A method for dividing and identifying temperature working conditions of a decomposition furnace under waste co-processing

Abstract

The application discloses a kind of decomposition furnace temperature working condition division and identification method under garbage collaborative disposal, its steps include:1) using WK-Means algorithm to carry out decomposition furnace temperature working condition division;2) using IPSO-SVM algorithm to carry out decomposition furnace temperature working condition identification.The application is aimed at the key parameter in the process of decomposition furnace temperature control under garbage collaborative disposal to carry out working condition division and identification, so as to improve the efficiency and accuracy of decomposition furnace temperature multi-working condition division and identification under garbage collaborative disposal.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing and pattern recognition, and in particular to a method for decomposing furnace temperature condition division and identification under garbage cooperative disposal. BACKGROUND

[0002] The cement industry has increasingly high requirements for production efficiency and product quality, and the decomposition furnace, as one of the core components of the cement production line, has an important influence on the calcination process. By accurately dividing and identifying the temperature condition of the decomposition furnace, the heat transfer, combustion and heat transfer process in the furnace can be better understood. In addition, the division and identification of temperature conditions can also help to monitor the running state of the decomposition furnace in real time and help to find potential faults or abnormal conditions. Research on temperature conditions can help optimize the distribution and transfer of heat energy in the furnace, reduce energy waste and loss. By reasonably adjusting parameters such as air flow direction and combustion conditions, energy consumption and carbon emissions can be reduced to achieve the goal of energy saving and emission reduction.

[0003] In the past, the division and identification of decomposition furnace temperature conditions mainly relied on experience and manual analysis. Operators judge the characteristics of temperature changes according to experience and observation, and then classify and identify the working conditions. This method has the problems of subjectivity and dependence on experience, and cannot handle large-scale data and complex temperature patterns. With the advancement of sensor technology, the division and identification of decomposition furnace temperature conditions increasingly rely on sensor data acquisition and processing. Using different types of sensors, such as thermocouples, infrared thermometers and optical fiber sensors, can monitor the temperature changes in the decomposition furnace in real time and transmit data to the computer system for analysis and processing of data-driven methods. However, sensor data acquisition and processing systems involve hardware devices, data transmission, data storage and algorithm models, etc. The complexity and dependence of the system increase the difficulty of maintenance and management, and also increase the risk of system failure or instability. Moreover, when dividing and identifying the working conditions, feature extraction and selection are needed to extract the most representative features to represent the temperature conditions. However, there may be subjectivity and fixed pattern limitations in the feature extraction and selection process, which may not accurately reflect the true temperature conditions. In recent years, data-driven methods, especially machine learning and deep learning techniques, have been widely used in the division and identification of decomposition furnace temperature conditions. However, a large amount of training data and a long training time are required, and feature data needs to be found. In addition to data-driven methods, there are also some methods based on physical models for the division and identification of decomposition furnace temperature conditions. However, this method requires a deep understanding of the physical properties of the decomposition furnace and modeling ability, and has high requirements for the accuracy of initial parameters and conditions. They are often limited by complexity, parameter estimation, computational complexity and real-time factors in practical applications. SUMMARY

[0004] The present application is to solve the above-mentioned deficiencies in the prior art, proposes a kind of garbage collaborative disposal under the temperature condition division and identification of decomposing furnace, to be able to quickly and accurately divide and identify the temperature condition of decomposing furnace, so as to realize the stable control of decomposing furnace temperature under the condition of garbage collaborative disposal, and can improve the efficiency and accuracy of decomposing furnace temperature multi-condition identification.

