Method and system for predicting pollutant concentration in solid waste incineration process

Abstract

The invention relates to a solid waste incineration process pollutant concentration prediction method and system. The method comprises the steps that related process variable historical data influencing solid waste incineration process pollutant concentration changes are collected to obtain a training set; establishing a supervision mechanism according to a hidden layer of a random configuration network based on the training set to obtain hidden layer nodes conforming to the supervision mechanism; determining an output weight of the random configuration network based on the obtained hidden layer node conforming to the supervision mechanism; and constructing a random configuration network prediction model based on the obtained output weight so as to predict the pollutant concentration in the solid waste incineration process. According to the method, on the premise that the accuracy of the model is guaranteed, the SCN prediction model can be rapidly constructed, accurate and rapid prediction of the pollutant concentration is achieved, a foundation is laid for optimization control of the solid waste incineration process, and the working efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of solid waste incineration, and in particular, to a method and system for predicting pollutant concentration during solid waste incineration. Background Art

[0002] With the acceleration of the urbanization process, the urban population density has been continuously rising, and the generation amount of urban solid waste has also been increasing year by year. The principles for urban solid waste treatment are reduction, harmlessness, and resource utilization. Currently, the methods for treating urban solid waste in China are landfill, composting, and incineration, among which landfill accounts for the largest proportion. However, landfill occupies a large amount of land resources and is not sustainable, which does not conform to the garbage treatment principles. Incineration can not only solve the land resource problem but also meet the requirements of harmlessness and light weight of solid waste. Therefore, the state strongly supports and advocates the incineration treatment of solid waste. However, the incineration brings about the problem of secondary pollution of solid waste, that is, solid waste incineration will produce pollutants, which not only damage the ecological environment but also have a great impact on people's physical and mental health. Therefore, quickly and accurately predicting the pollutant concentration can not only provide a theoretical basis for the control of the incineration process but also improve the working efficiency of power plants and achieve low-concentration emissions of pollutants.

[0003] Currently, urban solid waste incineration plants usually use a continuous flue gas monitoring system to measure the pollutant concentration. However, since the system works in a high-temperature and high-pressure environment for a long time, the internal sensors, probes, and optical components will age, corrode, and be damaged, thus reducing the monitoring ability. In addition, the equipment procurement and maintenance costs of the continuous flue gas monitoring system are relatively high, and the instrument measurement has a lag, making it difficult to meet the requirements of rapidity and accuracy for the optimization control of the solid waste incineration process. With the development of machine learning and big data technologies, data-driven parameter prediction methods have been widely applied to industrial processes. In particular, the recently emerged Stochastic Configuration Network (SCN) can effectively solve the problem of difficult determination of network structure, and thus has received extensive attention in the field of industrial production. However, when the parameter settings of the supervision mechanism in SCN are unreasonable, that is, when the constraint degree of the supervision mechanism is too strict, it may reduce the configuration efficiency of the hidden layer parameters. Moreover, the increase in the number of hidden layer nodes will increase the structural complexity and time complexity of the model, thereby affecting the rapidity and real-time performance of the model, reducing the model prediction accuracy, and further reducing the working efficiency.

[0004] Therefore, there is an urgent need to provide a method and system for predicting pollutant concentration during solid waste incineration to solve the above-mentioned technical problems. Summary of the Invention

[0005] This application provides a method and system for predicting pollutant concentrations during solid waste incineration. It can quickly construct a prediction model of SCN while ensuring the accuracy of the model, achieving accurate and rapid prediction of pollutant concentrations, thereby laying a foundation for the optimal control of the solid waste incineration process and improving work efficiency.

[0006] In the first aspect, this application provides a method for predicting pollutant concentrations during solid waste incineration. The method includes: collecting historical data of relevant process variables that affect the change of pollutant concentrations during solid waste incineration to obtain a training set; establishing a supervision mechanism based on the training set according to the hidden layer of the random configuration network to obtain hidden layer nodes that meet the supervision mechanism; determining the output weights of the random configuration network based on the obtained hidden layer nodes that meet the supervision mechanism; constructing a random configuration network prediction model based on the obtained output weights to realize the prediction of pollutant concentrations during solid waste incineration.

[0007] Optionally, the collecting historical data of relevant process variables that affect the change of pollutant concentrations during solid waste incineration to obtain a training set includes: collecting characteristic variable data of NO x , SO 2 , HCL, and CO, as well as the variable data of NO x , SO 2 , HCL, and CO at time K - 1, and using them as input variables; collecting the variable data of NO x , SO 2 , HCL, and CO at time K as output variables; forming a training set D with a sample size of N according to the obtained input variables and output variables, which is described by formula (1).

[0008]

[0009] In the formula, D is the training set with a sample size of N, X represents the input value of the training set, and Y is the output value of the pollutant concentration as the training set.

[0010] Optionally, the collecting historical data of relevant process variables that affect the change of pollutant concentrations during solid waste incineration to obtain a training set further includes: performing normalization processing on the obtained training set, which is described by formula (2).

[0011]

[0012] In the formula, m = 1, 2, …, M + 1; n = 1, 2, …, N; x n,m represents the value of the m-th characteristic variable of the n-th sample after normalization processing; the normalized training set is described by formula (3).

[0013]

[0014] Optionally, before establishing a supervision mechanism based on the training set according to the hidden layer of the random configuration network to obtain hidden layer nodes that meet the supervision mechanism, it further includes: setting initial parameters for the random configuration network; wherein, the initial parameters include the maximum number of hidden layer nodes L max , the maximum number of configurations T max , the tolerance error ε, the hidden layer parameter configuration range Υ = [λ min :Δλ:λ max , initializing the residual ε 0 = [y 1 ,..., y n T , setting the output matrix H of the hidden layer nodes to an empty set, given an empty set Ω to store the value ξ, and an empty set W to store the alternative node parameters w and b.

