Dioxin emission prediction method based on RF-PSO (Radio Frequency-Particle Swarm Optimization) integrated algorithm

Abstract

The invention provides a dioxin emission prediction method based on an RF-PSO integrated algorithm, and relates to the technical field of dioxin emission prediction. Comprising the following steps: S1, constructing a data set; s2, preprocessing a data set, performing standardization and dimension reduction processing on input features in the data set, and dividing the data set into a training set and a test set; s3, training and testing a random forest model, and taking the trained random forest model as a dioxin emission initial prediction model; s4, generating virtual sample input; expanding the range of the input features through multi-distribution overall trend diffusion, and constructing a virtual input data set; and S5, optimizing the generated virtual sample through an RF-PSO (Radio Frequency-Particle Swarm Optimization) integration algorithm. According to the method, even if the performance of the initial prediction model is weak, the reliable virtual sample can still be generated, and the high generalization ability and prediction precision of the final dioxin emission prediction model are ensured.

Description

Technical Field

[0001] The present application relates to the technical field of dioxin emission prediction, and in particular to a dioxin emission prediction method based on RF-PSO integrated algorithm. Background Art

[0002] Dioxins are extremely biotoxic, persistent, and bioaccumulative, making them difficult to degrade and treat. Solid waste incineration is one of the main sources of dioxin emissions in my country's environment. The conventional method for measuring dioxin emission levels in incineration flue gas is offline detection, which includes laboratory pretreatment steps such as Soxhlet extraction, purification, and concentration, and high-resolution gas chromatography / high-resolution mass spectrometry analysis. The detection cost is high and the required cycle is long. This method cannot reflect the operating status of waste incineration plants in a timely manner, nor can it meet the public's demand for real-time access to dioxin emission levels. With the increase in the proportion of domestic waste incineration and the increasingly stringent dioxin emission control standards, it is necessary to establish a high-precision dioxin emission prediction model, develop accurate online measurement methods, and provide rapid feedback to monitor and control dioxin emission levels.

[0003] Currently, machine learning has been applied to develop online dioxin emission measurement models. These models can predict dioxin concentrations in solid waste incineration flue gas using variables such as conventional pollutant concentrations, incinerator operating parameters, and flue gas purification system parameters. However, this approach requires sufficient data during training and validation to avoid overfitting. Dioxin sample collection is expensive and limited, and easy-to-collect variables are numerous. Therefore, dioxin emission prediction modeling is a typical small-sample, high-dimensional problem, making it difficult to obtain sufficient data to develop a robust and generalizable dioxin emission prediction model.

[0004] Virtual sample generation technology extracts data features from a small set of original samples, synthesizing a large number of new virtual samples. This effectively improves the generalization and prediction accuracy of machine learning models. In regression tasks, the generation of virtual output data generally relies on an initial prediction model trained on the original small-scale training set. To obtain more realistic and reliable virtual samples, the mean absolute percentage error (MAPE) of the initial prediction model is typically required to be ≤10%. However, in practical applications, MAPE >10% may occur due to the high-dimensional, sparse nature of data and noise interference. Therefore, it is necessary to consider how to generate reliable virtual samples from a weak initial prediction model. Summary of the Invention

[0005] The present application provides a dioxin emission prediction method based on the RF-PSO integrated algorithm to at least solve the above technical problems existing in the prior art.

[0006] According to a first aspect of the present application, a dioxin emission prediction method based on the RF-PSO integrated algorithm is provided, comprising the following steps: S1, constructing a data set, with dioxin toxicity equivalent as the output variable in the data set, and incinerator operating parameters and 124-TrCBz concentration as the input features in the data set; S2, dataset preprocessing, standardizes and reduces the dimension of the input features in the dataset, and divides the dataset into training set and test set; S3, random forest model training and testing, using the trained random forest model as the initial prediction model for dioxin emissions; S4, generate virtual sample input; expand the range of input features through multi-distribution overall trend diffusion, generate virtual input in the extended area of ​​the dataset and the actual observation area, and construct a virtual input dataset; S5, the generated virtual samples are optimized by the RF-PSO integrated algorithm.

