A coal-fired power plant dry electrostatic precipitator outlet flue dust emission concentration prediction model and method

Abstract

The application discloses a coal-fired power plant dry electric dust removal outlet flue dust emission concentration prediction model and method, the model is a data-driven online self-learning extreme learning machine model with a forgetting factor, the input of the model is data containing unit load, total air volume, total coal quantity, secondary voltage and secondary current intensity values of each electric field of the dust remover, and a rapping cycle, and the output of the model is a coal-fired power plant dry electric dust removal outlet flue dust emission concentration value. The application adopts the data-driven online self-learning extreme learning machine model with a forgetting factor, is faster in calculation speed, is reliable and effective in calculation method, is diversified in input data, is higher in calculation precision, can provide model support for optimization control of a dry electric dust remover of a coal-fired power plant, can provide guidance and reference for optimization operation of the dry electric dust remover by power plant operators, can solve problems such as high power consumption and high operation intensity of operators in actual operation of the dry electric dust removal system, and can improve the safety and economy of operation of the dry electric dust removal system of the unit.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of energy saving and environmental protection operation of coal-fired power plants, and particularly relates to a coal-fired power plant dry electrostatic precipitator outlet flue dust emission concentration prediction model and method. BACKGROUND

[0002] At present, most of the dry electrostatic precipitators of coal-fired power plants are manually operated, and automatic closed-loop control has not been realized. The main reason is that the flue dust concentration at the inlet of the dry electrostatic precipitator cannot be measured online, and simple closed-loop control of the outlet concentration of the dry electrostatic precipitator cannot quickly and effectively realize closed-loop control of the concentration, resulting in rough operation of the dry electrostatic precipitator of the power plant, high energy consumption, and great difficulty in operation adjustment. The key to solving this problem is to establish a relatively accurate dry electrostatic precipitator outlet concentration prediction model and prediction method. Based on the model, automatic and efficient operation of the dry electrostatic precipitator can be realized, and guidance and reference for the optimization operation of the dry electrostatic precipitator by the power plant operation personnel can be provided, the energy consumption of the dry electrostatic precipitator system can be reduced, and safe, economic, efficient, and environmentally friendly operation of the dry electrostatic precipitator system can be realized. However, there is currently no effective dry electrostatic precipitator outlet concentration prediction model and prediction method. SUMMARY

[0003] To solve the technical problems in the prior art, the purpose of the present application is to provide a coal-fired power plant dry electrostatic precipitator outlet flue dust emission concentration prediction model and method.

[0004] To achieve the above purposes and achieve the above technical effects, the technical solution adopted by the present application is as follows:

[0005] A coal-fired power plant dry electrostatic precipitator outlet flue dust emission concentration prediction model, the model is an online self-learning extreme learning machine model with a forgetting factor, the input of the model includes data such as unit load, total air volume, total coal volume, secondary voltage and secondary current intensity of each electric field of the precipitator, and dusting cycle, and the output of the model is the dry electrostatic precipitator outlet flue dust emission concentration value of the coal-fired power plant.

[0006] A prediction method for a coal-fired power plant dry electrostatic precipitator outlet flue dust emission concentration prediction model, comprising the following steps:

[0007] First, input data including unit load, total air volume, total coal volume, secondary voltage and secondary current intensity of each electric field of the precipitator, and dusting cycle are obtained, and then data screening is performed;

[0008] Subsequently, the screened data is input into a data-driven extreme learning machine model for model initialization calculation;

[0009] Finally, the online updated and screened data is input into an online self-learning extreme learning machine model with a forgetting factor for online model calculation, and the dry electrostatic precipitator outlet flue dust emission concentration value of the coal-fired power plant is output.

[0010] Further, the data screening includes data low-pass filtering, data trend item elimination, data singularity elimination, and steady-state data filtering.