[0015] Optionally, establishing a supervision mechanism based on the training set according to the hidden layer of the random configuration network to obtain hidden layer nodes that meet the supervision mechanism includes: training the random configuration network according to the training set input, randomly selecting hidden layer parameters w k and b k in the interval [-λ, λ], and saving the alternative nodes w k and b k that meet the supervision mechanism to the candidate hidden layer node set W, and the supervision mechanism is described by formula (4):

[0016]

[0017] wherein, g L represents the output of the L-th node in the hidden layer; 0 < ||g|| < b g , b g ∈ R + ; e L-1 is the network residual, r = 1 / (1+(α / L max ) L ) ∈ (0, 1), α = 1, 2,..., L max , L max is the maximum number of hidden layer nodes, and the non-negative sequence {u L} satisfies lim L→+∞ u L = 0 and u L ≤ (1 - r); if the randomly selected hidden layer parameters w k and b k do not meet the supervision mechanism, that is, W is an empty set, repeat the above steps to regenerate the hidden layer parameters, and then make a judgment on the supervision mechanism until a candidate set of nodes that meets the supervision mechanism is generated.

[0018] ​Optionally, a supervision mechanism is established based on the training set according to the hidden layer of the random configuration network to obtain hidden layer nodes that conform to the supervision mechanism. It further includes: calculating ξ of the candidate nodes that meet the conditions according to the candidate hidden layer node set W, and saving it to the empty set Ω. The ξ value is described by formulas (5) and (6). L value, and save it to the empty set Ω. The ξ L value is described by formulas (5) and (6):

[0019]

[0020]

[0021] Select the maximum value of ξ from the obtained Ω set, and the candidate node corresponding to this value is used as the k-th hidden layer node. Calculate the output value h of the corresponding node according to the k-th hidden layer node and save it to the output matrix H of the hidden layer nodes. The output value h is described by formula (7): L h

[0022] h k =[s(x 1 ; w k , b k ),..., g(x n ; w k , b k )] T (7)

[0023] where g(·) is the Sigmoid activation function.

[0024] Optionally, constructing a random configuration network prediction model based on the obtained output weights to achieve the prediction of pollutant concentration in the solid waste incineration process includes: calculating the output weights of the random configuration network using the QR decomposition method according to the hidden layer nodes; determining whether the obtained output weights meet the tolerance error; if so, constructing a random configuration network prediction model based on the output weights.

[0025] Optionally, calculating the output weights of the random configuration network using the QR decomposition method according to the number of hidden layer nodes includes: judging the number of hidden layer nodes. If the number of hidden layer nodes is equal to 1, calculate the network output weights using the QR decomposition method, which is described by formulas (8) and (9):

[0026] H = QR (8)

[0027]

[0028] where represents the hidden layer output weights; H = [g 1 , g 2 , …, g Lrepresents the output matrix of the hidden layer; represents the generalized inverse matrix of HL; if the number of hidden layer nodes is greater than 1, assume the added hidden layer node is the (k + 1)-th hidden layer node, and introduce auxiliary variables to update and calculate the output weights of the network.

[0029] Optionally, if the number of hidden layer nodes is greater than 1, assume the added hidden layer node is the (k + 1)-th hidden layer node, and introduce auxiliary variables to update and calculate the output weights of the network, including: introducing auxiliary variables to update Q, R, and R -1 to iteratively update and calculate the output weights, where the auxiliary variables are described by formulas (10), (11), (12), and (13), and the update of Q, R, and R -1 is described by formulas (14), (15), and (16):

[0030] H k+1 =[Q k ·R k |R k+1 (10)

[0031] c = Q k T ·h k+1 (11)

[0032]

[0033] b=(h k+1 -Q k ·c) / d (13),

[0034] Q k+1 ==[Q k |b] (14)

[0035]

[0036]

[0037] In the formula, H k+1 represents the output matrix of the (k + 1) hidden layer nodes, c, d, and b represent auxiliary variables;

[0038] According to the update of Q, R, and R -1 , the output weights are obtained, which is described by formula (17):

[0039]

[0040] where, α k+1 =d -1 ·b T ·Y.

[0041] Second aspect, the present application provides a system for predicting pollutant concentration during municipal solid waste incineration, which includes: a collection module, configured to collect historical data of relevant process variables that affect the change of pollutant concentration during solid waste incineration to obtain a training set; a data processing module, configured to establish a supervision mechanism based on the training set according to the hidden layer of the random configuration network to obtain hidden layer nodes that conform to the supervision mechanism; a determination module, configured to determine the output weight of the random configuration network based on the obtained hidden layer nodes that conform to the supervision mechanism; a construction module, configured to construct a random configuration network prediction model based on the obtained output weight to realize the prediction of pollutant concentration during solid waste incineration.

[0042] Third aspect, the present application further provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the above-mentioned method are implemented.

[0043] Fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned method are implemented.

