Industrial energy-saving AI intelligent control method and system for multi-device cooperation

Abstract

The invention discloses an industrial energy-saving AI intelligent control method and system for multi-device collaboration, and belongs to the technical field of energy saving.The method comprises the following steps that S1, global data perception is conducted, s2, collaborative state modeling: inputting the data acquired in the S1 into a dynamic coupling model based on a graph neural network, S3, inputting the global state vector into a reinforcement learning agent trained based on a near-end strategy optimization algorithm, and outputting a multi-dimensional collaborative control instruction set by the agent; s31, core heat state parameters of the electric arc furnace system are monitored in real time, the core heat state parameters serve as the standard of linkage regulation and control, and a unified multi-dimensional cooperative control instruction set is generated and executed through an AI cooperative controller; and S4, executing the multi-dimensional cooperative control instruction set by multiple devices, and dynamically adjusting the operation parameters of the multiple cooperative devices.

Description

Technical Field

[0001] This invention relates to the field of energy-saving technology, and in particular to an industrial energy-saving AI intelligent control method and system for multi-device collaboration. Background Technology

[0002] Electric arc furnace (EAF) steelmaking is an important short-process steelmaking technology in the modern steel industry, widely used for its high flexibility and environmental benefits. However, the EAF is also an extremely complex, multi-field coupled, and energy-intensive industrial system. Its operation involves the instantaneous conversion and transfer of electrical, chemical, and thermal energy, as well as the precise delivery of materials. For a long time, improving the overall energy efficiency and reducing operating costs of EAFs has been a core challenge that researchers in this field have continuously tackled.

[0003] Existing electric arc furnace control systems suffer from the following problems: Most current automated electric arc furnace systems employ an islanded control strategy targeting individual devices or subsystems. For example, the automatic electrode regulator aims only to stabilize the arc current or power; the oxygen lance and carbon injection system are controlled based on empirical settings or simple process models; and the dust removal fan typically operates at a fixed speed or is simply adjusted according to the furnace pressure. The lack of effective information exchange and coordination mechanisms between these subsystems leads to conflicting or even canceling control objectives. For instance, blindly increasing electrode power in pursuit of higher molten pool temperatures may simultaneously result in excessive electrode consumption, a surge in furnace lining heat load, and a spike in exhaust gas temperature, ultimately increasing the overall system energy consumption instead of decreasing it. Summary of the Invention

[0004] To address the aforementioned problems, this invention aims to solve the issues described above. One objective of this invention is to provide an industrial energy-saving AI intelligent control method and system for multi-device collaboration, which solves one of the problems described above.

[0005] The solution adopted in this invention is: an industrial energy-saving AI intelligent control method for multi-device collaboration, the method comprising the following steps: S1: Global Data Awareness: Real-time acquisition of operating data of multiple coordinating devices in the electric arc furnace system, including: electrode system, carbon powder injection system, dust removal fan system, heat exchanger system, and heat-conducting plate integrated into the electrode and furnace wall; S2: Cooperative state modeling: The data collected in S1 is input into a dynamic coupling model based on graph neural network. This model maps the multiple cooperative devices as nodes in the graph, establishes energy flow and material flow connection edges between nodes, and outputs a global state vector that characterizes the overall operating state of the system. S3: AI Collaborative Decision Making: The global state vector is input into a reinforcement learning agent trained based on a proximal policy optimization algorithm, and the agent outputs a multi-dimensional collaborative control instruction set. S31: Real-time monitoring of the core thermal state parameters of the electric arc furnace system, including the total system heat load H, the comprehensive heat exchange efficiency K of the heat exchanger system at the far end of the heat conduction plate, and the heat grade Q of the exhaust gas; using the core thermal state parameters as the benchmark for linkage control, a unified multi-dimensional collaborative control instruction set is generated and executed through an AI collaborative controller. S4: Multi-device collaborative execution and feedback: Execute the multi-dimensional collaborative control instruction set, dynamically adjust the operating parameters of the multiple collaborative devices, and collect system feedback data to form closed-loop control.

