Self-adaptive combustion parameter control method and device for waste incineration

Abstract

The invention provides a waste incineration self-adaptive combustion parameter control method and device, and relates to the technical field of computers. The method comprises the steps that waste incineration data related to control parameters are obtained; the waste incineration data are decomposed into a plurality of Gaussian distribution components, corresponding data probabilities are obtained, a sub-data set of each Gaussian distribution component is determined based on the data probabilities, and the data probabilities are used for indicating the probability value of any waste incineration data meeting the Gaussian distribution components; feature extraction is conducted on the sub-data sets through a feature extraction model, network parameters in the feature extraction model are updated when it is detected that an error result is larger than an error threshold value, so that a predicted value of the control parameters is output, and the error result is used for representing an error between the specified label and an output result of the feature extraction model; and based on the predicted value, a control strategy of the control parameters is determined, and the control strategy is used for indicating regulation and control of the waste incineration data. The method can improve the regulation and control precision.

Description

Technical Field

[0001] The invention relates to the field of computer technology, and in particular to a method and device for controlling self-adaptive combustion parameters of garbage incineration. Background Art

[0002] As an important method of modern waste treatment, waste incineration can not only effectively reduce the volume of waste, but also realize resource utilization through heat recovery. However, due to the complex composition of waste and large fluctuations in calorific value, traditional incineration parameter control methods are difficult to adapt to changes in real time, which can easily lead to low combustion efficiency, energy waste and excessive pollutant emissions.

[0003] For the parameter control of waste incineration, the traditional method mainly relies on manual experience. Since there are differences in the waste delivered each time, the parameters need to be adjusted at any time, which is not ideal in terms of adjustment efficiency and accuracy.

[0004] At present, deep learning technology is usually used to optimize waste incineration, such as the invention patent CN117308102A, which proposes a control system for efficient, clean and intelligent operation of the entire incinerator process. Specifically, the control process of the incinerator is disassembled into multiple modules, and some key indicators and parameters are predicted using a deep learning model, and finally the parameters of the incinerator are adjusted using the prediction results. However, this solution has the following problems: First, a key indicator depends on an independent model, resulting in high system complexity and redundancy, and independent models need to be trained separately, which requires a lot of time and cost, so the control efficiency is low; second, the model itself has errors, and the strong coupling of multiple independent models is prone to cause the accumulation of errors, resulting in reduced control accuracy of automated parameter adjustment, and strong coupling will lead to poor method transferability.

[0005] Therefore, the existing waste incineration control technology needs to rely on multiple strongly coupled independent models, resulting in low control accuracy and efficiency. Summary of the invention

[0006] In order to solve the above-mentioned problems of the prior art, the present invention provides a method and device for adaptive combustion parameter control of waste incineration, which solves the problem that the existing waste incineration control technology needs to rely on multiple strongly coupled independent models, and can improve the control accuracy and control efficiency.

[0007] In a first aspect, an embodiment of the present application provides a method for adaptively controlling combustion parameters of waste incineration, comprising: Obtain waste incineration data related to control parameters within a specified monitoring period; Decomposing the waste incineration data into a plurality of Gaussian distribution components, and obtaining corresponding data probabilities, and determining a sub-data set of each Gaussian distribution component based on the data probabilities, wherein the data probabilities are used to indicate a probability value that any waste incineration data satisfies the Gaussian distribution component; Extracting features from the sub-dataset using a pre-established feature extraction model, and updating network parameters in the feature extraction model when an error result is detected to be greater than a preset error threshold, so as to output a predicted value of the control parameter, wherein the error result is used to characterize an error between a specified label and an output result of the feature extraction model; Based on the predicted value, a control strategy for the control parameter is determined, and the control strategy is used to indicate and regulate current waste incineration data.

[0008] In a possible implementation, decomposing the waste incineration data into a plurality of Gaussian distribution components, obtaining corresponding data probabilities, and determining a sub-data set of each Gaussian distribution component based on the data probabilities includes: constructing the waste incineration data into a data matrix; Input the data matrix into a Gaussian mixture model containing multiple Gaussian distribution components, assign each element in the data matrix to the Gaussian distribution component, and output the corresponding data probability, wherein the number of the Gaussian distribution components is consistent with the number of data categories of the waste incineration data; Based on the elements and the corresponding data probabilities, a sub-dataset of each Gaussian distribution component is constructed.

[0009] In a possible implementation, the feature extraction model includes a first feature extraction model and a second feature extraction model arranged in sequence, and the first feature extraction model includes a plurality of multi-layer perceptrons; The method of extracting features from the sub-dataset using a pre-established feature extraction model and updating network parameters in the feature extraction model when an error result is detected to be greater than a preset error threshold to output a predicted value of the control parameter comprises: Inputting the sub-datasets into the first feature extraction model, performing feature extraction on the corresponding sub-datasets through the multi-layer perceptron, and using the data feature values ​​output by each of the multi-layer perceptrons as the output of the first feature extraction model; Vertically splicing the data feature values ​​to obtain fusion features; The time feature of the fusion feature is extracted through the second feature extraction model to output the predicted value of the control parameter.