[0044] The present application has at least the following advantages:

[0045] In the present application, historical data of relevant process variables that affect the change of pollutant concentration during solid waste incineration are collected to obtain a training set. The training set is used as input data. According to the random configuration network algorithm and the constructed supervision mechanism constraints, hidden layer nodes and hidden layer parameters that conform to the supervision mechanism are obtained. Furthermore, by constructing a new supervision mechanism, the configuration efficiency of the input weights and biases of the hidden layer is accelerated, and the transformation matrix is updated by combining QR decomposition and the SCN incremental mechanism, and the output weights are calculated iteratively, reducing the computational complexity of SCN model training. Thus, on the premise of ensuring the accuracy of the model, a prediction model of SCN can be quickly constructed to realize accurate and rapid prediction of pollutant concentration, solve the problem of the lag of measuring instruments, and further reduce the equipment procurement and maintenance costs, laying a foundation for the optimal control of the solid waste incineration process and improving work efficiency. Description of the Drawings

[0046] Figure 1 It is an application environment diagram showing the method for predicting pollutant concentration during solid waste incineration in an embodiment;

[0047] Figure 2 It is a flowchart showing the method for predicting pollutant concentration during solid waste incineration in an embodiment;

[0048] Figure 3 It is a flowchart showing the configuration of hidden layer nodes according to the supervision mechanism in an embodiment;

[0049] Figure 4 It is a schematic flow chart showing the calculation of the output value of the corresponding node according to the hidden layer nodes in an embodiment;

[0050] Figure 5 It is a schematic flow chart showing the determination of the output weight of the random configuration network according to the hidden layer nodes in an embodiment;

[0051] Figure 6 It is a schematic flow chart showing the calculation of the output weight of the random configuration network by using the QR decomposition method in an embodiment;

[0052] Figure 7 It is a schematic flow chart showing the obtaining of the output weight according to the update of Q, R, and R -1 in an embodiment;

[0053] Figure 8 It is a schematic diagram showing the comparison of NO x concentration prediction results in an embodiment;

[0054] Figure 9 It is a schematic diagram showing the comparison of SO 2 concentration prediction results in an embodiment;

[0055] Figure 10 It is a schematic diagram showing the comparison of HCL concentration prediction results in an embodiment;

[0056] Figure 11 It is a schematic diagram showing the comparison of CO concentration prediction results in an embodiment;

[0057] Figure 12 It is a schematic block diagram of a pollutant concentration prediction system in the process of municipal solid waste incineration in an embodiment;

[0058] Figure 13 It is a schematic structural diagram of a computer device in an embodiment. Specific embodiments

[0059] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0060] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0061] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will elaborate on each embodiment of this application in conjunction with the accompanying drawings. However, those of ordinary skill in the art can understand that in each embodiment of this application, many technical details are presented to help readers better understand this application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in this application can still be implemented. The division of the following embodiments is for convenience of description and should not constitute any limitation to the specific implementation of this application. The various embodiments can be combined with each other and cross-referenced on the premise of not being contradictory.

[0062] For ease of understanding, the system applicable to this application is first described. A method for predicting pollutant concentration in the solid waste incineration process provided by this application can be applied to a system architecture as Figure 1 shown. The system includes: a user space file server 103 and a terminal device 101. The terminal device 101 communicates with the user space file server 103 through a network. Among them, the user space file server 103 can be a file server based on the NFSv3\v4 protocol, running in a Linux environment. NFS (Network File System) is a network abstraction on top of the file system, allowing remote clients running on the terminal device 101 to access through the network in a manner similar to the local file system. The terminal device 101 can be, but is not limited to, various personal computers, laptop computers, smartphones, tablet computers, etc. The user space file server 103 can be implemented by an independent server or a server cluster composed of multiple servers.

[0063] Figure 2 The flowchart of a method for predicting pollutant concentration in the solid waste incineration process provided by the embodiments of this application. This method can be executed by the user space file server in the system as Figure 1 shown. As Figure 2 shown, this method can include the following steps:

[0064] S201. Collect historical data of relevant process variables affecting the change of pollutant concentration in the solid waste incineration process to obtain a training set;

[0065] S202. Based on the training set, establish a supervision mechanism according to the hidden layer of the random configuration network to obtain hidden layer nodes that meet the supervision mechanism;

[0066] S203. Determine the output weight of the random configuration network based on the obtained hidden layer nodes that meet the supervision mechanism;

[0067] S204. Construct a random configuration network prediction model based on the obtained output weight to achieve the prediction of pollutant concentration in the solid waste incineration process.

[0068] Based on the training process of the above steps, a random configuration network prediction model for predicting pollutant concentrations in the urban solid waste incineration process can be obtained. Through the random configuration network prediction model, the rapid and accurate prediction of its change trend can be realized.

[0069] The following will specifically expand and elaborate on each step:

[0070] Refer to Figure 2 As shown, in S201, historical data of relevant process variables that affect the change of pollutant concentrations in the solid waste incineration process are collected to obtain a training set;

[0071] In this embodiment, it should be noted that historical data of characteristic variables that affect pollutants and relevant process variables of corresponding pollutants are collected, and the above-mentioned variable historical data collected are preprocessed to unify all variable historical data for subsequent direct analysis and processing. The sample data can come from 1000 data generated during the incineration process of a certain urban solid waste incineration power plant, where the number of training set samples is 800 and the number of test set samples is 200.

[0072] In some embodiments, characteristic variables that affect NOx, SO 2 , HCL, and CO, as well as the variable data of NOx, SO 2 , HCL, and CO at time K - 1 are collected and used as input variables; the variable data of NOx, SO 2 , HCL, and CO at time K are collected and used as output variables; a training set D with a sample size of N is formed according to the obtained input variables and output variables, which is described by formula (1).

[0073]

[0074] In the formula, D is a training set with a sample size of N, X represents the input value of the training set, and Y is the pollutant concentration as the output value of the training set.

