Online Prediction Method and System for Coal Ash Generation Quantity of Power Plant Boilers Based on LSTM

Abstract

Online prediction method and system for coal ash generation amount of power station boilers based on LSTM. The method includes collecting historical data of coal ash generation amount of boilers and main influencing parameters to obtain correlation relation data and performing normalization processing; dividing the normalized correlation relation data into a training set and a test set; using the data in the training set to train and fit a pre-established LSTM network model for coal ash generation amount of boilers; combining the prediction data with the data in the test set to perform overfitting evaluation on the trained and fitted LSTM network model for coal ash generation amount of boilers; online sampling currently occurred data and inputting it into the LSTM network model for coal ash generation amount of boilers without overfitting evaluation to predict the coal ash generation amount value at future times. The present invention can record required data for a long time and can perform online prediction, with a long prediction period and high accuracy, meeting the requirements of power generation enterprises for dynamic prediction of coal ash generation amount of boilers.

Description

Technical Field

[0001] The present invention belongs to the technical field of solid waste monitoring, and particularly relates to an online prediction method and system for the coal ash generation amount of a power plant boiler based on LSTM. Background Art

[0002] When a thermal power plant uses coal as fuel, a large amount of coal ash will inevitably be generated after coal combustion in the boiler. Part of the coal ash (fly ash) in the flue gas is collected through dust removal for comprehensive utilization, and the other part is discharged into the atmosphere through the chimney along with the flue gas. With the improvement of national environmental protection requirements, the emission standards for the coal ash generated by boilers are undoubtedly becoming more and more stringent, which requires power generation enterprises to accurately control the operation of the coal ash emissions of boilers. The traditional coal ash generation amount of boilers is given at the design stage based on the maximum / minimum operating conditions of the boiler. During the actual operation of the power plant, due to the lack of a direct online coal ash amount sensor, the generated coal ash amount is mainly estimated by manual experience. In recent years, with the rapid development of new energy, in order to accommodate more intermittent and fluctuating wind power and photovoltaic power, thermal power production must participate in the load peak regulation of the power grid, and the daily power generation load changes accordingly. In addition, due to the imbalance between coal supply and demand and the high operation of coal prices, power plants burn a large amount of low-quality coal, and the coal quality is unstable, resulting in a large change in the coal ash generation amount of boilers, which brings many problems to the safe and economic operation of downstream dust collectors, pipeline transportation, and ash storage silos.

[0003] To calculate the boiler ash and slag amount conventionally, multiple parameters need to be involved. For example, except for the coal quality ash content, the carbon content in the ash is either ignored or compensated and calculated using the mechanical incomplete combustion heat loss Q4 coefficient (see the Technical Code for Ash Handling Design of Thermal Power Plants DL / T 5142-2012 for details). The existence of carbon content in coal ash increases the coal ash emissions and the greater the energy loss. The former is incomplete and has a large error; the latter formula requires matching different boiler forms and fuel types because the Q4 coefficient is an empirical value, and the calculation is cumbersome, complex, and inconvenient in practical applications.

[0004] Due to various factors such as different boiler forms, operating environments (combustion mode, flue gas velocity, furnace heat load, boiler operating load, etc.) and coal quality differences, the generated coal ash shows complex change characteristics, making it difficult to have a generally applicable online calculation model for the coal ash amount and difficult to accurately predict it. Summary of the Invention

[0005] The purpose of the present invention is to provide an online prediction method and system for the coal ash generation amount of a power plant boiler based on LSTM in view of the above problems in the prior art, which can predict the coal ash generation amount in the next time period according to the coal ash generation amount and main influencing parameter data in the previous time period, and improve the accuracy and practicability of the coal ash generation amount prediction.

[0006] To achieve the above object, the present invention has the following technical solutions:

[0007] An online prediction method for the coal ash generation amount of a power plant boiler based on LSTM, comprising:

[0008] Collect historical data of the coal ash generation amount of the boiler and the main influencing parameters to obtain correlation relationship data and perform normalization processing;

[0009] Divide the normalized correlation relationship data into a training set and a test set;

[0010] Use the data in the training set to train and fit a pre-established LSTM network model for the coal ash generation amount of the boiler;

[0011] Combine the prediction data with the data in the test set to evaluate the overfitting of the trained and fitted LSTM network model for the coal ash generation amount of the boiler;

[0012] Online sample the currently occurring data and input it into the LSTM network model for the coal ash generation amount of the boiler that is evaluated to have no overfitting to predict the coal ash generation amount value at a future time.

