Modeling method for multiple controlled variables in urban solid waste incineration process based on shared characteristics

Abstract

The invention provides an urban solid waste incineration process multi-controlled variable modeling method based on shared features. The urban solid waste incineration process multi-controlled variable modeling method comprises the steps that a plurality of manipulated variables and a plurality of controlled variables are collected in an urban solid waste incineration process; screening out at least one shared feature having significant influence on the plurality of controlled variables from the plurality of manipulated variables by using a correlation quantitative analysis method; improving a loss function and an evaluation index of the fuzzy neural network; and taking the screened shared features as input, taking the plurality of controlled variables as output, and training an improved fuzzy neural network so as to predict the plurality of controlled variables at the same time, thereby evaluating the urban solid waste incineration process. According to the method, breakthrough precision is achieved on the most important controlled variables, leading performance is kept on most of the controlled variables, and a solid foundation is laid for achieving more accurate and stable advanced control.

Description

Technical Field

[0001] This application belongs to the field of energy engineering technology, and in particular relates to a modeling method for multiple controlled variables in urban solid waste incineration processes based on shared characteristics. Background Technology

[0002] Municipal solid waste incineration (MSWI) is a typical complex industrial process characterized by multiple variables, strong coupling, nonlinearity, and large time lag. Unstable incineration processes can lead to secondary pollution and reduced energy efficiency. To achieve stable, environmentally friendly, and efficient operation, it is necessary to model several key controlled variables, such as furnace temperature and main steam flow, to provide a foundation for advanced control strategies. While traditional mechanistic models offer reliable extrapolation, they are complex to model and lack adaptability to nonlinear systems, making them unsuitable for the complex and variable MSWI process. In contrast, data-driven models excel at characterizing the nonlinear, time-varying, and coupled characteristics of complex systems lacking precise mechanisms. Their high development efficiency and scalability offer significant advantages in MSWI modeling. Existing technologies include Hu et al.'s proposal of a multi-objective robust modeling method based on an improved stochastic configuration network to accurately predict furnace temperature and flue gas oxygen content in MSWI processes; and Ding et al.'s proposal of a method such as "operating condition identification-feature reduction-shared membership function TS fuzzy neural network (MIMO-TSFNN)" to address the modeling challenges of multiple-input multiple-output (MIMO) systems. These methods lay a foundation for the modeling and control of MSWI processes.

[0003] However, existing technologies still have significant shortcomings: when models need to handle multiple controlled variables simultaneously, they often overlook the differences in numerical scales of these variables due to their different engineering dimensions and physical meanings. This difference can lead to a situation where, during model training and performance evaluation, a variable with a larger dimension dominates the entire model optimization process, while the performance of variables with smaller dimensions is masked, resulting in a decrease in the overall accuracy and applicability of the model. Therefore, how to construct a unified model that can accurately and comprehensively represent all controlled variables is a pressing technical problem that needs to be solved. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a modeling method for multiple controlled variables in the urban solid waste incineration process based on shared features. It achieves breakthrough accuracy on the most important controlled variable and maintains leading performance on the vast majority of controlled variables, laying a solid foundation for achieving more accurate and stable advanced control.

[0005] This application provides a modeling method for multiple controlled variables in urban solid waste incineration processes based on shared features, including: Several manipulated variables and multiple controlled variables were collected during the urban solid waste incineration process; Using correlation quantification analysis, at least one shared feature that has a significant impact on multiple controlled variables is selected from several manipulated variables. Improvements were made to the loss function and evaluation metrics of fuzzy neural networks; By using the selected shared features as input and multiple controlled variables as output, an improved fuzzy neural network is trained to simultaneously predict multiple controlled variables, thereby evaluating the urban solid waste incineration process.

[0006] Furthermore, the method of using correlation quantification analysis to screen out at least one shared feature from several manipulated variables that has a significant impact on multiple controlled variables includes: Calculate the Pearson correlation coefficient between each manipulated variable and multiple controlled variables; wherein the manipulated variables are variables closely related to the urban solid waste incineration process selected by experienced domain experts; Based on at least one pre-set correlation threshold, determine whether each manipulated variable is a shared feature.

