Converter oxygen consumption prediction method and system based on improved hybrid expert model

Abstract

The invention provides a converter oxygen consumption prediction method and system based on an improved hybrid expert model, and relates to the technical field of data processing.The method comprises the steps that the hybrid expert model is constructed, the hybrid expert model comprises an expert part and a gating part, and the expert part comprises a plurality of converter oxygen consumption prediction models; inputting prediction parameters to the expert part; according to the prediction parameters, through each converter oxygen consumption prediction model in the expert part, a converter oxygen consumption preliminary prediction value is output; inputting a prediction parameter and a converter oxygen consumption preliminary prediction value output by each converter oxygen consumption prediction model into a gating part; according to the prediction parameters and the converter oxygen consumption preliminary prediction value output by each converter oxygen consumption prediction model, selecting a preset number of converter oxygen consumption prediction models as a target converter oxygen consumption prediction model through a gating part; and according to the converter oxygen consumption preliminary prediction value of each target converter oxygen consumption prediction model, calculating a converter oxygen consumption final prediction value.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a method and system for predicting converter oxygen consumption based on an improved hybrid expert model. Background Art

[0002] With the widespread application of oxygen converter steelmaking process, converter steelmaking has become the mainstream steel production method. This process involves the redox reaction of oxygen with carbon, silicon, manganese, phosphorus, sulfur and other elements in molten iron, with the purpose of removing impurity elements and releasing a large amount of heat. In the converter steelmaking process, oxygen consumption is a key control parameter. Accurately predicting the oxygen consumption of the converter is of great practical significance for optimizing steelmaking process control, improving molten steel quality and saving production costs.

[0003] In current technology, a single mechanism or statistical model is usually used to calculate the oxygen consumption of the converter. However, due to the presence of a large number of idealized assumptions in the single mechanism or statistical model, it is unable to adapt to the reaction process under various operating conditions and is difficult to apply to converters under different operating conditions, which leads to deviations in the prediction results.

[0004] In addition, the existing technology often combines the mechanism model with the statistical model, identifies the production parameters that have a significant impact on the converter oxygen consumption through the reaction mechanism, simplifies the relationship between the oxygen consumption and these production parameters into a linear model, and fits it with linear regression. However, since the reaction mechanism in the converter steelmaking process is complex and the relationship between the oxygen consumption and the production parameters is usually nonlinear, the simplified assumptions of the linear model make it unable to fully reflect the real reaction mechanism, and the coefficients fitted by historical data also lack interpretability.

[0005] In the above technical scheme, since a single mechanism or statistical model has a large number of idealized assumptions, it cannot adapt to the reaction process under various operating conditions and is difficult to apply to converters under different operating conditions, which leads to deviations in the prediction results. By combining the mechanism model with the statistical model, the reaction mechanism in the converter steelmaking process is complex and the relationship between oxygen consumption and production parameters is usually nonlinear, making the simplified assumptions of the linear model unable to fully reflect the true reaction mechanism, and the coefficients fitted by historical data also lack interpretability. Summary of the invention

[0006] In order to solve the problem that traditional single mechanism or statistical models have a large number of idealized assumptions, cannot adapt to the reaction process under various working conditions, and are difficult to apply to converters under different working conditions, which leads to deviations in prediction results, the mechanism model and the statistical model are combined. Because the reaction mechanism in the converter steelmaking process is complex and the relationship between oxygen consumption and production parameters is usually nonlinear, the simplified assumptions of the linear model cannot fully reflect the real reaction mechanism, and the coefficients fitted by historical data also lack interpretability. The present invention provides a converter oxygen consumption prediction method and system based on an improved hybrid expert model.

[0007] The technical solution provided by the embodiment of the present invention is as follows:

[0008] First aspect:

[0009] An embodiment of the present invention provides a method for predicting converter oxygen consumption based on an improved hybrid expert model, comprising:

[0010] S1: constructing a hybrid expert model, wherein the hybrid expert model includes an expert part and a gating part, wherein the expert part includes a plurality of converter oxygen consumption prediction models;

[0011] S2: inputting prediction parameters into the expert part;

[0012] S3: outputting a preliminary predicted value of converter oxygen consumption according to the predicted parameters and using each converter oxygen consumption prediction model in the expert part;

[0013] S4: inputting the prediction parameters and the preliminary prediction values ​​of converter oxygen consumption output by each converter oxygen consumption prediction model into the gate control part;

[0014] S5: According to the prediction parameters and the preliminary predicted values ​​of converter oxygen consumption output by each converter oxygen consumption prediction model, a preset number of converter oxygen consumption prediction models are selected as target converter oxygen consumption prediction models through a gating part;

[0015] S6: Calculate the final predicted value of the converter oxygen consumption according to the preliminary predicted value of the converter oxygen consumption of each target converter oxygen consumption prediction model.