[0075] Specifically, the collected characteristic variables and NO at time K - 1 x , SO 2, The variable data of HCL and CO are shown in Table 1, including the cumulative flow of urea solvent supply, the cumulative feed amount of activated carbon storage bin, the cumulative amount of lime feeder, the flue gas temperature 1 on the left side of the primary combustion, the flue gas temperature 1 in the middle of the primary combustion chamber, the flue gas temperature 1 on the right side of the primary combustion, the flue gas temperature 2 on the left side of the primary combustion, the flue gas temperature 2 in the middle of the primary combustion chamber, the flue gas temperature 2 on the right side of the primary combustion, the flue gas temperature 3 on the left side of the primary combustion, the flue gas temperature 3 in the middle of the primary combustion chamber, the flue gas temperature 3 on the right side of the primary combustion, the temperature on the right side of the primary combustion chamber, the temperature on the left side of the primary combustion chamber, the temperature on the left inner side of the drying section grate, the temperature on the left outer side of the drying section grate, the temperature on the right inner side of the drying section grate, the temperature on the right outer side of the drying section grate, the temperature on the right inner side of the furnace wall between the drying section and the combustion section grate, the temperature on the right outer side of the furnace wall between the drying section and the combustion section grate, the temperature on the left outer side of the combustion section grate 2-1, the temperature on the right inner side of the combustion section grate 1-2, the furnace negative pressure, the flue gas pressure at the outlet of the induced draft fan, the flue gas pressure at the outlet of the FGD, the differential pressure A of the bag filter, the differential pressure B of the bag filter, the NOx concentration at the previous moment, the SO 2 concentration, the HCL concentration at the previous moment, the CO concentration at the previous moment, and the historical data of 31 variables. From this, the change trend of pollutant concentration can be accurately and quickly predicted based on input variables such as the cumulative flow of urea solvent supply, the cumulative feed amount of activated carbon storage bin, and the cumulative amount of lime feeder, so as to quickly understand the incineration situation and lay a foundation for the optimal control of the incineration process.

[0076] Table 1 Variable Details

[0077] Serial number Variable name Serial number Variable name <![CDATA[x 1 > Cumulative urea solvent supply flow rate (L) <![CDATA[x 17 > Temperature of the right inner side of the grate in the drying section (℃) <![CDATA[x 2 > Cumulative activated carbon storage bin feeding amount (Kg) <![CDATA[x 18 > Temperature of the right outer side of the grate in the drying section (℃) <![CDATA[x 3 > Cumulative lime feeder amount (Kg) <![CDATA[x 19 > Temperature of the right inner side of the furnace wall between the drying section and the combustion section grates (℃) <![CDATA[x 4 > Temperature of the flue gas on the left side of the primary combustion 1 (℃) <![CDATA[x 20 > Temperature of the right outer side of the furnace wall between the drying section and the combustion section grates (℃) <![CDATA[x 5 > Temperature of the flue gas in the middle of the primary combustion chamber 1 (℃) <![CDATA[x 21 > Temperature of the left outer side of the combustion section grate 2-1 (℃) <![CDATA[x 6 > Temperature of the flue gas on the right side of the primary combustion chamber 1 (℃) <![CDATA[x 22 > Temperature of the right inner side of the combustion section grate 1-2 (℃) <![CDATA[x 7 > Temperature of the flue gas on the left side of the primary combustion 2 (℃) <![CDATA[x 23 > Furnace negative pressure (Pa) <![CDATA[x 8 > Temperature of the flue gas in the middle of the primary combustion chamber 2 (℃) <![CDATA[x 24 > Flue gas pressure at the outlet of the induced draft fan (Pa) <![CDATA[x 9 > Temperature of the flue gas on the right side of the primary combustion chamber 2 (℃) <![CDATA[x 25 > <![CDATA[FGD outlet flue gas pressure ( KPa )]]> <![CDATA[x 10 > Temperature of the flue gas on the left side of the primary combustion 3 (℃) <![CDATA[x 26 > Differential pressure A of the bag filter (Pa) <![CDATA[x 11 > Temperature of the flue gas in the middle of the primary combustion chamber 3 (℃) <![CDATA[x 27 > Differential pressure B of the bag filter (Pa) <![CDATA[x 12 > Temperature of the flue gas on the right side of the primary combustion chamber 3 (℃) <![CDATA[x 28 > <![CDATA[Previous moment NO x Concentration (mg / m 3 N)]]> <![CDATA[x 13 > Temperature of the right side of the primary combustion chamber (℃) <![CDATA[x 29 > <![CDATA[Previous moment SO 2 Concentration (mg / m 3 N)]]> <![CDATA[x 14 > Temperature of the left side of the primary combustion chamber (℃) <![CDATA[x 30 > <![CDATA[HCL concentration at the previous moment (mg / m 3 N)]]> <![CDATA[x 15 > Temperature of the left inner side of the grate in the drying section (℃) <![CDATA[x 31 > <![CDATA[CO concentration at the previous moment (mg / m 3 N)]]> <![CDATA[x 16 > Temperature of the left outer side of the grate in the drying section (℃)

[0078] Refer to Figure 2 As shown, in some embodiments, in S201, the historical data of relevant process variables affecting the change of pollutant concentration in the solid waste incineration process is collected to obtain a training set, and it also includes: normalizing the obtained training set, which is described by formula (2),

[0079]

[0080] In the formula, m = 1, 2, …, M + 1; n = 1, 2, …, N; x n,m represents the value of the m-th feature variable of the n-th sample after normalization; the normalized training set is described by formula (3),

[0081]

[0082] By normalizing the training set, the influence of the dimension of different variables can be eliminated.

