real time. Further the computer readable program includes, extracting a set of features from the input data based on linear regression analysis of the plurality of real-time operating conditions. Furthermore, the computer readable program includes, classifying the set of features using a set of decision tree models to obtain a set of classified features and then identifying a set of real time regimes corresponding to the set of classified features. Further the computer readable program includes, identifying a set of relevant regimes using a pareto analysis technique performed on the set of regimes based on a domain knowledge and then detecting a set of anomalies. Finally, the computer readable program includes, performing a causality analysis for each relevant regime using the set of anomalies based on structure model learning on a pre-trained second Bayesian network model.
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 configured to perform regime based anomaly detection of assets in variable operated power plant according to some embodiments of the present disclosure.
FIG. 2 illustrates a broad level flow diagram for a method for regime based anomaly detection of assets in variable operated power plant according to some embodiments of the present disclosure.
FIG. 3 is an exemplary flow diagram depicting the steps of the method for regime based anomaly detection of assets in variable operated power plant according to some embodiments of the present disclosure.
FIG. 4 illustrates an exemplary flow diagram depicting the steps for training a first Bayesian network model for regime segmentation according to some embodiments of the present disclosure.
FIG. 5 illustrates an example frequency distribution of regimes of a sample asset in a coal based thermal power plant 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.
Existing power plants run in a steady state full load condition for majority of the time. Also, existing technology focuses to address such full load conditions. Adaptation of variable renewable energy and further dynamic changing condition in the grid brings new challenges to the industry which cannot be handled by the existing technology. The existing technologies use different types of solutions for conditional monitoring. Some frameworks primarily depend on statistical analysis along with threshold, rule-based approach, fuzzy logic based asset health score calculation logic and the like. These analysis uses a lot of membership function and need to be managed well. Statistical analysis of each sensor works well for one regime e.g., full load running condition. However, it will not work well for power plants which operate in a cyclic manner. In existing methods, outlier data is eliminated as part of pre-processing. However, such data preprocessing can also remove potential anomalies. However, such a step is legitimate for optimization.
Embodiments of the present disclosure provide an asset performance analysis (anomaly detection and causality analysis) of power plants which are operated in variable load conditions. The disclosed method uses feature engineering to extract parameters of input data and uses these extracted features for regime segmentation and regime detection of variable load operated power plants. The segmented regimes are used for anomaly detection in the power plant and further for causality analysis using the detected anomalies. The embodiments in the present disclosure are explained specifically for coal based power plants. However, the disclosed method may be utilized for regime based anomaly detection in other power plants such as solar, wind, thermal and the like.
Referring now to the drawings, and more particularly to FIG. 1 through FIG. 3, 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 configured to perform regime based anomaly detection of assets in variable operated power plant 102, according to some embodiments of the present disclosure. The system 100 provides a digital twin that can mimic the performance of the power plant 102 in real-time. Although the present disclosure is explained considering that the system 100 is implemented on a server, it may also be present elsewhere such as an on premise machine or a cloud. It may be understood that the system 100 comprises one or more computing devices 104, such as a laptop computer, a desktop computer, a notebook, a workstation, a cloud-based computing environment and the like. It will be understood that the system 100 may be accessed through one or more input/output interfaces collectively referred to as I/O interface 106. Examples of the I/O interface 106 may include, but are not limited to, a user interface, a portable computer, a personal digital assistant, a handheld device, a smartphone, a tablet computer, a workstation, and the like. The I/O interface 106 are communicatively coupled to the system 100 through a network 108.
In an embodiment, the network 108 may be a wireless or a wired network, or a combination thereof. In an example, the network 108 can be implemented as a computer network, as one of the different types of networks, such as virtual private network (VPN), intranet, local area network (LAN), wide area network (WAN), the internet, and such. The network 108 may either be a dedicated network or a shared network, which represents an association of the different types of networks that use a variety of protocols, for example, Hypertext Transfer Protocol (HTTP), Transmission Control Protocol/Internet Protocol (TCP/IP), and Wireless Application Protocol (WAP), to communicate with each other. Further, the network 108 may include a variety of network devices, including routers, bridges, servers, computing devices, storage devices. The network devices within the network 108 may interact with the system 100 through communication links.
The system 100 may be implemented in a workstation, a mainframe computer, a server, and a network server. In an embodiment, the computing device 104 further comprises one or more hardware processors 110, one or more memory 112, hereinafter referred as a memory 112 and a data repository 114, for example, a repository 114 or a database 114. The memory 112 is in communication with the one or more hardware processors 110, wherein the one or more hardware processors 110 are configured to execute programmed instructions stored in the memory 112, to perform various functions as explained in the later part of the disclosure. The repository 114 may store data processed, received, and generated by the system 100.
