Method and device for predicting concentration of sulfur dioxide at outlet of desulfurization system

Abstract

The invention provides a method and device for predicting the concentration of sulfur dioxide at an outlet of a desulfurization system, and belongs to the technical field of data processing. The method comprises the following steps: according to a sampling period, acquiring at least two input parameters related to the concentration of sulfur dioxide at the outlet of the desulfurization system; performing time delay compensation on each input parameter collected at each sampling moment to obtain a final input variable of the desulfurization system; and according to the final input variable of the desulfurization system, predicting the concentration of sulfur dioxide at the outlet of the desulfurization system at the target time. The method can accurately predict the outlet sulfur dioxide concentration.

Description

Technical Field

[0001] The present disclosure belongs to the field of data processing technology, and specifically relates to a method and device for predicting the sulfur dioxide concentration at the outlet of a desulfurization system, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] The current energy structure is complex and diverse, showing a diversified and vigorous development trend. However, although the fields of new energy power generation such as wind power and solar energy are gradually expanding, their share in the overall energy structure is still limited, and thermal power generation still occupies a dominant position in power supply. As an important part of thermal power generation, the sulfur oxides and nitrogen oxides produced during the operation of coal-fired power plants have become one of the sources of air pollution that cannot be ignored. At present, accurate measurement of sulfur dioxide (SO2) emission concentration is an important indicator to ensure the normal operation of coal-fired power plants. Not only that, due to the numerous chemical reactions inside the thermal power units, delayed interference and complex noise pollution, the staff faces huge challenges in the process of regulating the system parameters. Any carelessness will lead to the discharge concentration of polluted gases exceeding the standard. Therefore, it is very necessary to design an effective outlet SO2 concentration prediction model for coal-fired power plants. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art and to provide a prediction method and device, a computer-readable storage medium and a computer program product that can accurately predict the outlet sulfur dioxide concentration.

[0004] In order to achieve the above objectives, in a first aspect, an embodiment of the present disclosure provides a method for predicting the sulfur dioxide concentration at the outlet of a desulfurization system, the method comprising:

[0005] According to the sampling period, collecting input parameters related to the sulfur dioxide concentration at the outlet of the desulfurization system, wherein the input parameters include at least two types;

[0006] Performing time delay compensation on each input parameter collected at each sampling moment to obtain a final input variable of the desulfurization system;

[0007] The sulfur dioxide concentration at the outlet of the desulfurization system at the target time is predicted according to the final input variables of the desulfurization system.

[0008] In some embodiments, time delay compensation is performed on each input parameter collected at each sampling moment to obtain the final input variable of the desulfurization system, including:

[0009] Based on the time delay compensation value corresponding to each input parameter, time delay compensation is performed on the input parameter collected at each sampling moment to obtain the sampling time after compensation of each input parameter;

[0010] According to the target time and the sampling time after compensation of each input parameter, the final input variable of the desulfurization system is screened from the input parameters.

[0011] In some embodiments, predicting the sulfur dioxide concentration at the outlet of the desulfurization system at the target time according to the final input variable of the desulfurization system includes:

[0012] The final input variables of the desulfurization system are input into the prediction model to obtain the sulfur dioxide concentration at the outlet of the desulfurization system at the target time output by the prediction model, wherein the prediction model is obtained by training a preset model based on the historical input parameters and historical outlet sulfur dioxide concentration of the desulfurization system.

[0013] In some embodiments, the preset model is obtained by optimizing the hyperparameters of the long short-term memory neural network using an improved gray wolf optimization algorithm; wherein the improved gray wolf optimization algorithm is obtained by replacing the linear decreasing factor of the gray wolf optimization algorithm in the prey surrounding stage with a nonlinear decreasing factor, and / or replacing the mathematical model of hunting behavior in the gray wolf optimization algorithm with an adaptive step mathematical model, and / or replacing the mathematical model of hunting behavior in the gray wolf optimization algorithm with a dynamic weight mathematical model.

[0014] In some embodiments, the nonlinear decreasing factor is:

[0015] Among them, α1 represents the nonlinear decreasing factor, t represents the number of iterations, and T represents the total number of iterations.

[0016] In some embodiments, the adaptive stepping mathematical model is:

[0017] Among them, X(t+1) is the position of the gray wolf at the t+1th iteration, t represents the number of iterations, X1, X2, and X3 represent the position of the gray wolf at the first iteration, the second iteration, and the third iteration, respectively, and T represents the total number of iterations.

[0018] In some embodiments, the dynamic weight mathematical model is:

[0019] in,

[0020]

[0021] H = [X1|,X2|,X3];

[0022] X(t+1) represents the position of the gray wolf at the t+1th iteration; t represents the number of iterations; X1, X2, and X3 represent the position of the gray wolf at the first iteration, the second iteration, and the third iteration, respectively; T represents the total number of iterations; W1, W2, and W3 represent the dynamic weights of individuals α, β, and γ, respectively; H represents the sequence of the optimal solution, the second optimal solution, and the third optimal solution; ψ represents the middle value of the H sequence.

[0023] In some embodiments, the process of training the preset model to obtain the prediction model is as follows:

[0024] Obtaining historical input parameters and historical outlet sulfur dioxide concentrations of the desulfurization system at each historical sampling moment;

[0025] Based on the time delay compensation value corresponding to each of the historical input parameters, time delay compensation is performed on the historical input parameter collected at each of the historical sampling moments to obtain the compensated historical sampling time of each of the historical input parameters;

[0026] Constructing N training sample data according to the historical sampling time after compensation of each of the historical input parameters and the historical sampling time of the historical outlet sulfur dioxide concentration of the desulfurization system, wherein each training sample data includes at least two of the historical input parameters and at least one of the historical outlet sulfur dioxide concentrations;

[0027] The preset model is trained using the N training sample data to obtain the prediction model, where N is a positive integer greater than 1.

[0028] In some embodiments, obtaining the historical outlet sulfur dioxide concentration of the desulfurization system at each historical sampling time includes:

[0029] Collecting the original outlet sulfur dioxide concentration of the desulfurization system at each historical sampling moment;

[0030] The original outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments is subjected to noise reduction processing to obtain the historical outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments.

[0031] In some embodiments, performing noise reduction processing on the original outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments to obtain the historical outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments includes:

[0032] For each of the raw outlet sulfur dioxide concentrations, perform the following:

[0033] Decomposing the original outlet sulfur dioxide concentration into a first modal component and a second modal component;

[0034] Performing denoising processing on the first modal component to obtain a denoised first modal component;

[0035] The first modal component and the second modal component after noise reduction are superimposed to obtain the historical outlet sulfur dioxide concentration corresponding to the original outlet sulfur dioxide concentration.

[0036] In some embodiments, a wavelet denoising algorithm with improved wavelet coefficients is used to perform denoising on the first modal component to obtain a denoised first modal component, wherein the improved wavelet coefficients are:

[0037] Among them, S represents the improved wavelet coefficient; Z is the original wavelet coefficient in the wavelet denoising algorithm; θ represents the threshold.