[0010] In a possible implementation, the multilayer perceptron includes a plurality of hidden layers, the feature extraction of the sub-dataset is performed using a pre-established feature extraction model, and when an error result is detected to be greater than a preset error threshold, a network parameter in the feature extraction model is updated to output a predicted value of the control parameter, and further includes: For any multi-layer perceptron, detecting whether an error result of the multi-layer perceptron is greater than a preset error threshold, wherein the error result is used to indicate the difference between the specified label and a predicted value determined by taking the output of the current hidden layer as the data feature value currently output by the multi-layer perceptron; When it is detected that the error result is greater than the error threshold, and the number of nodes in each hidden layer included in the multilayer perceptron is less than the node number threshold, a new node is added to the hidden layer, and the added node satisfies a constraint condition, wherein the node number threshold is determined by the data length of the waste incineration data, and the constraint condition is determined by the error result before adding the node, the output of the last node in the hidden layer before adding the node, the output of the added node, and the error result after adding the node; Based on the specified label, the output weights of the nodes before the added node, and the outputs of the nodes after the added node, the output weights of the nodes in the hidden layer are updated; Based on the updated output weights, the data feature values ​​output by the multilayer perceptron are updated.

[0011] In a possible implementation, the extracting features of the sub-dataset using a pre-established feature extraction model, and updating the network parameters in the feature extraction model when it is detected that the error result is greater than a preset error threshold, so as to output the predicted value of the control parameter, further includes: The output of the hidden layer is determined by the following formula (1): (1) in, is the sub-dataset, represents the set of output weights of each node in the hidden layer before adding the node, represents the number of nodes in the hidden layer before adding the node, is the data length, Indicates the output weight of the jth node in the hidden layer before the increase node, represents the activation function of the jth node, represents the input weight of the jth node, represents the bias of the jth node; The error result is determined by the following formula (2): (2) in, represents the error result before adding the node, Specify a label for the Indicates the predicted value determined by taking the output of the current hidden layer as the data feature value of the current output of the multi-layer perceptron; The constraint condition is determined by the following formula (3): (3) in, represents the error result of the sub-dataset before adding the node, Indicates The output of the node, represents the output of the added node, Indicates the error result after adding nodes; The output weights of each node in the hidden layer are updated using the following formula (4): (4) in, represents the set of output weights of each node in the hidden layer after adding nodes, Represents the output of each node in the hidden layer after adding nodes; is the preset F norm; represents a pseudo-inverse operation; Indicates the value of the independent variable that minimizes the function.

[0012] In a possible implementation, the monitoring period is used to indicate a time interval formed by a current moment and a previous specified time period; and the control strategy for determining the control parameter based on the predicted value includes: Obtaining the observed value of the control parameter at the current moment; Determine a reinforcement learning evaluation value based on the garbage incineration data in the current monitoring period and the control reward value obtained in response to the regulation action on the garbage incineration data, wherein the control reward value includes a reward value corresponding to when the value of the control parameter at the next moment falls within the control target interval of the control parameter after the regulation action is executed, and the reinforcement learning evaluation value is used to evaluate the expected cumulative reward obtained by the executed regulation action; Based on the control action that maximizes the reinforcement learning evaluation value, a control strategy for the control parameter is determined.

[0013] In a possible implementation, the determining the control strategy of the control parameter based on the predicted value further includes: Processing a sub-dataset determined by the waste incineration data obtained after executing the control action through the feature extraction model to obtain a predicted value for the control parameter; Iteratively determine the control action that maximizes the reinforcement learning evaluation value based on the predicted value, the garbage incineration data obtained after executing the control action, and the corresponding control reward value, as the control strategy of the control parameter.

[0014] In a possible implementation, the control parameters include the furnace temperature of the incinerator, and the garbage incineration data include the grate speed, primary air supply, secondary air supply, fan pressure, grate temperature and the furnace temperature at the current moment of the incinerator.

[0015] In a second aspect, the embodiment of the present application provides a garbage incineration adaptive combustion parameter control device, comprising: A data acquisition module, used to acquire waste incineration data related to control parameters within a specified monitoring period; A data decomposition module, used to decompose the waste incineration data into multiple Gaussian distribution components, and obtain corresponding data probabilities, and determine a sub-data set of each Gaussian distribution component based on the data probabilities, wherein the data probability is used to indicate the probability value that any waste incineration data satisfies the Gaussian distribution component; A prediction module, used to extract features from the sub-dataset using a pre-established feature extraction model, and to update network parameters in the feature extraction model when an error result is detected to be greater than a preset error threshold, so as to output a predicted value of the control parameter, wherein the error result is used to characterize an error between a specified label and an output result of the feature extraction model; A strategy generation module is used to determine a control strategy for the control parameter based on the predicted value, wherein the control strategy is used to indicate and regulate the current waste incineration data.

[0016] The method and device for adaptive combustion parameter control of waste incineration provided in the embodiments of the present application can retain the correlation between data by decomposing waste incineration data related to control parameters into multiple Gaussian distribution components, and determining a sub-data set of each Gaussian distribution component based on the corresponding data probability obtained, and then extracting features from the sub-data set through a feature extraction model, and updating the network parameters in the feature extraction model when it is detected that the error result is greater than the error threshold, and outputting the predicted value of the control parameter, so as to determine the control strategy of the control parameter with the help of the predicted value, without relying on multiple strongly coupled independent models, thereby reducing the complexity of data processing, realizing adaptive configuration of the network parameters of the model, and improving the accuracy of model processing, thereby improving the control precision and control efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a method for adaptive combustion parameter control of waste incineration provided in an embodiment of the present application; Figure 2 A schematic diagram of a flow chart of network parameter updating provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a garbage incineration adaptive combustion parameter control device provided in an embodiment of the present application; DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Figure 1 A schematic flow chart of a method for adaptive combustion parameter control of waste incineration provided in an embodiment of the present application, the method comprising steps S101 to S104.

