Hydrogen utilization rate online regulation method and system for hydrogen metallurgical process of electric arc furnace

Abstract

This invention provides an online control method and system for hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process, comprising the following steps: acquiring parameter data from the smelting area and flue gas pipeline of the electric arc furnace, and setting time-series labels for the parameter data; acquiring and processing image data within the electric arc furnace in real time, and simultaneously setting time-series labels for the image data to obtain boiling intensity, combustion stability, and material melting progress; generating the hydrogen utilization rate under the corresponding time series, and simultaneously predicting the hydrogen utilization rate under future time series; dynamically adjusting the hydrogen flow rate and injection mode based on the hydrogen utilization rate and the predicted hydrogen utilization rate value; after each smelting, offline correction of the hydrogen utilization rate calculation model based on the actual total hydrogen consumption, smelting cycle, and detected hydrogen content in the molten steel, and updating it to the online model library. This invention employs the above-mentioned online control method and system for hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process to achieve precise, stable, and efficient control of hydrogen utilization rate.

Description

Technical Field

[0001] This invention relates to the field of metallurgical control, and in particular to a method and system for online control of hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process. Background Technology

[0002] Electric arc furnace hydrogen metallurgy is a key technology that uses hydrogen to replace traditional carbon-based reducing agents, achieving significant carbon reduction or even zero carbon emissions in the steelmaking process. However, the utilization efficiency of hydrogen in an electric arc furnace is affected by various factors such as interference from the high-temperature arc light inside the furnace, dynamic changes in the molten pool reaction, and the adaptability of the injection mode and flow rate. As a result, problems such as low hydrogen utilization and drastic fluctuations are common.

[0003] Existing control methods largely rely on offline experience or single signal feedback, with major drawbacks including: a lack of real-time quantitative sensing of the boiling intensity of the molten pool, flame stability, and material melting progress within the furnace, making it difficult to accurately judge the reaction process of hydrogen reduction and combustion; feedback control based solely on the current hydrogen utilization rate results in a delayed response, failing to address the nonlinear and large-hysteresis characteristics of metallurgical reactions; and hydrogen flow rate regulation and injection mode adjustment are independent, lacking a coordination mechanism, easily leading to hydrogen waste or insufficient reduction. Therefore, there is an urgent need for an online control method capable of real-time sensing of the furnace state, predicting trends in hydrogen utilization rate changes, and achieving intelligent coordination between flow rate and injection mode. Summary of the Invention

[0004] The purpose of this invention is to provide an online control method and system for hydrogen utilization rate in the hydrogen metallurgical process of electric arc furnace. By constructing a complete technical system of "multi-source sensing, time-series prediction, dual-dimensional collaborative control, and closed-loop iteration", the invention achieves accurate, stable, and efficient control of hydrogen utilization rate.

[0005] To achieve the above objectives, the present invention provides a method for online control of hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process, comprising the following steps: Acquire parameter data from the electric arc furnace smelting area and flue gas pipeline, and set time-series labels for the parameter data, which includes flow rate, temperature, and composition data. The image data inside the electric arc furnace is acquired and processed in real time, and time-series labels are set for the image data to obtain boiling intensity, combustion stability and material melting progress. Based on parameter data and boiling intensity, combustion stability and material melting progress, the hydrogen utilization rate under the corresponding time series is generated, and the hydrogen utilization rate under the future time series is predicted. The hydrogen flow rate and injection mode are dynamically adjusted based on the hydrogen utilization rate under the corresponding time series and the predicted hydrogen utilization rate under the future time series. After each smelting process, the hydrogen utilization rate calculation model is corrected offline based on the actual total hydrogen consumption, smelting cycle, and the measured hydrogen content of the molten steel, and then updated to the online model library.

[0006] Preferably, image data of the electric arc furnace is acquired in real time, and time-series labels are set for the image data. The image data of the electric arc furnace is classified according to the acquisition location, and corresponding processing is performed on each type of image data, including: Composite filtering technology is used to suppress strong arc interference in the electric arc furnace. By passively acquiring the infrared / visible light signal radiated by the molten pool itself, a clear dynamic image of the molten pool is obtained. The outline of the molten pool in the dynamic image is extracted. Combined with the dynamic image data, the dynamic characteristics of the molten pool surface are analyzed in time series to reflect the boiling intensity. The dynamic characteristics include fluctuation frequency and bubble generation rate. The texture features of flame images are extracted, and images of the same location under different time conditions are compared to obtain the changing, invariant, and protruding parts of the texture features. The changing, invariant, and protruding parts are classified, and texture features of the same category are depicted to accurately delineate the boundaries between the flame and the molten pool and droplets. By tracking the changes in flame texture features under different time periods, the movement trajectory and velocity field of the flame are tracked to evaluate the combustion stability. Image data processing of the molten pool surface yields RGB values ​​and texture features of the molten pool surface. The temperature distribution of the molten pool surface is obtained based on the RGB values, and the melting depth and solid-liquid phase ratio are obtained based on the texture features of the molten pool surface, thus obtaining the material melting progress.

