Electric furnace waste steel standard fingerprint spectrum generation method and system based on artificial intelligence

Abstract

The invention belongs to the technical field of electric furnace waste steel production, and discloses an electric furnace waste steel standard fingerprint spectrum generation method and system based on artificial intelligence. The method comprises the following steps: constructing a mechanism model, a key feature extraction model, a fingerprint generation model and a fingerprint analysis model; acquiring real-time production data, and writing the real-time production data into the mechanism model to obtain real-time mechanism dynamic data; performing key feature extraction on the real-time mechanism dynamic data by using a key feature extraction model; performing fingerprint spectrum generation on the plurality of real-time key features by using a fingerprint spectrum generation model; performing fingerprint analysis on the real-time fingerprint by using a fingerprint analysis model; and if the real-time fingerprint spectrum analysis result is abnormal, generating a real-time abnormal alarm. The problems of insufficient data comprehensiveness, limited data processing capability, low fingerprint spectrum generation quality and weak anomaly detection capability in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric furnace scrap steel production, and particularly relates to a method and system for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence. Background Art

[0002] The production of electric furnace scrap steel is an important link in the iron and steel industry, mainly involving the process of melting scrap steel into molten steel using an electric furnace. The fingerprint spectrum of electric furnace scrap steel is a data set that characterizes the composition, properties, and key features during the melting process. It is similar to the uniqueness of human fingerprints and can be used to identify and distinguish different scrap steel materials or production batches. The fingerprint spectrum is a comprehensive data characterization tool that provides strong support for the monitoring, optimization, and quality control of the scrap steel melting process.

[0003] The existing technologies for generating the standard fingerprint spectrum of electric furnace scrap steel have the following defects: 1) Insufficient data comprehensiveness: The existing technologies fail to consider the importance of production data including chemical composition, physical properties, and thermodynamic parameters during the production process of electric furnace scrap steel for the monitoring, optimization, and quality control of the scrap steel melting process, resulting in incomplete information in the fingerprint spectrum; 2) Limited data processing ability: Traditional data processing methods may not be able to effectively process a large amount of high-dimensional data, resulting in inaccurate generation of the standard fingerprint spectrum and low data processing efficiency; 3) Low quality of fingerprint spectrum generation: The existing technologies lack the analysis of key information in the production of electric furnace scrap steel and cannot accurately predict or reflect the complex changes during the scrap steel melting process, thus affecting the generation quality of the fingerprint spectrum; 4) Weak anomaly detection ability: The existing technologies lack the intelligent ability to generate and analyze the standard fingerprint spectrum and are insufficient in detecting abnormal situations during the production process, and cannot timely identify and respond to potential production problems. Summary of the Invention

[0004] In order to solve the problems of insufficient data comprehensiveness, limited data processing ability, low quality of fingerprint spectrum generation, and weak anomaly detection ability existing in the existing technologies, the purpose of the present invention is to provide a method and system for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence.

[0005] The technical solution adopted by the present invention is as follows: A method for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence, comprising the following steps: Construct a mechanism model for the production of electric furnace scrap steel, and based on the mechanism model, use artificial intelligence algorithms to construct a key feature extraction model, a fingerprint spectrum generation model, and a fingerprint spectrum analysis model; Collect the real-time production data during the production process of electric furnace scrap steel, and write the real-time production data into the mechanism model to obtain real-time mechanism dynamic data; Use the key feature extraction model to extract the key features from the real-time mechanism dynamic data, and obtain a number of corresponding real-time key features; Use the fingerprint spectrum generation model to generate the fingerprint spectrum from a number of real-time key features, and obtain the real-time fingerprint spectrum; Use the fingerprint spectrum analysis model to analyze the real-time fingerprint spectrum, and obtain the real-time fingerprint spectrum analysis result; If the real-time fingerprint spectrum analysis result shows an anomaly, generate a real-time anomaly alarm and return to the real-time production data collection step. Otherwise, directly return to the real-time production data collection step.

[0006] Furthermore, build a mechanism model for the production of electric furnace scrap steel, and based on the mechanism model, use artificial intelligence algorithms to build a key feature extraction model, a fingerprint spectrum generation model, and a fingerprint spectrum analysis model, including the following steps: Analyze the physical and chemical processes of the production of electric furnace scrap steel, set the mechanism parameters, and build a mechanism model for the production of electric furnace scrap steel according to the physical and chemical processes and the mechanism parameters; Collect a number of historical production data during the production process of electric furnace scrap steel, and preprocess the number of historical production data to obtain a number of preprocessed historical production data; Write the preprocessed historical production data into the mechanism model, and perform dynamic data extraction on the mechanism model written with the preprocessed historical production data to obtain a number of historical mechanism dynamic data; According to a number of historical mechanism dynamic data, use the integrated machine learning algorithm to build a key feature extraction model, and obtain a number of historical key features of each historical mechanism dynamic data; According to the number of historical key features of all historical mechanism dynamic data, use the deep learning algorithm to build a fingerprint spectrum generation model, and generate a number of historical fingerprint spectra; According to a number of historical fingerprint spectra, use the deep learning algorithm to build a fingerprint spectrum analysis model.

