A furnace temperature prediction method, system, device and medium for a solid waste incineration process

Abstract

The application relates to a furnace temperature prediction method, system, device and medium for a solid waste incineration process. The prediction method comprises the following steps: acquiring a plurality of groups of characteristic variables affecting furnace temperature and corresponding current furnace temperature values and corresponding historical data groups of next-time furnace temperature values to form case descriptions, and constructing a difference database D in the order of the case descriptions; performing data preprocessing based on the difference database D to obtain a training set; constructing a case difference prediction model based on a deep Q network algorithm according to the training set; obtaining case difference data of the furnace temperature according to the characteristic variables of the current furnace temperature, and taking the case difference data as input data of the case difference prediction model; and obtaining a predicted value of the next-time furnace temperature according to output data of the case difference prediction model, so that the furnace temperature change trend of the urban solid waste incineration process can be accurately predicted, the optimization control of the incineration process is ensured, and the work efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of solid waste incineration technology, and in particular to a method, system, equipment and medium for predicting furnace temperature in a solid waste incineration process. Background Technology

[0002] Solid waste incineration power generation not only reduces and renders waste harmless, but also utilizes the heat generated during incineration to produce electricity, thus realizing the resource recovery of waste. During incineration, furnace temperature is one of the key parameters for ensuring sufficient combustion of municipal solid waste and suppressing pollution. However, the lack of pre-incineration sorting of municipal solid waste makes it difficult to accurately predict temperature changes during the incineration process. Furthermore, the lag in the incineration process makes it difficult for operators to adjust their operations in a timely manner based on furnace temperature changes, leading to furnace temperatures easily falling within abnormal ranges. Therefore, establishing an accurate furnace temperature prediction model is fundamental to maintaining the stable and efficient operation of the solid waste incineration process.

[0003] Currently, methods for predicting furnace temperature in municipal solid waste incineration processes mainly include mechanistic modeling and data-driven prediction methods. Mechanistic modeling, starting from the reaction mechanism of the industrial process, establishes a mechanistic model of the controlled object based on the laws of energy conservation or material conservation. Parameter predictions are obtained through computational physical and chemical equations, offering good interpretability. However, the complexity of the incineration process, with its strong nonlinearity and coupling, makes it difficult to establish accurate mechanistic models. With advancements in sensor technology, a large amount of historical data on the production process has been preserved. This data implicitly contains operational patterns and information on process parameter changes, providing a guarantee for the establishment of data-driven models. Therefore, data-driven furnace temperature prediction methods have attracted considerable attention from researchers.

[0004] Data-driven modeling and prediction methods are mainly represented by machine learning methods such as support vector regression, back propagation neural networks, and stochastic configuration networks. However, due to the numerous and complex characteristic variables affecting furnace temperature during incineration, the above methods only obtain furnace temperature prediction results by learning the relationship between characteristic variables and furnace temperature changes. They do not consider the reference information provided by historical data on furnace temperature changes under the current incineration state, resulting in limited prediction accuracy. This makes it impossible to make timely judgments on furnace temperature changes based on the solid waste incineration process, affecting the optimized control of the operation process and reducing work efficiency. Summary of the Invention

[0005] This application provides a method, system, equipment, and medium for predicting furnace temperature in the solid waste incineration process, which aims to improve the accuracy of furnace temperature and flue gas oxygen content in the urban solid waste incineration process, provide a guarantee for the optimized control of the incineration process, and enable on-site operators to keep abreast of the incineration conditions of solid waste in the furnace, thereby improving work efficiency.

[0006] In a first aspect, this application provides a method for predicting furnace temperature in a solid waste incineration process. The method includes: acquiring multiple sets of characteristic variables affecting furnace temperature, their corresponding current furnace temperature values, and historical data of the furnace temperature values ​​at the next moment to form case descriptions; constructing a difference database D according to the order of the case descriptions; performing data preprocessing based on the difference database D to obtain a training set; constructing a case difference prediction model based on the training set using a deep Q-network algorithm; obtaining case difference data of the furnace temperature based on the characteristic variables of the current furnace temperature, and using this data as input data to the case difference prediction model; and obtaining the predicted value of the furnace temperature at the next moment based on the output data of the case difference prediction model.

[0007] Optionally, the step of obtaining multiple sets of characteristic variables affecting furnace temperature and their corresponding current furnace temperature values ​​and corresponding historical data of the next furnace temperature value to form a case description, and constructing a difference database D according to the case description order, includes: obtaining multiple sets of characteristic variables including but not limited to: grate speed, grate temperature, primary air flow, secondary air flow, blower pressure and their corresponding current furnace temperature values ​​as problem description features X of case c, and the furnace temperature value y of the next furnace temperature value as the solution description, together forming a case description, and taking N cases to construct a furnace temperature case library C, as shown in formula (1):

[0008]

[0009] Select a pair of cases c from the furnace temperature case library C in the order of their descriptions. i c j The case difference description is obtained through formula (2):

[0010] e = Δ(c) i ,c j )=(ΔX ij ,Δy ij (11);

[0011] Based on the obtained case difference descriptions, a case difference database D is constructed, containing N case differences. 2 It can be described by formula (3):

[0012]

[0013] Optionally, the data preprocessing based on the difference database D to obtain the training set includes: performing min-max standardization on the difference case problem features in the case difference database D, as described by formula (4):

[0014]

[0015] Where, ΔXk,m Let m be the feature variable in the k-th difference case, where k = 1, 2, ..., N. 2 N 2 The number of case differences is m = 1, 2, ..., M, where M represents the number of case features with differences, and here M = 66.