[0038] In some embodiments, decomposing the original outlet sulfur dioxide concentration into a first modal component and a second modal component comprises:

[0039] Decompose the original outlet sulfur dioxide concentration into n different modal components;

[0040] Sort n different modal components in order of frequency from low to high;

[0041] Determine whether the sum of the superposition of the first m modal components within the sampling period is less than 1;

[0042] If the superposition sum of the first m modal components within the sampling period is less than 1, the first m-1 modal components are determined as the first modal components, and the mth modal components to the nth modal components are determined as the second modal components; wherein n and m are both positive integers, and n>m, m>1.

[0043] In some embodiments, the method further comprises:

[0044] Obtaining the sulfur dioxide concentration at the outlet of the desulfurization system at at least two sampling moments and the input parameters at M sampling moments before each of the sampling moments;

[0045] Calculating the mutual information value between the sulfur dioxide concentration at the outlet of the desulfurization system at each sampling moment and each input parameter at M sampling moments before the sampling moment;

[0046] According to the mutual information value between the sulfur dioxide concentration at the outlet of the desulfurization system at each sampling moment and the input parameters at M sampling moments before the sampling moment, the time delay compensation value corresponding to each input parameter is determined, where M is a positive integer greater than or equal to 1.

[0047] In some embodiments, the input parameters include at least two of the following: booster fan original flue gas flow, desulfurization tower original flue gas temperature, booster fan original flue gas flow, booster fan current, limestone slurry pump phase current, desulfurization tower original flue gas dust, desulfurization tower PH average value, desulfurization tower original flue gas sulfur dioxide concentration, and limestone slurry density.

[0048] In a second aspect, the present disclosure further provides an outlet sulfur dioxide concentration prediction device, the device comprising:

[0049] A collection module, used for collecting input parameters related to the sulfur dioxide concentration at the outlet of the desulfurization system according to a sampling period, wherein the input parameters include at least two types;

[0050] A compensation module, used for performing time delay compensation on each input parameter collected at each sampling moment to obtain a final input variable of the desulfurization system;

[0051] The prediction module is used to predict the sulfur dioxide concentration at the outlet of the desulfurization system at a target time according to the final input variables of the desulfurization system.

[0052] In a third aspect, the present disclosure further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the computing method of the first aspect described above is implemented.

[0053] In a fourth aspect, the present disclosure further provides a computer-readable storage medium storing a computer program for executing the computing method of the first aspect.

[0054] In a fifth aspect, the present disclosure further provides a computer program product, including a computer program / instruction, which implements the steps of the computing method of the first aspect when the computer program / instruction is executed by a processor.

[0055] The embodiment of the present disclosure eliminates the influence of delay interference on the accuracy prediction based on the time delay between each input parameter and the output outlet sulfur dioxide concentration, thereby achieving the purpose of accurately predicting the outlet sulfur dioxide concentration. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 A flow chart of a method for predicting the sulfur dioxide concentration at the outlet of a desulfurization system provided in an embodiment of the present disclosure;

[0058] Figure 2 A schematic diagram of a decreasing trend of a decreasing factor provided in an embodiment of the present disclosure;

[0059] Figure 3 A process for predicting outlet sulfur dioxide concentration provided by an embodiment of the present disclosure;

[0060] Figure 4 A structural block diagram of a device for predicting the sulfur dioxide concentration at the outlet of a desulfurization system provided in an embodiment of the present disclosure;

[0061] Figure 5 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail and completely in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0063] With the continuous development of machine learning and deep learning, data-driven modeling technologies such as least squares support vector machine (LSSVM), artificial neural network, partial least squares, etc. are widely used in parameter prediction of coal-fired power plants. Among them, since the bidirectional long short-term memory neural network (BiLSTM) can memorize the signal characteristics of historical data for a long time, the BiLSTM network has great advantages in processing time series data. In order to further improve the prediction accuracy of the model, the prior art usually uses group intelligence algorithms such as particle swarm optimization (PSO), whale optimization algorithm (WOA), ant colony algorithm, grey wolf optimization algorithm (GWO) and fish swarm algorithm to optimize the parameters of BiLSTM network. However, when processing complex data or problems, the above algorithms are prone to fall into local optimization, resulting in poor parameter optimization effect of intelligent optimization algorithm on BiLSTM network.

[0064] In the prior art, an IPSO-LSTM prediction model based on an improved particle swarm algorithm (IPSO) to optimize a long short-term memory (LSTM) neural network is proposed, including: calculating the contribution rate of each variable based on principal component analysis (PCA) to screen out the auxiliary variables of the model and achieve dimensionality reduction of the auxiliary variables; using the improved particle swarm algorithm to determine the number of neurons, learning rate and number of iterations of the LSTM neural network; finally, using the selected auxiliary variables as the input of the IPSO-LSTM prediction model to predict the mass concentration of sulfur dioxide (SO2) at the outlet of a coal-fired power plant.

[0065] A coal-fired power plant is a system with large delay and large inertia characteristics. Field control experience shows that there is often a significant time delay between the input and output variables collected from the desulfurization system of a coal-fired power plant. In addition, the operating environment of a coal-fired power plant is relatively complex, and the data collected by the desulfurization system is usually subject to greater noise interference. Both of these factors have a direct impact on the prediction accuracy of the BiLSTM model. The outlet SO2 concentration prediction model proposed in the above-mentioned prior art does not consider the impact of time delay and noise interference on the model accuracy, and the global optimization performance of the particle swarm algorithm used is poor, and the optimal hyperparameter combination of the LSTM network cannot be obtained, resulting in low prediction accuracy of the model, which cannot meet the current coal-fired power plant's prediction requirements for outlet SO2 emission concentration.

[0066] First, in order to improve the prediction accuracy of SO2 emission concentration at the outlet of coal-fired power plants, the present invention discloses a method for predicting the sulfur dioxide concentration at the outlet of a desulfurization system. This method optimizes the input parameters of the desulfurization system by eliminating the time delay between the input parameters of the desulfurization system and the outlet sulfur dioxide concentration. Secondly, by using the noise-reduced outlet sulfur dioxide concentration to generate a prediction model, the outlet sulfur dioxide concentration predicted by the prediction model is more accurate, and the noise interference of the outlet sulfur dioxide concentration is reduced. At the same time, the gray wolf optimization algorithm improved by the hybrid strategy is combined to further optimize the hyperparameter combination of the BiLSTM network, thereby achieving accurate prediction of the outlet sulfur dioxide (SO2) concentration.

[0067] Figure 1 The present invention provides a flow chart of a method for predicting the sulfur dioxide concentration at the outlet of a desulfurization system. Figure 1 As shown, the method comprises the following steps:

[0068] S11. Collect input parameters related to the sulfur dioxide concentration at the outlet of the desulfurization system according to a sampling period, wherein the input parameters include at least two types.

[0069] Specifically, firstly, the parameters that have a greater impact on the concentration of sulfur dioxide (outlet SO2) at the outlet of the desulfurization system are screened out through mechanism analysis, and are collected based on multiple sampling cycles as input parameters.

