Data-driven smelting control optimization method

Abstract

The invention relates to a data-driven smelting control optimization method, and belongs to the technical field of smelting control. Comprising the steps that working condition classification is carried out through historical operation data of a current smelting furnace by adopting clustering analysis, association rule mining is carried out in each cluster, and a parameter rule base is established; obtaining the working condition category of the current smelting furnace temperature through the parameter rule base according to the smelting furnace operation production data; according to the current furnace temperature characteristic data and the working condition category, the change of the furnace temperature is predicted in real time; according to the current working condition type and the predicted furnace temperature data, a control parameter combination with the optimal energy consumption is selected from the parameter data set; and monitoring the state of the smelting furnace in real time, predicting the furnace temperature, adjusting the control parameter combination of smelting operation, and performing early warning through the furnace temperature fluctuation degree. Furnace temperature prediction is carried out through working condition recognition, then control parameter combinations are screened according to prediction results and working condition categories, and therefore the smelting process is regulated and controlled to achieve energy-saving optimization.

Description

Technical Field

[0001] The present invention belongs to the technical field of smelting control, and in particular relates to a data-driven smelting control optimization method. Background Art

[0002] With the continuous advancement of automation and intelligentization of industrial production, the optimization of the smelting process has become a key area to improve energy efficiency and reduce production costs. The traditional smelting process usually relies on experience and manual adjustment, and the control system's adjustment of the furnace temperature mostly relies on static or simple rules, which often leads to energy waste and insufficient control accuracy.

[0003] At present, analysis and prediction based on historical data have become an effective means to improve the efficiency of the smelting process. However, the existing technology lacks efficient methods for subdividing the operating condition categories and characterizing the state. The model for predicting furnace temperature only relies on historical data and cannot accurately reflect the complex relationship between all variables. At the same time, when screening the optimal control parameters, there is a lack of in-depth exploration of the complex relationship between the operating condition category and the control parameters, resulting in insufficient consideration of the diversity of all operating conditions, unsatisfactory control effects, and thus affecting energy-saving effects. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a data-driven smelting control optimization method, which predicts the furnace temperature through operating condition identification, and then screens the control parameter combination according to the prediction result and the operating condition category, so as to regulate the smelting process and achieve energy-saving optimization.

[0005] The purpose of the present invention can be achieved by the following technical solutions: A data-driven smelting control optimization method comprises the following steps: Establish a parameter rule base: Use cluster analysis to classify the operating conditions based on the historical operating data of the current furnace, and mine association rules within each cluster to establish a parameter rule base; Working condition acquisition: according to the production data of the furnace operation, the working condition category of the current furnace temperature is obtained through the parameter rule library; Furnace temperature trend prediction: Based on the current furnace temperature characteristic data and working condition category, the change of furnace temperature is predicted in real time; Obtain control parameters: Select the control parameter combination with the best energy consumption from the parameter data set according to the current operating condition category and predicted furnace temperature data; Abnormal operation monitoring: Real-time monitoring of the furnace status and furnace temperature prediction to adjust the control parameter combination of smelting operation, while issuing early warnings based on the degree of furnace temperature fluctuation.

[0006] Furthermore, the historical operation data includes sensor timing data, process parameters and equipment status logs, and the production data includes real-time furnace operation data and process parameters, wherein the process parameters include the heating time, furnace pressure, air-fuel ratio and heating power set for process production.

[0007] Furthermore, the establishment of the parameter rule base comprises the following steps: Data preprocessing: binning continuous control parameters to discretize them, remove noise and anomalies in time series data, and enhance the data; Working condition identification: By mining the characteristic data related to the furnace temperature change and performing cluster analysis, the working condition category of the current furnace temperature change can be identified; Rule mining: Construct an association rule mining algorithm in each working condition category, set the confidence level and minimum support level, and generate a parameter rule base for association rules with the working condition category as the result.

[0008] Furthermore, the operating condition identification comprises the following steps: Feature selection: The data is divided into grids and mutual information is calculated. The MIC value is obtained through different grid segmentation methods. The top 30 features with the strongest relationship with the target variable are selected to construct the furnace temperature feature data; Density clustering: Dynamically select the minimum number of samples according to the sampling period of the sliding window, and use the selected features and the minimum number of samples to perform clustering using the HDBSCAN algorithm; Working condition definition: Use UMAP to reduce the dimension of the clustered data and visualize the data points after dimension reduction. Use different colors to represent different clusters for each data point. Each cluster represents a furnace temperature working condition and defines the working condition category.