[0013] As a preferred scheme of the online prediction method for the coal ash generation amount of a power plant boiler based on LSTM of the present invention, the main influencing parameters include the unit's electric / thermal load, boiler type, coal consumption, coal quality of the coal, and carbon content of the coal ash.

[0014] As a preferred scheme of the online prediction method for the coal ash generation amount of a power plant boiler based on LSTM of the present invention, the correlation relationship between the historical data of the coal ash generation amount of the boiler and the main influencing parameters conforms to the following expression:

[0015]

[0016] In the formula: G h is the coal ash generation amount, with the unit of t / h;

[0017] G m is the actual coal consumption, with the unit of t / h;

[0018] A ar is the ash content of the coal as received;

[0019] C ar is the carbon content of the coal ash;

[0020] is the percentage of fly ash generated by the boiler in the ash and slag amount;

[0021] If the air-dried basis data is obtained, perform an online basis conversion according to the following formula:

[0022]

[0023] Where: A ad is the ash content of coal on an air-dried basis;

[0024] M ar is the moisture content of coal as-received;

[0025] M ad is the moisture content of coal on an air-dried basis.

[0026] As a preferred embodiment of the online prediction method for the coal ash generation amount of a power plant boiler based on LSTM according to the present invention, the step of collecting historical data of the coal ash generation amount of the boiler and the main influencing parameters to obtain correlation relationship data and performing normalization processing includes arranging the correlation relationship data according to time to obtain a sequence sample data set, and the normalization processing expression is as follows:

[0027] Xi = (Xi - Xmin) / (Xmax - Xmin)

[0028] Where: Xi represents the normalized value of the i-th value in the sequence sample data set;

[0029] Xi represents the i-th value in the sequence sample data set;

[0030] Xmin represents the minimum value in the sequence sample data set;

[0031] Xmax represents the maximum value in the sequence sample data set.

[0032] As a preferred embodiment of the online prediction method for the coal ash generation amount of a power plant boiler based on LSTM according to the present invention, in the step of dividing into a training set and a test set, after normalizing the data in the sequence sample data set, it is divided into a training set and a test set according to a time interval of 2:1, and the sampling period of each data set can represent the characteristic change samples of the same time period.

[0033] As a preferred embodiment of the online prediction method for the coal ash generation amount of a power plant boiler based on LSTM according to the present invention, the LSTM network model for the coal ash generation amount of the boiler includes a data sampling layer, a data preprocessing layer, an input layer, an LSTM layer, a Dropout layer, a fully connected layer, a SoftMax layer, and a classification output layer arranged in sequence; the data sampling layer is used to collect the coal ash generation amount of the boiler and the main influencing parameters, the data preprocessing layer is used to perform normalization processing on the data collected by the data sampling layer, the input layer is used to perform data diversity and format conversion, the LSTM layer is used to perform LSTM network training, the Dropout layer is used to supervise network overfitting, and the fully connected layer, the SoftMax layer, and the classification output layer are jointly used to output and optimize the results.

[0034] As a preferred embodiment of the online prediction method for the ash generation amount of power plant boilers based on LSTM according to the present invention, the establishment process of the LSTM network model for the ash generation amount of the boiler includes: inputting data of the main influencing parameters and adjusting the internal parameters of the network; the data of the main influencing parameters include the unit's electrical / thermal load, coal consumption, coal quality of the coal, and carbon content in the coal ash, and the step of adjusting the internal parameters of the network includes adjusting the learning rate and the number of iterations of the established LSTM network model for the ash generation amount of the boiler according to set values, obtaining the input weight W, the recurrent weight R, and the bias b, and respectively adjusting the hyperparameters of the corresponding input gate i, forget gate f, candidate gate g, and output gate o by using the gate activation function and the state activation function;

[0035] The internal learning weights of the LSTM network are:

[0036] W = [W i , W f , W g , W o T

[0037] R = [R i , R f , R g , R o T

[0038] b = [b i , b f , b g , b o T

[0039] The process of one iteration of the LSTM network model for the ash generation amount of the boiler is:

[0040]

[0041] In the formula, σ g is the gate activation function, and the sigmoid function is used, then σ(X) = (1 + e -x ) -1 ; σ c is the state activation function, and the tanh function is used, C t , h t are the output values of the network unit after one calculation.