[0005] To achieve the above-mentioned application purposes, the technical scheme is as follows:

[0006] The features of the present application, a kind of garbage collaborative disposal under the temperature condition division and identification method of decomposing furnace, are as follows:

[0007] Step 1. Temperature condition division of decomposing furnace:

[0008] Step 1.1. Obtain the decomposing furnace condition sample set U = {U1, U2, …, U k ,…,U n}, wherein, U k represents the kth decomposing furnace condition sample, and U k ={u k,1 ,u k,2 ,…,u k,q ,…,u k,Q}; wherein, u k,q represents the qth dimensional feature vector in the kth decomposing furnace condition sample U k in the decomposing furnace condition sample set U, Q represents dimension, and n represents sample number;

[0009] Step 1.2. Define the current iteration number as L, and initialize L = 1; the number of temperature conditions of decomposing furnace is used as the number of cluster centers, and is recorded as f, and the f cluster centers {S1 L ,S2 L ,…,S d L ,…,S f L} of the Lth iteration are initialized using HDBSCAN algorithm, wherein, represents the dth cluster center of the Lth iteration;

[0010] Step 1.3. Calculate the Euclidean distance of the kth decomposing furnace condition sample U k to each cluster center of the Lth iteration wherein, represents the Euclidean distance of the kth decomposing furnace condition sample U k to the dth cluster center of the Lth iteration Euclidean distances; thereby obtaining the Euclidean distances from n decomposition furnace working condition samples to f cluster centers of the Lth iteration respectively, and assigning each decomposition furnace working condition sample to the cluster of the cluster center of the Lth iteration with the nearest Euclidean distance, thereby obtaining f clusters;

[0011] Step 1.4. For each cluster, calculate the weighted Euclidean distance mean of all decomposition furnace working condition samples in the cluster and take it as the new cluster center, thereby obtaining f clusters of the (L+1)th iteration and their respective cluster centers wherein, represents the dth cluster center of the (L+1)th iteration;

[0012] Step 1.5. Determine whether is established, if yes; output f clusters of the (L+1)th iteration as the temperature working conditions of f decomposition furnaces respectively; otherwise, assign L+1 to L, and return to execute step 1.3 in sequence; wherein, Δ represents a threshold value;

[0013] Step 2. Identify the decomposition furnace temperature working conditions:

[0014] Step 2.1. According to the number f of temperature working conditions, set corresponding temperature working condition labels for f clusters of the (L+1)th iteration respectively, each cluster corresponding to a temperature working condition label, thereby obtaining a labeled decomposition furnace temperature working condition dataset, and divide it into a training set and a test set in proportion;

[0015] Step 2.2. Initialize L=1, and set the maximum number of iterations as L Max ;

[0016] Define and initialize two learning factors as c1 and c2, the maximum value of the inertia weight as ω max , and the minimum value as ω min ;

[0017] Define the size of the particle swarm as N;

[0018] Define the parameter size of the SVM model as M;

[0019] Let the position of each particle in the (L-1)th generation particle swarm consist of M parameters of the SVM model, and randomly initialize each parameter between 0 and 10, thereby obtaining the position of the hth particle in the (L-1)th generation particle swarm, denoted as wherein, represents the position of the hth particle in the (L-1)th generation particle swarm in the mth dimension, i.e., the mth parameter in the SVM model parameter; 1≤h≤N; 1≤m≤M;

[0020] Let the speed of the hth particle in the (L-1)th generation particle swarm consist of M random numbers between 0 and 1, and denote it as wherein, denotes the velocity of the hth particle in the (L-1)th population of particles in the mth dimension;

[0021] Step 2.3. Let the position of the hth particle in the (L-1)th population of particles be As the local optimal solution of the hth particle in the (L-1)th population of particles, let wherein, is the local optimal solution of the mth parameter in the SVM model parameters corresponding to the hth particle in the (L-1)th population of particles, let The fitness of is and initialize

[0022] Let the local optimal solution of the hth particle in the (L-1)th population of particles be As the global optimal solution of the (L-1)th population of particles, let wherein, is the global optimal solution of the mth parameter in the SVM model parameters corresponding to the (L-1)th population of particles; let L-1 The fitness of is and initialize

[0023] Step 2.4. Calculate the inertia weight ω of the Lth iteration using formula (1) L-1 :

[0024]

[0025] Step 2.5. Obtain the position and velocity of the hth particle in the Lth population of particles using formula (2) Thus, the position and velocity of each particle in the Lth population of particles are obtained.

