Agent-driven big data analysis and thermal power system operation control method and system

Abstract

The application relates to the technical field of power generation system control, and discloses an agent-driven big data analysis and thermal power generation system operation control method and system, which comprises the following steps: obtaining operation data of a thermal power generation system; inputting the operation data into a power system agent to analyze the operation state of the thermal power generation system and calculate optimal control parameters for adjusting the thermal power generation system to an optimal operation state; and controlling the operation of the thermal power generation system according to the optimal control parameters to adjust the operation state of the thermal power generation system in real time. Based on a large amount of operation data of the thermal power generation system, the power system agent can quickly extract high-quality samples from the massive data, realize in-depth analysis of the operation state of the thermal power generation system and accurate calculation of the optimal control parameters, improve the control efficiency and adjustment accuracy of the thermal power generation system, ensure that the system is always adjusted towards the optimal operation state, and improve the stability and reliability of system operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power generation system control, in particular to an Agent-driven big data analysis and thermal power generation system operation control method and system. BACKGROUND

[0002] The current thermal power generation system operation control method uses a simple mathematical model or fixed rules to analyze the thermal power generation system operation data and calculate the control parameters, which cannot fully consider the complexity and dynamic changes of the system, and it is difficult to make accurate and timely decisions in the face of complex working conditions and sudden situations, resulting in low system operation efficiency and even system failure; when facing massive data, it is difficult to accurately and effectively extract valuable information, resulting in low data processing efficiency and inability to quickly respond to actual operating conditions.

[0003] A thermal power generation unit control system and method are disclosed in a Chinese patent with the authorization announcement number CN109782712B, which includes a control cabinet, a database and a computing platform; the database is used to store the operation information of the unit in the DCS; the computing platform is used to find the operation information in the database, determine the optimal target information of the unit, and send the target information to the control cabinet; the control cabinet is used to collect the operation information of the unit in the DCS; and the control loop is adjusted according to the target information to control the operation of the unit.

[0004] The above patent has the problems raised in the background art: the simple mathematical model or fixed rules are used to analyze the thermal power generation system operation data and calculate the control parameters, which cannot fully consider the complexity and dynamic changes of the system, and it is difficult to make accurate and timely decisions in the face of complex working conditions and sudden situations, resulting in low system operation efficiency and even system failure; when facing massive data, it is difficult to accurately and effectively extract valuable information, resulting in low data processing efficiency and inability to quickly respond to actual operating conditions; to solve the above problems, the present application proposes an Agent-driven big data analysis and thermal power generation system operation control method and system. SUMMARY

[0005] In view of the deficiencies of the prior art, the main purpose of the present application is to provide an Agent-driven big data analysis and thermal power generation system operation control method and system, which can effectively solve the problems in the background art. The specific technical scheme of the present application is as follows:

[0006] The Agent-driven big data analysis and thermal power generation system operation control method comprises:

[0007] obtaining the operation data of the thermal power generation system;

[0008] The operation data is input into the power system agent for feature extraction and projection dimension reduction, and the operation state of the thermal power generation system is analyzed to calculate the optimal control parameter for adjusting the thermal power generation system to the optimal operation state.

[0009] According to the optimal control parameter, the operation of the thermal power generation system is controlled to adjust the operation state of the thermal power generation system in real time.

[0010] Specifically, the operation data is input into the power system agent for feature extraction and projection dimension reduction, and the operation state of the thermal power generation system is analyzed to calculate the optimal control parameter for adjusting the thermal power generation system to the optimal operation state, comprising:

[0011] The historical operation data is analyzed to calculate the features related to the operation state of the thermal power generation system, and a feature vector is constructed;

[0012] The feature vector is projected and mapped to a low-dimensional space for dimension reduction of the feature vector to obtain a low-dimensional feature vector;

[0013] According to the low-dimensional feature vector, the data samples are screened to obtain high-quality samples;

[0014] According to the high-quality samples, the relationship between the operation state of the thermal power generation system and the control parameter is analyzed, and a regression model is constructed;

[0015] According to the current operation state of the thermal power generation system, the optimal control parameter for adjusting the thermal power generation system to the optimal operation state is calculated through the regression model.

[0016] Specifically, the historical operation data is analyzed to calculate the features related to the operation state of the thermal power generation system, and a feature vector is constructed, comprising:

[0017] According to the historical operation data, the features related to the operation state of the thermal power generation system are calculated through thermodynamic formula to obtain a plurality of thermodynamic features;

[0018] The feature weight value of each thermodynamic feature is calculated through a pre-set feature weighting model;

[0019] According to the feature weight value, the corresponding thermodynamic feature is weighted to obtain a plurality of weighted features;

[0020] The plurality of weighted features are combined to obtain a feature vector.

[0021] Specifically, the feature vector is projected and mapped to a low-dimensional space for dimension reduction of the feature vector to obtain a low-dimensional feature vector, comprising:

[0022] The dependency relationship between each feature in the feature vector and the preset target variable is analyzed, and the mutual information value is calculated.

[0023] Based on the mutual information value, retain the corresponding features whose mutual information value is greater than a preset mutual information threshold to obtain the first feature vector;

[0024] The first feature vector is projected onto a low-dimensional space to obtain a low-dimensional feature vector.

[0025] Specifically, the step of projecting and mapping the first feature vector onto a low-dimensional space to obtain a low-dimensional feature vector includes:

[0026] Calculate the distance between each feature in the first feature vector to obtain the distance similarity.

[0027] Features with a distance similarity greater than a preset similarity threshold are connected to form a feature network, wherein each network point in the feature network is a feature, and network points are connected according to distance similarity.

[0028] Construct a projection plane based on the orthogonal vectors of the first feature vector;

[0029] Each network edge in the feature network is projected onto the projection plane to obtain multiple planar projection lines;

[0030] Based on the plane projection lines, calculate the intersection point or centroid of every two plane projection lines. When two plane projection lines intersect, calculate their intersection point; when two plane projection lines do not intersect, calculate their centroid.