[0009] Furthermore, the rule mining comprises the following steps: Construct a transaction database: convert the data in each working condition category into a transaction format, with each sample represented as a group item and the features of each sample as items; Mining association rules: Use the association rule mining algorithm to mine the data of each working condition category, set the minimum support and minimum confidence as the threshold for rule screening, and output the association rules that meet the conditions; Update association rules: According to the operation characteristics of smelting, set the sliding window to add new operation data, use the incremental learning method, update the association rules, and adjust the support and confidence.

[0010] Furthermore, the furnace temperature trend prediction includes the following steps: Model construction: Long short-term memory neural network is used as the furnace temperature prediction model, including input layer, LSTM layer, fully connected layer and output layer; Model training: The input variables are converted into time series data through preprocessing and divided into training set, validation set and test set with the corresponding furnace temperature values. The training data is input into the LSTM model, and the network is trained through back propagation and gradient descent algorithms. Real-time prediction: Obtain furnace temperature characteristic data and furnace temperature operating condition category in real time, input the acquired data into the trained furnace temperature prediction model, and perform real-time prediction of furnace temperature.

[0011] Furthermore, the input variables of the input layer include furnace temperature characteristic data and operating condition categories, and the output layer includes a single node representing the predicted furnace temperature.

[0012] Furthermore, the obtaining of control parameters comprises the following steps: Data merging: Match the control parameter combinations corresponding to all operating condition category data from historical operation data to establish parameter data sets; Energy consumption calculation: Calculate the comprehensive energy consumption data of each set of control parameter combinations during operation, including heating energy consumption, cooling energy consumption and other energy consumption, and store the energy consumption data together with its corresponding operating condition category; Parameter selection: Calculate similar operating conditions based on the current operating condition, and define a group of related operating conditions by predicting the difference between the furnace temperature and the target furnace temperature, and select the control parameter combination with the lowest energy consumption in this group of operating conditions.

[0013] Furthermore, in the abnormal operation monitoring, determining the stability of the furnace temperature change includes the following steps: Preset time window and monitor furnace temperature data within the time window; Calculate the deviation between the mean furnace temperature and the target furnace temperature, and calculate the standard deviation of the furnace temperature data as the furnace temperature fluctuation value; Set judgment thresholds for mean deviation and fluctuation value at different levels. When both mean deviation and fluctuation value exceed the set judgment threshold, an early warning will be issued according to the exceeded range set for the level.

[0014] Furthermore, the method further comprises the steps of: Operation data update: Collect all the operation data of the current furnace, and mark the equipment control parameter combination in the stable state according to the judgment of furnace temperature fluctuation, update the parameter rule base with the collected data, and update the parameter data set with the marked control parameter combination.

[0015] The beneficial effects of the present invention are: The present invention realizes energy-saving optimization in the smelting process by adjusting the furnace temperature control parameters. First, cluster analysis is performed on historical data to establish operating condition categories characterized by furnace temperature. Then, association rules are mined in each operating condition category. The condition data type and numerical range are used as the state representation of the operating condition category. The operating condition category and furnace temperature characteristic data are then input into the prediction model to obtain the predicted furnace temperature, thereby improving the accuracy of furnace temperature prediction. Finally, similar operating condition categories that meet the conditions are screened according to the predicted furnace temperature and operating condition category. The optimal control parameters are obtained through energy consumption calculation, and the smelting process is continuously regulated to achieve energy-saving optimization.