[0016] Optionally, the step of constructing a case difference prediction model based on the deep Q-network algorithm and the training set includes: setting the number of hidden layers L, the number of hidden layer nodes n, the learning rate lr, and the discount factor γ of the deep Q-network, and defining the relevant state space S, action space A, and reward function r, and initializing the parameters; wherein, the state space S is a series of state descriptions, which in this case are the problem features in the furnace temperature difference case, represented as S=[ΔX1,…,ΔX N 2 The action space is a series of action values ​​with intervals l, which here represents the solution size in the furnace temperature difference case, i.e., A = [Δy]. min ,Δy min +l,…,Δy max The reward function is described by formula (5):

[0017] r=-|Δy-a| (14),

[0018] In the formula, Δy represents the solution of the difference case, and a represents the action in action space A.

[0019] Optionally, the step of constructing a case difference prediction model based on the deep Q-network algorithm and the training set further includes: constructing a Q-network and a target Q-network, and defining the Q-function Q(s,a|θ) of the Q-network accordingly, which is described by formula (6):

[0020]

[0021] In the formula, θ represents the network weights, s represents the state in the state space S, and a represents the action in the action space A; the Q function corresponding to the target Q network is represented by Q'(s,a|θ'), where θ' represents the target network weights; the selection of action a is achieved according to the ε-greedy policy, which is described by formula (7):

[0022]

[0023] The training set is used as the input data for the Q(s,a|θ) network. Based on the reward values ​​corresponding to the training and learning of the Q network and the target Q network, a loss function L(θ) related to the reward value is set. According to the loss function L(θ), the parameters θ of the Q neural network are continuously updated until the loss function meets the set conditions, thus obtaining the case difference prediction model of furnace temperature.

[0024] Optionally, the training set is used as input data for the Q(s,a|θ) network, and a loss function L(θ) related to the reward value is set according to the reward value corresponding to the training and learning of the Q network and the target Q network. This includes setting the current state as s. t According to action a t Receive network rewards r t and enter the next state s t+1 and (s t ,a t ,r t ,s t+1 ) are stored in experience pool B; based on the samples (s) in experience pool B i ,a i ,r i ,s i+1 The target value t is calculated based on Q'(s,a|θ') of the target network. i It can be described by formula (8):

[0025] t i =r i +γmaxQ′(s t+1 ,a′|θ′) (16),

[0026] Among them, experience pool B is used to store the current state s. t Action a t Next state s t+1 And reward r t , a' represents the action value selected by the target network from the action space S; based on the constructed Q network and the target Q network, the loss function L(θ) is obtained by continuously learning and training according to the training set.

[0027] Optionally, the step of continuously updating the Q-neural network parameters θ according to the loss function L(θ) until the loss function meets the set conditions includes: updating the Q-network according to the Bellman equation to obtain the loss function L(θ), which is described by formula (9):

[0028] L i (θ i ) = E (s,a,r,s′) [(t i -Q(s i ,a i |θ i )) 2 ]

[0029] =E (s,a,r,s′) [(r i +γmaxQ′(s t+1 ,a′|θ′)-Q(s i ,ai |θ i )) 2 (17).

[0030] Optionally, the step of obtaining case difference data of furnace temperature based on the characteristic variables of the current furnace temperature and using it as input data for the case difference prediction model, and obtaining the predicted value of the furnace temperature at the next moment based on the output data of the case difference prediction model, includes: based on the characteristic variable X of the current furnace temperature... t The similarity measurement is calculated between the data and historical data in the furnace temperature case library C, and is described by formula (10):

[0031]

[0032] In the formula, X j M represents the problem characteristics of a case in the case library, where M is the number of feature variables; the most similar case c is retrieved based on similarity. xim =(X sim ,y sim ), the case difference data ΔX that constitutes the furnace temperature t,sim ; ΔX t,sim As input, the output action value a = Q(ΔX) is obtained through the constructed case difference prediction model. t,sim The value is determined, and the corresponding action value is selected from the action space as the case difference solution Δy. t,sim =A[a]; Based on the obtained case difference solution Δy t,sim The predicted value y of the furnace temperature is obtained and described by formula (11):

[0033] y = y t,sim +Δy t,sim (18).

[0034] Secondly, this application provides a system for predicting furnace temperature in a solid waste incineration process. The system includes: a data acquisition module, used to acquire multiple sets of characteristic variables affecting furnace temperature, corresponding current furnace temperature values, and historical data of the furnace temperature at the next moment, forming case descriptions, and constructing a difference database D according to the order of the case descriptions; a data preprocessing module, used to perform data preprocessing based on the difference database D to obtain a training set; a prediction model construction module, used to construct a case difference prediction model based on a deep Q-network algorithm; a training module, used to train the case difference prediction model according to the training set to obtain a case difference prediction model for furnace temperature; and a result prediction module, used to obtain case difference data of furnace temperature based on the characteristic variables of the current furnace temperature, and use this as input data for the case difference prediction model, and obtain the predicted value of the furnace temperature at the next moment based on the output data of the case difference prediction model.

[0035] Thirdly, this application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method described above when executing the computer program.

[0036] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method described above.

