Thermoelectric unit performance index auxiliary analysis system and method

Abstract

The invention discloses a thermoelectric unit performance index auxiliary analysis system and a thermoelectric unit performance index auxiliary analysis method, belongs to the technical field of thermoelectric units, and solves the problem that an existing method lacks deep analysis of real-time data and abnormal data identification capability. Performing difference feature extraction on the abnormal data set by the performance index evaluation model to obtain a difference feature set, weighting the difference feature set by the performance index evaluation model based on a principal component analysis method, and combining real-time working parameters and substituting the real-time working parameters into a Tops is method to compute a risk value of the unit equipment; according to the method, the real-time data is pre-identified based on the parameter variation coefficient, the abnormal working data can be rapidly extracted, meanwhile, the key index factors influencing the performance of the thermoelectric unit can be more accurately determined based on the abnormal working data, and the performance index evaluation model is trained through supervised learning, so that the performance of the thermoelectric unit is evaluated. And the accuracy and generalization ability of the performance index evaluation model are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of thermal power units, and particularly relates to a system and method for assisting in analyzing the performance indicators of thermal power units. Background Art

[0002] With the adjustment of the energy structure and the improvement of environmental protection requirements, the performance optimization and economic operation of thermal power units have become important topics in the power industry. As an efficient energy utilization system, thermal power units can provide both electricity and heat energy simultaneously. The accurate evaluation of their performance indicators is of great significance for optimizing operation, saving energy and reducing consumption, and improving economic benefits.

[0003] In recent years, with the development of computer technology, big data analysis, and machine learning technology, intelligent thermal power unit performance analysis systems have gradually become a research hotspot. By automatically collecting and analyzing operation data and combining machine learning algorithms, the performance of the units can be predicted and evaluated more accurately. Chinese Patent CN114997573A discloses a method and system for online performance evaluation of gas-steam combined cycle thermal power units, including the following steps: constructing a unit operating condition database based on the historical operation parameters of the thermal power unit; calculating the thermodynamic basic indicators of the unit based on the constructed unit operating condition database; collecting the operation data of the thermal power unit in real time, determining the current operating condition of the thermal power unit according to the constructed unit operating condition database, and calculating the current thermodynamic indicators of the unit; judging whether the difference between the current thermodynamic indicators of the unit and the thermodynamic basic indicators exceeds the indicator threshold. If so, it is judged as a weak point of energy exchange, and the online performance evaluation of the unit is completed. However, when the existing methods evaluate the performance of thermal power units, they use static historical data and fixed thermodynamic indicators. Although they can initially judge the weak points of energy exchange, they lack the ability to deeply analyze real-time data and identify abnormal data. To address the above problems, we propose a system and method for assisting in analyzing the performance indicators of thermal power units. Summary of the Invention

[0004] The purpose of the present invention is to provide a system and method for assisting in analyzing the performance indicators of thermal power units in view of the deficiencies of the existing technologies, and to solve the problem that when the existing methods evaluate the performance of thermal power units, they use static historical data and fixed thermodynamic indicators. Although they can initially judge the weak points of energy exchange, they lack the ability to deeply analyze real-time data and identify abnormal data.

[0005] The present invention is implemented as follows. A method for assisting in analyzing the performance indicators of a thermal power unit, the method for assisting in analyzing the performance indicators of a thermal power unit includes:

[0006] Collect sensors to collect the real-time working parameters of the thermal power unit based on a preset collection period, and pre-identify and extract the real-time working parameters based on the parameter coefficient of variation to identify abnormal working data in the real-time working parameters;

[0007] Feedback the abnormal working data in the real-time working parameters, retrieve the historical acquisition parameters of the acquisition sensors that uploaded the abnormal working data, and integrate the historical acquisition parameters and abnormal working data into an abnormal data set;

[0008] Pre-construct a performance index evaluation model, pick up modeling sample data based on web crawler technology, perform correlation analysis on the modeling sample data, determine the index factors affecting the performance of the thermal power unit, and perform index factor annotation processing on the modeling sample data. Perform outlier and normalization processing on the modeling sample data, convert the modeling sample data into a data type for supervised learning, and use the modeling sample data to train the performance index evaluation model;

[0009] Obtain the abnormal data set, the performance index evaluation model extracts differential features from the abnormal data set to obtain a differential feature set. The performance index evaluation model weights the differential feature set based on the principal component analysis method, combines the real-time working parameters and brings them into the Tops is method to calculate the equipment risk value of the unit;

[0010] Obtain the equipment risk value of the unit, judge whether the equipment risk value meets the preset equipment safety threshold. If it does not meet the preset safety threshold, trigger an equipment risk warning instruction.

[0011] Load the equipment risk value of the unit, and use the grey relational analysis method to calculate the associated risk value of the associated equipment associated with the unit equipment.

[0012] Preferably, the method for pre-identifying and extracting real-time working parameters based on the parameter coefficient of variation specifically includes:

[0013] Obtain the real-time working parameters, perform dimensionality reduction transformation on the real-time working parameters, and convert the real-time working parameters into low-dimensional real-time working parameters;

[0014] Load the real-time working parameters after dimensionality reduction, use the fuzzy clustering algorithm to generate parameter clustering clusters corresponding to the real-time working parameters, and determine the centroid of the parameter clustering clusters;

[0015] Calculate the distance between the real-time working parameters and the centroid, generate a parameter distance set, sort the parameter distance set, traverse the parameter distance set, and calculate the local density of the parameter distance set;

[0016] Load the local density of the parameter distance set, and combine the analytic hierarchy process to determine the initial coefficient of variation of the parameter;

[0017] Obtain the initial coefficient of variation of the parameter, and correct the initial coefficient of variation based on the index type weight and equipment weight corresponding to the real-time working parameters to determine the parameter coefficient of variation;

[0018] Pre-identify and extract the real-time working parameters based on the parameter coefficient of variation, and identify the abnormal working data in the real-time working parameters;

[0019] When determining the parameter variation coefficient by modifying the initial variation coefficient based on the index type weight and device weight corresponding to the real-time working parameters, the parameter variation coefficient is calculated by the following formula:

[0020]

[0021] where a c.v , respectively represent the parameter variation coefficient and the initial variation coefficient, W i represents the parameter judgment matrix constructed by the analytic hierarchy process based on the real-time working parameter X t , X dat is the local density of the parameter distance set, n is the number of the parameter distance set, q i , q j respectively represent the index type weight and device weight corresponding to the real-time working parameters, dist(i,z) represents the distance between the real-time working parameter and the centroid, μ z represents the centroid of the parameter clustering cluster, and m represents the number of index types.