[0016] In the process of obtaining the working condition, the present invention is implemented based on the established parameter rule base. The parameter rule base is constructed by clustering the relevant data of the furnace temperature change and then using the association rules to mine the associated data for each working condition as a condition. The inherent differences of the furnace temperature change are grasped more accurately through the working condition category and the working condition change is subdivided to achieve high-precision working condition division, establish basic conditions for furnace temperature prediction and adjustment of control parameters, and then select a group of working condition categories through the furnace temperature prediction result and the working condition category to obtain the control parameter combination, thereby realizing data-driven dynamic adaptation. This process mines the correlation between the data based on the big data of smelting, closely combines the furnace temperature state and related characteristics, and then uses the furnace temperature prediction to fine-tune the working condition category, and changes the control parameter combination to achieve the purpose of energy saving optimization while meeting the target furnace temperature. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0018] Figure 1 A schematic flow chart of a data-driven smelting control optimization method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0020] This embodiment provides a data-driven smelting control optimization method, such as Figure 1 As shown, the following steps are included: S1. Establishing a parameter rule base: Collect the historical operation data of the current furnace, classify the working conditions through cluster analysis after preprocessing, and mine the association rules in each cluster to establish a parameter rule base associated with the working condition categories, including the following steps: Among them, historical operation data includes sensor time series data, process parameters, and equipment status logs, etc. Sensor time series data refers to the state parameters monitored in real time by sensors installed during the smelting process, such as furnace temperature, gas flow, current, voltage, etc. These time series data reflect the dynamic characteristics of the smelting process; process parameters such as the heating time set by the process production, furnace pressure, air-fuel ratio, heating power, etc. are important factors affecting the smelting process; the equipment status log records the operating status of the equipment, such as equipment startup, shutdown, fault alarm and other information.

[0021] S11. Data preprocessing: The continuous control parameters are binned to discretize them, the noise and anomalies in the time series data are removed, and the data is enhanced.

[0022] In this embodiment, OPTICS adaptive binning is used for continuous control parameters, 30% window overlap is retained to ensure the effectiveness of association rules, STL decomposition is integrated to denoise working condition features, and data enhancement is achieved through SAE autoencoder.

[0023] It should be noted that the purpose of discretization is to convert continuous data into discrete intervals, so as to facilitate subsequent analysis. The specific method used is OPTICS adaptive binning, which can adapt to the distribution of data and maintain efficient association rule mining capabilities. OPTICS (Ordering Points To Identify the Clustering Structure) is a density-based clustering algorithm that can handle non-uniform distribution of data. STL (Seasonal and Trenddecomposition using Loess) decomposition is a decomposition method for time series data, which decomposes the original time series data into three parts: trend, seasonality, and residual. Through STL decomposition, the long-term trend and seasonal fluctuations in the smelting process can be separated, and the noise and abnormal data caused by these factors can be removed. For the data in the smelting process, the SAE (Sparse Autoencoder) autoencoder can perform nonlinear mapping on the input data, map the original data to a low-dimensional space through compression, and then decode and reconstruct it; during the training process, SAE learns the main features of the data, removes noise or outliers from the reconstructed data to achieve data denoising and enhancement, and through the autoencoder, it can compensate for missing values ​​and correct abnormal data, thereby enhancing the reliability and accuracy of the data.

[0024] S12, working condition identification: by mining the characteristic data related to the furnace temperature change and performing cluster analysis, the working condition category of the current furnace temperature change is identified, including the following steps: S121, feature selection: grid the data and calculate mutual information, obtain MIC values ​​through different grid segmentation methods, and select the top 30 features with the strongest relationship with the target variable to form furnace temperature feature data; It is understandable that MIC (Maximal Information Coefficient) is a statistic that measures the correlation between two variables. MIC aims to capture the nonlinear and complex relationship between variables. The calculation process of MIC selects appropriate grid segmentation, calculates mutual information, and maximizes the mutual information value to finally obtain a value that reflects the strength of the relationship between the two variables.

[0025] It should be noted that, in this embodiment, the purpose of classifying the working conditions is to better grasp the changes in furnace temperature so as to predict the furnace temperature. Therefore, the target variable selected when initially calculating the MIC value is the furnace temperature.

[0026] S122, density clustering: dynamically select the minimum number of samples according to the sampling period of the sliding window, and use the selected features and the minimum number of samples to perform clustering using the HDBSCAN algorithm.

[0027] It is understandable that HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm, and its core idea is to cluster data based on local density differences. Unlike traditional K-Means, HDBSCAN does not require the number of clusters to be specified in advance, and can handle clusters of any shape and noise points.