[0020] S101. Obtaining garbage incineration data related to control parameters within a specified monitoring period.

[0021] In the present application, the monitoring period is used to indicate the time interval formed by the current moment and the previous specified time period, so as to obtain the garbage incineration data within the time interval.

[0022] It should be noted that the goal of the adaptive combustion parameter control method for waste incineration provided in this application can be to minimize the emission of harmful substances during the incineration process, such as dioxin compounds (produced by incomplete combustion of chlorine-containing organic matter), carbon monoxide (produced when most substances are incompletely burned), nitrogen oxides (produced when the incineration temperature is too high), and heavy metal vapor (the incineration temperature is too high, causing metals such as mercury, lead, and cadmium to be released into the flue gas in the form of vapor or particles). The key factors in reducing the emission of harmful substances during the incineration process are sufficient oxygen and appropriate temperature. When the incineration temperature is 850°~950°, it can provide a temperature for complete combustion and avoid the generation of nitrogen oxides or heavy metal vapor. This application obtains the prediction results of the control parameters (such as incineration temperature, furnace temperature of the incinerator, etc.) through the current waste incineration data, and determines the control strategy in combination with the prediction results and the waste incineration data.

[0023] In a specific embodiment, the control parameters include the furnace temperature of the incinerator, and the garbage incineration data include the grate speed, primary air supply, secondary air supply, fan pressure, grate temperature and the furnace temperature at the current moment of the incinerator.

[0024] Exemplarily, the control parameter may be the furnace temperature, and the grate speed, the primary air supply, the secondary air supply, the fan pressure, the grate temperature and the furnace temperature at the current moment within the time interval formed by the current moment and the previous specified time period are obtained, and the prediction result about the furnace temperature is determined by the above-mentioned waste incineration data, and the control strategy is determined in combination with the prediction result and the waste incineration data to keep the furnace temperature at 850°~950°.

[0025] S102. Decompose the waste incineration data into multiple Gaussian distribution components, obtain corresponding data probabilities, and determine a sub-data set of each Gaussian distribution component based on the data probabilities, wherein the data probability is used to indicate the probability value that any waste incineration data satisfies the Gaussian distribution component.

[0026] It should be noted that the Gaussian mixture model consists of multiple Gaussian distributions (i.e., Gaussian distribution components) and a set of mixing coefficients. Each Gaussian distribution represents a cluster in the data space, and the mixing coefficient reflects the weight of each Gaussian component in the overall model.

[0027] In the present application, by decomposing the waste incineration data into multiple Gaussian distributions, a Gaussian mixture model is implemented to accurately describe the waste incineration data, and then according to the data probability that different waste incineration data satisfy different Gaussian distributions, a sub-data set of each Gaussian distribution is constructed, thereby achieving data decomposition while retaining the correlation between the data.

[0028] S103. Perform feature extraction on the sub-dataset using a pre-established feature extraction model, and update the network parameters in the feature extraction model when an error result is detected to be greater than a preset error threshold, so as to output a predicted value of the control parameter, wherein the error result is used to characterize the error between the specified label and the output result of the feature extraction model.

[0029] In the present application, the feature extraction model can be composed of one or more neural network models. In the feature extraction process, data features, data associations, etc. in the sub-data set can be extracted, and time features represented by the data can also be extracted, so as to extract the predicted value of the control parameter.

[0030] It should be noted that this application determines the error result of the feature extraction model based on the difference between the output result of the feature extraction model (such as the predicted value of the control parameter) and the specified label. Then, it detects whether the error result is greater than the error threshold. If so, it triggers the update of the network parameters of the feature extraction model, realizes the adaptive adaptation of the model parameters, reduces the error of the model and the model output results, and improves the accuracy of the prediction.

[0031] Exemplarily, the feature extraction model may include a multi-layer perceptron, and the residual (i.e., error result) during the training process of the multi-layer perceptron, especially in the back-propagation stage, represents the difference between the network output and the target value (i.e., the specified label in this application), and is back-propagated through the network to guide the update of the weight parameters.

[0032] S104. Determine a control strategy for the control parameter based on the predicted value, wherein the control strategy is used to indicate and regulate current waste incineration data.

[0033] In this application, the control strategy for controlling the garbage incineration data can be determined by combining the predicted value of the control parameter with reinforcement learning and other methods, so that the value of the control parameter is within the control target range. For example, for the furnace temperature, by adjusting the grate speed, the first air supply, the secondary air supply, the fan pressure, the grate temperature and the furnace temperature at the current moment of the incinerator, a corresponding control strategy is generated to keep the furnace temperature at 850°~950°.

[0034] The method and device for adaptive combustion parameter control of waste incineration provided in the present embodiment can retain the correlation between data by decomposing waste incineration data related to control parameters into multiple Gaussian distribution components, and determining a sub-data set of each Gaussian distribution component based on the corresponding data probability obtained, and then extracting features from the sub-data set through a feature extraction model, and updating the network parameters in the feature extraction model when it is detected that the error result is greater than the error threshold, and outputting the predicted value of the control parameter, so as to determine the control strategy of the control parameter with the help of the predicted value, without relying on multiple strongly coupled independent models, thereby reducing the complexity of data processing, realizing adaptive configuration of the network parameters of the model, and improving the accuracy of model processing, thereby improving the control precision and control efficiency.