[0016] Second aspect:

[0017] An embodiment of the present invention provides a converter oxygen consumption prediction system based on an improved hybrid expert model, comprising: a memory and one or more processors;

[0018] One or more application programs are stored in the memory, and the one or more application programs are suitable for being executed by the one or more processors to implement the above-mentioned converter oxygen consumption prediction method based on the improved hybrid expert model.

[0019] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0020] In the present invention, a hybrid expert model is constructed, which includes an expert part and a gating part. The expert part includes multiple converter oxygen consumption prediction models, and prediction parameters are first input into the expert part. Through each converter oxygen consumption prediction model in the expert part, a preliminary predicted value of converter oxygen consumption is output. Multiple experts can adapt to the reaction process under various operating conditions, avoiding the problem that a single mechanism or statistical model is difficult to apply to converters under different operating conditions, making the prediction result more accurate, and then the prediction parameters and the preliminary predicted value of converter oxygen consumption output by each converter oxygen consumption prediction model are input into the gating part, and a preset number of converter oxygen consumption prediction models are selected as target converter oxygen consumption prediction models, and then the final predicted value of converter oxygen consumption is calculated, which can simplify the reaction mechanism of converter steelmaking, solve the simplified assumption problem of linear models, and more comprehensively reflect the real reaction mechanism, so that the coefficients fitted by historical data have better interpretability. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 A schematic flow chart of a method for predicting converter oxygen consumption based on an improved hybrid expert model provided in an embodiment of the present invention;

[0023] Figure 2 A structural schematic diagram of a converter oxygen consumption prediction system based on an improved hybrid expert model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0025] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0026] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0027] Reference Manual Attached Figure 1 , showing a flow chart of a method for predicting converter oxygen consumption based on an improved hybrid expert model provided in an embodiment of the present invention.

[0028] The embodiment of the present invention provides a method for predicting converter oxygen consumption based on an improved hybrid expert model. The method can be implemented by a converter oxygen consumption prediction device based on an improved hybrid expert model. The converter oxygen consumption prediction device based on an improved hybrid expert model can be a terminal or a server. The processing flow of the method for predicting converter oxygen consumption based on an improved hybrid expert model can include the following steps:

[0029] S1: Construct a hybrid expert model. The hybrid expert model includes an expert part and a gating part. The expert part includes multiple converter oxygen consumption prediction models.

[0030] It should be noted that the framework of the mixed expert model (MoE) is "expert-independent", that is, there is no fixed requirement for the type, number and implementation of the expert part. Those skilled in the art can flexibly set the number and type of expert models according to actual needs.

[0031] Optionally, the method for constructing the converter oxygen consumption prediction model includes: random forest, support vector machine, multi-layer perceptron and KAN network.

[0032] Among them, Random Forest (RF) is an integrated learning method that builds multiple decision trees and votes or averages their results to improve the accuracy and stability of predictions. Each decision tree uses a random subset of the data set during training, and only considers a random subset of features when dividing nodes, thereby reducing the risk of overfitting. The final random forest model obtains the final prediction value by combining the prediction results of all trees.

[0033] Among them, Support Vector Machine (SVM) is a supervised learning model, mainly used for classification and regression tasks. It separates data points of different categories by finding an optimal hyperplane and maximizes the margin between categories. When dealing with nonlinear problems, SVM uses a kernel function to map data to a higher-dimensional space, making the data linearly separable in the new space.

[0034] Among them, the Multilayer Perceptron (MLP) is a feedforward neural network consisting of multiple layers of neurons, including input layer, hidden layer and output layer. Each layer of neurons undergoes nonlinear transformation through activation functions, enabling it to learn complex patterns and relationships. MLP is trained through the back-propagation algorithm, adjusting weights to minimize prediction errors, and is widely used in tasks such as classification and regression.