[0007] Preferably, the hydrogen utilization rate at the corresponding time sequence is generated based on parameter data and boiling intensity, combustion stability, and material melting progress, including the following steps: Align the parameter data with boiling intensity, combustion stability, and material melting progress in a unified time coordinate system; Then, based on the characteristic cycle of the hydrogen metallurgical reaction, it is divided into sliding time windows according to the time sequence; The hydrogen utilization rate within the corresponding window is calculated based on parameter data, boiling intensity, combustion stability, and material melting progress, i.e., the hydrogen utilization rate under the corresponding time series. ; In the formula, This represents the hydrogen utilization rate under the corresponding time series. For the conversion rate of the reduction reaction, The conversion rate of the combustion reaction is denoted as .

[0008] Preferably, a correction strategy is set to correct hydrogen utilization, specifically as follows: The emission loss rate was obtained from the hydrogen concentration and flue gas flow rate in the parameter data. Based on the hydrogen metallurgical reaction mechanism, hydrogen utilization includes , , , ; The image features and key parameters are input into the correction model, and the correction coefficient α is output to obtain the final hydrogen utilization rate: ; To correct the coefficients, under different time windows The values ​​of are different. For the final hydrogen utilization rate; contrast and Relationship, ensure .

[0009] Preferably, the prediction of hydrogen utilization rate in future time series includes: Extract parameter data, boiling intensity, combustion stability, material melting progress, and hydrogen utilization rate at the corresponding time series to extract time series derived features, and obtain the feature vector of each time series point. The time series derived features include differential features, interactive statistical features, and lag features. Extract local dependencies from temporally derived features and output the hidden state sequence. ; The self-attention mechanism is used to model the global dependency of time-series derived features. Parameter data, boiling intensity, combustion stability, material melting progress, and hydrogen utilization rate at the corresponding time series are encoded. The correlation weights of different time series points are calculated through multi-head self-attention, and the global feature representation is output. ; Weighting coefficients γ and β are dynamically generated based on the current smelting stage: Introducing a cross-attention mechanism at the feature level allows for the hidden state sequence to... With global feature representation The interaction is performed to calculate cross-modal attention weights, followed by weighted fusion, residual connection and layer normalization to obtain the predicted value of hydrogen utilization rate in future time series; Preferably, the dynamic control of hydrogen flow rate and injection mode includes: Hydrogen flow rate control: An adaptive PI controller combining prediction deviation and current status feedback is employed. When the current hydrogen utilization rate deviates from the set threshold and the prediction value shows a continuously increasing deviation, the hydrogen flow rate is quickly adjusted. In the short term, the deviation is corrected through a proportional-integral circuit, while in the medium to long term, the flow rate baseline is adjusted based on the predicted trend. At the same time, upper and lower limits for the flow rate and constraints on the rate of change are set to ensure equipment safety and reaction stability. Injection mode control: Based on hydrogen utilization status, predicted trends, and image characteristics, a rule-based control strategy is formulated. For example, when hydrogen utilization is below the threshold and molten pool boiling is weak, the vertical deep blowing mode is switched to enhance molten pool penetration; when hydrogen utilization is above the threshold and the flame is stable, the diffusion wide blowing mode is switched to reduce hydrogen waste; when hydrogen utilization fluctuates drastically and the flame is unstable, the pulse intermittent injection mode is switched to force disturbance, while setting mode switching priority and transition buffer to avoid drastic disturbance to the furnace reaction; Dynamic weighted fusion regulation: Introducing dynamic weighting factors to balance short-term fluctuation suppression and medium-to-long-term trend optimization. The greater the deviation of the current hydrogen utilization rate from the set value, the higher the short-term regulation weight, and the flow rate is adjusted first. When the current hydrogen utilization rate is close to the set value, the medium-to-long-term regulation weight is increased, and the injection mode is adjusted first, so as to achieve the synergistic cooperation of the two regulation methods. Closed-loop verification and iteration: Real-time collection of hydrogen utilization rate, hydrogen flow rate, and injection mode data after regulation, evaluation of regulation effect, and dynamic adjustment of PI controller parameters and weight factors; After each smelting, the injection rule base and weight generation model are updated based on full-process data to continuously optimize the accuracy and stability of the regulation strategy.