[0083] Refer to Figure 2As shown, in some embodiments, in S202, before establishing a supervision mechanism based on the training set according to the hidden layer of the randomly configured network to obtain hidden layer nodes that meet the supervision mechanism, it further includes:

[0084] Performing initial parameter setting on the randomly configured network; the initial parameters include the maximum number of hidden layer nodes L max , the maximum number of configuration times T max , the tolerance error ε, the hidden layer parameter configuration range Υ = [λ min :Δλ:λ max , initializing the residual ε 0 = [y 1 ,..., y n T , setting the output matrix H of the hidden layer nodes to an empty set, given an empty set Ω to store the value ξ, and an empty set W to store the alternative node parameters w and b;

[0085] In this embodiment, it should be noted that according to the SCN structure, the parameters are first initialized. In one example, by setting the maximum number of hidden layer nodes L of the randomly configured network max = 200, the maximum number of configuration times T max = 100, the tolerance error ε = 0.01, the hidden layer parameter configuration range Initializing the residual ε 0 = [y 1 ,..., y 800 T , setting the output matrix H of the hidden layer nodes to an empty set, given an empty set Ω to store the value ξ, and an empty set W to store the alternative node parameters w and b, so as to use the training set data as the input of the SCN algorithm, train the SCN, and obtain appropriate hidden layer nodes according to the constraints of the supervision mechanism.

[0086] Referring to Figure 2 、 Figure 3 shown, S202. Establishing a supervision mechanism based on the training set according to the hidden layer of the randomly configured network to obtain hidden layer nodes that meet the supervision mechanism;

[0087] ​​In this embodiment, it should be noted that the Stochastic Configuration Network (SCN) is a commonly used neural network model. Its basic idea is to randomly initialize the network weights and biases, use the stochastic gradient descent algorithm for optimization and adjustment, and make predictions through the trained model. Its structure includes an input layer, a hidden layer, and an output layer, mainly composed of several nodes, and each node represents a random variable. The connections between these nodes have random weights, which can be used to describe the dependence relationships between different nodes. The SCN model takes the input sequence as the values of random variables and uses random parameters for fitting and prediction. Therefore, by establishing a supervision mechanism to configure the hidden layer nodes, the configuration efficiency of the input weights and biases of the hidden layer can be accelerated, thereby improving the speed and real-time performance of the constructed model.

[0088] Step S202 specifically includes:

[0089] S2021. Train the randomly configured network according to the training set input, randomly select the hidden layer parameters w k and b k in the interval [-λ, λ], and save the alternative nodes w k and b k that meet the supervision mechanism to the candidate hidden layer node set W. The supervision mechanism is described by formula (4):

[0090]

[0091] where g L represents the output of the L-th node in the hidden layer; 0 < ||g|| < b g , b g ∈R + ; e L-1 is the network residual; r = 1 / (1+(α / L max ) L ) ∈ (0, 1), α = 1, 2, …, L max , L max is the maximum number of nodes in the hidden layer, and the non-negative sequence {u L} satisfies lim L→+∞ u L = 0 and u L ≤ (1 - r).

[0092] S2022. If the randomly selected hidden layer parameters w k and b k do not meet the supervision mechanism, that is, W is an empty set;

[0093] S2023. Repeat the above steps to regenerate the hidden layer parameters and then make a judgment on the supervision mechanism until a candidate set of nodes that meets the supervision mechanism is generated.

[0094] In this embodiment, it should be noted that the SCN model takes the input sequence as the value of a random variable and uses random parameters for fitting and prediction. Among them, according to the setting of the supervision mechanism, the hidden layer nodes are configured more reasonably. Since the weights between the input layer and the hidden nodes will affect the values from all hidden nodes to the output layer, the weight gradient information at this time should have an accumulative effect on the error between the hidden layer and the output layer. Therefore, configuring the hidden layer nodes well can accelerate the configuration efficiency of the input weights and biases of the hidden layer, thereby improving the speed and real-time performance of the constructed model.

[0095] Specifically, when configuring the k-th node of the hidden layer, randomly select the hidden layer parameters w in the interval [-10, 10] k and b k , and then according to the constraints of the preset supervision mechanism, discard the hidden layer parameters that do not meet the supervision mechanism, and save the alternative nodes w k and b k that meet the supervision mechanism to W to form a candidate node set.

[0096] In some embodiments, referring to Figure 2 、 Figure 4 shown, in S202, based on the training set, a supervision mechanism is established according to the hidden layer of the randomly configured network to obtain the hidden layer nodes that meet the supervision mechanism, and further includes:

[0097] S2024. Calculate the ξ L value of the candidate set nodes that meet the conditions according to the candidate hidden layer node set W, and save it to the empty set Ω. The ξ L value is described by formulas (5) and (6):

[0098]

[0099] ;

[0100] S2025. Select the maximum value of ξ L from the obtained Ω set, and the candidate node corresponding to this value is used as the k-th hidden layer node;

[0101] S2026. Calculate the output value h of the corresponding node according to the k-th hidden layer node and save it to the output matrix H of the hidden layer nodes. The output value h is described by formula (7):

[0102] h k =[s(x 1 ; w k , b k ),..., g(x n ; w k , b k )] T (7)

[0103] Among them, g(·) is the Sigmoid activation function.

[0104] In this embodiment, it should be noted that according to the candidate set nodes formed by the hidden layer parameters obtained by the supervision mechanism, calculate the ξ of the corresponding candidate set nodes that meet the conditions L value and find the largest ξ L value among all the ξ L value, and use the candidate node corresponding to this value as the k-th hidden layer node. g(·) is the Sigmoid activation function. According to the obtained hidden layer nodes and in cooperation with the activation function, obtain the output values of the corresponding hidden layer nodes, providing a data basis for subsequent calculations.

[0105] Refer to Figure 2 、 Figure 5 As shown, S203. Determine the output weights of the random configuration network based on the obtained hidden layer nodes that meet the supervision mechanism;

[0106] In this embodiment, it should be noted that from the perspective of the accuracy of the pollutant emission concentration prediction model for the incineration process, according to the reasonable configuration of the hidden layer nodes, and then calculate the output weights based on the hidden layer nodes, so as to construct a fast, real-time and efficient prediction model.