[0042] As a preferred embodiment of the online prediction method for the ash generation amount of power plant boilers based on LSTM according to the present invention, the LSTM network model for the ash generation amount of the boiler is trained and fitted using the mean absolute error loss function and the stochastic gradient descent Adam algorithm, the learning rates of the various parameters are dynamically modified, and the momentum method is introduced during the training and fitting process until the network converges.​​​

[0043] As a preferred embodiment of the online prediction method for the coal ash generation amount of power station boilers based on LSTM according to the present invention, the step of combining the prediction data with the data in the test set and evaluating the overfitting of the LSTM network model for the coal ash generation amount of the boiler after training and fitting includes inputting the data in the test set into the LSTM network model for the coal ash generation amount of the boiler for prediction, displaying the model loss of the training set and the model loss of the test set in one graph, and determining whether there is an overfitting phenomenon in the model;

[0044] The step of determining whether there is an overfitting phenomenon in the model includes:

[0045] Calculating the error score of the model through the initial prediction value and the actual value, using the root mean square error of the unit error identical to the variable; outputting the losses of training and testing at the end of each training epoch, and finally outputting the final root mean square error of the model for the test set data, and determining whether there is an overfitting phenomenon in the model according to the magnitude of the final root mean square error.

[0046] An online prediction system for the coal ash generation amount of power station boilers based on LSTM, comprising:

[0047] An association relationship data acquisition and processing module, configured to acquire historical data of the coal ash generation amount of the boiler and main influencing parameters to obtain association relationship data and perform normalization processing;

[0048] A data set construction module, configured to divide the normalized association relationship data into a training set and a test set;

[0049] A model training and fitting module, configured to use the data in the training set to train and fit a pre-established LSTM network model for the coal ash generation amount of the boiler;

[0050] A model overfitting evaluation module, configured to combine the prediction data with the data in the test set and evaluate the overfitting of the LSTM network model for the coal ash generation amount of the boiler after training and fitting;

[0051] A sampling prediction module, configured to online sample the currently occurred data and input it into the LSTM network model for the coal ash generation amount of the boiler evaluated to have no overfitting, and predict the value of the coal ash generation amount of the boiler at a future time.

[0052] Compared with the prior art, the present invention has at least the following beneficial effects:

[0053] The online prediction method for the coal ash generation amount of a power plant boiler based on LSTM predicts the coal ash generation amount in the next time period according to the coal ash generation amount and the data of main influencing parameters in the previous time period. First, historical data is collected. According to the time series curve of the unit's electricity / heat load and the coal consumption characteristics, the correlation relationship between the electricity / heat load and the coal ash generation amount of the boiler is established, arranged according to time, and normalized. To facilitate model training, the prepared data is divided into a training set and a test set according to a ratio. The training set data is input into the pre-established LSTM network model for the coal ash generation amount for model training to complete the training and fitting of the LSTM network model. The test set data is then input into the fitted LSTM network model. By comparing and displaying the model training and test loss data, it is judged whether there is an overfitting phenomenon in the model. After the model evaluation and verification, the currently generated data is sampled online and input into the verified LSTM network model to predict the value of the coal ash generation amount of the boiler in the future. By introducing the time series prediction LSTM model of multiple main influencing parameters, the present invention improves the accuracy and practicability of the coal ash generation amount prediction. Power generation enterprises can carry out the optimization operation of ash removal equipment according to the online calculated and predicted coal ash generation amount, which can not only reduce the production loss of equipment, but also has important significance for the safe storage of the planned ash silo capacity in advance and the comprehensive utilization in the later stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a time series diagram of the correlation between the electricity / heat load of a generator unit and the coal ash generation amount of the boiler on a certain day for establishing the LSTM network model;

[0055] Figure 2 is a schematic flow chart of the online prediction method for the coal ash generation amount of a power plant boiler based on LSTM of the present invention;

[0056] Figure 3 is a schematic structural diagram of the LSTM network model for the coal ash generation amount of the boiler provided by the present invention;

[0057] Figure 4 is a flow chart for adjusting the internal parameters of the LSTM network model for the coal ash generation amount of the boiler provided by the present invention;

[0058] Figure 5 is a graph of the overfitting evaluation result of the LSTM network model for the coal ash generation amount of the boiler of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The present invention will be further described in detail below with reference to the drawings and embodiments.