The system 100 supports various connectivity options such as BLUETOOTH®, USB, ZigBee and other cellular services. The network environment enables connection of various components of the system 100 using any communication link including Internet, WAN, MAN, and so on. In an exemplary embodiment, the system 100 is implemented to operate as a stand-alone device. In another embodiment, the system 100 may be implemented to work as a loosely coupled device to a smart computing environment. The components and functionalities of the system 100 are described further in detail.
FIG. 2 illustrates a broad level flow diagram for a method for regime based anomaly detection of assets in variable operated power plant according to some embodiments of the present disclosure. The disclosed method uses historical data for training decision tree models for extracting features initially. The extracted features are then used for regime segmentation using a Bayesian network model. The Bayesian network model is trained using the extracted features and a set of regimes obtained using a statistical information grid-based (STING) clustering technique. These trained models are further used in real time for feature extraction and regime segmentation of real time input data such as operating conditions data, asset running conditions data and environmental conditions data of the power plant. After obtaining regimes for the real time input data, anomaly detection is performed using on the collected real time input data to obtain a set of anomalies. The anomalies are detected by predicting target variables (process variables) of the power plant. These anomalies are used for causality analysis of the power plant.
FIG. 3 is an exemplary flow diagram depicting the steps of a method 300 for regime based anomaly detection of assets in variable operated power plant according to some embodiments of the present disclosure.
In an embodiment, the system 100 comprises of one or more data storage devices or the memory 112 operatively coupled to the one or more hardware processor(s) 110 and is configured to store instructions for execution of steps of the method 200 by the processor(s) or one or more hardware processors 110. The steps of the method 200 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 the steps of flow diagram as depicted in FIG.3. 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.
At step 302 of the method 300, the one or more hardware processors 102 are configured to receive (i) a plurality of real-time operating conditions, for example a power plant running in either full load, partial load or minimal load, (ii) a plurality of asset running conditions, for example, asset running in maximum or minimum permissible capacity with respect to asset design parameter suggested by original equipment maker, and (iii) a plurality of environment conditions such as input data associated with a power plant operating in real time. A few examples for the environment conditions associated with the power plant include ramp up-ramp down of load, day and night load consumption, seasonality load consumption such as rainy, summer, winter load consumption and so on. The power plant operates in a variable load condition.
At step 304 of the method 300, the one or more hardware processors 102 are configured to extract a set of features from the input data based on linear regression analysis of the plurality of real-time operating conditions. The set of features include (i) a set of trajectory parameters of a data distribution associated with the plurality of real-time operating conditions, (ii) a set of weather conditions from the plurality of environment conditions, and (iii) a set of seasonal demand features (for example, fuel type, demographic conditions and so on) from the plurality of real-time operating conditions. The set of trajectory parameters comprises a slope and a coefficient of determination of the data distribution associated with the plurality of real-time operating conditions. Feature extraction has been done taking into consideration the real-time operating conditions of the power plant i.e., load (MegaWatt), and asset running condition i.e., current (ampere). A linear regression model is used to extract the property like gradient (the slope or coefficient term) and r-square (the coefficient of determination) of the trajectory of the data distribution associated with the real-time operating conditions. Those properties are used along with start_load (loadt-n), instant_current (It). These are represented in term of function e.g., asset_regime = func (trajectory, current_state). Table 1 provides an example of extracted features and its values using the linear regression model.
S.No Start_load Gradient R-square Instant_current
0 371.848517 0.355 1.000 140.934
1 372.203373 0.355 1.000 141.380
2 372.558230 0.347 0.999 141.489
3 372.913086 0.334 0.996 141.598
4 373.267943 0.317 0.991 141.707
Table 1
At step 306 of the method 300, the one or more hardware processors 102 are configured to classify the set of features using a set of decision tree models to obtain a set of classified features. The set of features are classified using a learnt set of decision tree models. These learnt set of decision tree models are used further for regime segmentation for anomaly detection. The learning of the set of decision tree models is explained in conjunction with step 308.