[0022] Preferably, the method for training the performance index evaluation model using the modeling sample data specifically includes:

[0023] Taking the LSTM neural network model as the initial model of the performance index evaluation model, the LSTM neural network model consists of a set of input layers, three sets of intermediate layers and a set of output layers. The three sets of intermediate layers are the first intermediate layer, the second intermediate layer and the third intermediate layer respectively, and the input layer, the intermediate layer and the output layer are connected in sequence;

[0024] Introducing a U-shaped autoencoder in the input layer, the U-shaped autoencoder is used for upscaling and reconstructing the abnormal data set. The U-shaped autoencoder includes a downsampling layer, a max pooling layer and an upsampling layer, and the downsampling layer, the max pooling layer and the upsampling layer are connected in sequence;

[0025] Freezing the three sets of intermediate layers of the initial model, replacing the second intermediate layer with the diffusion decoder of the conditional time diffusion network model, and introducing the finite element algorithm. Based on the diffusion decoder combined with the finite element algorithm, differential feature extraction is performed on the abnormal data set;

[0026] Introducing a deformable convolutional neural network model between the third intermediate layer and the output layer. The deformable convolutional neural network model consists of three sets of convolutional layers, three sets of residual connection layers, a deformable convolutional neural network and a linear layer. The principal component analysis method is introduced in the linear layer. Freezing the output layer of the performance index evaluation model and replacing the output layer with a classifier, and the Topsis algorithm is introduced in the classifier;

[0027] The classifier consists of a group of global average pooling layers and three layers of multi-layer perceptrons. The risk classification results of the unit equipment are output through the multi-layer perceptrons, which is expressed as:

[0028]

[0029] Among them, Y p represents the output of the risk classification results of the unit equipment. l is the number of output nodes p of the multi-layer perceptron network. is the feature of the p-th output node of the multi-layer perceptron network. X avgpool represents the output feature of the global average pooling layer;

[0030] Load the modeling sample data, divide the modeling sample data into a training set and a validation set according to the allocation ratio of 3:1, set the number of iteration rounds for evaluating the performance of the model as 520, the training batch size for a single round as 25, and the learning rate as 0.01;

[0031] Obtain the training set. The training set uses the differential loss function and the metric optimization algorithm to train the performance metric evaluation model, and outputs the converged performance metric evaluation model;

[0032] Load the validation set, verify the detection accuracy of the performance metric evaluation model based on the validation set, and determine whether the detection accuracy of the performance metric evaluation model for the risk assessment of the unit equipment meets the preset accuracy threshold. If it meets the preset accuracy threshold, output the converged performance metric evaluation model.

[0033] Preferably, the method of combining real-time working parameters and bringing them into the Tops is method to calculate the risk value of the unit equipment specifically includes:

[0034] Load the abnormal data set, and the U-shaped autoencoder performs upscaling and reconstruction processing on the abnormal data set, and outputs the abnormal index matrix;

[0035] Load the abnormal index matrix, and the diffusion decoder combines the finite element algorithm to perform Fourier frequency analysis on the abnormal index matrix, screens out the index factors exceeding the preset frequency threshold based on the preset frequency threshold, and performs index factor annotation processing on the modeling sample data to obtain the index annotation set;

[0036] The diffusion decoder performs iterative solution on the index annotation set based on the Gauss-Newton method to realize the extraction of the differential features of the index annotation set, and obtains the differential feature set;

[0037] Among them, the differential feature set after iterative solution is expressed as:

[0038]

[0039] Among them, W k+1Denote the differential feature set after the (k + 1)-th iteration, ε is the total number of iterations based on the Gauss-Newton method, and α k Denote the iteration parameters at the k-th iteration, θ k is the gradient of the residual Denote the coefficient matrix of the normal equation of the differential feature set, W k , Are the input representations of the index annotation set and the variance of the index annotation set respectively;

[0040] Obtain the differential feature set, and the deformable convolutional neural network model assigns weights to the differential feature set based on the principal component analysis method to obtain the feature weight matrix of the index factors in the differential feature set and the index correlation matrix between the index factors;

[0041] Combine the feature weight matrix and index correlation matrix of the index factors to perform feature fusion and classification on the differential feature set, and calculate the fitting degree between the differential feature set and the evaluation classification level;

[0042] Load the differential feature set and the fitting degree between the differential feature set and the evaluation classification level, combine the real-time working parameters and bring them into the Topsis method to calculate the risk value of the unit equipment;

[0043] The fitting degree between the differential feature set and the evaluation classification level is calculated by the following formula:

[0044]

[0045] Where K(W k+1 , F x ) represents the fitting degree between the differential feature set and the evaluation classification level F x , B o , G o Represent the feature weight matrix of the index factor o and the index correlation matrix between the index factors o respectively, F represents the current evaluation classification level, and λ represents the membership degree of the index factor o to the evaluation classification level F x .

[0046] On the other hand, the present invention also provides a thermal power unit performance index auxiliary analysis system, and the thermal power unit performance index auxiliary analysis system includes:

[0047] A parameter acquisition module, which acquires the real-time working parameters of the thermal power unit based on a preset acquisition period, pre-identifies and extracts the real-time working parameters based on the parameter variation coefficient, and identifies the abnormal working data in the real-time working parameters;

[0048] An abnormal identification module, which feeds back the abnormal working data in the real-time working parameters, retrieves and uploads the historical acquisition parameters of the acquisition sensor that uploaded the abnormal working data, and integrates the historical acquisition parameters and the abnormal working data into an abnormal data set;

[0049] The device risk assessment module is used to pre - construct a performance index evaluation model, train the performance index evaluation model with modeling sample data, and obtain an abnormal data set. The performance index evaluation model extracts differential features from the abnormal data set to obtain a differential feature set. The performance index evaluation model weights the differential feature set based on the principal component analysis method, combines real - time working parameters and brings them into the Tops is method to calculate the risk value of the unit device;

[0050] The risk judgment module is used to obtain the risk value of the unit device, judge whether the device risk value meets the preset device safety threshold. If it does not meet the preset safety threshold, it triggers a device risk warning instruction.

