Coke oven operation state identification method

Abstract

The invention relates to a coke oven operation state identification method, which comprises the following steps: carrying out preprocessing and feature extraction on collected coke oven production process data to obtain feature representation suitable for model input; a self-organizing incremental neural network SOINN and a Gaussian mixture model GMM are utilized to construct a neural network model which automatically learns and adapts to data characteristics, the model receives feature representation as input, and training is performed by utilizing topological learning of the self-organizing incremental neural network SOINN and density modeling capability of the Gaussian mixture model GMM. Therefore, the operation state of the coke oven is classified and identified; further optimizing a result output by the model, adjusting model parameters by adopting a gradient descent method, evaluating and improving the performance of the model by utilizing technologies such as cross validation and the like, and carrying out judgment and fault diagnosis on the operation condition of the coke oven; according to the method, complex changes in coke oven operation can be more effectively identified and coped with, so that comprehensive and accurate monitoring and diagnosis of the operation state of the coke oven are realized.

Description

Technical Field

[0001] The invention relates to the field of coking technology in the field of coal chemical industry, and in particular to a method for identifying the operating status of a coke oven. Background Art

[0002] As one of the core processes in the coal chemical industry, coking production has extremely high performance requirements for coke oven equipment. However, coke ovens often face various faults and problems in long-term operation, such as insufficient suction, abnormal temperature, gas leakage, etc. These problems not only affect the production efficiency and stability of coke ovens, but may also lead to serious safety accidents, causing huge economic losses and safety hazards to corporate production.

[0003] Traditional coke oven fault diagnosis methods mainly rely on manual inspections and monitoring of single parameters, which have many problems, such as low efficiency of manual inspections, difficulty in analyzing monitoring data, and inability to achieve real-time monitoring. Although many algorithms have been applied in fault diagnosis, a single algorithm often cannot handle the complex process environment and variable data characteristics of coke ovens. For example, although the traditional neural network method performs well in pattern recognition, it lacks the ability to deeply model data distribution; and although the Gaussian mixture model is good at data density estimation, it has deficiencies in dynamic adjustment and topological learning. Therefore, a single algorithm is difficult to meet the high requirements of coke oven fault diagnosis for real-time and accuracy. In addition, due to the harsh operating environment of coke ovens and the complex equipment structure, fault diagnosis is often difficult and delayed, affecting the timeliness and accuracy of fault handling. Summary of the invention

[0004] The present invention provides a method for identifying the operating status of a coke oven. By combining a self-organizing incremental neural network and a Gaussian mixture model with the actual process flow of the coke oven, real-time monitoring and fault diagnosis of the operating status of the coke oven can be achieved, thereby effectively improving the stability and efficiency of the production process.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for identifying the operating status of a coke oven comprises the following steps:

[0007] S1. Preprocessing and feature extraction of the collected coke oven production process data to obtain feature representation suitable for model input;

[0008] S2. A neural network model that automatically learns and adapts to data characteristics is constructed using the self-organizing incremental neural network SOINN and the Gaussian mixture model GMM. The model accepts feature representation as input and is trained using the topological learning of the self-organizing incremental neural network SOINN and the density modeling capability of the Gaussian mixture model GMM, so as to classify and identify the operating status of the coke oven. The model output results are further optimized, and the model parameters are adjusted using the gradient descent method. At the same time, the model performance is evaluated and improved using cross-validation and other techniques, so as to judge the operating status of the coke oven and diagnose faults.

[0009] Furthermore, the feature extraction in step S1 is to extract effective features that can reflect the operating status of the coke oven from the original data, including temperature, pressure, and flow.

[0010] Furthermore, the step S2 specifically includes the following steps:

[0011] S2.1. Use Gaussian mixture model (GMM) to perform density estimation and probability modeling on the extracted features, reveal the distribution law and potential structure of the data, and combine the key parameters in the coke oven process to capture and analyze the important changes in the production process in real time;

[0012] S2.2. After preprocessing and feature extraction, a training sample set is constructed based on the obtained data. The sample set includes the normal operating state and various fault states of the coke oven. A self-organizing neural network model is constructed through the self-organizing incremental neural network SOINN algorithm. The model is trained and the topological structure is learned using the sample data, so that the network structure can be adaptively adjusted to better represent the data feature space.