[0107] Step S203 specifically includes:

[0108] S2031. Calculate the output weights of the random configuration network using the QR decomposition method according to the hidden layer nodes;

[0109] S2032. Determine whether the obtained output weights meet the tolerance error;

[0110] S2033. If so, construct a random configuration network prediction model based on the output weights.

[0111] In this embodiment, it should be noted that the QR (orthogonal triangular) decomposition method is the most effective and widely used method for finding all the eigenvalues of a general matrix. The general matrix is first orthogonally similar transformed into a Hessenberg matrix, and then the QR method is used to find the eigenvalues and eigenvectors. It decomposes the matrix into a normal orthogonal matrix Q and an upper triangular matrix R, so it is called the QR decomposition method, which is related to the general symbol Q of this normal orthogonal matrix.

[0112] Quickly obtain the output weights according to the QR algorithm decomposition. Compare the obtained output weights with the tolerance error and confirm. If the tolerance error is met, obtain the prediction model based on the selected hidden layer parameters and output weights.

[0113] In some embodiments, refer to Figure 5 、 Figure 6As shown, in S2031, according to the number of hidden layer nodes, the output weights of the random configuration network are calculated using the QR decomposition method, including:

[0114] S20311. Determine the number of hidden layer nodes. If the number of hidden layer nodes is equal to 1, use the QR decomposition method to calculate the network output weights, which are described by equations (8) and (9):

[0115] H = QR (8)

[0116]

[0117] Among them, represents the output weights of the hidden layer; H = [g 1 , g 2 , …, g L represents the output matrix of the hidden layer; represents the generalized inverse matrix of HL;

[0118] S20312. If the number of hidden layer nodes is greater than 1, set the added hidden layer node as the (k + 1)-th hidden layer node, and introduce auxiliary variables to update and calculate the network output weights.

[0119] In this embodiment, it should be noted that to determine whether the number of hidden layer nodes is greater than 1. In the case of being equal to 1, the output weights of the SCN model can be directly calculated according to the QR decomposition method. If the number of hidden layer nodes is greater than 1, then auxiliary variables need to be introduced to update Q, R, and R -1 so as to update the output weights, reducing the computational complexity of SCN model training.

[0120] In some embodiments, as shown in Figure 6 、 Figure 7 In S20312, if the number of hidden layer nodes is greater than 1, set the added hidden layer node as the (k + 1)-th hidden layer node, and introduce auxiliary variables to update and calculate the network output weights, including:

[0121] S203121. Introduce auxiliary variables to update Q, R, and R -1 to iteratively update and calculate the output weights. Among them, the auxiliary variables are described by formulas (10), (11), (12), and (13), and the update of Q, R, and R -1 is described by formulas (14), (15), and (16):

[0122] H k+1 = [Q k ·R k |h k+1 (10)

[0123] c = Q kT ·h k+1 (11)

[0124]

[0125] b = (h k+1 -Q k ·c) / d (13),

[0126] Q k+1 == Q k|b] (14)

[0127]

[0128]

[0129] Wherein, H k+1 represents the output matrix of k + 1 hidden layer nodes, c, d and b represent auxiliary variables;

[0130] S203122. According to the updates of Q, R and R -1 obtain the output weights, which are described by formula (17):

[0131]

[0132] where, α k+1 = d -1 ·b T ·Y.

[0133] In this embodiment, it should be noted that when the number of hidden layer nodes is greater than 1, the auxiliary variables introduced are used to update Q, R and R -1 are updated, and the QR decomposition and the SCN increment mechanism are combined to update the transformation matrix, and the output weights are calculated iteratively, so as to reduce the computational complexity of the SCN model training, and then construct a pollutant concentration prediction model to solve the problem of the lag of the measuring instrument.

[0134] S204. Based on the obtained output weights, construct a random configuration network prediction model to realize the prediction of pollutant concentration in the solid waste incineration process.

[0135] In this embodiment, it should be noted that the random configuration network prediction model is constructed by using the output weights obtained by the above operations, and the prediction of pollutant concentration in the solid waste incineration process is realized through the constructed random configuration network prediction model.

[0136] In some embodiments, the method for predicting pollutant concentration in the solid waste incineration process further includes: collecting historical data of relevant process variables that affect the change of pollutant concentration in the solid waste incineration process to obtain a test set; using the test set as input data of a prediction model to obtain output data to verify the prediction accuracy of the prediction model.

[0137] In this embodiment, it should be noted that the accuracy of the prediction model is evaluated by calculating the root mean square error (RMSE) and the mean absolute percentage error (MAPE); the smaller the RMSE and MAPE, the higher the prediction accuracy.

[0138] Specifically, referring to Figures 8 - 11 As shown, the prediction model is tested according to the test set, and the test results show that the root mean square errors of NO x , SO 2 , HCL, and CO are 4.9004, 1.7141, 0.1564, and 9.5265 respectively, and the mean absolute errors are 3.7370, 1.2641, 0.1183, and 6.5059 respectively, which can achieve accurate prediction of pollutant concentration; its training time is 0.8624 s, meeting the rapidity requirement for pollutant concentration prediction in the urban solid waste incineration process, indicating that the prediction model is conducive to accurate and rapid prediction of pollutant concentration and can lay a foundation for the optimal control of the solid waste incineration process.