[0051] Preferably, the parameter acquisition module includes:

[0052] The acquisition sensor is used to obtain real - time working parameters, perform dimensionality reduction transformation on the real - time working parameters, and transform the real - time working parameters into low - dimensional real - time working parameters;

[0053] The local density calculation unit is used to load the dimensionality - reduced real - time working parameters, generate parameter clustering clusters corresponding to the real - time working parameters using the fuzzy clustering algorithm, determine the centroid of the parameter clustering clusters, calculate the distance between the real - time working parameters and the centroid, generate a parameter distance set, sort the parameter distance set, traverse the parameter distance set, and calculate the local density of the parameter distance set;

[0054] The coefficient of variation determination unit is used to load the local density of the parameter distance set, determine the initial coefficient of variation of the parameter in combination with the analytic hierarchy process, and correct the initial coefficient of variation based on the index type weight and device weight corresponding to the real - time working parameters to determine the parameter coefficient of variation;

[0055] The pre - extraction unit pre - identifies and extracts the real - time working parameters based on the parameter coefficient of variation, and identifies abnormal working data in the real - time working parameters.

[0056] Compared with the prior art, the embodiments of the present application mainly have the following beneficial effects:

[0057] In the embodiments of the present invention, pre - identifying the real - time data based on the parameter coefficient of variation can quickly extract abnormal working data. At the same time, based on the abnormal working data, the key index factors affecting the performance of the thermal power unit can be determined more accurately. And by supervised learning to train the performance index evaluation model, the accuracy and generalization ability of the performance index evaluation model are improved. It overcomes the problem that when the prior art evaluates the performance of the thermal power unit, using static historical data and fixed thermodynamic indicators, although it can initially judge the weak points of energy exchange, it lacks the ability to deeply analyze real - time data and identify abnormal data, and is difficult to adapt to complex dynamic operating conditions.

[0058] In the embodiments of the present invention, based on the coefficient of variation of parameters, the real-time working parameters are pre-identified and extracted. By calculating the coefficient of variation of the real-time working parameters, the parameters with large fluctuations can be quickly identified, so as to screen out potential abnormal data. Moreover, by screening abnormal data through the coefficient of variation, noise and abnormal fluctuations can be effectively removed, and the reliability and accuracy of the data can be improved. In the present invention, the initial coefficient of variation is corrected based on the index type weight and equipment weight corresponding to the real-time working parameters to determine the final coefficient of variation of the parameters. Thus, the initial coefficient of variation can be dynamically adjusted according to different index type weights and equipment weights, enabling the real-time adjustment of the initial coefficient of variation to dynamically reflect the actual operating state of the equipment and avoiding the problem of misjudgment caused by the fixed initial coefficient of variation.

[0059] In the embodiments of the present invention, a performance index evaluation model for training modeling sample data is provided. The performance index evaluation model is based on the LSTM neural network model, and the Tops is algorithm is introduced in the output layer. Combining with a classifier, the equipment risk is evaluated. The Tops is algorithm can scientifically evaluate the equipment risk by calculating the distances between each feature and the ideal solution and the negative ideal solution. By combining with the U-shaped autoencoder and the conditional time diffusion network model, the model can dynamically adapt to the changes in the data, improve the adaptability to the complex industrial environment, and the performance index evaluation model realizes the efficient processing of abnormal data and the extraction of differential features through the collaborative combination of the U-shaped autoencoder and the conditional time diffusion network model. This method not only improves the quality of the features, but also further optimizes the feature extraction process through the finite element algorithm, improving the accuracy of the performance index evaluation model. Brief Description of the Drawings

[0060] Figure 1 It shows a schematic diagram of the implementation process of the performance index auxiliary analysis method for a thermal power unit.

[0061] Figure 2 It shows a schematic diagram of the implementation process of the method for pre-identifying and extracting real-time working parameters based on the coefficient of variation of parameters.

[0062] Figure 3 It shows a schematic diagram of the implementation process of the method for training a performance index evaluation model using modeling sample data.

[0063] Figure 4 It shows a schematic diagram of the method process for combining real-time working parameters and bringing them into the Tops is method to calculate the equipment risk value of the unit.

[0064] Figure 5 It shows a schematic diagram of the structure of the performance index auxiliary analysis system for a thermal power unit. Detailed Description of the Embodiments

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs; the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above drawings are intended to cover non-exclusive inclusion. The terms "first", "second", etc. in the specification and claims of this application or the above drawings are used to distinguish different objects and not to describe a specific order.

[0066] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in connection with the embodiments can be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0067] When evaluating the performance of a thermal power unit using existing methods, static historical data and fixed thermodynamic indicators are adopted. Although the weak points of energy exchange can be initially judged, there is a lack of in-depth analysis of real-time data and the ability to identify abnormal data. To address the above problems, we propose a system and method for assisting in analyzing the performance indicators of a thermal power unit. When implementing the method for assisting in analyzing the performance indicators of a thermal power unit, first, the real-time operating parameters of the thermal power unit are collected by sensors based on a preset collection period, and the real-time operating parameters are pre-identified and extracted based on the parameter coefficient of variation. At the same time, a performance indicator evaluation model is pre-constructed, and the performance indicator evaluation model is trained using the modeling sample data. Finally, the performance indicator evaluation model extracts the differential features from the abnormal data set to obtain a differential feature set. The performance indicator evaluation model assigns weights to the differential feature set based on the principal component analysis method, combines the real-time operating parameters, and brings them into the Tops is method to calculate the equipment risk value of the unit. In the embodiments of the present invention, pre-identifying the real-time data based on the parameter coefficient of variation can quickly extract abnormal operating data. At the same time, based on the abnormal operating data, the key indicator factors affecting the performance of the thermal power unit can be determined more accurately, and the performance indicator evaluation model is trained through supervised learning, improving the accuracy and generalization ability of the performance indicator evaluation model. It overcomes the deficiencies of existing methods in evaluating the performance of thermal power units, where static historical data and fixed thermodynamic indicators are used. Although the weak points of energy exchange can be initially judged, there is a lack of in-depth analysis of real-time data and the ability to identify abnormal data, making it difficult to adapt to complex dynamic operating conditions.

[0068] The embodiments of the present invention provide a method for assisting in analyzing the performance indicators of a thermal power unit. Figure 1The figure shows a schematic diagram of the implementation process of the auxiliary analysis method for the performance indicators of a combined heat and power unit. The auxiliary analysis method for the performance indicators of a combined heat and power unit specifically includes:

[0069] Step S10: The acquisition sensor collects the real-time operating parameters of the combined heat and power unit based on a preset acquisition period, pre-identifies and extracts the real-time operating parameters based on the parameter variation coefficient, and identifies the abnormal operating data in the real-time operating parameters;

[0070] It should be noted that in this embodiment, the combined heat and power unit includes unit equipment such as boilers, steam turbines, generators, condensers, feedwater systems, condensate systems, fuel supply systems, and heat exchange systems. The real-time operating parameters include, but are not limited to, boiler parameters (steam pressure, temperature, fuel flow, flue gas temperature), steam turbine parameters (speed, extraction steam flow), generator parameters (power generation, power generation load), heat supply parameters, heat network dynamic parameters, equipment temperature, equipment pressure, equipment vibration parameters, current and voltage of electrical equipment. The real-time operating parameters are used to characterize the following performance indicators of the combined heat and power unit: thermal efficiency, power generation efficiency, heat supply capacity, peak shaving capacity, and energy conversion efficiency.