[0007] Furthermore, we determine whether each manipulated variable is a shared feature using the following method: If the correlation threshold is one, then the absolute value of the Pearson correlation coefficient between the manipulated variable and multiple controlled variables is compared with the correlation threshold. If there are multiple correlation thresholds, then for each controlled variable, the absolute value of the Pearson correlation coefficient between the manipulated variable and the controlled variable is compared with the correlation threshold of the controlled variable. When the absolute values ​​of the Pearson correlation coefficients between the manipulated variable and multiple controlled variables all exceed their respective correlation thresholds, the manipulated variable is determined to be a shared feature.

[0008] Furthermore, the improvements to the loss function and evaluation metrics for fuzzy neural networks include: The Huber function is used as the loss function to update the parameters of the fuzzy neural network, and the relative root mean square error is used as the evaluation metric as the learning target of the fuzzy neural network. This avoids the training process being dominated by a large number of controlled variables, and ensures that the predictive ability is balanced for all controlled variables.

[0009] Furthermore, the improved fuzzy neural network includes a pre-processor network and a post-processor network; The preceding network uses Gaussian membership functions to calculate the membership degree of the input shared features to each fuzzy rule, then calculates the activation intensity of each fuzzy rule based on the fuzzy multiplication operator, and outputs the weight of each fuzzy rule after normalization and defuzzification. The consequent network obtains the consequent parameters corresponding to each fuzzy rule to perform local predictions of multiple controlled variables. Then, based on the weights of each fuzzy rule, it performs a weighted average of the local prediction results to obtain the global prediction results of multiple controlled variables.

[0010] Furthermore, the manipulated variables include: primary air temperature, primary air volume, primary air pressure, secondary air temperature, secondary air volume, drying section air volume, combustion section 1 air volume, combustion section 2 air volume, combustion section air volume, feeding speed, and drying section grate speed. The controlled variables include: furnace temperature, flue gas oxygen content, main steam flow rate, and ignition point temperature; The shared features include: primary air temperature, primary air volume, drying section air volume, secondary air volume, feeding speed, and drying grate speed.

[0011] The modeling method for multiple controlled variables in the urban solid waste incineration process based on shared features provided in this application has achieved breakthrough accuracy on the most important controlled variable and maintained leading performance on the vast majority of controlled variables, laying a solid foundation for achieving more accurate and stable advanced control. Attached Figure Description

[0012] Figure 1 A flowchart is shown below illustrating a modeling method for multiple controlled variables in an urban solid waste incineration process based on shared features, as provided in an embodiment of this application. Figure 2 This paper shows a trend graph of RRMSE changes during the model training process provided in the embodiments of this application; Figure 3 The following diagram illustrates the fitting effect of the training samples provided in the embodiments of this application; Figure 4 The following diagram illustrates the fitting effect of the verification sample provided in an embodiment of this application; Figure 5 The diagram shows the test fitting effect of the four controlled variables provided in the embodiments of this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this technical solution clearer, the following detailed description, in conjunction with specific embodiments, further illustrates this technical solution. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of this technical solution.

[0014] Example 1: Please see as follows Figure 1 The flowchart shown is a modeling method for multiple controlled variables in the urban solid waste incineration process based on shared features. Figure 1 As shown, the method includes: S101. Collect several manipulated variables and multiple controlled variables during the urban solid waste incineration process.

[0015] The manipulated variables include: primary air temperature, primary air volume, primary air pressure, secondary air temperature, secondary air volume, drying section air volume, combustion section 1 air volume, combustion section 2 air volume, combustion section air volume, feeding rate, and drying section grate speed; the controlled variables include: furnace temperature, flue gas oxygen content, main steam flow rate, and burnout point temperature.

[0016] In this step, operational datasets can be collected from actual urban solid waste incineration power plants to obtain several manipulated variables and multiple controlled variables.