[0013] S2.3. Based on the constructed self-organizing incremental neural network SOINN model, the Gaussian mixture model GMM is used to perform density estimation and probability modeling on each node, and the similarity and connection relationship between nodes are quantified through the parameter information of the Gaussian mixture model GMM model to improve the topological structure of the SOINN network;

[0014] S2.4, input the preprocessed and feature-extracted data into the constructed SOINN-GMM model for training and learning, adopt a supervised learning method with a penalty loss function, and use cross-validation technology to adjust and optimize the model parameters to obtain the coke oven operation status recognition model with the best performance;

[0015] S2.5, the data after preprocessing and feature extraction are input into the trained SOINN-GMM model, and the recognition result of the coke oven operation status is obtained through the learning and training of the model;

[0016] S2.6, according to the results of the SOINN-GMM model output in step S2.5, the operating status information of each coke oven partition is counted, and the coke oven status is divided into normal and abnormal categories according to the preset threshold value. By analyzing the operating status of each coke oven partition, the existing abnormal conditions and fault types are identified, and combined with the actual process flow of the coke oven, an accurate judgment basis for the current process state is provided, and real-time feedback is given to the coke oven control system to dynamically adjust the production parameters;

[0017] S2.7. Upon receiving the output alarm or warning information, the staff shall immediately take appropriate measures to deal with the current situation. The corresponding measures include dispatching maintenance personnel to conduct on-site maintenance, starting backup equipment to ensure that production is not affected, or taking emergency shutdown measures to prevent possible accidents. After solving the problem, the system automatically evaluates the measures taken and updates the system database in a timely manner.

[0018] Furthermore, the step S2.1 uses a Gaussian mixture model GMM to perform density estimation and probability modeling on the extracted features, including calculation of Gaussian membership and evolved Gaussian mixture model in the SOINN-GMM model:

[0019]

[0020] Among them, μ i (t) represents the Gaussian membership of the i-th node, A N represents the number of neighboring nodes, d i (t) and d j (t) represent the Euclidean distances between the input sample X(t) and the i-th node and the j-th node at time t, respectively. In the Gaussian membership-based self-organizing incremental neural network, these distances are used to calculate the Gaussian membership value μ i (t), thereby determining which node the input sample X(t) belongs to;

[0021] Gaussian membership indicates the degree of a node's belonging to an input sample. Gaussian membership is used to identify nodes that have distinguishing features from other data points and is used for anomaly detection. A node's Gaussian membership lower than a set value indicates that it is inconsistent with the distribution pattern of the input data, representing anomalies or outliers. The denominators in formula (1) are the sum of the reciprocals of the distances between a node and all neighboring nodes and the sum of the distances between a node and its neighboring nodes multiplied by the reciprocals of the sum of the Gaussian memberships of the node. The influence of neighboring nodes on a node and the distance between a node and the input sample are analyzed to describe the distribution of nodes in the feature space.

[0022] Furthermore, the self-organizing incremental neural network SOINN algorithm formula is as follows:

[0023]

[0024] Among them, S i Indicates the node closest to sample X, i.e. the winning node, B N is the set of nodes in the current network, W i is the weight vector of node c, ||·|| represents the Euclidean norm of the vector;

[0025] Formula (2) is used to calculate the Euclidean distance between the input sample X and the nodes in the network, so as to find the nearest node as the winning node. By continuously updating the node weights and topological structure in the network, the topological learning of the data feature space is realized.

[0026] Furthermore, the weight vector is defined as the weighted mean W of the samples X(t) belonging to the i-th node i (t), Gaussian membership μ i (t) represents the weight, and the calculation formula is as follows:

[0027]

[0028] The data parameters of the operating characteristics of the internal reaction system of the coke oven are dynamically changing. In order to adapt the model to the changing data distribution and sample characteristics, the weight vector is dynamically updated. The update formula is as follows:

[0029] W i (t+1)=W i (t)+ΔW i (t+1) (4)

[0030]

[0031] Among them, W i (t+1) is defined as the weighted mean of the sample X(t+1) belonging to the i-th node, μ i (t) represents the Gaussian membership value of the i-th node.