[0035] In some embodiments, decomposing the waste incineration data into a plurality of Gaussian distribution components, obtaining corresponding data probabilities, and determining a sub-data set of each Gaussian distribution component based on the data probabilities includes: constructing the waste incineration data into a data matrix; Input the data matrix into a Gaussian mixture model containing multiple Gaussian distribution components, assign each element in the data matrix to the Gaussian distribution component, and output the corresponding data probability, wherein the number of the Gaussian distribution components is consistent with the number of data categories of the waste incineration data; Based on the elements and the corresponding data probabilities, a sub-dataset of each Gaussian distribution component is constructed.

[0036] In this embodiment, since the garbage incineration data may contain multiple indicators and parameters, the value length of the indicator / parameter (i.e., data length) is set to L, and the data matrix is ​​constructed based on the data length and the number of indicators / parameters in the garbage incineration data. For example, the indicators / parameters include the grate speed, primary air supply, secondary air supply, fan pressure, grate temperature, and furnace temperature at the current moment of the incinerator, a total of 6, and the generated The matrix M.

[0037] It should be noted that the Gaussian mixture model is used to convert the matrix into a linear combination of multiple Gaussian distributions (i.e., Gaussian distribution components), expressed as ,in, represents the Gaussian mixture model about the matrix M, represents the number of Gaussian distribution components; represents the mixing coefficient of the kth Gaussian distribution component, which reflects the weight of the corresponding Gaussian distribution component in the overall model and satisfies and ; Represents the probability density function of the kth Gaussian distribution component.

[0038] In this embodiment, the number of Gaussian distribution components is consistent with the number of data categories of the waste incineration data. For example, the waste incineration data includes the grate speed, primary air supply, secondary air supply, fan pressure, grate temperature and furnace temperature at the current moment of the incinerator. There are 6 data categories in total, so the number of Gaussian distribution components is 6.

[0039] More specifically, the probability density function of the kth Gaussian distribution component is expressed as follows: ; in, represents the matrix mean of the kth Gaussian distribution component, represents the row covariance matrix of the kth Gaussian distribution component, represents the column variance matrix of the kth Gaussian distribution component, Represents the trace of the matrix.

[0040] Furthermore, each element in the data matrix is ​​substituted into each Gaussian distribution component in turn to obtain the probability value that different elements satisfy different Gaussian distributions, so as to obtain the corresponding data probability. Then, each element is multiplied by the corresponding data probability (such as the probability value that the element satisfies a certain Gaussian distribution component) to obtain the sub-dataset of the Gaussian distribution component. Among them, the sub-dataset corresponds to the Gaussian distribution component one by one, that is, the number of the sub-dataset is consistent with the number of Gaussian distribution components.

[0041] Therefore, this embodiment achieves data decomposition by allocating elements in the data matrix to each Gaussian distribution component and determining a sub-data set of each Gaussian distribution component based on the element and the corresponding data probability, while being able to retain the correlation between the data.

[0042] In some embodiments, the feature extraction model includes a first feature extraction model and a second feature extraction model arranged in sequence, and the first feature extraction model includes a plurality of multi-layer perceptrons; The method of extracting features from the sub-dataset using a pre-established feature extraction model and updating network parameters in the feature extraction model when an error result is detected to be greater than a preset error threshold to output a predicted value of the control parameter comprises: Inputting the sub-datasets into the first feature extraction model, performing feature extraction on the corresponding sub-datasets through the multi-layer perceptron, and using the data feature values ​​output by each of the multi-layer perceptrons as the output of the first feature extraction model; Vertically splicing the data feature values ​​to obtain fusion features; The time feature of the fusion feature is extracted through the second feature extraction model to output the predicted value of the control parameter.

[0043] In this embodiment, the first feature extraction module is used to extract data features, and the second feature module is used to extract time features. It should be noted that each multi-layer perceptron configured in the first feature extraction model corresponds to a sub-data set, that is, the number of the multi-layer perceptrons is consistent with the number of sub-data sets (that is, the number of Gaussian distribution components). In this regard, the sub-data sets are input into the first feature extraction model, that is, each sub-data set is input into the multi-layer perceptron corresponding to it for feature extraction, and the data feature values ​​of each multi-layer perceptron for its respective sub-data set are used as the output of the first feature extraction model.

[0044] It should be noted that in this embodiment, multiple multi-layer perceptrons are set to process the corresponding sub-data sets, but the multi-layer perceptrons are not independent of each other. Since the output of the hidden layer of any multi-layer perceptron is used as the data feature value output by the multi-layer perceptron, and the data feature value output by other multi-layer perceptrons is used as the output of the first feature extraction model, and then the output is processed by the second feature extraction model to output the predicted value, the error result is determined by the difference between the specified label and the predicted value, and the predicted value is determined by the output of multiple multi-layer perceptrons. By determining whether the error result is greater than the error threshold, the network parameters of the multi-layer perceptron are updated to achieve adaptive adjustment of the model parameter configuration. Therefore, this embodiment overcomes the defect that the existing deep learning method needs to rely on multiple strongly coupled independent models by sharing losses (i.e., error results) and adaptively adjusting the model parameter configuration, reduces the model error and data processing complexity, and improves the accuracy and efficiency of the model, thereby effectively improving the control precision and efficiency of the parameters.

[0045] Furthermore, the data feature values ​​of multiple multi-layer perceptrons output by the first feature extraction model are vertically spliced ​​to obtain fused features. The fused features are extracted by the second feature extraction model to obtain the time features in the data, thereby obtaining the predicted values ​​for the control parameters.

[0046] Exemplarily, the second feature extraction model can be LSTM (Long Short-Term Memory), which is a special recurrent neural network (RNN) designed to solve the gradient vanishing and gradient exploding problems encountered by traditional RNN when processing long-term dependencies.