At step 308 of the method 300, the one or more hardware processors 102 are configured to identify a set of real time regimes corresponding to the set of classified features using a directed acyclic graph generated from a pre-trained first Bayesian network model. The first Bayesian network model is trained using the learnt set of decision tree models. FIG. 4 illustrates an exemplary flow diagram depicting the steps for training the first Bayesian network model for regime segmentation according to some embodiments of the present disclosure. At step 402, the one or more hardware processors 102 are configured to receive (i) a plurality of historical operating conditions, (ii) a plurality of historical asset running conditions, and (iii) a plurality of historical environment conditions as historical input data corresponding to the power plant. Further, a set of historical features are extracted from the historical input data at step 404. At step 406, the set of decision tree models are obtained by a learning process. For learning the set of decision tree models, one decision tree model is learnt using at least one feature from the set of historical features. Each decision tree model is learnt based on a comparison of the at least one feature with a set of predefined binning thresholds. The predefined binning thresholds are obtained based on the domain knowledge. As an example, for a current dataset, the four features being considered are load, ID fan current, gradient and r-square. For learning the decision tree model for the feature ‘ID fan current’, it has been divided into multiple bins or classes based on domain denoted by “low current: LC”,” normal current: NC”, “high current: HC”. This binning is entirely based on domain understanding specific to operation and may change with power plants. As an instance, the below conditions which is being compared with the binning thresholds are considered for learning the decision tree model for the feature ID fan current denoted as feature_0.
feature_0 <=157.58
feature_0 <=98.50
class: low current
feature_0 >98.50
class: normal current
feature_0 > 157.58
class: high current
The feature ‘load’ will be divided into multiple bins or classes based on the domain knowledge, and denoted by “no load: NL”, “low load: LL”, “partial load: PL”, “half load: HL”, “partial full load: PFL”, “full load: FL”. As an instance, the below conditions are considered for learning the decision tree model for the feature load denoted as feature_1.
feature_1 <=352.00
feature_1 <=263.39
feature_1<=88.40
feature_1<=3.63
class: no load
feature_1 >3.63
class: low load
feature_0 > 88.40
feature_1<=175.16
class: partial load
feature_1>175.16
class: half load
feature_1> 263.39
class: partial full load
feature_1 > 352.00
class: full load
For the features ‘gradient’ and ‘r-square’ the below conditions are considered for learning the decision tree model.
r-square <=0.92
r-square <=0.00
gradient<=0.00
class: -ve steady state
gradient>0.00
class: +ve steady state
r-square >0.00
class: on transit
r-square >0.92
gradient<=0.00
gradient<=-1.00
gradient<=-2.00
class: high ramp down
gradient>-2.00
class: ramp down
gradient>-1.00
class: -ve steady state
gradient>0.00
gradient<=1.00
class: +ve steady state
gradient>1.00
gradient<=2.00
class: ramp up
gradient>2.00
class: high ramp up
At step 408, the set of decision tree models are combined to generate a hierarchical grid structure corresponding to the set of historical features. The grid structure contains several grid layers which has been defined using each decision tree model created using the step 406. For example, considering the current dataset with the extracted features mentioned above, the features ID fan current and load shall be part of higher layers of the hierarchical grid structure. And the lower layers correspond to the gradient and r-square features.
At step 410, regime segmentation is performed on the hierarchical grid structure using a statistical information grid-based (STING) clustering technique to obtain a set of regimes. In STING clustering, the data distribution or probability distribution, is transformed into grid-based clustering method considering a space-driven approach by partitioning the embedding space into cells or bins independent of the distribution of the features. STING is a grid-based multiresolution clustering technique in which the features are divided into rectangular cells. Now, space can be divided into hierarchical and recursive way, based on the features. In the present disclosure, another dimension is added to the hierarchical grid structure as a timestep and trajectory to it. Hence a concept of Temporal STING is provided, where the position of cells in the grid may change and will allow to change the hierarchical grid with time and can adopt to changes. By this STING clustering on the set of historical features, a set of regimes are identified.
Further at step 412, the first Bayesian network model is trained using the set of regimes and the set of historical features. A directed acyclic graph (DAG) generated from a pre-trained first Bayesian network model. Bayesian Network Hill Climb Search techniques with BIC score estimator has been used to extract the relationship graph (the directed acyclic graph) for all the features considered for the regime segmentation. Bayesian Network classifier works well even in imbalanced dataset with discreate value. Here in the present disclosure, it has been tried to adopt the explainable AI with feature transformation step of binning, decision Tree followed by clustering with STING segmentation and then finally applying Bayesian Network Hill Climb search technique to demonstrate DAG and adopt explainable AI.
Once the first Bayesian network model is trained at step 308 (in accordance to steps 402 through 412), at step 310 of the method 300, the one or more hardware processors 102 are configured to identify the set of relevant regimes using a pareto analysis technique performed on the set of regimes based on domain knowledge. FIG. 5 illustrates an example of frequency distribution of regimes of a sample asset in a coal based thermal power plant in accordance with some embodiments of the present disclosure. As an instance, considering the ‘current’ dataset mentioned above for feature classification, hierarchical structure formed under layer of trajectory ? {on transit} i.e., r-square <0.92 and >0 based on the decision tree, has been ignored. After the first level rejection of those regimes, based on pareto analysis technique, minority regimes are rejected for model building. And hence the set of relevant regimes are obtained. Considering the FIG. 5, for experimental purpose, a threshold has been considered which is close to the median of the frequency distribution and after removing the on-transit regimes, the datapoints that will be covered by any machine learning based anomaly detection is approximately 59%. Alternatively, 80-20 principle of the pareto analysis technique also is adopted, and on top of that domain knowledge can be applied to ignore the regime with changing state.
