Intelligent optimization method and system for operation of garbage combustion equipment and medium

Abstract

The invention provides an intelligent optimization method and system for operation of garbage combustion equipment and a medium. The method comprises the following steps: acquiring a preset number of historical input variable data and historical output variable data of garbage combustion equipment, and then classifying the historical input variable data and the corresponding historical output variable data to obtain a training data set and a test data set, key parameters of a gradient boosting tree are obtained, an updated gradient boosting tree model is obtained through training, input variable importance indexes are obtained through calculation according to the updated gradient boosting tree model, optimized input variables are obtained through threshold comparison, and a thermal efficiency prediction model is obtained through processing of a multi-layer BP neural network and a particle swarm optimization algorithm. The thermal efficiency prediction model is processed by a reinforcement learning algorithm to obtain optimal input variable feature data; therefore, intelligent optimization of the garbage combustion equipment is realized through the gradient boosting tree model, the multilayer BP neural network and the particle swarm optimization algorithm technology, and the power generation cost of the garbage combustion equipment is reduced.

Description

Technical Field

[0001] The present application relates to the field of waste combustion, and in particular to an intelligent optimization method, system and medium for the operation of waste combustion equipment. Background Art

[0002] With the acceleration of urbanization, the amount of garbage is increasing, and garbage incineration is widely used as an effective way to dispose of garbage. However, there are many problems in the operation of current garbage incineration equipment. For example, the operation of garbage furnaces mainly relies on personal experience, which leads to insufficient garbage combustion and low effective conversion rate. Moreover, the operation control method of traditional garbage incineration equipment is often based on fixed parameters or simple feedback mechanisms, which is difficult to adapt to the dynamic changes of garbage characteristics, and cannot effectively adjust the key parameters such as temperature, air supply, and combustion time in the combustion process in real time and accurately, thus affecting the efficiency and environmental protection effect of garbage combustion. In addition, there are many parameters in the combustion process of garbage incineration equipment, and how to optimize and control more parameters in the combustion process is also one of the difficulties. In response to the above problems, effective technical solutions are urgently needed. Summary of the invention

[0003] The purpose of the present application is to provide an intelligent optimization method, system and medium for the operation of a waste combustion equipment. The present application first analyzes the parameters in the combustion process through a gradient boosting tree model to screen out important parameters, thereby reducing the number of parameters to be optimized; in addition, the intelligent optimization of the waste combustion equipment is achieved through a multi-layer BP neural network and a particle swarm optimization algorithm technology, and further optimizes the parameters in the combustion process through reinforcement learning, thereby obtaining the highest boiler thermal efficiency and reducing the power generation cost of the waste combustion equipment.

[0004] The present application also provides an intelligent optimization method for the operation of a garbage combustion device, comprising the following steps:

[0005] Obtaining a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set;

[0006] Obtain the key parameters of the preset gradient boosting tree, and train the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model;

[0007] Obtaining an input variable importance index according to the updated gradient boosting tree model, and comparing the input variable importance index with a preset optimized input variable threshold to obtain an optimized input variable;

[0008] An initial thermal efficiency prediction model is established based on the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and the thermal efficiency prediction model is obtained through processing with a preset particle swarm optimization algorithm;

[0009] The thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables.

[0010] Optionally, in the intelligent optimization method for the operation of the garbage incineration equipment described in the present application, the obtaining of a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data and classifying them to obtain a training data set and a test data set, specifically includes:

[0011] Establish a corresponding relationship between historical input variable data and corresponding historical output variable data, and group them according to a preset ratio to obtain a training data set and a test data set;

[0012] The training data set includes training history input variable data and corresponding training history output variable data;

[0013] The test data set includes test history input variable data and corresponding test history output variable data.

[0014] Optionally, in the intelligent optimization method for the operation of the garbage incineration equipment described in the present application, the key parameters of the preset gradient boosting tree are obtained, and the gradient boosting tree is trained according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model, specifically including:

[0015] Get the key parameters of the preset gradient boosting tree, including the number of decision trees, learning rate data, and maximum depth data of the decision tree;

[0016] Get the average value of the training history output variable data and mark it as the model initial value;

[0017] Iterate according to the model initial value, training history input variable data and corresponding training history output variable data combined with the number of decision trees, and obtain an updated gradient boosting tree model after testing the test history input variable data and the corresponding test history output variable data.

[0018] Optionally, in the intelligent optimization method for the operation of the garbage incineration equipment described in the present application, the input variable importance index is calculated according to the updated gradient boosting tree model, and the input variable importance index is compared with a preset optimization input variable threshold to obtain the optimized input variable, specifically including:

[0019] The input variable importance index of each input variable is obtained through gradient boosting tree model processing;

[0020] Calculate the sum of the importance indexes of the input variables in each input variable and record it as the sum of importance;

[0021] The input variable importance index of each input variable is divided by the total importance to obtain the input variable importance ratio data of each input variable;

[0022] Sort the important proportion data of the input variables from large to small;

[0023] Select input variables with a value greater than the preset importance threshold and mark them as optimized input variables.

[0024] Optionally, in the intelligent optimization method for the operation of the garbage incineration equipment described in the present application, the optimization input variable data and the output variable data corresponding to the optimization input variable are combined with a preset multi-layer BP neural network to establish an initial thermal efficiency prediction model, and the thermal efficiency prediction model is obtained by processing with a preset particle swarm optimization algorithm, specifically including:

[0025] The initial thermal efficiency prediction model is established by presetting a multi-layer BP neural network combined with optimized input variable data and output variable data;

[0026] Obtain the predicted value of the output variable and the actual value of the output variable, calculate the mean square error based on the predicted value of the output variable and the actual value of the output variable, and mark it as the fitness value;

[0027] The initial thermal efficiency prediction model is optimized in a preset manner according to the fitness value to obtain a thermal efficiency prediction model.

[0028] Optionally, in the intelligent optimization method for the operation of the garbage combustion equipment described in the present application, the thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables, specifically including:

[0029] S51, initializing the input variables to be optimized;

[0030] S52, the agent obtains the predicted output variable according to the input variable to be optimized;

[0031] S53, obtaining the output variable reward value of each output variable data and the sum of the reward values ​​of the output variables respectively through a preset reward function; the intelligent agent of the reinforcement learning is the thermal efficiency prediction model;

[0032] S54. Adjust the optimized input variable data according to the sum of the output variable reward values, and then re-execute steps S51-S53 until the optimal input variable characteristic data is obtained.