[0026]

[0027] In formula (2), r1 and r2 are two random numbers between 0 and 1.

[0028] Step 2.6. Initialize h = 1.

[0029] Step 2.7. Let the position of the hth particle in the Lth population of particles be As the parameters of the SVM model, and after initializing the SVM model, use the training set to train the initialized SVM model, and obtain the hth trained SVM model in the Lth population of particles.

[0030] The test set is input into the h-th trained SVM model in the L-th generation particle swarm for classification prediction. The prediction results are compared with the actual temperature condition labels to obtain the classification accuracy as the fitness of the h-th particle in the L-th generation particle swarm.

[0031] Step 2.8. Judgment If true, determine the position of the h-th particle in the L-th generation particle swarm. As the local optimal solution for the h-th particle in the L-th generation particle swarm. Right now And the fitness of the local optimal solution of the h-th particle in the L-th generation particle swarm The fitness of the h-th particle in the L-th generation particle swarm. Right now Otherwise, As the local optimal solution for the h-th particle in the L-th generation particle swarm. Right now Will As fitness Right now

[0032] Step 2.9. Judgment If it holds true, then find the local optimal solution of the h-th particle in the L-th generation particle swarm. R, the global optimal solution of the Lth generation particle swarm. L ,Right now Will fitness R, the global optimal solution of the Lth generation particle swarm. L fitness Right now Otherwise, R L-1 R, the global optimal solution of the Lth generation particle swarm. L R L =R L-1 ,Will Fitness as the global optimal solution Right now

[0033] Step 2.10. After assigning h+1 to h, repeat step 2.7 sequentially until h>N, thereby obtaining the local optimal solution P of N particles in the Lth generation particle swarm. L ={P1 L P2 L ,…,P h L ,…,P N L} and its fitness, the global optimal solution R of the Lth generation particle swarmL and fitness of the individual;

[0034] Step 2.11. After L+1 is assigned to L, it is judged whether L>L Max is established, step 2.12 is executed, otherwise, step 2.4 is returned to be executed in sequence;

[0035] Step 2.12. The global optimal solution of the L Max th particle swarm is output and the global optimal solution is taken as the optimal parameter of the SVM model, so that the SVM model with the optimal parameter is used to realize the working condition recognition of the decomposition furnace temperature.

[0036] The electronic device comprises a memory and a processor, and the feature lies in that the memory is used to store a program supporting the processor to execute the decomposition furnace temperature working condition division and recognition method, and the processor is configured to execute the program stored in the memory.

[0037] The computer readable storage medium comprises a computer program stored thereon, and the feature lies in that the computer program is run by the processor to execute the steps of the decomposition furnace temperature working condition division and recognition method.

[0038] Compared with the prior art, the beneficial effects of the present application are as follows:

[0039] 1.The present application adopts WK-Means algorithm to divide the temperature condition of the decomposition furnace, WK-Means algorithm is an improved version of the traditional K-Means algorithm, its main idea is to introduce a weighting factor to weight different features, in order to reduce the weight gap between different features. It does not require a large amount of computing resources and complex parameter settings, can adapt to complex furnace characteristics to automatically extract useful features, and automatically cluster temperature data; the present application has improved on the basis of the existing WK-Means algorithm, HDBSCAN algorithm is used when initializing the cluster center, HDBSCAN is a density-based clustering algorithm, which can effectively identify clusters with different densities, and can automatically determine the number of cluster centers without prior specification, by using HDBSCAN algorithm to determine the initial cluster center, the local structure and clustering characteristics of the data can be better captured. This helps to avoid the problem of WK-Means algorithm falling into local optimal solution due to poor initial clustering. And the decomposition furnace temperature condition data may contain noise and outliers, these abnormal points have a greater impact on the clustering results. HDBSCAN algorithm can identify and exclude noise and outliers in low-density areas, and process them as separate clusters or noise points, thereby improving the robustness and reliability of the decomposition furnace condition temperature data clustering. In the initial stage, noise and outliers are excluded from the clustering process, reducing the interference of these data on the subsequent clustering results.