[0035] Among them, the KAN network (Kolmogorov-Arnold Network, KAN) is a feedforward neural network based on the theories of Russian mathematicians Kolmogorov and Arnold, and is designed to efficiently approximate nonlinear functions. The KAN network has a specific structure and activation function, which enables it to show a high ability in approximating complex functions, and is particularly suitable for processing high-dimensional data and nonlinear mapping tasks. Its core is to use neurons with specific properties to simplify the network training process and improve the model's expressiveness.

[0036] In the present invention, different models may perform differently on different data features and tasks. By combining these different types of prediction models in a hybrid expert model, the advantages of each model can be utilized, and their respective advantages and disadvantages can be complemented, so as to further improve the stability and accuracy of the overall prediction, and make up for the limitations of a single model on a specific data set, thereby improving the accuracy and robustness of the converter oxygen consumption prediction, while reducing the risk of overfitting.

[0037] S2: Input prediction parameters to the expert part.

[0038] Among them, the prediction parameters include: molten iron weight, scrap steel weight, element content in molten iron and element content during converter auxiliary gun detection.

[0039] It should be noted that in actual production, the element content data of the converter sub-lance detection cannot be directly obtained before the start of blowing, because the sub-lance detection usually occurs near the end of blowing. Therefore, in actual production, in order to predict the oxygen consumption, it is necessary to use the expected target element content value instead of the actual element content value of the converter sub-lance detection.

[0040] In the present invention, by using the expected element content value in the model instead of the actual detection data, the present invention can improve the practicality and stability of the prediction model in actual production, enhance the robustness of the model, optimize the prediction results, reduce production costs, and improve production flexibility. This method effectively addresses the problem of data acquisition difficulties that may occur in actual production and provides a more reliable and stable solution for converter oxygen consumption prediction.

[0041] S3: According to the prediction parameters, the preliminary prediction value of the converter oxygen consumption is output through each converter oxygen consumption prediction model in the expert part.

[0042] In the present invention, in the expert part, a variety of prediction models (such as random forest, support vector machine, multi-layer perceptron and KAN network) are combined, and each model has different prediction capabilities and specialties. By integrating the prediction results of these models, the advantages of each model can be fully utilized to obtain more comprehensive and accurate prediction results. The preliminary prediction values ​​of multiple models can cover different characteristics and patterns of data, thereby improving the accuracy of the overall prediction. The complementarity between models makes the final prediction results more reliable. Different models have different sensitivities to data anomalies. Integrating the prediction results of multiple models can better handle abnormal situations in the data and improve the robustness of the model to noise.

[0043] S4: inputting the prediction parameters and the preliminary prediction value of the converter oxygen consumption output by each converter oxygen consumption prediction model into the gate control part.

[0044] In the present invention, by simultaneously inputting the prediction parameters and the preliminary prediction values ​​of each model, the gating part is able to receive more information. This enables the gating part to comprehensively consider the input features and the model prediction results, thereby making more accurate model selection and weighted decisions. The gating part assigns weights based on the prediction parameters and the preliminary prediction values ​​output by the model. By comprehensively considering this information, the gating part can assign weights more accurately, ensuring that more trust is given to the most appropriate model, thereby optimizing the final prediction results.

[0045] S5: According to the prediction parameters and the preliminary prediction values ​​of the converter oxygen consumption output by each converter oxygen consumption prediction model, a preset number of converter oxygen consumption prediction models are selected as target converter oxygen consumption prediction models through a gating part.

[0046] The gating part includes classifiers, and the preset number is K.

[0047] Optionally, the classifier comprises a random forest classifier.

[0048] In a possible implementation, S5 specifically includes sub-steps S501 to S502:

[0049] S501: Output the selection probability of each converter oxygen consumption prediction model through the classifier.

[0050] S502: Using the Top-K strategy, retain the K converter oxygen consumption prediction models with the greatest selection probability as target converter oxygen consumption prediction models.

[0051] Among them, the Top-K strategy is a commonly used technology in sorting or selection tasks. Its core idea is to select the K items with the highest probability or score from a set of prediction results. Specifically, for a set of model output probabilities or scores, the Top-K strategy will filter out the top K maximum values, retain the items corresponding to these maximum values, and set the values ​​of the remaining items to 0. This strategy helps to reduce the impact of low-probability or low-score items on the final results and improve the prediction accuracy and stability of the model.