[0006] The preferred technical solution is that, in step S31, the total system heat load H is a comprehensive index calculated based on the power of the electrode system and the heat exchanger system; the comprehensive heat exchange efficiency K is a performance index determined based on the heat output per unit time, the effective area of ​​the heat conduction plate, and the temperature difference between the heat exchanger system at the hot end and the far end; and the waste gas heat grade Q is a heat grade evaluation value that comprehensively considers the physical sensible heat of the waste gas and the chemical energy of its combustible components. Using the three core thermal state parameters as a unified and coordinated control benchmark, a unified multi-dimensional collaborative control instruction set is generated through an AI collaborative controller. This instruction set synchronously coordinates the operating status of the following devices: based on the total system heat load H, synchronously adjusts the input power of the electrode system and the carbon powder injection rate of the carbon powder injection system; based on the comprehensive heat exchange efficiency K, dynamically adjusts the heat dissipation intensity and heat distribution path of the heat exchanger system; and based on the waste gas heat grade Q, collaboratively controls the operating air volume and waste heat recovery mode of the dust removal fan system. The preferred technical solution is that, in step S311: the input electrical power of the electrode system is synchronously adjusted according to the total system heat load H, including the coordinated temperature control of the heat-conducting plate inside the electrode. The specific steps are as follows: The electrode body operating temperature Te and the electric arc furnace molten pool temperature Tf are monitored in real time. Based on the deviation between the electric arc furnace molten pool temperature Tf and the target smelting temperature, the required electrode power adjustment ΔP is predicted. According to the electrode power adjustment ΔP and the electrode body operating temperature Te, the thermal conductivity intensity adjustment ΔΦ of the heat-conducting plate inside the electrode is determined. By adjusting the heat dissipation power of the heat exchanger at the far end of the heat-conducting plate, the thermal conductivity intensity adjustment ΔΦ is controlled so that the electrode body operating temperature Te is maintained within the preset optimized temperature range [Te_min, Te_max]. The thermal conductivity intensity adjustment ΔΦ is positively correlated with the electrode power adjustment ΔP, and when Te exceeds the optimized temperature range, priority is given to adjustment. In this way, the comprehensive heat exchange efficiency K is enhanced through precise control of the electrode temperature, achieving the dual effect of reducing electrode loss and improving energy utilization efficiency.

[0007] The preferred technical solution is that, in step S312, the carbon powder injection rate and the heat dissipation intensity of the heat-conducting plate system are dynamically coordinated according to the total heat load H of the system and the temperature distribution inside the furnace. When local overheating is detected in the furnace, the carbon powder injection rate and the heat exchange intensity of the corresponding heat-conducting plate are increased simultaneously. Temperature balance control is achieved by strengthening the formation of the foam slag layer and enhancing heat dissipation. Establish a correlation model between toner consumption and heat dissipation, and optimize the ratio between the two with the goal of minimizing the total system energy consumption.

[0008] The preferred technical solution is that the correlation model between the carbon powder consumption and the heat dissipation is as follows: an energy consumption optimization function E is constructed with the carbon powder injection rate V and the equivalent heat dissipation Q of the heat conduction plate system as input variables; the calculation expression of the energy consumption optimization function E is: E = a·V + b·Q + c·(T1 - T2)², where a, b, and c are weighting coefficients, T1 is the target furnace temperature, and T2 is the actual furnace temperature; Based on real-time collected data of total system heat load H and furnace temperature distribution, the optimal solution set [V1, Q1] that minimizes E is dynamically solved by gradient descent method; the optimal solution set is then converted into control commands to synchronously adjust the feeder speed of the carbon powder injection system and the medium circulation rate of the heat-conducting plate system. Establish an optimization effect feedback mechanism, and periodically adjust the weight coefficients a, b, and c by comparing the difference between actual energy consumption and predicted energy consumption to achieve self-calibration of model parameters.

[0009] The preferred technical solution is that the heat-conducting plate is a sealed vacuum cavity structure, the shell is made of nickel-based high-temperature alloy material, and the wall thickness is 1.0-1.5mm; the vacuum cavity is filled with FUD inorganic superconducting medium, the composition of which includes Li2CO3-K2CO3-Cs2CO3 ternary eutectic salt matrix, 0.6wt% graphene thermal conductivity enhancer and 0.4wt% CeO2 thermal stabilizer; the inner surface of the shell is formed with a micro-capillary channel structure by laser cladding process, the channel width is 50-80μm and the depth is 30-50μm; The heat-conducting plate is bonded to the electrode and furnace wall substrate through a high-temperature welding process, and forms a closed loop with the remote heat exchanger; the remote heat exchanger adopts a finned tube structure, and the heat dissipation intensity is continuously controlled by adjusting the speed of the cooling fan.

[0010] The preferred technical solution is that, in step S313, the operating air volume and waste heat recovery mode of the dust removal fan system are controlled in a coordinated manner based on the waste gas calorific value Q. The specific steps are as follows: The system monitors the exhaust gas temperature (Tg), exhaust gas component concentration (Cg), and exhaust gas flow rate (Fg) in real time. Based on these parameters, it calculates the real-time values ​​of the recoverable waste heat (Er) and the exhaust gas calorific value (Q). When the exhaust gas calorific value (Q) is higher than a preset threshold (Q_th), it activates a high-efficiency waste heat recovery mode, simultaneously increasing the operating airflow of the dust removal fan system to enhance exhaust gas capture efficiency and adjusting the heat exchange rate of the heat exchanger system to maximize waste heat utilization. When the exhaust gas calorific value (Q) is lower than the preset threshold (Q_th), it switches to an energy-saving operation mode, reducing the operating airflow of the dust removal fan system to decrease energy consumption while maintaining basic exhaust gas treatment requirements. A correlation model is established between the exhaust gas calorific value (Q) and the total system heat load (H). An AI collaborative controller dynamically optimizes the dust removal fan airflow setpoint and waste heat recovery parameters to achieve a balance between exhaust gas energy recovery and system energy consumption.