[0032] Furthermore, the calculation formula of the penalty loss function in step S2.5 is as follows:

[0033] L penalty =L supervised +λ· L topology (6)

[0034] Among them, L penalty represents the loss function after adding the penalty term; L supervised represents the loss function of supervised learning, which is used to measure the prediction error of the SOINN-GMM model on the training set; L topologyrepresents the loss function of topological learning, which is used to measure the fitting degree of the SOINN-GMM model on the topological structure; λ represents the penalty coefficient, which is used to balance the importance between supervised learning and topological learning; the larger the λ, the higher the importance of topological learning, and vice versa;

[0035] L supervised It is expressed by the cross entropy loss function or the mean square error loss function, and its calculation formula is as follows:

[0036]

[0037] Where N represents the number of samples, and They represent the true label and model prediction value of the i-th sample, respectively, and Loss(·) represents the loss function;

[0038] L topology Represents the loss function of topological learning, which is used to measure the degree of fit of the model in the topological structure. The calculation formula is as follows:

[0039]

[0040] Among them, edges represents the edge set in the network, d ij represents the distance between node i and node j, d max Indicates the maximum distance.

[0041] Furthermore, the step S2.6 includes the following process to process the model output result:

[0042] Classification judgment: According to the output results of the SOINN-GMM model, the operating status of the coke oven is divided into different categories or states;

[0043] Fault diagnosis: According to the output results of the SOINN-GMM model, the operating status of the coke oven is diagnosed;

[0044] Real-time monitoring: The output of the SOINN-GMM model is used to monitor the operation status of the coke oven in real time;

[0045] Decision support: Provide decision support for the operation of coke ovens based on the output results of the SOINN-GMM model.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention combines the self-organizing incremental neural network SOINN and the Gaussian mixture model GMM for data processing. The combination of the two not only overcomes the limitations of a single algorithm, but also can achieve comprehensive monitoring and accurate diagnosis of the complex process status of the coke oven. The feature representation obtained by observation processing is input into the neural network model. The obtained model can accurately classify and identify the operating status of the coke oven, and further optimize the diagnosis results, including: using the self-organizing incremental neural network SOINN to perform topological learning of the data feature space, which well solves the shortcomings of traditional methods in dealing with complex data distribution and dynamically changing coke oven operating status identification problems. The present invention strongly supports the coke oven production Real-time monitoring and fault diagnosis of the production process are of great significance for helping production enterprises improve the production efficiency and equipment stability of coke ovens. The present invention not only innovates in the design of algorithm models, but also breaks through the limitations of traditional methods in the combination of process and algorithm. By integrating real-time process data with advanced algorithm models, the system can more effectively identify and respond to complex changes in coke oven operation, thereby realizing comprehensive and accurate monitoring and diagnosis of the coke oven operation status; realizing real-time monitoring and multi-parameter comprehensive analysis of the coke oven operation status, accurately identifying potential fault hazards and issuing early warnings; and providing corresponding response measures, which greatly improves the accuracy and efficiency of fault diagnosis. The system can not only help enterprises to timely discover and deal with various fault problems in the operation of coke ovens, reduce the risk of production interruptions and safety accidents, but also improve production efficiency and equipment utilization, saving a lot of manpower and material costs for enterprises, and has great economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of the method described in the present invention. DETAILED DESCRIPTION

[0049] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings:

[0050] See Figure 1 , is a schematic diagram of the structure of the present invention. A method for identifying the operating state of a coke oven of the present invention processes the coke oven production process data in advance, inputs the feature representation obtained by the observation processing into the neural network model, and obtains the model result to identify and classify the operating state of the coke oven, and finally realizes accurate judgment and fault diagnosis of the operating state of the coke oven, which specifically includes the following steps:

[0051] S1. Preprocess and extract features of the collected coke oven production process data. The preprocessing includes steps such as data cleaning, denoising and normalization to ensure the quality and availability of the data, and obtain feature representation suitable for model input. Feature extraction is to extract effective features that can reflect the operating status of the coke oven from the original data, including indicators such as temperature, pressure, and flow.

[0052] S2. A neural network model that can automatically learn and adapt to data characteristics is constructed using the self-organizing incremental neural network SOINN and the Gaussian mixture model GMM. The model accepts feature representation as input and is trained using the topological learning of the self-organizing incremental neural network SOINN and the density modeling capability of the Gaussian mixture model GMM, so as to classify and identify the operating status of the coke oven. These algorithms are closely integrated with the coke oven process parameters, especially the real-time data input of key process parameters such as temperature, pressure, and airflow, and can dynamically adjust the learning process of the model so that it can not only adapt to data changes in the production process, but also effectively identify abnormal conditions in the process, thereby providing more accurate operating status monitoring and fault diagnosis. The results of the model output are further optimized, and the model parameters are adjusted using the gradient descent method. At the same time, the model performance is evaluated and improved using cross-validation and other technologies to achieve accurate judgment of the operating status of the coke oven and fault diagnosis.