[0040] 2.IPSO-SVM algorithm is a kind of optimization algorithm based on particle swarm algorithm, which can optimize the parameter setting of the classifier to achieve better working condition recognition effect. Compared with the traditional POS-SVM algorithm, IPSO-SVM algorithm can overcome the limitations of traditional POS-SVM algorithm in dealing with complex problems and high-dimensional data, and can also identify and classify different clusters through SVM classifier to achieve more refined identification of temperature condition. It uses an adaptive way to adjust and optimize parameters, without long training time and modeling ability; and the present application has improved on the basis of the existing IPSO-SVM algorithm, the parameter combination of the SVM model is used as the position attribute of the particle, and the classification accuracy of the trained SVM model on the test set is used as the fitness of the particle, which can help to find a better parameter combination, and can realize parameter optimization, search parameter space, adaptability and reduce computational complexity in IPSO-SVM algorithm, so as to improve the recognition accuracy of the decomposition furnace temperature condition, can more reliably identify unknown conditions, adapt to different scenes and problems, and reduce the amount of manual work, improve the safety and efficiency of industrial processes.

[0041] 3.Overall, the application adopts WK-Means and IPSO-SVM algorithms to automatically divide and identify the working conditions through data analysis and optimization algorithms, reduces the dependence on artificial experience, can adapt to complex furnace temperature characteristics and nonlinear change modes, improves the accuracy and stability of working condition division and identification, fully utilizes a large amount of temperature data and statistical characteristics, can mine hidden information in the data and decompose the furnace temperature characteristics, and does not require a long training time, so it has higher efficiency for working condition division and identification. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The flowchart of the WK-Means algorithm of the application is shown in the figure.

[0043] Figure 2 The accuracy rate change graph of the SVM model in the iteration process of the IPSO-SVM algorithm of the application is shown in the figure.

[0044] Figure 3 The comparison graph before and after the improvement of the PSO-SVM of the application is shown in the figure.

[0045] Figure 4 The flowchart of the IPSO-SVM algorithm of the application is shown in the figure. DETAILED DESCRIPTION

[0046] In this embodiment, a method for dividing and identifying the working conditions of the decomposition furnace temperature under the waste cooperative disposal is performed according to the following steps:

[0047] Step 1. Divide the working conditions of the decomposition furnace temperature:

[0048] Step 1.1. Obtain the decomposition furnace working condition sample set U = {U1, U2, …, U k ,…,U n}, wherein U k represents the kth decomposition furnace working condition sample, and U k = {u k,1 ,u k,2 ,…,u k,q ,…,u k,Q}; wherein u k,q represents the kth decomposition furnace working condition sample U kthe q-th dimensional eigenvector, Q represents the dimension, and n represents the number of samples; in this embodiment, the outlet temperature of the decomposition furnace, the derivative of the outlet temperature of the decomposition furnace, and the CO concentration are taken as the eigenvectors of the samples in the decomposition furnace working condition sample set. Because the outlet temperature of the decomposition furnace, the derivative of the outlet temperature of the decomposition furnace, and the CO concentration are taken as the eigenvectors, the working conditions of the decomposition furnace can be comprehensively reflected to a certain extent. These features can provide information about temperature distribution, temperature change rate, and combustion and cracking quality effect, which is helpful for monitoring, identifying, and dividing the running state of the decomposition furnace, so as to realize the analysis and optimization of the working conditions of the decomposition furnace.