[0037] This application has at least the following advantages:

[0038] According to the technical content provided in the embodiments of this application, by collecting historical data on characteristic variables generated during the incineration process of solid waste incineration power plants, as well as the corresponding current furnace temperature value and the furnace temperature value at the next moment, case descriptions are formed. A difference database D is constructed according to the order of the case descriptions to obtain a training set. The adaptation process of the case inference prediction model based on the deep Q-network algorithm is improved. Using the training set as input data, the model is continuously trained to obtain a prediction model. The study of furnace temperature changes from the case difference data not only includes the influence of various parameters on the furnace temperature but also utilizes historical data for reference information on furnace temperature changes. Finally, a furnace temperature prediction model is obtained, which can accurately predict the furnace temperature change trend during the urban solid waste incineration process, providing a guarantee for the optimized control of the incineration process. This lays the foundation for timely judgment of furnace temperature changes and optimized control of the solid waste incineration process, improving work efficiency. Attached Figure Description

[0039] Figure 1 This is a diagram illustrating the application environment of a furnace temperature prediction method for solid waste incineration in one embodiment.

[0040] Figure 2 This diagram illustrates the application environment of a furnace temperature prediction method for solid waste incineration in one embodiment.

[0041] Figure 3 This is a schematic diagram illustrating the furnace temperature prediction model structure for the solid waste incineration process in one embodiment.

[0042] Figure 4 This is a schematic diagram illustrating the process of building a difference database in one embodiment;

[0043] Figure 5 This is a schematic diagram illustrating the process of building a case difference prediction model in one embodiment;

[0044] Figure 6 This is a flowchart illustrating the setting of the loss function in one embodiment;

[0045] Figure 7 This is a schematic diagram illustrating the process of obtaining furnace temperature prediction values ​​based on a case difference prediction model in one embodiment.

[0046] Figure 8 This is a line graph illustrating the comparison of furnace temperature prediction results for a solid waste incineration process in one embodiment;

[0047] Figure 9 This is a structural block diagram illustrating a furnace temperature prediction system for a solid waste incineration process in one embodiment;

[0048] Figure 10 This is a schematic structural diagram of a computer device in one embodiment. Detailed Implementation

[0049] 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 merely illustrative of the present application and are not intended to limit the scope of the present application.

[0050] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when used in this specification, the words “comprising” and / or “including” indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0051] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details are presented in the various embodiments of this application to facilitate a better understanding of the application. However, the technical solutions claimed in this application can be implemented even without these technical details and various variations and modifications based on the following embodiments. The division of the various embodiments below is for ease of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0052] For ease of understanding, the system to which this application applies will first be described. The furnace temperature prediction method for a solid waste incineration process provided in this application can be applied to, for example... Figure 1The system architecture shown includes a user-space file server 103 and a terminal device 101. The terminal device 101 communicates with the user-space file server 103 via a network. 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 a file system, allowing remote clients running on the terminal device 101 to access the file system over the network in a manner similar to a local file system. The terminal device 101 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, etc. The user-space file server 103 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0053] Figure 2 This is a flowchart illustrating a method for predicting furnace temperature in a solid waste incineration process, provided in an embodiment of this application. Figure 3 This is a schematic diagram of the structure of a furnace temperature prediction model for a solid waste incineration process provided in an embodiment of this application. The method can be derived from, for example... Figure 1 The user-space file server in the system shown executes this. For example... Figure 2 , Figure 3 As shown, the method may include the following steps:

[0054] S201. Obtain historical data of multiple sets of characteristic variables affecting furnace temperature, the corresponding current furnace temperature value, and the corresponding furnace temperature value at the next moment to form a case description, and construct a difference database D according to the case description order;

[0055] S202. Preprocess the data based on the difference database D to obtain the training set;

[0056] S203. Based on the deep Q-network algorithm, a case difference prediction model is constructed according to the training set.

[0057] S204. Based on the characteristic variables of the current furnace temperature, obtain the case difference data of the furnace temperature, and use it as the input data of the case difference prediction model. Based on the output data of the case difference prediction model, obtain the predicted value of the furnace temperature at the next moment.

[0058] The following is a detailed explanation of each step:

[0059] S201. Obtain historical data of multiple sets of characteristic variables affecting furnace temperature, the corresponding current furnace temperature value, and the corresponding furnace temperature value at the next moment to form a case description, and construct a difference database D according to the case description order;

[0060] In this embodiment, it should be noted that multiple sets of characteristic variables affecting furnace temperature, along with corresponding current furnace temperature values ​​and historical data of the next furnace temperature value, are acquired to form case descriptions. These case descriptions are then compared in order to obtain case difference descriptions, resulting in a difference database D. This database provides a data foundation for the prediction model. By collecting these characteristic variables and their corresponding furnace temperature values, the differences between historical and current furnace temperature data are learned, facilitating subsequent predictive analysis based on the collected historical data. Specifically, a training set of 1000 samples and a test set of 200 samples can be collected, with each sample interval being 10 seconds. The training set is used to train the constructed prediction model, and the test set is used to test whether the final prediction model can accurately predict, thus determining its accuracy.

[0061] S202. Preprocess the data based on the difference database D to obtain the training set;

[0062] In this embodiment, it should be noted that the constructed difference database D is preprocessed to unify all historical data so that it can be directly analyzed and processed later.

[0063] S203. Based on the deep Q-network algorithm, a case difference prediction model is constructed according to the training set.

[0064] In this embodiment, it should be noted that the Deep Q-Network algorithm, or DQN for short, is a Q-learning algorithm implemented using deep learning. Q-learning is a reinforcement learning method based on evaluation values, which guides the agent's decision-making by learning an action-value function Q(state, action). The difference between Deep Q-Network and traditional Q-learning algorithms is that it does not require a predefined state-action value function. Instead, it automatically learns a function approximator representing this function through a neural network. During the training of this neural network, the parameters of the action-value function are updated using the Q-learning algorithm, thus enabling the handling of more complex control problems. Based on historical data generated during the incineration process, and using the DQN function approximator, the parameters of the action-value function are updated using the Q-learning algorithm during the training of the neural network to obtain the optimal result, leading to the final case-difference prediction model for furnace temperature, which enables accurate prediction of furnace temperature.