[0031] Continue until the calculation of the intersection point or centroid between every two plane projection lines is completed;

[0032] The calculated intersection points and centroids are concatenated to obtain a low-dimensional feature vector.

[0033] Specifically, the step of filtering data samples based on the low-dimensional feature vector to obtain high-quality samples includes:

[0034] Based on the similarity between low-dimensional feature vectors, the data samples are clustered to obtain the first high-quality samples;

[0035] The first high-quality sample is expanded using a preset sample expansion model to obtain a second high-quality sample.

[0036] The first high-quality sample and the second high-quality sample are combined to obtain a high-quality sample.

[0037] Specifically, the step of clustering data samples based on the similarity between low-dimensional feature vectors to obtain the first high-quality sample includes:

[0038] According to the preset performance index, a performance threshold range of high-quality samples is set;

[0039] According to the low-dimensional feature vectors, sample data satisfying the performance threshold range is screened out to obtain initial high-quality samples;

[0040] According to the similarity between the low-dimensional feature vectors, the data samples are clustered to obtain first-class samples;

[0041] The distance between the cluster center of each class of samples and the initial high-quality samples is calculated to obtain a first distance;

[0042] The class corresponding to the cluster center with the first distance less than a preset first distance threshold is retained to obtain second-class samples;

[0043] In the second-class samples, the distance between the samples in each class of samples and the cluster center is calculated to obtain a second distance;

[0044] The samples with the second distance less than a preset second distance threshold are retained to obtain similar high-quality samples;

[0045] The initial high-quality samples, the cluster centers of the second-class samples, and the similar high-quality samples are combined to obtain first high-quality samples.

[0046] Specifically, the first high-quality samples are expanded by a preset sample expansion model to obtain second high-quality samples, including:

[0047] According to the first high-quality samples, a generated sample is obtained by a preset sample expansion model;

[0048] The generated sample satisfying the performance threshold range of high-quality samples is retained by calculating the performance index of the generated sample to obtain second high-quality samples.

[0049] Specifically, the relationship between the operation state of the thermal power generation system and the control parameters is analyzed according to the high-quality samples, and a regression model is constructed, including:

[0050] According to the high-quality samples, the relationship between the operation state of the thermal power generation system and the control parameters is preliminarily analyzed to obtain a first regression model;

[0051] The prediction result obtained by the first regression model is used as a new sample;

[0052] The high-quality samples and the new sample are combined to fit the relationship between the operation state of the thermal power generation system and the control parameters to obtain a regression model.

[0053] The application discloses an Agent-driven big data analysis and thermal power system operation control system.

[0054] A data acquisition module is arranged to acquire operation data of the thermal power system.

[0055] An optimal control parameter calculation module is arranged to input the operation data into the power system agent to perform feature extraction and dimension reduction, and analyze the operation state of the thermal power system, so as to calculate optimal control parameters for adjusting the thermal power system to an optimal operation state.

[0056] A thermal power system control module is arranged to control the operation of the thermal power system according to the optimal control parameters, so as to perform real-time adjustment on the operation state of the thermal power system.

[0057] Compared with the prior art, the application has the following beneficial effects:

[0058] Based on a large amount of operation data of the thermal power system, the application can quickly extract high-quality samples from massive data through the power system agent, realize in-depth analysis on the operation state of the thermal power system and accurate calculation of optimal control parameters, improve the control efficiency and adjustment accuracy of the thermal power system, continuously adapt to dynamic changes of the system, continuously optimize the control strategy, ensure that the system is always adjusted to an optimal operation state, and improve the stability and reliability of system operation. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 A work flow chart of the Agent-driven big data analysis and thermal power system operation control method in the embodiment 1 of the application is shown in the figure.

[0060] Figure 2 A work flow chart of the optimal control parameter calculation process in the embodiment 1 of the application is shown in the figure.

[0061] Figure 3 A schematic diagram of the feature network construction in the embodiment 1 of the application is shown in the figure.

[0062] Figure 4 A schematic diagram of the feature network projection in the embodiment 1 of the application is shown in the figure.

[0063] Figure 5 A schematic diagram of the first high-quality sample screening process in the embodiment 1 of the application is shown in the figure.

[0064] Figure 6 A structural schematic diagram of the Agent-driven big data analysis and thermal power system operation control system in the embodiment 2 of the application is shown in the figure. DETAILED DESCRIPTION

[0065] In order to make the above objectives, characteristics and advantages of the present application more apparent, a detailed description of the specific embodiments of the present application will be given below with reference to the accompanying drawings.

[0066] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application, however, can be practiced in a variety of ways beyond the specific embodiments described herein without departing from the scope of the present application, which is set forth in the appended claims. It should be recognized, therefore, that the specific embodiments disclosed are illustrative only and not limiting of the present application.

[0067] Secondly, the term "one embodiment" or "an embodiment" as used herein means that a particular implementation can include a particular feature, structure, or characteristic, but every embodiment can not necessarily include the particular feature, structure, or characteristic. Furthermore, the following claims can refer to "one embodiment" or "an embodiment" in the sense of claiming a particular feature, structure, or characteristic of more than one embodiment.

[0068] Embodiment 1

[0069] The embodiment provides an agent-driven big data analysis and thermal power system operation control method, which comprises the following steps of: Figure 1 The agent-driven big data analysis and thermal power system operation control method comprises the following steps of:

[0070] S101, acquiring operation data of a thermal power system;

[0071] S102, inputting the operation data into a power system agent for feature extraction and dimension reduction projection, and analyzing the operation state of the thermal power system to calculate optimal control parameters for adjusting the thermal power system to an optimal operation state;

[0072] S103, controlling the operation of the thermal power system according to the optimal control parameters to adjust the operation state of the thermal power system in real time.