[0047] Therefore, this embodiment configures a first feature extraction model with multiple multi-layer perceptrons to extract features from the sub-dataset, thereby solving the problem that the existing waste incineration control technology needs to rely on multiple strongly coupled independent models, achieving shared loss and adaptive configuration of model network parameters, and improving the accuracy and efficiency of model processing. Subsequently, by feature splicing and configuring the second feature extraction model to extract time features, the time dependency between data is captured, the accuracy of prediction is improved, and thus the control accuracy and control efficiency are improved.

[0048] In some embodiments, the multilayer perceptron includes a plurality of hidden layers. Figure 2 This is a flow chart of network parameter updating provided in an embodiment of the present application. This embodiment provides a method for updating network parameters, which includes steps S201 to S204.

[0049] S201, for any multilayer perceptron, detecting whether an error result of the multilayer perceptron is greater than a preset error threshold, wherein the error result is used to indicate a difference between the specified label and a predicted value determined by taking the output of the current hidden layer as a data feature value currently output by the multilayer perceptron; S202, when it is detected that the error result is greater than the error threshold, and the number of nodes in each hidden layer included in the multilayer perceptron is less than the node number threshold, adding a new node in the hidden layer, and making the added node meet the constraint condition, wherein the node number threshold is determined by the data length of the waste incineration data, and the constraint condition is determined by the error result before adding the node, the output of the last node in the hidden layer before adding the node, the output of the added node, and the error result after adding the node; S203, based on the specified label, the output weights of the nodes before the added node, and the outputs of the nodes after the added node, updating the output weights of the nodes in the hidden layer; S204: Based on the updated output weights, update the data feature values ​​output by the multilayer perceptron.

[0050] In this embodiment, the multilayer perceptron includes N hidden layers, N is greater than or equal to 1, and each hidden layer includes multiple nodes, and the maximum number of nodes is the data length L. Specifically, if the number of nodes in the hidden layer before adding the node is , the hidden layer performs the following operation, that is, the output of the hidden layer is determined by the following formula (1): (1) in, is the sub-dataset, represents the set of output weights of each node in the hidden layer before adding the node, represents the number of nodes in the hidden layer before adding the node, is the data length, Indicates the output weight of the jth node in the hidden layer before the increase node, represents the activation function of the jth node, represents the input weight of the jth node, represents the bias of the jth node.

[0051] Then, it is detected whether the error result of the multilayer perceptron is greater than the error threshold, wherein the error result is determined by the following formula (2): (2) in, represents the error result before adding the node, Specify a label for the It indicates the predicted value determined by taking the output of the current hidden layer as the data feature value of the current output of the multilayer perceptron.

[0052] It should be noted that the output of the hidden layer in any multi-layer perceptron replaces the data feature value of the multi-layer perceptron output by the first feature extraction model, the output of the first feature extraction model is input into the second feature extraction model, and the predicted value of the control parameter is output, then the predicted value is combined with the specified label to determine the residual (i.e., the error result).

[0053] For example, the output of the hidden layer in the first multi-layer perceptron is As the output of the first feature extraction model One of the components, namely , so that the output of the first feature extraction model The predicted value is obtained through the second feature extraction model, and the error result is determined by the predicted value and the specified label.

[0054] Further, when the residual is greater than the error threshold and the current number of nodes in the hidden layer of the multilayer perceptron has not reached the maximum value L, new nodes are added to the hidden layer according to the constraint conditions.

[0055] Specifically, the constraint condition is determined by the following formula (3): (3) in, represents the error result of the sub-dataset before adding the node, Indicates The output of the last node in the hidden layer before adding the node, represents the output of the added node, Represents the error result after adding nodes.

[0056] Furthermore, based on the specified label, the output weights of the nodes before the added node, and the outputs of the nodes after the added node, the output weights of the nodes in the hidden layer are updated. Specifically, the output weights of the nodes in the hidden layer are updated by the following formula (4): (4) in, represents the set of output weights of each node in the hidden layer after adding nodes, Represents the output of each node in the hidden layer after adding nodes; is the preset F norm. Among them, Indicates a pseudo-inverse operation on the matrix. represents the set of output weights of each node in the hidden layer after adding nodes, Represents the output weight of the nodes added in the hidden layer; Indicates the value of the independent variable that makes the function reach the minimum value, that is, for the independent variable ,make The function gets the minimum value value.

[0057] Furthermore, based on the updated output weights, the data feature values ​​output by the multilayer perceptron are updated.

[0058] Therefore, this embodiment detects whether the error result exceeds the error threshold, thereby triggering the addition of nodes in the hidden layer of the multilayer perceptron, and updating the output weights of the nodes in the hidden layer, as well as the data feature values ​​output by the multilayer perceptron. It does not need to rely on multiple strongly coupled independent models, thereby improving the adaptability and accuracy of the model, being suitable for a variety of complex and real-time changing data environments, and improving the accuracy and efficiency of data processing.

[0059] In some embodiments, the monitoring period is used to indicate a time interval formed by the current moment and a previous specified time period; the control strategy for determining the control parameter based on the predicted value includes: Obtaining the observed value of the control parameter at the current moment; Determine a reinforcement learning evaluation value based on the garbage incineration data in the current monitoring period and the control reward value obtained in response to the regulation action on the garbage incineration data, wherein the control reward value includes a reward value corresponding to when the value of the control parameter at the next moment falls within the control target interval of the control parameter after the regulation action is executed, and the reinforcement learning evaluation value is used to evaluate the expected cumulative reward obtained by the executed regulation action; Based on the control action that maximizes the reinforcement learning evaluation value, a control strategy for the control parameter is determined.