[0070] In some embodiments, the input parameters include at least two of the following: booster fan original flue gas flow, desulfurization tower original flue gas temperature, booster fan original flue gas flow, booster fan current, limestone slurry pump phase current, desulfurization tower original flue gas dust, desulfurization tower PH average value, desulfurization tower original flue gas sulfur dioxide concentration, and limestone slurry density.

[0071] S12. Perform time delay compensation on each input parameter collected at each sampling moment to obtain a final input variable of the desulfurization system.

[0072] Specifically, the desulfurization system of a coal-fired power plant is a large delay system. There is a large delay between each input parameter and the output parameter (export sulfur dioxide concentration), and the delay time between different types of input parameters is inconsistent. Therefore, eliminating the delay interference between each input parameter and the output parameter is very important for analyzing and predicting the changing trend of the outlet SO2 concentration. Among them, time delay compensation for input parameters refers to adjusting the time series corresponding to the input parameters based on the time delay, while the input parameters themselves do not change. That is, the difference between the final input variable and the input parameter of the desulfurization system is only that the time series corresponding to the two are different.

[0073] S13. Predicting the sulfur dioxide concentration at the outlet of the desulfurization system at a target time according to the final input variables of the desulfurization system.

[0074] Among them, the target time refers to the time corresponding to the sulfur dioxide concentration at the outlet of the desulfurization system to be predicted. The target time can be a moment or a time series, and the present disclosure does not limit it here. The final input variable of the desulfurization system has a certain corresponding relationship with the sulfur dioxide concentration at the outlet of the desulfurization system at the target time. That is, after the final input variable of the desulfurization system is input into the desulfurization system as input data, the outlet sulfur dioxide concentration output by the desulfurization system at the target time can be obtained.

[0075] The embodiment of the present disclosure eliminates the influence of delay interference on the accuracy prediction based on the time delay between each input parameter and the output outlet sulfur dioxide concentration, thereby achieving the purpose of accurately predicting the outlet SO2 emission concentration.

[0076] In some embodiments, step S12, performing time delay compensation on each input parameter collected at each sampling moment to obtain the final input variable of the desulfurization system, specifically comprises the following steps:

[0077] S121. Based on the time delay compensation value corresponding to each input parameter, time delay compensation is performed on the input parameter collected at each sampling time to obtain a sampling time after compensation for each input parameter.

[0078] That is, different types of input parameters correspond to different time delay compensation values. For example, the time delay compensation value corresponding to the average pH value of the desulfurization tower is different from the time delay compensation value corresponding to the original flue gas temperature of the desulfurization tower.

[0079] S122. Filter the final input variables of the desulfurization system from the input parameters according to the target time and the sampling time after each input parameter is compensated.

[0080] Specifically, the sampling time after the input parameter compensation refers to the time when the input parameter passes through the desulfurization system to obtain the sulfur dioxide concentration at the outlet of the desulfurization system. Among them, the sampling time after the input parameter compensation has a certain corresponding relationship with the target time. Optionally, the sampling time after the input parameter compensation has a one-to-one correspondence with the target time. Optionally, the sampling time after the input parameter compensation may not have a one-to-one correspondence with the target time, which can be specifically determined according to the prediction method adopted. The purpose of step S122 is to remove the parameters in the input parameters that are not related to the sulfur dioxide concentration at the outlet of the target time.

[0081] In some embodiments, the method includes not only steps S11 to S13, but also S14, a process of determining the time delay compensation value corresponding to each input parameter. Step S14 specifically includes the following steps:

[0082] S141. Obtain the sulfur dioxide concentration at the outlet of the desulfurization system at at least two sampling moments and the input parameters at M sampling moments before each of the sampling moments.

[0083] Specifically, there is a time delay between the input parameters of the desulfurization system and the output outlet sulfur dioxide concentration, that is, the outlet sulfur dioxide concentration of the desulfurization system obtained by sampling at the sampling time is obtained based on the input parameters before the sampling time.

[0084] S142, calculating the mutual information value between the sulfur dioxide concentration at the outlet of the desulfurization system at each sampling moment and each input parameter at M sampling moments before the sampling moment.

[0085] Specifically, the mutual information theory is an algorithm for measuring the degree of mutual dependence between variables. The present disclosure determines the correlation and degree of dependence between the sulfur dioxide concentration at the outlet of the desulfurization system and each input parameter by calculating the mutual information value between the two.

[0086] S143. Determine the time delay compensation value corresponding to each input parameter according to the mutual information value between the sulfur dioxide concentration at the outlet of the desulfurization system at each sampling moment and the input parameters at M sampling moments before the sampling moment, where M is a positive integer greater than or equal to 1.

[0087] Optionally, for each input parameter, the maximum value of the mutual information value between the sulfur dioxide concentration at the outlet of the desulfurization system at each sampling moment and the input parameter at N sampling moments before the sampling moment can be determined, and the difference between the sampling moment of the input parameter corresponding to the maximum value and the sampling moment of the outlet sulfur dioxide concentration corresponding to the input parameter is calculated, and the difference is used as the time delay compensation value of the input parameter.

[0088] Optionally, the maximum delay time between the variables of the desulfurization system does not exceed 600 seconds, that is, the maximum time delay compensation value is 600 seconds.

[0089] For example, take the sampling period of 2s as an example. Since the delay time between the input parameters and the output outlet sulfur dioxide concentration is kept within 600s, in order to eliminate the delay interference between the input parameters and the outlet sulfur dioxide concentration, 300 sets of input parameters are used for timing adjustment. Then, using the mutual information algorithm, according to the sampling period, the mutual information value between each input parameter and the outlet SO2 concentration at each sampling moment in the first 600 seconds is calculated in sequence. At the same time, the maximum value of the mutual information value between each input parameter and the outlet SO2 concentration is recorded, indicating that the correlation and dependence between the input parameters and the output parameters are the strongest at this time, and the sampling time of the input parameter corresponding to the maximum mutual information value and the sampling time corresponding to the outlet sulfur dioxide concentration corresponding to the input parameter are recorded. Finally, the time delay compensation value of the input parameter is determined based on the input parameter corresponding to the maximum mutual information value and the sampling time of the outlet sulfur dioxide concentration, thereby effectively eliminating the interference caused by the delay.

[0090] Of course, you can also select the average of multiple mutual information values ​​or the mutual information value with the highest frequency among the multiple mutual information values, and calculate the difference between the sampling time of the input parameter corresponding to the average of the mutual information values ​​or the mutual information value with the highest frequency among the multiple mutual information values ​​and the sampling time of the outlet sulfur dioxide concentration corresponding to the input parameter, and use the difference as the time delay compensation value of the input parameter. The present disclosure does not impose any restrictions on this.

