Electroslag remelting smelting endpoint element dynamic forecasting and process decision optimization method and system

Abstract

The invention provides an electroslag remelting smelting endpoint element dynamic forecasting and process decision optimization method and system, and the method comprises the steps: obtaining a historical data set and a real-time data flow in a smelting process, and carrying out the preprocessing of data; an online stochastic gradient descent and incremental learning algorithm is introduced, an element dynamic prediction model is constructed, sensor data in the smelting process are received in real time, and an element concentration prediction model is dynamically optimized; on the basis of a depth deterministic strategy gradient algorithm, an optimization decision model is constructed, and a smelting operation strategy is adjusted in real time according to element components, smelting working conditions and other environment parameters predicted online by the element dynamic forecasting model; and a real-time feedback mechanism is set, so that the element dynamic prediction model and the optimization decision model work cooperatively, and element concentration prediction and real-time synchronous adjustment of a control strategy are realized. The smelting parameters can be adjusted in real time, element prediction precision is improved, energy consumption is reduced, manual intervention is reduced, and smelting efficiency and quality stability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of metallurgical technology, and in particular, to a method and system for dynamically predicting the end-point elements and optimizing the process decision-making in electroslag remelting smelting. Background Art

[0002] Electroslag Remelting (ESR) is an important metallurgical process widely used in the production of high-end steel and alloy materials. In this process, the electrode is melted by an electric current, and the metal drops and solidifies in a water-cooled mold to form an ingot, thereby realizing the refining and shaping of the metal. During the electroslag remelting process, precise control of the element concentration is crucial for ensuring the compositional uniformity and quality stability of the final product.

[0003] Currently, the prediction of element concentration in the electroslag remelting process mainly relies on static mathematical models or machine learning models based on offline training. Traditional thermodynamic prediction methods are mainly based on the Molecular-Ion Coexistence Theory (MICT) and the principle of material conservation, while kinetic-based modeling methods describe the element transport and interfacial reaction processes through complex partial differential equations and numerical calculations. In addition, existing intelligent control systems usually adopt fixed-rule control (such as PID control) or manual experience control, as well as some models based on simple machine learning or deep learning to achieve the control of the smelting process.

[0004] However, these existing technologies have significant limitations. Static mathematical models and machine learning models based on offline training are difficult to adapt to the complex multi-physical field coupling effects and dynamic changes during the smelting process, resulting in low prediction accuracy. Traditional thermodynamic and kinetic models have deficiencies in timeliness, accuracy, and computational complexity, and are difficult to meet the requirements of precise composition control in the electroslag remelting process. Fixed-rule control and manual experience control lack the adaptive ability to non-linear dynamic changes and are difficult to standardize and quantify, resulting in limited control accuracy and production process consistency. In addition, existing intelligent control systems lack self-learning and optimization capabilities, cannot adapt to the dynamic changes during the smelting process in real time, and require a large amount of resources for model update and maintenance. Reinforcement learning and online learning methods also face problems such as low sample efficiency, long training time, high cost, and dependence on environmental disturbances in practical applications. Summary of the Invention

[0005] In view of the above-mentioned technical problems, a method and system for dynamically predicting the end-point elements and optimizing the process decision-making in electroslag remelting smelting are provided. The present invention can adjust the smelting parameters in real time, improve the element prediction accuracy, reduce energy consumption, and reduce manual intervention, significantly enhancing the smelting efficiency and quality stability.

[0006] The technical means adopted by the present invention are as follows:

[0007] A dynamic prediction method for end - point elements and an optimization method for process decision - making in electroslag remelting smelting, comprising:

[0008] S1. Obtain the historical data set and real - time data stream during the smelting process, and pre - process the obtained data;

[0009] S2. Introduce the Online Stochastic Gradient Descent (Online SGD) and incremental learning algorithms to construct an element dynamic prediction model, receive the sensor data during the smelting process in real - time, and dynamically optimize the element concentration prediction model;

[0010] S3. Based on the Deep Deterministic Policy Gradient (DDPG) algorithm, construct an optimization decision - making model, and adjust the smelting operation strategy in real - time according to the element composition, smelting conditions, and other environmental parameters predicted online by the element dynamic prediction model;

[0011] S4. Set up a real - time feedback mechanism to enable the element dynamic prediction model and the optimization decision - making model to work together, realize the real - time synchronous adjustment of element concentration prediction and control strategy, and ensure the accuracy of element concentration prediction and the stable control of the smelting process.

[0012] Further, step S1 specifically includes:

[0013] S11. Clean the data to remove invalid, abnormal, or incorrect data to ensure the quality of the input data;

[0014] S12. Perform normalization processing on the data, and use the normalization method to standardize or scale the data to keep it on a unified scale to ensure the stability of model training and prediction accuracy;

[0015] S13. Time - series processing, through the sliding window method, input a fixed - length historical data sequence each time, so that the model can use past information for real - time prediction and optimization;

[0016] S14. Feature extraction, extract key features from the original smelting data, including equipment operation status, raw and auxiliary material ratio, process smelting parameters, and environmental data;

[0017] S15. Store the processed data in a relational database and a time - series database to achieve efficient data management.

