Thermal power station water supply system performance evaluation method based on LSTM neural network

Abstract

The invention relates to a thermal power station water supply system performance evaluation method based on an LSTM (Long Short Term Memory) neural network, which comprises the following steps of: establishing a key parameter prediction model of related equipment of a water supply system based on the LSTM neural network by selecting related parameters influencing intermediate variation of the related equipment of the water supply system as characteristic quantities for representing the health degree of the equipment; and taking the real-time operation data as an input variable of the LSTM neural network prediction model to predict a water supply system related equipment parameter prediction value, and fusing similarity values of a plurality of parameters according to a difference similarity judgment function between the prediction value and an actual value to obtain a comprehensive health degree evaluation index. All-direction monitoring of related equipment of the thermal power station water supply system is achieved, and the prediction result precision is high.

Description

Technical Field

[0001] The invention relates to a performance evaluation method for a water supply system of a thermal power plant, and in particular to a performance evaluation method for a water supply system of a thermal power plant based on an LSTM neural network. Background Art

[0002] With the proposal of the "dual carbon" goal and the development of clean energy, thermal power units need to frequently carry out deep peak regulation under the premise of ensuring safety and environmental protection. The water supply control system, as a very important subsystem in the direct current boiler control system, is related to the safe and stable operation of the entire unit. At the same time, it plays an important role in improving the efficiency of the unit and reducing equipment loss. Therefore, it is crucial to monitor the operating status of the water supply system and early detection and discovery of faults under deep peak regulation conditions. Most of the existing neural network models select existing measurement points of the unit such as fixed temperature, pressure, and current, and cannot accurately reflect the changes in intermediate variables in deep regulation conditions. In addition, the input data and output data in traditional machine learning algorithms are independent of each other, and there is no strong connection between the previous and next data. It is difficult to solve the sequence problem with a strong relationship between the previous and next data. Therefore, according to the real-time value trend of the monitored physical parameters, it is predicted whether the equipment will fail within a certain period of time in the future and what kind of failure will occur, so as to fundamentally eliminate the occurrence of failures. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the purpose of the present invention is to provide a performance evaluation method for the water supply system of a thermal power plant based on an LSTM neural network, which can effectively solve the problem of performance evaluation of the water supply system of a thermal power plant based on an LSTM neural network.

[0004] The technical solution provided by the present invention is:

[0005] A performance evaluation method for a thermal power plant water supply system based on an LSTM neural network comprises the following steps:

[0006] Step 1: Obtain historical data

[0007] Collect historical data such as load, pressure, valve position, valve opening, temperature, and flow stored in the DCS system of thermal power units during real-time operation of the units;

[0008] The loads include feedwater pump load, No. 1 high-pressure heater load, No. 2 high-pressure heater load, and No. 3 high-pressure heater load;

[0009] The pressure includes the feedwater pump inlet pressure, the first stage steam extraction pressure of the No. 1 high pressure heater, the second stage steam extraction pressure of the No. 2 high pressure heater, and the third stage steam extraction pressure of the No. 3 high pressure heater;

[0010] The valve position includes the valve position of the small water supply pump;

[0011] The valve opening includes the opening of the No. 1 high pressure heater drain valve of the No. 1 high pressure heater, the opening of the No. 2 high pressure heater drain valve of the No. 2 high pressure heater, and the opening of the No. 3 high pressure heater drain valve of the No. 3 high pressure heater;

[0012] The temperatures include the feedwater pump outlet temperature, the first stage extraction temperature of the No. 1 high pressure heater, the second stage extraction temperature of the No. 2 high pressure heater, the No. 2 high pressure heater outlet water temperature, the third stage extraction temperature of the No. 3 high pressure heater, the No. 3 high pressure heater outlet water temperature and the No. 4 high pressure heater outlet water temperature, and the intermediate point temperature of the power plant;

[0013] The flow rate includes the feed water flow rate of the No. 1 high-pressure heater, the feed water flow rate of the No. 2 high-pressure heater, and the feed water flow rate of the No. 3 high-pressure heater;

[0014] Step 2: Classify historical data according to unit load conditions

[0015] (1) Shutdown condition: P<0.1P n ;

[0016] (2) Deep adjustment condition: 0.1P n ≤P<0.3P n ;

[0017] (3) Low load condition: 0.3P n ≤P<0.5P n ;

[0018] (4) Medium load condition: 0.5P n ≤P<0.8P n ,;

[0019] (5) High load condition: 0.8P n ≤P<P n ;

[0020] Among them, P is the current load of the thermal power unit, P n is the maximum capacity of the unit;

[0021] Step 3: Determine system input / output variables

[0022] The water supply system selects different neural network model parameters according to different working conditions. Before establishing the neural network model, the input / output variables of the model are determined first.