[0007] Furthermore, the key feature extraction model is built based on the RF algorithm, and the key feature extraction model includes a feature key score selection module and a key feature extraction module connected in sequence. The feature key score selection module is provided with a number of integrated CART trees; The fingerprint spectrum generation model is built based on the cGAN-MLP algorithm, and the fingerprint spectrum generation model includes a conditional information embedder and a conditional information processor both built based on the MLP algorithm, and a generator and a discriminator both built based on the RNN algorithm. The generator is respectively connected to the discriminator and the conditional information embedder, and the discriminator is connected to the conditional information processor; The fingerprint spectrum analysis model is constructed based on the GCN-DBN algorithm, and the fingerprint spectrum analysis model includes a graph structure feature extraction module constructed based on the GCN algorithm and a fingerprint spectrum analysis module constructed based on the DBN algorithm that are connected in sequence.

[0008] Furthermore, according to a number of historical mechanism dynamic data, using an integrated machine learning algorithm, a key feature extraction model is constructed, and a number of historical key features of each historical mechanism dynamic data are obtained, including the following steps: Using the RF algorithm, an initial key feature extraction model is constructed; the initial key feature extraction model includes an initial feature key score selection module and an initial key feature extraction module; According to a number of historical mechanism dynamic data, the initial feature key score selection module is trained to obtain a final feature key score selection module, and a number of feature key scores and historical dynamic data features of each historical mechanism dynamic data are obtained; Based on a number of feature key scores, according to a number of historical mechanism dynamic data, the initial key feature extraction module is trained to obtain a final key feature extraction module, and a number of historical key features of each historical mechanism dynamic data are obtained.

[0009] Furthermore, according to a number of historical key features of all historical mechanism dynamic data, using a deep learning algorithm, a fingerprint spectrum generation model is constructed, and a number of historical fingerprint spectra are generated, including the following steps: Using the cGAN-MLP algorithm, an initial fingerprint spectrum generation model is constructed; the initial fingerprint spectrum generation model includes an initial generator and an initial discriminator; A corresponding real fingerprint spectrum is set for each historical mechanism dynamic data, and a comprehensive loss function is constructed by combining the first loss function of the generator and the second loss function of the discriminator; Using the conditional information embedder of the initial generator, conditional information embedding is performed on a number of historical key features and random noise of each historical mechanism dynamic data to obtain a number of historical conditional information embedding features; According to a number of historical conditional information embedding features, the initial generator is trained to obtain an optimized generator, and a number of historical fingerprint spectra are generated; Using the conditional information processor of the initial discriminator, conditional information processing is performed on a number of historical key features of each historical mechanism dynamic data to obtain a number of historical conditional information processing features; According to the real fingerprint spectrum, historical fingerprint spectrum, and historical conditional information processing features of each historical mechanism dynamic data, the initial discriminator is trained to obtain an optimized discriminator, and a number of historical authenticity discrimination results are generated; According to the historical fingerprint spectrum generated by the generator and the historical authenticity discrimination result generated by the discriminator, using the comprehensive loss function, the historical loss value during the training process is obtained; If the historical loss value is less than the loss value threshold, then combining the optimized generator and the optimized discriminator, the final fingerprint spectrum generation model is obtained.

[0010] Furthermore, according to a number of historical fingerprint spectra, using a deep learning algorithm, a fingerprint spectrum analysis model is constructed, including the following steps: Using the GCN-DBN algorithm, an initial fingerprint spectrum analysis model is constructed; the initial fingerprint spectrum analysis model includes an initial graph structure feature extraction module and an initial fingerprint spectrum analysis module; According to a number of historical fingerprint spectra, the initial graph structure feature extraction module is trained to obtain the final graph structure feature extraction module, and a number of historical graph structure features are generated; According to a number of historical graph structure features, the initial fingerprint spectrum analysis module is trained to obtain the final fingerprint spectrum analysis module; Integrating the final graph structure feature extraction module and the final fingerprint spectrum analysis module, the final fingerprint spectrum analysis model is obtained.

[0011] Furthermore, using the key feature extraction model, key features are extracted from the real-time mechanism dynamic data to obtain the corresponding number of real-time key features, including the following steps: Preprocess the real-time mechanism dynamic data to obtain the preprocessed real-time mechanism dynamic data, and input the preprocessed real-time mechanism dynamic data into the key feature extraction model; Using the feature key scoring selection module of the key feature extraction model, extract the real-time dynamic data features of the real-time mechanism dynamic data; Using the key feature extraction module of the key feature extraction model, perform key feature extraction on the real-time dynamic data features to obtain the corresponding number of real-time key features.