[0139] The implementation principle of this embodiment: Collect historical data of relevant process variables that affect the change of pollutant concentration in the solid waste incineration process to obtain a training set. Through data preprocessing of the training set, a unified training set is obtained. Using the training set as input data, according to the random configuration network algorithm and the constructed supervision mechanism constraint, the hidden layer nodes and hidden layer parameters that meet the supervision mechanism are obtained. Furthermore, by constructing a new supervision mechanism, the configuration efficiency of the input weights and biases of the hidden layer is accelerated, and the transformation matrix is updated by combining QR decomposition with the SCN increment mechanism, and the output weights are iteratively calculated to reduce the computational complexity of SCN model training. Thus, on the premise of ensuring the accuracy of the model, the prediction model of SCN can be quickly constructed to achieve accurate and rapid prediction of pollutant concentration, solve the problem of the lag of measuring instruments, and further reduce the equipment procurement and maintenance costs, lay a foundation for the optimal control of the solid waste incineration process, and improve work efficiency.

[0140] Referring to Figure 12 As shown, the present application also provides a system for predicting pollutant concentration in the urban solid waste incineration process. The system may include: a collection module, a data processing module, and a model construction module. The main functions of each component module are as follows:

[0141] The collection module 301 is used to collect historical data of relevant process variables that affect the change of pollutant concentration in the solid waste incineration process to obtain a training set;

[0142] A data processing module 302, configured to establish a supervision mechanism based on a training set according to hidden layers of a random configuration network, so as to obtain hidden layer nodes conforming to the supervision mechanism;

[0143] A determination module 303, configured to determine output weights of the random configuration network based on the obtained hidden layer nodes conforming to the supervision mechanism;

[0144] A construction module 304, configured to construct a random configuration network prediction model based on the obtained output weights, so as to implement prediction of pollutant concentrations in the solid waste incineration process.

[0145] According to an embodiment of the present application, the present application further provides a computer device and a computer-readable storage medium.

[0146] As Figure 13 shown, it is a block diagram of a computer device according to an embodiment of the present application. The computer device is intended to represent various forms of digital computers or mobile devices. Among them, the digital computer may include a desktop computer, a portable computer, a workbench, a personal digital assistant, a server, a mainframe computer, and other suitable computers. The mobile device may include a tablet computer, a smart phone, a wearable device, etc.

[0147] As Figure 13 shown, the device 600 includes a computing unit 601, a ROM 602, a RAM 603, a bus 604, and an input / output (I / O) interface 605. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other through the bus 604. The input / output (I / O) interface 605 is also connected to the bus 604.

[0148] The computing unit 601 may execute various processes in the method embodiments of the present application according to computer instructions stored in the read-only memory (ROM) 602 or computer instructions loaded from a storage unit 608 into the random access memory (RAM) 603. The computing unit 601 may be various general-purpose and / or dedicated processing components having processing and computing capabilities. The computing unit 601 may include, but is not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. In some embodiments, the method provided by the embodiments of the present application may be implemented as a computer software program, which is tangibly included in a computer-readable storage medium, such as the storage unit 608.

[0149] The RAM 603 can also store various programs and data required for the operation of the device 600. Part or all of the computer programs can be loaded and / or installed onto the device 600 via the ROM 602 and / or the communication unit 609.

[0150] The input unit 606, output unit 607, storage unit 608, and communication unit 609 in the device 600 can be connected to the I / O interface 605. Among them, the input unit 606 can be, for example, a keyboard, mouse, touch screen, microphone, etc.; the output unit 607 can be, for example, a display, speaker, indicator light, etc. The device 600 can exchange information, data, etc. with other devices through the communication unit 609.

[0151] It should be noted that the device may also include other components necessary for normal operation. It may also only include the components necessary to implement the solution of this application, and does not necessarily include all the components shown in the figure.

[0152] The various embodiments of the systems and technologies described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof.

[0153] The computer instructions for implementing the methods of this application can be written in any combination of one or more programming languages. These computer instructions can be provided to the computing unit 601, such that when the computer instructions are executed by the computing unit 601, such as a processor, the steps involved in the method embodiments of this application are executed.

[0154] The computer-readable storage medium provided by this application can be a tangible medium that can contain or store computer instructions for executing the steps involved in the method embodiments of this application. The computer-readable storage medium can include, but is not limited to, storage media in the forms of electronic, magnetic, optical, electromagnetic, etc.

[0155] The above specific embodiments do not constitute a limitation on the protection scope of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. A method for predicting pollutant concentration in a solid waste incineration process, characterized in that: The following steps are involved: Collect historical data of relevant process variables that affect the change of pollutant concentration in the solid waste incineration process to obtain a training set; Establishing a supervision mechanism according to the hidden layer of the randomly configured network based on the training set to obtain hidden layer nodes that meet the supervision mechanism; Determining the output weights of the randomly configured network based on the obtained hidden layer nodes that conform to the supervision mechanism; A random configuration network prediction model is constructed based on the obtained output weights to predict the pollutant concentration in the solid waste incineration process.

2. The method for predicting pollutant concentration in solid waste incineration process according to claim 1, characterized in that: The historical data of relevant process variables that affect the change of pollutant concentration in the solid waste incineration process are collected to obtain a training set, including: Collection impact NO x , SO2, HCL and CO characteristic variable data and NO at K-1 time x , SO2, HCL and CO variable data and use them as input variables; Collect NO at time K x , SO2, HCL and CO variable data as output variables; A training set D with a sample size of N is formed according to the input variables and the output variables obtained, which is described by formula (1): Where D is the training set with a sample size of N, X represents the input value of the training set, and Y is the pollutant concentration as the output value of the training set.