[0073] At present, the operation process control method of the thermal power system cannot dynamically adjust according to complex working conditions and sudden situations, it is difficult to make accurate and timely decisions, it is difficult to accurately and effectively extract valuable information from massive data, and the data processing efficiency is low. The present application can automatically adjust the control strategy and parameters according to the change of the system operation state, adapt to different operation conditions and environmental changes, and has stronger adaptability and flexibility compared with the traditional control method.

[0074] In this embodiment, first, data related to the operation state of the thermal power generation system during operation of the thermal power generation system is acquired. Various sensors are installed at key parts of the thermal power generation system, such as the boiler, steam turbine, generator, and pipeline. For example, a temperature sensor is installed in the boiler furnace to monitor the combustion temperature, and a pressure sensor and a flow sensor are installed in the steam pipeline to measure the pressure and flow of the steam. The operation data of the thermal power generation system is continuously collected at a preset time interval, and rich operation data is acquired to comprehensively reflect the operation of each part of the thermal power generation system.

[0075] Specifically, after receiving the operation data, the power system agent processes and analyzes the input operation data based on big data analysis technology and machine learning algorithms. First, through data cleaning, feature extraction, and other steps, the original system operation data is converted into a data format suitable for model analysis. Then, the operation state of the thermal power generation system is evaluated and classified according to the extracted features, and the current system state is determined. Next, according to the relationship model between the system operation state and the control parameters, the control parameters that can make the system achieve the best operation state are calculated, wherein the relationship model is trained based on high-quality sample data in a large amount of historical data and can accurately reflect the relationship between the system operation state and the control parameters. Through the driving of the power system agent, the best control parameters can be calculated based on the real-time operation state to realize intelligent operation control. Through dimensionality reduction and high-quality sample selection of massive data, useful information can be quickly extracted from a large amount of data, improving the data processing speed. High-quality samples can improve the accuracy and generalization ability of the model, thereby more comprehensively and accurately analyzing the operation state of the thermal power generation system and calculating the control parameters.

[0076] Specifically, according to the best control parameters calculated by the power system agent, the actuator of the thermal power generation system is adjusted through the control system. The actuator adjusts the operation parameters of the equipment according to the received control signal, such as adjusting the fuel valve opening to change the fuel supply amount, adjusting the steam turbine inlet valve opening to change the steam flow, and other adjustments, thereby realizing real-time adjustment of the operation state of the thermal power generation system and making the system approach the best operation state. According to the best control parameters calculated in real time, the thermal power generation system can be quickly and accurately adjusted, improving the response speed and adjustment accuracy of the system and ensuring that the system is always in the best operation state.

[0077] The application is based on a large amount of thermal power generation system operation data, and a high-quality sample is quickly extracted from massive data by an electric power system intelligent agent, the operation state of the thermal power generation system is deeply analyzed, and the optimal control parameters are accurately calculated, the control efficiency and adjustment accuracy of the thermal power generation system are improved, the system can continuously adapt to the dynamic changes of the system, continuously optimize the control strategy, ensure that the system always adjusts to the best operation state, and improve the stability and reliability of the system operation.

[0078] Further, as Figure 2 The operation data is input into the electric power system intelligent agent for feature extraction and dimensionality reduction, and the operation state of the thermal power generation system is analyzed, and the optimal control parameters for adjusting the thermal power generation system to the optimal operation state are calculated, including:

[0079] S201, analyzing the historical operation data, calculating the features related to the operation state of the thermal power generation system, and constructing a feature vector;

[0080] S202, projecting and mapping the feature vector to a low-dimensional space to reduce the dimensionality of the feature vector, and obtaining a low-dimensional feature vector;

[0081] S203, filtering the data samples according to the low-dimensional feature vector to obtain high-quality samples;

[0082] S204, analyzing the relationship between the operation state of the thermal power generation system and the control parameters according to the high-quality samples, and constructing a regression model;

[0083] S205, calculating the optimal control parameters for adjusting the thermal power generation system to the optimal operation state according to the current operation state of the thermal power generation system and the regression model.

[0084] According to the historical operation data, the features related to the operation state of the thermal power generation system are extracted and reduced in dimension, and the high-quality sample data is filtered from a large amount of operation data, and the relationship model between the operation state of the thermal power generation system and the control parameters is fitted based on the high-quality sample data, which can quickly respond to the changing operation state of the thermal power generation system, and adjust the control strategy in real time. Compared with the traditional fixed mode control method, it has stronger response speed and flexibility, and can better meet the complex needs in actual production.

[0085] In this embodiment, the electric power system intelligent agent extracts features from the historical operation data of the thermal power generation system, obtains a large number of features that can effectively represent the operation state of the system, combines these representative features together to form a feature vector, and the extracted features can highlight the key information of the operation state of the thermal power generation system, making the analysis more targeted and helping to find potential problems and rules in the system operation.

[0086] Specifically, by projecting and mapping the feature vector into a low-dimensional space, the dimensionality of the feature vector is reduced, which can prevent the initial extraction of a large number of features from containing a large amount of redundant information and increasing the computational burden. Through dimensionality reduction, the dimensionality of the feature vector is reduced while retaining key information, making the data easier to process and analyze, removing features that have little distinction or strong correlation with the system operating state, highlighting the main features that affect the system operating state, and helping to improve the generalization ability of the model and avoid overfitting. The reduced feature vector contains data samples under different operating states of the thermal power generation system, and the high-quality samples are selected from the mass of data samples, which can reflect the best operating state of the thermal power generation system. Using high-quality samples to build a regression model can reduce the influence of noise and abnormal data, improve the accuracy and reliability of the model, and thus more accurately predict the best control parameters.

[0087] Specifically, after the current operating state data of the thermal power generation system is processed by the feature extraction and dimensionality reduction steps, a low-dimensional feature vector of the current state is obtained, and the low-dimensional feature vector is input into the trained regression model. The regression model calculates the corresponding best control parameters according to the relationship between the operating state and the control parameters learned before. The best control parameters can adjust the system from the current operating state to the best operating state. According to the operating state of the thermal power generation system, the corresponding best control parameters are calculated in real time to realize intelligent adaptive control, which is difficult to achieve by traditional fixed model control methods.