[0010] Preferably, after each smelting process, a closed-loop correction system is constructed to achieve continuous evolution of the control model: Collect core data such as actual total hydrogen consumption, smelting cycle, and hydrogen content detection values ​​in molten steel; calculate the actual hydrogen utilization rate; compare it with model calculation values ​​and predicted values; and analyze the reasons for deviations. Reinforce the algorithm logic, adjust the structure and parameters of the prediction model, and update the spray pattern rule base; The revised calculation model, prediction model, and control rules are synchronized to the online model library to replace the original model, ensuring that the control system for subsequent smelting processes continues to adapt to changes in operating conditions and forming a data-driven closed-loop improvement.

[0011] An online control system for hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process includes: The data processing module is used to acquire parameter data from the electric arc furnace smelting area and flue gas pipeline, and to set time-series labels for the parameter data, which includes flow rate, temperature and composition data. The hydrogen utilization calculation module is used to acquire and process the image data inside the electric arc furnace in real time, and set time tags for the image data to obtain boiling intensity, combustion stability and material melting progress. The trend prediction module is used to generate the hydrogen utilization rate for the corresponding time series based on parameter data, boiling intensity, combustion stability and material melting progress, and to predict the hydrogen utilization rate for the future time series. The dynamic control module is used to dynamically adjust the hydrogen flow rate and injection mode based on the hydrogen utilization rate under the corresponding time series and the predicted value of the hydrogen utilization rate under the future time series. The model iteration and update module is used to offline correct the hydrogen utilization rate calculation model after each smelting process, based on the actual total hydrogen consumption, smelting cycle, and the detected value of hydrogen content in the molten steel, and then update it to the online model library.

[0012] Therefore, the present invention employs the above-mentioned method and system for online control of hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process, and the technical effects are as follows: Strong sensing capability: Through composite filtering and image processing technology, it can clearly extract key features such as boiling intensity of molten pool, flame stability and material melting progress under strong arc light interference, providing rich and reliable input for regulation.

[0013] Good predictive foresight: By integrating local temporal dependence and global self-attention mechanism, and combining cross-attention feature interaction, it can achieve high-precision prediction of future multi-step changes in hydrogen utilization rate and support advanced regulation.

[0014] Optimized Coordination and Control: By dynamically weighting and integrating rapid-response flow control with sustained jetting mode adjustment, it balances short-term fluctuation suppression with long-term efficiency optimization, avoiding the limitations of single-control mechanism.

[0015] Strong self-evolution capability: Constructing a two-layer update mechanism of "online closed-loop verification + offline iterative optimization" enables the model and rules to continuously adapt to changes in furnace conditions, and the hydrogen utilization rate can be stably maintained at over 85%. Attached Figure Description

[0016] Figure 1 The diagram shows the dynamic response and prediction effect of hydrogen utilization. Figure 2 To coordinate and control the execution of: hydrogen flow rate and injection mode; Figure 3 To aid in the perception of characteristics: molten pool boiling intensity and combustion stability. Detailed Implementation

[0017] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0018] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0019] Example 1 A method for online control of hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process includes the following steps: Acquire parameter data from the electric arc furnace smelting area and flue gas pipeline, and set time-series labels for the parameter data. The parameter data includes flow rate, temperature composition data, and other parameters. Install a laser gas analyzer at the furnace top flue to collect H2, CO, and CO2 concentrations once per second and record the collection time. Install a mass flow meter at the hydrogen inlet pipeline to record the instantaneous flow rate at a period of 0.5 seconds.

[0020] The system acquires and processes image data from the electric arc furnace in real time, and sets time-series labels on the image data to obtain boiling intensity, combustion stability, and material melting progress. Using a high-temperature resistant industrial endoscope, it captures 30 frames of images of the molten pool per second, and timestamps each frame. The image processing algorithm outputs the current boiling intensity as 0.65 (dimensionless), combustion stability as 0.82, and material melting progress as 45%.

[0021] Real-time acquisition of image data within the electric arc furnace, simultaneous assignment of time-series labels to the image data, classification of the image data according to acquisition location, and corresponding processing of each type of image data, including: A composite filtering technique was employed to suppress strong arc light interference within the electric arc furnace. Clear dynamic images of the molten pool were obtained by passively acquiring the infrared / visible light signals radiated by the molten pool itself. The central molten pool contour was extracted from the dynamic image, and combined with the dynamic image data, the dynamic characteristics of the molten pool surface were analyzed in time series to reflect the boiling intensity. These dynamic characteristics included fluctuation frequency and bubble formation rate. A 950nm narrowband filter and a neutral density filter were added in front of the lens. After acquiring the infrared image of the molten pool, the molten pool contour was extracted using Canny edge detection. The pixel grayscale variance of the central region of the molten pool was statistically analyzed over 10 consecutive frames, yielding a fluctuation frequency of 12 Hz and a bubble formation rate of 8 bubbles / second, thus calculating the boiling intensity to be 0.78.