[0028] It should be noted that HDBSCAN determines whether each data point belongs to a cluster by calculating the local density, including core distance and reachable distance. First, the "connection" relationship between points is calculated through the reachable distance, and a minimum spanning tree is generated. The minimum spanning tree ensures that all points are connected through the shortest reachable distance; different cluster structures are gradually formed based on the descending order of reachable distance, and then a hierarchical clustering tree is constructed by merging and splitting clusters. The final cluster structure is determined by calculating the stability of clusters at different levels; finally, by traversing the hierarchical tree, the cluster splitting points are automatically selected based on the stability threshold to determine the range of each cluster.

[0029] S123, working condition definition: Use UMAP to reduce the dimension of the clustered data and visualize the data points after dimension reduction. Use different colors to represent different clusters for each data point. Each cluster represents a furnace temperature working condition and defines the working condition category.

[0030] Among them, UMAP (Uniform Manifold Approximation and Projection) is a dimensionality reduction technology that is widely used to map high-dimensional data into low-dimensional space, and is usually used for data visualization and feature learning. Its main advantage is that it can preserve the local structure of the data while preserving the global structure, so that the reduced-dimensional data has a better visualization effect in two-dimensional or three-dimensional space.

[0031] S13, rule mining: construct an association rule mining algorithm in each working condition category, set the confidence and minimum support, generate an association rule construction parameter rule base with the working condition category as the result, including the following steps: S131, constructing a transaction database: converting the data in each working condition category into a transaction format, wherein each sample is represented as a group item, and the feature of each sample is used as an item; It should be noted that the sample characteristics include at least one of the 30 characteristics related to furnace temperature changes, such as air-fuel ratio, cooling water flow, etc. By mining the association rules of each type of operating condition, the final conditional data type and value range are not the same, which is also the characteristic manifestation of different furnace temperature conditions.

[0032] S132. Mining association rules: Use association rule mining algorithms (such as Apriori, FP-Growth, Eclat, etc.) to mine the data of each working condition category (cluster), set the minimum support (Minimum Support) and minimum confidence (Minimum Confidence) as the thresholds for rule screening, and output association rules that meet the conditions.

[0033] In this embodiment, the minimum support is initially set to 0.15 and the minimum confidence is set to 0.8. They can be adjusted in subsequent applications to meet the needs of data mining. In addition, the lift can be set to 1.5 to limit the positive correlation between the premise and the result of the rule.

[0034] S133, updating association rules: according to the operation characteristics of smelting, setting a sliding window to add new operation data, using an incremental learning method, updating association rules, and adjusting support and confidence.

[0035] It is understandable that by monitoring the operating status of the furnace, continuously collecting status data, adding the collected data to the historical operating data through a sliding window, and using the incremental learning method to process the new data to complete the update of the association rules, the consumption of computing resources can be reduced and efficiency can be improved.

[0036] It should be noted that through historical operating data, the data related to furnace temperature changes are clustered to define the operating condition categories of furnace temperature changes, and then in each operating condition category, through association analysis, the parameters associated with the operating condition are mined from the historical operating data and a parameter rule library is established, with the associated parameters as the rule antecedent and the operating condition category as the rule consequence, so as to obtain the current furnace temperature operating condition category through real-time data in the production process. The mining of association rules is mainly to further associate the furnace temperature operating condition with related data, so as to more accurately grasp the inherent differences in furnace temperature changes and subdivide the operating condition changes, thereby improving the accuracy of furnace temperature prediction.

[0037] S2. Working condition acquisition: The working condition category of the current furnace temperature is obtained through the parameter rule library according to the real-time data of the furnace operation and the process parameters.

[0038] It should be noted that the parameter rule library contains association rules based on the operating condition category. When applied, the operating condition category is obtained by matching the antecedents (conditions) of the association rules with real-time production data. Furthermore, under the premise of each operating condition category, the data type composition and specific parameters of the antecedents are different. The production data includes real-time furnace operation data and process parameters. The real-time data is specifically sensor timing data. The process parameters include the heating time set by the process production, furnace pressure, air-fuel ratio, heating power, etc.