[0052] In the present invention, the optimal K models are determined by the selection probability output by the classifier, and the model with the most predictive ability can be effectively selected. The prediction results of these models are considered more reliable, thereby improving the accuracy of the final prediction. The Top-K strategy can avoid the negative impact of low-probability or low-score models on the final prediction results. This helps to filter out those models with poor performance and reduce the interference of erroneous predictions on the final results. The Top-K strategy reduces the model set to K optimal models, making the model combination more streamlined and efficient. This can reduce the computational complexity while focusing on the most valuable model combination and improving computational efficiency.

[0053] S6: Calculate the final predicted value of the converter oxygen consumption according to the preliminary predicted value of the converter oxygen consumption of each target converter oxygen consumption prediction model.

[0054] In a possible implementation, S6 specifically includes sub-steps S601 to S603:

[0055] S601: retaining the selection probability of each target converter oxygen consumption prediction model, and setting the selection probability of the unselected converter oxygen consumption prediction model to 0.

[0056] In a possible implementation manner, S601 specifically includes:

[0057] According to the following formula, the selection probability of each target converter oxygen consumption prediction model is retained, and the selection probability of the unselected converter oxygen consumption prediction model is set to 0:

[0058]

[0059] Among them, q i represents the selection probability of each converter oxygen consumption prediction model after using the top-k strategy, p i represents the selection probability of the i-th converter oxygen consumption prediction model output by the classifier, and TopK( ) represents the Top-K strategy.

[0060] In the present invention, by setting the probability of the unselected models to 0, it is ensured that only the valid models selected by the Top-K strategy are used in the final calculation. This can ensure that the final prediction result is based on the most valuable and reliable model prediction. This strategy reduces the impact of low-probability models on the final prediction result, thereby improving the reliability and accuracy of the prediction.

[0061] S602: Normalizing the selection probability of each target converter oxygen consumption prediction model.

[0062] In a possible implementation manner, S602 specifically includes:

[0063] According to the following formula, the selection probability of each target converter oxygen consumption prediction model is normalized:

[0064]

[0065] Among them, G i represents the weight coefficient of each target converter oxygen consumption prediction model, and n represents the total number of target converter oxygen consumption prediction models.

[0066] In the present invention, the selection probability is normalized, and the selection probability of each target model can be converted into a weight coefficient, so that the sum of these weight coefficients is 1. This normalization process can ensure that the contribution ratio of different models is reasonable, which is conducive to accurate weighting. Through the normalized weight coefficients, the influence of each target model can be more accurately allocated, thereby improving the accuracy of the final prediction value.

[0067] S603: Using the normalized selection probability as the weight coefficient of each target converter oxygen consumption prediction model, the final predicted value of the converter oxygen consumption is calculated.

[0068] In a possible implementation manner, S603 specifically includes:

[0069] The final predicted value of converter oxygen consumption is calculated according to the following formula:

[0070]

[0071] in, It represents the final predicted value of converter oxygen consumption, E i represents the preliminary predicted value of converter oxygen consumption output by the i-th target converter oxygen consumption prediction model, and x represents the prediction parameter.

[0072] In the present invention, the normalized weight coefficients are used to perform weighted averaging on the preliminary prediction values ​​of the target model, and the prediction results of each model can be comprehensively considered. This weighted averaging method combines the prediction advantages of multiple models and can effectively improve the accuracy of the final prediction value. Through weighted averaging, the influence of any single model on the final prediction result is reduced, and the influence of model prediction error on the final result is avoided, thereby enhancing the stability of the prediction result.

[0073] In a possible implementation, the training method of the hybrid expert model specifically includes:

[0074] Get the sample dataset.

[0075] The sample dataset is divided into an expert training dataset and a gated training dataset.

[0076] The expert part is trained through the expert training dataset.

[0077] The gated part is trained through the gated training dataset.

[0078] It should be noted that, by training separately, first training the expert part and then training the gating part, it is achieved that the hybrid expert model can support converter oxygen consumption prediction models with different implementation methods.

[0079] Furthermore, dividing the sample data set into an expert training data set and a gating training data set can avoid overfitting of some converter oxygen consumption prediction models on the training set and causing the gating part to make wrong choices.

[0080] It should be noted that the gating part is implemented based on a supervised classifier, and the classification label used in classifier training is the converter oxygen consumption prediction model with the smallest error between the converter oxygen consumption prediction value and the actual converter oxygen consumption value.

[0081] Among them, the converter oxygen consumption prediction model with the smallest error between the converter oxygen consumption prediction value and the converter oxygen consumption true value is automatically generated after the expert part training is completed and before the gating part training starts.