[0065] S204. Based on the characteristic variables of the current furnace temperature, obtain the case difference data of the furnace temperature and use it as the input data of the case difference prediction model. Based on the output data of the case difference prediction model, obtain the predicted value of the furnace temperature at the next moment.

[0066] In this embodiment, it should be noted that a difference database D is constructed based on the collected historical data. The difference data forms a training set. The training set is used as the input data for the prediction model to train the prediction model and obtain the output value. Based on the difference pattern between the historical furnace temperature data and the current furnace temperature data, the furnace temperature prediction model of the urban solid waste incineration process is realized by combining the output value with the retrieved similar cases, so as to accurately predict the trend of furnace temperature change.

[0067] Reference Figure 4 As shown, in some embodiments, in S201, multiple sets of historical data on characteristic variables affecting furnace temperature, the corresponding current furnace temperature value, and the corresponding furnace temperature value at the next moment are obtained to form a case description, and a difference database is constructed according to the order of the case descriptions, including:

[0068] S2011. Obtain multiple sets of characteristic variables affecting furnace temperature, such as grate speed, grate temperature, primary air flow, secondary air flow, and blower pressure, as well as the corresponding current furnace temperature value, as the problem description feature X of case c. The furnace temperature value y at the next moment is used as the solution description. Together, they form the case description. A total of N cases are selected to construct the furnace temperature case library C, as shown in formula (1):

[0069]

[0070] S2012. Select a pair of cases c from the case library in the order of case description. i c j The case difference description is obtained through formula (2):

[0071] e = Δ(c) i ,c j )=(ΔX ij ,Δy ij (20);

[0072] S2013. Based on the obtained case difference descriptions, construct a case difference database D, with N case differences. 2 It can be described by formula (3):

[0073]

[0074] In this embodiment, it should also be noted that multiple sets of data such as grate speed, grate temperature, primary air flow, secondary air flow, and blower pressure are collected, specifically as shown in Table 1, totaling 65 feature variables and their corresponding current furnace temperature values. The historical data of the acquired feature variables and the current furnace temperature value are used as the problem description feature X of case C, and the furnace temperature value y at the next moment is used as the solution description, together forming the case description. A total of N cases are selected to construct the furnace temperature case library C, which is used to provide a data foundation for obtaining the case difference description.

[0075] Table 1 is a data table showing the characteristics of historical data collected;

[0076]

[0077]

[0078]

[0079] In this embodiment, it should also be noted that the above 65 feature variables are all process variables of the solid waste incineration process and the corresponding current furnace temperature values, which are also necessary conditions for predicting the furnace temperature. By collecting historical data of these 65 feature variables and the corresponding current furnace temperature values, case difference descriptions are obtained to facilitate training of the prediction model.

[0080] Reference Figure 2 As shown, in some embodiments, in S202, data preprocessing is performed based on the difference database D to obtain a training set, including:

[0081] The features of the differential cases in the case difference database D are standardized using a min-max method, which is described by formula (4):

[0082]

[0083] Where, ΔX k,m Let m be the feature variable in the k-th difference case, where k = 1, 2, ..., N. 2 N 2 The number of case differences is m = 1, 2, ..., M, where M represents the number of case features with differences, and here M = 66.

[0084] In this embodiment, it should be noted that normalization is used to eliminate the influence of different dimensions between variables. Here, all feature variables in the training set D are normalized.

[0085] Reference Figure 2 , Figure 5 As shown, in some embodiments, in S203, a case difference prediction model is constructed based on the deep Q-network algorithm according to the training set, including:

[0086] S2031. Based on a deep Q-network, set the number of hidden layers L, the number of hidden layer nodes n, the learning rate lr, and the discount factor γ of the deep Q-network, and define the relevant state space S, action space A, and reward function r, and initialize the parameters; where,

[0087] The state space S is a series of state descriptions, which in this case are the problem characteristics in the furnace temperature difference case, represented as S=[ΔX1,…,ΔX N2 The action space is a series of action values ​​with intervals l, which here represents the solution size in the furnace temperature difference case, i.e., A = [Δy]. min ,Δy min +l,…,Δy max ];

[0088] The reward function is described by formula (5):

[0089] r=-|Δy-a| (23),

[0090] In the formula, Δy represents the solution of the difference case, and a represents the action in action space A.

[0091] In this embodiment, it should be noted that the Deep Q-Network (DQN) algorithm is a deep reinforcement learning algorithm based on value functions. By constructing a prediction model based on the DQN algorithm, DQN effectively combines deep learning and reinforcement learning. It introduces three core technologies: First, the objective function, which uses a convolutional neural network combined with a fully connected layer as an approximator for the action value function, achieving an end-to-end effect. The input is video footage, and the output is a finite number of action value functions. Second, the target network, which is set up to separately handle TD error, making the target value relatively stable. TD error stands for Temporal Difference Error, which is an error calculation method based on time differences and is commonly used in reinforcement learning. Third, the experience playback mechanism, which effectively solves the problems of correlation and non-static relationships between data, ensuring that the information input to the network satisfies the condition of independent and identically distributed data.