[0060] The present invention proposes an online prediction method for the ash generation amount of power plant boilers based on LSTM. The LSTM neural network is suitable for processing data related to time series and can reflect the main influencing parameters of the ash generation amount of boilers, thereby making the prediction of the ash generation amount of boilers more accurate. The prediction method for the ash generation amount of boilers based on LSTM deep learning can record the required data for a long time and can perform online prediction, with a long prediction period and high accuracy, meeting the dynamic prediction requirements of power generation enterprises for the ash generation amount of boilers.

[0061] Please refer to Figure 2 , an online prediction method for the ash generation amount of power plant boilers based on LSTM, comprising the following steps:

[0062] Step 1: Collect historical data on the ash generation amount of boilers and the main influencing parameters, establish the correlation between the electrical / thermal load and the ash generation amount of boilers, arrange them in time to obtain a sequence sample data set, and then normalize the data so that its value is between [0, 1]. This step prepares the data set for the LSTM model. Collect data related to the ash generation amount of boilers from the historical data of the enterprise production monitoring system, including the ash generation amount data of boilers and the main influencing parameter data. The main influencing parameter data includes: unit electrical / thermal load, coal consumption, coal quality of the coal burned, and carbon content in the coal ash, etc.

[0063] First of all, it is necessary to collect historical data on the ash generation amount of boilers and the main influencing parameters. These data are generally obtained from the enterprise production monitoring system (DCS / SIS / ERP) through API / SDK interfaces. For example Figure 1 shown is the sequence sample set of the ash generation amount of a certain generator set in a day. Select the electrical / thermal load of the generator set, which is the main influencing parameter of the ash generation amount of boilers, as the associated data. The basis for data calculation is Formula 1, that is, the calculation basis for the ash generation amount:

[0064]

[0065] In the formula: G h is the ash generation amount, with the unit of t / h;

[0066] G m is the actual coal consumption, with the unit of t / h;

[0067] A ar is the as-received ash content of the coal;

[0068] C ar is the carbon content in the coal ash;

[0069] is the percentage of fly ash generated by the boiler in the ash and slag amount;

[0070] Among them, the as-received ash content of coal and the carbon content in coal ash can be obtained from the digital coal yard. If the air-dried basis (analysis basis) data is obtained, the online basis conversion needs to be completed according to formula (2):

[0071]

[0072] In the formula: A ad is the air-dried basis ash of coal;

[0073] M ar is the as-received moisture of coal;

[0074] M ad is the air-dried basis moisture of coal.

[0075] Then, normalize this data to obtain the original acquisition sequence:

[0076] St = {Xt(1), Xt(2), Xt(3),... Xt(u)}, Xt ∈ {U, I},

[0077] The formula for normalizing the original acquisition data is as follows:

[0078] Xi = (Xi - Xmin) / (Xmax - Xmin) (3)

[0079] In the formula: Xi represents the normalized value of the i-th value in the sequence sample dataset;

[0080] Xi represents the i-th value in the sequence sample dataset;

[0081] Xmin represents the minimum value in the sequence sample dataset;

[0082] Xmax represents the maximum value in the sequence sample dataset.

[0083] Step 2: Divide the normalized correlation relationship data into a training set and a test set. For the convenience of model training, use the data of the recent 60 days to fit the model, and then use the data of the recent 30 days for evaluation, that is, the data scale of the training set and the test set is 2:1, and ensure that the sampling period of each dataset can represent the characteristic change samples of the same time period.