[0082] In the present invention, by training the expert part and the gating part separately, the hybrid expert model can support converter oxygen consumption prediction models of different implementation modes, such as random forest, support vector machine, multi-layer perceptron and KAN network. This training method provides greater flexibility, so that the model can select and combine different expert models according to specific needs and data characteristics. Dividing the sample data set into an expert training data set and a gating training data set can effectively avoid overfitting of certain converter oxygen consumption prediction models on the training set. By training separately, it is ensured that the training processes of the expert model and the gating model are relatively independent, thereby reducing the risk of overfitting. By using the expert model with the smallest error as the classification label in the training of the gating part, it can be ensured that the gating part can select the optimal expert model. This method can more accurately identify and select those experts who perform best in actual predictions, thereby improving the accuracy of the final prediction results.

[0083] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0084] In the present invention, a hybrid expert model is constructed, which includes an expert part and a gating part. The expert part includes multiple converter oxygen consumption prediction models, and prediction parameters are first input into the expert part. Through each converter oxygen consumption prediction model in the expert part, a preliminary predicted value of converter oxygen consumption is output. Multiple experts can adapt to the reaction process under various operating conditions, avoiding the problem that a single mechanism or statistical model is difficult to apply to converters under different operating conditions, making the prediction result more accurate, and then the prediction parameters and the preliminary predicted value of converter oxygen consumption output by each converter oxygen consumption prediction model are input into the gating part, and a preset number of converter oxygen consumption prediction models are selected as target converter oxygen consumption prediction models, and then the final predicted value of converter oxygen consumption is calculated, which can simplify the reaction mechanism of converter steelmaking, solve the simplified assumption problem of linear models, and more comprehensively reflect the real reaction mechanism, so that the coefficients fitted by historical data have better interpretability.

[0085] Reference Manual Attached Figure 2 , showing a structural schematic diagram of a converter oxygen consumption prediction system based on an improved hybrid expert model provided by the present invention.

[0086] The present invention further provides a converter oxygen consumption prediction system 30 based on an improved hybrid expert model, comprising: a memory 303 and one or more processors 301 .

[0087] One or more application programs are stored in the memory 303 , and the one or more application programs are suitable for being executed by the one or more processors 301 to implement the method for predicting converter oxygen consumption based on the improved hybrid expert model described in the method embodiment.

[0088] The converter oxygen consumption prediction system 30 based on the improved hybrid expert model includes: a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, through a bus 302.

[0089] The structure of the converter oxygen consumption prediction system 30 based on the improved hybrid expert model does not constitute a limitation on the embodiment of the present invention.

[0090] Processor 301 may be a CPU, a general purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. Processor 301 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0091] The bus 302 may include a path to transmit information between the above components. The bus 302 may be a PCI bus or an EISA bus, etc. The bus 302 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0092] The memory 303 can be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disk storage (including a compressed optical disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0093] It should be noted that the converter oxygen consumption prediction system 30 based on the improved hybrid expert model can implement the above-mentioned converter oxygen consumption prediction method based on the improved hybrid expert model, and can achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0094] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0095] In the present invention, a hybrid expert model is constructed, which includes an expert part and a gating part. The expert part includes multiple converter oxygen consumption prediction models, and prediction parameters are first input into the expert part. Through each converter oxygen consumption prediction model in the expert part, a preliminary predicted value of converter oxygen consumption is output. Multiple experts can adapt to the reaction process under various operating conditions, avoiding the problem that a single mechanism or statistical model is difficult to apply to converters under different operating conditions, making the prediction result more accurate, and then the prediction parameters and the preliminary predicted value of converter oxygen consumption output by each converter oxygen consumption prediction model are input into the gating part, and a preset number of converter oxygen consumption prediction models are selected as target converter oxygen consumption prediction models, and then the final predicted value of converter oxygen consumption is calculated, which can simplify the reaction mechanism of converter steelmaking, solve the simplified assumption problem of linear models, and more comprehensively reflect the real reaction mechanism, so that the coefficients fitted by historical data have better interpretability.

[0096] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program can be loaded and executed by a processor to implement the method for predicting converter oxygen consumption based on an improved hybrid expert model as described in the first aspect.

[0097] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0098] There are a few points to note:

[0099] (1) The drawings of the embodiments of the present invention only involve structures related to the embodiments of the present invention. Other structures may refer to conventional designs.