[0023] The input / output variables are:

[0024] a) Feedwater pump:

[0025] Neural network input points: load, feedwater pump inlet pressure, feedwater pump inlet temperature, small machine valve position;

[0026] Neural network output points: feed water pump flow, feed water pump outlet temperature, feed water pump outlet temperature, feed water pump speed;

[0027] b) No. 1 high pressure heater:

[0028] Neural network input points: load, first stage extraction temperature, first stage extraction pressure, No. 2 high pressure heater outlet water temperature, No. 1 high pressure heater drain valve opening, feed water flow rate;

[0029] Neural network output points: No. 1 high pressure heater drain temperature, No. 1 high pressure heater outlet water temperature;

[0030] c) No. 2 high pressure heater:

[0031] Neural network input points: load, second stage extraction steam temperature, second stage extraction steam pressure, No. 3 high pressure heater outlet water temperature, No. 2 high pressure heater drain valve opening, feed water flow;

[0032] Neural network output points: No. 2 high pressure heater drain temperature, No. 2 high pressure heater outlet water temperature;

[0033] d) No. 3 high pressure heater:

[0034] Neural network input points: load, three-stage extraction steam temperature, three-stage extraction steam pressure, No. 4 high-pressure heater outlet water temperature, No. 3 high-pressure heater drain valve opening, feed water flow;

[0035] Neural network output points: No. 3 high pressure heater drain temperature, No. 3 high pressure heater outlet water temperature;

[0036] Under shutdown conditions, the neural network model does not perform data reading training, and under deep adjustment conditions and low load conditions, the intermediate point temperature of the power plant is added as the input point of the neural network model;

[0037] Step 4: Training data acquisition and preprocessing

[0038] a) The frequency of data collection is 1 second per record, and the collection range is data within 6 months before the current time;

[0039] b) Use normality test analysis method to detect outliers, and then perform linear regression on the outliers;

[0040] c) Replace samples with data deviation greater than 10% by using an approximate function fitted to normal data;

[0041] d) Delete the data items with multiple missing items in the data set, and fill the data items with less than 60 seconds in the data set by cubic spline interpolation;

[0042] Step 5: Training of LSTM neural network model

[0043] a) Divide the input and output columns in the training set into x and y, and convert the input and output columns into arrays of corresponding dimensions according to the input points and output points of the corresponding device neural network model;

[0044] b) Initialize the LSTM neural network model and start training, set the number of neuron cores, set the format of input data during training, adjust the number of hidden layers, number of nodes, weights, learning rate, activation function and other parameters of the deep neural network model according to the number of system input / output variables and system complexity, and perform repeated training until the accuracy of the deep neural network prediction model meets the requirements;

[0045] Step 6: Prediction of relevant output parameters of water supply system

[0046] The deep neural network prediction model obtained after training is used to take the real-time value of the system input variable as the input data of the LSTM neural network prediction model. The LSTM neural network prediction model is used to make real-time predictions on the system output value. After the predicted value is obtained, it is inversely scaled and inversely differentiated to restore it to the original value range.

[0047] Step 7: Estimation of water supply system equipment health

[0048] For the comparison of a single parameter, the similarity calculation method based on distance and trend is adopted, as follows:

[0049] Define the distance calculation formula based on normal distribution:

[0050]

[0051] Where: x i is the predicted value of the LSTM network model, x0 is the actual value of the current unit, and σ is the standard deviation.

[0052] Define the Pearson correlation coefficient for calculating linear similarity of vectors:

[0053]

[0054] Where: X is the predicted value of the LSTM network model, Y is the actual value of the current unit, σ is the standard deviation, and cov is the covariance;

[0055] Combining distance similarity and trend similarity, we can get the similarity judgment function H(x i ,x j ):

[0056]

[0057] Where: xi，xj is the Pearson correlation coefficient between the predicted value and the actual value, g(x i, x j ) is the distance between the predicted value and the actual value based on normal distribution;

[0058] The similarity values ​​of multiple parameters are combined to obtain a comprehensive health evaluation index to evaluate the performance of the water supply system of the thermal power plant.