[0092] Reinforcement learning has four elements: state, action, policy, and reward. Here, a prediction model is constructed based on the DQN structure. First, its parameters are configured by setting the number of hidden layers L=3, the number of hidden layer nodes n=64, the learning rate lr=0.001, and the discount factor γ=0.95. The relevant state space S, action space A, and reward function r are defined. The reward function is described by formula (5): r=-|Δy-a|(24). Parameter initialization is performed to configure the parameter settings of the algorithm used in the prediction model for subsequent calculations. The state space S is a series of state descriptions, which here represents the problem features in the furnace temperature difference case: S=[ΔX1,…,ΔX…]. N 2 The action space is a series of action values ​​with intervals l = 0.5, which here represents the solution size in the furnace temperature difference case, A = [Δy]. min ,Δy min +0.5,…,Δy max], where Δy represents the solution of the difference case, and a represents the action in action space A. A prediction model is constructed based on the DNQ structure and the specific furnace temperature prediction environment.

[0093] Reference Figure 2 , Figure 5 As shown, in some embodiments, S203, based on the deep Q-network algorithm, constructing a case difference prediction model according to the training set, further includes:

[0094] S2032. Construct the Q-network and the target Q-network, and define the Q-function Q(s,a|θ) of the Q-network accordingly, which is described by formula (6):

[0095]

[0096] In the formula, θ represents the network weights, s represents the state in the state space S, a represents the action in the action space A, the Q function corresponding to the target Q network is represented by Q'(s,a|θ'), and θ' represents the target network weights;

[0097] S2033. The selection of action a is achieved according to the ε-greedy strategy, which is described by formula (7):

[0098]

[0099] S2034. Use the training set as input data for the Q(s,a|θ) network, and set a loss function L(θ) related to the reward value based on the reward value corresponding to the training and learning of the Q network and the target Q network.

[0100] S2035. Based on the loss function L(θ), continuously update the Q neural network parameters θ until the loss function meets the set conditions to obtain the case difference prediction model of furnace temperature.

[0101] In this embodiment, it should be noted that the deep Q-network is a combination of neural networks in deep learning and Q-learning algorithms in reinforcement learning for solving optimal action-value functions. By constructing a Q-network and a target Q-network, the Q-function Q(s,a|θ) of the Q-network is defined accordingly. The Q(s,a|θ) framework is a nonlinear function approximation model. θ represents the parameters of the Q-neural network. The deep Q-network is initialized by constructing the neural network framework, i.e., the Q(s,a|θ) framework, and initializing the θ parameters. The training set is used as the input data for the Q(s,a|θ) network. The Q-network and the target Q-network are trained to learn the corresponding reward values. A loss function L(θ) related to the reward values ​​is set, and the Q-neural network parameters θ are continuously updated until the loss function meets the set conditions, resulting in a prediction model. Here, the loss function is used to optimize the network. As the Q-network is continuously updated, this loss function decreases until a preset condition is reached, at which point learning stops. Specifically, the preset condition can be a preset number of training iterations, such as 1000, or a preset training error, such as 0.001.

[0102] Furthermore, the ε-greedy strategy is used to select action 'a'. The ε-greedy method primarily balances exploration and exploitation; that is, whether to expend effort exploring to obtain a more accurate estimate of the reward, or to choose the action with the highest expected reward based on currently available information. In the ε-greedy algorithm, at each decision point, the agent selects a random action with probability ε and the action currently considered best with probability 1-ε. The formula for the ε-greedy algorithm is as follows: if the random number r < ε, then a random action is selected; if the random number r >= ε, then the action currently considered best is selected, where ε is a decimal between 0 and 1, called the exploration rate. Specifically, at each decision point, we first randomly generate a random number between 0 and 1. If r is less than the exploration rate, a random action is selected; otherwise, the action currently considered best is selected. Moreover, in the ε-greedy algorithm, the exploration rate ε can be gradually reduced over time so that the agent can make better use of previous experience and knowledge. Here, the ε-greedy strategy is used to select action a, where the initial ε is set to 0.5, and then the selection of action a in the Q network is updated to obtain the case difference solution Δy.

[0103] In some embodiments, refer to Figure 3 , Figure 5 , Figure 6 As shown in S2034, the training set is used as the input data for the Q(s,a|θ) network. Based on the reward values ​​corresponding to the training and learning of the Q network and the target Q network, a loss function L(θ) related to the reward value is set, including:

[0104] S20341, Set the current state to s t According to action a t Receive network rewards r t and enter the next state s t+1 and (s t ,a t ,r t ,s t+1 Store it in experience pool B;

[0105] S20342, Based on the sampling (s) in experience pool B i ,a i ,r i ,s i+1 The target value t is calculated based on Q'(s,a|θ') of the target network. i It can be described by formula (8):

[0106] t i =r i +γmaxQ′(s t+1 ,a′|θ′) (26),

[0107] Among them, experience pool B is used to store the current state s. t Action a t Next state s t+1 And reward r t , where a' represents the action value selected by the target network from the action space S;

[0108] S20343. Based on the constructed Q-network and the target Q-network, the loss function L(θ) is obtained by continuously learning and training according to the training set.

[0109] In this embodiment, it should be noted that the obtained case difference data is used as the training set and input into the Q-network. The target network Q'(s,a|θ') is used to separately process the TD error, making the target value relatively stable. Techniques such as experience replay and the target network are used to optimize the training process. Here, TD error stands for Temporal Difference Error, a method for calculating error based on time differences, commonly used in reinforcement learning. The formula for TD error is a commonly used formula in machine learning, used to measure the difference between the predicted value and the true value. The formula for TD error is: TD Error = Predicted Value - True Value + Discount Factor * Predicted Value of the Next State. Where the predicted value is the predicted value in the current state, the true value is the true value in the current state, and the discount factor is a value between 0 and 1 used to measure the importance of future rewards. The formula for TD error can help us evaluate the predictive ability of the model. By calculating the difference between the predicted value and the true value, we can understand the magnitude of the model's error and thus optimize the model.