[0071] Step S20: Feedback the abnormal operating data in the real-time operating parameters, retrieve and upload the historical acquisition parameters of the acquisition sensor that uploaded the abnormal operating data, and integrate the historical acquisition parameters and the abnormal operating data into an abnormal data set;

[0072] Step S30: Pre-construct a performance indicator evaluation model, pick up the modeling sample data based on web crawler technology, perform a correlation analysis on the modeling sample data, determine the indicator factors affecting the performance of the combined heat and power unit, and perform label processing on the indicator factors in the modeling sample data. Perform outlier and normalization processing on the modeling sample data, convert the modeling sample data into a data type for supervised learning, and use the modeling sample data to train the performance indicator evaluation model;

[0073] Step S40: Obtain the abnormal data set, the performance indicator evaluation model extracts the differential features from the abnormal data set to obtain a differential feature set. The performance indicator evaluation model weights the differential feature set based on the principal component analysis method, combines the real-time operating parameters, and brings them into the Tops is method to calculate the risk value of the unit equipment;

[0074] Step S50: Obtain the risk value of the unit equipment, and judge whether the equipment risk value meets the preset equipment safety threshold;

[0075] In this embodiment, the equipment safety threshold can be set to 0.2 - 0.25. When the equipment risk value exceeds the preset safety threshold, it indicates that the current operating risk of the mechanical and electrical equipment is relatively large. Therefore, it is necessary to trigger an equipment risk warning instruction.

[0076] Step S60: If it does not meet the preset safety threshold, trigger an equipment risk warning instruction;

[0077] If it meets the preset safety threshold, it is determined that the unit equipment is operating normally, and the equipment risk warning instruction is not triggered.

[0078] Step S70, when the equipment risk warning instruction is triggered, load the risk value of the unit equipment, and use the grey relational analysis method to calculate the associated risk value of the associated equipment associated with the unit equipment.

[0079] In this embodiment, when using the grey relational analysis method to calculate the associated risk value of the associated equipment associated with the unit equipment, the associated equipment associated with the unit equipment refers to the boiler system, steam turbine system, power generation system, heat network system, auxiliary system and other related equipment. When calculating the associated risk value of these equipment's operating states and performance indicators through the grey relational analysis method, it can comprehensively reflect their impact on the overall performance and safe operation of the unit.

[0080] In the embodiment of the present invention, pre-identifying the real-time data based on the parameter variation coefficient can quickly extract abnormal working data. At the same time, based on the abnormal working data, it can more accurately determine the key index factors affecting the performance of the thermoelectric unit, and train the performance index evaluation model through supervised learning, improving the accuracy and generalization ability of the performance index evaluation model. It overcomes the problem that when evaluating the performance of thermoelectric units by the existing methods, using static historical data and fixed thermodynamic indicators, although it can initially judge the weak points of energy exchange, it lacks the ability to deeply analyze real-time data and identify abnormal data, and is difficult to adapt to complex dynamic operating conditions.

[0081] The embodiment of the present invention provides a method for pre-identifying and extracting real-time working parameters based on the parameter variation coefficient. Figure 2 The schematic diagram of the implementation process of the method for pre-identifying and extracting real-time working parameters based on the parameter variation coefficient is shown. The method for pre-identifying and extracting real-time working parameters based on the parameter variation coefficient specifically includes:

[0082] Step S101, obtain the real-time working parameters, perform dimensionality reduction transformation on the real-time working parameters, and transform the real-time working parameters into low-dimensional real-time working parameters.

[0083] In this embodiment, when performing dimensionality reduction transformation on the real-time working parameters, the locally linear embedding algorithm can be used for dimensionality reduction processing. The locally linear embedding algorithm is a method based on manifold learning, which believes that high-dimensional data is actually on a low-dimensional manifold, and by maintaining the local neighborhood relationship of data points in the high-dimensional space and reconstructing these relationships in the low-dimensional space, dimensionality reduction is achieved.

[0084] Step S102, load the real-time working parameters after dimensionality reduction, use the fuzzy clustering algorithm to generate the parameter clustering clusters corresponding to the real-time working parameters, and determine the centroid of the parameter clustering clusters.

[0085] Step S103: Calculate the distance between the real-time working parameters and the centroid, generate a set of parameter distances, sort the set of parameter distances, traverse the set of parameter distances, and calculate the local density of the set of parameter distances.

[0086] Step S104: Load the local density of the set of parameter distances, and determine the initial coefficient of variation of the parameters by combining the analytic hierarchy process.

[0087] Step S105: Obtain the initial coefficient of variation of the parameters, and correct the initial coefficient of variation based on the index type weight and equipment weight corresponding to the real-time working parameters to determine the coefficient of variation of the parameters.

[0088] It should be noted that in this embodiment, the coefficient of variation of the parameters is used to measure the degree of abnormality of the parameter data. Among the real-time working parameters of the thermal power unit, the coefficient of variation of the parameters can be used to measure the stability of each parameter (such as temperature, pressure, flow rate, etc.). For example, a higher coefficient of variation may indicate that a certain parameter has large fluctuations or abnormalities. Therefore, a higher coefficient of variation has high reference and analysis value in the evaluation of the performance indicators of the thermal power unit.

[0089] Among them, when correcting the initial coefficient of variation based on the index type weight and equipment weight corresponding to the real-time working parameters to determine the coefficient of variation of the parameters, the coefficient of variation of the parameters is calculated by the following formula:

[0090]

[0091] where a c.v , respectively represent the coefficient of variation of the parameters and the initial coefficient of variation, W i represents the parameter judgment matrix constructed by the analytic hierarchy process based on the real-time working parameter X t X is the local density of the set of parameter distances, n is the number of the set of parameter distances, q dat i , q j respectively represent the index type weight and equipment weight corresponding to the real-time working parameters. In the embodiment of the present invention, the index type weight and equipment weight can be determined by the expert evaluation method, the analytic hierarchy process or the Delphi method. dist(i,z) represents the distance between the real-time working parameter and the centroid, μ z represents the centroid of the parameter clustering cluster, m represents the number of index types. In this embodiment, the number of index types can be 4 - 15 groups.