[0017] S102. Using correlation quantification analysis, select at least one shared feature from several manipulated variables that has a significant impact on multiple controlled variables.

[0018] The shared features include: primary air temperature, primary air volume, drying section air volume, secondary air volume, feeding speed, and drying grate speed.

[0019] In practice, at least one shared feature that has a significant impact on multiple controlled variables can be selected through the following methods: Step 1021: Calculate the Pearson correlation coefficient between each manipulated variable and multiple controlled variables.

[0020] The operational variables are selected by experienced domain experts and are closely related to the urban solid waste incineration process.

[0021] In this step, the strength of the linear association between the manipulated variable and the controlled variable under different operating conditions is quantified by calculating the Pearson correlation coefficient (PCC).

[0022] As an example, Table 1 below shows the absolute values ​​of the Pearson correlation coefficients between the manipulated variable and the controlled variable.

[0023] Table 1. Statistical table of absolute values ​​of Pearson correlation coefficients between manipulated and controlled variables.

[0024] Step 1022: Determine whether each manipulated variable is a shared feature based on at least one pre-set correlation threshold.

[0025] In this step, experienced domain experts set one or more correlation thresholds. That is, all manipulated variables can share a single correlation threshold, several manipulated variables can share a single correlation threshold, or each manipulated variable can have a different correlation threshold. The specific settings can be made according to the actual situation, and this application does not impose any limitations on them.

[0026] In practical implementation, the following methods can be used to determine whether each manipulated variable is a shared feature: Step 201: If the correlation threshold is one, then compare the absolute value of the Pearson correlation coefficient between the manipulated variable and multiple controlled variables with the correlation threshold.

[0027] Step 202: If there are multiple correlation thresholds, then for each controlled variable, compare the absolute value of the Pearson correlation coefficient between the manipulated variable and the controlled variable with the correlation threshold of the controlled variable.

[0028] Step 203: When the absolute value of the Pearson correlation coefficient between the manipulated variable and multiple controlled variables exceeds their respective correlation thresholds, the manipulated variable is determined to be a shared feature.

[0029] The method described above for screening shared features using a correlation threshold ensures that shared features have a broad and critical influence on the overall manipulated variable, rather than being effective only for a single manipulated variable.

[0030] S103. Improve the loss function and evaluation index for fuzzy neural networks.

[0031] In practical implementation, the loss function and evaluation metrics of fuzzy neural networks can be improved in the following ways: Step 1031: Use the Huber function as the loss function to update the parameters of the fuzzy neural network, and use the relative root mean square error as the evaluation index as the learning target of the fuzzy neural network, so as to avoid the training process being dominated by a large number of controlled variables and ensure that the predictive ability of all controlled variables is balanced.

[0032] In this step, the different units of the controlled variables (such as temperature in "°C" and flow rate in "t / h") lead to distortion in model performance evaluation. Therefore, this application uses the Huber function as the loss function for parameter updates and redefines the model evaluation metrics, abandoning the traditional single, unit-sensitive evaluation metrics. During the model training phase, the evaluation metrics are redefined or combined, using the relative root mean square error (RRMSE) as the learning objective of the fuzzy neural network. This ensures that the model treats each controlled variable equally during optimization, avoiding large-scale controlled variables from dominating the training process, thereby ensuring balanced and accurate predictive ability for all controlled variables.

[0033] S104. Using the selected shared features as input and multiple controlled variables as output, train the improved fuzzy neural network to simultaneously predict multiple controlled variables, thereby evaluating the urban solid waste incineration process.

[0034] The improved fuzzy neural network includes a pre-processor network and a post-processor network. The preceding network uses Gaussian membership functions to calculate the membership degree of the input shared features to each fuzzy rule, then calculates the activation intensity of each fuzzy rule based on the fuzzy multiplication operator, and outputs the weight of each fuzzy rule after normalization and defuzzification.