[0049] Step 1.2. Define the current iteration number as L, and initialize L = 1; take the number of temperature working conditions of the decomposition furnace as the number of cluster centers and mark it as f, and initialize f cluster centers {S1 L ,S2 L ,…,S d L ,…,S f L} of the L-th iteration by using the HDBSCAN algorithm, wherein, represents the d-th cluster center of the L-th iteration; in this embodiment, initializing the cluster center by using the HDBSCAN algorithm can reduce parameter dependence, adapt to various cluster shapes and sizes, automatically process noise and outliers, and provide efficient and scalable cluster analysis, which is helpful for avoiding the problem that the WK-Means algorithm falls into a local optimal solution due to poor initial clustering.

[0050] Step 1.3. Calculate the Euclidean distance of the k-th decomposition furnace working condition sample U k to each cluster center of the L-th iteration wherein, represents the Euclidean distance of the k-th decomposition furnace working condition sample U k to the d-th cluster center of the L-th iteration; thereby obtaining the Euclidean distances of n decomposition furnace working condition samples to f cluster centers of the L-th iteration, and distributing each decomposition furnace working condition sample to the cluster to which the cluster center of the L-th iteration with the nearest Euclidean distance belongs, thereby obtaining f clusters;

[0051] Step 1.4. For each cluster, calculate the weighted Euclidean distance mean of all decomposition furnace working condition samples in the cluster and take it as a new cluster center, set the weights of the three features as [0.1410, 0.7510, 0.1080] according to the process characteristics, thereby obtaining f clusters and their respective cluster centers of the L+1-th iteration wherein, represents the d-th cluster center of the L+1-th iteration;

[0052] Step 1.5. Determine If true, output f class clusters of the L+1th iteration as the temperature conditions of f decomposition furnaces respectively; otherwise, assign L+1 to L, and return to execute step 1.3 sequentially; wherein Δ represents a threshold value; the flowchart of the WK-Means algorithm is shown in Figure 1 .

[0053] Step 2. Identify the temperature conditions of the decomposition furnace:

[0054] Step 2.1. According to the number f of temperature conditions, set corresponding temperature condition labels for f class clusters of the L+1th iteration, each class cluster corresponding to a temperature condition label, so as to obtain a labeled decomposition furnace temperature condition data set, and divide it into a training set and a test set in proportion;

[0055] Step 2.2. Initialize L=1, and set the maximum number of iterations as L Max . Figure 2 The change of SVM accuracy rate in the iteration process. It can be seen from Figure 2 that the accuracy rate of SVM increases with the increase of the number of iterations, and the increase of the accuracy rate of SVM becomes smaller and smaller with the increase of the number of iterations; in this embodiment, L Max is 100.

[0056] Define and initialize two learning factors as c1 and c2, the maximum value of the inertia weight as ω max , and the minimum value as ω min .

[0057] Define the size of the particle swarm as N;

[0058] Define the parameter size of the SVM model as M;

[0059] In this embodiment, the improved PSO-SVM algorithm is used, and under the premise of ensuring the same parameters, it is compared with the unimproved algorithm. The parameters are shown in Table 1.

[0060] Table 1 Comparison of parameters before and after IPSO-SVM improvement

[0061]

[0062] The flowchart of the IPSO-SVM algorithm is shown in Figure 4 .

[0063] Let the position of each particle in the L-1th generation particle swarm consist of M parameters of the SVM model, and randomly initialize each parameter between 0 and 10, so that the position of the hth particle in the L-1th generation particle swarm is recorded as wherein, represents the position of the hth particle in the (L-1)th generation particle group in the mth dimension, that is, the mth parameter in the SVM model parameter; 1≤h≤N; 1≤m≤M; combining the parameters of the SVM model as the particle position can better model the parameter space, realize parameter optimization, maintain the continuity of the solution space, and provide an index for evaluating and ranking the performance of the particles. This helps the IPSO-SVM algorithm to find the best solution in the parameter search of the SVM model, thereby improving the recognition accuracy of the temperature working condition of the decomposition furnace.