[0022] Texture features of flame images were extracted, and images of the same location under different time conditions were compared to identify changing, invariant, and abrupt parts of the texture features. These parts were then categorized, and texture features of the same category were depicted to accurately delineate the boundaries between the flame, molten pool, and droplets. By tracking the changes in flame texture features at different time intervals, the flame's trajectory and velocity field were assessed to evaluate combustion stability. Comparing two flame images at t=1 second and t=2 seconds, using the LBP texture descriptor, it was found that the texture at the flame root changed only slightly (invariant part), the texture in the middle of the flame spread outward (changing part), and a bright area suddenly appeared at t=2 seconds (abrupt part). The trajectory of the abrupt area was tracked using optical flow, and the flame front velocity was found to be 0.3 m / s, with a calculated combustion stability of 0.91.

[0023] Image data processing of the molten pool surface yields RGB values ​​and texture features. The temperature distribution on the molten pool surface is obtained from the RGB values, and the melting depth and solid-liquid phase ratio are derived from the texture features, thus determining the material melting progress. RGB values ​​are extracted from the molten pool image, and the temperature field is inverted using colorimetric thermometry (R / G ratio), revealing a center temperature of 1650℃ and an edge temperature of 1420℃. Contrast, energy, and other texture features are extracted using the gray-level co-occurrence matrix (GLCM), and the solid-liquid phase ratio is inverted, resulting in a melting depth of 120 mm and a material melting progress of 68%.

[0024] Based on parameter data and boiling intensity, combustion stability, and material melting progress, the hydrogen utilization rate for the corresponding time series is generated, and the hydrogen utilization rate for future time series is predicted; the hydrogen flow rate of 0.8 Nm³ in the first 5 seconds is used. 3 Based on data such as / s, flue gas CO concentration of 2.3%, boiling intensity of 0.72, and combustion stability of 0.88, the current hydrogen utilization rate is calculated to be 87%. At the same time, based on the LSTM prediction model, the predicted hydrogen utilization rates for the next 3 seconds are 86.5%, 86.0%, and 85.8%, respectively.

[0025] Based on parameter data and boiling intensity, combustion stability, and material melting progress, the hydrogen utilization rate for the corresponding time series is generated, including the following steps: Align the parameter data, boiling intensity, combustion stability, and material melting progress in a unified time coordinate system; resample all data from the hydrogen flow sensor (sampling period 0.5 seconds), flue gas analyzer (sampling period 1 second), and boiling intensity output from image processing (0.033 seconds per frame) to a unified 1-second time grid to ensure that all data at t=10.0 seconds are measured values ​​under the same smelting conditions.

[0026] Then, based on the characteristic cycle of the hydrogen metallurgical reaction, a sliding time window is divided according to the time sequence; the characteristic cycle of the hydrogen reduction reaction is 5 seconds, so the window length L = 5 seconds and the sliding step size step = 1 second are set. When t = 10 seconds, the window covers the data in the interval [10, 15] seconds.

[0027] The hydrogen utilization rate within the corresponding window is calculated based on parameter data, boiling intensity, combustion stability, and material melting progress, i.e., the hydrogen utilization rate under the corresponding time series. ; In the formula, This represents the hydrogen utilization rate under the corresponding time series. For the conversion rate of the reduction reaction, Let be the combustion reaction conversion rate. Within a [10, 15] second window, based on the increase in H2O in the flue gas, the amount of hydrogen consumed in reduction is estimated to be 70% of the total input. Based on the CO2 concentration in the flame image, the amount of hydrogen consumed in combustion is estimated to be 15%. Therefore, the calculated value is... .

[0028] The correction strategy is set to correct hydrogen utilization, specifically as follows: The emission loss rate was obtained from the hydrogen concentration and flue gas flow rate in the parameter data. Based on the hydrogen metallurgical reaction mechanism, hydrogen utilization includes , , , The measured H2 concentration in the flue gas was 0.8% (volume fraction), and the flue gas flow rate was 10 Nm³. 3 The calculated unreacted hydrogen emission rate is 0.08 Nm³ / s. 3 / s, while the total hydrogen input is 1.0 Nm³. 3 / s, then =8%. At this time... + =85%, plus =8%, totaling 93%, close to 100%, with the remaining 7% being measurement error and hydrogen dissolution loss.

[0029] The image features and key parameters are input into the correction model, and the correction coefficient α is output to obtain the final hydrogen utilization rate: ; To correct the coefficients, under different time windows The values ​​of are different. For the final hydrogen utilization rate; contrast and Relationship, ensure .