[0018] Further, in step S12, the method of standardizing the data is adopted, and the specific formula is:

[0019]

[0020] where \(x\) is the original value, \(\mu\) is the feature mean, and \(\sigma\) is the feature standard deviation.

[0021] Further, step S2 specifically includes:

[0022] S21. When new data arrives each time, use the online stochastic gradient descent algorithm for local update, gradually adjust the parameters of the element concentration prediction model to minimize the prediction error;

[0023] S22. Use the incremental learning algorithm to regularly update the parameters of the element concentration prediction model, so that the element concentration prediction model adapts to the accumulation of the data stream and ensures a high degree of fit with the actual data in the current smelting environment;

[0024] S23. Through the error feedback mechanism, compare the prediction errors of the online stochastic gradient descent algorithm and the incremental learning algorithm, and automatically select the element concentration prediction model with the smallest error.

[0025] Further, the bottom layer of the element dynamic prediction model is a multi-layer feedforward neural network, which consists of an input layer, multiple hidden layers, and an output layer. Among them, the key parameters of the model adjusted by the coupling strategy of the online stochastic gradient descent (Online SGD) algorithm and the incremental learning algorithm include the learning rate, the number of neurons in the hidden layer, and the selection of the activation function.

[0026] Further, the element concentration prediction model supports the online AdaGrad or online Adam optimization algorithm to adaptively adjust the learning rate and accelerate the convergence of the model. At the same time, combined with the sliding window or forgetting mechanism, it avoids the interference of old data on the element concentration prediction model, so that the element concentration prediction model focuses on the latest data stream, thereby improving the prediction accuracy and real-time response ability.

[0027] Further, step S3 specifically includes:

[0028] S31. Based on the deep deterministic policy gradient (DDPG) algorithm, construct an optimization decision model, including a policy network (Actor) and a value network (Critic). Among them, the policy network (Actor) generates continuous control actions (such as melting speed, current) according to the current smelting state; the value network (Critic) evaluates the impact of the state-action pair on the model performance and calculates the Q value to optimize the smelting operation parameters;

[0029] S32. Define the state space (smelting state parameters) and the action space (continuous operation parameters such as melting speed, current, voltage, etc.);

[0030] S33. Design a reward function that comprehensively considers the element content error, energy consumption, and efficiency, as follows:

[0031]

[0032] where, R tis the value of the reward function, representing the reward obtained by the agent at time t; the reward value is used to guide the agent to optimize its strategy; α1, α2, and α3 all represent weight coefficients, which are used to balance the influence of different parts in the reward function and determine the relative importance of the target element content error, smelting energy consumption, and production efficiency in the reward; C pred,i represents the predicted content of the i-th element, that is, the element content predicted by the agent according to the current smelting state; C target,i represents the target content of the i-th element, that is, the element content target that the smelting process hopes to achieve.

[0033] S34. Use the experience replay and target network mechanisms to train the model to achieve dynamic adjustment of smelting parameters and improve the stability and convergence of the training process.

[0034] Furthermore, in step S33, the designed reward function specifically includes:

[0035] When the content of the end element is close to the target value, a positive reward is given;

[0036] When the smelting energy consumption is too high or the efficiency decreases, a negative reward is given. Thus, the algorithm is guided to gradually adjust the smelting operation and optimize the smelting process.

[0037] The present invention also provides an electroslag remelting end element dynamic prediction and process decision optimization system implemented based on the above-mentioned electroslag remelting end element dynamic prediction and process decision optimization method, including: a data preprocessing module, an online learning module, a reinforcement learning module, and a synergy module, where:

[0038] The data preprocessing module is used to obtain the historical data set and real-time data stream in the smelting process and preprocess the obtained data;

[0039] The online learning module introduces the online stochastic gradient descent and incremental learning algorithms to construct an element dynamic prediction model, receives the sensor data in the smelting process in real time, and dynamically optimizes the element concentration prediction model;

[0040] The reinforcement learning module constructs an optimization decision model based on the deep deterministic policy gradient algorithm, and adjusts the smelting operation strategy in real time according to the element composition, smelting conditions, and other environmental parameters predicted online by the element dynamic prediction model;

[0041] The synergy module sets a real-time feedback mechanism to enable the element dynamic prediction model and the optimization decision model to work together to achieve real-time synchronous adjustment of the element concentration prediction and the control strategy, ensuring the accuracy of the element concentration prediction and the stable control of the smelting process.

[0042] Compared with the prior art, the present invention has the following advantages:

[0043] 1. Through the online stochastic gradient descent (Online SGD) and incremental learning algorithms, the present invention can receive sensor data during the smelting process in real time and dynamically optimize the element concentration prediction model, which can significantly improve the accuracy and timeliness of element concentration prediction and reduce the calculation cost. Through the error feedback mechanism and optimization algorithms (such as online AdaGrad, Adam), the model can continuously adaptively adjust, ensuring the stability and accuracy of the smelting process. By combining a sliding window or forgetting mechanism, the interference of old data to the model is avoided, making the model more focused on the latest data stream, thereby improving the prediction accuracy and real-time response ability. Through these technical means, the present invention can achieve high-precision dynamic prediction of element concentration during the smelting process, providing accurate data support for subsequent intelligent control.