[0064] Let the velocity of the hth particle in the (L-1)th generation particle group be composed of M random numbers between 0 and 1, denoted as wherein, represents the velocity of the hth particle in the (L-1)th generation particle group in the mth dimension;

[0065] Step 2.3. Let the position of the hth particle in the (L-1)th generation particle group be as the local optimal solution of the hth particle in the (L-1)th generation particle group, denoted as wherein, is the local optimal solution of the mth parameter in the SVM model parameter corresponding to the hth particle in the (L-1)th generation particle group, let The fitness of is and initialize

[0066] Let the local optimal solution of the hth particle in the (L-1)th generation particle group be as the global optimal solution of the (L-1)th generation particle group, denoted as wherein, is the global optimal solution of the mth parameter in the SVM model parameter corresponding to the (L-1)th generation particle group; let R L-1 The fitness of is and initialize

[0067] Step 2.4. Calculate the inertia weight ω of the (L-1)th iteration using formula (1) L-1 :

[0068]

[0069] Step 2.5. Obtain the position and velocity of the hth particle in the Lth generation particle group using formula (2) , thereby obtaining the position and velocity of each particle in the Lth generation particle group;

[0070]

[0071] In formula (2), r1 and r2 are two random numbers between 0 and 1;

[0072] Step 2.6. Initialize h = 1;

[0073] Step 2.7. Position the h-th particle in the L-th generation particle swarm. The parameters are used as parameters for the SVM model. After initializing the SVM model, the training set is used to train the initialized SVM model to obtain the h-th trained SVM model in the L-th generation particle swarm.

[0074] The test set is input into the h-th trained SVM model in the L-th generation particle swarm for classification prediction. The prediction results are compared with the actual temperature condition labels to obtain the classification accuracy as the fitness of the h-th particle in the L-th generation particle swarm. In the IPSO-SVM algorithm, using the classification accuracy of the SVM model as the fitness function allows the optimization algorithm to search for the optimal solution in a goal-oriented manner. This effectively measures the model's performance, guides the evolutionary search process, and aligns with the optimization objective. This improves the algorithm's efficiency, performance, and interpretability, making it better suited for solving classification problems.

[0075] Step 2.8. Judgment If true, determine the position of the h-th particle in the L-th generation particle swarm. As the local optimal solution for the h-th particle in the L-th generation particle swarm. Right now And the fitness of the local optimal solution of the h-th particle in the L-th generation particle swarm The fitness of the h-th particle in the L-th generation particle swarm. Right now Otherwise, As the local optimal solution for the h-th particle in the L-th generation particle swarm. Right now Will As fitness Right now Update the local optimum and fitness for each particle.

[0076] Step 2.9. Judgment If it holds true, then find the local optimal solution of the h-th particle in the L-th generation particle swarm. R, the global optimal solution of the Lth generation particle swarm. L ,Right now Will fitness R, the global optimal solution of the Lth generation particle swarm. L fitness Right now Otherwise, R L-1R as the global optimal solution of the Lth generation particle group L , i.e. R L = R L-1 , and fitness of the global optimal solution , i.e. updating the global optimal solution and its fitness of each generation particle group.

[0077] Step 2.10. After assigning h+1 to h, repeat step 2.7 sequentially until h>N, thereby obtaining the local optimal solution P L = {P1 L , P2 L , …, P h L , …, P N L} and its fitness of N particles in the Lth generation particle group, the global optimal solution R L and its fitness of the Lth generation particle group.

[0078] Step 2.11. After assigning L+1 to L, determine whether L>L Max is true, if true, execute step 2.12, otherwise, return to step 2.4 for sequential execution.

[0079] Step 2.12. output the global optimal solution R Max of the Lth generation particle group and the global optimal solution R as the optimal parameters of the SVM model, thereby realizing the working condition recognition of the decomposition furnace temperature by the SVM model with the optimal parameters. Figure 3 Comparison chart of PSO-SVM before and after improvement.