[0012] Furthermore, using the fingerprint spectrum generation model, fingerprint spectra are generated for a number of real-time key features to obtain real-time fingerprint spectra, including the following steps: Input a number of real-time key features into the fingerprint spectrum generation model; Using the conditional information embedder of the fingerprint spectrum generation model, perform conditional information embedding on a number of real-time key features to obtain real-time conditional information embedding features; Using the generator of the fingerprint spectrum generation model, perform fingerprint spectrum generation on the real-time conditional information embedding features to obtain real-time fingerprint spectra.

[0013] Further, using a fingerprint spectrum analysis model to perform fingerprint spectrum analysis on the real-time fingerprint spectrum to obtain the real-time fingerprint spectrum analysis result, which includes the following steps: Input the real-time fingerprint spectrum into the fingerprint spectrum analysis model; Use the graph structure feature extraction module of the fingerprint spectrum analysis model to extract the real-time graph structure features of the real-time fingerprint spectrum; Use the fingerprint spectrum analysis module of the fingerprint spectrum analysis model to perform fingerprint spectrum analysis on the real-time graph structure features to obtain the real-time fingerprint spectrum analysis result.

[0014] An artificial intelligence-based system for generating standard fingerprint spectra of electric furnace scrap steel, which is used to implement the method for generating standard fingerprint spectra of electric furnace scrap steel. The system includes a model construction unit, a production data writing unit, a key feature extraction unit, a fingerprint spectrum generation unit, a fingerprint spectrum analysis unit, and an abnormal alarm generation unit that are connected in sequence.

[0015] The beneficial effects of the present invention are as follows: The artificial intelligence-based method and system for generating standard fingerprint spectra of electric furnace scrap steel provided by the present invention consider the importance of production data for the monitoring, optimization, and quality control of the scrap steel melting process, collect production data such as chemical composition, physical properties, and thermodynamic parameters as the data basis for generating fingerprint spectra, improve data comprehensiveness, and ensure the integrity of fingerprint spectrum information; utilize advanced artificial intelligence algorithms with powerful data processing capabilities, which can efficiently process a large amount of high-dimensional data, improve the accuracy and efficiency of generating standard fingerprint spectra; through the combination of artificial intelligence models and mechanism models, analyze the key information in the production of electric furnace scrap steel, improve the prediction accuracy of the model, accurately predict or reflect the complex changes in the scrap steel melting process, and improve the quality of generating fingerprint spectra; construct a key feature extraction model, a fingerprint spectrum generation model, and a fingerprint spectrum analysis model, realize systematic key feature extraction, fingerprint spectrum generation, and fingerprint spectrum analysis, effectively detect abnormal situations in the production process, timely identify and respond to potential problems, reduce manual operations, and reduce operation complexity and error rates.

[0016] Other beneficial effects of the present invention will be further described in the specific implementation manner. Description of the Drawings

[0017] Figure 1 is a flowchart of the artificial intelligence-based method for generating standard fingerprint spectra of electric furnace scrap steel in the present invention.

[0018] Figure 2 is a structural block diagram of the artificial intelligence-based system for generating standard fingerprint spectra of electric furnace scrap steel in the present invention. Specific Embodiment