[0091] In some embodiments, step S13, predicting the sulfur dioxide concentration at the outlet of the desulfurization system at the target time based on the final input variables of the desulfurization system, specifically includes: inputting the final input variables of the desulfurization system into a prediction model to obtain the sulfur dioxide concentration at the outlet of the desulfurization system at the target time output by the prediction model, wherein the prediction model is obtained by training a preset model based on the historical input parameters and historical outlet sulfur dioxide concentrations of the desulfurization system.

[0092] In some embodiments, in step S13, the process of training the preset model to obtain the prediction model is as follows:

[0093] Step S131, obtaining historical input parameters and historical outlet sulfur dioxide concentration of the desulfurization system at each historical sampling time.

[0094] Step S132: Based on the time delay compensation value corresponding to each of the historical input parameters, time delay compensation is performed on the historical input parameter collected at each of the historical sampling moments to obtain the compensated historical sampling time of each of the historical input parameters.

[0095] Step S133, constructing N training sample data according to the historical sampling time after compensation of each of the historical input parameters and the historical sampling time of the historical outlet sulfur dioxide concentration of the desulfurization system, wherein each training sample data includes at least two of the historical input parameters and at least one of the historical outlet sulfur dioxide concentrations.

[0096] Step S134: Use the N training sample data to train the preset model to obtain the prediction model, where N is a positive integer greater than 1.

[0097] Specifically, the present disclosure can use a distributed control system (DCS system) of a coal-fired power plant to collect 10,300 sets of data, the collected data includes historical input parameters and historical outlet sulfur dioxide concentrations, and the data sampling period is 2s. Based on the above description, since the delay time between the input parameters and the output outlet sulfur dioxide concentration is kept within 600s, in order to eliminate the delay interference between the input parameters and the output outlet sulfur dioxide concentration, the first 300 sets of data can be used for timing adjustment, and the remaining 10,000 sets of data are divided into training sets and test sets according to 8:2 for neural network training to obtain a prediction model.

[0098] The disclosed embodiment firstly performs time delay compensation on the historical input parameters of the desulfurization system at each historical sampling moment, and uses the data after time delay compensation as samples for neural network training, so that the prediction model obtained by such training is more accurate.

[0099] In some embodiments, in step S131, obtaining the historical outlet sulfur dioxide concentration of the desulfurization system at each historical sampling time specifically includes the following steps:

[0100] S1311. Collect the original outlet sulfur dioxide concentration of the desulfurization system at each historical sampling time.

[0101] S1312. Perform noise reduction processing on the original outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments to obtain the historical outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments.

[0102] Specifically, since the operating conditions of coal-fired power plants are particularly complex, even slight noise interference during the operation of the units may cause changes in the outlet SO2 concentration. Therefore, there will be more or less noise interference in the collected original outlet SO2 concentration data. Since noise interference seriously affects the prediction accuracy, the present disclosure requires denoising the original outlet SO2 concentration. The neural network training is performed based on the historical outlet sulfur dioxide concentration obtained after denoising, and the obtained prediction model can be more accurate in predicting the outlet SO2 concentration.

[0103] In some embodiments, step S1312, performing noise reduction processing on the original outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments to obtain the historical outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments, includes: for each of the original outlet sulfur dioxide concentrations, performing the following operations:

[0104] S1. Decomposing the original outlet sulfur dioxide concentration into a first modal component and a second modal component.

[0105] S2. Perform denoising on the first modal component to obtain a denoised first modal component.

[0106] S3. Superimpose the first modal component and the second modal component after noise reduction to obtain a historical outlet sulfur dioxide concentration corresponding to the original outlet sulfur dioxide concentration.

[0107] Specifically, Empirical Mode Decomposition (EMD) is a common algorithm for processing non-stationary signals and data, which can decompose the original signal into intrinsic mode functions (IMF) of different time scales. In the present disclosure, first, each original outlet SO2 concentration is decomposed into n different IMF components (IMF1, IMF2, ..., IMFn) by EMD. Among them, the first modal component can be a high-frequency modal component (referred to as high-frequency IMF component), and the second modal component can be a low-frequency modal component (referred to as low-frequency IMF component). The fluctuation of the high-frequency modal component is complex, which not only contains most of the characteristic information in the output variable, but also contains more noise interference in the output variable. The present disclosure only performs noise reduction processing on the high-frequency modal component (that is, the first modal component, referred to as the high-frequency IMF component), which can reduce the amount of calculation and improve the calculation efficiency while realizing the noise reduction of the original outlet SO2 concentration.

[0108] In some embodiments, step S1, decomposing the original outlet sulfur dioxide concentration into a first modal component and a second modal component, specifically comprises the following steps:

[0109] S101. Decompose the original outlet sulfur dioxide concentration into n different modal components.

[0110] S102. Sort n different modal components in order of frequency from low to high.

[0111] S103. Determine whether the sum of the superposition of the first m modal components within the sampling period is less than 1; if the sum of the superposition of the first m modal components within the sampling period is less than 1, determine the first m-1 modal components as the first modal components, and determine the mth modal component to the nth modal component as the second modal components; wherein n and m are both positive integers, and n>m, m>1.

[0112] Specifically, the main purpose of step S1 is to determine whether the decomposed IMF component is a high-frequency signal. The basis for the determination is: the frequency of high-frequency signals changes rapidly during the entire time period, and the envelope formed by the high-frequency signals is relatively symmetrical about the time axis. When the high-frequency signals in the entire period are superimposed, the value obtained is small or close to 0. However, the frequency of low-frequency signals changes slowly during the entire period, and the envelope formed by the low-frequency signals is asymmetrical about the time axis. The superposition value of low-frequency signals in the entire period is large or not close to 0. Therefore, the original outlet SO2 concentration is decomposed into n IMF components by this method and judged in turn. When the sum of the superposition of the lower m IMF components in the period is calculated to be lower than zero (exceeding 1), IMFm to IMFn are judged as low-frequency IMF components, and IMF1 to IMFm-1 are judged as high-frequency IMF components.

[0113] In some embodiments, in step S2, a wavelet denoising algorithm with improved wavelet coefficients is used to perform denoising on the first modal component to obtain a denoised first modal component, wherein the improved wavelet coefficients are:

[0114] Among them, S represents the improved wavelet coefficient; Z is the original wavelet coefficient in the wavelet denoising algorithm; θ represents the threshold.

[0115] Specifically, the wavelet denoising algorithm can effectively reduce the noise interference in the signal, but the choice of the threshold function directly affects the denoising effect on the characteristic signal. Common threshold functions include hard threshold and soft threshold functions. However, due to the discontinuity of the signal of the hard threshold function, the signal of the feature reconstruction is not smooth, and there is a fixed deviation between the noise estimation value and the true value of the soft threshold function, which not only causes incomplete noise processing, but also causes the loss of some high-frequency feature information. Therefore, the denoising based on the above-mentioned improved wavelet threshold function disclosed in the present invention can not only achieve better denoising effect, but also can avoid losing high-frequency feature information.