[0080] In this embodiment, an electronic device includes a memory for storing a program supporting a processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0081] In this embodiment, a computer readable storage medium has a computer program stored thereon, and the computer program is executed by a processor to perform the steps of the above method.

[0082] Table 2 is a comparison of the final results of the accuracy of the two algorithms in recognizing the working condition of the decomposition furnace temperature. Figure 3 Table 2 IPSO-SVM and PSO-SVM algorithm result comparison

[0083]

[0084]

[0085] ​As shown in Table 2, IPSO-SVM reduces the number of abnormal points in the identification of the temperature working condition of the decomposing furnace and improves the accuracy of the identification of the temperature working condition of the decomposing furnace.

Claims

1. A method for temperature condition division and identification of a decomposition furnace under waste co-processing, characterized in that, is carried out as follows: Step 1. Temperature of the decomposition furnace is divided into working conditions: Step 1.

1. Obtain a set of decomposition furnace working condition samples U = {U1, U2, …, U k ,…,U n}, wherein U k represents the kth decomposition furnace working condition sample, and U k = {u k,1 ,u k,2 ,…,u k,q ,…,u k,Q}; wherein u k,q represents the qth dimensional feature vector in the kth decomposition furnace working condition sample U k in the set of decomposition furnace working condition samples U, Q represents the dimension, and n represents the sample quantity; Step 1.

2. Define the current iteration number as L, and initialize L = 1; take the number of kinds of temperature conditions of the decomposition furnace as the number of cluster centers and record it as f, and initialize f cluster centers {S1 L ,S2 L ,…,S d L ,…,S f L} of the Lth iteration using the HDBSCAN algorithm, wherein, represents the dth cluster center of the Lth iteration; Step 1.

3. Calculate the Euclidean distance of the kth decomposition furnace working condition sample U to the dth cluster center of the Lth iteration k Euclidean distance of the kth decomposition furnace working condition sample U to the dth cluster center of the Lth iteration wherein, represents the kth decomposition furnace working condition sample U k Euclidean distance of the kth decomposition furnace working condition sample U to the dth cluster center of the Lth iteration ; thus obtaining the Euclidean distances of the n decomposition furnace working condition samples to the f cluster centers of the Lth iteration respectively, and assigning each decomposition furnace working condition sample to the cluster to which the cluster center of the Lth iteration with the closest Euclidean distance belongs, thus obtaining f clusters. Step 1.

4. For each cluster, calculate the weighted Euclidean distance mean of all the decomposition furnace operating condition samples within the cluster and take it as the new cluster center, so as to obtain f clusters and their respective cluster centers in the L+1th iteration wherein, represents the dth cluster center in the L+1th iteration; Step 1.

5. judging whether the following condition is established: if yes, output the f class clusters of the L+1th iteration as the temperature conditions of the f decomposition furnaces; otherwise, assign L+1 to L, and return to execute step 1.3 sequentially; wherein Δ represents a threshold value. Step 2. The decomposition furnace temperature working condition is identified: Step 2.

1. According to the number of temperature working conditions f, f class clusters of the L+1th iteration are respectively set with corresponding temperature working condition labels, each class cluster corresponds to a temperature working condition label, thereby obtaining a labeled decomposition furnace temperature working condition data set, and dividing it into a training set and a test set in proportion; Step 2.

2. Initialize L = 1, let the maximum number of iterations be L Max ; Define and initialize two learning factors as c1 and c2, maximum value of inertia weight as ω max and minimum value of inertia weight as ω min ; The size of the particle swarm is defined as N; The size of the SVM model parameters is defined as M; Let the position of each particle in the L-1th generation particle group consist of M parameters of the SVM model, and each parameter is randomly initialized between 0 and 10, so that the position of the hth particle in the L-1th generation particle group is denoted as wherein, indicates the position of the hth particle in the L-1th generation particle group in the mth dimension, that is, the mth parameter in the SVM model parameter; 1≤h≤N; 1≤m≤M. Let the velocity of the hth particle in the (L-1)th generation particle swarm be composed of M random numbers between 0 and 1, and denoted as wherein, denotes the velocity of the hth particle in the (L-1)th generation particle swarm in the mth dimension. Step 2.