[0019] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0020] Embodiment 1: As Figure 1 shown, this embodiment provides a method for generating a standard fingerprint spectrum of electric furnace scrap based on artificial intelligence, including the following steps: S1: Construct a mechanism model for the production of electric furnace scrap, and based on the mechanism model, use artificial intelligence algorithms to construct a key feature extraction model, a fingerprint spectrum generation model, and a fingerprint spectrum analysis model, including the following steps: S1-1: Analyze the physical and chemical processes of the production of electric furnace scrap, set mechanism parameters, and construct a mechanism model for the production of electric furnace scrap according to the physical and chemical processes and mechanism parameters; by studying the physical and chemical changes in the melting process of electric furnace scrap, determine the key parameters affecting these processes, and establish a mathematical model to describe the relationship between these parameters; The physical and chemical processes include heat transfer process, mass transfer process, chemical reaction process, and molten bath dynamics process; The mechanism model is based on the understanding of the physical, chemical, and thermodynamic principles of the melting process of electric furnace scrap, and describes and simulates these processes through mathematical equations, which is used to understand and predict the behavior in the production process, and provides data support for the subsequent artificial intelligence model. The constructed mechanism model includes the following models: Heat transfer model: Arc heating: Describes the heat exchange process between the arc and the scrap; Convection and radiation: The heat in the molten bath of the scrap is transferred to other parts of the scrap and the furnace wall through convection and radiation; Heat conduction: The heat conduction in the scrap and the furnace lining material; Mass transfer model: Mass transfer in the gas phase and liquid phase: Includes the dissolution and reaction of gases such as oxygen, nitrogen, and carbon dioxide in the molten bath; Oxidation and reduction reactions of elements in the scrap; Chemical reaction model: Oxidation reactions of elements such as carbon, silicon, manganese, phosphorus, and sulfur; Removal processes of impurity elements, such as desulfurization and dephosphorization; Burnout and recovery rates of alloy elements; Molten bath dynamics model: Formation, development, and shape change of the molten bath; Flow characteristics in the molten bath, including eddy currents, circulation, and stirring effects; Arc model: Arc stability, shape, and temperature distribution; Influence of the arc on the heating of the scrap; Electrode consumption model: Melting and consumption rate of the electrode under the action of the arc; Thermodynamic model: Prediction of the melting point of the scrap; Thermodynamic behavior of phase changes (such as solid-liquid phase changes); Electric furnace operation parameter model: Influence of power supply parameters (such as current, voltage, power) on the melting process; Influence of operation parameters such as feeding strategy, melting time, and electrode adjustment; Interaction model of scrap steel and slag: Formation, properties and functions of slag; Chemical reactions and mass transfer between slag and scrap steel; S1-2: Collect a number of historical production data during the production of EAF scrap steel, and preprocess the number of historical production data to obtain a number of preprocessed historical production data; The production data includes the following components: Arc current, voltage, and power data: These data reflect the energy input of the electric furnace and the stability of the arc; Melting time data: The time required for scrap steel to melt, which affects production efficiency; Temperature data: The molten bath temperature and the furnace lining temperature, which are crucial for the melting process; Scrap steel composition data: The chemical composition of scrap steel, which affects the melting process and the quality of the final product; Oxygen flow rate data: The role of oxygen in the melting process, such as promoting the oxidation of scrap steel; Slag composition data: The chemical composition of slag, which affects the impurity removal and melting effect of scrap steel; The preprocessing includes data cleaning, denoising, and standard processing to ensure data quality, eliminate noise and outliers, and improve the efficiency and accuracy of model training; S1-3: Write the preprocessed historical production data into the mechanism model, and perform dynamic data extraction on the mechanism model written with the preprocessed historical production data to obtain a number of historical mechanism dynamic data; S1-4: According to a number of historical mechanism dynamic data, use an integrated machine learning algorithm to construct a key feature extraction model, and obtain a number of historical key features for each historical mechanism dynamic data; The key feature extraction model is constructed based on the Random Forest (RF) algorithm, and the key feature extraction model includes a feature key score selection module and a key feature extraction module connected in sequence. The feature key score selection module is provided with a number of integrated Classification And Regression Tree (CART) trees; The CART trees of the feature key score selection module screen the features of the mechanism dynamic data to obtain the key score of each feature. The key score is used to evaluate the key degree of the feature. The key feature extraction module performs key feature extraction according to the feature key score, which can process a large number of features and select the most stable and most discriminative key features; reduce the data dimension, and improve the interpretability and prediction ability of the model; The key features include: Arc stability features (such as fluctuations in current and voltage); Distribution and change features of molten bath temperature; Scrap steel melting rate features; Slag formation and property features; Oxygen consumption features; Electrode consumption rate features; According to a number of historical mechanism dynamic data, an integrated machine learning algorithm is used to construct a key feature extraction model, and a number of historical key features of each historical mechanism dynamic data are obtained, including the following steps: S1-4-1: Use the RF algorithm to construct an initial key feature extraction model; the initial key feature extraction model includes an initial feature key score selection module and an initial key feature extraction module; S1-4-2: According to a number of historical mechanism dynamic data, train the initial feature key score selection module to obtain a final feature key score selection module, and obtain a number of feature key scores and historical dynamic data features of each historical mechanism dynamic data; Including the following steps: S1-4-2-1: The initial feature key score selection module extracts the historical dynamic data features of the historical mechanism dynamic data, divides the historical dynamic data features into M alternative features, and obtains the feature contribution degrees of the M alternative features; The formula is: In the formula, is the feature contribution degree of the j'th alternative feature; is the feature contribution degree of the j'th alternative feature in the i'th tree of the random forest; i' is the CART tree indicator; j' is the alternative feature indicator; n' is the total number of CARTs; In the formula, is the CART tree node of the random forest m 、node i' and the Gini index of node r"; is the CART tree node m The proportion of class k" in; k' is the total number of classes; is the node indicator; k" is the class indicator; S1-4-2-2: Normalize the feature contribution degrees of the M alternative features to obtain corresponding normalized feature contribution degrees; The formula is: In the formula, is the normalized feature contribution degree; J is the total number of alternative features; S1-4-2-3: Generate feature selection standard values for a number of alternative features according to the normalized feature contribution degrees; The formula is: In the formula, is the feature selection standard value of the j'th alternative feature; is the feature contribution degree after the normalization process of the alternative feature; is the alternative feature indicator; S1-4-2-4: Obtain several key feature scores according to the feature selection standard value as the key feature scores; S1-4-2-5: Traverse all historical mechanism dynamic data, train the initial key feature score selection module, and obtain the final key feature score selection module; S1-4-3: Based on several key feature scores and several historical mechanism dynamic data, train the initial key feature extraction module to obtain the final key feature extraction module, and obtain several historical key features of each historical mechanism dynamic data; By extracting key features, the model can reduce the influence of noise and irrelevant features, improve the accuracy and efficiency of prediction; Selecting fewer key features helps to reduce the complexity of the model, thereby reducing the risk of overfitting and improving the generalization ability of the model; S1-5: According to several historical key features of all historical mechanism dynamic data, use a deep learning algorithm to construct a fingerprint map generation model and generate several historical fingerprint maps; The fingerprint map generation model is constructed based on the conditional generative adversarial network (cGAN, Conditional Generative Adversarial Network)-multilayer perceptron (MLP, Multilayer Perceptron) algorithm, and the fingerprint map generation model includes a conditional information embedder and a conditional information processor