[0116] The disclosed embodiment uses an improved wavelet threshold algorithm to perform noise reduction on the decomposed high-frequency IMF; secondly, the noise-reduced high-frequency IMF component and the low-frequency IMF component are superimposed, and finally the outlet SO2 concentration after noise reduction is obtained, that is, the historical outlet sulfur dioxide concentration. The disclosed embodiment provides a noise reduction method of empirical mode decomposition combined with an improved wavelet algorithm to reduce the noise of each original outlet sulfur dioxide concentration obtained, which can achieve better noise reduction effect without losing high-frequency feature information, so that the model training is performed according to the historical outlet sulfur dioxide concentration obtained after noise reduction, and the obtained prediction model is more accurate in predicting the outlet SO2 concentration.

[0117] In some embodiments, the preset model in step S13 is obtained by optimizing the hyperparameters of the long short-term memory neural network using an improved grey wolf optimization algorithm.

[0118] Among them, the improved gray wolf optimization algorithm is obtained by replacing the linear decreasing factor of the gray wolf optimization algorithm in the prey surrounding stage with a nonlinear decreasing factor, and / or replacing the mathematical model of hunting behavior in the gray wolf optimization algorithm with an adaptive step mathematical model, and / or replacing the mathematical model of hunting behavior in the gray wolf optimization algorithm with a dynamic weight mathematical model.

[0119] Specifically, the GWO algorithm is an intelligent optimization algorithm developed under the inspiration of wolf hunting behavior. The algorithm's parameter optimization process includes four stages: social hierarchy division, encircling prey, hunting, and attacking prey.

[0120] In the social hierarchy stage, the gray wolf population is divided into α, β, γ and ω wolves from top to bottom. During the hunting process, α is the leader wolf, representing the optimal solution and responsible for the decision-making of the wolf pack; β and γ represent the second and third optimal solutions respectively, used to assist the leader wolf in making decisions; and the remaining wolves are ω, responsible for surrounding and attacking prey.

[0121] In the stage of encircling prey, the gray wolf will gradually approach and encircle the prey during the search process. The model formula for encircling prey is:

[0122] D=|C·X p (t)-X(t)|

[0123] X(t+1)=X p (t)-A·D

[0124] Where D is the distance between the wolf and its prey; X p represents the position of the prey; X(t) represents the position of the gray wolf at the tth iteration; A represents the convergence factor; t represents the current number of iterations; C represents the swing factor. The formulas of A and C are:

[0125] A=2α·r1-α

[0126] C=2r2

[0127] Among them, α decreases linearly between [0,2]; r represents a random value in [0,1].

[0128] Since the linear decreasing factor has a constant rate of change, it cannot fully reflect the optimization search process of the GWO algorithm, causing the algorithm to easily fall into local optimization. Therefore, the present disclosure introduces a nonlinear decreasing factor to adjust the local and global optimization process of the algorithm.

[0129] In some embodiments, the nonlinear decreasing factor is:

[0130] Among them, α1 represents the nonlinear decreasing factor, t represents the number of iterations, and T represents the total number of iterations.

[0131] Figure 2 A schematic diagram of a decreasing trend of a decreasing factor provided by an embodiment of the present disclosure. Figure 2 As shown, L1 represents the linear decreasing factor, and L2 represents the nonlinear decreasing factor. The rate of change of the nonlinear decreasing factor is not constant. The rate of change of the nonlinear decreasing factor is slower in the early (local search) and late (global search) stages of the iteration, and faster in the middle of the iteration cycle. It is proved that in the entire iteration cycle, the local contraction and global search time of the gray wolf algorithm are longer, and the conversion from local optimization to global optimization process can be realized quickly, making it easier for the algorithm to jump out of the local optimization process.

[0132] In the hunting phase, α, β, and γ jointly guide the wolf pack to surround the prey, and update the position of the ω wolf based on the information of the three wolves. The formula for the individual gray wolf to track the prey is:

[0133]

[0134] In order to further balance the global search and local search processes of the algorithm, the present disclosure introduces an adaptive step-shifting strategy.

[0135] In some embodiments, the adaptive stepping mathematical model is:

[0136]

[0137] Among them, X(t+1) represents the position of the gray wolf at the t+1th iteration, t represents the number of iterations, X1, X2, and X3 represent the position of the gray wolf at the first iteration, the position of the gray wolf at the second iteration, and the position of the gray wolf at the third iteration, respectively, and T represents the total number of iterations.

[0138] At the same time, due to the averaging effect of the standard GWO algorithm, the leading role of the optimal solution in the entire search process is not highlighted, because a dynamic weight strategy is introduced to enhance the important role of the optimal solution in the optimization process.

[0139] In some embodiments, the dynamic weight mathematical model is:

[0140] in,

[0141]

[0142] H=[|X1|,|X2|,|X3|]

[0143] Among them, X(t+1) represents the position of the gray wolf at the t+1th iteration; t represents the number of iterations; X1, X2, and X3 represent the position of the gray wolf at the first iteration, the second iteration, and the third iteration, respectively; T represents the total number of iterations; W1, W2, and W3 represent the dynamic weights of individuals α, β, and γ, respectively; H represents the sequence of the optimal solution, the second optimal solution, and the third optimal solution; ψ represents the middle value of the H sequence.

[0144] Finally, combining the adaptive stepping strategy and dynamic weight strategy, the formula for the next step of the gray wolf position update process is:

[0145]

[0146] Where: D α , D β , D γ Respectively represent the distances between individuals α, β, and γ at the tth iteration; X is the position of the gray wolf at the tth iteration; X(t+1) represents the position of the gray wolf at the t+1th iteration; C1, C2, C3 represent the random vectors of individuals α, β, and γ, respectively; W1, W2, W3 represent the dynamic weights of individuals α, β, and γ, respectively; H represents the sequence of the optimal solution, the second optimal solution, and the third optimal solution; ψ represents the middle value of the H sequence.

[0147] In the attack and search phase, in the process of simulating the gray wolf attacking and searching for prey, the fluctuation of the convergence factor will gradually decrease as the decreasing factor decreases. When |A|>1, the ω wolf will search extensively away from the prey, and the algorithm is in the global optimization phase; when |A|<1, the gray wolf will enter the attack phase, and the algorithm will be in the local optimization phase.

[0148] It can be understood that in the embodiments of the present disclosure, only any one of the above-mentioned improvement methods can be used to improve the Grey Wolf Optimization Algorithm, or two or three improvement methods can be mixed to improve the Grey Wolf Optimization Algorithm, and the present disclosure does not limit this.

[0149] The present invention uses a nonlinear decreasing factor, an adaptive position update strategy, a dynamic weight strategy, at least one or more of the mixed improved gray wolf optimization algorithm to improve the global optimization ability and search rate of the standard gray wolf algorithm. The improved gray wolf algorithm is used to optimize the hyperparameters of the long short-term memory neural network (BiLSTM network), specifically optimizing the learning rate, the number of hidden layers, and the number of training times in the BiLSTM to obtain the optimal hyperparameter combination of the preset model. According to multiple rounds of simulation experience, the parameter optimization ranges of the learning rate, the number of hidden layer neurons, and the number of training times of the BiLSTM network are respectively between [0.001, 0.005], [1,300], and [30, 200].