[0033] In a second aspect, the present application provides an intelligent optimization system for the operation of a garbage combustion device, the system comprising: a memory and a processor, the memory comprising an intelligent optimization method program for the operation of the garbage combustion device, the intelligent optimization method program for the operation of the garbage combustion device being executed by the processor to implement the following steps:

[0034] Obtaining a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set;

[0035] Obtain the key parameters of the preset gradient boosting tree, and train the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model;

[0036] Obtaining an input variable importance index according to the updated gradient boosting tree model, and comparing the input variable importance index with a preset optimized input variable threshold to obtain an optimized input variable;

[0037] An initial thermal efficiency prediction model is established based on the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and the thermal efficiency prediction model is obtained through processing with a preset particle swarm optimization algorithm;

[0038] The thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables.

[0039] The method of obtaining a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set, specifically includes:

[0040] Establish a corresponding relationship between historical input variable data and corresponding historical output variable data, and group them according to a preset ratio to obtain a training data set and a test data set;

[0041] The training data set includes training history input variable data and corresponding training history output variable data;

[0042] The test data set includes test history input variable data and corresponding test history output variable data.

[0043] The step of obtaining the key parameters of the preset gradient boosting tree and training the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model specifically includes:

[0044] Get the key parameters of the preset gradient boosting tree, including the number of decision trees, learning rate data, and maximum depth data of the decision tree;

[0045] Get the average value of the training history output variable data and mark it as the model initial value;

[0046] Iterate according to the model initial value, training history input variable data and corresponding training history output variable data combined with the number of decision trees, and obtain an updated gradient boosting tree model after testing the test history input variable data and the corresponding test history output variable data.

[0047] The step of calculating the input variable importance index according to the updated gradient boosting tree model and comparing the input variable importance index with a preset optimized input variable threshold to obtain the optimized input variable specifically includes:

[0048] The input variable importance index of each input variable is obtained through gradient boosting tree model processing;

[0049] Calculate the sum of the importance indexes of the input variables in each input variable and record it as the sum of importance;

[0050] The input variable importance index of each input variable is divided by the total importance to obtain the input variable importance ratio data of each input variable;

[0051] Sort the important proportion data of the input variables from large to small;

[0052] Select input variables with a value greater than the preset importance threshold and mark them as optimized input variables.

[0053] The step of calculating the input variable importance index according to the updated gradient boosting tree model and comparing the input variable importance index with a preset optimized input variable threshold to obtain the optimized input variable specifically includes:

[0054] The input variable importance index of each input variable is obtained through gradient boosting tree model processing;

[0055] Calculate the sum of the importance indexes of the input variables in each input variable and record it as the sum of importance;

[0056] The input variable importance index of each input variable is divided by the total importance to obtain the input variable importance ratio data of each input variable;

[0057] Sort the important proportion data of the input variables from large to small;

[0058] Select input variables with a value greater than the preset importance threshold and mark them as optimized input variables.

[0059] The method of establishing an initial thermal efficiency prediction model based on the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and obtaining the thermal efficiency prediction model through a preset particle swarm optimization algorithm specifically includes:

[0060] The initial thermal efficiency prediction model is established by presetting a multi-layer BP neural network combined with optimized input variable data and output variable data;

[0061] Obtain the predicted value of the output variable and the actual value of the output variable, calculate the mean square error based on the predicted value of the output variable and the actual value of the output variable, and mark it as the fitness value;

[0062] The initial thermal efficiency prediction model is optimized in a preset manner according to the fitness value to obtain a thermal efficiency prediction model.

[0063] The thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables, specifically including:

[0064] Initializing the input variables to be optimized;

[0065] The intelligent agent obtains the predicted output variable according to the input variable to be optimized;

[0066] The output variable reward value and the sum of the output variable reward values ​​of each output variable data are obtained respectively by processing with a preset reward function; the intelligent agent of the reinforcement learning is the thermal efficiency prediction model;

[0067] The optimized input variable data is adjusted according to the sum of the output variable reward values, and then the above steps are re-executed until the sum of the reward values ​​no longer changes, at which time the optimal input variable characteristic data is obtained.

[0068] Optionally, in the intelligent optimization system for the operation of the garbage incineration equipment described in the present application, the obtaining of a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data and classifying them to obtain a training data set and a test data set, specifically includes:

[0069] Establish a corresponding relationship between historical input variable data and corresponding historical output variable data, and group them according to a preset ratio to obtain a training data set and a test data set;

[0070] The training data set includes training history input variable data and corresponding training history output variable data;

[0071] The test data set includes test history input variable data and corresponding test history output variable data.

[0072] Optionally, in the intelligent optimization system for the operation of the garbage incineration equipment described in the present application, the key parameters of the preset gradient boosting tree are obtained, and the gradient boosting tree is trained according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model, specifically including:

[0073] Get the key parameters of the preset gradient boosting tree, including the number of decision trees, learning rate data, and maximum depth data of the decision tree;

[0074] Get the average value of the training history output variable data and mark it as the model initial value;

[0075] Iterate according to the model initial value, training history input variable data and corresponding training history output variable data combined with the number of decision trees, and obtain an updated gradient boosting tree model after testing the test history input variable data and the corresponding test history output variable data.

[0076] In a third aspect, the present application also provides a readable storage medium, in which a program for an intelligent optimization method for the operation of a waste incineration device is stored. When the program for the intelligent optimization method for the operation of a waste incineration device is executed by a processor, the steps of an intelligent optimization method for the operation of a waste incineration device as described in any one of the above items are implemented.

[0077] As can be seen from the above, the intelligent optimization method, system and medium for the operation of garbage combustion equipment provided by the present application. The method obtains a preset number of historical input variable data and historical output variable data of the garbage combustion equipment, and then classifies the historical input variable data and the corresponding historical output variable data to obtain a training data set and a test data set, obtains the key parameters of the gradient boosting tree and trains to obtain an updated gradient boosting tree model, calculates the input variable importance index according to the updated gradient boosting tree model and obtains the optimized input variable through threshold comparison, obtains the thermal efficiency prediction model through multi-layer BP neural network and particle swarm optimization algorithm processing, and obtains the optimal input variable feature data after the thermal efficiency prediction model is processed by the reinforcement learning algorithm; thereby, the intelligent optimization of the garbage combustion equipment is realized through the gradient boosting tree model, the multi-layer BP neural network, and the particle swarm optimization algorithm technology, and the power generation cost of the garbage combustion equipment is reduced.