[0060] In this embodiment, the control strategy can be determined by reinforcement learning. Specifically, reinforcement learning is performed by the following formula: ; in, express The state at the moment, specifically the control parameters The observed value at time, such as the furnace temperature at time t ; express The control action executed at any time is used to represent the control of the waste incineration data related to the control parameters, such as ,in, represents the grate speed of the incinerator at time t, Indicates the first gas supply at time t, Indicates the secondary air supply at time t, represents the fan pressure at time t, Indicates the grate temperature at time t. is the attenuation rate, is the learning rate, is a reward function (used to determine the control reward value). For example, the reward function can be The reward value is 1 when 850°~950° is met, and the rest of the reward values ​​are 0. Represents the reinforcement learning evaluation value, that is, The total reward value expected to be obtained under the current state and action.

[0061] Then, each control action is traversed, and the control action with the maximum reinforcement learning evaluation value is selected, and the control strategy corresponding to the control action is determined, and the control strategy keeps the control parameter within the control target range.

[0062] Therefore, this embodiment performs reinforcement learning through the observed values ​​of the control parameters at the current moment and the current waste incineration data, and selects the control action that maximizes the reinforcement learning evaluation value to generate a corresponding control strategy for indicating the control of the waste incineration data, so as to achieve an effective balance between the control efficiency and the control accuracy, thereby achieving reasonable scheduling, effectively improving the control efficiency and accuracy, and maintaining the stability of the control.

[0063] In some embodiments, the control strategy for determining the control parameter based on the predicted value further includes: Processing a sub-dataset determined by the waste incineration data obtained after executing the control action through the feature extraction model to obtain a predicted value for the control parameter; Iteratively determine the control action that maximizes the reinforcement learning evaluation value based on the predicted value, the garbage incineration data obtained after executing the control action, and the corresponding control reward value, as the control strategy of the control parameter.

[0064] In this embodiment, based on the above reinforcement learning, the control action at the next moment is output , that is, the grate speed, the first gas supply, the second gas supply, the fan pressure, and the grate temperature at time t+1, that is, It should be noted that in order to make the value of the control parameter at the next moment within the control target range, As Operation at the moment (real The operation at the moment is set to The prediction results are then output to the Gaussian mixture model and feature extraction model. , and take it as the state of the next moment Then, the prediction result is Iterate so that The value of is maximized, and the resulting control action That is the way to regulate the incinerator parameters, that is, the output control strategy.

[0065] Therefore, this embodiment combines the Gaussian mixture model, feature extraction model and reinforcement learning results for prediction, and iterates the prediction results through reinforcement learning, so as to screen out the control action that maximizes the reinforcement learning evaluation value as a control strategy, thereby improving the accuracy and efficiency of the control action, and ensuring that the value of the control parameter falls within the control target range, thereby improving the reliability and accuracy of the control.

[0066] Figure 3 A schematic diagram of the structure of a garbage incineration adaptive combustion parameter control device provided in an embodiment of the present application, the garbage incineration adaptive combustion parameter control device 300 comprises: The data acquisition module 301 is used to acquire the waste incineration data related to the control parameters within a specified monitoring period; A data decomposition module 302 is used to decompose the waste incineration data into multiple Gaussian distribution components, obtain corresponding data probabilities, and determine a sub-data set of each Gaussian distribution component based on the data probabilities, wherein the data probability is used to indicate the probability value of any waste incineration data satisfying the Gaussian distribution component; A prediction module 303 is used to extract features from the sub-dataset using a pre-established feature extraction model, and to update network parameters in the feature extraction model when an error result is detected to be greater than a preset error threshold, so as to output a predicted value of the control parameter, wherein the error result is used to characterize an error between a specified label and an output result of the feature extraction model; The strategy generation module 304 is used to determine the control strategy of the control parameter based on the predicted value, and the control strategy is used to indicate the regulation of the current waste incineration data.

[0067] In some embodiments, the data decomposition module 302 includes: A matrix construction unit, used for constructing the waste incineration data into a data matrix; an allocation unit, configured to input the data matrix into a Gaussian mixture model comprising a plurality of Gaussian distribution components, allocate each element in the data matrix to the Gaussian distribution component, and output a corresponding data probability, wherein the number of the Gaussian distribution components is consistent with the number of data categories of the waste incineration data; The sub-dataset construction unit is used to construct a sub-dataset of each Gaussian distribution component based on the elements and the corresponding data probabilities.

[0068] In some embodiments, the feature extraction model includes a first feature extraction model and a second feature extraction model arranged in sequence, and the first feature extraction model includes a plurality of multi-layer perceptrons; The prediction module 303 includes: A first feature extraction unit, configured to input the sub-dataset into the first feature extraction model, perform feature extraction on the corresponding sub-dataset through the multi-layer perceptron, and use the data feature values ​​output by each of the multi-layer perceptrons as the output of the first feature extraction model; A splicing unit, used for vertically splicing the data feature values ​​to obtain fusion features; The second feature extraction unit is used to extract the time feature of the fusion feature through the second feature extraction model to output the predicted value of the control parameter.