[0150] Figure 3 A flow chart of prediction of outlet SO2 emission concentration is provided in the embodiment of the present disclosure. Figure 3 As shown in the figure, firstly, the input variables are screened after the start, and the input variables related to the outlet SO2 concentration are screened out, including the original flue gas flow of the booster fan, the original flue gas temperature of the desulfurization tower, the original flue gas flow of the booster fan, the current of the booster fan, the phase current of the limestone slurry pump, the original flue gas dust of the desulfurization tower, the average pH value of the desulfurization tower, the original flue gas sulfur dioxide concentration of the desulfurization tower, and the density of the limestone slurry. Secondly, the delay interference of these input variables is eliminated. Then, the denoising algorithm based on EMD combined with the improved wavelet threshold is used to denoise the outlet sulfur dioxide concentration data of the desulfurization system collected, and the input variables and the outlet sulfur dioxide concentration are normalized. Then, the BiLSTM network is optimized based on the hybrid improved gray wolf algorithm, and the improved gray wolf algorithm is used to optimize the learning rate, number of hidden layers and number of training times of the BiLSTM network to obtain the optimal hyperparameter combination of the model, and the prediction model is obtained according to the optimal hyperparameter combination. Finally, based on the prediction model, the outlet sulfur dioxide concentration at the target moment is accurately predicted.

[0151] In some embodiments, the method further includes determining the prediction effect of the prediction model based on the actual value of the outlet SO2 concentration and the pre-test root mean square error (RMSE), absolute percentage error (MAPE) and mean absolute error (MAE).

[0152] Specifically, the root mean square error (RMSE) between the actual value and the pre-test of the outlet SO2 concentration is:

[0153]

[0154] The absolute percentage error (MAPE) between the actual value and the pre-test value of outlet SO2 concentration is:

[0155]

[0156] The mean absolute error (MAE) between the actual and pre-tested values ​​of outlet SO2 concentration is:

[0157]

[0158] Among them, y i represents the actual value of SO2 concentration at the ith outlet, Represents the predicted value, and n represents the total number of predicted samples.

[0159] In some embodiments, the method further includes optimizing the prediction model according to the prediction effect of the prediction model, so that the optimized prediction model can be more accurate in predicting the outlet SO2 concentration.

[0160] In the embodiment of the present disclosure, the mutual information algorithm is first used to eliminate the delayed interference between the input parameters and the outlet sulfur dioxide concentration, and then the input parameters after time series adjustment are used to predict the outlet sulfur dioxide concentration at the target time based on the prediction model. Among them, the prediction model is obtained through neural network training. In the process of obtaining the prediction model through neural network training, the original outlet sulfur dioxide concentration in the training sample is subjected to noise reduction processing by using the empirical mode decomposition combined with the improved wavelet threshold algorithm to reduce the noise interference of the original outlet SO2 concentration; at the same time, the hybrid improved gray wolf algorithm is used to optimize the preset model used to generate the prediction model. The present disclosure uses the prediction model obtained in this way to predict the outlet SO2 concentration at the target time, so that the prediction result is more accurate. By accurately predicting the outlet SO2 concentration, the staff can adjust the parameters of the desulfurization system in time to ensure that the SO2 emission concentration remains within a reasonable range, which is of great significance to ensure the economic and stable operation of the coal-fired unit.

[0161] On the second aspect, based on the same inventive concept, the disclosed embodiment also provides a device for predicting the sulfur dioxide concentration at the outlet of a desulfurization system.

[0162] Figure 4 This is a structural block diagram of a device 400 for predicting the sulfur dioxide concentration at the outlet of a desulfurization system provided in an embodiment of the present disclosure. Figure 4 As shown, the prediction device 400 includes a collection module 401 , a compensation module 402 and a prediction module 403 .

[0163] The acquisition module 401 is used to acquire input parameters related to the sulfur dioxide concentration at the outlet of the desulfurization system according to a sampling period, wherein the input parameters include at least two types.

[0164] In some embodiments, the input parameters include at least two of the following: booster fan original flue gas flow, desulfurization tower original flue gas temperature, booster fan original flue gas flow, booster fan current, limestone slurry pump phase current, desulfurization tower original flue gas dust, desulfurization tower PH average value, desulfurization tower original flue gas sulfur dioxide concentration, and limestone slurry density.

[0165] The compensation module 402 is used to perform time delay compensation on each input parameter collected at each sampling moment to obtain the final input variable of the desulfurization system.

[0166] The prediction module 403 predicts the sulfur dioxide concentration at the outlet of the desulfurization system at the target time according to the final input variables of the desulfurization system.

[0167] In some embodiments, the compensation module 402 includes a first compensation unit 412 and a screening unit 422 .

[0168] The first compensation unit 412 is used to perform time delay compensation on each input parameter collected at each sampling time based on the time delay compensation value corresponding to each input parameter, so as to obtain the sampling time of each input parameter after compensation.

[0169] The screening unit 422 is used to screen the final input variable of the desulfurization system from the input parameters according to the target time and the sampling time after each input parameter is compensated.

[0170] In some embodiments, the first compensation unit 412 includes a first acquisition unit 51 , a calculation unit 52 , and a determination unit 53 .

[0171] The first acquisition unit 51 is used to acquire the sulfur dioxide concentration at the outlet of the desulfurization system at at least two sampling moments and the input parameters at M sampling moments before each of the sampling moments.

[0172] The calculation unit 52 is used to calculate the mutual information value between the sulfur dioxide concentration at the outlet of the desulfurization system at each sampling moment and the input parameters at M sampling moments before the sampling moment.

[0173] The determination unit 53 is used to determine the time delay compensation value corresponding to each input parameter according to the mutual information value between the sulfur dioxide concentration at the outlet of the desulfurization system at each sampling moment and the input parameters at M sampling moments before the sampling moment.

[0174] In some embodiments, the prediction module 404 is specifically used to input the final input variables of the desulfurization system into a prediction model to obtain the sulfur dioxide concentration at the outlet of the desulfurization system at the target time output by the prediction model, wherein the prediction model is obtained by training a preset model based on the historical input parameters and historical outlet sulfur dioxide concentrations of the desulfurization system.

[0175] In some implementations, the prediction device 400 includes not only a collection module 401, a compensation module 402 and a prediction module 404, but also a training module 405. The training module 405 is used to train the preset model to obtain the prediction model.

[0176] In some embodiments, the training module 405 includes a second acquisition unit 415 , a second compensation unit 425 , a construction unit 435 , and a training unit 445 .

[0177] The second acquisition unit 415 is used to acquire the historical input parameters and the historical outlet sulfur dioxide concentration of the desulfurization system at each historical sampling moment;

[0178] The second compensation unit 425 is used to perform time delay compensation on each historical input parameter collected at the historical sampling time based on the time delay compensation value corresponding to each historical input parameter, so as to obtain the compensated historical sampling time of each historical input parameter.