[0011] Further, the step of model initialization calculation of the data-driven extreme learning machine model includes:

[0012] (1) setting the number of hidden layer nodes of the extreme learning machine as lo, and setting the output prediction mean square error expectation threshold as η;

[0013] (2) randomly setting the input weight matrix W and the bias matrix b of the data-driven extreme learning machine model;

[0014] (3) increasing the number of hidden layer nodes, and the increment is l a , then lo = lo + l a , and the input weight w and the bias b0 of the newly added hidden layer node are also randomly set, and the input weight matrix and the bias matrix after the increase of the number of hidden layer nodes are iteratively calculated according to the following formula:

[0015]

[0016] (4) calculating the output of the hidden layer node using the activation function, and the calculation formula is:

[0017]

[0018] and then the output matrix T of the hidden layer node is obtained.

[0019] (5) the output weight matrix β of the hidden layer node is updated according to the formula: β = YT -1 ;

[0020] (6) calculating the mean square error Mse after the increase of the hidden layer node, if the mean square error Mse is within the expectation threshold η, the model initialization calculation is ended, otherwise returning to step (3) for iterative calculation.

[0021] Further, the step of model initialization calculation of the data-driven extreme learning machine model includes:

[0022] (1) first, the online data after data screening is used as the model calculation data sample, assuming that the number of newly added data samples is k1, and n samples in the single calculation selection data sample are selected to obtain the output update matrix H of the extreme learning machine hidden layer;

[0023] (2) iteratively and recursively calculating the output weight matrix β (k+1) of the extreme learning machine hidden layer:

[0024]

[0025]

[0026] In the formula, P is a temporary intermediate variable matrix in the self-learning recursive calculation process, the subscripts k and k+1 represent the kth and (k+1)th iteration calculation (the same below); H is an implicit layer output matrix in the self-learning recursive calculation process; β represents a connection weight matrix of the model implicit layer and the output layer in the self-learning recursive calculation process; and T represents a calculation output matrix of the online model.

[0027] (3) when the number of samples increased in the iteration calculation reaches k1, the final output weight β is determined k+1 ; otherwise, return to step (2) for iteration calculation.

[0028] Further, the calculation formula of the implicit layer node output matrix T is:

[0029]

[0030] wherein T1 represents the first implicit layer node output, T2 represents the second implicit layer node output, and T l0 represents the l0th implicit layer node output.

[0031] Compared with the prior art, the present application has the following beneficial effects:

[0032] The application discloses a coal-fired power plant dry electric dust removal outlet flue dust emission concentration prediction model and method. The model is a data-driven online self-learning extreme learning machine model with a forgetting factor. The input of the model is data containing unit load, total air volume, total coal volume, secondary voltage and secondary current intensity values of each electric field of a dust remover, and a rapping cycle. The output of the model is a dry electric dust removal outlet flue dust emission concentration value of the coal-fired power plant. The coal-fired power plant dry electric dust removal outlet flue dust emission concentration prediction model and method provided by the application adopt a data-driven online self-learning extreme learning machine model with a forgetting factor, and have the advantages of faster calculation speed, reliable and effective calculation method, diversified input data, higher calculation accuracy, and the like. The input data can contain unit load, total air volume, total coal volume, secondary voltage and secondary current intensity values of each electric field of a dust remover, and a rapping cycle. After data screening, the input data is input into the model, so that the problem of distortion of the model calculation value caused by noise and mutation of the input data is reduced. The model can provide model support for optimization control of the dry electric dust remover of the coal-fired power plant, and can provide guidance and reference for optimization operation of the dry electric dust remover by the power plant operation personnel. The model can solve the problems of high power consumption and high operation intensity of the operation personnel in the actual operation of the dry electric dust removal system, and can improve the safety and economy of the operation of the dry electric dust removal system of the unit. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The schematic diagram of the present application. DETAILED DESCRIPTION

[0034] The present application will be described in detail below so that the advantages and features of the present application can be more easily understood by those skilled in the art, and the scope of protection of the present application can be more clearly defined.