[0084] Step 3: Define and establish the LSTM network model for the coal ash generation amount of the boiler, and adjust the internal parameters of the network. Establish Figure 3The shown LSTM network model for the boiler coal ash generation amount, the LSTM model includes: a data sampling layer a, a data preprocessing layer b, an input layer c, an LSTM layer d, a Dropout layer e, a fully connected layer f, a SoftMax layer g, and a classification output layer h. Among them: layer a is used to collect the boiler coal ash generation amount and the main influencing parameters, layer b is used to normalize the data of layer a, layer c is used for data diversity and format conversion, layer d is used for LSTM network training, layer e is used to supervise network overfitting, and finally layers f, g, and h are used to output and optimize the results. After the collected data is subjected to format conversion through the input layer, it is transmitted to the LSTM layer for LSTM network training. After passing through the fully connected layer and the SoftMax layer in sequence, the classification result is output by the classification output layer. The model inputs multiple main influencing parameter data, including boiler type, coal consumption, coal quality of the coal, and carbon content in the coal ash, etc. Adjusting the internal parameters of the network means adjusting the learning rate and the number of iterations according to the set values, obtaining the input weight, the recurrent weight, and the bias, and respectively using the gate activation function and the state activation function to adjust the parameters of the corresponding input gate, forget gate, candidate gate, and output gate.

[0085] The established model adjusts the learning rate and the number of iterations according to the set values, obtains the input weight W, the recurrent weight R, and the bias b, and respectively uses the gate activation function and the state activation function to adjust the hyperparameters of the corresponding input gate i, forget gate f, candidate gate g, and output gate o, specifically as Figure 4 shown. The internal learning weights of the LSTM network are:

[0086] W = [W i , W f , W g , W o T

[0087] R = [R i , R f , R g , R o T

[0088] b = [b i , b f , b g , b o T

[0089] The process of one iteration of the LSTM network model for the boiler coal ash generation amount is:

[0090]

[0091] Among them, σ g is the gate activation function. Using the sigmoid function, then σ(X) = (1 + e -x )​​​-1 ; σ c is the state activation function, using the tanh function, C t , h t are the output values of the network unit after one calculation.

[0092] Step 4: Model training and fitting. Input the training set data obtained in Step 2 into the boiler coal ash generation LSTM network model for training until the network converges. During this process, use the mean absolute error (MAE) loss function and the Adam algorithm of efficient stochastic gradient descent. On the one hand, dynamically modify the learning rate of each parameter, and on the other hand, introduce the momentum method, so that the parameter update has more opportunities to jump out of the local optimum, accelerating and optimizing the network convergence.

[0093] Step 5: Model overfitting evaluation.

[0094] After fitting, use the model to predict the test set data obtained in Step 2, input it into the fitted LSTM model, display the model loss of the training set and the model loss of the test set in one graph, and judge whether the model has overfitting phenomenon.

[0095] Through the initial prediction value and the actual value, calculate the error score of the model using the root mean square error (RMSE) of the unit error of the same variable as the variable. Output the losses of training and testing at the end of each training epoch. Finally, output the final RMSE of the model for the test data set. As Figure 5 shown, it can be seen that the model has obtained good RMSE test results.

[0096] Step 6: Model prediction.

[0097] After the model evaluation and verification, input the currently generated data sampled online into the verified LSTM neural network to predict the boiler coal ash generation value in the future time, and set the accuracy, balance score and accuracy rate to evaluate the results.

[0098] Another embodiment of the present invention also proposes an online prediction system for boiler coal ash generation based on LSTM, including:

[0099] The correlation relationship data acquisition and processing module is used to collect the historical data of the boiler coal ash generation amount and the main influencing parameters to obtain the correlation relationship data and perform normalization processing;

[0100] The data set construction module is used to divide the normalized correlation relationship data into a training set and a test set;

[0101] The model training and fitting module is used to train and fit the pre-established boiler coal ash generation LSTM network model using the data in the training set;

[0102] The model overfitting evaluation module is used to combine the prediction data with the data in the test set to evaluate the overfitting of the LSTM network model for the boiler coal ash generation amount after training and fitting.

[0103] The sampling prediction module is used to online sample the currently occurred data and input it into the LSTM network model for the boiler coal ash generation amount that is evaluated to have no overfitting, so as to predict the boiler coal ash generation amount value in the future.