[0092] Step S106: Pre-identify and extract the real-time working parameters based on the coefficient of variation of the parameters, and identify the abnormal working data in the real-time working parameters.

[0093] ​In the embodiments of the present invention, real-time working parameters are pre-identified and extracted based on the coefficient of variation of parameters. By calculating the coefficient of variation of real-time working parameters, parameters with large fluctuations can be quickly identified, thereby screening out potential abnormal data. Moreover, by screening abnormal data through the coefficient of variation, noise and abnormal fluctuations can be effectively removed, improving the reliability and accuracy of the data. In the present invention, the initial coefficient of variation is corrected based on the index type weight and equipment weight corresponding to the real-time working parameters to determine the final coefficient of variation of the parameters. Thus, the initial coefficient of variation can be dynamically adjusted according to different index type weights and equipment weights, enabling the real-time adjustment of the initial coefficient of variation to dynamically reflect the actual operating state of the equipment and avoiding misjudgment problems caused by a fixed initial coefficient of variation.

[0094] The embodiments of the present invention provide a method for training a performance index evaluation model using modeling sample data. Figure 3 The schematic diagram of the implementation process of the method for training a performance index evaluation model using modeling sample data is shown. The method for training a performance index evaluation model using modeling sample data specifically includes:

[0095] Step S201: Use the LSTM neural network model as the initial model of the performance index evaluation model. The LSTM neural network model consists of a set of input layers, three sets of intermediate layers, and a set of output layers. The three sets of intermediate layers are the first intermediate layer, the second intermediate layer, and the third intermediate layer, and the input layer, intermediate layer, and output layer are connected in sequence.

[0096] In this embodiment, the number of neural network nodes in the first intermediate layer and the third intermediate layer can be 20 - 80. Moreover, the neural networks in the first intermediate layer and the third intermediate layer are composed of an input gate, a hidden gate, and an output gate. The input gate is used to extract features from the data after the abnormal data set is upscaled and reconstructed, and the feature extraction is realized through the Sigmoid activation function.

[0097] Step S202: Introduce a U-shaped autoencoder in the input layer. The U-shaped autoencoder is used to upscale and reconstruct the abnormal data set. The U-shaped autoencoder includes a downsampling layer, a max pooling layer, and an upsampling layer, and the downsampling layer, max pooling layer, and upsampling layer are connected in sequence.

[0098] In the embodiments of the present invention, introducing a U-shaped autoencoder to upscale and reconstruct the abnormal data set can effectively extract the key features in the data, while removing noise and abnormal fluctuations. This processing method not only improves the data quality but also enhances the robustness of the model to abnormal data, avoiding misjudgment caused by data quality problems.

[0099] Step S203: Freeze three sets of intermediate layers of the initial model, replace the second intermediate layer with the diffusion decoder of the conditional temporal diffusion network model, and introduce the finite element algorithm. Based on the diffusion decoder and in combination with the finite element algorithm, extract differential features from the abnormal data set.

[0100] Step S204: Introduce a deformable convolutional neural network model between the third intermediate layer and the output layer. The deformable convolutional neural network model consists of three sets of convolutional layers, three sets of residual connection layers, a deformable convolutional neural network, and a linear layer. Introduce the principal component analysis method in the linear layer, freeze the output layer of the performance metric evaluation model, and replace the output layer with a classifier. Introduce the Tops is algorithm into the classifier.

[0101] In this embodiment, the U-shaped autoencoder is trained using the mean squared error loss function, and the architecture of the classifier can be a Diff-PNA network. After the model is pre-constructed, use the ablation experiment method to remove a certain part or function of the performance metric evaluation model and observe its impact on the performance of the performance metric evaluation model. Specifically, the performance metric evaluation model focuses on three key modules: the U-shaped autoencoder, the diffusion decoder, and the classifier.

[0102] In the embodiment of the present invention, by combining the diffusion decoder of the conditional temporal diffusion network model with the finite element algorithm to extract differential features, the dynamic changes in the data can be accurately captured. And a deformable convolutional neural network model is introduced in the third intermediate layer, and in combination with the principal component analysis method (PCA) for dimensionality reduction processing, the feature extraction process is further optimized. This method not only reduces data redundancy but also improves the computational efficiency of the model.

[0103] The classifier consists of a set of global average pooling layers and three layers of multi-layer perceptrons. The risk classification result of the unit equipment is output through the multi-layer perceptrons, expressed as:

[0104]

[0105] where Y p represents the risk classification result output of the unit equipment, l is the number of output nodes p of the multi-layer perceptron network, is the feature of the p-th output node of the multi-layer perceptron network, and X avgpool represents the output feature of the global average pooling layer.

[0106] Step S205: Load the modeling sample data, divide the modeling sample data into a training set and a validation set according to a ratio of 3:1, set the number of iteration rounds of the performance metric evaluation model to 520, the training batch size per round to 25, and the learning rate to 0.01.

[0107] Step S206: Obtain a training set. The training set is used to train a performance metric evaluation model by means of a differential loss function and a metric optimization algorithm, and an converged performance metric evaluation model is output.

[0108] Step S207: Load a validation set, and verify the detection accuracy of the performance metric evaluation model based on the validation set.

[0109] In the embodiment of the present invention, the modeling sample data can be data obtained from the historical operation records of a thermal power unit, including the real-time operation parameters of the equipment (such as temperature, pressure, flow rate, etc.), the equipment status (such as normal operation, fault status, etc.), and the corresponding performance metrics. It can also be data generated by the Virtual Sample Generation (VSG) technology, which is used to expand the number of samples and solve the small sample problem.

[0110] Step S208: Determine whether the detection accuracy of the performance metric evaluation model for the risk assessment of unit equipment meets a preset accuracy threshold. In this embodiment, the accuracy threshold can be set to 0.9 - 0.95.

[0111] Step S209: If it meets the preset accuracy threshold, output the converged performance metric evaluation model.

[0112] If it does not meet the preset accuracy threshold, return to Step S206 to continue iterative training of the model.

[0113] In the embodiment of the present invention, a modeling sample data is provided to train a performance metric evaluation model. The performance metric evaluation model is based on an LSTM neural network model, and the Topsis algorithm is introduced in the output layer. Combining with a classifier to evaluate the equipment risk, the Topsis algorithm can scientifically evaluate the equipment risk by calculating the distances between each feature and the ideal solution and the negative ideal solution. By combining with a U-shaped autoencoder and a conditional time diffusion network model, the model can dynamically adapt to data changes, improve the adaptability to complex industrial environments, and the performance metric evaluation model realizes the efficient processing of abnormal data and the extraction of differential features through the collaborative combination of the U-shaped autoencoder and the conditional time diffusion network model. The method not only improves the quality of features, but also further optimizes the feature extraction process through the finite element algorithm, improving the accuracy of the performance metric evaluation model.