[0035] The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.

[0036] The present application discloses a coal-fired power plant dry electrostatic precipitator outlet flue dust emission concentration prediction model, which is a data-driven online self-learning extreme learning machine model with a forgetting factor. The input of the model includes data such as unit load, total air volume, total coal volume, secondary voltage and secondary current intensity of each electric field of the precipitator, and rapping cycle. The output of the model is the dry electrostatic precipitator outlet flue dust emission concentration value of the coal-fired power plant.

[0037] As shown in Figure 1 The present application also discloses a coal-fired power plant dry electrostatic precipitator outlet flue dust emission concentration prediction method, which comprises the following steps:

[0038] Firstly, the input data including unit load, total air volume, total coal volume, secondary voltage and secondary current intensity of each electric field of the precipitator, and rapping cycle are selected from the DCS historical data of the coal-fired power plant precipitator system, and then data screening is performed, including data low-pass filtering, data trend item elimination, data singularity elimination, and steady-state data filtering. Existing data screening technology can be used;

[0039] Subsequently, the screened data is sent to the data-driven extreme learning machine model for model initialization calculation. The model initialization training sample data obtained by data screening of the DCS historical data of the power plant precipitator system is used, the sample data input data matrix U and the output data matrix Y are calculated; the steps of the data-driven extreme learning machine model for model initialization calculation include:

[0040] (1) The number of hidden layer nodes of the extreme learning machine is set to l0, and the output prediction mean square error expectation threshold is set to η;

[0041] (2) The input weight matrix W and the bias matrix b of the data-driven extreme learning machine model are randomly set;

[0042] (3) The number of hidden layer nodes is increased, and the incremental node number is la Then, l0 = l0 + 1 a Similarly, the input weight w and bias b0 of the newly added hidden layer node are randomly set, and the iterative calculation formula of the input weight matrix and the bias matrix after the number of hidden layer nodes is increased is:

[0043]

[0044] (4) The output of the hidden layer node is calculated by the activation function, and the calculation formula is:

[0045]

[0046] Based on the output value T i of a single hidden layer node, the output matrix T of the hidden layer node is obtained:

[0047]

[0048] Where T1 represents the output of the first hidden layer node, T2 represents the output of the second hidden layer node, and T l0 represents the output of the l0th hidden layer node.

[0049] (5) The output weight matrix β of the hidden layer node is updated as follows: β = YT -1

[0050] (6) Calculate the mean square error Mse after adding the hidden layer node, if the mean square error Mse is within the expected threshold η, the model initialization calculation is completed, otherwise return to step (3) iterative calculation, wherein the calculation formula of the mean square error Mse is:

[0051]

[0052] Finally, the data selected by online updating is input into the online self-learning extreme learning machine model with a forgetting factor to perform online model calculation, and the output of the coal-fired power plant dry electrostatic precipitator outlet dust emission concentration value is output. The calculation steps include:

[0053] (1) First, the online data after data screening is used as model calculation data samples, assuming that the number of newly added data samples is k1, and n samples in the single calculation selection data sample are selected to obtain the output update matrix H of the extreme learning machine hidden layer.

[0054] (2) Online iterative recursion is performed to calculate the output weight matrix β (k+1) of the extreme learning machine hidden layer:

[0055]

[0056] ​

[0057] In the formula, the temporary intermediate variable matrix in the self-learning recursive calculation process is P, the subscripts k and k+1 represent the kth and (k+1)th iteration calculation (the same below without repeated enumeration); H is the output matrix of the hidden layer in the self-learning recursive calculation process; β represents the connection weight matrix of the model hidden layer and the output layer in the self-learning recursive calculation process; T represents the calculation output matrix of the online model;

[0058] (3) When the number of samples increased by iteration reaches k1, the final output weight β is determined. k+1 Otherwise, return to step (2) for iteration calculation.

[0059] The parts or structures not specifically described in the present application can adopt the prior art or existing products, which will not be described here.