[0088] Further, the analysis of the historical operating data and the calculation of the features related to the operating state of the thermal power generation system include:

[0089] S301, according to the historical operating data, the features related to the operating state of the thermal power generation system are calculated by the thermodynamic formula, and a plurality of thermodynamic features are obtained;

[0090] S302, the feature weight value of each thermodynamic feature is calculated by a pre-set feature weighting model;

[0091] S303, the corresponding thermodynamic features are weighted according to the feature weight value, and a plurality of weighted features are obtained;

[0092] S304, the plurality of weighted features are combined to obtain a feature vector.

[0093] In the present embodiment, the thermal power generation system involves a complex energy conversion process, and thermodynamic principles run through it. By analyzing the temperature, pressure, flow, and other parameters in historical operation data, a series of characteristics reflecting the energy state and conversion efficiency of the system can be calculated through thermodynamic formulas. For example, the enthalpy and entropy of steam are closely related to the work capacity and energy quality of the system. These thermodynamic characteristics intuitively describe the operating state of the thermal power generation system from an energy perspective.

[0094] Specifically, data related to thermodynamic calculations, such as steam temperature, pressure, flow, etc., are extracted from historical operation data. These data are usually collected by sensors distributed in key equipment such as boilers and steam turbines. According to specific thermodynamic characteristic calculation requirements, appropriate thermodynamic formulas are selected. For example, in calculating the enthalpy of steam, for water vapor, the IAPWS-IF97 formula published by the International Association for the Properties of Water and Steam is used in the present embodiment (this formula is well-known in the field and will not be described here). Based on the temperature and pressure of the steam, the formula calculates the accurate enthalpy value. The entropy value can also be calculated according to related thermodynamic formulas combined with temperature, pressure, and other parameters. The calculated multiple thermodynamic characteristics are organized to form a set of thermodynamic characteristics including enthalpy, entropy, specific volume, etc.

[0095] Specifically, different thermodynamic characteristics have different degrees of influence on the operating state of the thermal power generation system. Through a feature weighting model, each thermodynamic characteristic is assigned a weight value according to the importance difference of each characteristic. In the present embodiment, the feature weighting model is a feature importance evaluation model based on random forest. This model is trained with a large amount of historical data, taking the thermodynamic characteristics in historical operation data as input variables and the key performance indicators of the system as output variables to build a random forest model. The model evaluates the importance contribution of each input feature to the output variable and outputs the weight value corresponding to each thermodynamic characteristic. These weight values reflect the relative importance of each characteristic in describing the operating state of the thermal power generation system.

[0096] Specifically, according to the calculated feature weight values, the thermodynamic characteristics are weighted. Each thermodynamic characteristic is multiplied by its corresponding feature weight value. After weighting, the influence of important features can be highlighted, and the influence of relatively unimportant features can be weakened. This makes the constructed feature vector more accurately reflect the comprehensive effect of each thermodynamic characteristic on the operating state of the thermal power generation system, enhancing the representativeness and effectiveness of the feature vector. The multiple thermodynamic characteristics after weighting are combined in the order of feature weight values from high to low to form a multi-dimensional feature vector. The feature vector integrates the information of multiple thermodynamic characteristics and highlights the influence of key features through weighting. It can comprehensively and accurately describe the operating state of the thermal power generation system.

[0097] Further, the dimension reduction of the feature vector by projecting and mapping the feature vector to a low-dimensional space to obtain a low-dimensional feature vector comprises:

[0098] S401, analyzing the dependency relationship between each feature in the feature vector and the preset target variable, and calculating a mutual information value;

[0099] S402, according to the mutual information value, retaining the corresponding feature whose mutual information value is greater than a preset mutual information threshold value, to obtain a first feature vector;

[0100] S403, projecting and mapping the first feature vector to a low-dimensional space to obtain a low-dimensional feature vector.

[0101] In this embodiment, the dependency relationship between each feature in the feature vector and the preset target variable is calculated by the mutual information value. Each feature in the feature vector represents some aspect of information of the operation state of the thermal power generation system, and the preset target variable reflects the key performance indicators of the system, such as power generation efficiency, energy consumption, etc. By calculating the mutual information value between the feature and the target variable, the information contribution degree of each feature to the target variable can be quantified. The higher the mutual information value, the stronger the dependency relationship between the feature and the target variable, and the more valuable the feature is in describing the change of the target variable. Conversely, the lower the mutual information value, the weaker the correlation between the feature and the target variable. A mutual information calculation method based on entropy is used, and the calculation formula is:

[0102] I(X; Y) = H(X) + H(Y) - H(X, Y);

[0103]

[0104]

[0105] In the formula, I(X; Y) is the mutual information value between the feature variable X and the target variable Y, H(X) is the entropy value of the feature variable X, H(Y) is the entropy value of the target variable Y, H(X, Y) is the joint entropy value of the feature variable X and the target variable Y, p(x i ) is the probability that the feature variable X takes the value x i , n is the number of all values of the feature variable X, p(y j ) is the probability that the target variable Y takes the value y j , m is the number of all values of the target variable Y, p(x i , y j ) is the joint probability that the feature variable X takes the value x i and the target variable Y takes the value y j .

[0106] Specifically, according to the historical data characteristics of the thermal power generation system and the actual calculation precision requirement, a suitable mutual information threshold is set, and according to the mutual information threshold, the corresponding features with the mutual information value greater than the preset mutual information threshold are retained, and the corresponding features with the mutual information value less than or equal to the preset mutual information threshold are removed, to obtain the features after preliminary screening. The retained features are recombined according to the order in the original feature vector to form a first feature vector. Through preliminary dimension reduction, the dimension of the feature vector can be effectively reduced, a large number of features with weak influence on the target variable are removed, and only those features with important influence on the target variable are focused on, the interference of noise and redundant information is reduced, and thus the analysis efficiency and the convergence speed of the model are improved.