[0104] It should be noted that for the information interaction, execution process, etc. between the above-mentioned module units, since they are based on the same concept as the method embodiment, for their specific functions and the technical effects brought, reference can be made to the method embodiment part, and details will not be elaborated here.

[0105] Those skilled in the art can clearly understand that for the convenience and conciseness of description, only the above-mentioned division of each functional unit and module is used as an example. In actual application, the above-mentioned functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of the present application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the foregoing method embodiment, and details will not be elaborated here.

[0106] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above-described embodiment methods of this application, a computer program can be used to instruct relevant hardware to complete. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the photographing device / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk, or an optical disc, etc.

[0107] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0108] The above-described embodiments are only used to illustrate the technical solutions of this application, rather than to limit it; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. An online prediction method for the ash generation amount of power plant boilers based on LSTM, characterized in that, Including: Collecting historical data of boiler coal ash generation amount and main influencing parameters to obtain correlation relationship data and performing normalization processing; the main influencing parameters include unit electric / thermal load, boiler type, coal consumption, coal quality of the coal, and carbon content in the coal ash. Dividing the normalized correlation relationship data into a training set and a test set. Using the data in the training set to train and fit a pre-established LSTM network model for boiler coal ash generation amount; the LSTM network model for boiler coal ash generation amount includes a data sampling layer, a data preprocessing layer, an input layer, an LSTM layer, a Dropout layer, a fully connected layer, a SoftMax layer, and a classification output layer arranged in sequence; the data sampling layer is used to collect the boiler coal ash generation amount and main influencing parameters, the data preprocessing layer is used to perform normalization processing on the data collected by the data sampling layer, the input layer is used to perform data diversity and format conversion, the LSTM layer is used to perform LSTM network training, the Dropout layer is used to supervise network overfitting, and the fully connected layer, the SoftMax layer, and the classification output layer are jointly used to output and optimize the results. Combining the prediction data with the data in the test set to perform overfitting evaluation on the trained and fitted LSTM network model for boiler coal ash generation amount. Online sampling the currently occurring data and inputting it into the LSTM network model for boiler coal ash generation amount that is evaluated to have no overfitting to predict the boiler coal ash generation amount value in the future time. The step of combining the prediction data with the data in the test set to perform overfitting evaluation on the trained and fitted LSTM network model for boiler coal ash generation amount includes inputting the data in the test set into the LSTM network model for boiler coal ash generation amount for prediction, displaying the model loss of the training set and the model loss of the test set in one graph, and determining whether the model has an overfitting phenomenon. The step of determining whether the model has an overfitting phenomenon includes: Calculating the error score of the model by using the initial prediction value and the actual value and adopting the root mean square error of the unit error same as the variable. Outputting the losses of training and testing at the end of each training epoch, and finally outputting the final root mean square error of the model for the test set data, and determining whether the model has an overfitting phenomenon according to the magnitude of the final root mean square error.

2. The online prediction method for the ash generation amount of a power plant boiler based on LSTM according to claim 1, characterized in that The correlation relationship between the historical data of the boiler coal ash generation amount and the main influencing parameters conforms to the following expression: Where: G h is the coal ash generation amount, with the unit of t / h; G m is the actual coal consumption, with the unit of t / h; A ar is the as-received ash content of coal; C ar is the carbon content of coal ash; The percentage of fly ash generated by the boiler in the ash and slag volume; If the air-dried basis data is obtained, perform online reference conversion according to the following formula: Where: A ad is the ash content of coal on an air-dried basis; M ar is the moisture content of the as-received coal; M ad is the moisture content of coal on an air-dried basis.

3. The online prediction method for the ash generation amount of a power plant boiler based on LSTM according to claim 1, characterized in that, The step of collecting historical data of boiler coal ash generation amount and main influencing parameters to obtain correlation relationship data and performing normalization processing includes arranging the correlation relationship data in time to obtain a sequence sample data set, and the normalization processing expression is as follows: Xi = (Xi - Xmin) / (Xmax - Xmin) Where: Xi represents the normalized value of the i-th value in the sequence sample data set; Xi represents the i-th value in the sequence sample data set; Xmin represents the minimum value in the sequence sample data set; Xmax represents the maximum value in the sequence sample data set.