[0060] The above description is only an embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation using the content of the present application specification, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A prediction method of a prediction model of an outlet fly ash emission concentration of a dry electrostatic precipitator of a coal-fired power plant, characterized by, The coal-fired power plant dry electrostatic precipitator outlet flue dust emission concentration prediction model is an online self-learning extreme learning machine model with a forgetting factor, the input of the model is data including unit load, total air volume, total coal volume, secondary voltage and secondary current intensity value of each electric field of the precipitator, and rapping cycle, and the output of the model is the dry electrostatic precipitator outlet flue dust emission concentration value of the coal-fired power plant; The prediction method comprises the following steps: Firstly, input data including unit load, total air volume, total coal volume, secondary voltage and secondary current intensity value of each electric field of the precipitator, and rapping cycle are obtained, and then data screening is performed; Then, the screened data are input into a data-driven extreme learning machine model for model initialization calculation; Finally, the online updated screened data are input into an online self-learning extreme learning machine model with a forgetting factor for online model calculation, and the dry electrostatic precipitator outlet flue dust emission concentration value of the coal-fired power plant is output; The model initialization calculation step of the data-driven extreme learning machine model comprises: (1) the number of hidden layer nodes of the extreme learning machine is set to , and the output prediction mean square error expectation threshold is set to ; (2) randomly setting an input weight matrix of a data-driven extreme learning machine model W and a bias matrix b ; (3) The number of nodes in the hidden layer is increased, and the incremental number of nodes is Then The input weight of the newly added hidden layer node is also randomly set And the bias Then, the input weight matrix and the bias matrix after the number of hidden layer nodes is increased are calculated iteratively as follows: ； (4) an implicit layer node output calculation excitation function is used to calculate the implicit layer node output, and the calculation formula is: ； Further, the hidden layer node output matrix T ; (5) the output weight matrix of the hidden layer node The update formula is: ; (6) Calculate the mean square error after increasing the nodes of the hidden layer Mse If the mean square error Mse If the mean square error is within the expected threshold , the model initialization calculation ends, otherwise return to step (3) for iterative calculation.

2. The method of claim 1, wherein the method is characterized by: The data screening comprises data low-pass filtering, data trend item elimination, data singularity elimination, and steady-state data filtering.

3. The method of claim 1, wherein the method further comprises: determining the concentration of the dry electrostatic precipitator outlet flue dust emissions of the coal-fired power plant based on the determined concentration of the dry electrostatic precipitator outlet flue dust emissions of the coal-fired power plant. The hidden layer node output matrix T The calculation formula is: ； wherein, T 1 represents the output of the first hidden layer node, T 2 represents the output of the second hidden layer node, T l0 represents the output of the 10th hidden layer node.

4. The method of claim 1, wherein the method further comprises: determining the concentration of the dry electrostatic precipitator outlet flue dust emissions of the coal-fired power plant based on the determined concentration of the dry electrostatic precipitator outlet flue dust emissions of the coal-fired power plant. The online model calculation step of inputting the online updated screened data into the online self-learning extreme learning machine model with a forgetting factor comprises: (1) First, the online data after data screening is taken as the model calculation data sample, and it is assumed that the number of newly added data samples is , n in each calculation is selected from the data sample, and the output update matrix H of the extreme learning machine hidden layer is obtained; (2) Online iterative recursive computation of the output weight matrix of the hidden layer of the extreme learning machine : ； In the formula, the temporary intermediate variable matrix in the self-learning recursive calculation process is , the subscripts k and k+1 respectively represent the k and k+1 times of iterative calculation; is the hidden layer output matrix in the self-learning recursive calculation process; represents the connection weight matrix of the model hidden layer and the output layer in the self-learning recursive calculation process; represents the calculation output matrix of the online model; (3) When the increased sample number reaches , the final output weight is determined; otherwise, go back to step (2) for iterative calculation.

Images