[0078] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures particularly pointed out in the written description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0080] Figure 1 A flow chart of an intelligent optimization method for the operation of a garbage combustion equipment provided in an embodiment of the present application;

[0081] Figure 2 A flowchart of obtaining a training data set and a test data set for an intelligent optimization method for operation of a garbage combustion device provided in an embodiment of the present application;

[0082] Figure 3 A flowchart of obtaining and updating a gradient boosting tree model for an intelligent optimization method for operation of a garbage combustion device provided in an embodiment of the present application;

[0083] Figure 4 A flowchart of optimizing input variables of an intelligent optimization method for operation of a garbage combustion device provided in an embodiment of the present application;

[0084] Figure 5 A schematic diagram of the importance of input parameters of the incinerator of the embodiment of the present application;

[0085] Figure 6 It is a schematic diagram of the number of iterations of reinforcement learning in an embodiment of the present application;

[0086] Figure 7 Schematic diagram of the fitness value of particles in the embodiment of the present application. DETAILED DESCRIPTION

[0087] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0088] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0089] Please refer to Figure 1 , Figure 1 The flowchart of the intelligent optimization method of the operation of the garbage combustion equipment in some embodiments of the present application. The intelligent optimization method of the operation of the garbage combustion equipment is used in terminal devices, such as computers, mobile phones, etc. The intelligent optimization method of the operation of the garbage combustion equipment includes the following steps:

[0090] S11, obtaining a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set;

[0091] S12, obtaining key parameters of a preset gradient boosting tree, and training the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model;

[0092] S13, calculating and obtaining an input variable importance index according to the updated gradient boosting tree model, and comparing the input variable importance index with a preset optimized input variable threshold to obtain an optimized input variable;

[0093] S14, establishing an initial thermal efficiency prediction model according to the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and obtaining the thermal efficiency prediction model through a preset particle swarm optimization algorithm;

[0094] S15. Processing the thermal efficiency prediction model through a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables.

[0095] It should be noted that the operation of the garbage combustion equipment is a relatively complex process, so it involves many factors. The neural network model is needed in the data processing process. First, a preset number of historical operation data of the garbage combustion equipment must be obtained, including historical input variable data and historical output variable data. There are multiple historical input variable data and multiple historical output variable data. There is a corresponding relationship between the historical input data and the historical output variable data. In order to better utilize the historical data, classification is performed to obtain a training data set and a test data set, and the key parameters of the preset gradient boosting tree are obtained. After training, an updated gradient boosting tree model is obtained. There will be some factors in the input variables that have an insignificant effect. Therefore, it is necessary to calculate the input variable importance index through the updated gradient boosting tree model, and then obtain the optimized input variable according to the input variable importance index threshold comparison. The optimized input variable has multiple influencing factors, but the number of influencing factors is less than the number of influencing factors in the input variable data, and then the thermal efficiency prediction model is obtained through the preset multi-layer BP neural network and the preset particle swarm optimization algorithm; the thermal efficiency prediction model is processed according to the preset reinforcement learning algorithm to obtain the optimal input variable feature data, and then the adjustment target of the input variable is clarified.

[0096] Please refer to Figure 2 , Figure 2 The present invention provides a flowchart of obtaining a training data set and a test data set in an intelligent optimization method for the operation of a garbage combustion device provided by an embodiment of the present invention. According to an embodiment of the present invention, the acquisition of a preset amount of historical operation data of the garbage combustion device includes historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set, specifically including:

[0097] S21, establishing a corresponding relationship between historical input variable data and corresponding historical output variable data, and grouping them according to a preset ratio to obtain a training data set and a test data set;

[0098] S22, the training data set includes training history input variable data and corresponding training history output variable data;

[0099] S23. The test data set includes test history input variable data and corresponding test history output variable data.

[0100] It should be noted that the historical input variable data is the independent variable, and the historical output variable data is the dependent variable, so there is a corresponding relationship between the two. In this embodiment, the historical input variable data includes the historical feeder speed data, the historical main steam flow data, the historical main steam temperature data, the historical main steam pressure data, the historical furnace temperature data, the historical feed water temperature data, the historical feed water flow data, the historical primary air temperature data, the historical induced draft fan frequency data, the historical secondary air volume data, the historical exhaust gas temperature data, the historical primary air volume data of the first grate, the historical primary air volume data of the second grate, and the historical primary air volume data of the third grate. , historical primary air volume data of the 4th grate, historical primary air volume data of the 5th grate, historical operating speed data of the 1st grate, historical operating speed data of the 2nd grate, historical operating speed data of the 3rd grate, historical operating speed data of the 4th grate and historical operating speed data of the 5th grate; historical output variable data include historical flue gas oxygen content data, historical furnace negative pressure data, historical furnace outlet temperature data and historical unit steam volume data; the preset ratio can be set according to user needs. In this embodiment, the ratio of the corresponding data number in the training data set to the corresponding data number in the test data set is 7:3.

[0101] The combustion parameters (i.e., incinerator combustion parameters) in the combustion process of this embodiment have 21 input variables, and the output has 4 variables (i.e., flue gas oxygen content, furnace negative pressure, furnace outlet temperature, and unit steam volume). This embodiment expects to adjust the incinerator combustion parameters so that the unit steam volume data is maximized, while the flue gas oxygen content, furnace negative pressure, and furnace outlet temperature are maintained within a reasonable range. However, there are 21 incinerator combustion parameters. If all parameters are used, the subsequent training time of neural networks, reinforcement learning, etc. will be lengthened, that is, the amount of calculation is large. This implementation expects to reduce the incinerator parameters to reduce the amount of calculation.

[0102] Please refer to Figure 3 , Figure 3 The flowchart of obtaining an updated gradient boosting tree model of an intelligent optimization method for the operation of a garbage combustion device provided by an embodiment of the present application. According to an embodiment of the present invention, the key parameters of the preset gradient boosting tree are obtained, and the gradient boosting tree is trained according to the training data set, the test data set and the key parameters to obtain the updated gradient boosting tree model, specifically including:

[0103] S31, obtaining key parameters of the preset gradient boosting tree, including the number of decision trees, learning rate data, and maximum depth data of the decision tree;

[0104] S32, obtaining the average value of the training history output variable data and marking it as the model initial value;

[0105] S33, iterating according to the model initial value, the training history input variable data and the corresponding training history output variable data in combination with the number of decision trees, and testing with the test history input variable data and the corresponding test history output variable data to obtain an updated gradient boosting tree model.