[0011] The preferred technical solution is that, in step S314: based on the comprehensive heat exchange efficiency K and the ambient temperature Ta, the heat dissipation intensity and heat distribution path of the heat exchanger system are dynamically adjusted, and the specific steps are as follows: The system monitors the inlet temperature Tin, outlet temperature Tout, and vaporization coefficient Fm of the liquid medium in real time. Based on these parameters, the actual overall heat exchange efficiency K is calculated. The actual overall heat exchange efficiency K is compared with the target heat exchange efficiency K_target to generate a heat dissipation intensity adjustment ΔK. Simultaneously, the ambient temperature Ta is monitored, and the priority of the heat distribution path is adjusted according to the ambient temperature Ta. When the ambient temperature Ta is low, heat is preferentially distributed to the plant's hot water system. When the ambient temperature Ta is high, heat dissipation is preferentially enhanced to protect the equipment. The heat dissipation intensity adjustment ΔK is controlled by adjusting the cooling fan speed of the heat exchanger system to maintain the actual overall heat exchange efficiency K within the target range. The heat output path is dynamically switched using a heat distribution valve.

[0012] The preferred technical solution further includes step S5: Adaptive model update and optimization: Based on the system feedback data collected in step S4, the dynamically coupled model based on graph neural networks and the reinforcement learning agent trained based on the proximal policy optimization algorithm are updated periodically. The specific steps are as follows: Historical operational data is collected, including equipment operating parameters, control command sets, and corresponding system performance indicators; the deviation between actual and predicted energy efficiency indicators is calculated to evaluate model performance; when the deviation exceeds a preset tolerance, a model update process is triggered; the graph neural network dynamic coupling model is retrained using historical data to optimize the connection weights of energy and material flows between nodes; simultaneously, a reinforcement learning agent interacts with the updated dynamic coupling model to adjust policy parameters to adapt to system changes; the accuracy and stability of the updated model are verified, and it is deployed to the real-time control system after simulation testing; a model version management mechanism is established to ensure the continuity and reliability of the control system.

[0013] This invention also relates to an industrial energy-saving AI intelligent control system for multi-device collaboration using the above-mentioned method. The system includes: a global data sensing module, comprising a sensor network deployed on multiple collaborative devices in the electric arc furnace system, for real-time acquisition of the operating data of the collaborative devices; the collaborative devices include: an electrode system, a carbon powder injection system, a dust removal fan system, a heat exchanger system, and a heat-conducting plate integrated into the electrode and the furnace wall. The collaborative state modeling and decision-making center, communicatively connected to the global data perception module, includes: a dynamic coupling model processing unit based on a graph neural network, configured to map the multiple collaborative devices as nodes in a graph, establish energy flow and material flow connection edges between nodes, and output a global state vector characterizing the overall operating state of the system; and an AI collaborative controller trained based on a near-end strategy optimization algorithm, configured to receive the global state vector and output a unified multi-dimensional collaborative control instruction set; the AI ​​collaborative controller uses the total system heat load H, the comprehensive heat exchange efficiency K of the heat exchanger system at the far end of the heat conduction plate, and the waste gas heat grade Q as core thermal state parameters as the benchmark for linkage regulation; The collaborative execution and feedback network is communicatively connected to the collaborative state modeling and decision-making center. It includes actuators connected to the multiple collaborative devices for receiving and executing the multi-dimensional collaborative control instruction set and dynamically adjusting the operating parameters of each device. The sensor network is further used to collect system feedback data after instruction execution and send the feedback data back to the collaborative state modeling and decision-making center to form closed-loop control.

[0014] Compared with existing technologies, the industrial energy-saving AI intelligent control method and system for multi-device collaboration of the present invention has the following technical effects: 1. This application relates to an industrial energy-saving AI intelligent control method and system for multi-device collaboration. This invention achieves global dynamic energy optimization and significantly reduces overall energy consumption. Through a dynamic coupling model based on graph neural networks, it precisely characterizes the complex energy and material flow coupling relationships between devices such as electrodes, toner, heat exchangers, and dust collectors at the control level. The AI ​​decision center performs collaborative decision-making based on this global state, breaking down energy consumption barriers between devices and finding the optimal operating point with the lowest total system energy consumption. Compared to traditional independent subsystem control, this avoids the increase in total system energy consumption caused by individual devices pursuing their own optimization, thus reducing overall energy consumption.