3. Let the position of the hth particle in the Lth generation particle swarm be Let the local optimum solution of the hth particle in the Lth generation particle swarm be wherein, is the local optimum solution of the mth parameter corresponding to the hth particle in the Lth generation particle swarm, and let The fitness of is and initialize Let the local optimal solution of the hth particle in the L-1th generation particle group be denoted as Let the global optimal solution of the L-1th generation particle group be denoted as Wherein, is the global optimal solution of the mth parameter in the SVM model parameter corresponding to the L-1th generation particle group; let R L-1 The fitness of the Lth generation particle group is And initialize Step 2.

4. Calculate the inertial weight ω for the L-lth iteration using formula (1) L-1 : Step 2.

5. Obtain the position of the hth particle in the Lth generation of particles using formula (2) and velocity to obtain the position and velocity of each particle in the Lth generation of particles. In formula (2), r1 and r2 are two random numbers between 0 and 1; Step 2.

6. Initialize h = 1; Step 2.

7. obtaining the position of the hth particle in the Lth generation particle swarm After the SVM model is initialized as the parameter and initialized, the initialized SVM model is trained using the training set to obtain the hth trained SVM model in the Lth generation particle swarm. The test set is input into the hth trained SVM model in the Lth generation of particle swarm for classification prediction, the obtained prediction result is compared with the real temperature condition label, and a classification accuracy is obtained as the fitness of the hth particle in the Lth generation of particle swarm Step 2.

8. judging whether the following equation is established, if yes, the position X h L the local optimal solution of the hth particle in the Lth generation of particle swarm i.e. and the fitness of the local optimal solution of the hth particle in the Lth generation of particle swarm is the fitness corresponding to the hth particle in the Lth generation of particle swarm i.e. otherwise, the hth particle in the Lth generation of particle swarm is replaced by the new particle the local optimal solution of the hth particle in the Lth generation of particle swarm i.e. the hth particle in the Lth generation of particle swarm is replaced by the new particle as the fitness of the hth particle in the Lth generation of particle swarm i.e. Step 2.

9. judging whether or not is established, if so, the local optimal solution P h L of the hth particle in the Lth generation particle swarm is taken as the global optimal solution R L of the Lth generation particle swarm, i.e. the fitness of R is taken as the fitness of the global optimal solution R L of the Lth generation particle swarm, i.e. Otherwise, R L-1 L is taken as the global optimal solution R L of the Lth generation particle swarm, i.e. R L-1 ​​​​​​ Step 2.

10. After h+1 is assigned to h, repeat the step 2.7 sequence until h>N, thereby obtaining the local optimal solution of the N particles in the Lth generation particle swarm and the fitness thereof, the global optimal solution R of the Lth generation particle swarm L and the fitness thereof; Step 2.

11. After L+1 is assigned to L, it is determined whether L>L Max is true, if true, step 2.12 is executed, otherwise, the procedure returns to step 2.4 for sequential execution; Step 2.

12. Output the Lth Max global optimal solution of the particle swarm and the global optimal solution as the optimal parameters of the SVM model, so that the SVM model with the optimal parameters realizes the working condition recognition of the decomposition furnace temperature.

2. An electronic device comprising a memory and a processor, characterized in that The memory is used to store a program supporting the processor to execute the decomposition furnace temperature working condition division and identification method of claim 1, and the processor is configured to execute the program stored in the memory.

3. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is run by the processor to execute the steps of the decomposition furnace temperature working condition division and identification method of claim 1.

Images