[0007] In certain embodiments of the first aspect of the present application, the specific method of S1 is as follows: S11: Use EPA23a method to detect dioxin emission data at the end of solid waste incineration system, and use dioxin toxicity equivalent as the output variable in the data set; S12: The 124-TrCBz concentration and the concentration of other conventional pollutants at the end of the solid waste incineration system are simultaneously detected, and the incinerator operating parameters are collected through the DCS system, and the above parameters are used as input features in the original data set.

[0008] In certain embodiments of the first aspect of the present application, the other conventional pollutants include carbon monoxide, nitrogen oxides, sulfur dioxide, ammonia, hydrogen chloride and carbon dioxide.

[0009] In certain embodiments of the first aspect of the present application, in S2, the method for performing dimensionality reduction processing on the input features in the data set is as follows: The standardized data set is subjected to dimensionality reduction processing using the Boruta algorithm; Randomly arrange each input feature to generate the corresponding shadow feature; concatenate the input feature and the shadow feature to obtain the expanded feature matrix; Training a random forest model on the expanded feature matrix and calculating the Z score of each feature, including input features and shadow features; Record the maximum Z score of all shadow features, mark the input features with a score greater than the maximum Z score as important features and retain them, and mark the input features with a score less than the maximum Z score as unimportant features and permanently delete them from the feature set; Remove all shadow features.

[0010] In certain embodiments of the first aspect of the present application, in S3, the method for training and testing the random forest model is as follows: S31: Based on the training set, the random forest model was selected to construct the dioxin emission prediction model, and the grid search method and five-fold random cross-validation method were used for hyperparameter optimization; S32, tests the generalization ability of the random forest model based on the test set, using the coefficient of determination, root mean square error, and mean absolute percentage error as evaluation indicators.

[0011] In certain embodiments of the first aspect of the present application, in S31, the method for performing hyperparameter optimization using a grid search method and a five-fold random cross-validation method is as follows: A grid search was performed to traverse the predefined hyperparameter combinations of the random forest model: the number of decision trees in the random forest and the maximum depth of the decision tree. All other parameters were set to their default values. The optimal hyperparameters were selected using a five-fold randomized cross-validation method. The five-fold randomized cross-validation method was used as follows: The entire training set is randomly divided into 5 non-overlapping subsets, each with 20% of the data; Each time, 4 subsets are used as training sets to train the random forest model, and the remaining 1 subset is used as the validation set, for a total of 5 cycles; For each cycle, set the parameter optimization target to negative mean square error; calculate the negative mean square error based on the validation set; Calculate the average of the negative mean square error of 5 cycles as the final score of the hyperparameter combination, and select the hyperparameter combination with the highest average negative mean square error.

[0012] In certain embodiments of the first aspect of the present application, in S4, the method for generating the virtual sample input is as follows: S41, calculate the data center point CL;

[0013] Among them, n is the number of samples, CL is the center point, X （.） is the order statistic; S42, calculating the left and right skewness and variance of the center point CL;

[0014]

[0015]

[0016] Where S KL is the left skewness, S KR is the right skewness, N L 、NR is the number of samples whose values ​​are less than or greater than the center point CL, is the variance of the input feature X; S43, calculating the upper and lower bounds of the acceptable range;

[0017]

[0018] Where LB is the lower bound of the acceptable range, and RB is the upper bound of the acceptable range; S44, calculating the number of virtual samples to be generated within each range;

[0019]

[0020]

[0021]

[0022] Where, are the number of virtual samples to be generated in the range of (LB, min), (min, max), and (max, RB), respectively. The total number of virtual samples generated for the request; S45, generating a virtual input; Generate virtual sample input using uniform distribution within the extended range: (LB, min) range:

[0023] In the range of (max, RB):

[0024] Generate virtual sample input using triangular distribution within the direct observation range (min, max):

[0025] Where s is a random number that obeys uniform distribution within (0,1); For each input feature Expand and finally generate a virtual input dataset .