[0030] Input the current boiling intensity (0.72), combustion stability (0.88), material melting progress (68%), and flue gas H2 concentration (0.8%) into the trained random forest modified model, and input α=0.05. .at this time The deviation from 100% was 11.05%, triggering a system alarm and requiring the sensor to be recalibrated.

[0031] Predicting hydrogen utilization rates for future time series, including: Extracting parameter data, boiling intensity, combustion stability, material melting progress, and hydrogen utilization rate at corresponding time points yields time-series derived features, resulting in a feature vector for each time point. These derived features include difference features, interactive statistical features, and lag features. The first-order difference (flow rate change rate) of the hydrogen flow sequence is calculated. The Pearson correlation coefficient (interactive statistical) within a sliding window is calculated for boiling intensity and hydrogen utilization rate. The hydrogen utilization rate at the first three time points is used as the lag feature. Finally, a feature vector with a dimension of 32 is obtained for each time point.

[0032] Extract local dependencies from temporally derived features and output the hidden state sequence. The feature vector is input into a single-layer LSTM network (64 hidden units), the input sequence length is 10 seconds, and the output is the hidden state sequence. , dimension 10×64.

[0033] The self-attention mechanism is used to model the global dependency of time-series derived features. Parameter data, boiling intensity, combustion stability, material melting progress, and hydrogen utilization rate at the corresponding time series are encoded. The correlation weights of different time series points are calculated through multi-head self-attention, and the global feature representation is output. Using four-head self-attention to encode the same temporal features, the correlation weight between the hydrogen flow rate changes at second 1 and second 8 is as high as 0.85, indicating that early flow rate fluctuations significantly affect later utilization. The final output is the global feature. It is a vector of 1×256.

[0034] The weighting coefficients γ and β are dynamically generated based on the current smelting stage: when the material melting progress is <30%, it is considered to be in the initial melting stage, and the model generates γ=0.7 (emphasizing local dependence) and β=0.3 (emphasizing global dependence); when the melting progress is >80%, γ=0.2 and β=0.8.

[0035] Introducing a cross-attention mechanism at the feature level allows for the hidden state sequence to... With global feature representation The system interacts with other systems to calculate cross-modal attention weights, then performs weighted fusion, residual connections, and layer normalization to obtain predicted values ​​for hydrogen utilization in future time series. (10×64) as the query, (1×256) are used as the key and value. After cross-attention calculation, the fused features are obtained. Then, after residual connection and LayerNorm, the hydrogen utilization rate prediction value for the next 5 seconds is output through a fully connected layer: [86.2%, 85.7%, 85.1%, 84.6%, 84.0%].

[0036] The system dynamically adjusts the hydrogen flow rate and injection mode based on the hydrogen utilization rate under the corresponding time sequence and the predicted hydrogen utilization rate under the future time sequence. The current hydrogen utilization rate is 78%, and the predicted values ​​for the next 3 seconds are 75%, 72%, and 70%. The system determines that the deviation from the set value is 85% and the trend is deteriorating, and immediately performs adjustment.

[0037] Dynamically controlling hydrogen flow rate and injection mode, specifically including: Hydrogen flow rate control: An adaptive PI controller combining prediction deviation and current state feedback is employed. When the current hydrogen utilization rate deviates from the set threshold and the predicted deviation continues to widen, the hydrogen flow rate is rapidly adjusted. In the short term, the deviation is corrected through a proportional-integral (PI) circuit. In the medium to long term, the flow rate baseline is adjusted based on the predicted trend. Simultaneously, upper and lower limits for flow rate and rate of change constraints are set to ensure equipment safety and reaction stability. For example, if the current deviation e = 85% - 78% = 7%, the predicted deviation ê(t+1) = 10%, and the controller output ΔQ = 20%, then the hydrogen flow rate will decrease from 1000 Nm³. 3 / h increased to 1200 Nm 3 / h; however, the rate of change is limited to ≤10% / step, and the actual adjustment is 1100 Nm. 3 / h. Simultaneously set the flow rate limit to 1500 Nm. 3 / h, lower limit 500 Nm 3 / h.

[0038] Injection mode control: Based on hydrogen utilization status, predicted trends, and image characteristics, a rule-based control strategy is formulated. For example, when hydrogen utilization is below the threshold and molten pool boiling is weak, the system switches to vertical deep blowing mode to enhance molten pool penetration; when hydrogen utilization is above the threshold and the flame is stable, the system switches to diffusion wide blowing mode to reduce hydrogen waste; when hydrogen utilization fluctuates drastically and the flame is unstable, the system switches to pulse intermittent injection to force disturbance, while setting mode switching priority and transition buffer to avoid drastic disturbance to the furnace reaction; currently, the hydrogen utilization is 78% (<80%), the boiling intensity is 0.25 (low), and it is predicted to continue to decrease. The system triggers the rule, gradually adjusting the injection angle from 45° to 90° and the speed from 10 m / s to 18 m / s within 3 seconds, switching to "vertical deep blowing" mode, and changing 5° every 0.5 seconds.