[0039] S3. Furnace temperature trend prediction: Establish a furnace temperature prediction model to make real-time predictions on furnace temperature changes based on current furnace temperature characteristic data and furnace temperature operating condition categories, including the following steps: S31. Model building: Long short-term memory neural network (LSTM) is used as the furnace temperature prediction model, including input layer, LSTM layer, fully connected layer and output layer; The input variables of the input layer include furnace temperature feature data and operating condition categories. The LSTM layer uses LSTM units to process time series data. One or more LSTM layers can be set according to the characteristics of the data. The number of units in the LSTM layer can be adjusted according to actual conditions. Generally, 64, 128 or 256 units are used. The fully connected layer (Dense layer) is set after the LSTM layer. Feature fusion and further learning are performed by adding one or more fully connected layers. The output layer is usually a single node, which represents the predicted furnace temperature.

[0040] S32, model training: convert the input variables into time series data through preprocessing and divide them into training set, validation set and test set with the corresponding furnace temperature values, input the training data into the LSTM model, and train the network through back propagation and gradient descent algorithm; S33, real-time prediction: real-time acquisition of furnace temperature characteristic data and furnace temperature operating condition category, input of the acquired data into a trained furnace temperature prediction model, and real-time prediction of the furnace temperature.

[0041] It should be noted that, in this embodiment, by adding the identified operating condition category of the current furnace temperature to the input of the model, the changing characteristics of the furnace temperature operating conditions are more finely reflected in the prediction model, thereby improving the accuracy of the furnace temperature prediction and providing a basis for subsequent energy-saving optimization and regulation.

[0042] S4, obtaining control parameters: selecting the control parameter combination with the best energy consumption from the parameter data set according to the current operating condition category and the predicted furnace temperature data, including the following steps: S41. Data merging: Match the control parameter combinations corresponding to all operating condition category data from the historical operating data to establish a parameter data set.

[0043] Based on the operating condition categories of the parameter rule library, the corresponding control parameter combinations are supplemented for each operating condition using time series, including heating power, cooling flow, furnace gas flow, furnace charge amount, etc., to build a mapping relationship between the operating condition and the control parameters, thereby establishing a complete data set.

[0044] S42, energy consumption calculation: calculate the comprehensive energy consumption data of each set of control parameter combinations during operation, including heating energy consumption, cooling energy consumption and other energy consumption, and store the energy consumption data together with its corresponding operating condition category.

[0045] It is understandable that for each set of control parameter combinations, the comprehensive energy consumption is calculated based on the corresponding operation time, furnace temperature change, external conditions, etc. Other energy consumption includes the impact of furnace gas flow, furnace charge amount, etc. on energy consumption.

[0046] S43, parameter selection: Calculate similar operating condition categories based on the current operating condition category, and define a group of related operating condition categories by predicting the difference between the furnace temperature and the target furnace temperature, and select the control parameter combination with the lowest energy consumption in the group of operating condition categories.

[0047] Specifically, similar operating condition categories are obtained by calculating the similarity of the operating conditions and screening them with a similarity threshold, and then the deviation range between the predicted furnace temperature and the target furnace temperature is defined, samples with excessive deviations are extracted from similar operating conditions, and the operating condition category with the lowest energy consumption is selected from the retained samples to obtain the control parameters.

[0048] It should be noted that the operating condition categories obtained by clustering the HDBSCAN algorithm set the metric distance during the operation of HDBSCAN. Clusters with smaller distances are close in the feature space. Therefore, a series of operating condition categories can be searched by calculating the center points between adjacent clusters, thereby providing some control parameter combinations for the control parameter selection. In the process of obtaining the operating condition, it is relatively accurate to determine the operating condition type through the condition data of the association rule. Its purpose is to accurately reflect the state of the furnace temperature through the operating condition category. In this step, a group of similar operating conditions are searched based on the current operating condition category to provide a basis for the selection of control parameters. By further limiting the similar operating condition categories based on the furnace temperature deviation, the operating condition categories that do not meet the furnace temperature regulation are eliminated. The control parameter combination is obtained by comparing the energy consumption in the retained operating condition category samples, so that the smelting is regulated in advance and in accordance with the furnace temperature conditions to achieve energy-saving optimization.

[0049] S5. Abnormal operation monitoring: Real-time monitoring of furnace temperature data and identification of working conditions, and adjustment of control parameter combination of smelting operation through furnace temperature prediction. At the same time, setting the furnace temperature fluctuation threshold to determine the stability of furnace temperature changes, and issuing an early warning when the set threshold is exceeded.