[0114] The embodiment of the present invention provides a method for combining real-time working parameters and substituting them into the Topsis method to calculate the risk value of unit equipment. Figure 4 The flow chart of the method for combining real-time working parameters and substituting them into the Topsis method to calculate the risk value of unit equipment is shown. The method for combining real-time working parameters and substituting them into the Topsis method to calculate the risk value of unit equipment specifically includes:

[0115] Step S301, load the abnormal data set, and the U-shaped autoencoder performs upscaling and reconstruction processing on the abnormal data set, and outputs an abnormal index matrix;

[0116] Step S302, load the abnormal index matrix, and the diffusion decoder combines the finite element algorithm to perform Fourier frequency analysis on the abnormal index matrix, screen out the index factors exceeding the preset frequency threshold based on the preset frequency threshold, and perform index factor annotation processing on the modeling sample data to obtain an index annotation set;

[0117] Step S303, the diffusion decoder performs iterative solution on the index annotation set based on the Gauss-Newton method to realize the extraction of the differential characteristics of the index annotation set and obtain a differential characteristic set;

[0118] Among them, the differential characteristic set after iterative solution is expressed as:

[0119]

[0120] Among them, W k+1 represents the differential characteristic set after k + 1 iterations, ε is the total number of iterations based on the Gauss-Newton method, which can be 5-10, α k represents the iteration parameter at the kth iteration, θ k is the gradient of the residual, represents the coefficient matrix of the normal equation of the differential characteristic set, W k , are the input representations of the index annotation set and the variance of the index annotation set respectively;

[0121] Step S304, obtain the differential characteristic set, and the deformable convolutional neural network model weights the differential characteristic set based on the principal component analysis method to obtain the characteristic weight matrix of the index factors in the differential characteristic set and the index correlation matrix between the index factors;

[0122] Step S305, combine the characteristic weight matrix and the index correlation matrix of the index factors to perform feature fusion and classification on the differential characteristic set, and calculate the fitting degree between the differential characteristic set and the evaluation classification level;

[0123] Step S306, load the differential characteristic set and the fitting degree between the differential characteristic set and the evaluation classification level, combine the real-time working parameters and substitute them into the Tops is method to calculate the risk value of the unit equipment.

[0124] In this embodiment, the fitting degree between the differential characteristic set and the evaluation classification level is calculated by the following formula:

[0125]

[0126] Among them, K(W k+1 , F x ) represents the fitting degree between the differential characteristic set and the evaluation classification level Fx The fitting degree of B o , G o respectively represent the characteristic weight matrix of the index factor o and the index correlation matrix between the index factors o. F represents the current evaluation classification level, and the current evaluation classification level can be 1-8. The λ index factor o represents the membership degree of the evaluation classification level F x .

[0127] In this embodiment, when combining real-time working parameters and bringing them into the Tops is method to calculate the risk value of the unit equipment, the Topsis method can calculate the distances between each differential feature set and the ideal solution and the negative ideal solution. The ideal solution and the negative ideal solution are determined by the fitting degree between the differential feature set and the evaluation classification level. Among them, the ideal solution refers to the maximum value in the fitting degree of the evaluation classification level, and the negative ideal solution refers to the minimum value in the fitting degree of the evaluation classification level. For each index factor in the differential feature set, the distances between it and the ideal solution and the negative ideal solution are calculated respectively to determine the risk value of the unit equipment. The commonly used method for calculating the distances between it and the ideal solution and the negative ideal solution to determine the risk value of the unit equipment is the Euclidean distance, which can scientifically evaluate the equipment risk and accurately rank the risk values. This method not only considers the positive and negative impacts of each characteristic factor, but also optimizes the characteristic weights through the entropy weight method, further improving the accuracy of the risk assessment. It helps to give priority early warning for high-risk equipment.

[0128] On the other hand, the embodiment of the present invention also provides a thermal power unit performance index auxiliary analysis system Figure 5 shows a structural schematic diagram of the thermal power unit performance index auxiliary analysis system. The thermal power unit performance index auxiliary analysis system specifically includes

[0129] A parameter acquisition module 100, which acquires the real-time working parameters of the thermal power unit based on a preset acquisition period, pre-identifies and extracts the real-time working parameters based on the parameter variation coefficient, and identifies the abnormal working data in the real-time working parameters

[0130] An abnormal identification module 200, which feeds back the abnormal working data in the real-time working parameters, retrieves and uploads the historical acquisition parameters of the acquisition sensor that uploaded the abnormal working data, and integrates the historical acquisition parameters and the abnormal working data into an abnormal data set

[0131] A device risk assessment module 300, which is used to pre-construct a performance index evaluation model, train the performance index evaluation model with modeling sample data, and obtain an abnormal data set. The performance index evaluation model extracts differential features from the abnormal data set to obtain a differential feature set. The performance index evaluation model assigns weights to the differential feature set based on the principal component analysis method, combines the real-time working parameters and brings them into the Tops is method to calculate the risk value of the unit equipment

[0132] A risk judgment module 400 is configured to obtain the risk value of the unit equipment, determine whether the equipment risk value meets a preset equipment safety threshold, and if it meets the preset safety threshold, trigger an equipment risk warning instruction.

[0133] It should be noted that it can be understood that the thermal power unit performance index auxiliary analysis system provided by the embodiments of the present invention corresponds to the above-mentioned thermal power unit performance index auxiliary analysis method. For the explanations, examples, beneficial effects, etc. of the relevant content, reference can be made to the corresponding content in the thermal power unit performance index auxiliary analysis method, which will not be elaborated here.

[0134] In this embodiment, the parameter acquisition module 100 includes:

[0135] An acquisition sensor 110 is configured to obtain real-time working parameters, perform dimensionality reduction transformation on the real-time working parameters, and transform the real-time working parameters into low-dimensional real-time working parameters;

[0136] A local density calculation unit 120 is configured to load the real-time working parameters after dimensionality reduction, generate a parameter clustering cluster corresponding to the real-time working parameters by using a fuzzy clustering algorithm, determine the centroid of the parameter clustering cluster, calculate the distance between the real-time working parameters and the centroid, generate a parameter distance set, sort the parameter distance set, traverse the parameter distance set, and calculate the local density of the parameter distance set;

[0137] A coefficient of variation determination unit 130 is configured to load the local density of the parameter distance set, determine the initial coefficient of variation of the parameter in combination with the analytic hierarchy process, and correct the initial coefficient of variation based on the index type weight and equipment weight corresponding to the real-time working parameters to determine the parameter coefficient of variation;

[0138] A pre-extraction unit 140 pre-identifies and extracts the real-time working parameters based on the parameter coefficient of variation, and identifies the abnormal working data in the real-time working parameters.