[0044] 2. The present invention uses the deep deterministic policy gradient (DDPG) algorithm to adjust the smelting operation strategy, such as melting speed, current, and electrode position, in real time according to the element composition, smelting conditions, and other environmental parameters predicted online by the element dynamic prediction model. The DDPG algorithm can efficiently handle continuous action spaces, ensuring precise control of the smelting process under complex conditions. Through the combination of a policy network (Actor) and a value network (Critic), as well as the experience replay and target network mechanisms, the stability and convergence of the training process are improved. The reward function comprehensively considers the target element content error, smelting energy consumption, and efficiency improvement, guiding the algorithm to optimize the operation strategy. Through these technical means, the present invention can achieve precise control of the smelting process, optimize operation parameters, improve production efficiency, reduce energy consumption, reduce resource waste, and promote the development of the smelting process towards energy conservation and environmental protection.

[0045] 3. Through the real-time feedback mechanism, the online learning algorithm and the DDPG policy network can continuously self-optimize, ensuring the efficient operation and precise control of the smelting process in various dynamic environments. This collaborative optimization closed-loop mechanism makes the smelting process not only more efficient but also has strong adaptability and stability, greatly enhancing the intelligence and self-regulation ability of the entire process.

[0046] For the above reasons, the method of the present invention is not only applicable to the electroslag remelting process but can also be extended to other metallurgical smelting processes, such as blast furnaces, electric furnaces, vacuum induction furnaces, and vacuum consumable furnaces, etc. Only by appropriately adjusting the model and optimizing the input and output parameters to adapt to the characteristics of different smelting processes. Through precise parameter adjustment and intelligent control strategies, efficient control of the smelting process can be achieved, ensuring the maximization of the utilization of various resources during the production process while reducing the environmental burden. Precise control strategies can significantly reduce resource waste during the smelting process and promote the development of the smelting process towards energy conservation and environmental protection. Description of the Drawings

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0048] Figure 1 This is the flowchart of the method of the present invention.

[0049] Figure 2 This is the operation flowchart of the system of the present invention. Detailed implementation manners

[0050] In order to enable those skilled in the art of the present technology to better understand the solution of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0051] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above accompanying drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0052] As Figure 1 shown, the present invention provides a method for dynamic prediction of end-point elements and optimization of process decision-making in electroslag remelting smelting, including:

[0053] S1. Obtain the historical data set and real-time data stream during the smelting process, and preprocess the obtained data;

[0054] S2. Introduce the Online Stochastic Gradient Descent (Online SGD) and incremental learning algorithms to construct an element dynamic prediction model, receive the sensor data during the smelting process in real time, and dynamically optimize the element concentration prediction model;

[0055] S3. Based on the Deep Deterministic Policy Gradient (DDPG) algorithm, an optimization decision model is constructed. According to the element composition, smelting conditions, and other environmental parameters predicted online by the element dynamic prediction model, the smelting operation strategy is adjusted in real time;

[0056] S4. A real-time feedback mechanism is set up to enable the element dynamic prediction model and the optimization decision model to work together, realizing real-time synchronous adjustment of element concentration prediction and control strategy, and ensuring the accuracy of element concentration prediction and the stable control of the smelting process.

[0057] Specifically in implementation, as a preferred implementation manner of the present invention, step S1 specifically includes:

[0058] S11. Clean the data to remove invalid, abnormal, or incorrect data, and ensure the quality of the input data;

[0059] In this embodiment, it specifically includes: for historical data, for fields with fewer missing values, use the mean, median, or interpolation method to fill; if there are too many missing data and they cannot be reasonably filled, then remove this part of the data; for abnormal data, use the 3σ principle and the Isolation Forest statistical method to detect and process abnormal data; for real-time data streams, since the API may collect data with the same timestamp repeatedly, it is necessary to remove duplicates to ensure that the records at each time point are unique. And based on the Isolation Forest algorithm trained with historical data, alarm and process abnormal data.

[0060] S12. Perform normalization processing on the data. Use the normalization method to standardize or scale the data to keep it on a unified scale, and ensure the stability of model training and prediction accuracy;

[0061] In this embodiment, the method of standardizing the data is adopted, and the specific formula is:

[0062]

[0063] Among them, x is the original value, μ is the feature mean, and σ is the feature standard deviation. The normalized data can effectively avoid the influence of unbalanced numerical scales on the model learning process, and improve the training efficiency and prediction accuracy. The normalized data can effectively avoid the influence of unbalanced numerical scales on the model learning process, and improve the training efficiency and prediction accuracy.

[0064] S13. Time series processing. By means of the sliding window method, input a historical data sequence of a fixed length each time, so that the model can use past information for real-time prediction and optimization;

[0065] In this embodiment, since the smelting data is time-series data and there is a time dependence between the data at different time points, it is necessary to maintain the temporal consistency of the data during data preprocessing. The sliding window processing is an important part of this method, and its window size and step size need to be set in advance according to the sensor characteristics of the user and the sampling measurement conditions to ensure that the data can effectively capture the dynamic changes in the smelting process. Through the sliding window method, a historical data sequence of a fixed length is input each time, enabling the model to utilize past information for real-time prediction and optimization. In addition, the time series processing also includes steps such as timestamp alignment and feature extraction (such as mean, trend, and periodicity analysis) to enhance the model's prediction ability and adapt to the continuous changes in the smelting process.