[0110] Specifically, obtain the current state as s t Δy is obtained based on action a. t Receive network rewards r t That is, r = -|Δy - a|, and then enter the next state s. t+1 and (s t ,a t ,r t ,s t+1 ) are stored in experience pool B. Samples are taken from experience pool B (s) i ,a i ,r i ,s i+1 The target value t is calculated using Q'(s,a|θ') of the target network. i , i.e. t i =r i +γmaxQ′(s t+1 ,a′|θ′), where the experience pool B is an element used in the experience replay technique of DQN reinforcement learning, used to store continuously updated elements, namely the current state, action, next state, reward, etc., and a′ represents the action value selected by the target network from the action space. Based on the training set as the input data of the Q(s,a|θ) network, and utilizing the experience replay mechanism, the obtained data (s) are processed... t ,a t ,r t ,s t+1 The reward value t is stored in the experience pool B and continuously trained based on the Q-network and the target Q-network to obtain the corresponding reward value. i We obtain the loss function L(θ) and continuously update the parameters of the Q function and the objective Q function to maximize the total reward.

[0111] In some embodiments, refer to Figure 3 , Figure 5 As shown in S2035, the Q-neural network parameters θ are continuously updated according to the loss function L(θ) until the loss function meets the set conditions. This includes updating the Q-neural network according to the Bellman equation to obtain the loss function L(θ), which is described by formula (9):

[0112] L i (θ i ) = E (s,a,r,s′) [(t i -Q(s i ,a i |θ i )) 2 ]

[0113] =E (s,a,r,s′) [(r i +γmaxQ′(s t+1 ,a′|θ′)-Q(s i ,a i |θ i )) 2 (27).

[0114] In this embodiment, it should be noted that the Bellman equation is a system of functional equations concerning the objective function. The method of establishing a system of functional equations using optimization and embedding principles is called the functional equation method. In practical applications, specific solutions must be sought for each specific problem. The significance of the Bellman equation lies in the fact that the value function of the current state can be calculated using the value function of the next state. Similarly, the state-action value function has a similar relationship. The purpose of calculating the state value function is to construct a learning algorithm to obtain the optimal policy from the data. Each policy corresponds to a state value function, and the optimal policy corresponds to the optimal state value function, in order to maximize the total reward.

[0115] Here, the Bellman equation is used to update the Q-network, thereby determining the current furnace temperature change value based on the furnace temperature change value of the next state. Finally, the constructed Q-network is continuously updated according to the state value function corresponding to each policy until the loss function L(θ) satisfies the preset condition. Here, the preset condition is that when the number of training times reaches 1000, the optimal policy corresponding to the optimal state value function under the preset condition is obtained.

[0116] In some embodiments, refer to Figure 2 , Figure 7As shown in S204, based on the characteristic variables of the current furnace temperature, case difference data of the furnace temperature is obtained and used as input data for the case difference prediction model. Based on the output data of the case difference prediction model, the predicted value of the furnace temperature at the next moment is obtained, including:

[0117] S2041, Based on the characteristic variable X of the current furnace temperature t The similarity measurement is calculated between the data and historical data in the furnace temperature case library C, and is described by formula (10):

[0118]

[0119] In the formula, X j M represents the problem characteristics of a case in the case library, where M is the number of feature variables.

[0120] S2042. The most similar case c is obtained based on similarity. xim =(X sim ,y sim ), the case difference data ΔX that constitutes the furnace temperature t,sim ;

[0121] S2043, ΔX t,sim As input, the output action value a = Q(ΔX) is obtained through the constructed model for predicting furnace temperature changes. t,sim The value is determined, and the corresponding action value is selected from the action space as the case difference solution, i.e., Δy. t,sim =A[a];

[0122] S2044. Based on the obtained case difference solution Δy t,sim The predicted value y of the furnace temperature is obtained and described by formula (11):

[0123] y = y t,sim +Δy t,sim (28).

[0124] In this embodiment, it should be noted that the characteristic variable X of the current furnace temperature in the test set is used. t Similarity measurement calculations are performed between the data and historical data in the furnace temperature case library C, specifically using the formula... Find the case c that has the closest similarity to the feature variables and corresponding furnace temperature values ​​in the furnace temperature case library C. sim =(X sim ,y sim ), and obtained the furnace temperature case difference data ΔX t,simThis uses a distance metric formula because the more similar two cases are, the smaller their distance. After performing a similarity metric on all cases, a ranking is generated based on the similarity score (D). The case with the smallest D value is then selected as the most similar case. In this case, this refers to obtaining the furnace temperature case difference data ΔX. t,sim .

[0125] Based on furnace temperature case difference data ΔX t,sim As input data to the furnace temperature prediction model, the output action value a = Q(ΔX) is obtained. t,sim The value is selected as the case difference solution Δy based on the action space. t,sim =A[a], based on the case difference solution and the current furnace temperature value, the predicted furnace temperature y is obtained, that is, the furnace temperature value at the next moment is y = y t,sim +Δy t,sim To predict furnace temperature change trends.

[0126] In some embodiments, refer to Figure 2 , Figure 7 As shown, based on the characteristic variables of the current furnace temperature, the case difference data of the furnace temperature is obtained and used as the input data of the case difference prediction model. Based on the output data of the case difference prediction model, the predicted value of the furnace temperature at the next moment is obtained. It also includes: evaluating the prediction accuracy of the model based on the test set and the predicted furnace temperature change model.