[0053] The specific steps include:

[0054] S2.1. Use GMM to perform density estimation and probability modeling on the extracted features to reveal the distribution law and potential structure of the data;

[0055] The calculation formulas of Gaussian membership and evolved Gaussian mixture model in Gm-SOINN are as follows:

[0056]

[0057] Among them, μ i (t) represents the Gaussian membership of the i-th node, A N represents the number of neighboring nodes, d i (t) and d j (t) represent the Euclidean distances between the input sample X(t) and the i-th node and the j-th node at time t, respectively. In the Gaussian membership-based self-organizing incremental neural network, these distances are used to calculate the Gaussian membership value μ i (t), thereby determining which node the input sample X(t) belongs to;

[0058] Formula (9) is used to calculate the Gaussian membership of a node. Gaussian membership is used to identify nodes that have distinguishing features from other data points, and is therefore used for anomaly detection. A node whose Gaussian membership is lower than a set value indicates that it is inconsistent with the distribution pattern of the input data, representing anomalies or outliers. The denominators in formula (9) are the sum of the reciprocals of the distances between the node and all neighboring nodes and the sum of the distances between the node and its neighboring nodes multiplied by the reciprocals of the sum of the Gaussian memberships of the node. This takes into account the influence of neighboring nodes on the node and the distance between the node and the input sample, thereby more accurately describing the distribution of the node in the feature space.

[0059] S2.2. After preprocessing and feature extraction, a training sample set is constructed based on the obtained data; the sample set includes the normal operation state of the coke oven and various fault states. A self-organizing neural network model is constructed by the self-organizing incremental neural network SOINN algorithm. The model is trained and the topological structure is learned using the sample data, so that the network structure can be adaptively adjusted to better represent the data feature space. The formula of the self-organizing incremental neural network SOINN algorithm is as follows:

[0060]

[0061] Among them, S i Indicates the node closest to sample X, i.e. the winning node, B N is the set of nodes in the current network, W i is the weight vector of node c, ||·|| represents the Euclidean norm of the vector;

[0062] Formula (10) is used to calculate the Euclidean distance between the input sample X and the nodes in the network, so as to find the nearest node as the winning node. By continuously updating the node weights and topological structure in the network, the topological learning of the data feature space is realized, so that the network can better represent the distribution and characteristics of the data.

[0063] The weight vector is defined as the weighted mean W of the sample X(t) belonging to the i-th node i (t), Gaussian membership μ i (t) represents the weight, and the calculation formula is as follows:

[0064]

[0065] Since the data parameters of the operating characteristics of the internal reaction system of the coke oven are dynamically changing, in order to adapt the model to the changing data distribution and sample characteristics, it is important to dynamically update the weight vector. Specifically, the update formula is as follows:

[0066] W i (t+1)=W i (t)+ΔWi (t+1) (12)

[0067]

[0068] Among them, W i (t+1) is defined as the weighted mean of the sample X(t+1) belonging to the i-th node, μ i (t) represents the Gaussian membership value of the i-th node.

[0069] S2.3. Based on the constructed self-organizing incremental neural network SOINN model, the Gaussian mixture model GMM is used to perform density estimation and probability modeling for each node. In actual application cases, when the operating conditions of the coke oven change, the method of using the self-organizing incremental neural network SOINN or the Gaussian mixture model GMM alone cannot adjust the model in time to correctly identify the abnormal state. However, when the self-organizing incremental neural network SOINN and GMM are combined, the system can simultaneously use the real-time topological learning of SOINN and the precise density modeling of the Gaussian mixture model GMM to respond to abnormal changes in the coke oven state. The final diagnosis result is not only more accurate than when the two algorithms are used alone, but also significantly reduces the misjudgment and missed judgment. The superposition of this effect makes the combined algorithm perform much better than the simple addition of the two algorithms in processing complex process environments. The similarity and connection relationship between nodes are quantified through the parameter information of the GMM model, and the topological structure of the self-organizing incremental neural network SOINN network is further improved. In this way, the distribution and category information of the data can be more accurately described, providing a more reliable basis for subsequent state recognition.