[0059] There is a strong causal relationship between the input and output of the coal-fired power unit system equipment under deep adjustment conditions, and the single parameter early warning mode cannot reflect the abnormal operation of the equipment in time. The LSTM neural network will use the output of the previous moment as the input of this moment, that is, the output of this moment is closely related to the output of all previous moments, and realize the "memory" function of previous data in the process of continuous weight parameter update, and realize the all-round monitoring of thermal power plant equipment through continuous update. The present invention selects key variables of water supply system related equipment through expert experience, calculates the correlation coefficient between the output of the neural network model and the actual value through the LSTM model, and evaluates the health of the equipment based on this, and realizes the health diagnosis of the equipment. Applying LSTM to the performance evaluation of the water supply system of thermal power units is also one of the main innovations of the present invention. This method solves the problem that the traditional neural network model cannot monitor the early evolution of faults. The variables that characterize the water supply operation status of thermal power plants have strong coupling and complex and changeable nonlinear characteristics. Often, the change of one variable triggers the change of another variable. The single parameter early warning mode cannot reflect the abnormal operation of the equipment in time. The present invention selects the intermediate variables and related parameters that affect the water supply system related equipment as the characteristic quantities to characterize the health of the equipment, establishes a prediction model for the key parameters of the water supply system related equipment based on the LSTM neural network, uses the real-time operation data as the input variable of the LSTM neural network prediction model to predict the predicted values ​​of the parameters of the water supply system related equipment, and combines the similarity values ​​of multiple parameters according to the difference similarity judgment function between the predicted value and the actual value to obtain a comprehensive health evaluation index. The comprehensive monitoring of the water supply system related equipment of the thermal power station is realized, and the prediction result is highly accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is the data preprocessing process of the present invention.

[0061] Figure 2 This is a flow chart of the LSTM prediction model of the present invention.

[0062] Figure 3 The figure is a flow chart of the water supply system health assessment process of the present invention.

[0063] Figure 4 The figure is a flow chart of the overall evaluation of the present invention. DETAILED DESCRIPTION

[0064] The specific implementation modes of the present invention are further described in detail below in conjunction with the embodiments.

[0065] A method for evaluating the performance of a water supply system of a thermal power plant based on an LSTM neural network, characterized in that it comprises the following steps:

[0066] Step 1: Obtain historical data

[0067] Collect historical data such as load, pressure, valve position, valve opening, temperature, and flow stored in the DCS system of thermal power units during real-time operation of the units;

[0068] The loads include feedwater pump load, No. 1 high-pressure heater load, No. 2 high-pressure heater load, and No. 3 high-pressure heater load;

[0069] The pressure includes the feedwater pump inlet pressure, the first stage steam extraction pressure of the No. 1 high pressure heater, the second stage steam extraction pressure of the No. 2 high pressure heater, and the third stage steam extraction pressure of the No. 3 high pressure heater;

[0070] The valve position includes the valve position of the small water supply pump;

[0071] The valve opening includes the opening of the No. 1 high pressure heater drain valve of the No. 1 high pressure heater, the opening of the No. 2 high pressure heater drain valve of the No. 2 high pressure heater, and the opening of the No. 3 high pressure heater drain valve of the No. 3 high pressure heater;

[0072] The temperatures include the feedwater pump outlet temperature, the first stage extraction temperature of the No. 1 high pressure heater, the second stage extraction temperature of the No. 2 high pressure heater, the No. 2 high pressure heater outlet water temperature, the third stage extraction temperature of the No. 3 high pressure heater, the No. 3 high pressure heater outlet water temperature and the No. 4 high pressure heater outlet water temperature, and the intermediate point temperature of the power plant;

[0073] The flow rate includes the feed water flow rate of the No. 1 high-pressure heater, the feed water flow rate of the No. 2 high-pressure heater, and the feed water flow rate of the No. 3 high-pressure heater;

[0074] Historical data must be able to cover the full operating range of the unit to ensure data traversability;

[0075] Step 2: Classify historical data according to unit load conditions

[0076] (1) Shutdown condition: P<0.1P n ;

[0077] (2) Deep adjustment condition: 0.1P n ≤P<0.3P n ;

[0078] (3) Low load condition: 0.3P n ≤P<0.5P n ;

[0079] (4) Medium load condition: 0.5Pn ≤P<0.8P n ;

[0080] (5) High load condition: 0.8P n ≤P<P n ;

[0081] Among them, P is the current load of the thermal power unit, P n is the maximum capacity of the unit;

[0082] Step 3: Determine system input / output variables

[0083] The water supply system selects different neural network model parameters according to different working conditions. Before establishing the neural network model, the input / output variables of the model are determined first.