[0139] In this embodiment, the acquisition sensor 110 can be a temperature sensor, a humidity sensor, a smoke sensor, a pressure sensor, a vibration sensor, an acceleration sensor, and data interaction and transmission between the acquisition sensor 110, the local density calculation unit 120, the coefficient of variation determination unit 130, and the pre-extraction unit 140 can be realized in a way of electrical sequential connection, parallel connection, or feedback connection.

[0140] In summary, the present invention provides a system and method for auxiliary analysis of the performance indicators of a thermoelectric unit. In the embodiments of the present invention, pre-identification of real-time data is performed based on the parameter coefficient of variation, which can quickly extract abnormal working data. At the same time, based on the abnormal working data, the key index factors affecting the performance of the thermoelectric unit can be determined more accurately, and a performance index evaluation model is trained through supervised learning, improving the accuracy and generalization ability of the performance index evaluation model. It overcomes the deficiencies of existing methods in evaluating the performance of thermoelectric units. When using static historical data and fixed thermodynamic indicators, although the weak points of energy exchange can be initially judged, there is a lack of in-depth analysis of real-time data and the ability to identify abnormal data, making it difficult to adapt to complex dynamic operating conditions.

[0141] It should be noted that for the foregoing embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps may be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0142] In several embodiments provided by the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the above division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection between devices or units can be in the form of telecommunications or other forms.

[0143] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the protection scope of the invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on these embodiments, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art can still, without conflict, make combinations, additions, deletions, or other adjustments to the features in the embodiments of the present invention according to the circumstances without creative efforts, so as to obtain different technical solutions that are essentially not divorced from the concept of the present invention. These technical solutions also belong to the scope of protection of the present invention.

Claims

1. A method for auxiliary analysis of thermal power unit performance indicators, characterized in that: The auxiliary analysis method for the thermal power unit performance index comprises: The acquisition sensor acquires the real-time working parameters of the thermal power unit based on a preset acquisition cycle, pre-identifies and extracts the real-time working parameters based on the parameter variation coefficient, and identifies abnormal working data in the real-time working parameters; Feedback abnormal working data in real-time working parameters, retrieve historical acquisition parameters of acquisition sensors that upload the abnormal working data, and integrate historical acquisition parameters and abnormal working data into an abnormal data set; Pre-build the performance index evaluation model, pick up the modeling sample data based on the crawler technology, perform correlation analysis on the modeling sample data, determine the index factors that affect the performance of the thermal power unit, and perform index factor labeling on the modeling sample data. Perform outlier and normalization processing on the modeling sample data, convert the modeling sample data into a data type for supervised learning, and use the modeling sample data to train the performance index evaluation model; Obtain the abnormal data set, and the performance indicator evaluation model extracts the differential features of the abnormal data set to obtain the differential feature set. The performance indicator evaluation model assigns weights to the differential feature set based on the principal component analysis method, combines the real-time working parameters and brings them into the Topsis method to calculate the risk value of the equipment group; Obtain the equipment risk value of the unit and determine whether the equipment risk value meets the preset equipment safety threshold. If it does not meet the preset safety threshold, trigger the equipment risk warning instruction.

2. The auxiliary analysis method for thermal power unit performance index according to claim 1, characterized in that: The auxiliary analysis method for the thermal power unit performance index also includes: The risk value of the unit equipment is loaded, and the grey correlation analysis method is used to calculate the associated risk value of the associated equipment associated with the unit equipment.

3. The auxiliary analysis method for thermal power unit performance index according to claim 1, characterized in that: The method for pre-identifying and extracting real-time working parameters based on parameter variation coefficient specifically includes: Acquire real-time working parameters, perform dimensionality reduction transformation on the real-time working parameters, and transform the real-time working parameters into low-dimensional real-time working parameters; Load the real-time working parameters after dimension reduction, generate parameter clusters corresponding to the real-time working parameters using a fuzzy clustering algorithm, and determine the centroid of the parameter clusters; Calculate the distance between the real-time working parameter and the centroid, generate a parameter distance set, sort the parameter distance set, traverse the parameter distance set, and calculate the local density of the parameter distance set; The local density of the parameter distance set is loaded, and the initial coefficient of variation of the parameter is determined by combining the analytic hierarchy process; Obtain the initial coefficient of variation of the parameter, and modify the initial coefficient of variation based on the indicator type weight and equipment weight corresponding to the real-time working parameter to determine the coefficient of variation of the parameter; The real-time working parameters are pre-identified and extracted based on the parameter variation coefficient to identify abnormal working data in the real-time working parameters.

4. The auxiliary analysis method for thermal power unit performance index according to claim 3, characterized in that: When the initial coefficient of variation is corrected based on the indicator type weight and the equipment weight corresponding to the real-time working parameter and the parameter coefficient of variation is determined, the parameter coefficient of variation is calculated by the following formula: Among them, a c.v , Respectively represent the parameter variation coefficient, the initial variation coefficient, W i Represents the analytic hierarchy process based on real-time working parameters X t The constructed parameter judgment matrix, X dat is the local density of the parameter distance set, n is the number of parameter distance sets, q i ,q j They represent the indicator type weight and device weight corresponding to the real-time working parameters, dist(i,z) represents the distance between the real-time working parameters and the centroid, μ z represents the centroid of the parameter clustering cluster, and m represents the number of indicator types.