[0106] It should be noted that the number of decision trees refers to the number of decision trees to be generated, that is, the number of iterations of the entire training process; the learning rate data can control how much influence each newly generated decision tree can have on the part that was not predicted correctly before (residual); the maximum depth data of the decision tree, how deep the decision tree can grow at most, that is, the maximum depth of the decision tree; the average value of the training history output variable data is to obtain the average value of the historical flue gas oxygen content data, historical furnace negative pressure data, historical furnace outlet temperature data and historical unit steam volume data in the training data set respectively; in the iterative process, the residual needs to be calculated, and the difference obtained by subtracting the value predicted by the gradient boosting tree from the actual target value is the residual to be processed in this round; at the same time, a new decision tree will be trained, and a new decision tree will be trained according to the historical input variable data or the training history input variable data, the corresponding residual and the maximum depth data of the decision tree, and then the new prediction value is obtained by adding the predicted value of the gradient boosting tree model to the learning rate data multiplied by the predicted value of the new decision tree, and the updated gradient boosting tree model is obtained after the number of decision trees is cycled.

[0107] Please refer to Figure 4 and Figure 5 , Figure 4 This is a flowchart of obtaining optimized input variables of an intelligent optimization method for the operation of a garbage combustion device provided by an embodiment of the present application. The input variable importance index is calculated according to the updated gradient boosting tree model, and the input variable importance index is compared with a preset optimized input variable threshold to obtain the optimized input variable, specifically including:

[0108] S41, obtaining an input variable importance index of each input variable through gradient boosting tree model processing;

[0109] S42, calculating the sum of the importance indexes of the input variables among the input variables, and recording it as the sum of importance;

[0110] S43, dividing the input variable importance index of each input variable by the total importance to obtain the input variable importance ratio data of each input variable;

[0111] S44, sorting the important proportion data of the input variables from large to small;

[0112] S45. Select input variables that are greater than a preset importance threshold and mark them as optimized input variables.

[0113] It should be noted that, in this embodiment, the initial value of the importance index of each input variable is set to 0, and then the updated gradient boosting tree model is traversed in a preset manner to calculate the contribution of each input variable to reducing the error, that is, the input variable importance index corresponding to each input variable, and then the important proportion of the input variable data is calculated and the threshold is compared to obtain the optimized input variable, that is, the input variable that has a greater impact on the prediction result.

[0114] refer to Figure 5 , this embodiment sets a preset importance threshold for the importance percentage of the 21 parameters of the incinerator, and marks the input variables greater than the preset importance threshold as optimized input variables. Figure 5 This embodiment reduces the above 21 variables to 13 variables, namely: primary air flow rate of left unit 1 (Nm3 / h), primary air flow rate of left unit 2 (Nm3 / h), primary air flow rate of left unit 3 (Nm3 / h), primary air flow rate of left unit 4 (Nm3 / h), primary air flow rate of left unit 5 (Nm3 / h), left primary fan suction pipe temperature measurement (°C), secondary fan frequency (%), feeder speed (mm / s), unit conveying speed (mm / s), 2 unit conveying speed (mm / s), 3 unit conveying speed (mm / s), 4 unit conveying speed (mm / s), 5 unit conveying speed (mm / s).

[0115] This embodiment first analyzes the parameters in the combustion process through the gradient boosting tree model, screens out important parameters, thereby reducing the number of parameters to be optimized, thereby reducing the number of input parameters of the subsequent multi-layer BP neural network, thereby reducing the amount of training. Further, in the subsequent reinforcement learning, because the number of parameters to be optimized is reduced, it is also beneficial to reduce the selection space of the reinforcement learning action and reduce the number of training times. This implementation obtains important parameters through the gradient boosting tree model, on the one hand, reduces the number of parameters to be optimized, and on the other hand, even if the number of parameters to be optimized is reduced, the combustion process of the combustion equipment can be well predicted through the BP neural network. Due to the nonlinear relationship between the parameters to be optimized (i.e. the parameters of the above-mentioned incinerator) and the output variables (i.e. the flue gas oxygen content, furnace negative pressure, furnace outlet temperature and unit steam volume) in the combustion process (if dimensionality reduction is not performed, there are too many parameters to be optimized, the required calculation amount is large, and it will also lead to falling into the local optimal solution during optimization), this embodiment first establishes the contribution of the parameters to be optimized in the combustion process to the output variables through the gradient boosting tree model, thereby screening the parameters to be optimized with greater contribution, thereby reducing the number of input variables of the subsequent BP neural network model and the selection space of the action of reinforcement learning, which can greatly reduce the amount of training. On the other hand, it can also reduce the reinforcement learning action selection space caused by the small contribution of the parameters to be optimized, and the local optimal solution caused by the learning of the parameters to be optimized with small contribution.

[0116] According to an embodiment of the present invention, the initial thermal efficiency prediction model is established based on the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and the thermal efficiency prediction model is obtained by processing with a preset particle swarm optimization algorithm, specifically including:

[0117] The initial thermal efficiency prediction model is established by presetting a multi-layer BP neural network combined with optimized input variable data and output variable data;

[0118] Obtain the predicted value of the output variable and the actual value of the output variable, calculate the mean square error based on the predicted value of the output variable and the actual value of the output variable, and mark it as the fitness value;

[0119] The initial thermal efficiency prediction model is optimized in a preset manner according to the fitness value to obtain a thermal efficiency prediction model.

[0120] It should be noted that the BP neural network is a feedforward neural network widely used in the field of machine learning. It uses a back propagation algorithm to adjust the connection weights between neurons to minimize the error between the predicted output and the actual output. The multi-layer BP neural network includes an input layer, a hidden layer (which can be one or more layers) and an output layer. The input layer receives external data, which undergoes complex nonlinear transformations through the hidden layer, and finally obtains the prediction results in the output layer; in this embodiment, the input layer refers to the data corresponding to the optimized input variables, and the output layer refers to the output variable data corresponding to the output variables; the number of neurons and the number of layers in the hidden layer can be adjusted according to the actual situation, and the network structure that is most suitable for modeling the operation data of the garbage incineration equipment is found through continuous trials and optimization; the particle swarm optimization algorithm is an optimization algorithm based on swarm intelligence. In this embodiment, the fitness value is obtained by calculating the mean square error, and the mean square error (MSE) is an indicator for measuring the "mean square error". It is used to evaluate the performance of the prediction model. It calculates the average value of the square of the difference (error) between the predicted value and the true value. The smaller the fitness value, the better the corresponding neural network parameters. The preset method refers to the PSO update mechanism and particle optimal value update method commonly used in the particle swarm optimization algorithm. The thermal efficiency prediction model is obtained when the mean square error reaches the set accuracy target.

[0121] refer to Figure 7 , Figure 7 is the fitness value of the particle in this embodiment, that is, after 200 iterations, the fitness value of the particle drops to the minimum value, and the BP neural parameter is optimal at this time.