[0084] The principal component analysis (PCA) dimensionality reduction method is used to reduce the data dimension, and representative typical types of data are selected as input / output variables of the neural network prediction model;

[0085] The input / output variables are:

[0086] a) Feedwater pump:

[0087] Neural network input points: load, feedwater pump inlet pressure, feedwater pump inlet temperature, small machine valve position;

[0088] Neural network output points: feed water pump flow, feed water pump outlet temperature, feed water pump outlet temperature, feed water pump speed;

[0089] b) No. 1 high pressure heater:

[0090] Neural network input points: load, first stage extraction temperature, first stage extraction pressure, No. 2 high pressure heater outlet water temperature, No. 1 high pressure heater drain valve opening, feed water flow rate;

[0091] Neural network output points: No. 1 high pressure heater drain temperature, No. 1 high pressure heater outlet water temperature;

[0092] c) No. 2 high pressure heater:

[0093] Neural network input points: load, second stage extraction steam temperature, second stage extraction steam pressure, No. 3 high pressure heater outlet water temperature, No. 2 high pressure heater drain valve opening, feed water flow;

[0094] Neural network output points: No. 2 high pressure heater drain temperature, No. 2 high pressure heater outlet water temperature;

[0095] d) No. 3 high pressure heater:

[0096] Neural network input points: load, three-stage extraction steam temperature, three-stage extraction steam pressure, No. 4 high-pressure heater outlet water temperature, No. 3 high-pressure heater drain valve opening, feed water flow;

[0097] Neural network output points: No. 3 high pressure heater drain temperature, No. 3 high pressure heater outlet water temperature;

[0098] Under shutdown conditions, the neural network model does not perform data reading training, and under deep adjustment conditions and low load conditions, the intermediate point temperature of the power plant is added as the input point of the neural network model;

[0099] Step 4: Training data acquisition and preprocessing

[0100] a) The frequency of data collection is 1 second per record, and the collection range is the data within 6 months before the current time. In this embodiment, there are 7776000 data records in total;

[0101] b) Use normality test analysis method to detect outliers, and then perform linear regression on the outliers;

[0102] c) Replace samples with data deviation greater than 10% by using an approximate function fitted to normal data;

[0103] d) Delete the data items with multiple missing items in the data set, and fill the data items with less than 60 seconds in the data set by cubic spline interpolation;

[0104] Step 5: Training of LSTM neural network model

[0105] a) Divide the input and output columns in the training set into x and y, and convert the input and output columns into arrays of corresponding dimensions according to the input points and output points of the corresponding device neural network model;

[0106] b) Initialize the LSTM neural network model and start training, set the number of neuron cores, set the format of input data during training, adjust the number of hidden layers, number of nodes, weights, learning rate, activation function and other parameters of the deep neural network model according to the number of system input / output variables and system complexity, and perform repeated training until the accuracy of the deep neural network prediction model training data is greater than 95%;

[0107] Step 6: Prediction of relevant output parameters of water supply system

[0108] The deep neural network prediction model obtained after training is used to take the real-time value of the system input variable as the input data of the LSTM neural network prediction model. The LSTM neural network prediction model is used to make real-time predictions on the system output value. After the predicted value is obtained, it is inversely scaled and inversely differentiated to restore it to the original value range.

[0109] Step 7: Estimation of water supply system equipment health

[0110] For the comparison of a single parameter, the similarity calculation method based on distance and trend is adopted, as follows:

[0111] Define the distance calculation formula based on normal distribution:

[0112]

[0113] Where: x i is the predicted value of the LSTM network model, x0 is the actual value of the current unit, and σ is the standard deviation.

[0114] Define the Pearson correlation coefficient for calculating linear similarity of vectors:

[0115]

[0116] Where: X is the predicted value of the LSTM network model, Y is the actual value of the current unit, σ is the standard deviation, and cov is the covariance;

[0117] Combining distance similarity and trend similarity, we can get the similarity judgment function H(x i ,x j ):

[0118]

[0119] Where: xi，xj is the Pearson correlation coefficient between the predicted value and the actual value, g(x i , x j ) is the distance between the predicted value and the actual value based on normal distribution;

[0120] The similarity values ​​of multiple parameters are combined to obtain a comprehensive health evaluation index to evaluate the performance of the water supply system of the thermal power plant.