[0070] S2.4. The data after preprocessing and feature extraction are input into the constructed SOINN-GMM model for training and learning. During the learning process, a supervised learning method with a penalty loss function is used to ensure that the model can fully consider the distribution characteristics and structural information of the data, thereby improving the generalization ability and recognition accuracy of the model. At the same time, the model parameters are adjusted and optimized using cross-validation technology to obtain the coke oven operation status recognition model with the best performance. In this way, real-time monitoring and fault diagnosis of the coke oven operation status can be achieved, and production efficiency and equipment stability can be improved;

[0071] The calculation formula of the penalty loss function is as follows:

[0072] L penalty =L supervised +λ· L topology (14)

[0073] Among them, L penalty represents the loss function after adding the penalty term; L supervisedrepresents the loss function of supervised learning, which is used to measure the prediction error of the SOINN-GMM model on the training set; L topology represents the loss function of topological learning, which is used to measure the fitting degree of the SOINN-GMM model on the topological structure; λ represents the penalty coefficient, which is used to balance the importance between supervised learning and topological learning; the larger the λ, the higher the importance of topological learning, and vice versa;

[0074] L supervised It is expressed by the cross entropy loss function or the mean square error loss function, and its calculation formula is as follows:

[0075]

[0076] Where N represents the number of samples, and They represent the true label and model prediction value of the i-th sample, respectively, and Loss(·) represents the loss function;

[0077] L topology Represents the loss function of topological learning, which is used to measure the degree of fit of the model in the topological structure. The calculation formula is as follows:

[0078]

[0079] Among them, edges represents the edge set in the network, d ij represents the distance between node i and node j, d max Indicates the maximum distance.

[0080] S2.5, input the preprocessed and feature-extracted data into the trained SOINN-GMM model, and obtain the recognition result of the coke oven operation status through model learning and training;

[0081] S2.6, perform further subsequent operations according to the results output by the SOINN-GMM model in step S2.5;

[0082] Based on the results output by the SOINN-GMM model, the operating status information of each coke oven partition is counted, and the coke oven status is divided into normal and abnormal categories according to the preset threshold. In actual cases, by combining the self-organizing incremental neural network SOINN and the Gaussian mixture model GMM, the system can accurately identify and distinguish the operating status of each coke oven partition. In the partition with drastic temperature fluctuations, the density modeling of the Gaussian mixture model GMM alone may lead to misjudgment, while the use of SOINN alone may not fully consider the small fluctuations in temperature changes and lead to misjudgment. However, the combined algorithm can combine the advantages of these two aspects and show higher sensitivity and accuracy when dealing with drastic changes. The superposition of this effect significantly improves the diagnostic ability of the entire system under different working conditions, proving the superiority of the combined algorithm under complex process conditions. By analyzing the operating status of each coke oven partition, possible abnormal conditions and fault types can be identified. Specifically, the following methods can be used to process the results of the model output:

[0083] Classification judgment: According to the output of the model, the operation status of the coke oven is divided into different categories or states. For example, the operation status of the coke oven is divided into different categories such as normal, abnormal, and fault, so as to facilitate subsequent processing and analysis;

[0084] Fault diagnosis: According to the output of the model, the operation status of the coke oven is diagnosed. By analyzing the output of the model, it is determined whether there is any abnormality or fault in the coke oven, and the type and cause of the fault are further determined.

[0085] Real-time monitoring: Use the model output results to monitor the operation status of the coke oven in real time. By continuously inputting new data and identifying and analyzing it, the operation status of the coke oven can be monitored in real time, and potential problems and abnormal situations can be discovered and handled in a timely manner;

[0086] Decision support: Based on the output of the model, provide decision support for the operation of the coke oven. By identifying and analyzing the operating status of the coke oven, provide reference opinions and suggestions for the operation and management of the coke oven, and help improve the operating efficiency and stability of the coke oven.