[0026] In certain embodiments of the first aspect of the present application, in S5, the method for optimizing the generated virtual samples by using the RF-PSO integrated algorithm is as follows: S51: Generate a virtual output corresponding to a virtual input data set based on the random forest model; S52: Dynamically adjust the input features of virtual samples based on the PSO algorithm; S53: Mix the virtual sample data set with the original training set, re-establish the random forest model, and use the grid search method and five-fold random cross-validation method to optimize the hyperparameters; if the model performance is improved, return to step S4, generate virtual sample input again, and execute S51 and S52 in sequence; otherwise, go to S54; S54, retain the current virtual sample as the qualified virtual sample dataset, mix it with the original training set, establish the random forest model again, and use the grid search method and five-fold random cross-validation method to optimize the hyperparameters.

[0027] In certain embodiments of the first aspect of the present application, in S52, the method for dynamically adjusting the input features of the virtual sample based on the PSO algorithm is as follows: S521, particle swarm initialization: Input feature matrix with pre-generated virtual samples As the initial particle swarm position, each particle represents the input feature vector of a virtual sample; Generate a random initial particle velocity matrix in the interval [-1,1], with a dimension matching the input feature number; Limit the particle positions to the characteristic upper and lower bounds (LB, RB) of the original data set; S522, fitness function calculation: A pre-trained initial prediction model is used to map particle input features to output features to form a complete virtual sample; Merge a single virtual sample with the original training data to construct an augmented dataset, and retrain the random forest model based on the augmented dataset; Use the trained random forest model to calculate MAPE on the original training set as the fitness value of the particle; S523, particle swarm dynamic update: The inertia weight is calculated by the number of iterations from Linearly down to , balancing global exploration and local development capabilities:

[0028] in, is the updated inertia weight, is the current iteration number, is the maximum number of iterations; Update particle velocity:

[0029] in, is a random vector, are the individual and group learning factors, and are the optimal positions of individuals and populations, is the particle velocity, is the particle position; Adjust the particle position according to the updated velocity and reapply the boundary constraints to ensure validity; S524, optimal solution iteration: Compare the current fitness with the historical optimal value and retain the optimal position of each particle , achieving individual optimal update; After each iteration, the particle with the best fitness among all particles is selected as , achieving global optimal update; S525, output optimization results: Output the complete virtual samples corresponding to the global optimal particle and its fitness value to obtain a set of virtual samples; Perform PSO optimization once to obtain a set of virtual samples and repeat the above process times, get A virtual sample dataset.

[0030] Compared with the prior art, this application has the following beneficial effects: The dioxin emission prediction method presented in this application, based on an integrated algorithm combining data dimensionality reduction, virtual sample generation (VSG), and random forest-particle swarm optimization (RF-PSO), can generate a large number of virtual samples from a small number of real samples, alleviating data scarcity and avoiding overfitting. Even when the initial prediction model performance is weak (e.g., due to noise interference or high-dimensional data sparsity resulting in a MAPE greater than 10%), reliable virtual samples can still be generated, ensuring the high generalization and prediction accuracy of the final dioxin emission prediction model.

[0031] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and other objects, features and advantages of the exemplary embodiments of the present application will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present application are shown in an illustrative and non-limiting manner, in which: In the drawings, the same or corresponding reference numerals denote the same or corresponding parts.

[0033] Figure 1 This is a schematic diagram of the overall design idea architecture of this application.

[0034] Figure 2 This is the MD-MTD schematic diagram.

[0035] Figure 3 is the prediction performance graph of the final random forest model on the training set.

[0036] Figure 4 is the prediction performance graph of the final random forest model on the test set.

[0037] Figure 5 This is a comparison chart of RMSE and MAPE of the random forest model on the test set during the initial and 4 iterations.

[0038] Figure 6 This is a comparison chart of the predicted values ​​and true values ​​of the test set of the random forest model during the initial and 4 iterations. DETAILED DESCRIPTION

[0039] In order to make the purpose, features, and advantages of this application more obvious and easy to understand, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of this application.

[0040] Example 1: This embodiment provides a dioxin emission prediction method based on RF-PSO integrated algorithm. Figure 1 , including the following steps: S1: Construct a data set; collect and analyze dioxin emissions from the end of the solid waste incineration system, and simultaneously collect incinerator operating parameters, 124-trichlorobenzene (124-TrCBz) concentrations, and other conventional pollutant concentrations.