[0039] Dynamic weighted fusion regulation: A dynamic weighting factor is introduced to balance short-term fluctuation suppression with medium- to long-term trend optimization. The greater the deviation of the current hydrogen utilization rate from the set value, the higher the short-term regulation weight, prioritizing flow rate adjustment. When the current hydrogen utilization rate is close to the set value, the medium- to long-term regulation weight increases, prioritizing injection mode adjustment, achieving synergistic cooperation between the two regulation methods. If the current hydrogen utilization rate is 78%, deviating from 85% by 7%, the short-term weight is calculated. =0.98, long-term weight =0.02, the final control command was mainly to increase the flow rate; when the utilization rate recovered to 84%, =0.55, =0.45, at this point we should start considering adjusting the jetting mode to tilted side blowing.

[0040] Closed-loop verification and iteration: Real-time data collection of hydrogen utilization rate, hydrogen flow rate, and injection mode after regulation to evaluate the regulation effect and dynamically adjust PI controller parameters and weight factors; after each smelting, the injection rule base and weight generation model are updated based on full-process data to continuously optimize the accuracy and stability of the regulation strategy. After 10 seconds of regulation, the hydrogen utilization rate rose to 86%, exceeding the target of 85%, but the overshoot was 2%. Therefore, the PI controller's Kp was reduced from 1.2 to 1.0, and Ki was reduced from 0.2 to 0.15; after smelting, the successful case of "vertical deep blowing and 20% flow rate" was stored in the rule base, and the weight generation network was retrained.

[0041] After each smelting cycle, the hydrogen utilization rate calculation model is offline revised based on the actual total hydrogen consumption, smelting cycle, and measured hydrogen content in the molten steel, and then updated to the online model library. The actual total hydrogen input for a certain furnace cycle was 1200 Nm³. 3 The smelting cycle was 45 minutes, and the hydrogen content of the molten steel was measured at 4 ppm, resulting in an actual hydrogen utilization rate of 82%. However, the utilization rate calculated by the online model was 86%, a deviation of 4%. The deviation data was used to correct the reduction reaction coefficient and combustion reaction coefficient in the model, and the updated model was uploaded to the online model library for use in the next furnace.

[0042] After each smelting process, a closed-loop correction system is constructed to achieve continuous evolution of the control model: Collect core data such as actual total hydrogen consumption, smelting cycle, and hydrogen content detection values ​​in molten steel; calculate the actual hydrogen utilization rate; compare it with model calculation values ​​and predicted values; and analyze the reasons for deviations. Reinforce the algorithm logic, adjust the structure and parameters of the prediction model, and update the spray pattern rule base; The revised calculation model, prediction model, and control rules are synchronized to the online model library to replace the original model, ensuring that the control system for subsequent smelting processes continues to adapt to changes in operating conditions and forming a data-driven closed-loop improvement.

[0043] Analysis of data from five consecutive heats revealed that when the proportion of light and thin scrap exceeded 40%, the model's predicted hydrogen utilization rate was generally 5% to 8% higher. This was attributed to the rapid melting and vigorous agitation of this type of scrap, leading to increased hydrogen escape. Accordingly, a "sparse coding of scrap type" feature was added to the prediction model, along with a new rule: "When light and thin scrap > 40%, the injection pulse frequency is increased by default." The revised model and rule base were deployed in the next heat, reducing the prediction error to within 2%.

[0044] like Figure 1 , Figure 2 and Figure 3 As shown, the results of specific embodiments are analyzed as follows: Advanced prediction + dynamic control: The initial hydrogen utilization rate was 78%, and it was predicted that it would continue to decline in the future (75% to 70%). The system immediately implemented a flow rate increase (+20.4%) and switched to the "vertical deep blowing" mode. Only 10 seconds later, the utilization rate quickly rose to 86% and stabilized in the target range [85%-92%].

[0045] Uncontrolled comparison: Without intervention, hydrogen leakage intensifies, utilization rate continues to decline to around 65%, and hydrogen consumption per ton of steel increases significantly.

[0046] Prediction accuracy: Based on the prediction model that combines local dependency (LSTM) and global self-attention, the rolling prediction of future time series values ​​is highly consistent with the actual values, with MAE ≤ 1.9%, providing a reliable basis for proactive regulation.