[0087] S2.7. When the staff receives the output alarm or warning information, they immediately take corresponding measures to deal with the current situation, including dispatching maintenance personnel to conduct on-site maintenance, starting backup equipment to ensure that production is not affected, or taking emergency shutdown measures to prevent possible accidents. After solving the problem, the system automatically evaluates the measures taken and updates the system database in a timely manner. Ultimately, the application of the combined algorithm not only provides more accurate real-time monitoring and fault diagnosis, but also significantly improves the decision-making support capabilities of the entire system. This superposition effect far exceeds the simple addition of the two algorithms when used separately, ensuring the stability and safety of the coke oven operation and further improving production efficiency and economic benefits.

[0088] The present invention combines a self-organizing incremental neural network SOINN and a Gaussian mixture model GMM for data processing, and inputs the feature representation obtained by observation processing into the neural network model. The obtained model can accurately classify and identify the operating status of the coke oven, and further optimize the diagnosis result, including: using the self-organizing incremental neural network SOINN to perform topological learning of the data feature space, which well solves the shortcomings of traditional methods in dealing with complex data distribution and dynamically changing coke oven operating status identification problems. The present invention strongly supports real-time monitoring and fault diagnosis of the coke oven production process, and is of great significance for helping production enterprises improve coke oven production efficiency and equipment stability.

[0089] The above embodiments are implemented based on the technical solution of the present invention, and detailed implementation methods and specific operation processes are given, but the protection scope of the present invention is not limited to the above embodiments. The methods used in the above embodiments are conventional methods unless otherwise specified.

Claims

1. A method for identifying the operating status of a coke oven, characterized in that: The steps include: S1. Preprocessing and feature extraction of the collected coke oven production process data to obtain feature representation suitable for model input; S2. A neural network model that automatically learns and adapts to data characteristics is constructed using the self-organizing incremental neural network SOINN and the Gaussian mixture model GMM. The model accepts feature representation as input and is trained using the topological learning of the self-organizing incremental neural network SOINN and the density modeling capability of the Gaussian mixture model GMM, so as to classify and identify the operating status of the coke oven. The model output results are further optimized, and the model parameters are adjusted using the gradient descent method. At the same time, the model performance is evaluated and improved using cross-validation and other techniques, so as to judge the operating status of the coke oven and diagnose faults.

2. A method for identifying the operating status of a coke oven according to claim 1, characterized in that: The feature extraction in step S1 is to extract effective features that can reflect the operating status of the coke oven from the original data, including temperature, pressure, and flow.

3. A method for identifying the operating status of a coke oven according to claim 1, characterized in that: The step S2 specifically includes the following steps: S2.

1. Use Gaussian mixture model (GMM) to perform density estimation and probability modeling on the extracted features, reveal the distribution law and potential structure of the data, and combine the key parameters in the coke oven process to capture and analyze the important changes in the production process in real time; S2.

2. After preprocessing and feature extraction, a training sample set is constructed based on the obtained data. The sample set includes the normal operating state and various fault states of the coke oven. A self-organizing neural network model is constructed through the self-organizing incremental neural network SOINN algorithm. The model is trained and the topological structure is learned using the sample data, so that the network structure can be adaptively adjusted to better represent the data feature space. S2.

3. Based on the constructed self-organizing incremental neural network SOINN model, the Gaussian mixture model GMM is used to perform density estimation and probability modeling on each node, and the similarity and connection relationship between nodes are quantified through the parameter information of the Gaussian mixture model GMM model to improve the topological structure of the SOINN network; S2.4, input the preprocessed and feature-extracted data into the constructed SOINN-GMM model for training and learning, adopt a supervised learning method with a penalty loss function, and use cross-validation technology to adjust and optimize the model parameters to obtain the coke oven operation status recognition model with the best performance; S2.5, the data after preprocessing and feature extraction are input into the trained SOINN-GMM model, and the recognition result of the coke oven operation status is obtained through the learning and training of the model; S2.6, according to the results of the SOINN-GMM model output in step S2.5, the operating status information of each coke oven partition is counted, and the coke oven status is divided into normal and abnormal categories according to the preset threshold value. By analyzing the operating status of each coke oven partition, the existing abnormal conditions and fault types are identified, and combined with the actual process flow of the coke oven, an accurate judgment basis for the current process state is provided, and real-time feedback is given to the coke oven control system to dynamically adjust the production parameters; S2.

7. Upon receiving the output alarm or warning information, the staff shall immediately take appropriate measures to deal with the current situation. The corresponding measures include dispatching maintenance personnel to conduct on-site maintenance, starting backup equipment to ensure that production is not affected, or taking emergency shutdown measures to prevent possible accidents. After solving the problem, the system automatically evaluates the measures taken and updates the system database in a timely manner.