5. The auxiliary analysis method for thermal power unit performance index according to claim 1, characterized in that: The method of using modeling sample data to train a performance indicator evaluation model specifically includes: The LSTM neural network model is used as the initial model of the performance indicator evaluation model. The LSTM neural network model consists of a group of input layers, three groups of intermediate layers and a group of output layers. The three groups of intermediate layers are the first intermediate layer, the second intermediate layer, and the third intermediate layer. The input layer, the intermediate layer, and the output layer are connected in sequence. A U-shaped autoencoder is introduced into the input layer. The U-shaped autoencoder is used to upgrade and reconstruct the abnormal data set. The U-shaped autoencoder includes a downsampling layer, a maximum pooling layer, and an upsampling layer. The downsampling layer, the maximum pooling layer, and the upsampling layer are connected in sequence. The three groups of middle layers of the initial model are frozen, the second middle layer is replaced by the diffusion decoder of the conditional time diffusion network model, and the finite element algorithm is introduced to extract differential features of the abnormal data set based on the diffusion decoder combined with the finite element algorithm; A deformable convolutional neural network model is introduced between the third intermediate layer and the output layer. The deformable convolutional neural network model consists of three groups of convolutional layers, three groups of residual connection layers, a deformable convolutional neural network, and a linear layer. The principal component analysis method is introduced in the linear layer, the output layer of the performance index evaluation model is frozen, and the output layer is replaced by a classifier. The Topsis algorithm is introduced in the classifier. The classifier consists of a set of global average pooling layers and three layers of multilayer perceptrons. The multilayer perceptron is used to output the risk classification results of the unit equipment, which can be expressed as: Among them, Y p represents the risk classification result output of the unit equipment, l is the number of output nodes p of the multilayer perceptron network, is the feature of the p-th output node of the multilayer perceptron network, X avgpool Represents the output features of the global average pooling layer.

6. The auxiliary analysis method for thermal power unit performance index according to claim 5, characterized in that: The method of using modeling sample data to train a performance indicator evaluation model specifically includes: Load the modeling sample data, divide the modeling sample data into training set and validation set according to the distribution ratio of 3:1, set the iteration rounds of the performance indicator evaluation model to 520, the single round training batch to 25, and the learning rate to 0.01; Obtain a training set, use a differential loss function and an indicator optimization algorithm to train a performance indicator evaluation model, and output a converged performance indicator evaluation model; Load the validation set, verify the detection accuracy of the performance indicator evaluation model based on the validation set, and determine whether the detection accuracy of the performance indicator evaluation model for unit equipment risk assessment meets the preset accuracy threshold. If it meets the preset accuracy threshold, output the converged performance indicator evaluation model.

7. The auxiliary analysis method for thermal power unit performance index according to claim 6, characterized in that: The method of combining real-time working parameters and introducing the Topsis method to calculate the risk value of group equipment specifically includes: Load the abnormal data set, and the U-shaped autoencoder upgrades and reconstructs the abnormal data set, and outputs the abnormal indicator matrix; Load the abnormal indicator matrix, and use the diffusion decoder combined with the finite element algorithm to perform Fourier frequency analysis on the abnormal indicator matrix. Based on the preset frequency threshold, the indicator factors that exceed the preset frequency threshold are screened, and the modeling sample data is annotated with the indicator factors to obtain the indicator annotation set. The diffusion decoder iteratively solves the indicator annotation set based on the Gauss-Newton method to extract the differential features of the indicator annotation set and obtain the differential feature set; Among them, the differential feature set after iterative solution is expressed as: Among them, W k+1 represents the differential feature set after iteration k+1 times, ε is the total number of iterations based on the Gauss-Newton method, and α k represents the iteration parameter when iterating k times, θ k is the gradient of the residual, The coefficient matrix of the normal equation of the difference feature set, W k , are the input representations of the indicator annotation set and the variance of the indicator annotation set respectively; The differential feature set is obtained, and the deformable convolutional neural network model weights the differential feature set based on the principal component analysis method to obtain the feature weight matrix of the indicator factors in the differential feature set and the indicator correlation matrix between the indicator factors; Combine the feature weight matrix of the indicator factors and the indicator correlation matrix to perform feature fusion and classification on the differential feature set, and calculate the fit between the differential feature set and the evaluation classification level; Load the differential feature set, the fit between the differential feature set and the assessment classification level, combine the real-time working parameters and bring in the Topsis method to calculate the risk value of the group equipment.

8. The auxiliary analysis method for thermal power unit performance index according to claim 7, characterized in that: The degree of fit between the differential feature set and the evaluation classification level is calculated by the following formula: Among them, K(W k+1 ,F x ) represents the difference feature set and the evaluation classification level F x The fit, B o ,G o They represent the feature weight matrix of indicator factor o and the indicator association matrix between indicator factors o, F represents the current evaluation classification level, and λ indicator factor o is related to the evaluation classification level F x The degree of membership.

9. A thermal power unit performance index auxiliary analysis system, used to implement the thermal power unit performance index auxiliary analysis method as claimed in any one of claims 1 to 8, characterized in that: The thermal power unit performance index auxiliary analysis system comprises: The parameter acquisition module acquires the real-time working parameters of the thermal power unit based on a preset acquisition cycle, pre-identifies and extracts the real-time working parameters based on the parameter variation coefficient, and identifies abnormal working data in the real-time working parameters; The abnormality identification module feeds back abnormal working data in the real-time working parameters, retrieves the historical acquisition parameters of the acquisition sensor that uploaded the abnormal working data, and integrates the historical acquisition parameters and the abnormal working data into an abnormal data set; Equipment risk assessment module, used to pre-build a performance indicator assessment model, use modeling sample data to train the performance indicator assessment model, and obtain an abnormal data set. The performance indicator assessment model extracts differential features from the abnormal data set to obtain a differential feature set. The performance indicator assessment model weights the differential feature set based on the principal component analysis method, combines real-time working parameters and brings in the Topsis method to calculate the risk value of the computer group equipment. The risk judgment module is used to obtain the risk value of the unit equipment and judge whether the equipment risk value meets the preset equipment safety threshold. If it does not meet the preset safety threshold, the equipment risk warning instruction is triggered.

10. The auxiliary analysis system for thermal power unit performance indicators according to claim 9, characterized in that: The parameter acquisition module comprises: The acquisition sensor is used to obtain the real-time working parameters, perform dimensionality reduction transformation on the real-time working parameters, and transform the real-time working parameters into low-dimensional real-time working parameters; A local density calculation unit is used to load the real-time working parameters after dimension reduction, generate parameter clusters corresponding to the real-time working parameters using a fuzzy clustering algorithm, determine the centroid of the parameter clusters, calculate the distance between the real-time working parameters and the centroid, generate a parameter distance set, sort the parameter distance set, traverse the parameter distance set, and calculate the local density of the parameter distance set; The coefficient of variation determination unit is used to load the local density of the parameter distance set, determine the initial coefficient of variation of the parameter in combination with the hierarchical analysis method, correct the initial coefficient of variation based on the indicator type weight and equipment weight corresponding to the real-time working parameter, and determine the coefficient of variation of the parameter; The pre-extraction unit pre-identifies and extracts the real-time working parameters based on the parameter variation coefficient, and identifies abnormal working data in the real-time working parameters.

Images