[0100] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0101] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0102] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for predicting converter oxygen consumption based on an improved hybrid expert model, characterized in that: include: S1: constructing a hybrid expert model, wherein the hybrid expert model includes an expert part and a gating part, wherein the expert part includes a plurality of converter oxygen consumption prediction models; S2: inputting prediction parameters into the expert part; S3: outputting a preliminary predicted value of converter oxygen consumption according to the predicted parameters and using each converter oxygen consumption prediction model in the expert part; S4: inputting the prediction parameters and the preliminary prediction values ​​of converter oxygen consumption output by each converter oxygen consumption prediction model into the gate control part; S5: According to the prediction parameters and the preliminary predicted values ​​of converter oxygen consumption output by each converter oxygen consumption prediction model, a preset number of converter oxygen consumption prediction models are selected as target converter oxygen consumption prediction models through a gating part; S6: Calculate the final predicted value of the converter oxygen consumption according to the preliminary predicted value of the converter oxygen consumption of each target converter oxygen consumption prediction model.

2. The method for predicting converter oxygen consumption based on an improved hybrid expert model according to claim 1, characterized in that: The method for constructing the converter oxygen consumption prediction model includes: random forest, support vector machine, multi-layer perceptron and KAN network.

3. The method for predicting converter oxygen consumption based on an improved hybrid expert model according to claim 1, characterized in that: The prediction parameters include: molten iron weight, scrap steel weight, the content of each element in the molten iron, and the content of each element during converter auxiliary gun detection.

4. The method for predicting converter oxygen consumption based on an improved hybrid expert model according to claim 1, characterized in that: The gating part includes a classifier, the preset number is K, and S5 specifically includes: S501: Outputting the selection probability of each converter oxygen consumption prediction model through the classifier; S502: Using the Top-K strategy, retaining the K converter oxygen consumption prediction models with the greatest selection probabilities as target converter oxygen consumption prediction models.

5. The method for predicting converter oxygen consumption based on the improved hybrid expert model according to claim 4, characterized in that: The S6 specifically includes: S601: retaining the selection probability of each target converter oxygen consumption prediction model, and setting the selection probability of the unselected converter oxygen consumption prediction model to 0; S602: normalizing the selection probability of each target converter oxygen consumption prediction model; S603: Using the normalized selection probability as the weight coefficient of each target converter oxygen consumption prediction model, the final predicted value of the converter oxygen consumption is calculated.

6. The method for predicting converter oxygen consumption based on improved hybrid expert model according to claim 5, characterized in that: The S601 is specifically as follows: According to the following formula, the selection probability of each target converter oxygen consumption prediction model is retained, and the selection probability of the unselected converter oxygen consumption prediction model is set to 0: ； Among them, q i represents the selection probability of each converter oxygen consumption prediction model after using the top-k strategy, p i represents the selection probability of the i-th converter oxygen consumption prediction model output by the classifier, and TopK( ) represents the Top-K strategy.

7. The method for predicting converter oxygen consumption based on improved hybrid expert model according to claim 5, characterized in that: The S602 is specifically as follows: The selection probability of each target converter oxygen consumption prediction model is normalized according to the following formula: ； Among them, G i represents the weight coefficient of each target converter oxygen consumption prediction model, and n represents the total number of target converter oxygen consumption prediction models.

8. The method for predicting converter oxygen consumption based on improved hybrid expert model according to claim 5, characterized in that: The S603 is specifically as follows: The final predicted value of converter oxygen consumption is calculated according to the following formula: ； in, It represents the final predicted value of converter oxygen consumption, E i represents the preliminary predicted value of converter oxygen consumption output by the i-th target converter oxygen consumption prediction model, and x represents the prediction parameter.

9. The method for predicting converter oxygen consumption based on an improved hybrid expert model according to claim 1, characterized in that: The training method of the hybrid expert model specifically includes: Get a sample dataset; Dividing the sample data set into an expert training data set and a gated training data set; Training the expert part using the expert training data set; The gated part is trained using the gated training data set.

10. A converter oxygen consumption prediction system based on an improved hybrid expert model, characterized in that: include: memory and one or more processors; One or more application programs are stored in the memory, and the one or more application programs are suitable for being executed by the one or more processors to implement the method for predicting converter oxygen consumption based on the improved hybrid expert model as described in any one of claims 1 to 9.

Images