[0050] Among them, judging the stability of furnace temperature change includes the following steps: S51, preset a time window, and monitor the furnace temperature data within the time window; S52, calculating the deviation between the furnace temperature mean and the target furnace temperature, and calculating the standard deviation of the furnace temperature data as the furnace temperature fluctuation value; S53. Setting judgment thresholds of mean deviation and fluctuation value at different levels. When both the mean deviation and fluctuation value exceed the set judgment threshold, an early warning is issued according to the exceeding range set for the level.

[0051] It should be noted that the energy-saving optimization in this embodiment is based on the predicted furnace temperature. The optimal energy consumption control parameter combination is selected through different furnace temperature conditions, and the smelting process control parameters are fine-tuned in time, so as to continuously optimize the smelting energy consumption. At the same time, a fault-tolerant mechanism is set for the optimization process through early warning prompts.

[0052] S6. Operation data update: Collect all the operation data of the current furnace, and mark the equipment control parameter combination in the stable state according to the judgment of the furnace temperature fluctuation, update the parameter rule base with the collected data, and update the parameter data set with the marked control parameter combination.

[0053] It should be noted that the data during the operation of the furnace is used to fill the historical operation data, so as to establish a more complete and accurate parameter rule base. At the same time, in the process of obtaining the control parameters, the parameter data set is established based on the time series to complete the data mapping, and there is a furnace temperature lag in the data matching. Therefore, the parameter data set is updated by marking the furnace temperature in the steady state in the subsequent data collection, thereby improving the accuracy of obtaining the control parameters and reducing the impact of the furnace temperature lag.

[0054] The present invention realizes energy-saving optimization in the smelting process by adjusting the furnace temperature control parameters. First, cluster analysis is performed on historical data to establish operating condition categories characterized by furnace temperature. Then, association rules are mined in each operating condition category. The condition data type and numerical range are used as the state representation of the operating condition category. The operating condition category and furnace temperature characteristic data are then input into the prediction model to obtain the predicted furnace temperature, thereby improving the accuracy of furnace temperature prediction. Finally, similar operating condition categories that meet the conditions are screened according to the predicted furnace temperature and operating condition category. The optimal control parameters are obtained through energy consumption calculation, and the smelting process is continuously regulated to achieve energy-saving optimization.

[0055] In the process of obtaining the working condition, the present invention is implemented based on the established parameter rule base. The parameter rule base is constructed by clustering the relevant data of the furnace temperature change and then using the association rules to mine the associated data for each working condition as a condition. The inherent differences of the furnace temperature change are grasped more accurately through the working condition category and the working condition change is subdivided to achieve high-precision working condition division, establish basic conditions for furnace temperature prediction and adjustment of control parameters, and then select a group of working condition categories through the furnace temperature prediction result and the working condition category to obtain the control parameter combination, thereby realizing data-driven dynamic adaptation. This process mines the correlation between the data based on the big data of smelting, closely combines the furnace temperature state and related characteristics, and then uses the furnace temperature prediction to fine-tune the working condition category, and changes the control parameter combination to achieve the purpose of energy saving optimization while meeting the target furnace temperature.

[0056] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A data-driven smelting control optimization method, characterized in that: The following steps are involved: Establish a parameter rule base: Use cluster analysis to classify the operating conditions based on the historical operating data of the current furnace, and mine association rules within each cluster to establish a parameter rule base; Working condition acquisition: according to the production data of the furnace operation, the working condition category of the current furnace temperature is obtained through the parameter rule library; Furnace temperature trend prediction: Based on the current furnace temperature characteristic data and working condition category, the change of furnace temperature is predicted in real time; Obtain control parameters: Select the control parameter combination with the best energy consumption from the parameter data set according to the current operating condition category and predicted furnace temperature data; Abnormal operation monitoring: Real-time monitoring of the furnace status and furnace temperature prediction to adjust the control parameter combination of smelting operation, while issuing early warnings based on the degree of furnace temperature fluctuation.

2. A data-driven smelting control optimization method according to claim 1, characterized in that: The historical operation data includes sensor timing data, process parameters and equipment status logs, and the production data includes real-time furnace operation data and process parameters, wherein the process parameters include the heating time, furnace pressure, air-fuel ratio and heating power set for process production.