[0107] Specifically, by mapping the first feature vector to a low-dimensional space, the data dimension is further reduced under the premise of retaining the main features and structure of the data, which can avoid the problem that a single dimension reduction method cannot combine the linear and nonlinear characteristics of feature data, and is not conducive to further data analysis and model training. Through mutual information analysis and feature selection and a reasonable mapping method, efficient feature vector dimension reduction is realized, which reduces the data dimension while retaining the information related to the key performance indicators of the thermal power generation system to the greatest extent, and provides a better data basis for subsequent data analysis and model training.

[0108] Further, the projecting and mapping the first feature vector to a low-dimensional space to obtain a low-dimensional feature vector comprises:

[0109] S501, calculate the distance between each feature in the first feature vector to obtain a distance similarity;

[0110] S502, connect the features with a distance similarity greater than a preset similarity threshold to form a feature network, wherein each network point in the feature network is a feature, and the network points are connected according to the distance similarity;

[0111] S503, constructing a projection plane according to the orthogonal vectors of the first feature vector;

[0112] S504, projecting each network edge in the feature network onto the projection plane to obtain a plurality of plane projection lines;

[0113] S505, calculating the intersection point or centroid of each two plane projection lines according to the plane projection lines, wherein when two plane projection lines intersect, the intersection point is calculated, and when two plane projection lines do not intersect, the centroid is calculated;

[0114] S506, until the calculation of the intersection point or centroid between each two plane projection lines in all plane projection lines is completed;

[0115] S507, splice the calculated intersection and the centroid to obtain a low-dimensional feature vector.

[0116] In the embodiment, the distance between each feature in the first feature vector is calculated to measure the similarity between the features. The closer the distance, the more similar the change trend of the two features in the numerical value, and the more similar the representation of the thermal power generation system operation state. The distance similarity between each feature is obtained by calculating the Euclidean distance. According to the historical data characteristics of the thermal power generation system and the actual calculation accuracy requirement, the similarity threshold is set.

[0117] Specifically, when the distance similarity of two features is greater than the threshold, it means that they are more similar in numerical performance and influence on the system operation state. They are connected to form a feature network. In this network, each node represents a feature, and the weight of the edge is the distance similarity between the corresponding nodes. The existence of the edge indicates that the features have strong similarity. By constructing the feature network, features with similar properties can be integrated together. From the perspective of network structure, analyzing the relationship between features helps to discover the connection and distribution rule between features. The length of the connected edge in the feature network reflects the closeness between features. The shorter the connected edge, the higher the distance similarity, indicating that the two features are more closely related and have a higher synergistic effect in describing the operation state of the thermal power generation system. By calculating the comprehensive situation of the length of the connected edge between each feature and other connected features, a weight is assigned to each feature. The higher the weight, the more closely related the feature is to other features in the feature network, and the greater the comprehensive influence on the system operation state.

[0118] As Figure 3 , the distance similarity threshold is set to 0.5. The features with a distance similarity greater than 0.5 are connected, and the weight of the edge represents the similarity. The higher the similarity, the shorter the length of the corresponding edge. The feature network is constructed. According to the feature network, a projection plane is calculated. In the embodiment, the orthogonal vectors of the first feature vector are calculated to construct a two-dimensional projection plane. Specifically, principal component analysis is performed on the first feature vector to obtain the first two principal component vectors in the principal component analysis result. The two principal component vectors are orthogonal to each other. According to the two principal component vectors, a projection plane is constructed. Using the principal component vectors obtained by principal component analysis to construct the projection plane can ensure that the projected features can retain the main variance information of the data, thereby reducing information loss.

[0119] Specifically, the feature network is projected onto the projection plane, as Figure 4The constructed feature network is projected, and a plurality of plane projection lines are obtained on the projection plane, so that the relationship between the feature points can be directly observed, the feature points are dimensionally reduced according to the plane projection lines after projection, the plane projection lines are calculated two by two, and the intersection points or centroids between each two plane projection lines are calculated, when two plane projection lines intersect, the intersection point is directly selected as the feature point after dimensional reduction, the intersection point can comprehensively reflect the feature information between the two plane projection lines, and can reflect the relationship information between the original feature vectors; when two plane projection lines do not intersect, the centroid is calculated, and the centroid is calculated according to the end points of the two projection lines; new feature information is extracted from the plane projection lines by calculating the intersection points and the centroids, the feature information can better reflect the relationship between the original feature vectors, and the feature dimensional reduction is realized, the high-dimensional feature vector is mapped to a low-dimensional space, the feature dimension is reduced, and the subsequent calculation speed is improved.

[0120] Further, all the plane projection lines are traversed, the intersection points or centroids between two plane projection lines are calculated, the geometric features of the feature network on the projection plane are comprehensively extracted, rich feature information is extracted, and the relationship between the original feature vectors is more accurately represented; the coordinates of the calculated intersection points and centroids are spliced in order to form a new vector, and the vector is a low-dimensional feature vector. By splicing, a plurality of feature points on a two-dimensional plane are combined into a vector, the intersection points and the centroids are spliced into a low-dimensional feature vector, the geometric features on the two-dimensional plane can be converted into a one-dimensional vector, meanwhile, the low-dimensional feature vector can reduce the dimension of data, improve the calculation efficiency and the generalization ability of the model, this combination mode considers the similarity and synergistic effect between features, and also considers the diversity of features, so that the low-dimensional feature vector can more effectively describe the operation state of the thermal power generation system.

[0121] Further, the low-dimensional feature vector is used to filter data samples to obtain high-quality samples, including:

[0122] S601, the data samples are clustered according to the similarity between the low-dimensional feature vectors to obtain first high-quality samples;

[0123] S602, the first high-quality samples are expanded by using a preset sample expansion model to obtain second high-quality samples;

[0124] S603, the first high-quality samples and the second high-quality samples are combined to obtain high-quality samples.