[0069] In some embodiments, the multilayer perceptron comprises a plurality of hidden layers; The prediction module 303 also includes: A detection unit, used for detecting, for any multilayer perceptron, whether an error result of the multilayer perceptron is greater than a preset error threshold, wherein the error result is used to indicate a difference between the specified label and a predicted value determined by taking the output of the current hidden layer as the data feature value currently output by the multilayer perceptron; A node adding unit, configured to add a new node in the hidden layer when it is detected that the error result is greater than the error threshold and the number of nodes in each hidden layer included in the multilayer perceptron is less than the node number threshold, and make the added node meet the constraint condition, wherein the node number threshold is determined by the data length of the waste incineration data, and the constraint condition is determined by the error result before adding the node, the output of the last node in the hidden layer before adding the node, the output of the added node, and the error result after adding the node; A network parameter updating unit, configured to update the output weight of each node in the hidden layer based on the specified label, the output weight of each node before the added node, and the output of each node after the added node; The output updating unit is used to update the data feature value output by the multilayer perceptron based on the updated output weight.

[0070] In some embodiments, the prediction module 303 further includes: The data feature value calculation unit is used to determine the output of the hidden layer by the following formula (1): (1) in, is the sub-dataset, represents the set of output weights of each node in the hidden layer before adding the node, represents the number of nodes in the hidden layer before adding the node, is the data length, Indicates the output weight of the jth node in the hidden layer before the increase node, represents the activation function of the jth node, represents the input weight of the jth node, represents the bias of the jth node; The error result calculation unit is used to determine the error result by the following formula (2): (2) in, represents the error result before adding the node, Specify a label for the Indicates the predicted value determined by taking the output of the current hidden layer as the data feature value of the current output of the multi-layer perceptron; The constraint condition calculation unit is used to determine the constraint condition by the following formula (3): (3) in, represents the error result of the sub-dataset before adding the node, Indicates The output of the node, represents the output of the added node, Indicates the error result after adding nodes; The output weight calculation unit is used to update the output weight of each node in the hidden layer through the following formula (4): (4) in, represents the set of output weights of each node in the hidden layer after adding nodes, Represents the output of each node in the hidden layer after adding nodes; is the preset F norm; represents a pseudo-inverse operation; Indicates the value of the independent variable that minimizes the function.

[0071] In some embodiments, the policy generation module 304 includes: An observation value acquisition unit, used to acquire the observation value of the control parameter at the current moment; A reinforcement learning unit, for determining a reinforcement learning evaluation value based on the garbage incineration data in the current monitoring period and a control reward value obtained in response to a control action on the garbage incineration data, wherein the control reward value includes a reward value corresponding to when the value of the control parameter at the next moment after the control action is performed falls within a control target interval of the control parameter, and the reinforcement learning evaluation value is used to evaluate the expected cumulative reward obtained by the executed control action; A strategy determination unit is used to determine the control strategy of the control parameter based on the control action that maximizes the reinforcement learning evaluation value.

[0072] In some embodiments, the policy generation module 304 further includes: A model processing unit, configured to process a sub-data set determined by the waste incineration data obtained after executing the control action through the feature extraction model to obtain a predicted value for the control parameter; An iterative unit is used to iteratively determine the control action that maximizes the reinforcement learning evaluation value based on the predicted value, the garbage incineration data obtained after executing the control action, and the corresponding control reward value, as the control strategy of the control parameter.

[0073] In some embodiments, the control parameters include the furnace temperature of the incinerator, and the waste incineration data include the grate speed, primary air supply, secondary air supply, fan pressure, grate temperature and the furnace temperature at the current moment of the incinerator.

[0074] The device of the embodiments of the present application can execute the method provided by the embodiments of the present application, and the implementation principles are similar. The actions performed by each module in the device of each embodiment of the present application correspond to the steps in the method of each embodiment of the present application. For the detailed functional description of each module of the device, please refer to the description in the corresponding method shown in the previous text, which will not be repeated here.

[0075] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for adaptive combustion parameter control of waste incineration, characterized in that: include: Obtain waste incineration data related to control parameters within a specified monitoring period; Decomposing the waste incineration data into a plurality of Gaussian distribution components, and obtaining corresponding data probabilities, and determining a sub-data set of each Gaussian distribution component based on the data probabilities, wherein the data probabilities are used to indicate a probability value that any waste incineration data satisfies the Gaussian distribution component; Performing feature extraction on the sub-dataset using a pre-established feature extraction model, and updating network parameters in the feature extraction model when an error result is detected to be greater than a preset error threshold, so as to output a predicted value of the control parameter, wherein the error result is used to characterize an error between a specified label and an output result of the feature extraction model; Based on the predicted value, a control strategy for the control parameter is determined, and the control strategy is used to indicate and regulate current waste incineration data.

2. The method for adaptive combustion parameter control of waste incineration according to claim 1, characterized in that: Decomposing the waste incineration data into a plurality of Gaussian distribution components, obtaining corresponding data probabilities, and determining a sub-data set of each Gaussian distribution component based on the data probabilities includes: constructing the waste incineration data into a data matrix; Input the data matrix into a Gaussian mixture model containing multiple Gaussian distribution components, assign each element in the data matrix to the Gaussian distribution component, and output the corresponding data probability, wherein the number of the Gaussian distribution components is consistent with the number of data categories of the waste incineration data; Based on the elements and the corresponding data probabilities, a sub-dataset of each Gaussian distribution component is constructed.