both constructed based on the MLP algorithm, and a generator and a discriminator both constructed based on the recurrent neural network (RNN, Recurrent Neural Network) algorithm. The generator is respectively connected to the discriminator and the conditional information embedder, and the discriminator is connected to the conditional information processor; The conditional information embedding module is used to process the unsequenced conditional information of the key features and random noise to obtain the conditional information embedding features in sequence format, and realize the integration of conditional information into the generation process; The generator processes the conditional information embedding features output by the conditional information embedding module to generate a fingerprint map; The conditional information processing module is used to process the additional conditional information of the key features to help the discriminator more accurately judge the authenticity of the fingerprint map; The discriminator analyzes the generated fingerprint map and the real fingerprint map to judge whether the generated fingerprint map is real and conforms to the given conditional information; The generator and the discriminator compete with each other through an adversarial training process. The generator tries to generate a fingerprint map that can deceive the discriminator, while the discriminator tries to better identify the real and generated fingerprint maps; According to several historical key features of all historical mechanism dynamic data, use deep learning algorithms to construct a fingerprint spectrum generation model and generate several historical fingerprint spectra, including the following steps: S1-5-1: Use the cGAN-MLP algorithm to construct an initial fingerprint spectrum generation model; the initial fingerprint spectrum generation model includes an initial generator and an initial discriminator; S1-5-2: Set corresponding real fingerprint spectra for each historical mechanism dynamic data, and combine the first loss function of the generator and the second loss function of the discriminator to construct a comprehensive loss function; provide the goal of training the generator and the discriminator to ensure that the generator can generate fingerprints close to the real ones. S1-5-3: Use the conditional information embedder of the initial generator to perform conditional information embedding on several historical key features and random noise of each historical mechanism dynamic data to obtain several historical conditional information embedded features; ensure that the generator considers specific conditions of the input data when generating fingerprint spectra. S1-5-4: Train the initial generator according to several historical conditional information embedded features to obtain an optimized generator and generate several historical fingerprint spectra; the generator gradually learns to generate historical fingerprint spectra similar to the real fingerprint spectra. S1-5-5: Use the conditional information processor of the initial discriminator to perform conditional information processing on several historical key features of each historical mechanism dynamic data to obtain several historical conditional information processed features; improve the discriminator's ability to distinguish between real and generated data. S1-5-6: Train the initial discriminator according to the real fingerprint spectra, historical fingerprint spectra and historical conditional information processed features of each historical mechanism dynamic data to obtain an optimized discriminator and generate several historical authenticity discrimination results; the discriminator learns to identify the differences between the fingerprint spectra generated by the generator and the real fingerprint spectra. S1-5-7: According to the historical fingerprint spectra generated by the generator and the historical authenticity discrimination results generated by the discriminator, use the comprehensive loss function to obtain the historical loss value during the training process; S1-5-8: If the historical loss value is less than the loss value threshold, combine the optimized generator and the optimized discriminator to obtain the final fingerprint spectrum generation model; S1-6: According to several historical fingerprint spectra, use deep learning algorithms to construct a fingerprint spectrum analysis model; The fingerprint spectrum analysis model is constructed based on the Graph Convolutional Network (GCN)-Deep Belief Nets (DBN) algorithm, and the fingerprint spectrum analysis model includes a graph structure feature extraction module constructed based on the GCN algorithm and a fingerprint spectrum analysis module constructed based on the DBN algorithm, which are connected in sequence; The GCN network performs feature propagation on the fingerprint spectrum through operations similar to convolution, extracts node features and edge features between entities, and constitutes the graph structure features of the fingerprint spectrum; the DBN network performs label prediction based on the graph structure features; According to a number of historical fingerprint spectra, use deep learning algorithms to construct a fingerprint spectrum analysis model, including the following steps: S1-6-1: Use the GCN-DBN algorithm to construct an initial fingerprint spectrum analysis model; the initial fingerprint spectrum analysis model includes an initial graph structure feature extraction module and an initial fingerprint spectrum analysis module; S1-6-2: According to a number of historical fingerprint spectra, train the initial graph structure feature extraction module to obtain a final graph structure feature extraction module, and generate a number of historical graph structure features; training the graph structure feature extraction module effectively extracts key features from the graph structure of the fingerprint spectrum, which is crucial for understanding the internal relationship of the data; S1-6-3: According to a number of historical graph structure features, train the initial fingerprint spectrum analysis module to obtain a final fingerprint spectrum analysis module; training the fingerprint spectrum analysis module learns the deep features in the fingerprint spectrum, and these features may be very crucial for distinguishing normal and abnormal patterns; S1-6-4: Integrate the final graph structure feature extraction module and the final fingerprint spectrum analysis module to obtain a final fingerprint spectrum analysis model; S2: Collect real-time production data during the production process of electric furnace scrap steel, and write the real-time production data into the mechanism model to obtain real-time mechanism dynamic data; S3: Use the key feature extraction model to extract key features from the real-time mechanism dynamic data to obtain corresponding real-time key features, including the following steps: S3-1: Preprocess the real-time mechanism dynamic data to obtain preprocessed real-time mechanism dynamic data, and input the preprocessed real-time mechanism dynamic data into the key feature extraction model; S3-2: Use the feature key scoring selection module of the key feature extraction model to extract real-time dynamic data features of the real-time mechanism dynamic data; S3-3: Use the key feature extraction module of the key feature extraction model to extract key features from the real-time dynamic data features to obtain corresponding real-time key features; S4: Use the fingerprint spectrum generation model to generate fingerprint spectra for a number of real-time key features to obtain real-time fingerprint spectra, including the following steps: S4-1: Input a number of real-time key features into the fingerprint spectrum generation model; S4-2: Use the conditional information embedder of the fingerprint spectrum generation model to embed conditional information into a number of real-time key features to obtain real-time conditional information embedded features; S4-3: Use the generator of the fingerprint spectrum generation model to generate fingerprint spectra for the real-time conditional information embedded features to obtain real-time fingerprint spectra; S5: Use the fingerprint spectrum analysis model to analyze the real-time fingerprint spectra to obtain real-time fingerprint spectrum analysis results, including the following steps: S5-1: Input the real-time fingerprint spectra into the fingerprint spectrum analysis model; S5-2: Use the graph structure feature extraction module of the fingerprint spectrum analysis model to extract the real-time graph structure features of the real-time fingerprint spectra; S5-3: Use the fingerprint spectrum analysis module of the fingerprint spectrum analysis model to analyze the real-time graph structure features to obtain real-time fingerprint spectrum analysis results; The fingerprint spectrum analysis results include component analysis results (element composition: identifying the types and proportions of elements contained in the scrap steel; impurity content: detecting the types and contents of impurities in the scrap steel), quality assessment results (scrap steel grade: grading the scrap steel according to the fingerprint spectrum analysis results; quality consistency: evaluating the quality stability of different batches of scrap steel), and anomaly detection results (production anomalies: detecting anomalies in the production process, such as temperature fluctuations, composition deviations, etc.; process stability: analyzing the stability and trend of the production process); S6: If the real-time fingerprint spectrum analysis result indicates an anomaly, generate a real-time anomaly alarm and return to the production data real-time acquisition step; otherwise, directly return to the production data real-time acquisition step; By analyzing the fingerprint spectra, the model can detect anomalies in the electric furnace scrap steel production process, take timely measures, and reduce losses.