[0121] The method for fusing the similarity values ​​of multiple parameters to obtain the health evaluation index adopts a probability fusion method based on DS evidence theory, and the synthesis rule is:

[0122]

[0123] in, is the conflict factor. The larger the k value, the greater the evidence conflict. The final output diagnosis result is m(A) is the support level for the occurrence of proposition A, m1 and m2 are two basic probability distributions on the identification framework;

[0124] The equipment condition is divided according to the percentage system. When the health evaluation index is 80-100 points, it is good, and the operator does not need to frequently check the equipment operation. When the health evaluation index is 60-80 points, it is passed. The operator needs to check the parameter change trend in time to prevent the rapid deterioration of the parameters from causing damage to the equipment. When the health evaluation index is below 60 points, it is unsuccessful. It is recommended to shut down the equipment in time and repair the equipment.

[0125] The method of the present invention is compared with the existing BP neural network and SVM methods, and the results are as follows:

[0126] method error Accuracy Advance warning time / min BP Neural Network 0.15 0.75 15 SVM 0.18 0.70 10 The present invention 0.1 0.85 30

[0127] The superior performance of the method in this paper is mainly attributed to its deep learning framework, which can learn and extract complex time dynamic features from a large amount of historical data, combine expert experience with LSTM neural network model, and consider the time dynamics of equipment to achieve more accurate performance evaluation prediction. As can be seen from the above table, in contrast, due to the lack of a design specifically for processing such data, BP neural network and SVM cannot fully utilize the information of time series data. Although the BP neural network model is a deep learning model, it mainly relies on the linear autoregression of the data and cannot capture complex nonlinear dynamics well. Although SVM can handle nonlinear relationships, it does not consider the sequential information of the time series. In addition, the experimental results show that the early warning time of the present invention is significantly better than other methods. This means that after using the present invention, the operation and maintenance personnel have more time to prepare and respond to possible equipment failures, greatly improving the operating efficiency and safety of the power plant. In general, the present invention has the characteristics of high accuracy, low error, and early warning time.