[0179] The construction unit 435 is used to construct N training sample data according to the historical sampling time after compensation of each of the historical input parameters and the historical sampling time of the historical outlet sulfur dioxide concentration of the desulfurization system, wherein each training sample data includes at least two of the historical input parameters and at least one of the historical outlet sulfur dioxide concentrations.

[0180] The training unit 445 is used to train the preset model using the N training sample data to obtain the prediction model, where N is a positive integer greater than 1.

[0181] In some embodiments, the second acquisition unit 415 includes a collection subunit 4151 and a noise reduction unit 4152 .

[0182] The collection subunit 4151 is used to collect the original outlet sulfur dioxide concentration of the desulfurization system at each historical sampling moment.

[0183] The noise reduction unit 4152 is used to perform noise reduction processing on the original outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments to obtain the historical outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments.

[0184] In some embodiments, the noise reduction unit 4152 includes a decomposition unit 61 , a noise reduction sub-unit 62 and a superposition unit 63 .

[0185] The decomposition unit 61 is used to decompose the original outlet sulfur dioxide concentration into a first modal component and a second modal component.

[0186] The denoising subunit 62 is used to perform denoising processing on the first modal component to obtain a denoised first modal component.

[0187] The superposition unit 63 is used to superimpose the first modal component and the second modal component after noise reduction to obtain the historical outlet sulfur dioxide concentration corresponding to the original outlet sulfur dioxide concentration.

[0188] In some embodiments, the denoising subunit 62 is specifically configured to perform denoising on the first modal component using a wavelet denoising algorithm with improved wavelet coefficients to obtain a denoised first modal component, wherein the improved wavelet coefficients are:

[0189] Among them, S represents the improved wavelet coefficient; Z is the original wavelet coefficient in the wavelet denoising algorithm; θ represents the threshold.

[0190] In some embodiments, the decomposition unit 61 includes a decomposition sub-unit 611 , a sorting unit 612 and a judgment unit 613 .

[0191] Among them, the original outlet sulfur dioxide concentration of the decomposition sub-is decomposed into n different modal components.

[0192] The sorting unit 612 is used to sort the n different modal components in order of frequency from low to high.

[0193] The judgment unit 613 is used to judge whether the superposition sum of the first m modal components within the sampling period is less than 1; and when the superposition sum of the first m modal components within the sampling period is less than 1, the first m-1 modal components are determined as first modal components, and the mth modal component to the nth modal component are determined as second modal components; wherein n and m are both positive integers, and n>m, m>1.

[0194] In some embodiments, the preset model is obtained by optimizing the hyperparameters of the long short-term memory neural network using an improved gray wolf optimization algorithm; wherein the improved gray wolf optimization algorithm is obtained by replacing the linear decreasing factor of the gray wolf optimization algorithm in the prey surrounding stage with a nonlinear decreasing factor, and / or replacing the mathematical model of hunting behavior in the gray wolf optimization algorithm with an adaptive step mathematical model, and / or replacing the mathematical model of hunting behavior in the gray wolf optimization algorithm with a dynamic weight mathematical model.

[0195] In some embodiments, the nonlinear decreasing factor is:

[0196] Among them, α1 represents the nonlinear decreasing factor, t represents the number of iterations, and T represents the total number of iterations.

[0197] In some embodiments, the adaptive stepping mathematical model is:

[0198] Among them, X(t+1) represents the position of the gray wolf at the t+1th iteration, t represents the number of iterations, X1, X2, and X3 represent the position of the gray wolf at the first iteration, the position of the gray wolf at the second iteration, and the position of the gray wolf at the third iteration, respectively, and T represents the total number of iterations.

[0199] In some embodiments, the dynamic weight mathematical model is:

[0200] in,

[0201]

[0202]

[0203] H = [|X1|,X2|,|X3|];

[0204] X(t+1) represents the position of the gray wolf at the t+1th iteration; t represents the number of iterations; X1, X2, and X3 represent the position of the gray wolf at the first iteration, the second iteration, and the third iteration, respectively; T represents the total number of iterations; W1, W2, and W3 represent the dynamic weights of individuals α, β, and γ, respectively; H represents the sequence of the optimal solution, the second optimal solution, and the third optimal solution; ψ represents the middle value of the H sequence.

[0205] In some embodiments, the prediction device 400 also includes a prediction effect determination module 407, which is used to determine the prediction effect of the optimized prediction model based on the actual value of the outlet SO2 concentration and the pre-tested root mean square error (RMSE), absolute percentage error (MAPE) and mean absolute error (MAE).

[0206] In some embodiments, the prediction device 400 further includes an optimization module 406 for optimizing the optimized prediction model according to the prediction effect of the optimized prediction model.

[0207] The prediction device 400 provided by the embodiment of the present disclosure, the compensation module 402 uses the mutual information algorithm to eliminate the delay interference between the input parameters and the outlet sulfur dioxide concentration, and obtains the final input parameters of the prediction model. Among them, the prediction module 404 uses the hybrid improved gray wolf algorithm to optimize the BiLSTM network to obtain the preset model to be trained. The noise reduction unit 4152 uses the empirical mode decomposition combined with the improved wavelet threshold algorithm to reduce the noise interference of the original outlet SO2 concentration, and obtains the historical outlet sulfur dioxide concentration as a training sample for the neural network model. Based on the preset model, the training samples obtained after processing are used to perform neural network training to obtain the final prediction model, and realize the effective prediction of the outlet SO2 concentration. By accurately predicting the outlet SO2 concentration, the staff can adjust the parameters of the desulfurization system in time to ensure that the SO2 emission concentration remains within a reasonable range, which is of great significance to ensure the economic and stable operation of the coal-fired unit.

[0208] It can be understood that the specific details and corresponding technical effects of the prediction device provided in the embodiment of the present disclosure can refer to the specific details and technical effects of the prediction method embodiment of the first aspect mentioned above, and will not be repeated here.

[0209] In a third aspect, an embodiment of the present disclosure further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method described in any one of the first aspects is implemented.

[0210] Specifically, Figure 5 A schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as Figure 5 As shown, the electronic device 005 includes a processor 501, a memory 502 and a bus 503. The processor 501 and the memory 502 communicate with each other via the bus 503.

[0211] The processor 501 is used to call the program instructions in the memory 502 to execute the methods provided by the above-mentioned method embodiments.

[0212] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program of the method described in any one of the first aspects above.

[0213] In a fifth aspect, an embodiment of the present disclosure further provides a computer program product, comprising a computer program / instructions, which, when executed by a processor, implement the steps of any one of the methods described in the first aspect above.

[0214] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0215] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0216] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0217] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0218] The present disclosure uses specific embodiments to illustrate the principles and implementation methods of the present disclosure. The description of the above embodiments is only used to help understand the method of the present disclosure and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present disclosure, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present disclosure.

Claims

1. A method for predicting the sulfur dioxide concentration at the outlet of a desulfurization system, characterized in that: The method comprises: According to the sampling period, collecting input parameters related to the sulfur dioxide concentration at the outlet of the desulfurization system, wherein the input parameters include at least two types; Performing time delay compensation on each input parameter collected at each sampling moment to obtain a final input variable of the desulfurization system; The sulfur dioxide concentration at the outlet of the desulfurization system at the target time is predicted according to the final input variables of the desulfurization system.