[0035] The preceding network specifically includes: 1) Input layer: Used to receive shared features from the input. The shared features of each sample are represented as an input vector. : ; (1) 2) Membership Function Layer: Used to calculate the membership degree of the shared features of the input to each fuzzy rule using Gaussian membership functions. The input pair of the th _ ... The membership degree values ​​of the fuzzy rules are: ; (2) In the formula, It is the first The input corresponds to the first The membership function center of a fuzzy rule It is the first The input corresponds to the first The width of the membership function of a fuzzy rule. and It is a numerical stability technique to prevent the exponent from exponent from exponent explosion due to an excessively small denominator, or underflow due to an excessively large exponent. This is to ensure that the membership degree is never zero, thus avoiding errors in subsequent division calculations.

[0036] Further obtain the first The membership matrix of each sample is : ; (3) By using a membership function layer, the precise input (shared features) is transformed into a fuzzy set, thereby achieving fuzzy partitioning of the feature space.

[0037] 3) Rule Layer: Used to treat fuzzy multiplication operators as fuzzy logic rules and calculate the activation strength of each fuzzy rule. The activation strength of each fuzzy rule is: ; (4) In the formula, These are measures to ensure numerical stability, ensuring... .

[0038] Further obtain the first The activation intensity vectors of each sample for each fuzzy rule are: ; (5) Activation intensity vectors for each fuzzy rule After performing normalization and defuzzification, the weights of each fuzzy rule are obtained as follows: ; (6) Among them, the The activation weights of a fuzzy rule are represented as follows: (7) The consequent network obtains the consequent parameters corresponding to each fuzzy rule to perform local predictions of multiple controlled variables. Then, based on the weights of each fuzzy rule, it performs a weighted average of the local prediction results to obtain the global prediction results of multiple controlled variables.

[0039] The consequent network specifically includes: 1) Consequence layer: Obtain the consequent parameters corresponding to each fuzzy rule. The parameter matrix of the assembly is represented as follows: ; (8) Based on the The first fuzzy rule corresponds to the th fuzzy rule. The consequent parameter yields multiple controlled variables in the first... Local prediction results under each consequent parameter: ; (9) 2) Output layer: Activation weights based on various fuzzy rules For each local prediction result Perform a weighted average to obtain the first... The global prediction results for multiple controlled variables of a sample are as follows: ; (10) Further obtain the first The global prediction result for each sample is expressed as: (11) This part combines the antecedent network and consequent network of fuzzy rules to complete the mapping from input to output.

[0040] Example 2: The effectiveness of the proposed modeling method for multiple controlled variables in urban solid waste incineration processes based on shared features was verified by collecting operational datasets from actual urban solid waste incineration power plants.

[0041] Please see as follows Figure 2The graph shows the trend of RRMSE changes for each controlled variable during model training. A significantly low RRMSE indicates more precise and stable control, which has direct engineering value for improving boiler efficiency, ensuring safety, and reducing losses.

[0042] Please see as follows Figure 3 The training sample fitting effect diagram shown compares the predicted values ​​obtained using the training samples with the target values ​​during model training. Figure 4 The graph showing the fitting effect of the validation samples compares the predicted values ​​obtained using the validation samples with the target values ​​during the model validation process. Figure 5 The test fit graphs for the four controlled variables shown are obtained by comparing the predicted values ​​of each controlled variable with the target values ​​during the model testing process.

[0043] The model achieved breakthrough performance on the core controlled variable of furnace temperature, with R² consistently exceeding 0.90 and RRMSE consistently below 0.006, demonstrating extremely high levels of accuracy and reliability in predicting the most critical parameter. The improved fuzzy neural network exhibited strong and balanced predictive capabilities when facing controlled variables with different characteristics. Whether predicting noisy flue gas oxygen content or dynamic main steam flow, its performance surpassed or rivaled other advanced models, showcasing excellent comprehensive strength and generalization ability. The most significant advantage of the improved fuzzy neural network is its extremely low RRMSE. Especially in ignition point temperature prediction, its RRMSE maintains an order-of-magnitude advantage, directly translating into a more stable and reliable control basis in engineering applications, effectively avoiding system instability caused by large fluctuations in predicted values. The high-precision and high-stability predictive output provides a direct and reliable model foundation for combustion optimization, improving boiler efficiency, ensuring system safety, and reducing equipment wear, demonstrating clear engineering application value.