[0122] According to an embodiment of the present invention, the thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables, specifically including:

[0123] S51, initializing the input variables to be optimized;

[0124] The input variables to be optimized in this embodiment are the above-mentioned 13 incinerator combustion parameters.

[0125] S52, the agent obtains the predicted output variable according to the input variable to be optimized;

[0126] The intelligent agent of this embodiment is the thermal efficiency prediction model;

[0127] S53, obtaining the output variable reward value and the sum of the output variable reward values ​​of each output variable data respectively through a preset reward function; the intelligent agent of the reinforcement learning is the thermal efficiency prediction model;

[0128] S54. Adjust the optimized input variable data according to the sum of the output variable reward values, and then re-execute steps S51-S53 until the optimal input variable characteristic data is obtained.

[0129] It should be noted that the requirements for output variables are generally relatively fixed. In this embodiment, the output variable standard data include standard data of flue gas oxygen content (data variation range), standard data of furnace negative pressure (data variation range), standard data of furnace outlet temperature (data variation range) and standard data of unit steam volume (data variation range). The standard data of flue gas oxygen content is [3, 8], the standard data of furnace negative pressure is [20, 70], and the standard data of furnace outlet temperature is [950, 1150]. The unit steam volume data refers to the steam volume data generated after the unit weight of garbage is burned. The unit steam volume standard data is [1.5, 3.5]. When the flue gas oxygen content data, the furnace negative pressure data and the furnace outlet temperature data are within the standard data range, the reward value is -2, indicating that the current state is acceptable, but there is still room for improvement. Otherwise, the reward value is 0, indicating that the current state deviates greatly from the ideal range and needs to be adjusted. The unit steam volume data needs to be calculated in a preset manner according to its average value and standard deviation to obtain its reward value, as detailed below.

[0130] The reward values ​​corresponding to the flue gas oxygen content data, furnace negative pressure data, furnace outlet temperature data and unit steam volume data are added together to obtain the sum of the output variable reward values. When the sum of the reward values ​​no longer changes, it can be considered that the optimal solution has been obtained. By continuously trying different combinations of input variable data, the BP neural network algorithm gradually learns the strategy that can maximize the reward value, that is, the optimal input variable characteristic data corresponding to the optimal input variable.

[0131] refer to Figure 6 In this embodiment, if the total reward value does not change after 60 iterations, the iteration can be terminated at the 80th time, thereby obtaining the optimal input variable feature data corresponding to the termination of the iteration.

[0132] This embodiment needs to optimize to maximize the unit steam volume. Therefore, this embodiment uses the first three output variables to calculate the reward value, and the reward value range is as follows:

[0133] 1. Oxygen content of boiler outlet flue gas (%) range: 3-8

[0134] 2. The first channel flue gas outlet pressure (furnace outlet negative pressure (Pa) variation range: 20-70

[0135] 3. The average temperature of the upper and middle parts of the furnace (furnace temperature) (℃) range: 950----1150. If the value of the output variable is within the given upper and lower limits: then reward = -2, otherwise reward = 0;

[0136] The fourth variable (steam production) is maximized as the optimization target:

[0137] According to professional heat balance calculation: the maximum steam output is 3.5 and the minimum is 1.5; the reward values ​​are as follows:

[0138] Reward consists of four parts: reward 1, reward 2, reward 3, reward 4

[0139] Reward_tota l=reward 1+reward 2+reward 3+reward 4

[0140] Reward4 = -norm(Z4-3.5); Z4 is the predicted value of the fourth variable (steam production), that is, the predicted value output by the agent based on the input parameters;

[0141] When Z4<(1.5-2.0871) / 0.1574, Reward_tota l=-10; when Z4=(3-2.0871) / 0.1574, Reward_tota l=0.

[0142] (Z4-2.0871) / 0.1574mean=2.0871,std=0.1574

[0143] All calculations are done using normalized values: If the predicted value of the fourth item is Z4, then enter (Z4-2.0871) / 0.1574

[0144] goa l State=(3-2.0871) / 0.1574=5.8;

[0145] reward4=-norm((Z4-2.0871) / 0.1574-goal State);

[0146] Minimum reward4 value: -norm((1.5-2.0871) / 0.1574-goal State) = -norm(-3.72-5.8) = -9.53

[0147] Maximum value of reward4: -norm((3-2.0871) / 0.1574-goal State)=0

[0148] Oxygen content of flue gas at furnace outlet, reward 1:

[0149] When Z1 (normalized value) > Z1 max (normalized value) or Z1 (normalized value) < Z1 min (normalized value), Reward 1 = -10; where Z1 is the predicted value of the oxygen content of the flue gas at the furnace outlet.

[0150] When Z1 min (normalized value) < Z1 (normalized value) < Z1 max (normalized value), Reward1 = 0.

[0151] reward 2:

[0152] When Z2 (normalized value) > Z2 max (normalized value) or Z2 < Z2 min (normalized value), Reward 2 = -10;

[0153] When Z2 min (normalized value) < Z2 (normalized value) < Z2 max (normalized value), Reward2 = 0. Where Z2 is the predicted value of the flue gas outlet pressure of the first channel.

[0154] reward 3:

[0155] When Z3 (normalized value) > Z3 max (normalized value) or Z3 < Z3 min (normalized value), Reward 3 = -10;

[0156] When Z3 min (normalized value) < Z3 (normalized value) < Z3 max (normalized value), Reward3 = 0. Where Z3 is the predicted value of the furnace temperature.

[0157] Specifically, according to the oxygen content of the flue gas at the furnace outlet, the negative pressure at the furnace outlet, the furnace temperature, and the steam output output by the agent, calculate their corresponding reward values Reward 1, Reward 2, Reward 3, and Reward 4 respectively, and sum the reward values Reward 1, Reward 2, Reward 3, and Reward 4 as the total reward value Reward total.

[0158] Specifically, the calculation of Reward4 is as follows: when Z4 < (1.5 - 2.0871) / 0.1574, Reward_total = -10; when Z4 = (3 - 2.0871) / 0.1574, Reward_total = 0; otherwise goal State = (3 - 2.0871) / 0.1574 = 5.8; reward4 = -norm((Z4 - 2.0871) / 0.1574 - goal State).

[0159] In calculating the reward value, this embodiment considers steam production on the one hand, and also considers other output variables (i.e., oxygen content of flue gas at furnace outlet, negative pressure at furnace outlet, furnace temperature) as the total reward value, so that when maximizing the output of the incinerator, the incinerator can be guaranteed to work in a normal state, and the incinerator cannot work in a normal state due to reinforcement learning when maximizing the reward value of steam production. Furthermore, in this embodiment, when the predicted steam production is no longer within the preset range, the total reward value is directly set to 0, indicating that optimization is required.