[0041] The specific method is as follows: S11: Use EPA23a method to detect dioxin emission data at the end of solid waste incineration system, and use dioxin toxicity equivalent as the output variable in the data set; S12: The 124-TrCBz concentration and the concentrations of other conventional pollutants (such as carbon monoxide, nitrogen oxides, sulfur dioxide, ammonia, hydrogen chloride, and carbon dioxide) at the end of the solid waste incineration system are simultaneously detected, and the incinerator operating parameters are collected through the DCS system. The above parameters are used as input features in the original data set.

[0042] For example, an experiment conducted at a large-scale municipal solid waste incinerator in Ningbo, Zhejiang, was used to construct a long-term dioxin emission dataset. Samples were collected from the flue gas purification system's final stack. EPA 23a method was used to measure dioxin emissions. Simultaneously, 124-TrCBz concentrations were measured online at the same sampling point using thermal desorption-gas chromatography-tunable laser ionization-time-of-flight mass spectrometry. Conventional pollutant concentrations were simultaneously collected at the stack using a continuous emissions monitoring system, and incineration parameters were collected using a DCS system. The resulting long-term dioxin emission dataset consisted of 36 samples and 27 input features. The prediction target was dioxin emission levels from the municipal solid waste incineration process, expressed in International Toxicity Equivalents (I-TEQs). Input features included incinerator operating parameters, 124-TrCBz concentrations, and concentrations of other conventional pollutants.

[0043] S2: Dataset preprocessing: standardize the input features in the dataset, then perform dimensionality reduction on the standardized dataset using the Boruta algorithm, and finally randomly divide the reduced dimensionality dataset into training and test sets according to the specified ratio.

[0044] S21: The method for standardizing the input features of the dataset is as follows: Normalize the input features by removing the mean and scaling to unit variance, and perform the following calculation on each input feature individually:

[0045] in and Represent the standardized input features and original input features respectively, and represent the mean and standard deviation of the original input features respectively.

[0046] S22: Dimensionality reduction is performed on the standardized dataset using the Boruta algorithm: For each input feature Randomly arrange and generate corresponding shadow features It is worth mentioning that the input features here refer to the input features after S21 normalization. The input features and shadow features are concatenated to obtain the expanded feature matrix:

[0047] in, and The distribution of is the same, and there are 2p features after expansion; Train the random forest model on the expanded feature matrix and calculate the Z score (Z_score) of each feature (including input features and shadow features); Record the maximum Z_score of all shadow features as Z_score max , input features greater than Z_score max The features with a Z score less than the maximum Z score are marked as important features and retained, and the input features with a Z score less than the maximum Z score are marked as unimportant features and permanently deleted from the feature set; Remove all shadow features.

[0048] This embodiment also provides another method for performing dimensionality reduction on a standardized data set using the Boruta algorithm: Randomly arrange each input feature to generate the corresponding shadow feature; concatenate the input feature and the shadow feature to obtain the expanded feature matrix; A random forest model is trained on the expanded feature matrix, the random forest model is trained N times (N=1000 in the embodiment), and N Z scores of each feature are calculated; Each time the random forest model is run, the maximum Z score of all shadow features is recorded, and the input features with a Z score higher than the maximum Z score are marked once; After running the random forest model N times, the number of times each input feature is marked is calculated, recorded as Hits, and Hits obeys the Bernoulli distribution law:

[0049] When the Hits value of an input feature is significantly greater than the expected value, the feature is marked as an important feature; when it is significantly less than the expected value, the feature is deleted.

[0050] S23: Randomly divide the dataset after dimensionality reduction into training set and test set according to the ratio of training set group number: test set group number = 7:3.

[0051] S3: Random forest model training and testing, using the trained random forest model as the initial prediction model.

[0052] The specific methods of S3 are as follows: S31: Based on the training set, the random forest model was selected to construct a dioxin emission prediction model, and the grid search method and five-fold random cross-validation method were used for hyperparameter optimization.

[0053] Random forest is an ensemble learning method that improves generalization ability by building multiple decision trees and combining their predictions. For regression problems, the prediction value of random forest is the average of the prediction values ​​of all decision trees:

[0054] Where K is the number of trees, is the predicted value of the kth tree.