At step 312 of the method 300, the one or more hardware processors 102 are configured to detect a set of anomalies associated with the set of relevant regimes. Detecting the set of anomalies includes initially identifying a set of actual process variables (PV), (ii) a set of manipulated variables (MV), and (iii) a set of disturbance variables (DV) associated with the set of relevant regimes. Based on domain knowledge and operation experience, PV, MV and DV are selected for anomaly detection. Anomalies can be of two type- controllable and uncontrollable, so variables have been grouped based on it. PV are variables that are monitored as key process indicator, it depends on other features which may be MV or DV. Deviation or residual error in process variable (target variable) are anomaly. MV are variables which are controllable and operation engineer can control it to recover or avoid anomaly. DV are variables which are uncontrollable.
Further using a trained regression model a set of predicted process variables associated with the set of relevant regimes are predicted. The below equations are used for predicting PVs for a specific regime i.
For regime i, predicted value:
y ^_(?i,pv?_1 )=b_(0,?pv?_1 )+b_(1,?pv?_1 ) x_(?pv?_1,1)+b_(2,?pv?_1 ) x_(?pv?_1,2)+?+b_(k,?pv?_1 ) x_(?pv?_1,k) ?1
y ^_(?i,pv?_2 )=b_(0,?pv?_2 )+b_(1,?pv?_2 ) x_(?pv?_2,1)+b_(2,?pv?_2 ) x_(?pv?_2,2)+?+b_(k,?pv?_2 ) x_(?pv?_2,k) ?2
…….
y ^_(?i,pv?_n )=b_(0,?pv?_n )+b_(1,?pv?_n ) x_(?pv?_n,1)+b_(2,?pv?_n ) x_(?pv?_n,2)+?+b_(k,?pv?_n ) x_(?pv?_n,k) ?3
Here, b is used to represent a sample estimate of a ß coefficient. Thus, b_0 is the sample estimate of ß_0, b_1 is the sample estimate of ß_1, and so on. y ^_(?i,pv?_n ) is predictor variable for each p_v of ith regime.
Further after predicting PVs, a residual error associated with each relevant regime is calculated based on a comparison between the set of predicted process variables and the set of actual process variables. For regime i, residual error ? for each pv:
?_(i,?pv?_1 )= ?(y?_(?i,pv?_1 )- y ^_(?i,pv?_1 )) ?4
?_(i,?pv?_2 )= ?(y?_(?i,pv?_2 )- y ^_(?i,pv?_2 )) ?5
….
?_(i,?pv?_n )= ?(y?_(?i,pv?_n )- y ^_(?i,pv?_n )) ?6
Further an anomaly associated with each relevant regime is detected by comparing the residual error with a predefined regime threshold to obtain a set of anomalies. Based on the standard deviation of each target variable (pv) within specific regime, threshold value to consider as anomaly may differ.
anomaly against a pv for a specific regime={¦(p,&?_(i,?pv?_n )t_2 )¦ ?7
t_1 and t_2 are the threshold value of error to consider as p and q type of anomaly respectively and can be treated as lower threshold limit or high threshold limit.
At step 314 of the method 300, the one or more hardware processors 102 are configured to perform a causality analysis for each relevant regime using the set of anomalies based on structure model learning on a pre-trained second Bayesian network model. Structure Model learning holds directed edges, describing a cause of anomaly due to effect relationship with different magnitude of dependent parameter, resulting to weighted edge of graph adjacency matrix. The second Bayesian network model has been created on top of the structured learning and then the second Bayesian network model has been fit with the data and Maximum Likelihood Estimator. Inference models are built with Variable Elimination algorithms, on which different queries related to causal inference can be fired to extracted probabilistic inference. The second Bayesian network model is pre-trained on historical anomaly data which represents probability of the causality of various operational state.
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.
Embodiments of present disclosure herein addresses unresolved problem of asset performance analysis of variable load operated power plant. The embodiment thus provides a method for regime segmentation based anomaly detection in power plant, especially in coal based power plant. The disclosed method uses trained decision tree models for classifying features extracted from operating conditions, environmental conditions and asset running conditions of the power plant. The regimes associated with these features are identified using a trained Bayesian network model. These regimes are further utilized for anomaly detection and causality analysis in the power plant.
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.
, C , Claims:We Claim:
1. A processor implemented method (300) comprising:
receiving, via one or more hardware processors, (i) a plurality of real-time operating conditions, (ii) a plurality of asset running conditions, and (iii) a plurality of environment conditions as input data associated with a power plant operating in real time (302);
extracting, via the one or more hardware processors, a set of features from the input data based on linear regression analysis of the plurality of real-time operating conditions (304);
classifying, via the one or more hardware processors, the set of features using a set of decision tree models to obtain a set of classified features (306);
identifying, via the one or more hardware processors, a set of real time regimes corresponding to the set of classified features using a directed acyclic graph generated from a pre-trained first Bayesian network model (308);
identifying, via the one or more hardware processors, a set of relevant regimes using a pareto analysis technique performed on the set of regimes based on a domain knowledge (310);
detecting, via the one or more hardware processors, a set of anomalies associated with the set of relevant regimes (312); and
performing, via the one or more hardware processors, a causality analysis for each relevant regime using the set of anomalies based on structure model learning on a pre-trained second Bayesian network model (314).