[0047] Synergistic effect: The hydrogen flow rate is smoothly adjusted according to the current deviation and the predicted trend, and the injection mode is intelligently switched according to the boiling intensity / flame stability (deep injection, side injection, wide injection) to achieve short-cycle fluctuation suppression and long-cycle efficiency optimization.

[0048] Therefore, the present invention adopts the above-mentioned method and system for online control of hydrogen utilization rate in the electric arc furnace hydrogen metallurgical process. By constructing a complete technical system of "multi-source sensing, time-series prediction, dual-dimensional collaborative control, and closed-loop iteration", the invention achieves accurate, stable, and efficient control of hydrogen utilization rate.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An on-line method for regulating the hydrogen utilization ratio in a hydrogen metallurgical process in an electric arc furnace, characterized in that, Includes the following steps: Acquire parameter data from the electric arc furnace smelting area and flue gas pipeline, and set time-series labels for the parameter data, which includes flow rate, temperature, and composition data. The image data inside the electric arc furnace is acquired and processed in real time, and time-series labels are set for the image data to obtain boiling intensity, combustion stability and material melting progress. Based on parameter data and boiling intensity, combustion stability and material melting progress, the hydrogen utilization rate under the corresponding time series is generated, and the hydrogen utilization rate under the future time series is predicted. The hydrogen flow rate and injection mode are dynamically adjusted based on the hydrogen utilization rate under the corresponding time series and the predicted hydrogen utilization rate under the future time series. After each smelting process, the hydrogen utilization rate calculation model is corrected offline based on the actual total hydrogen consumption, smelting cycle, and the measured hydrogen content of the molten steel, and then updated to the online model library.

2. The method of on-line regulation of hydrogen utilization in a hydrogen metallurgical process in an electric arc furnace according to claim 1, characterized in that, Real-time acquisition of image data within the electric arc furnace, simultaneous assignment of time-series labels to the image data, classification of the image data according to acquisition location, and corresponding processing of each type of image data, including: Composite filtering technology is used to suppress strong arc interference in the electric arc furnace. By passively acquiring the infrared / visible light signal radiated by the molten pool itself, a clear dynamic image of the molten pool is obtained. The outline of the molten pool in the dynamic image is extracted. Combined with the dynamic image data, the dynamic characteristics of the molten pool surface are analyzed in time series to reflect the boiling intensity. The dynamic characteristics include fluctuation frequency and bubble generation rate. The texture features of flame images are extracted, and images of the same location under different time conditions are compared to obtain the changing, invariant, and protruding parts of the texture features. The changing, invariant, and protruding parts are classified, and texture features of the same category are depicted to accurately delineate the boundaries between the flame and the molten pool and droplets. By tracking the changes in flame texture features under different time periods, the movement trajectory and velocity field of the flame are tracked to evaluate the combustion stability. Image data processing of the molten pool surface yields RGB values ​​and texture features of the molten pool surface. The temperature distribution of the molten pool surface is obtained based on the RGB values, and the melting depth and solid-liquid phase ratio are obtained based on the texture features of the molten pool surface, thus obtaining the material melting progress.

3. The method of on-line regulation of hydrogen utilization in a hydrogen metallurgy process in an electric arc furnace according to claim 1, characterized in that, Based on parameter data and boiling intensity, combustion stability, and material melting progress, the hydrogen utilization rate for the corresponding time series is generated, including the following steps: Align the parameter data with boiling intensity, combustion stability, and material melting progress in a unified time coordinate system; Then, based on the characteristic cycle of the hydrogen metallurgical reaction, it is divided into sliding time windows according to the time sequence; The hydrogen utilization rate within the corresponding window is calculated based on parameter data, boiling intensity, combustion stability, and material melting progress, i.e., the hydrogen utilization rate under the corresponding time series. ； wherein is the hydrogen utilization for the respective timing, is the reduction reaction conversion rate, is the combustion reaction conversion rate.

4. The method of on-line regulation of hydrogen utilization in a hydrogen metallurgy process in an electric arc furnace according to claim 3, characterized in that, The correction strategy is set to correct hydrogen utilization, specifically as follows: The emission loss rate is obtained from the hydrogen concentration in the flue gas composition and the flue gas flow in the parameter data , based on the hydrogen metallurgical reaction mechanism, hydrogen utilization includes 、 、 , ; The image features and key parameters are input into the correction model, and the correction coefficient α is output to obtain the final hydrogen utilization rate: ； To correct the coefficients, under different time windows The values ​​of are different. For the final hydrogen utilization rate; contrast and Relationship, ensure .