[0055] Each tree uses the Bootstrap method to randomly extract samples from the training set multiple times with replacement. The root of the decision tree is a randomly selected training sample. When the node is split, m features are randomly selected ( , p is the total number of features) to perform optimal splitting. For each feature, try different splitting points (thresholds). Assume that the feature The split point is t, the data is divided into left child node and right child node, and the feature The weighted mean square error (MSE) when splitting with the split point t:

[0056]

[0057]

[0058] in, and are the number of samples of the left child node and the right child node respectively, is the total number of samples of the current node, and is the mean of the target values ​​in the left and right child nodes. Select the features and split point that minimize the MSE after the split. If the decrease in the MSE of the current node is less than the set minimum split threshold, or the number of samples is less than the minimum number of samples to split, the split is stopped and the node becomes a leaf. Repeat the above process for the left and right child nodes until the stopping condition (such as reaching the maximum tree depth or the minimum number of samples for a node) is met. Repeat the above steps K times to obtain K decision trees, forming the random forest model.

[0059] This method uses the RandomForestRegressor model from the Python sklearn library, setting the random_state value to 42. A grid search is performed on predefined hyperparameter combinations of the random forest model: n_estimators (the number of decision trees in the random forest) and max_depth (the maximum depth of the decision tree). All other parameters are set to their default values. Five-fold randomized cross-validation is also used to select the optimal hyperparameters. The five-fold randomized cross-validation method is as follows: 1. Randomly divide the entire training set into 5 non-overlapping subsets, each with 20% of the data; 2. Each time, four subsets are used as training sets to train the random forest model, and the remaining subset is used as the validation set, for a total of five cycles; 3. For each loop, set the parameter optimization target to negative mean square error (neg-MSE); calculate the negative mean square error (neg-MSE) based on the validation set; 4. Calculate the average negative mean square error (neg-MSE) of the five cycles as the final score for the hyperparameter combination, and select the hyperparameter combination with the highest average neg-MSE (i.e., the combination with the smallest actual MSE).

[0060] S32: Test the generalization ability of the random forest model based on the test set, using the coefficient of determination (R 2 ), root mean square error (RMSE) and mean absolute percentage error (MAPE) as evaluation indicators:

[0061]

[0062]

[0063] Where N is the number of samples, and represent the predicted value and actual value of the target respectively, express The average value of .

[0064] In this embodiment, the n_estimators of the initial random forest model established is 100 and the max_depth is 9. The performance on the original dioxin emission dataset is shown in Table 1 below: Table 1: Performance on the original dioxin emission dataset

[0065] S4: Generate virtual sample input; expand the range of input features through the multi-distribution total trend diffusion (MD-MTD) technology, and generate virtual input in the expanded area of ​​the dataset and the actual observation area respectively.

[0066] The specific method of S4 is as follows: Each input feature is To expand, the principle of MD-MTD technology is as follows Figure 2 As shown in the figure, MD-MTD expands the acceptable range of the sample set from (min, max) to (LB, RB), where (min, max) represents the actual observation area of ​​the sample set, and (LB, min) and (max, RB) represent the extended area. Within the extended area, since the extended data is unknown, a uniform distribution is used to generate virtual samples. Within the actual observation area, a triangular distribution is used to generate virtual samples, i.e., more virtual samples are generated closer to the data center, and fewer virtual samples are generated farther away from the data center.

[0067] By characteristics For example: (1) Calculate the data center point CL

[0068] Among them, n is the number of samples, CL is the center point, X （.） is an order statistic.

[0069] (2) Calculate the left and right skewness and variance of the center point CL

[0070]

[0071]

[0072] Where S KL is the left skewness, S KR is the right skewness, N L 、N R is the number of samples whose values ​​are less than or greater than the center point CL, is the variance of the input feature X.

[0073] (3) Calculate the upper and lower bounds of the acceptable range

[0074]

[0075] Where LB is the lower bound of the acceptable range, and RB is the upper bound of the acceptable range.