2. The method as claimed in claim 1, wherein the power plant operates in a variable load condition.

3. The method as claimed in claim 1, wherein the set of features include (i) a set of trajectory parameters of a data distribution associated with the plurality of real-time operating conditions, (ii) a set of weather conditions from the plurality of environment conditions, and (iii) a set of seasonal demand features from the plurality of real-time operating conditions.

4. The method as claimed in claim 3, wherein the set of trajectory parameters comprises a slope and a coefficient of determination of the data distribution associated with the plurality of real-time operating conditions.

5. The method as claimed in claim 1, wherein training the first Bayesian network model (400) comprises:
receiving, via the one or more hardware processors, (i) a plurality of historical operating conditions, (ii) a plurality of historical asset running conditions, and (iii) a plurality of historical environment conditions as historical input data corresponding to the power plant (402);
extracting, via the one or more hardware processors, a set of historical features from the historical input data (404);
obtaining, via the one or more hardware processors, the set of decision tree models by learning at least one decision tree model among the set of decision tree models using at least one feature among the set of historical features, wherein the at least one decision tree model is learnt based on a comparison of the at least one feature with a set of predefined binning thresholds (406);
combining, via the one or more hardware processors, the set of decision tree models to generate a hierarchical grid structure corresponding to the set of historical features (408);
performing, via the one or more hardware processors, regime segmentation on the hierarchical grid structure using a statistical information grid-based (STING) clustering technique to obtain a set of regimes (410); and
training the first Bayesian network model using the set of regimes and the set of historical features (412).
6. The method as claimed in claim 5, comprises:
generating, via the one or more hardware processors, the directed acyclic graph for the set of historical features from the first Bayesian network model.