3. The method for adaptive combustion parameter control of waste incineration according to claim 2, characterized in that: The feature extraction model includes a first feature extraction model and a second feature extraction model arranged in sequence, wherein the first feature extraction model includes a plurality of multi-layer perceptrons; The method of extracting features from the sub-dataset using a pre-established feature extraction model and updating a network parameter in the feature extraction model when an error result is detected to be greater than a preset error threshold to output a predicted value of the control parameter comprises: Inputting the sub-datasets into the first feature extraction model, performing feature extraction on the corresponding sub-datasets through the multi-layer perceptron, and using the data feature values ​​output by each of the multi-layer perceptrons as the output of the first feature extraction model; Vertically splicing the data feature values ​​to obtain fusion features; The time feature of the fusion feature is extracted through the second feature extraction model to output the predicted value of the control parameter.

4. The method for adaptive combustion parameter control of waste incineration according to claim 3 is characterized in that: The multilayer perceptron includes a plurality of hidden layers, and the feature extraction of the sub-dataset is performed using a pre-established feature extraction model, and when an error result is detected to be greater than a preset error threshold, a network parameter in the feature extraction model is updated to output a predicted value of the control parameter, and further includes: For any multi-layer perceptron, detecting whether an error result of the multi-layer perceptron is greater than a preset error threshold, wherein the error result is used to indicate the difference between the specified label and a predicted value determined by taking the output of the current hidden layer as the data feature value currently output by the multi-layer perceptron; When it is detected that the error result is greater than the error threshold, and the number of nodes in each hidden layer included in the multilayer perceptron is less than the node number threshold, a new node is added to the hidden layer, and the added node satisfies a constraint condition, wherein the node number threshold is determined by the data length of the waste incineration data, and the constraint condition is determined by the error result before adding the node, the output of the last node in the hidden layer before adding the node, the output of the added node, and the error result after adding the node; Based on the specified label, the output weights of the nodes before the added node, and the outputs of the nodes after the added node, the output weights of the nodes in the hidden layer are updated; Based on the updated output weights, the data feature values ​​output by the multilayer perceptron are updated.

5. The method for adaptive combustion parameter control of waste incineration according to claim 4 is characterized in that: The method further comprises: extracting features from the sub-dataset using a pre-established feature extraction model, and updating network parameters in the feature extraction model when an error result is detected to be greater than a preset error threshold, so as to output a predicted value of the control parameter. The output of the hidden layer is determined by the following formula (1): （1） in, is the sub-dataset, represents the set of output weights of each node in the hidden layer before adding the node, represents the number of nodes in the hidden layer before adding the node, is the data length, Indicates the output weight of the jth node in the hidden layer before the increase node, represents the activation function of the jth node, represents the input weight of the jth node, represents the bias of the jth node; The error result is determined by the following formula (2): （2） in, represents the error result before adding the node, Specify a label for the Indicates the predicted value determined by taking the output of the current hidden layer as the data feature value of the current output of the multi-layer perceptron; The constraint condition is determined by the following formula (3): （3） in, represents the error result of the sub-dataset before adding the node, Indicates The output of the node, represents the output of the added node, Indicates the error result after adding nodes; The output weights of each node in the hidden layer are updated using the following formula (4): （4） in, represents the set of output weights of each node in the hidden layer after adding nodes, Represents the output of each node in the hidden layer after adding nodes; is the preset F norm; represents a pseudo-inverse operation; Indicates the value of the independent variable that minimizes the function.

6. The method for adaptive combustion parameter control of waste incineration according to claim 5, characterized in that: The monitoring period is used to indicate the time interval formed by the current time and the previous specified time period; The step of determining the control strategy of the control parameter based on the predicted value comprises: Obtaining the observed value of the control parameter at the current moment; Determine a reinforcement learning evaluation value based on the garbage incineration data in the current monitoring period and the control reward value obtained in response to the regulation action on the garbage incineration data, wherein the control reward value includes a reward value corresponding to when the value of the control parameter at the next moment falls within the control target interval of the control parameter after the regulation action is executed, and the reinforcement learning evaluation value is used to evaluate the expected cumulative reward obtained by the executed regulation action; Based on the control action that maximizes the reinforcement learning evaluation value, a control strategy for the control parameter is determined.

7. The method for adaptive combustion parameter control of waste incineration according to claim 6, characterized in that: The control strategy for determining the control parameter based on the predicted value further includes: Processing a sub-dataset determined by the waste incineration data obtained after executing the control action through the feature extraction model to obtain a predicted value for the control parameter; Iteratively determine the control action that maximizes the reinforcement learning evaluation value based on the predicted value, the garbage incineration data obtained after executing the control action, and the corresponding control reward value, as the control strategy of the control parameter.

8. The method for controlling the adaptive combustion parameters of waste incineration according to any one of claims 1 to 7, characterized in that: The control parameters include the furnace temperature of the incinerator, and the garbage incineration data include the grate speed, the primary air supply, the secondary air supply, the fan pressure, the grate temperature and the furnace temperature at the current moment of the incinerator.

9. A garbage incineration adaptive combustion parameter control device, characterized in that: include: A data acquisition module, used to acquire waste incineration data related to control parameters within a specified monitoring period; A data decomposition module, used to decompose the waste incineration data into multiple Gaussian distribution components, and obtain corresponding data probabilities, and determine a sub-data set of each Gaussian distribution component based on the data probabilities, wherein the data probability is used to indicate the probability value that any waste incineration data satisfies the Gaussian distribution component; A prediction module, used to extract features from the sub-dataset using a pre-established feature extraction model, and to update network parameters in the feature extraction model when an error result is detected to be greater than a preset error threshold, so as to output a predicted value of the control parameter, wherein the error result is used to characterize an error between a specified label and an output result of the feature extraction model; A strategy generation module is used to determine a control strategy for the control parameter based on the predicted value, wherein the control strategy is used to indicate and regulate the current waste incineration data.

Images