[0066] S14. Feature extraction, extracting key features from the original smelting data, including equipment operating status, raw material and auxiliary material ratio, process smelting parameters, and environmental data;

[0067] In this embodiment, in order to improve the prediction accuracy and generalization ability of the model, it is necessary to extract key features from the original smelting data, mainly including equipment operating status, raw material and auxiliary material ratio, process smelting parameters, and environmental data. Among them, equipment features cover core parameters such as current, voltage, and melting speed, raw material and auxiliary material features involve the feeding ratio and chemical composition of key materials, process features include furnace temperature change, slag layer thickness, etc., and environmental features consider external influencing factors such as humidity and air flow. Among them, parameters such as furnace temperature change that cannot be directly or indirectly measured and sampled can be calculated through mechanism models such as kinetics. In addition, to reduce redundant data and optimize calculation efficiency, filtering methods (Spearman correlation coefficient analysis screening), embedding methods (algorithms such as principal component analysis, Lasso regression, etc.), and wrapper methods (algorithms such as recursive feature elimination) can be used to screen the most representative features, and at the same time, algorithms such as PCA and KPCA can be used for eigenvalue dimensionality reduction. Ensure that the model can more accurately learn the key patterns of the smelting process and achieve intelligent optimization and dynamic adjustment of the smelting state.

[0068] S15. Store the processed data in a relational database and a time-series database to achieve efficient data management.

[0069] In this embodiment, in order to efficiently manage this data, a relational database is used to store historical data, and a time-series database is used to process real-time data streams. Combined with distributed storage and caching mechanisms, it ensures efficient access and real-time processing of data. In addition, real-time monitoring is achieved through a data stream management platform, and stream processing technology is used to timely capture changes in the smelting process. Finally, data security and permission management need to be emphasized in database deployment, and measures such as encryption, access control, and security auditing are adopted to ensure the security and privacy protection of data, thereby providing a solid data foundation for the dynamic optimization of the smelting process.

[0070] In specific implementation, as a preferred implementation manner of the present invention, step S2 specifically includes:

[0071] S21. When new data arrives each time, the Online Stochastic Gradient Descent (Online SGD) algorithm is used for local update, and the parameters of the element concentration prediction model are gradually adjusted to minimize the prediction error;

[0072] In this embodiment, the Online Stochastic Gradient Descent (Online SGD) algorithm minimizes the prediction error by gradually adjusting the model parameters and performs local update when new data arrives each time, avoiding the high computational cost of full-data training. This enables the model to closely follow the changes in the data stream in real time and timely reflect the actual situation in the smelting process.

[0073] S22. The incremental learning algorithm is used to regularly update the parameters of the element concentration prediction model, so that the element concentration prediction model adapts to the accumulation of the data stream and ensures a high degree of fit with the actual data of the current smelting environment;

[0074] In this embodiment, the incremental learning algorithm enables the model to automatically adjust during the accumulation of the data stream by regularly updating the model parameters, thereby ensuring that the model can highly fit the actual data of the smelting environment at any time. This method avoids frequent retraining for model update and improves the real-time performance and adaptability of the system.

[0075] S23. The prediction errors of the Online Stochastic Gradient Descent algorithm and the incremental learning algorithm are compared through an error feedback mechanism, and the element concentration prediction model with the smallest error is automatically selected.

[0076] In this embodiment, in order to further improve the accuracy and stability of the model, an error feedback mechanism is built, that is, by comparing the errors between the prediction results of the two strategies of Online SGD and incremental learning and the actual values in real time, the model with the smallest error is automatically selected for application. Specifically, the error feedback mechanism is used to compare the prediction errors of the two algorithms, and the parameters of the model are adjusted according to the optimal algorithm. If the adjustment result of Online SGD is better than that of incremental learning, the SGD optimization strategy is preferentially applied, and vice versa, to ensure the accuracy of the model in a dynamic environment. This process continuously corrects the model weights through error feedback, thereby improving the real-time performance and accuracy of the model.

[0077] In specific implementation, as a preferred implementation manner of the present invention, the bottom layer of the element dynamic prediction model is a multi-layer feedforward neural network, which consists of an input layer, multiple hidden layers and an output layer. Among them, the key parameters of the model adjusted by the coupling strategy of the Online Stochastic Gradient Descent (Online SGD) algorithm and the incremental learning algorithm include the learning rate, the number of neurons in the hidden layer, and the selection of the activation function.

[0078] In specific implementation, as a preferred implementation manner of the present invention, the element concentration prediction model supports online AdaGrad or online Adam optimization algorithms to adaptively adjust the learning rate, accelerate model convergence, and at the same time, in combination with a sliding window or forgetting mechanism, avoid the interference of old data on the element concentration prediction model, so that the element concentration prediction model focuses on the latest data stream, thereby improving the prediction accuracy and real-time response ability.