2. The method according to claim 1, characterized in that Perform time delay compensation on each input parameter collected at each sampling moment to obtain the final input variables of the desulfurization system, including: Based on the time delay compensation value corresponding to each input parameter, time delay compensation is performed on the input parameter collected at each sampling moment to obtain the sampling time after compensation of each input parameter; According to the target time and the sampling time after compensation of each input parameter, the final input variable of the desulfurization system is screened from the input parameters.

3. The method according to claim 1, characterized in that: The step of predicting the sulfur dioxide concentration at the outlet of the desulfurization system at a target time according to the final input variable of the desulfurization system comprises: The final input variables of the desulfurization system are input into the prediction model to obtain the sulfur dioxide concentration at the outlet of the desulfurization system at the target time output by the prediction model, wherein the prediction model is obtained by training a preset model based on the historical input parameters and historical outlet sulfur dioxide concentration of the desulfurization system.

4. The method according to claim 3, characterized in that The preset model is obtained by optimizing the hyperparameters of the long short-term memory neural network using an improved gray wolf optimization algorithm; wherein the improved gray wolf optimization algorithm is obtained by replacing the linear decreasing factor of the gray wolf optimization algorithm in the prey encirclement stage with a nonlinear decreasing factor, and / or replacing the mathematical model of hunting behavior in the gray wolf optimization algorithm with an adaptive step mathematical model, and / or replacing the mathematical model of hunting behavior in the gray wolf optimization algorithm with a dynamic weight mathematical model.

5. The method according to claim 4, characterized in that The nonlinear decreasing factor is: Among them, α1 represents the nonlinear decreasing factor, t represents the number of iterations, and T represents the total number of iterations.

6. The method according to claim 4, characterized in that The adaptive stepping mathematical model is: Among them, X(t+1) is the position of the gray wolf at the t+1th iteration, t represents the number of iterations, X1, X2, and X3 represent the position of the gray wolf at the first iteration, the second iteration, and the third iteration, respectively, and T represents the total number of iterations.

7. The method according to claim 4, characterized in that The dynamic weight mathematical model is: in, H = [|X1|,|X2|,|X3|]; X(t+1) represents the position of the gray wolf at the t+1th time; t represents the number of iterations; X1, X2, and X3 represent the position of the gray wolf at the first iteration, the second iteration, and the third iteration, respectively; T represents the total number of iterations; W1, W2, and W3 represent the dynamic weights of individuals α, β, and γ, respectively; H represents the sequence of the optimal solution, the second optimal solution, and the third optimal solution; ψ represents the middle value of the H sequence.

8. The method according to claim 3, characterized in that The process of training the preset model to obtain the prediction model is as follows: Obtaining historical input parameters and historical outlet sulfur dioxide concentrations of the desulfurization system at each historical sampling moment; Based on the time delay compensation value corresponding to each of the historical input parameters, time delay compensation is performed on the historical input parameter collected at each of the historical sampling moments to obtain the compensated historical sampling time of each of the historical input parameters; Constructing N training sample data according to the historical sampling time after compensation of each of the historical input parameters and the historical sampling time of the historical outlet sulfur dioxide concentration of the desulfurization system, wherein each training sample data includes at least two of the historical input parameters and at least one of the historical outlet sulfur dioxide concentrations; The preset model is trained using the N training sample data to obtain the prediction model, where N is a positive integer greater than 1.

9. The method according to claim 8, characterized in that Obtain the historical outlet sulfur dioxide concentration of the desulfurization system at each historical sampling time, including: Collecting the original outlet sulfur dioxide concentration of the desulfurization system at each historical sampling moment; The original outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments is subjected to noise reduction processing to obtain the historical outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments.

10. The method according to claim 9, characterized in that The original outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments is subjected to noise reduction processing to obtain the historical outlet sulfur dioxide concentration of the desulfurization system at each of the historical sampling moments, including: For each of the raw outlet sulfur dioxide concentrations, perform the following: Decomposing the original outlet sulfur dioxide concentration into a first modal component and a second modal component; Performing denoising processing on the first modal component to obtain a denoised first modal component; The first modal component and the second modal component after noise reduction are superimposed to obtain the historical outlet sulfur dioxide concentration corresponding to the original outlet sulfur dioxide concentration.

11. The method according to claim 10, characterized in that The first modal component is subjected to denoising by using a wavelet denoising algorithm with improved wavelet coefficients to obtain a denoised first modal component, wherein the improved wavelet coefficients are: Among them, S represents the improved wavelet coefficient; Z is the original wavelet coefficient in the wavelet denoising algorithm; θ represents the threshold.

12. The method according to claim 10, characterized in that The step of decomposing the original outlet sulfur dioxide concentration into a first modal component and a second modal component comprises: Decompose the original outlet sulfur dioxide concentration into n different modal components; Sort n different modal components in order of frequency from low to high; Determine whether the sum of the superposition of the first m modal components within the sampling period is less than 1; If the superposition sum of the first m modal components within the sampling period is less than 1, the first m-1 modal components are determined as the first modal components, and the mth modal components to the nth modal components are determined as the second modal components; wherein n and m are both positive integers, and n>m, m>1.

13. The method according to claim 2 or 8, characterized in that: The method further comprises: Obtaining the sulfur dioxide concentration at the outlet of the desulfurization system at at least two sampling moments and the input parameters at M sampling moments before each of the sampling moments; Calculating the mutual information value between the sulfur dioxide concentration at the outlet of the desulfurization system at each sampling moment and each input parameter at M sampling moments before the sampling moment; According to the mutual information value between the sulfur dioxide concentration at the outlet of the desulfurization system at each sampling moment and the input parameters at M sampling moments before the sampling moment, the time delay compensation value corresponding to each input parameter is determined, where M is a positive integer greater than or equal to 1.

14. The method according to claim 1, characterized in that The input parameters include at least two of the following: booster fan original flue gas flow, desulfurization tower original flue gas temperature, booster fan original flue gas flow, booster fan current, limestone slurry pump phase current, desulfurization tower original flue gas dust, desulfurization tower PH average value, desulfurization tower original flue gas sulfur dioxide concentration, limestone slurry density.

15. A device for predicting sulfur dioxide concentration at the outlet of a desulfurization system, characterized in that: The device comprises: A collection module, used for collecting input parameters related to the sulfur dioxide concentration at the outlet of the desulfurization system according to a sampling period, wherein the input parameters include at least two types; A compensation module, used for performing time delay compensation on each input parameter collected at each sampling moment to obtain a final input variable of the desulfurization system; The prediction module is used to predict the sulfur dioxide concentration at the outlet of the desulfurization system at a target time according to the final input variables of the desulfurization system.

16. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 14 is implemented.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program for executing the method according to any one of claims 1 to 14.

18. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 14 are implemented.

Images