[0160] Through the method of this embodiment, the final optimized variables are:

[0161] (1) Input variable optimization value:

[0162] 1. Left unit 1 primary air flow (Nm3 / h) = 770.35 Reasonable change range: 500-----1800

[0163] 2. Left unit 2 primary air flow (Nm3 / h) = 540.8 Reasonable change range: 500-----1500

[0164] 3. Left unit 3 primary air flow (Nm3 / h) = 7160.24 Reasonable change range: 3000------10000

[0165] 4. Left unit 4 primary air flow (Nm3 / h) = 4322.86 Reasonable change range: 2000------10000

[0166] 5. Left unit 5 primary air flow (Nm3 / h) = 2556.49 Reasonable change range: 2000------5000

[0167] 6. Temperature measurement of the left primary fan suction pipe (℃) = 170.69 Reasonable range of change: 150----230

[0168] 7. Secondary fan frequency (%) = 47.59 Reasonable variation range: 30------86

[0169] 8. Feeder speed mm / s=0.74 Reasonable variation range: 0.3----0.8

[0170] 9.1 Unit transmission speed (mm / s) = 0.75 Reasonable variation range: 0.5----2

[0171] 10.2 Unit transmission speed (mm / s) = 1.56 Reasonable variation range: 0.7----2.5

[0172] 11.3 Unit transmission speed (mm / s) = 1.38 Reasonable variation range: 0.3----1.5

[0173] 12.4 Unit transmission speed (mm / s) = 1.17 Reasonable variation range: 0.3----1.2

[0174] 13.5 Unit transmission speed (mm / s) = 0.43 Reasonable variation range: 0.35----1.3

[0175] (2) Optimal value of output variable:

[0176] 1. Oxygen content of boiler outlet flue gas (%) = 6.92 Reasonable variation range: 3----8

[0177] 2. The first channel flue gas outlet pressure (furnace outlet negative pressure, Pa) = 68.36 Reasonable range of change: 20---70

[0178] 3. Average temperature of the upper middle part of the furnace (furnace temperature) (℃) = 1086.86 Reasonable range of change: 950----1150

[0179] 4. Gas production per unit of garbage = 2.41

[0180] The current average value in DCS data is 2.01, an increase of about 20%.

[0181] The present invention also discloses an intelligent optimization system for the operation of a garbage combustion device, comprising a memory and a processor, wherein the memory stores an intelligent optimization method program for the operation of the garbage combustion device, and when the intelligent optimization method program for the operation of the garbage combustion device is executed by the processor, the following steps are implemented:

[0182] Obtaining a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set;

[0183] Obtain the key parameters of the preset gradient boosting tree, and train the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model;

[0184] Obtaining an input variable importance index according to the updated gradient boosting tree model, and comparing the input variable importance index with a preset optimized input variable threshold to obtain an optimized input variable;

[0185] An initial thermal efficiency prediction model is established based on the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and the thermal efficiency prediction model is obtained through processing with a preset particle swarm optimization algorithm;

[0186] The thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables.

[0187] It should be noted that the operation of the garbage combustion equipment is a relatively complex process, so it involves many factors. The neural network model is needed in the data processing process. First, a preset number of historical operation data of the garbage combustion equipment must be obtained, including historical input variable data and historical output variable data. There are multiple historical input variable data and multiple historical output variable data. There is a corresponding relationship between the historical input data and the historical output variable data. In order to better utilize the historical data, classification is performed to obtain a training data set and a test data set, and the key parameters of the preset gradient boosting tree are obtained. After training, an updated gradient boosting tree model is obtained. There will be some factors in the input variables that have an insignificant effect. Therefore, it is necessary to calculate the input variable importance index through the updated gradient boosting tree model, and then obtain the optimized input variable according to the input variable importance index threshold comparison. The optimized input variable has multiple influencing factors, but the number of influencing factors is less than the number of influencing factors in the input variable data, and then the thermal efficiency prediction model is obtained through the preset multi-layer BP neural network and the preset particle swarm optimization algorithm; the thermal efficiency prediction model is processed according to the preset reinforcement learning algorithm to obtain the optimal input variable feature data, and then the adjustment target of the input variable is clarified.

[0188] According to an embodiment of the present invention, the acquisition of a preset amount of historical operation data of the garbage incineration equipment includes historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data and classifying them to obtain a training data set and a test data set, specifically including:

[0189] Establish a corresponding relationship between historical input variable data and corresponding historical output variable data, and group them according to a preset ratio to obtain a training data set and a test data set;

[0190] The training data set includes training history input variable data and corresponding training history output variable data;

[0191] The test data set includes test history input variable data and corresponding test history output variable data.

[0192] It should be noted that the historical input variable data is the independent variable, and the historical output variable data is the dependent variable, so there is a corresponding relationship between the two. In this embodiment, the historical input variable data includes the historical feeder speed data, the historical main steam flow data, the historical main steam temperature data, the historical main steam pressure data, the historical furnace temperature data, the historical feed water temperature data, the historical feed water flow data, the historical primary air temperature data, the historical induced draft fan frequency data, the historical secondary air volume data, the historical exhaust gas temperature data, the historical primary air volume data of the first grate, the historical primary air volume data of the second grate, and the historical primary air volume data of the third grate. , historical primary air volume data of the 4th grate, historical primary air volume data of the 5th grate, historical operating speed data of the 1st grate, historical operating speed data of the 2nd grate, historical operating speed data of the 3rd grate, historical operating speed data of the 4th grate and historical operating speed data of the 5th grate; historical output variable data include historical flue gas oxygen content data, historical furnace negative pressure data, historical furnace outlet temperature data and historical unit steam volume data; the preset ratio can be set according to user needs. In this embodiment, the ratio of the corresponding data number in the training data set to the corresponding data number in the test data set is 7:3.

[0193] According to an embodiment of the present invention, the step of obtaining the key parameters of a preset gradient boosting tree and training the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model specifically includes:

[0194] Get the key parameters of the preset gradient boosting tree, including the number of decision trees, learning rate data, and maximum depth data of the decision tree;

[0195] Get the average value of the training history output variable data and mark it as the model initial value;

[0196] Iterate according to the model initial value, training history input variable data and corresponding training history output variable data combined with the number of decision trees, and obtain an updated gradient boosting tree model after testing the test history input variable data and the corresponding test history output variable data.