[0079] In specific implementation, as a preferred implementation manner of the present invention, step S3 specifically includes:

[0080] S31. Based on the Deep Deterministic Policy Gradient (DDPG) algorithm, construct an optimized decision model, including a policy network (Actor) and a value network (Critic). Among them, the policy network (Actor) generates continuous control actions (such as melting speed, current) according to the current smelting state; the value network (Critic) evaluates the impact of the state-action pair on the model performance and calculates the Q value to optimize the smelting operation parameters;

[0081] In this embodiment, DDPG, as a continuous control reinforcement learning algorithm based on the Actor-Critic structure, is more suitable for high-dimensional and continuous control problems, such as the optimization and adjustment of key smelting parameters such as melting speed and current. DDPG can output a smooth and differentiable action space adjustment strategy, making the optimization of smelting operations more stable.

[0082] S32. Define the state space (smelting state parameters) and the action space (continuous operation parameters such as melting speed, current, voltage, etc.);

[0083] In this embodiment, the state space defines the smelting state parameters at a certain moment during the smelting process. These parameters reflect the specific situation of the smelting process, including physical quantities collected by sensors, smelting historical states, environmental parameters, and auxiliary raw material parameters, etc. These parameters usually do not change easily during the smelting process, so as the dimension of the state space, they can relatively stably reflect the current smelting condition. The action space is the operation parameters that the intelligent control system needs to adjust according to the current smelting state, belonging to the continuous action space, including parameters such as melting speed, voltage, current, cooling system temperature control, and electrode parameters.

[0084] S33. Design a reward function that comprehensively considers element content error, energy consumption, and efficiency, as follows:

[0085]

[0086] Wherein, R tis the value of the reward function, representing the reward obtained by the agent at time t; the reward value is used to guide the agent to optimize its policy; α1, α2, and α3 all represent weight coefficients, which are used to balance the influence of different parts of the reward function and determine the relative importance of the target element content error, smelting energy consumption, and production efficiency in the reward; C pred,i represents the predicted content of the i-th element, that is, the element content predicted by the agent according to the current smelting state; C target,i represents the target content of the i-th element, that is, the target of the element content that the smelting process hopes to achieve.

[0087] In this embodiment, the reward function will guide the agent to select appropriate actions during the learning process to achieve the optimization goal. In this model, the reward function needs to ensure that the content of the end element always remains within the target range, and at the same time, it is necessary to minimize the smelting energy consumption and improve the smelting efficiency as much as possible. Therefore, a reward function that comprehensively considers the target element content error, smelting energy consumption, and efficiency improvement is designed.

[0088] S34. Use the experience replay and target network mechanisms to train the model to achieve dynamic adjustment of smelting parameters. Improve the stability and convergence of the training process.

[0089] In summary, during the training process of the DDPG model, the agent continuously optimizes the policy by maximizing the long-term reward, and gradually learns how to adjust the smelting operation parameters to ensure that the content of the end element is within a reasonable range. To improve the training efficiency and stability, an experience replay pool is used to store the state-action-reward triples, and combined with the prioritized experience replay mechanism, select those experiences that contribute more to policy optimization for learning. Finally, optimization suggestions are made for the action space to achieve process optimization. In this way, DDPG can adjust the operation parameters in real time during the smelting process, ensure the precise control of the end element content, and at the same time optimize the energy efficiency and production efficiency of the smelting process, realizing the intelligent optimization and precise control of the smelting process.

[0090] Specifically, as a preferred embodiment of the present invention, in step S33, the designed reward function specifically includes:

[0091] When the content of the end element approaches the target value, a positive reward is given;

[0092] When the smelting energy consumption is too large or the efficiency decreases, a negative reward is given. Thus, the algorithm is guided to gradually adjust the smelting operation and optimize the smelting process.

[0093] Corresponding to the electro-slag remelting smelting end element dynamic prediction and process decision-making optimization method in this application, this application also provides an electro-slag remelting smelting end element dynamic prediction and process decision-making optimization system, as Figure 2As shown in the figure, it includes: a data preprocessing module, an online learning module, a reinforcement learning module, and a synergy module, where:

[0094] The data preprocessing module is used to obtain the historical data set and real-time data stream in the smelting process, and preprocess the obtained data;

[0095] The online learning module introduces online stochastic gradient descent (Online SGD) and incremental learning algorithms, constructs an element dynamic prediction model, receives sensor data in the smelting process in real time, and dynamically optimizes the element concentration prediction model;

[0096] The reinforcement learning module constructs an optimization decision model based on the deep deterministic policy gradient (DDPG) algorithm, and adjusts the smelting operation strategy in real time according to the element composition, smelting conditions, and other environmental parameters predicted online by the element dynamic prediction model;

[0097] The synergy module sets a real-time feedback mechanism, enabling the element dynamic prediction model and the optimization decision model to work together, realizing real-time synchronous adjustment of element concentration prediction and control strategy, and ensuring the accuracy of element concentration prediction and the stable control of the smelting process.

[0098] For the embodiments of the present invention, since it corresponds to the above embodiments, the description is relatively simple. For relevant similarities, please refer to the description in the above embodiments, and details are not repeated here.