3. The data-driven smelting control optimization method according to claim 1, characterized in that: The step of establishing a parameter rule base comprises the following steps: Data preprocessing: binning continuous control parameters to discretize them, remove noise and anomalies in time series data, and enhance the data; Working condition identification: By mining the characteristic data related to the furnace temperature change and performing cluster analysis, the working condition category of the current furnace temperature change can be identified; Rule mining: Construct an association rule mining algorithm in each working condition category, set the confidence level and minimum support level, and generate a parameter rule base for association rules with the working condition category as the result.

4. A data-driven smelting control optimization method according to claim 3, characterized in that: The working condition identification comprises the following steps: Feature selection: The data is divided into grids and mutual information is calculated. The MIC value is obtained through different grid segmentation methods. The top 30 features with the strongest relationship with the target variable are selected to construct the furnace temperature feature data; Density clustering: Dynamically select the minimum number of samples according to the sampling period of the sliding window, and use the selected features and the minimum number of samples to perform clustering using the HDBSCAN algorithm; Working condition definition: Use UMAP to reduce the dimension of the clustered data and visualize the data points after dimension reduction. Use different colors to represent different clusters for each data point. Each cluster represents a furnace temperature working condition and defines the working condition category.

5. A data-driven smelting control optimization method according to claim 4, characterized in that: The rule mining comprises the following steps: Construct a transaction database: convert the data in each working condition category into a transaction format, with each sample represented as a group item and the features of each sample as items; Mining association rules: Use the association rule mining algorithm to mine the data of each working condition category, set the minimum support and minimum confidence as the threshold for rule screening, and output the association rules that meet the conditions; Update association rules: According to the operation characteristics of smelting, set the sliding window to add new operation data, use the incremental learning method, update the association rules, and adjust the support and confidence.

6. The data-driven smelting control optimization method according to claim 1, characterized in that: The furnace temperature trend prediction comprises the following steps: Model construction: Long short-term memory neural network is used as the furnace temperature prediction model, including input layer, LSTM layer, fully connected layer and output layer; Model training: The input variables are converted into time series data through preprocessing and divided into training set, validation set and test set with the corresponding furnace temperature values. The training data is input into the LSTM model, and the network is trained through back propagation and gradient descent algorithms. Real-time prediction: Obtain furnace temperature characteristic data and furnace temperature operating condition category in real time, input the acquired data into the trained furnace temperature prediction model, and perform real-time prediction of furnace temperature.

7. A data-driven smelting control optimization method according to claim 6, characterized in that: The input variables of the input layer include furnace temperature characteristic data and operating condition categories, and the output layer includes a single node representing the predicted furnace temperature.

8. The data-driven smelting control optimization method according to claim 1, characterized in that: The obtaining of control parameters comprises the following steps: Data merging: Match the control parameter combinations corresponding to all operating condition category data from historical operation data to establish parameter data sets; Energy consumption calculation: Calculate the comprehensive energy consumption data of each set of control parameter combinations during operation, including heating energy consumption, cooling energy consumption and other energy consumption, and store the energy consumption data together with its corresponding operating condition category; Parameter selection: Calculate similar operating conditions based on the current operating condition, and define a group of related operating conditions by predicting the difference between the furnace temperature and the target furnace temperature, and select the control parameter combination with the lowest energy consumption in this group of operating conditions.

9. The data-driven smelting control optimization method according to claim 1, characterized in that: In the abnormal operation monitoring, judging the stability of furnace temperature change includes the following steps: Preset time window and monitor furnace temperature data within the time window; Calculate the deviation between the mean furnace temperature and the target furnace temperature, and calculate the standard deviation of the furnace temperature data as the furnace temperature fluctuation value; Set judgment thresholds for mean deviation and fluctuation value at different levels. When both mean deviation and fluctuation value exceed the set judgment threshold, an early warning will be issued according to the exceeded range set for the level.

10. A data-driven smelting control optimization method according to claim 1, characterized in that: Also includes the steps: Operation data update: Collect all the operation data of the current furnace, and mark the equipment control parameter combination in the stable state according to the judgment of furnace temperature fluctuation, update the parameter rule base with the collected data, and update the parameter data set with the marked control parameter combination.

Images