[0125] In the embodiment, according to the low-dimensional feature vectors, the data samples are dimensionally reduced, the high-quality samples capable of accurately reflecting the state of the thermal power generation system are screened from a large number of data samples, and through clustering screening and sample expansion, the quantity and diversity of the high-quality samples are greatly enriched, more sufficient and comprehensive data support is provided for data analysis and optimization of the thermal power generation system, compared with the single or simple combined sample processing manner of the prior art, the high-quality samples can be more effectively screened and expanded, and more high-quality data samples are provided for data analysis and optimization of the thermal power generation system.

[0126] Specifically, first, the original data is clustered, the data samples with similar operation state characteristics are classified into a class by calculating the similarity between the low-dimensional feature vectors, each class represents a similar operation state mode, the high-quality samples are retained in each class, the data is preliminarily dimensionally reduced, the first high-quality samples are obtained, through the clustering algorithm, the data with similar high-quality operation characteristics can be quickly screened from a large number of data samples, the efficiency of high-quality sample screening is greatly improved, and the workload of manual screening is reduced. Secondly, the first high-quality samples are expanded to increase the richness and diversity of the high-quality samples, and the more rich high-quality sample data is used for data analysis and model training to improve the generalization ability and accuracy of the model.

[0127] Specifically, the first high-quality samples and the second high-quality samples are combined to obtain high-quality sample data, the first high-quality samples are real high-quality operation samples screened from the original data and have high reliability and representativeness; the second high-quality samples are new samples generated on the basis of the first high-quality samples by the sample expansion model, and the quantity and diversity of the samples are increased; the combination of the two can make full use of the authenticity of the first high-quality samples and the expansibility of the second high-quality samples, form a more comprehensive and rich high-quality sample set, and contain not only the high-quality state samples appearing in actual operation but also the high-quality state samples generated by the model, thereby providing more sufficient data support for analyzing the relationship between the operation state of the thermal power generation system and the control parameters.

[0128] Further, the clustering of the data samples according to the similarity between the low-dimensional feature vectors to obtain the first high-quality samples comprises:

[0129] S701, setting a performance threshold range of high-quality samples according to a preset performance index;

[0130] S702, screening sample data satisfying the performance threshold range according to the low-dimensional feature vectors to obtain initial high-quality samples;

[0131] S703, clustering the data samples according to the similarity between the low-dimensional feature vectors to obtain first-class samples;

[0132] S704, calculate the distance between the cluster center of each class of samples and the initial high-quality samples, to obtain a first distance;

[0133] S705, reserve the class corresponding to the cluster center with the first distance less than a preset first distance threshold, to obtain second-class samples;

[0134] S706, in the second-class samples, calculate the distance between the samples in each class and the cluster center, to obtain a second distance;

[0135] S707, reserve the samples with the second distance less than a preset second distance threshold, to obtain similar high-quality samples;

[0136] S708, combine the initial high-quality samples, the cluster centers of the second-class samples, and the similar high-quality samples, to obtain first high-quality samples.

[0137] In this embodiment, as Figure 5 , the initial high-quality samples are first screened out, the samples are clustered, the distance between the cluster center of each class and the initial high-quality samples is calculated, the high-quality classes are selected, and the sample data close to the cluster center is selected from the high-quality classes to jointly constitute the first high-quality samples.

[0138] Specifically, according to the preset performance index, the performance threshold range of the high-quality samples is set, and the threshold range can be set according to the running state of the thermal power generation system and the actual calculation precision requirement. For example, the power generation efficiency is set to be greater than 90%, the coal consumption is set to be lower than 300 grams / kilowatt hour, the sulfur dioxide emission concentration is set to be lower than 50 milligrams / cubic meter, etc. The low-dimensional feature vector is matched with the preset performance threshold range, and the sample data that performs well in the key performance index is screened out as the initial high-quality samples. Through strict performance threshold screening, it is ensured that the initial high-quality samples perform well in the key performance index.

[0139] Specifically, according to the similarity between low-dimensional feature vectors, the data samples are clustered to obtain first-class samples. By calculating the distance between each sample data feature vector, data samples with similar running state characteristics and close distance are divided into the same category. The cluster center of each class is extracted from each class, and the distance between the cluster center and the initial high-quality sample is calculated to obtain a first distance. The closeness of each class sample to the known high-quality running state is evaluated. The smaller the distance, the closer the running state represented by the sample to the high-quality running state, and vice versa. According to actual calculation requirements, a first distance threshold is set. When the first distance is less than the threshold, it means that the category corresponding to the cluster center is close to the high-quality running state represented by the initial high-quality sample, and has high potential quality. The samples of these categories are retained to obtain second-class samples, which can further focus on data related to the high-quality running state and remove categories that are significantly different from the high-quality running state, thereby narrowing the data range and improving the accuracy of the screening.

[0140] Specifically, in the second-class samples, each category represents a running state mode that is close to the high-quality running state. The distance between each sample in the class and the cluster center is calculated to evaluate the deviation of each sample in its own category. The smaller the distance, the more representative the sample is of the typical characteristics of its own category, and the higher the similarity to the high-quality running state. According to the actual calculation accuracy requirements, a second distance threshold is set. Samples with a second distance less than the threshold are retained. A distance less than the threshold means that the sample is close to the typical characteristics of its own category and represents the high-quality running state. Retaining these samples can filter out samples similar to the high-quality running state from the second-class samples, further enriching the set of high-quality samples and improving the quality and representativeness of the high-quality samples. The initial high-quality samples, the cluster centers of the second-class samples, and the similar high-quality samples are combined to obtain first high-quality samples. The combined high-quality samples contain more high-quality sample information.

[0141] Further, the first high-quality samples are expanded by a preset sample expansion model to obtain second high-quality samples, including:

[0142] S801, according to the first high-quality samples, a generated sample is obtained by a preset sample expansion model;

[0143] S802, by calculating the performance indicators of the generated samples, the generated samples that meet the performance threshold range of the high-quality samples are retained to obtain second high-quality samples.