[0099] Embodiment

[0100] In the verification stage of this embodiment, data of a certain electroslag remelting furnace of a special steel plant in China in the past year was used to construct a dynamic prediction and optimization decision method for the final titanium content, as follows:

[0101] I. Data acquisition and preprocessing

[0102] First, data is obtained. Based on the production database provided by the factory, the data includes equipment parameters (such as mold dimensions), slag system parameters (such as slag volume, slag system composition content, etc.), types and dosages of alloy additives, environmental parameters (such as atmospheric humidity, atmospheric temperature, etc.), cooling system parameters (such as inlet water temperature, outlet water temperature), smelting parameters during electroslag remelting (such as current, voltage, melting speed, protective gas state parameters, etc.), electrode state parameters (such as electrode dimensions, shape, and state, etc.), ingot parameters (such as composition content, dimensions, shape, final ingot temperature, etc.), and ingot end element content.

[0103] Then data cleaning was carried out, which included two parts: historical data cleaning and real-time data cleaning. In historical data cleaning, for fields with fewer missing values, the mean filling method was used for processing. Fields with more missing values that could not be reasonably filled were excluded. At the same time, the 3σ principle and the Isolation Forest algorithm were used to handle outliers. In real-time data cleaning, duplicate records were removed through timestamps to ensure the uniqueness of each record, and the Isolation Forest model trained based on historical data was used to detect and alarm outliers in real-time data, thus ensuring the accuracy and consistency of the data.

[0104] Then all data was normalized, and at the same time, real-time data was sequenced. To further improve the performance of the model and reduce costs, the mutual information method was used for model screening. The specific principle and formula are as follows:

[0105]

[0106] Finally, 21 input variables were selected, including ingot weight, slag amount, initial titanium, tungsten, copper, aluminum, ferrosilicon powder, silicon, carbon, aluminum powder, sulfur, carbon powder, manganese, oxygen, phosphorus, inlet temperature, outlet temperature, ambient humidity, ambient temperature, real-time current and voltage, and melting rate. The output variable is the final titanium content.

[0107] II. Building an online learning module

[0108] In the online learning module, an 8-layer deep neural network (DNN) was constructed to predict the dynamic changes of element concentrations during the smelting process in real-time. This model combines the online stochastic gradient descent (SGD) and incremental learning (IncrementalLearning) methods, and introduces an error feedback mechanism. According to the prediction error of the model, it dynamically selects whether to use online learning or incremental learning for updating to ensure that the model can adapt to the real-time data changes during the smelting process. In addition, the Adam adaptive optimization algorithm is introduced to dynamically adjust the learning rate to further optimize the hyperparameters and improve the convergence speed and generalization ability of the model.

[0109] The 8-layer deep neural network (DNN) constructed in the online learning module has input features covering smelting parameters such as current, voltage, melting rate, furnace temperature, slag layer thickness, and key element concentrations. The 6 hidden layers use the ReLU activation function, with the number of layers being 128 - 256 - 512 - 512 - 256 - 128 in sequence, and finally output the predicted element concentration values. During the training process, the mean squared error (MSE) is used as the loss function, and data sampling is carried out through a sliding window mechanism (window length 10 minutes, step size 1 minute) to ensure that the model can capture the dynamic changes of the smelting conditions.

[0110] For the optimization of the model, a set of error feedback mechanisms is designed. Based on the data nodes of every 30 furnace runs, the prediction errors (MSE) of two strategies, online SGD and incremental learning, are calculated, and then the model with the smallest error is automatically selected for application. That is, the system regularly uses the error feedback mechanism to compare the prediction errors of the two algorithms and adjusts the model parameters according to the optimal algorithm. If the adjustment result of online SGD is better than that of incremental learning (lower MSE), the SGD optimization strategy is preferentially applied for rapid update; if the prediction error of incremental learning is lower, the incremental learning strategy is adopted for stable update. This error feedback process ensures the accuracy of the model in a dynamic environment, improves real-time performance and adaptability by continuously correcting the model weights. In addition, the Adam optimization algorithm is introduced, which combines momentum optimization (Momentum) and RMSprop to adaptively adjust the learning rate and improve the convergence speed and stability of the model. The dynamic adjustment mechanism of Adam can effectively avoid the problems of too large or too small learning rate in the case of large changes in the smelting data stream, ensuring the efficient update of the model.

[0111] To verify the effectiveness of the model, traditional machine learning methods (random forest, support vector machine, XGBoost), using only neural networks (without online / incremental learning), and methods using incremental learning and online learning respectively on the basis of neural networks are compared. The experimental results show that:

[0112] Traditional machine learning methods perform well on static data, but have weak generalization ability in a dynamic smelting environment, with too high MSE, reaching up to 53.21 at most and 0.21 at least. As the data stream is input, a single machine learning model cannot update itself autonomously and cannot guide production smelting for prediction;

[0113] When only using neural networks (without online / incremental learning), the model is difficult to adapt to real-time data changes, resulting in large prediction errors. As the data flows in, the MSE gradually increases from small, and the fluctuation has great uncertainty, with the maximum value of MSE reaching 12.231.