[0076] (4) Calculate the number of virtual samples to be generated within each range

[0077]

[0078]

[0079]

[0080] Where, are the number of virtual samples to be generated in the range of (LB, min), (min, max), and (max, RB), respectively. The total number of virtual samples generated for the request.

[0081] (5) Generate virtual input Generate virtual sample input using uniform distribution within the extended range: (LB, min) range:

[0082] In the range of (max, RB):

[0083] Generate virtual sample input using triangular distribution within the direct observation range (min, max):

[0084] Where s is a random number that obeys the uniform distribution in (0,1).

[0085] For each input feature Expand and finally generate a virtual input dataset .

[0086] S5: Optimize the generated virtual samples through the RF-PSO integrated algorithm, and repeat this step until the generated virtual samples are mixed with the original training set and modeled to obtain satisfactory model performance (R on the test set). 2 ≥0.9, RMSE≤0.0350, MAPE≤25%).

[0087] Please refer to Figure 1 , S5 specifically includes the following: S51: Generate a virtual input dataset based on the random forest model built in S3 Corresponding virtual output ; S52: Dynamically adjust the input features of virtual samples based on the PSO algorithm to improve data augmentation effects. The specific steps are as follows: S521, particle swarm initialization: Input feature matrix with pre-generated virtual samples As the initial particle swarm position, each particle represents the input feature vector of a virtual sample; Generate a random initial particle velocity matrix in the interval [-1,1], with a dimension matching the input feature number; The particle positions are restricted to the characteristic upper and lower bounds (LB, RB) of the original data set to ensure that the virtual samples are physically meaningful.

[0088] S522, fitness function calculation: A pre-trained initial prediction model is used to map particle input features to output features to form a complete virtual sample; Merge a single virtual sample with the original training data to construct an augmented dataset, and retrain the random forest model based on the augmented dataset; The MAPE of the predicted value is calculated using the trained random forest model on the original training set as the fitness value of the particle. A smaller MAPE indicates a higher particle quality.

[0089] S523, particle swarm dynamic update: The inertia weight is calculated by the number of iterations from Linearly down to , balancing global exploration and local development capabilities:

[0090] in, is the updated inertia weight, is the current iteration number, is the maximum number of iterations.

[0091] Update particle velocity:

[0092] in, is a random vector, are the individual and group learning factors, and are the optimal positions of individuals and populations, is the particle velocity, is the particle position.

[0093] The particle position is adjusted based on the updated velocity and the bounding constraints are reapplied to ensure validity.

[0094] S524, optimal solution iteration: Compare the current fitness with the historical optimal value and retain the optimal position of each particle , achieving individual optimal update; After each iteration, the particle with the best fitness among all particles is selected as , achieving global optimal update.

[0095] S525, output optimization results: Output the complete virtual sample corresponding to the global optimal particle and its fitness value (minimum MAPE) to obtain a set of virtual samples; Perform PSO optimization once to obtain a set of virtual samples and repeat the above process times, get A virtual sample dataset.

[0096] S53: Set the number of samples to The virtual sample dataset is mixed with the original training set, and the random forest model is rebuilt. Hyperparameter optimization is performed using grid search and five-fold randomized cross-validation. The MAPE is calculated based on the same test set to evaluate the performance of the random forest model. If the model performance improves, return to S4, generate virtual sample input again, and execute S51 and S52 in sequence (the model in step 2 of S52 needs to be updated, and the particle fitness function needs to be calculated using the better-performing random forest model). Otherwise, proceed to S54.

[0097] S54: Retain the current virtual sample as the qualified virtual sample dataset and mix it with the original training set. Rebuild the random forest model and optimize hyperparameters using grid search and five-fold randomized cross-validation. Calculate the MAPE based on the same test set to evaluate model performance.

[0098] In this example, after 4 iterations of the RF-PSO ensemble algorithm and adding 120 virtual samples to the original training set, the random forest model was rebuilt. The performance of the final random forest model on the mixed training set and test set is shown in Table 2 below. The prediction performance graphs on the mixed training set and test set are respectively Figure 3 and Figure 4 .