[0144] In the embodiment, the first high-quality sample is expanded to obtain more abundant high-quality sample data, and the model trained based on the abundant high-quality sample has higher accuracy and generalization, thereby improving the accuracy of the control parameter calculation. According to the first high-quality sample, a generated sample is obtained through a preset sample expansion model. In the embodiment, the sample expansion model is a GAN model. The first high-quality sample is used to train the sample expansion model to obtain a pre-trained sample expansion model, and a generated sample close to the real first high-quality sample is generated. Through the generated sample, the number of high-quality samples can be effectively expanded, more abundant data resources are provided for subsequent data analysis and model training, and the generalization ability and accuracy of the model are improved.

[0145] Specifically, the generated sample is screened, the performance indicators of the generated sample are calculated, the generated sample meeting the performance threshold range of the high-quality sample is retained to obtain a second high-quality sample, and the performance indicators of the generated sample, such as power generation efficiency, coal consumption, and pollutant emission, are calculated and compared with the preset performance threshold range of the high-quality sample. The sample that truly meets the high-quality operation standard is screened from the generated sample, which avoids data pollution caused by inferior samples in the generated sample. Through strict performance threshold screening, it is ensured that the second high-quality sample meets the high-quality operation standard in key performance indicators, and the quality of the expanded high-quality sample set is improved.

[0146] Further, the relationship between the operation state of the thermal power generation system and the control parameter is analyzed according to the high-quality sample, and a regression model is constructed, including:

[0147] S901, a first regression model is obtained by preliminarily analyzing the relationship between the operation state of the thermal power generation system and the control parameter according to the high-quality sample;

[0148] S902, a prediction result obtained by prediction through the first regression model is used as a new sample;

[0149] S903, the high-quality sample and the new sample are combined to fit the relationship between the operation state of the thermal power generation system and the control parameter, and a regression model is obtained.

[0150] In the embodiment, the relationship between the operation state of the thermal power generation system and the control parameter is analyzed according to the high-quality sample, and a regression model is constructed. Multiple models are fused, the information in the high-quality sample data can be more comprehensively and deeply mined and utilized, compared with the single modeling method in the prior art, the regression model can be more effectively constructed, and the complex relationship between the operation state of the thermal power generation system and the control parameter can be more accurately described, thereby providing accurate mathematical model support for the optimization control of the system.

[0151] Specifically, according to the high-quality sample data, the running state parameters are taken as independent variables, the control parameters are taken as dependent variables, a random forest regression method is used for fitting, a first regression model is constructed, and the complex relationship between the running state and the control parameters of the thermal power generation system is preliminarily quantified by constructing the first regression model; the running state parameters in the high-quality sample are input into the first regression model, the first regression model predicts according to the relationship learned by it, and the corresponding control parameter prediction value is obtained; the prediction result is taken as a new sample, which can test the prediction ability of the first regression model on the one hand, and provide more data points for further fitting the relationship on the other hand, enrich the sample set, and help to more comprehensively depict the relationship between the running state and the control parameters.

[0152] Specifically, the high-quality sample and the new sample are combined to form a more rich and representative sample set, the combined sample set is used for further fitting, the regression model can learn a more comprehensive and accurate relationship between the running state and the control parameters, and the regression model is obtained; the new sample set has increased in data quantity and data diversity, so that the regression model can better capture the complex patterns and rules in the data, thereby obtaining a more optimized and accurate regression model.

[0153] Embodiment 2:

[0154] In this embodiment, as Figure 6 , an Agent-driven big data analysis and thermal power generation system operation control system is provided for implementing the Agent-driven big data analysis and thermal power generation system operation control method, comprising:

[0155] A data acquisition module acquires running data of the thermal power generation system;

[0156] An optimal control parameter calculation module inputs the running data into the power system agent for feature extraction and projection dimension reduction, analyzes the running state of the thermal power generation system, and calculates the optimal control parameter for adjusting the thermal power generation system to the optimal running state;

[0157] A thermal power generation system control module controls the running of the thermal power generation system according to the optimal control parameter to adjust the running state of the thermal power generation system in real time.

[0158] In this embodiment, the data acquisition module includes a sensor network, a data transmission line, a data collection interface, and a data cache unit. The module acquires real-time and accurate operation data of each link of the thermal power generation system through various sensors, comprehensively reflects the operation state of the system, converts the original data collected by the sensors into a format, and stably transmits the data to the subsequent processing module through the data transmission line, thereby providing a data basis for system analysis and decision-making. The data cache unit is used for temporarily storing data, ensuring data integrity when data processing fluctuates, and facilitating preliminary management and monitoring of data.

[0159] Specifically, the optimal control parameter calculation module includes a power system agent. Through the powerful analysis capability of the power system agent, the operation state of the thermal power generation system is analyzed in depth in combination with the preprocessed operation data, and the difference between the current state and the optimal operation state is identified. Based on the reasoning and decision-making of the agent, the control parameters required for the thermal power generation system to reach the optimal operation state are calculated, such as fuel supply amount, valve opening degree, equipment speed, and other adjustment values. Through model training and updating components, the performance and accuracy of the power system agent are continuously improved, so that it can better adapt to the dynamic changes of the thermal power generation system and continuously provide better control parameter calculation results.

[0160] Specifically, the thermal power generation system control module includes an actuator control unit, a feedback sensor and monitoring unit, a control strategy adjustment unit, and a communication interface and coordination unit. According to the parameters provided by the optimal control parameter calculation module, the actuator action of the thermal power generation system is accurately controlled to realize real-time adjustment of the system operation state, so that the system is adjusted towards the optimal operation state. The control effect is monitored in real time through the feedback sensor, and the actual operation data is fed back to the control strategy adjustment unit to provide a basis for further optimizing the control strategy. According to the feedback data, the control strategy is adjusted to continuously optimize the control effect, and at the same time, it works collaboratively with other system modules to ensure the overall stable and efficient operation of the thermal power generation system.