[0197] It should be noted that the number of decision trees refers to the number of decision trees to be generated, that is, the number of iterations of the entire training process; the learning rate data can control how much influence each newly generated decision tree can have on the part that was not predicted correctly before (residual); the maximum depth data of the decision tree, how deep the decision tree can grow at most, that is, the maximum depth of the decision tree; the average value of the training history output variable data is to obtain the average value of the historical flue gas oxygen content data, historical furnace negative pressure data, historical furnace outlet temperature data and historical unit steam volume data in the training data set respectively; in the iterative process, the residual needs to be calculated, and the difference obtained by subtracting the value predicted by the gradient boosting tree from the actual target value is the residual to be processed in this round; at the same time, a new decision tree will be trained, and a new decision tree will be trained according to the historical input variable data or the training history input variable data, the corresponding residual and the maximum depth data of the decision tree, and then the new prediction value is obtained by adding the predicted value of the gradient boosting tree model to the learning rate data multiplied by the predicted value of the new decision tree, and the updated gradient boosting tree model is obtained after the number of decision trees is cycled.

[0198] The step of calculating the input variable importance index according to the updated gradient boosting tree model and comparing the input variable importance index with a preset optimized input variable threshold to obtain the optimized input variable specifically includes:

[0199] The input variable importance index of each input variable is obtained through gradient boosting tree model processing;

[0200] Calculate the sum of the importance indexes of the input variables in each input variable and record it as the sum of importance;

[0201] The input variable importance index of each input variable is divided by the total importance to obtain the input variable importance ratio data of each input variable;

[0202] Sort the important proportion data of the input variables from large to small;

[0203] Select input variables with a value greater than the preset importance threshold and mark them as optimized input variables.

[0204] It should be noted that, in this embodiment, the initial value of the importance index of each input variable is set to 0, and then the updated gradient boosting tree model is traversed in a preset manner to calculate the contribution of each input variable to reducing the error, that is, the input variable importance index corresponding to each input variable, and then the important proportion of the input variable data is calculated and the threshold is compared to obtain the optimized input variable, that is, the input variable that has a greater impact on the prediction result.

[0205] According to an embodiment of the present invention, the initial thermal efficiency prediction model is established based on the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and the thermal efficiency prediction model is obtained by processing with a preset particle swarm optimization algorithm, specifically including:

[0206] The initial thermal efficiency prediction model is established by presetting a multi-layer BP neural network combined with optimized input variable data and output variable data;

[0207] Obtain the predicted value of the output variable and the actual value of the output variable, calculate the mean square error based on the predicted value of the output variable and the actual value of the output variable, and mark it as the fitness value;

[0208] The initial thermal efficiency prediction model is optimized in a preset manner according to the fitness value to obtain a thermal efficiency prediction model.

[0209] It should be noted that the BP neural network is a feedforward neural network widely used in the field of machine learning. It uses a back propagation algorithm to adjust the connection weights between neurons to minimize the error between the predicted output and the actual output. The multi-layer BP neural network includes an input layer, a hidden layer (which can be one or more layers) and an output layer. The input layer receives external data, which undergoes complex nonlinear transformations through the hidden layer, and finally obtains the prediction results in the output layer; in this embodiment, the input layer refers to the data corresponding to the optimized input variables, and the output layer refers to the output variable data corresponding to the output variables; the number of neurons and the number of layers in the hidden layer can be adjusted according to the actual situation, and the network structure that is most suitable for modeling the operation data of the garbage incineration equipment is found through continuous trials and optimization; the particle swarm optimization algorithm is an optimization algorithm based on swarm intelligence. In this embodiment, the fitness value is obtained by calculating the mean square error, and the mean square error (MSE) is an indicator for measuring the "mean square error". It is used to evaluate the performance of the prediction model. It calculates the average value of the square of the difference (error) between the predicted value and the true value. The smaller the fitness value, the better the corresponding neural network parameters. The preset method refers to the PSO update mechanism and particle optimal value update method commonly used in the particle swarm optimization algorithm. The thermal efficiency prediction model is obtained when the mean square error reaches the set accuracy target.

[0210] According to an embodiment of the present invention, the thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables, specifically including:

[0211] Obtain output variable standard data and output variable real-time data;

[0212] The output variable reward value and the sum of the output variable reward values ​​of each output variable data are obtained by processing with a preset reward function;

[0213] The thermal efficiency prediction model adjusts the optimized input variable data corresponding to the output variable reward value and the sum of the output variable reward values ​​in a preset manner to obtain the optimal input variable characteristic data.

[0214] It should be noted that the requirements for output variables are generally relatively fixed. In this embodiment, the standard data of output variables include standard data of flue gas oxygen content, standard data of furnace negative pressure, standard data of furnace outlet temperature and standard data of unit steam volume. The standard data of flue gas oxygen content is [3,8], the standard data of furnace negative pressure is [20,70], and the standard data of furnace outlet temperature is [950,1150]. The unit steam volume data refers to the steam volume data generated after the unit weight of garbage is burned. The standard data of unit steam volume is [1.5,3.5]. When the flue gas oxygen content data, furnace negative pressure data and furnace outlet temperature data are within the standard data range, the reward value is If the reward value is -2, it means that the current state is acceptable, but there is still room for improvement. Otherwise, the reward value is 0, which means that the current state deviates greatly from the ideal range and needs to be adjusted. The reward value of the unit steam volume data is calculated in a preset way according to the reward values ​​of the flue gas oxygen content data, the furnace negative pressure data and the furnace outlet temperature data. The reward values ​​corresponding to the flue gas oxygen content data, the furnace negative pressure data, the furnace outlet temperature data and the unit steam volume data are added to obtain the sum of the output variable reward values. By continuously trying different combinations of input variable data, the BP neural network algorithm gradually learns the strategy that can maximize the reward value, that is, the optimal input variable characteristic data corresponding to the optimal input variable.

[0215] A third aspect of the present invention provides a readable storage medium, which includes a program for an intelligent optimization method for the operation of a waste incineration device. When the program for the intelligent optimization method for the operation of a waste incineration device is executed by a processor, the steps of an intelligent optimization method for the operation of a waste incineration device as described in any one of the above items are implemented.