[0099] Table 2: Performance of the final random forest model on the mixed training and test sets

[0100] The RMSE and MAPE of the random forest model on the test set were reduced by 42.98% and 36.95%, respectively, and the performance was significantly improved. More than 80% of the dioxin concentration prediction errors were within [-40%, 40%], and the dioxin emission concentration prediction errors were within [-0.071, 0.058] ng I-TEQ / Nm 3 The results show that the established RF-PSO integrated algorithm and modeling framework can be well applied to the prediction of dioxin emissions from small sample domestic waste incineration process. At the same time, the comparison of RMSE and MAPE of the initial random forest model and the random forest model in the 4-iteration process on the test set is shown in Figure 5 , the comparison between the test set prediction value and the true value is shown in Figure 6 As can be seen from the figure, the performance of the model gradually improves during the iteration process, which also proves that this method can reduce the impact of the weak performance of the initial prediction model and generate reliable virtual samples during the iteration process.

[0101] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this application can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of this application can be achieved. This is not limited herein.

[0102] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0103] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A dioxin emission prediction method based on RF-PSO integrated algorithm, characterized in that: The following steps are involved: S1, constructing a data set, with dioxin toxicity equivalent as the output variable in the data set, and incinerator operating parameters and 124-TrCBz concentration as the input features in the data set; S2, dataset preprocessing, standardizes and reduces the dimension of the input features in the dataset, and divides the dataset into training set and test set; S3, random forest model training and testing, using the trained random forest model as the initial prediction model for dioxin emissions; S4, generate virtual sample input; The range of input features is expanded through multi-distribution overall trend diffusion, virtual inputs are generated in the extended area of ​​the dataset and the actual observation area, and a virtual input dataset is constructed; S5, the generated virtual samples are optimized by the RF-PSO integrated algorithm.

2. The dioxin emission prediction method based on the RF-PSO integrated algorithm according to claim 1 is characterized in that: The specific method of S1 is as follows: S11: Use EPA23a method to detect dioxin emission data at the end of solid waste incineration system, and use dioxin toxicity equivalent as the output variable in the data set; S12: The 124-TrCBz concentration and the concentration of other conventional pollutants at the end of the solid waste incineration system are simultaneously detected, and the incinerator operating parameters are collected through the DCS system, and the above parameters are used as input features in the original data set.

3. The dioxin emission prediction method based on the RF-PSO integrated algorithm according to claim 2 is characterized in that: The other conventional pollutants include carbon monoxide, nitrogen oxides, sulfur dioxide, ammonia, hydrogen chloride and carbon dioxide.

4. The dioxin emission prediction method based on RF-PSO integrated algorithm according to claim 1, characterized in that: In S2, the method for performing dimensionality reduction processing on the input features in the data set is as follows: The standardized data set is subjected to dimensionality reduction processing using the Boruta algorithm; Randomly arrange each input feature to generate the corresponding shadow feature; concatenate the input feature and the shadow feature to obtain the expanded feature matrix; Training a random forest model on the expanded feature matrix and calculating the Z score of each feature, including input features and shadow features; Record the maximum Z score of all shadow features, mark the input features with a score greater than the maximum Z score as important features and retain them, and mark the input features with a score less than the maximum Z score as unimportant features and permanently delete them from the feature set; Remove all shadow features.

5. The dioxin emission prediction method based on RF-PSO integrated algorithm according to claim 1, characterized in that: In S3, the random forest model training and testing methods are as follows: S31: Based on the training set, the random forest model was selected to construct the dioxin emission prediction model, and the grid search method and five-fold random cross-validation method were used for hyperparameter optimization; S32, tests the generalization ability of the random forest model based on the test set, using the coefficient of determination, root mean square error, and mean absolute percentage error as evaluation indicators.

6. The dioxin emission prediction method based on RF-PSO integrated algorithm according to claim 5, characterized in that: In S31, the method for performing hyperparameter optimization using the grid search method and the five-fold random cross-validation method is as follows: A grid search was performed to traverse the predefined hyperparameter combinations of the random forest model: the number of decision trees in the random forest and the maximum depth of the decision tree. All other parameters were set to their default values. The optimal hyperparameters were selected using a five-fold randomized cross-validation method. The five-fold randomized cross-validation method was used as follows: The entire training set is randomly divided into 5 non-overlapping subsets, each with 20% of the data; Each time, 4 subsets are used as training sets to train the random forest model, and the remaining 1 subset is used as the validation set, for a total of 5 cycles; For each cycle, set the parameter optimization target to negative mean square error; calculate the negative mean square error based on the validation set; Calculate the average of the negative mean square error of 5 cycles as the final score of the hyperparameter combination, and select the hyperparameter combination with the highest average negative mean square error.