[0021] Example 2: As Figure 2 shown, this embodiment provides an artificial intelligence-based electric furnace scrap steel standard fingerprint spectrum generation system for implementing the electric furnace scrap steel standard fingerprint spectrum generation method. The system includes a model construction unit, a production data writing unit, a key feature extraction unit, a fingerprint spectrum generation unit, a fingerprint spectrum analysis unit, and an anomaly alarm generation unit that are connected in sequence; A model construction unit for constructing a mechanism model of electric furnace scrap production, and based on the mechanism model, using artificial intelligence algorithms to construct a key feature extraction model, a fingerprint spectrum generation model, and a fingerprint spectrum analysis model; A production data writing unit for collecting real-time production data during the production process of electric furnace scrap and writing the real-time production data into the mechanism model to obtain real-time mechanism dynamic data; A key feature extraction unit for using the key feature extraction model to extract key features from the real-time mechanism dynamic data to obtain a corresponding number of real-time key features; A fingerprint spectrum generation unit for using the fingerprint spectrum generation model to generate a fingerprint spectrum from a number of real-time key features to obtain a real-time fingerprint spectrum; A fingerprint spectrum analysis unit for using the fingerprint spectrum analysis model to analyze the real-time fingerprint spectrum to obtain a real-time fingerprint spectrum analysis result; An abnormal alarm generation unit for generating a real-time abnormal alarm when the real-time fingerprint spectrum analysis result indicates an abnormality, and returning to the real-time production data collection step.