[0216] The present invention discloses an intelligent optimization method, system and medium for the operation of a garbage combustion device. The method obtains a preset number of historical input variable data and historical output variable data of the garbage combustion device, and then classifies the historical input variable data and the corresponding historical output variable data to obtain a training data set and a test data set, obtains key parameters of a gradient boosting tree and trains to obtain an updated gradient boosting tree model, calculates the input variable importance index according to the updated gradient boosting tree model and obtains the optimized input variable through threshold comparison, obtains a thermal efficiency prediction model through multi-layer BP neural network and particle swarm optimization algorithm processing, and obtains the optimal input variable feature data after the thermal efficiency prediction model is processed by a reinforcement learning algorithm; thereby, the intelligent optimization of the garbage combustion device is realized through the gradient boosting tree model, the multi-layer BP neural network and the particle swarm optimization algorithm technology, and the power generation cost of the garbage combustion device is reduced.

[0217] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0218] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0219] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0220] Those skilled in the art can understand that all or part of the steps of the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0221] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

Claims

1. An intelligent optimization method for the operation of a garbage combustion device, characterized in that: include: Obtaining a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set; Obtain the key parameters of the preset gradient boosting tree, and train the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model; Obtaining an input variable importance index according to the updated gradient boosting tree model, and comparing the input variable importance index with a preset optimized input variable threshold to obtain an optimized input variable; An initial thermal efficiency prediction model is established based on the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and the thermal efficiency prediction model is obtained through processing with a preset particle swarm optimization algorithm; The thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables.

2. The intelligent optimization method for the operation of a garbage incineration device according to claim 1 is characterized in that: The method of obtaining a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set, specifically includes: Establish a corresponding relationship between historical input variable data and corresponding historical output variable data, and group them according to a preset ratio to obtain a training data set and a test data set; The training data set includes training history input variable data and corresponding training history output variable data; The test data set includes test history input variable data and corresponding test history output variable data.

3. The intelligent optimization method for the operation of a garbage incineration device according to claim 2 is characterized in that: The step of obtaining the key parameters of the preset gradient boosting tree and training the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model specifically includes: Get the key parameters of the preset gradient boosting tree, including the number of decision trees, learning rate data, and maximum depth data of the decision tree; Get the average value of the training history output variable data and mark it as the model initial value; Iterate according to the model initial value, training history input variable data and corresponding training history output variable data combined with the number of decision trees, and obtain an updated gradient boosting tree model after testing the test history input variable data and the corresponding test history output variable data.

4. The intelligent optimization method for the operation of a garbage combustion equipment according to claim 3 is characterized in that: The step of calculating the input variable importance index according to the updated gradient boosting tree model and comparing the input variable importance index with a preset optimized input variable threshold to obtain the optimized input variable specifically includes: The input variable importance index of each input variable is obtained through gradient boosting tree model processing; Calculate the sum of the importance indexes of the input variables in each input variable and record it as the sum of importance; The input variable importance index of each input variable is divided by the total importance to obtain the input variable importance ratio data of each input variable; Sort the important proportion data of the input variables from large to small; Select input variables with a value greater than the preset importance threshold and mark them as optimized input variables.

5. The intelligent optimization method for the operation of a garbage combustion equipment according to claim 4 is characterized in that: The method of establishing an initial thermal efficiency prediction model based on the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and obtaining the thermal efficiency prediction model through a preset particle swarm optimization algorithm specifically includes: The initial thermal efficiency prediction model is established by presetting a multi-layer BP neural network combined with optimized input variable data and output variable data; Obtain the predicted value of the output variable and the actual value of the output variable, calculate the mean square error based on the predicted value of the output variable and the actual value of the output variable, and mark it as the fitness value; The initial thermal efficiency prediction model is optimized in a preset manner according to the fitness value to obtain a thermal efficiency prediction model.

6. The intelligent optimization method for the operation of a garbage incineration device according to claim 5, characterized in that: The thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables, specifically including: S51, initializing the input variables to be optimized; S52, the agent obtains the predicted output variable according to the input variable to be optimized; S53, obtaining the output variable reward value and the sum of the output variable reward values ​​of each output variable data respectively through a preset reward function; the intelligent agent of the reinforcement learning is the thermal efficiency prediction model; S54, adjusting the optimized input variable data according to the sum of the output variable reward values, and then re-executing steps S51-S53 until the sum of the reward values ​​no longer changes, at which time the optimal input variable characteristic data is obtained.

7. An intelligent optimization system for the operation of garbage combustion equipment, characterized in that: The invention comprises a memory and a processor, wherein the memory comprises an intelligent optimization method program for the operation of a garbage combustion device, and when the intelligent optimization method program for the operation of the garbage combustion device is executed by the processor, the following steps are implemented: Obtaining a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set; Obtain the key parameters of the preset gradient boosting tree, and train the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model; Obtaining an input variable importance index according to the updated gradient boosting tree model, and comparing the input variable importance index with a preset optimized input variable threshold to obtain an optimized input variable; An initial thermal efficiency prediction model is established based on the optimized input variable data and output variable data corresponding to the optimized input variable in combination with a preset multi-layer BP neural network, and the thermal efficiency prediction model is obtained through processing with a preset particle swarm optimization algorithm; The thermal efficiency prediction model is processed by a preset reinforcement learning algorithm to obtain optimal input variable characteristic data corresponding to the optimized input variables.

8. The intelligent optimization system for the operation of garbage combustion equipment according to claim 7 is characterized in that: The method of obtaining a preset amount of historical operation data of the garbage incineration equipment, including historical input variable data and historical output variable data, establishing a corresponding relationship between the historical input variable data and the corresponding historical output variable data, and classifying them to obtain a training data set and a test data set, specifically includes: Establish a corresponding relationship between historical input variable data and corresponding historical output variable data, and group them according to a preset ratio to obtain a training data set and a test data set; The training data set includes training history input variable data and corresponding training history output variable data; The test data set includes test history input variable data and corresponding test history output variable data.

9. The intelligent optimization system for the operation of garbage combustion equipment according to claim 8, characterized in that: The step of obtaining the key parameters of the preset gradient boosting tree and training the gradient boosting tree according to the training data set, the test data set and the key parameters to obtain an updated gradient boosting tree model specifically includes: Get the key parameters of the preset gradient boosting tree, including the number of decision trees, learning rate data, and maximum depth data of the decision tree; Get the average value of the training history output variable data and mark it as the model initial value; Iterate according to the model initial value, training history input variable data and corresponding training history output variable data combined with the number of decision trees, and obtain an updated gradient boosting tree model after testing the test history input variable data and the corresponding test history output variable data.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a program for an intelligent optimization method for the operation of a waste incineration device. When the program for the intelligent optimization method for the operation of a waste incineration device is executed by a processor, the steps of the intelligent optimization method for the operation of a waste incineration device as described in any one of claims 1 to 6 are implemented.

Images