Claims

1. A performance evaluation method for a thermal power plant water supply system based on an LSTM neural network, characterized in that: The following steps are involved: Step 1: Obtain historical data Collect historical data such as load, pressure, valve position, valve opening, temperature, and flow stored in the DCS system of thermal power units during real-time operation of the units; The loads include feedwater pump load, No. 1 high-pressure heater load, No. 2 high-pressure heater load, and No. 3 high-pressure heater load; The pressure includes the feedwater pump inlet pressure, the first stage steam extraction pressure of the No. 1 high pressure heater, the second stage steam extraction pressure of the No. 2 high pressure heater, and the third stage steam extraction pressure of the No. 3 high pressure heater; The valve position includes the valve position of the small water supply pump; The valve opening includes the opening of the No. 1 high pressure heater drain valve of the No. 1 high pressure heater, the opening of the No. 2 high pressure heater drain valve of the No. 2 high pressure heater, and the opening of the No. 3 high pressure heater drain valve of the No. 3 high pressure heater; The temperatures include the feedwater pump outlet temperature, the first stage extraction temperature of the No. 1 high pressure heater, the second stage extraction temperature of the No. 2 high pressure heater, the No. 2 high pressure heater outlet water temperature, the third stage extraction temperature of the No. 3 high pressure heater, the No. 3 high pressure heater outlet water temperature and the No. 4 high pressure heater outlet water temperature, and the intermediate point temperature of the power plant; The flow rate includes the feed water flow rate of the No. 1 high-pressure heater, the feed water flow rate of the No. 2 high-pressure heater, and the feed water flow rate of the No. 3 high-pressure heater; Step 2: Classify historical data according to unit load conditions (1) Shutdown condition: P<0.1P n ; (2) Deep adjustment condition: 0.1P n ≤P<0.3P n ; (3) Low load condition: 0.3P n ≤P<0.5P n ; (4) Medium load condition: 0.5P n ≤P<0.8P n ; (5) High load condition: 0.8P n ≤P <P n ; Among them, P is the current load of the thermal power unit, P n is the maximum capacity of the unit; Step 3: Determine system input / output variables The water supply system selects different neural network model parameters according to different working conditions. Before establishing the neural network model, the input / output variables of the model are determined first. The input / output variables are: a) Feedwater pump: Neural network input points: load, feedwater pump inlet pressure, feedwater pump inlet temperature, small machine valve position; Neural network output points: feed water pump flow, feed water pump outlet temperature, feed water pump outlet temperature, feed water pump speed; b) No. 1 high pressure heater: Neural network input points: load, first stage extraction temperature, first stage extraction pressure, No. 2 high pressure heater outlet water temperature, No. 1 high pressure heater drain valve opening, feed water flow rate; Neural network output points: No. 1 high pressure heater drain temperature, No. 1 high pressure heater outlet water temperature; c) No. 2 high pressure heater: Neural network input points: load, second stage extraction steam temperature, second stage extraction steam pressure, No. 3 high pressure heater outlet water temperature, No. 2 high pressure heater drain valve opening, feed water flow; Neural network output points: No. 2 high pressure heater drain temperature, No. 2 high pressure heater outlet water temperature; d) No. 3 high pressure heater: Neural network input points: load, three-stage extraction steam temperature, three-stage extraction steam pressure, No. 4 high-pressure heater outlet water temperature, No. 3 high-pressure heater drain valve opening, feed water flow; Neural network output points: No. 3 high pressure heater drain temperature, No. 3 high pressure heater outlet water temperature; Under shutdown conditions, the neural network model does not perform data reading training, and under deep adjustment conditions and low load conditions, the intermediate point temperature of the power plant is added as the input point of the neural network model; Step 4: Training data acquisition and preprocessing a) The frequency of data collection is 1 second per record, and the collection range is data within 6 months before the current time; b) Use normality test analysis method to detect outliers, and then perform linear regression on the outliers; c) Replace samples with data deviation greater than 10% by using an approximate function fitted to normal data; d) Delete the data items with multiple missing items in the data set, and fill the data items with less than 60 seconds in the data set by cubic spline interpolation; Step 5: Training of LSTM neural network model a) Divide the input and output columns in the training set into x and y, and convert the input and output columns into arrays of corresponding dimensions according to the input points and output points of the corresponding device neural network model; b) Initialize the LSTM neural network model and start training, set the number of neuron cores, set the format of input data during training, adjust the number of hidden layers, number of nodes, weights, learning rate, activation function and other parameters of the deep neural network model according to the number of system input / output variables and system complexity, and perform repeated training until the accuracy of the deep neural network prediction model meets the requirements; Step 6: Prediction of relevant output parameters of water supply system The deep neural network prediction model obtained after training is used to take the real-time value of the system input variable as the input data of the LSTM neural network prediction model. The LSTM neural network prediction model is used to make real-time predictions on the system output value. After the predicted value is obtained, it is inversely scaled and inversely differentiated to restore it to the original value range. Step 7: Estimation of water supply system equipment health For the comparison of a single parameter, the similarity calculation method based on distance and trend is adopted, as follows: Define the distance calculation formula based on normal distribution: Where: x i is the predicted value of the LSTM network model, x0 is the actual value of the current unit, and σ is the standard deviation. Define the Pearson correlation coefficient for calculating linear similarity of vectors: Where: X is the predicted value of the LSTM network model, Y is the actual value of the current unit, σ is the standard deviation, and cov is the covariance; Combining distance similarity and trend similarity, we can get the similarity judgment function H(x i ,x j ): Where: xi,xj is the Pearson correlation coefficient between the predicted value and the actual value, g(x i ,x j ) is the distance between the predicted value and the actual value based on normal distribution; The similarity values ​​of multiple parameters are combined to obtain a comprehensive health evaluation index to evaluate the performance of the water supply system of the thermal power plant.

2. The method for evaluating the performance of a thermal power plant water supply system based on an LSTM neural network according to claim 1, characterized in that: The method of fusing the similarity values ​​of multiple parameters to obtain the health evaluation index in step 7 adopts a probabilistic fusion method based on DS evidence theory, and the synthesis rule is: in, is the conflict factor. The larger the k value, the greater the evidence conflict. The final output diagnosis result is m(A) is the support level for the occurrence of proposition A, m1 and m2 are two basic probability distributions on the identification framework; The equipment condition is divided according to the percentage system. When the health evaluation index is 80-100 points, it is good, and the operator does not need to frequently check the equipment operation. When the health evaluation index is 60-80 points, it is passed. The operator needs to check the parameter change trend in time to prevent the rapid deterioration of the parameters from causing damage to the equipment. When the health evaluation index is below 60 points, it is unsuccessful. It is recommended to shut down the equipment in time and repair the equipment.

Images