5. The method for online control of hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process according to claim 1, characterized in that, Predicting hydrogen utilization rates for future time series, including: Extract parameter data, boiling intensity, combustion stability, material melting progress, and hydrogen utilization rate at the corresponding time series to extract time series derived features, and obtain the feature vector of each time series point. The time series derived features include differential features, interactive statistical features, and lag features. Extract local dependencies from temporally derived features and output the hidden state sequence. ; The self-attention mechanism is used to model the global dependency of time-series derived features. Parameter data, boiling intensity, combustion stability, material melting progress, and hydrogen utilization rate at the corresponding time series are encoded. The correlation weights of different time series points are calculated through multi-head self-attention, and the global feature representation is output. ; Weighting coefficients γ and β are dynamically generated based on the current smelting stage: Introducing a cross-attention mechanism at the feature level allows for the hidden state sequence to... With global feature representation The system interacts with other systems to calculate cross-modal attention weights, then performs weighted fusion, and obtains predicted values ​​for hydrogen utilization in future time series through residual connections and layer normalization.

6. The method for online control of hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process according to claim 1, characterized in that, Dynamically controlling hydrogen flow rate and injection mode, specifically including: Hydrogen flow rate control: An adaptive PI controller combining prediction deviation and current status feedback is adopted. When the current hydrogen utilization rate deviates from the set threshold and the prediction value shows that the deviation continues to increase, the hydrogen flow rate is quickly adjusted. In the short term, the deviation is corrected by the proportional-integral stage. In the medium and long term, the flow rate benchmark value is adjusted by combining the prediction trend. At the same time, upper and lower limits of flow rate and change rate constraints are set to ensure equipment safety and reaction stability. Injection mode control: Based on hydrogen utilization status, predicted trends and image characteristics, a rule-based control strategy is formulated; when hydrogen utilization is below the threshold and molten pool boiling is weak, switch to vertical deep blowing mode to enhance molten pool penetration; when hydrogen utilization is above the threshold and flame is stable, switch to diffusion wide blowing mode to reduce hydrogen waste; when hydrogen utilization fluctuates drastically and flame is unstable, switch to pulse intermittent injection to force disturbance, while setting mode switching priority and transition buffer to avoid drastic disturbance to the furnace reaction; Dynamic weighted fusion regulation: Introducing dynamic weighting factors to balance short-term fluctuation suppression and medium-to-long-term trend optimization. The greater the deviation of the current hydrogen utilization rate from the set value, the higher the short-term regulation weight, and the flow rate is adjusted first. When the current hydrogen utilization rate is close to the set value, the medium-to-long-term regulation weight is increased, and the injection mode is adjusted first, so as to achieve the synergistic cooperation of the two regulation methods. Closed-loop verification and iteration: Real-time collection of hydrogen utilization rate, hydrogen flow rate, and injection mode data after regulation, evaluation of regulation effect, and dynamic adjustment of PI controller parameters and weight factors; After each smelting, the injection rule base and weight generation model are updated based on full-process data to continuously optimize the accuracy and stability of the regulation strategy.

7. The method for online control of hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process according to claim 1, characterized in that, After each smelting process, a closed-loop correction system is constructed to achieve continuous evolution of the control model: Collect core data such as actual total hydrogen consumption, smelting cycle, and hydrogen content detection values ​​in molten steel; calculate the actual hydrogen utilization rate; compare it with model calculation values ​​and predicted values; and analyze the reasons for deviations. Reinforce the algorithm logic, adjust the structure and parameters of the prediction model, and update the spray pattern rule base; The revised calculation model, prediction model, and control rules are synchronized to the online model library to replace the original model, ensuring that the control system for subsequent smelting processes continues to adapt to changes in operating conditions and forming a data-driven closed-loop improvement.

8. An online control system for hydrogen utilization rate in an electric arc furnace hydrogen metallurgical process, used to execute the method of claim 1, characterized in that, include: The data processing module is used to acquire parameter data from the electric arc furnace smelting area and flue gas pipeline, and to set time-series labels for the parameter data, which includes flow rate, temperature and composition data. The hydrogen utilization calculation module is used to acquire and process the image data inside the electric arc furnace in real time, and set time tags for the image data to obtain boiling intensity, combustion stability and material melting progress. The trend prediction module is used to generate the hydrogen utilization rate for the corresponding time series based on parameter data, boiling intensity, combustion stability and material melting progress, and to predict the hydrogen utilization rate for the future time series. The dynamic control module is used to dynamically adjust the hydrogen flow rate and injection mode based on the hydrogen utilization rate under the corresponding time series and the predicted value of the hydrogen utilization rate under the future time series. The model iteration and update module is used to offline correct the hydrogen utilization rate calculation model after each smelting process, based on the actual total hydrogen consumption, smelting cycle, and the detected value of hydrogen content in the molten steel, and then update it to the online model library.

Images