[0114] When using incremental learning and online learning alone, the use of both strategies reduces the error of the model, but the error range is within (0.02 - 10.2). After combining online learning and incremental learning and adopting the error feedback mechanism, the model can dynamically select the optimal update strategy, making the overall value of MSE within 0.02, and the prediction accuracy is significantly improved.

[0115] The experimental results show that this method can effectively reduce the prediction error (the MSE is reduced from 0.02 to within 0.008), and through the error feedback mechanism and dynamic model update strategy, the adaptability and stability of the model are improved. During the smelting process, the model realizes more accurate prediction and control of element concentration, meeting the process requirements.

[0116] III. Building the Reinforcement Learning Module

[0117] Based on the online learning module, a Deep Deterministic Policy Gradient (DDPG) reinforcement learning algorithm was built to optimize the smelting operation strategy in real time. DDPG adopts the Actor-Critic structure, where the Actor network adjusts the operation parameters (such as electrode position, feeding rate, etc.) in real time according to the online predicted elemental composition, smelting conditions, and environmental parameters; while the Critic network evaluates the quality of the operation strategy through the reward function and continuously optimizes the Actor network through the feedback mechanism.

[0118] The state space was defined as 7 variables, including the inlet water temperature, outlet water temperature, environmental humidity, environmental temperature, real-time current, voltage, and melting rate, thus constructing a 7-dimensional state space. Based on this state space, a 3-dimensional action space was defined to optimize the operation parameters according to the state changes. Specifically, the action space includes the melting rate, current, and voltage, which is a 3-dimensional continuous space.

[0119] The input dimension of the Actor network is 7 (corresponding to 7 state variables), and the output dimension is 3 (corresponding to 3 actions: melting rate, current, and voltage). A 9-layer fully connected neural network (FCN) is used, with the number of neurons in each layer being 256, 128, and 64 (repeated layers). The ReLU activation function is adopted, and an L2 regularization layer is added to reduce overfitting.

[0120] The input of the Critic network includes the concatenation of state variables and action variables, with an input dimension of 10 and the output being the Q value (scalar). A 5-layer fully connected neural network (FCN) is used, with the number of neurons in each layer being 256, 128, 128, 64, 64. The ReLU activation function is adopted, and an L2 regularization layer is also added.

[0121] Hyperparameter settings:

[0122] Learning rate: Using the Adam optimizer, the learning rates of both the Actor and Critic networks are set to 0.001, β1 = 0.9, β2 = 0.999. Size of the experience replay pool: 10000. Batch size: Each time an update is made, 64 samples are randomly sampled. Target network update frequency: The target network is updated every 50 steps. Discount factor (gamma): 0.99. Exploration rate: The initial exploration rate is 0.1 and decays gradually during training.

[0123] Reward function design:

[0124] The reward function combines multiple indicators in the smelting process, aiming to minimize the deviation of titanium element content and improve the electrical efficiency. The specific design is as follows:

[0125] Component error: The reward is calculated based on the error between the final titanium content and the target value during the smelting process. The smaller the error, the higher the reward.

[0126] The target concentration is set within a tolerance range of ±1%. A negative reward is given when the range is exceeded.

[0127] When the content error of the final titanium element is lower than the set threshold (e.g., 1%), a positive reward is given; otherwise, a negative reward is given.

[0128] Temperature error: The target range of temperature is set as:

[0129] Inlet water temperature: ±2°C, outlet water temperature: ±3°C.

[0130] If the temperature deviation exceeds the set threshold, the reward is negative. The smaller the error, the higher the reward.

[0131] Electric efficiency: When the electric efficiency increases by 1%, the reward increases by +10; if the electric efficiency decreases, a negative reward is given. The smaller the error, the higher the reward.

[0132] Abnormal operation alarm: When the following abnormal operations occur during the smelting process, the alarm mechanism is triggered and penalties are imposed on the abnormal operations:

[0133] When the final titanium content exceeds the specified range, the inlet and outlet water temperatures exceed ±5°C, or the component deviation exceeds ±2%, the system will trigger the alarm mechanism and perform automatic adjustment.

[0134] Optimization results:

[0135] This method can significantly improve the operation effect during the smelting process, specifically manifested as:

[0136] The deviation of the final titanium content is reduced by 3%, the number of abnormal operation alarms is reduced by 47%, and the electric efficiency is increased by 11.5%.

[0137] Through optimized control, the temperature and component deviations during the smelting process can be maintained within the target range, significantly improving the stability and energy efficiency of the smelting process, while reducing the losses caused by abnormal operations.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic prediction of endpoint elements and process decision optimization of electroslag remelting, characterized in that: Including: S1. Obtain the historical data set and real-time data stream during the smelting process, and preprocess the obtained data; S2. Introduce the online stochastic gradient descent and incremental learning algorithms to construct an element dynamic prediction model, receive the sensor data during the smelting process in real time, and dynamically optimize the element concentration prediction model; S3. Based on the deep deterministic policy gradient algorithm, construct an optimization decision model, and adjust the smelting operation strategy in real time according to the element composition, smelting conditions and other environmental parameters predicted online by the element dynamic prediction model; S4. Set up a real-time feedback mechanism to enable the element dynamic prediction model and the optimization decision model to work together, realize the real-time synchronous adjustment of the element concentration prediction and the control strategy, and ensure the accuracy of the element concentration prediction and the stable control of the smelting process.