[0127] In this embodiment, it should be noted that steps S2041 to S2044 are repeated to perform a similarity measurement calculation between the current furnace temperature feature variable of the test set data and the historical data in the furnace temperature case library C, obtaining furnace temperature case difference data. This furnace temperature case difference data is used as input data, and the output data, i.e., the case difference solution, is obtained according to the prediction model. Based on the case difference solution and the current furnace temperature value, the predicted furnace temperature value y, i.e., the furnace temperature value at the next moment, is obtained. Then, the obtained predicted value y is compared with the actual value of the test set. The root mean square error (RMSE) and mean absolute percentage error (MAPE) are used to evaluate the overall predictive performance of the model. The smaller the RMSE and MAPE, the higher the prediction accuracy.

[0128] Specifically, in one example, such as Figure 8As shown, 200 sets of historical data from the incineration process were collected as a test set. The mean absolute error and root mean square error were used to evaluate the predicted and actual values. The calculated mean absolute error and root mean square error were 1.791 (°C) and 2.325 (°C), respectively, which met the target error. This indicates that the method can accurately estimate the furnace temperature in the urban solid waste incineration process. The furnace temperature change prediction model constructed based on the case reasoning prediction method improved by deep Q network can accurately predict its change trend, providing a guarantee for the optimized control of the incineration process. This allows on-site operators to keep abreast of the incineration conditions of solid waste in the furnace and improve work efficiency.

[0129] The implementation principle of this embodiment is as follows: The above steps mainly involve collecting historical data on characteristic variables generated during the incineration process of solid waste incineration power plants, as well as the corresponding current furnace temperature and the furnace temperature at the next moment, to form case descriptions. Following the order of these case descriptions, a difference database D is constructed to obtain the training set. The adaptation process of the case inference prediction model based on the deep Q-network algorithm uses the training set as input data, continuously learning and training to obtain the prediction model. By studying the furnace temperature changes from the case difference data, it not only includes the influence of various parameters on the furnace temperature but also utilizes historical data for reference information on furnace temperature changes. Finally, a furnace temperature prediction model is obtained, which can accurately predict the furnace temperature change trend during the urban solid waste incineration process, providing a guarantee for the optimized control of the incineration process. This lays the foundation for timely judgment of furnace temperature changes and optimized control of the solid waste incineration process, improving work efficiency.

[0130] Reference Figure 9 As shown, this application also provides a system for predicting furnace temperature in a solid waste incineration process. This system may include: a data acquisition module, a data preprocessing module, a prediction model construction module, and a result prediction module. The main functions of each component module are as follows:

[0131] The data acquisition module 301 is used to acquire multiple sets of characteristic variables affecting furnace temperature, the corresponding current furnace temperature value, and the historical data of the corresponding next furnace temperature value to form a case description, and to construct a difference database D according to the case description order;

[0132] Data preprocessing module 302 is used to preprocess data based on the differential database D to obtain the training set;

[0133] The prediction model construction module 303 is used to construct a case difference prediction model based on the deep Q-network algorithm and the training set.

[0134] The result prediction module 304 is used to obtain case difference data of furnace temperature based on the characteristic variables of the current furnace temperature, and use it as input data for the case difference prediction model. Based on the output data of the case difference prediction model, the predicted value of the furnace temperature at the next moment is obtained.

[0135] According to embodiments of this application, this application also provides a computer device and a computer-readable storage medium.

[0136] like Figure 10 The diagram shown is a block diagram of a computer device according to an embodiment of this application. The term "computer device" is intended to represent various forms of digital computers or mobile devices. 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 smartphone, a wearable device, etc.

[0137] like Figure 10 As shown, 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, ROM 602, and RAM 603 are interconnected via the bus 604. The input / output (I / O) interface 605 is also connected to the bus 604.

[0138] The computing unit 601 can execute various processes in the method embodiments of this application according to computer instructions stored in the read-only memory (ROM) 602 or computer instructions loaded from the storage unit 608 into the random access memory (RAM) 603. The computing unit 601 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. The computing unit 601 can 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 methods provided in the embodiments of this application can be implemented as computer software programs, which are tangibly contained in a computer-readable storage medium, such as the storage unit 608.

[0139] RAM 603 may also store various programs and data required for the operation of device 600. Part or all of the computer program may be loaded and / or installed on device 600 via ROM 602 and / or communication unit 609.

[0140] The input unit 606, output unit 607, storage unit 608, and communication unit 609 in device 600 can be connected to I / O interface 605. The input unit 606 can be, for example, a keyboard, mouse, touchscreen, or microphone; the output unit 607 can be, for example, a display, speaker, or indicator light. Device 600 can exchange information and data with other devices through the communication unit 609.

[0141] It should be noted that the device may also include other components necessary for normal operation. It may also include only the components necessary for implementing the solution of this application, without necessarily including all the components shown in the figures.

[0142] Various implementations of the systems and techniques 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), payload programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof.

[0143] The computer instructions used to implement the methods of this application may be written in any combination of one or more programming languages. These computer instructions may be provided to the computing unit 601 such that when executed by the computing unit 601, such as a processor, the computer instructions cause the execution of the steps involved in the embodiments of the methods of this application.

[0144] The computer-readable storage medium provided in this application can be a tangible medium that can contain or store computer instructions for performing the steps involved in the method embodiments of this application. The computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, and other forms of storage media.

[0145] The specific embodiments described above do not constitute a limitation on the scope of protection 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 should be included within the scope of protection of this application.