[0022] A method and system for generating a standard fingerprint spectrum of electric furnace scrap based on artificial intelligence provided by the present invention, considering the importance of production data for the monitoring, optimization, and quality control of the scrap melting process, collecting production data such as chemical composition, physical properties, and thermodynamic parameters as the data basis for fingerprint spectrum generation, improving data comprehensiveness, and ensuring the integrity of fingerprint spectrum information; using advanced artificial intelligence algorithms with powerful data processing capabilities, capable of efficiently processing a large amount of high-dimensional data, improving the accuracy and efficiency of standard fingerprint spectrum generation; through the combination of artificial intelligence models and mechanism models, analyzing the key information of electric furnace scrap production, improving the prediction accuracy of the model, accurately predicting or reflecting the complex changes during the scrap melting process, and improving the quality of fingerprint spectrum generation; constructing a key feature extraction model, a fingerprint spectrum generation model, and a fingerprint spectrum analysis model, realizing systematic key feature extraction, fingerprint spectrum generation, and fingerprint spectrum analysis, effectively detecting abnormal situations in the production process, timely identifying and responding to potential problems, reducing manual operations, and reducing operation complexity and error rates.

[0023] The present invention is not limited to the above optional embodiments, and anyone can obtain other various forms of products under the inspiration of the present invention. The above specific embodiments should not be construed as limiting the protection scope of the present invention, and the protection scope of the present invention should be defined by the claims, and the description can be used to interpret the claims.

Claims

1. A method for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence, characterized in that: The steps include: Construct a mechanism model for scrap steel production in electric furnaces, and based on the mechanism model, use artificial intelligence algorithms to construct key feature extraction models, fingerprint generation models, and fingerprint analysis models; Collect real-time production data during the scrap steel production process of the electric furnace, and write the real-time production data into the mechanism model to obtain real-time mechanism dynamic data; Use the key feature extraction model to extract key features from real-time mechanism dynamic data and obtain several corresponding real-time key features; Use the fingerprint generation model to generate fingerprints for several real-time key features to obtain real-time fingerprints; Using the fingerprint analysis model, the real-time fingerprint analysis is performed to obtain the real-time fingerprint analysis result; If the real-time fingerprint analysis result shows that there is an abnormality, a real-time abnormality alarm is generated and the process returns to the real-time production data collection step; otherwise, the process directly returns to the real-time production data collection step.

2. The method for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence according to claim 1, characterized in that: Construct a mechanism model for scrap steel production in an electric furnace, and based on the mechanism model, use artificial intelligence algorithms to construct a key feature extraction model, a fingerprint generation model, and a fingerprint analysis model, including the following steps: Analyze the physical and chemical process of scrap steel production in electric furnaces, set mechanism parameters, and construct a mechanism model of scrap steel production in electric furnaces based on the physical and chemical process and mechanism parameters; Collecting some historical production data in the process of scrap steel production in an electric furnace, and preprocessing some historical production data to obtain some preprocessed historical production data; Writing the preprocessed historical production data into the mechanism model, and performing dynamic data extraction on the mechanism model into which the preprocessed historical production data is written, to obtain a number of historical mechanism dynamic data; Based on several historical mechanism dynamic data, an integrated machine learning algorithm is used to build a key feature extraction model, and several historical key features of each historical mechanism dynamic data are obtained; Based on several historical key features of all historical mechanism dynamic data, a deep learning algorithm is used to build a fingerprint generation model and generate several historical fingerprint maps; Based on several historical fingerprint maps, a fingerprint analysis model is constructed using a deep learning algorithm.

3. The method for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence according to claim 2, characterized in that: The key feature extraction model is constructed based on the RF algorithm, and the key feature extraction model includes a feature key score selection module and a key feature extraction module connected in sequence, and the feature key score selection module is provided with a plurality of integrated CART trees; The fingerprint generation model is constructed based on the cGAN-MLP algorithm, and the fingerprint generation model includes a conditional information embedder and a conditional information processor both constructed based on the MLP algorithm, and a generator and a discriminator both constructed based on the RNN algorithm, the generator is connected to the discriminator and the conditional information embedder respectively, and the discriminator is connected to the conditional information processor; The fingerprint analysis model is constructed based on the GCN-DBN algorithm, and the fingerprint analysis model includes a graph structure feature extraction module constructed based on the GCN algorithm and a fingerprint analysis module constructed based on the DBN algorithm, which are connected in sequence.