[0044] In conclusion, the improved fuzzy neural network, with its superior accuracy of core metrics, robust overall performance, and absolute advantage in predictive stability, has been proven to be the best solution for multivariate modeling of MSWI processes.

[0045] The above content is only a preferred embodiment of the present invention. For those skilled in the art, many changes can be made in the specific implementation and application scope based on the ideas of the present invention. As long as these changes do not depart from the concept of the present invention, they all fall within the protection scope of the present invention.

Claims

1. A modeling method for multiple controlled variables in urban solid waste incineration processes based on shared features, characterized in that, The method includes: Several manipulated variables and multiple controlled variables were collected during the urban solid waste incineration process; Using correlation quantification analysis, at least one shared feature that has a significant impact on multiple controlled variables is selected from several manipulated variables. Improvements were made to the loss function and evaluation metrics of fuzzy neural networks; By using the selected shared features as input and multiple controlled variables as output, an improved fuzzy neural network is trained to simultaneously predict multiple controlled variables, thereby evaluating the urban solid waste incineration process.

2. The method as described in claim 1, characterized in that, The method of using correlation quantification analysis to screen out at least one shared feature from several manipulated variables that has a significant impact on multiple controlled variables includes: Calculate the Pearson correlation coefficient between each manipulated variable and multiple controlled variables; wherein the manipulated variables are variables closely related to the urban solid waste incineration process selected by experienced domain experts; Based on at least one pre-set correlation threshold, determine whether each manipulated variable is a shared feature.

3. The method as described in claim 2, characterized in that, Determine whether each manipulated variable is a shared feature using the following method: If the correlation threshold is one, then the absolute value of the Pearson correlation coefficient between the manipulated variable and multiple controlled variables is compared with the correlation threshold. If there are multiple correlation thresholds, then for each controlled variable, the absolute value of the Pearson correlation coefficient between the manipulated variable and the controlled variable is compared with the correlation threshold of the controlled variable. When the absolute values ​​of the Pearson correlation coefficients between the manipulated variable and multiple controlled variables all exceed their respective correlation thresholds, the manipulated variable is determined to be a shared feature.

4. The method as described in claim 1, characterized in that, The improvements to the loss function and evaluation metrics for fuzzy neural networks include: The Huber function is used as the loss function to update the parameters of the fuzzy neural network, and the relative root mean square error is used as the evaluation metric as the learning target of the fuzzy neural network. This avoids the training process being dominated by a large number of controlled variables, and ensures that the predictive ability is balanced for all controlled variables.

5. The method as described in claim 1, characterized in that, The improved fuzzy neural network includes a pre-processor network and a post-processor network; The preceding network uses Gaussian membership functions to calculate the membership degree of the input shared features to each fuzzy rule, then calculates the activation intensity of each fuzzy rule based on the fuzzy multiplication operator, and outputs the weight of each fuzzy rule after normalization and defuzzification. The consequent network obtains the consequent parameters corresponding to each fuzzy rule to perform local predictions of multiple controlled variables. Then, based on the weights of each fuzzy rule, it performs a weighted average of the local prediction results to obtain the global prediction results of multiple controlled variables.

6. The method as described in claim 1, characterized in that, The manipulated variables include: primary air temperature, primary air volume, primary air pressure, secondary air temperature, secondary air volume, drying section air volume, combustion section 1 air volume, combustion section 2 air volume, combustion section air volume, feeding rate, and drying section grate speed. The controlled variables include: furnace temperature, flue gas oxygen content, main steam flow rate, and ignition point temperature; The shared features include: primary air temperature, primary air volume, drying section air volume, secondary air volume, feeding speed, and drying grate speed.

Images