[0161] The basic principles and main features of the present application and the advantages of the present application have been shown and described. It should be understood by those skilled in the art that the present application is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for agent-driven big data analysis and operation control of a thermal power generation system, characterized by, The method comprises the following steps: acquiring operation data of a thermal power generation system; analyzing historical operation data to calculate features related to the operation state of the thermal power generation system and construct a feature vector; dimensionally reducing the feature vector by projecting and mapping the feature vector to a low-dimensional space to obtain a low-dimensional feature vector; screening data samples according to the low-dimensional feature vector to obtain high-quality samples; analyzing the relationship between the operation state of the thermal power generation system and control parameters according to the high-quality samples to construct a regression model; calculating the optimal control parameters for adjusting the thermal power generation system to the optimal operation state through the regression model according to the current operation state of the thermal power generation system; controlling the operation of the thermal power generation system according to the optimal control parameters to adjust the operation state of the thermal power generation system in real time.

2. The agent-driven big data analytics and thermal power plant operation control method according to claim 1, characterized in that, The method of analyzing historical operation data to calculate features related to the operation state of the thermal power generation system and construct a feature vector comprises the following steps: calculating features related to the operation state of the thermal power generation system through thermodynamic formulas according to historical operation data to obtain a plurality of thermodynamic features; calculating the feature weight values of each thermodynamic feature through a preset feature weighting model; weighting the corresponding thermodynamic features according to the feature weight values to obtain a plurality of weighted features; combining the plurality of weighted features to obtain a feature vector.

3. The agent-driven big data analytics and thermal power plant operation control method of claim 1, wherein, The method of dimensionally reducing the feature vector by projecting and mapping the feature vector to a low-dimensional space to obtain a low-dimensional feature vector comprises the following steps: analyzing the dependency relationship between each feature in the feature vector and a preset target variable to calculate mutual information values; retaining the corresponding features with mutual information values greater than a preset mutual information threshold value according to the mutual information values to obtain a first feature vector; projecting and mapping the first feature vector to a low-dimensional space to obtain a low-dimensional feature vector.

4. The agent-driven big data analytics and thermal power plant operation control method according to claim 3, characterized in that, The method of projecting and mapping the first feature vector to a low-dimensional space to obtain a low-dimensional feature vector comprises the following steps: calculating the distance between each feature in the first feature vector to obtain a distance similarity; connecting the features with a distance similarity greater than a preset similarity threshold value to form a feature network, wherein each network point in the feature network is a feature, and the network points are connected according to the distance similarity; constructing a projection plane according to the orthogonal vectors of the first feature vector; projecting each network edge in the feature network onto the projection plane to obtain a plurality of plane projection lines; calculating the intersection points or centroids of each two plane projection lines according to the plane projection lines, wherein when two plane projection lines intersect, the intersection point is calculated, and when two plane projection lines do not intersect, the centroid is calculated; calculating the intersection points or centroids between each two plane projection lines in all plane projection lines until the calculation is completed; splicing the calculated intersection points and centroids to obtain a low-dimensional feature vector.

5. The agent-driven big data analytics and thermal power plant operation control method of claim 1, wherein, The method of screening data samples according to the low-dimensional feature vector to obtain high-quality samples comprises the following steps: clustering data samples according to the similarity between low-dimensional feature vectors to obtain first high-quality samples; expanding the first high-quality samples through a preset sample expansion model to obtain second high-quality samples; Combine the first high-quality sample and the second high-quality sample to obtain a high-quality sample.

6. The agent-driven big data analytics and thermal power plant operation control method of claim 5, wherein, The first high-quality sample is obtained by clustering data samples according to similarities between low-dimensional feature vectors. A performance threshold range of the high-quality sample is set according to a preset performance index. Initial high-quality samples are obtained by screening sample data satisfying the performance threshold range according to the low-dimensional feature vectors. The first distance is obtained by calculating distances between clustering centers of each class of samples and the initial high-quality samples. The second class of samples is obtained by retaining classes corresponding to clustering centers with the first distance less than a preset first distance threshold. The second distance is obtained by calculating distances between sample data in each class of samples and the clustering centers in the second class of samples. Similar high-quality samples are obtained by retaining sample data with the second distance less than a preset second distance threshold. The first high-quality sample is obtained by combining the initial high-quality samples, clustering centers of the second class of samples, and the similar high-quality samples. The second high-quality sample is obtained by expanding the first high-quality sample through a preset sample expansion model.

7. The agent-driven big data analytics and thermal power plant operation control method of claim 5, wherein, The second high-quality sample is obtained by retaining generated samples satisfying the performance threshold range of the high-quality sample by calculating performance indexes of the generated samples. The regression model is constructed by analyzing a relationship between the operation state of the thermal power generation system and the control parameter according to the high-quality sample. The first regression model is obtained by preliminarily analyzing the relationship between the operation state of the thermal power generation system and the control parameter according to the high-quality sample.

8. The agent-driven big data analytics and thermal power plant operation control method of claim 1, wherein, The regression model is obtained by fitting the relationship between the operation state of the thermal power generation system and the control parameter by combining the high-quality sample and the new sample. The Agent-driven big data analysis and thermal power generation system operation control method is implemented, and includes: A data acquisition module acquires operation data of the thermal power generation system. An optimal control parameter calculation module analyzes historical operation data, calculates features related to the operation state of the thermal power generation system, constructs a feature vector, projects and maps the feature vector to a low-dimensional space to reduce the feature vector to obtain a low-dimensional feature vector, filters data samples according to the low-dimensional feature vector to obtain high-quality samples, analyzes a relationship between the operation state of the thermal power generation system and the control parameter according to the high-quality samples, and constructs a regression model.

9. Agent-driven big data analysis and thermal power generation system operation control system, characterized by, The thermal power generation system control module controls the operation of the thermal power generation system according to the optimal control parameter to adjust the operation state of the thermal power generation system in real time. ​ ​ ​

Images