3. The method for predicting pollutant concentration in solid waste incineration process according to claim 2 is characterized in that: The collecting of historical data of relevant process variables that affect the change of pollutant concentration in the solid waste incineration process to obtain a training set also includes: The obtained training set is normalized and described by formula (2): Where, m = 1, 2, ..., M + 1; n = 1, 2, ..., N; x n,m Represents the mth characteristic variable value of the nth sample after normalization; The normalized training set is described by formula (3):

4. The method for predicting pollutant concentration in solid waste incineration process according to any one of claims 1 to 3, characterized in that: Before establishing a supervision mechanism based on the training set according to the hidden layer of the randomly configured network to obtain hidden layer nodes that meet the supervision mechanism, the method further includes: Set initial parameters for the random configuration network; The initial parameters include the maximum number of nodes in the hidden layer L max , the maximum number of configuration times T max , tolerance error ε, hidden layer parameter configuration range Υ=[λ min :Δλ:λ max ], initialize the residual ε0=[y1,...,y n ] T , the output matrix H of the hidden layer nodes is set to the empty set, the empty set Ω is given to store the value ξ, and the empty set W stores the candidate node parameters w and b.

5. The method for predicting pollutant concentration in solid waste incineration process according to claim 4, characterized in that: The step of establishing a supervision mechanism based on the training set and randomly configuring the hidden layer of the network to obtain hidden layer nodes that conform to the supervision mechanism includes: According to the training set input, the network is randomly configured for training, and the hidden layer parameter w is randomly selected in the interval [-λ,λ] k and b k , and the candidate node w that satisfies the supervision mechanism k and b k Save to the candidate hidden layer node set W, the supervision mechanism is described by formula (4): Among them, g L represents the output of the Lth node in the hidden layer; 0<||g|| g , b g ∈R + ;e L-1 is the network residual; r = 1 / (1+(α / L max ) L )∈(0,)1,α=1,2,…,L ma x , L max is the maximum number of nodes in the hidden layer, the non-negative sequence {u L }Satisfy lim L→+∞ u L =0 and u L ≤(1-r);​ If the hidden layer parameter w is randomly selected k and b k The supervision mechanism is not satisfied, that is, W is an empty set; Repeat the above steps to regenerate the hidden layer parameters, and then make the judgment of the supervision mechanism until the candidate set nodes that meet the supervision mechanism are generated.

6. The method for predicting pollutant concentration in solid waste incineration process according to claim 5, characterized in that: The step of establishing a supervision mechanism based on the training set and randomly configuring the hidden layer of the network to obtain hidden layer nodes that conform to the supervision mechanism also includes: According to the candidate hidden layer node set W, calculate the ξ of the candidate set nodes that meet the conditions L value, and save it to the empty set Ω, ξ L The value is described by formula (5) and (6): Select ξ from the obtained Ω set L The maximum value of , then the candidate node corresponding to this value is used as the kth hidden layer node; The output value h of the corresponding node is calculated according to the kth hidden layer node and saved in the output matrix H of the hidden layer node. The output value h is described by formula (7): h k =[g(x1;w k ,b k ),...,g(x n ;w k ,b k )] T (7) Among them, g(·) is the Sigmoid activation function.

7. The method for predicting pollutant concentration in solid waste incineration process according to claim 6, characterized in that: The randomly configured network prediction model is constructed based on the obtained output weights to realize the prediction of pollutant concentration in the solid waste incineration process, including: According to the hidden layer nodes, the QR decomposition method is used to calculate the output weights of the random configuration network; Determining whether the obtained output weight satisfies a tolerance error; If so, a random configuration network prediction model is constructed based on the output weights.

8. The method for predicting pollutant concentration in solid waste incineration process according to claim 7, characterized in that: The method of calculating the output weight of the random configuration network by using the QR decomposition method according to the number of hidden layer nodes includes: Determine the number of hidden layer nodes. If the number of hidden layer nodes is equal to 1, the QR decomposition method is used to calculate the network output weight, which can be described by equations (8) and (9): H=QR (8) in, represents the output weight of the hidden layer; H = [g1, g2, ..., g L ] represents the hidden layer output matrix; represents the generalized inverse matrix of HL; If the number of hidden layer nodes is greater than 1, the added hidden layer node is set as the k+1th hidden layer node, and auxiliary variables are introduced to update the network output weights.

9. The method for predicting pollutant concentration in solid waste incineration process according to claim 8, characterized in that: If the number of hidden layer nodes is greater than 1, the added hidden layer node is set as the k+1th hidden layer node, and an auxiliary variable is introduced to update the network output weight, including: Introduce auxiliary variables to Q, R and R -1 Update to iteratively update the calculated output weights, where the auxiliary variables are described by formulas (10), (11), (12), and (13) for Q, R, and R -1 The update is described by formulas (14), (15), and (16): H k+1 =[Q k ·R k |h k+1 ] (10) c=Q k T ·h k+1 (11) b=(h k+1 -Q k ·c) / d (13), Q k+1 ==[Q k |b] (14) In the formula, H k+1 represents the output matrix of k+1 hidden layer nodes, c, d and b represent auxiliary variables; According to the Q, R and R -1 The output weight is obtained by updating , which is described by formula (17): Among them, a k+1 =d -1 ·b T ·Y.

10. A pollutant concentration prediction system for the process of municipal solid waste incineration, characterized in that: The system comprises: A collection module, used to collect historical data of relevant process variables that affect the change of pollutant concentration in the solid waste incineration process to obtain a training set; A data processing module, used to establish a supervision mechanism according to the hidden layer of the randomly configured network based on the training set, so as to obtain hidden layer nodes that meet the supervision mechanism; A determination module, used for determining the output weights of the randomly configured network based on the obtained hidden layer nodes that conform to the supervision mechanism; A construction module is used to construct a random configuration network prediction model based on the obtained output weights to predict the pollutant concentration in the solid waste incineration process.

Images