4. A method for identifying the operating status of a coke oven according to claim 2, characterized in that: The step S2.1 uses the Gaussian mixture model GMM to perform density estimation and probability modeling on the extracted features, including the calculation of Gaussian membership and evolved Gaussian mixture model in the SOINN-GMM model: Among them, μ i (t) represents the Gaussian membership of the i-th node, A N represents the number of neighboring nodes, d i (t) and d j (t) represent the Euclidean distances between the input sample X(t) and the i-th node and the j-th node at time t, respectively. In the Gaussian membership-based self-organizing incremental neural network, these distances are used to calculate the Gaussian membership value μ i (t), thereby determining which node the input sample X(t) belongs to; Gaussian membership indicates the degree of a node's belonging to an input sample. Gaussian membership is used to identify nodes that have distinguishing features from other data points and is used for anomaly detection. A node's Gaussian membership lower than a set value indicates that it is inconsistent with the distribution pattern of the input data, representing anomalies or outliers. The denominators in formula (1) are the sum of the reciprocals of the distances between a node and all neighboring nodes and the sum of the distances between a node and its neighboring nodes multiplied by the reciprocals of the sum of the Gaussian memberships of the node. The influence of neighboring nodes on a node and the distance between a node and the input sample are analyzed to describe the distribution of nodes in the feature space.

5. A method for identifying the operating status of a coke oven according to claim 2, characterized in that: The self-organizing incremental neural network SOINN algorithm formula is as follows: Among them, S i Indicates the node closest to sample X, i.e. the winning node, B N is the set of nodes in the current network, W i is the weight vector of node c, ||·|| represents the Euclidean norm of the vector; Formula (2) is used to calculate the Euclidean distance between the input sample X and the nodes in the network, so as to find the nearest node as the winning node. By continuously updating the node weights and topological structure in the network, the topological learning of the data feature space is realized.

6. A method for identifying the operating status of a coke oven according to claim 5, characterized in that: The weight vector is defined as the weighted mean W of the sample X(t) belonging to the i-th node i (t), Gaussian membership μ i (t) indicates that the weight calculation formula is as follows: The data parameters of the operating characteristics of the internal reaction system of the coke oven are dynamically changing. In order to adapt the model to the changing data distribution and sample characteristics, the weight vector is dynamically updated. The update formula is as follows: W i (t+1)=W i (t)+ΔW i (t+1) (4) Among them, W i (t+1) is defined as the weighted mean of the sample X(t+1) belonging to the i-th node, μ i (t) represents the Gaussian membership value of the i-th node.

7. A method for identifying the operating status of a coke oven according to claim 2, characterized in that: The calculation formula of the penalty loss function in step S2.4 is as follows: L penalty =L supervised +λ·L topology (6) Among them, L penalty represents the loss function after adding the penalty term; L supervised represents the loss function of supervised learning, which is used to measure the prediction error of the SOINN-GMM model on the training set; L topology represents the loss function of topological learning, which is used to measure the fitting degree of the SOINN-GMM model on the topological structure; λ represents the penalty coefficient, which is used to balance the importance between supervised learning and topological learning; the larger the λ, the higher the importance of topological learning, and vice versa; L supervised It is expressed by the cross entropy loss function or the mean square error loss function, and its calculation formula is as follows: Where N represents the number of samples, and They represent the true label and model prediction value of the i-th sample, respectively, and Loss(·) represents the loss function; L topology Represents the loss function of topological learning, which is used to measure the degree of fit of the model in the topological structure. The calculation formula is as follows: Among them, edges represents the edge set in the network, d ij represents the distance between node i and node j, d max Indicates the maximum distance.

8. A method for identifying the operation status of a coke oven according to claim 2, characterized in that: The step S2.6 includes the following steps to process the model output results: Classification judgment: According to the output results of the SOINN-GMM model, the operating status of the coke oven is divided into different categories or states; Fault diagnosis: According to the output results of the SOINN-GMM model, the operating status of the coke oven is diagnosed; Real-time monitoring: The output of the SOINN-GMM model is used to monitor the operation status of the coke oven in real time; Decision support: Provide decision support for the operation of coke ovens based on the output results of the SOINN-GMM model.

Images