Claims

1. A method for predicting furnace temperature in a solid waste incineration process, characterized in that, Includes the following steps: Collect historical data of multiple sets of characteristic variables affecting furnace temperature, the corresponding current furnace temperature value, and the corresponding furnace temperature value at the next moment to form a case description, and construct a difference database D according to the case description order; Data preprocessing is performed based on the differential database D to obtain the training set; A case difference prediction model is constructed based on the deep Q-network algorithm and the training set. The deep Q-network-based algorithm, which constructs a case difference prediction model based on the training set, includes: Based on a deep Q-network, the number of hidden layers L, the number of hidden layer nodes n, the learning rate lr, and the discount factor γ are set. The relevant state space S, action space A, and reward function r are also defined, and the parameters are initialized. The state space S is a series of state descriptions, which here represent the problem characteristics in the furnace temperature difference case, expressed as S=[Δ X 1,…, Δ X N 2 The action space is the interval. l A series of action values, here representing the solution magnitude in the furnace temperature difference case, i.e., A = [Δ y min ,Δ y min + l ,…,Δ y max ]; The reward function is described by formula (5): （5）， In the formula, Δy represents the solution of the difference case, and a represents the action in action space A; The method for constructing a case difference prediction model based on the deep Q-network algorithm and the training set also includes: Construct a Q-network and a target Q-network, and define the Q-function Q(s, a|θ) of the Q-network accordingly, which is described by formula (6): （6）， In the formula, θ represents the network weights, s represents the state in the state space S, and a represents the action in the action space A; The Q-function corresponding to the target Q-network is denoted by Q'(s, a|θ'), where θ' represents the target network weights; The selection of action a is achieved according to the ε-greedy strategy, which is described by formula (7): The training set is used as the input data of the Q(s, a|θ) network. Based on the reward value corresponding to the training and learning of the Q network and the target Q network, a loss function L(θ) related to the reward value is set. Based on the loss function L(θ), the Q neural network parameters θ are continuously updated until the loss function meets the set conditions, thus obtaining the case difference prediction model for furnace temperature; Based on the characteristic variables of the current furnace temperature, the case difference data of the furnace temperature is obtained and used as the input data of the case difference prediction model. Based on the output data of the case difference prediction model, the predicted value of the furnace temperature at the next moment is obtained.

2. The method for predicting furnace temperature in the solid waste incineration process according to claim 1, characterized in that, The process involves acquiring historical data of multiple sets of characteristic variables affecting furnace temperature, their corresponding current furnace temperature values, and their corresponding next-time furnace temperature values ​​to form case descriptions. A difference database D is then constructed according to the order of these case descriptions, including: Multiple sets of characteristic variables, including but not limited to grate speed, grate temperature, primary air flow, secondary air flow, and blower pressure, along with their corresponding current furnace temperature values, are obtained as the problem description feature X for case c. The furnace temperature value y at the next moment is used as the solution description. Together, they form the case description. A total of N cases are selected to construct the furnace temperature case library C, as shown in formula (1): （1）； Select a pair of cases c from the furnace temperature case library C in the order of their descriptions. i c j The case difference description is obtained through formula (2): （2）； Based on the obtained case difference descriptions, a case difference database D is constructed, containing the number of case differences. N 2 It can be described by formula (3): （3）。 3. The furnace temperature prediction method for solid waste incineration process according to claim 2, characterized in that, The data preprocessing based on the difference database D yields a training set, including: The features of the differential cases in the case difference database D are standardized using a min-max method, and described by formula (4): （4）； Where, Δ Let m be the feature variable in the k-th difference case, where k = 1, 2, ... N 2 , N 2 The number of case differences is m = 1, 2, ..., M, where M represents the number of case features with differences, and here M = 66.

4. The method for predicting furnace temperature in the solid waste incineration process according to claim 1, characterized in that, The step of using the training set as input data for the Q(s, a|θ) network, and setting a loss function L(θ) related to the reward value based on the reward value corresponding to the training and learning of the Q network and the target Q network, includes: Set the current state to According to action a t Receive network rewards r t and enter the next state. and will ( , , , Store it in experience pool B; Based on sampling from experience pool B ( , , , The target value is calculated based on Q'(s, a|θ') of the target network. It can be described by formula (8): （8）， Among them, experience pool B is used to store the current state. Action a t Next state Rewards r t , where a' represents the action value selected by the target network from the action space S; Based on the constructed Q-network and the target Q-network, the loss function L(θ) is obtained by continuously learning and training according to the training set.

5. The furnace temperature prediction method for solid waste incineration process according to claim 4, characterized in that, The step of continuously updating the Q-neural network parameters θ according to the loss function L(θ) until the loss function meets the set conditions includes: The Q-network is updated according to the Bellman equation, and the loss function L(θ) is obtained, which is described by formula (9): （9）。 6. The furnace temperature prediction method for solid waste incineration process according to claim 5, characterized in that, The step of obtaining case difference data of furnace temperature based on the characteristic variables of the current furnace temperature and using it as input data for the case difference prediction model, and obtaining the predicted value of the furnace temperature at the next moment based on the output data of the case difference prediction model, includes: Based on the characteristic variables of the current furnace temperature The similarity measurement is calculated between the data and historical data in the furnace temperature case library C, and is described by formula (10): （10）， In the formula, M represents the problem characteristics of a case in the case library, where M is the number of feature variables. The most similar case was retrieved based on similarity. =( , ), the case differences in furnace temperature Δ X t,sim ; Δ X t,sim As input, the output action value a=Q(Δ) is obtained through the constructed case difference prediction model. X t,sim The value is determined, and the corresponding action value is selected from the action space as the case difference solution Δ. y t,sim =A[a]; Based on the obtained case difference solution Δ y t,sim The predicted value y of the furnace temperature is obtained and described by formula (11): （11）。 7. A system for performing a furnace temperature prediction method for a solid waste incineration process as described in any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images