4. The online prediction method for the coal ash generation amount of a power plant boiler based on LSTM according to claim 3, characterized in that In the step of dividing into a training set and a test set, after normalizing the data in the sequence sample dataset, it is divided into a training set and a test set at a time interval of 2:1, and the sampling period of each dataset can represent the characteristic change samples of the same time period.

5. The online prediction method for the ash generation amount of a power plant boiler based on LSTM according to claim 1, characterized in that, The establishment process of the boiler coal ash generation LSTM network model includes: inputting the data of the main influencing parameters and adjusting the internal parameters of the network; the data of the main influencing parameters include the unit's electricity / heat load, coal consumption, coal quality of the coal burned, and carbon content in the coal ash. The step of adjusting the internal parameters of the network includes adjusting the learning rate and the number of iterations of the established boiler coal ash generation LSTM network model according to the set values, obtaining the input weight W, the recurrent weight R, and the bias b, and respectively adjusting the hyperparameters of the corresponding input gate i, forget gate f, candidate gate g, and output gate o by using the gate activation function and the state activation function. The internal learning weights of the LSTM network are: W = [W i , W f , W g , W o T ​ R = [R i , R f , R g , R o T ​ b = [b i , b f , b g , b o T ​ The process of one iteration of the boiler coal ash generation LSTM network model is: where σ g is the gate activation function. Using the sigmoid function, then σ(X) = (1 + e -x ) -1 ; σ c is the state activation function. Using the tanh function, C t , h t are the output values of the network unit after one calculation.

6. The online prediction method for the coal ash generation amount of a power station boiler based on LSTM according to claim 5, characterized in that, Using the mean absolute error loss function and the stochastic gradient descent Adam algorithm to train and fit the boiler coal ash generation LSTM network model, dynamically modifying the learning rates of each parameter, and introducing the momentum method during the training and fitting process until the network converges.

7. An on-line prediction system for the ash generation amount of a power plant boiler based on LSTM, characterized in that, It includes: The correlation relationship data acquisition and processing module is used to collect the historical data of the boiler coal ash generation amount and the main influencing parameters to obtain the correlation relationship data and perform normalization processing; the main influencing parameters include the unit's electricity / heat load, boiler type, coal consumption, coal quality of the coal burned, and carbon content in the coal ash. The dataset construction module is used to divide the normalized correlation relationship data into a training set and a test set. The model training and fitting module is used to train and fit the pre-established boiler coal ash generation LSTM network model by using the data in the training set; the boiler coal ash generation LSTM network model includes a data sampling layer, a data preprocessing layer, an input layer, an LSTM layer, a Dropout layer, a fully connected layer, a SoftMax layer, and a classification output layer arranged in sequence; the data sampling layer is used to collect the boiler coal ash generation amount and the main influencing parameters, the data preprocessing layer is used to perform normalization processing on the data collected by the data sampling layer, the input layer is used to perform data diversity and format conversion, the LSTM layer is used to perform LSTM network training, the Dropout layer is used to supervise network overfitting, and the fully connected layer, the SoftMax layer, and the classification output layer are jointly used to output and optimize the results. The model overfitting evaluation module is used to combine the predicted data with the data in the test set to evaluate the overfitting of the trained and fitted boiler coal ash generation LSTM network model. The sampling prediction module is used to online sample the currently occurred data and input it into the boiler coal ash generation LSTM network model without overfitting evaluation to predict the boiler coal ash generation amount value in the future time. The steps of combining the predicted data with the data in the test set to evaluate the overfitting of the LSTM network model for the boiler coal ash generation amount after training and fitting include: inputting the data in the test set into the LSTM network model for the boiler coal ash generation amount for prediction, displaying the model losses of the training set and the test set in one graph, and judging whether there is an overfitting phenomenon in the model; The steps of judging whether there is an overfitting phenomenon in the model include: Calculating the error score of the model by using the root mean square error of the unit error identical to the variable through the initial predicted value and the actual value; Outputting the losses of training and testing at the end of each training epoch, and finally outputting the final root mean square error of the model for the test set data, and judging whether there is an overfitting phenomenon in the model according to the magnitude of the final root mean square error.

Images