7. The method as claimed in claim 1, wherein detecting the set of anomalies comprises:
identifying, via the one or more hardware processors, (i) a set of actual process variables, (ii) a set of manipulated variables, and (iii) a set of disturbance variables associated with the set of relevant regimes;
predicting, via the one or more hardware processors, a set of predicted process variables using a trained regression model associated with the set of relevant regimes;
calculating, via the one or more hardware processors, a residual error associated with each relevant regime based on a comparison between the set of predicted process variables and the set of actual process variables; and
detecting, via the one or more hardware processors, an anomaly associated with each relevant regime by comparing the residual error with a predefined regime threshold to obtain a set of anomalies.

8. A system (100), comprising:
a memory (112) storing instructions;
one or more communication interfaces (106); and
one or more hardware processors (110) coupled to the memory (112) via the one or more communication interfaces (106), wherein the one or more hardware processors (110) are configured by the instructions to:
receive (i) a plurality of real-time operating conditions, (ii) a plurality of asset running conditions, and (iii) a plurality of environment conditions as input data associated with a power plant operating in real time;
extract a set of features from the input data based on linear regression analysis of the plurality of real-time operating conditions;
classify the set of features using a set of decision tree models to obtain a set of classified features;
identify a set of real time regimes corresponding to the set of classified features using a directed acyclic graph generated from a pre-trained first Bayesian network model;
identify a set of relevant regimes using a pareto analysis technique performed on the set of regimes based on a domain knowledge;
detect a set of anomalies associated with the set of relevant regimes; and
perform a causality analysis for each relevant regime using the set of anomalies based on structure model learning on a pre-trained second Bayesian network model.

9. The system as claimed in claim 8, wherein the power plant operates in a variable load condition.

10. The system as claimed in claim 8, wherein the set of features include (i) a set of trajectory parameters of a data distribution associated with the plurality of real-time operating conditions, (ii) a set of weather conditions from the plurality of environment conditions, and (iii) a set of seasonal demand features from the plurality of real-time operating conditions.

11. The system as claimed in claim 10, wherein the set of trajectory parameters comprises a slope and a coefficient of determination of the data distribution associated with the plurality of real-time operating conditions.

12. The system as claimed in claim 8, wherein training the first Bayesian network model comprises:
receive (i) a plurality of historical operating conditions, (ii) a plurality of historical asset running conditions, and (iii) a plurality of historical environment conditions as historical input data corresponding to the power plant;
extract a set of historical features from the historical input data;
obtain the set of decision tree models by learning at least one decision tree model among the set of decision tree models using at least one feature among the set of historical features, wherein the at least one decision tree model is learnt based on a comparison of the at least one feature with a set of predefined binning thresholds;
combine the set of decision tree models to generate a hierarchical grid structure corresponding to the set of historical features;
perform regime segmentation on the hierarchical grid structure using a statistical information grid-based (STING) clustering technique to obtain a set of regimes; and
train the first Bayesian network model using the set of regimes and the set of historical features.

13. The system as claimed in claim 12, comprises:
generating, via the one or more hardware processors, the directed acyclic graph for the set of historical features from the first Bayesian network model.

14. The system as claimed in claim 8, wherein detecting the set of anomalies comprising,
identify (i) a set of actual process variables, (ii) a set of manipulated variables, and (iii) a set of disturbance variables associated with the set of relevant regimes;
predict a set of predicted process variables using a trained regression model associated with the set of relevant regimes;
calculate a residual error associated with each relevant regime based on a comparison between the set of predicted process variables and the set of actual process variables; and
detect an anomaly associated with each relevant regime by comparing the residual error with a predefined regime threshold to obtain a set of anomalies.

Dated this 22nd Day of September 2023

Tata Consultancy Services Limited
By their Agent & Attorney

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