2. The method for dynamic prediction of endpoint elements and process decision optimization of electroslag remelting according to claim 1 is characterized in that: Step S1 specifically includes: S11. Clean the data to remove invalid, abnormal or incorrect data to ensure the quality of the input data; S12. Normalize the data. Use the normalization method to standardize or scale the data so that it remains on a unified scale to ensure the stability of model training and prediction accuracy; S13. Time series processing. Through the sliding window method, input a historical data sequence of a fixed length each time, so that the model can use past information for real-time prediction and optimization; S14. Feature extraction. Extract key features from the original smelting data, including equipment operating status, raw material ratio, process smelting parameters and environmental data; S15. Store the processed data in a relational database and a time series database to achieve efficient data management.

3. The method for dynamic prediction of endpoint elements and process decision optimization of electroslag remelting according to claim 1 is characterized in that: In step S12, the data is standardized, and the specific formula is: where x is the original value, μ is the feature mean, and σ is the feature standard deviation.

4. The dynamic prediction method for end-point elements and the optimization method for process decision-making in electroslag remelting according to claim 1, characterized in that Step S2 specifically includes: S21. When new data arrives each time, use the online stochastic gradient descent algorithm for local update, and gradually adjust the parameters of the element concentration prediction model to minimize the prediction error; S22. Use the incremental learning algorithm to regularly update the parameters of the element concentration prediction model, so that the element concentration prediction model adapts to the accumulation of the data stream and ensures a high degree of fit with the actual data of the current smelting environment; S23. Compare the prediction errors of the online stochastic gradient descent algorithm and the incremental learning algorithm through the error feedback mechanism, and automatically select the element concentration prediction model with the smallest error.

5. A dynamic prediction method for end-point elements and an optimization method for process decision-making in electroslag remelting smelting according to claim 4, characterized in that The bottom layer of the element dynamic prediction model is a multi-layer feedforward neural network, which consists of an input layer, multiple hidden layers and an output layer. Among them, the key parameters of the model adjusted by the coupling strategy of the online stochastic gradient descent algorithm and the incremental learning algorithm include the learning rate, the number of neurons in the hidden layer, and the selection of the activation function.

6. The dynamic prediction method for end-point elements and the optimization method for process decision-making in electroslag remelting according to claim 4, characterized in that The element concentration prediction model supports the online AdaGrad or online Adam optimization algorithm to adaptively adjust the learning rate, accelerate the convergence of the model, and at the same time combine the sliding window or forgetting mechanism to avoid the interference of old data on the element concentration prediction model, so that the element concentration prediction model focuses on the latest data stream.

7. The dynamic prediction method for end-point elements and the optimization method for process decision-making in electroslag remelting according to claim 1, characterized in that Step S3 specifically includes: S31. Based on the Deep Deterministic Policy Gradient algorithm, construct an optimization decision-making model, including a policy network and a value network. Among them, the policy network generates continuous control actions according to the current smelting state; the value network evaluates the impact of the state-action pair on the model performance and calculates the Q value to optimize the smelting operation parameters; S32. Define the state space and the action space; S33. Design a reward function that comprehensively considers the element content error, energy consumption, and efficiency, as follows: Among them, R t is the value of the reward function, which represents the reward obtained by the agent at time t; α1, α2, and α3 all represent weight coefficients; C pred,i represents the predicted content of the i-th element; C target,i Represents the target content of the i-th element. S34. Use the experience replay and target network mechanisms to train the model to dynamically adjust the smelting parameters.

8. The method for dynamic prediction of endpoint elements and process decision optimization of electroslag remelting according to claim 7, characterized in that: In step S33, the designed reward function specifically includes: When the content of the end-point element is close to the target value, a positive reward is given; When the smelting energy consumption is too high or the efficiency decreases, a negative reward is given.

9. A system for dynamic prediction of endpoint elements and process decision optimization of electroslag remelting, implemented based on the method for dynamic prediction of endpoint elements and process decision optimization of electroslag remelting as described in any one of claims 1 to 8, characterized in that: Including: A data preprocessing module, an online learning module, a reinforcement learning module, and a synergy module, where: The data preprocessing module is used to obtain the historical data set and real-time data stream during the smelting process and preprocess the obtained data; The online learning module introduces the online stochastic gradient descent and incremental learning algorithms to construct an element dynamic prediction model, receives the sensor data during the smelting process in real time, and dynamically optimizes the element concentration prediction model; The reinforcement learning module, based on the Deep Deterministic Policy Gradient algorithm, constructs an optimization decision-making model, and adjusts the smelting operation strategy in real time according to the element composition, smelting conditions, and other environmental parameters predicted online by the element dynamic prediction model; The synergy module sets a real-time feedback mechanism to enable the element dynamic prediction model and the optimization decision-making model to work together, realizing the real-time synchronous adjustment of the element concentration prediction and the control strategy, and ensuring the accuracy of the element concentration prediction and the stable control of the smelting process.

Images