7. The dioxin emission prediction method based on RF-PSO integrated algorithm according to claim 1, characterized in that: In S4, the method for generating virtual sample input is as follows: S41, calculate the data center point CL; Among them, n is the number of samples, CL is the center point, X （.） is the order statistic; S42, calculating the left and right skewness and variance of the center point CL; Where S KL is the left skewness, S KR is the right skewness, N L 、N R is the number of samples whose values ​​are less than or greater than the center point CL, is the variance of the input feature X; S43, calculating the upper and lower bounds of the acceptable range; Where LB is the lower bound of the acceptable range, and RB is the upper bound of the acceptable range; S44, calculating the number of virtual samples to be generated within each range; Where, are the number of virtual samples to be generated in the range of (LB, min), (min, max), and (max, RB), respectively. The total number of virtual samples generated for the request; S45, generating a virtual input; Generate virtual sample input using uniform distribution within the extended range: (LB, min) range: In the range of (max, RB): Generate virtual sample input using triangular distribution within the direct observation range (min, max): Where s is a random number that obeys uniform distribution within (0,1); For each input feature Expand and finally generate a virtual input dataset .

8. The dioxin emission prediction method based on RF-PSO integrated algorithm according to claim 1, characterized in that: In S5, the method for optimizing the generated virtual samples by the RF-PSO integrated algorithm is as follows: S51: Generate a virtual output corresponding to a virtual input data set based on the random forest model; S52: Dynamically adjust the input features of virtual samples based on the PSO algorithm; S53: Mix the virtual sample data set with the original training set, re-establish the random forest model, and use the grid search method and five-fold random cross-validation method to optimize the hyperparameters; if the model performance is improved, return to step S4, generate virtual sample input again, and execute S51 and S52 in sequence; otherwise, go to S54; S54, retain the current virtual sample as the qualified virtual sample dataset, mix it with the original training set, establish the random forest model again, and use the grid search method and five-fold random cross-validation method to optimize the hyperparameters.

9. The dioxin emission prediction method based on RF-PSO integrated algorithm according to claim 1, characterized in that: In S52, the method for dynamically adjusting the input features of the virtual sample based on the PSO algorithm is as follows: S521, particle swarm initialization: Input feature matrix with pre-generated virtual samples As the initial particle swarm position, each particle represents the input feature vector of a virtual sample; Generate a random initial particle velocity matrix in the interval [-1,1], with a dimension matching the input feature number; Limit the particle positions to the characteristic upper and lower bounds (LB, RB) of the original data set; S522, fitness function calculation: A pre-trained initial prediction model is used to map particle input features to output features to form a complete virtual sample; Merge a single virtual sample with the original training data to construct an augmented dataset, and retrain the random forest model based on the augmented dataset; Use the trained random forest model to calculate MAPE on the original training set as the fitness value of the particle; S523, particle swarm dynamic update: The inertia weight is calculated by the number of iterations from Linearly down to , balancing global exploration and local development capabilities: in, is the updated inertia weight, is the current iteration number, is the maximum number of iterations; Update particle velocity: in, is a random vector, are the individual and group learning factors, and are the optimal positions of individuals and populations, is the particle velocity, is the particle position; Adjust the particle position according to the updated velocity and reapply the boundary constraints to ensure validity; S524, optimal solution iteration: Compare the current fitness with the historical optimal value and retain the optimal position of each particle , achieving individual optimal update; After each iteration, the particle with the best fitness among all particles is selected as , achieving global optimal update; S525, output optimization results: Output the complete virtual samples corresponding to the global optimal particle and its fitness value to obtain a set of virtual samples; Perform PSO optimization once to obtain a set of virtual samples and repeat the above process times, get A virtual sample dataset.

Images