4. The method for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence according to claim 3 is characterized in that: According to several historical mechanism dynamic data, an integrated machine learning algorithm is used to build a key feature extraction model, and several historical key features of each historical mechanism dynamic data are obtained, including the following steps: Using the RF algorithm, constructing an initial key feature extraction model; the initial key feature extraction model includes an initial feature key score selection module and an initial key feature extraction module; According to a number of historical mechanism dynamic data, the initial feature key score selection module is trained to obtain a final feature key score selection module, and a number of feature key scores and historical dynamic data features of each historical mechanism dynamic data are obtained; Based on several feature key scores and several historical mechanism dynamic data, the initial key feature extraction module is trained to obtain a final key feature extraction module, and several historical key features of each historical mechanism dynamic data are obtained.

5. The method for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence according to claim 3 is characterized in that: According to several historical key features of all historical mechanism dynamic data, a deep learning algorithm is used to build a fingerprint generation model and generate several historical fingerprint maps, including the following steps: Using the cGAN-MLP algorithm, an initial fingerprint generation model is constructed; the initial fingerprint generation model includes an initial generator and an initial discriminator; The real fingerprint map corresponding to each setting of historical mechanism dynamic data is constructed, and the comprehensive loss function is constructed by combining the first loss function of the generator and the second loss function of the discriminator; Using the conditional information embedder of the initial generator, conditional information is embedded into several historical key features and random noise of each historical mechanism dynamic data to obtain several historical conditional information embedding features; According to several historical condition information embedding features, the initial generator is trained to obtain an optimized generator and generate several historical fingerprint maps; Using the conditional information processor of the initial discriminator, conditional information processing is performed on a number of historical key features of each historical mechanism dynamic data to obtain a number of historical conditional information processing features; According to the real fingerprint map, historical fingerprint map and historical condition information processing characteristics of each historical mechanism dynamic data, the initial discriminator is trained to obtain an optimized discriminator and generate several historical authenticity discrimination results; According to the historical fingerprint map generated by the generator and the historical authenticity judgment result generated by the discriminator, a comprehensive loss function is used to obtain the historical loss value in the training process; If the historical loss value is less than the loss value threshold, the optimized generator and the optimized discriminator are combined to obtain the final fingerprint generation model.

6. The method for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence according to claim 3 is characterized in that: Based on several historical fingerprint maps, a fingerprint analysis model is constructed using a deep learning algorithm, including the following steps: Using the GCN-DBN algorithm, an initial fingerprint analysis model is constructed; the initial fingerprint analysis model includes an initial graph structure feature extraction module and an initial fingerprint analysis module; According to several historical fingerprint graphs, the initial graph structure feature extraction module is trained to obtain the final graph structure feature extraction module, and several historical graph structure features are generated; According to several historical graph structural features, the initial fingerprint spectrum analysis module is trained to obtain the final fingerprint spectrum analysis module; The final graph structure feature extraction module and the final fingerprint spectrum analysis module are integrated to obtain the final fingerprint spectrum analysis model.

7. The method for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence according to claim 4, characterized in that: Using the key feature extraction model, key features are extracted from real-time mechanism dynamic data to obtain several corresponding real-time key features, including the following steps: Preprocessing the real-time mechanism dynamic data to obtain preprocessed real-time mechanism dynamic data, and inputting the preprocessed real-time mechanism dynamic data into a key feature extraction model; Use the feature key score selection module of the key feature extraction model to extract the real-time dynamic data features of the real-time mechanism dynamic data; The key feature extraction module of the key feature extraction model is used to extract key features of real-time dynamic data features to obtain corresponding real-time key features.

8. The method for generating standard fingerprint of electric furnace scrap steel based on artificial intelligence according to claim 5, characterized in that: Using the fingerprint generation model, fingerprint generation is performed on several real-time key features to obtain a real-time fingerprint, including the following steps: Inputting several real-time key features into the fingerprint generation model; Use the conditional information embedder of the fingerprint spectrum generation model to embed conditional information into several real-time key features to obtain real-time conditional information embedding features; The generator of the fingerprint generation model is used to generate the fingerprint of the real-time condition information embedded features to obtain the real-time fingerprint.

9. The method for generating a standard fingerprint spectrum of electric furnace scrap steel based on artificial intelligence according to claim 6, characterized in that: Using the fingerprint analysis model, the real-time fingerprint analysis is performed on the real-time fingerprint to obtain the real-time fingerprint analysis result, including the following steps: Inputting the real-time fingerprint into the fingerprint analysis model; Use the graph structure feature extraction module of the fingerprint analysis model to extract the real-time graph structure features of the real-time fingerprint; The fingerprint analysis module of the fingerprint analysis model is used to perform fingerprint analysis on the real-time graph structure features to obtain real-time fingerprint analysis results.

10. An artificial intelligence-based standard fingerprint generation system for electric furnace scrap steel, used to implement the standard fingerprint generation method for electric furnace scrap steel as claimed in any one of claims 1 to 9, characterized in that: The system comprises a model building unit, a production data writing unit, a key feature extraction unit, a fingerprint spectrum generating unit, a fingerprint spectrum analyzing unit and an abnormal alarm generating unit which are connected in sequence.

Images