[0015] By deeply exploring and efficiently utilizing waste heat resources, the system improves overall energy utilization efficiency. Through real-time calculation and tracking of waste gas heat grade Q and overall heat exchange efficiency K, the system can intelligently switch waste heat recovery modes. When the waste gas quality is high, it actively enhances capture and recovery; at the same time, it dynamically allocates heat paths according to ambient temperature, enabling precise and efficient utilization of thermal energy and greatly improving energy utilization.

[0016] Traditional control methods respond to changes in furnace conditions with lag and in a localized manner. This invention uses the total system heat load H as a core benchmark to achieve synchronous and coordinated adjustment of electrode power, carbon powder injection, and heat dissipation intensity. When an increase in furnace heat load is detected, not only is the electrode power adjusted, but the carbon powder and heat dissipation system are also coordinated simultaneously. This rapidly smooths temperature fluctuations, effectively preventing localized overheating or uneven temperature distribution, significantly improving the temperature stability of the smelting process, and thus ensuring consistent product quality.

[0017] By precisely monitoring the electrode body's operating temperature Te and coordinating control with the superconducting heat-conducting plate, the electrode can be maintained within an optimized temperature range, significantly reducing the risk of oxidation, loss, and breakage due to overheating and extending its service life. Monitoring the overall heat exchange efficiency and intelligently adjusting the heat exchanger also prevents the heat exchange system from operating under overload or inefficient conditions for extended periods, thus reducing its failure rate.

[0018] Other features and advantages of the invention will become clear when reading the following description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention. In these drawings, similar reference numerals are used to denote similar elements. The drawings described below are some embodiments of the invention, but not all embodiments. Other drawings will be readily available to those skilled in the art based on these drawings without any inventive effort.

[0020] Figure 1This is a schematic diagram of the main process of the industrial energy-saving AI intelligent control method for multi-device collaboration provided in a specific embodiment of the present invention; Figure 2 This is a schematic diagram of the specific process of step S31 of the industrial energy-saving AI intelligent control method for multi-device collaboration provided in a specific embodiment of the present invention. Detailed Implementation To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other.

[0021] The following description, in conjunction with the accompanying drawings and embodiments, details the industrial energy-saving AI intelligent control method and system for multi-device collaboration.

[0022] like Figure 1-2 As shown, an industrial energy-saving AI intelligent control method for multi-device collaboration includes the following steps: S1: Global data perception: Real-time acquisition of operating data from multiple collaborative devices in the electric arc furnace system. These collaborative devices include: an electrode system, a toner injection system, a dust removal fan system, a heat exchanger system, and heat-conducting plates integrated into the electrodes and furnace wall. A corresponding sensor network is deployed on each collaborative device, and the data is transmitted to the central control unit in real-time via an industrial bus or industrial Ethernet. The acquired data specifically includes: for the electrode system, the voltage, current, active power, electrode position, and the electrode body operating temperature Te acquired by thermocouples embedded within the electrodes. For the furnace wall heat-conducting plate, the furnace wall temperature, the temperature at the end of the heat-conducting plate that is attached to the furnace wall, and the temperature at the end of the heat-conducting plate that is located in the heat exchanger system are acquired. For the toner injection system, the toner hopper level and feeder speed are acquired to calculate the toner injection rate V and injection pressure. Dust removal fan system: Collects fan speed, motor current, damper opening, and exhaust gas temperature Tg, exhaust gas component concentration Cg, and exhaust gas flow rate Fg via a flue gas analyzer and thermocouples. Heat exchanger system: Collects inlet temperature Tin, outlet temperature Tout of the remote heat exchanger connected to the superconducting heat plate, and cooling fan speed. Electric arc furnace body: Collects molten pool temperature Tf and temperature distribution in different areas of the furnace via a radiation thermometer or thermocouples embedded in the furnace lining. This provides comprehensive, real-time, and high-precision data support for subsequent collaborative modeling and decision-making, and is a prerequisite for achieving global optimization.

[0023] S2: Collaborative State Modeling: The data collected in S1 is input into a dynamic coupling model based on a graph neural network. This model maps the multiple collaborative devices as nodes in a graph and establishes energy and material flow connections between nodes, outputting a global state vector representing the overall operating state of the system. Graph Structure Construction: Nodes: Each collaborative device, such as the electrode, toner spray gun, dust collector fan, heat exchanger, and superconducting heat transfer plate, is defined as a node. Node Attributes: The attribute vector of each node is the operating data of that device collected in S1. Connection Edges: Directed edges are established between nodes where energy or material exchange occurs, based on physical and technological knowledge. For example: Electrode → Molten Pool: Energy flow. Toner Spray Gun → Molten Pool: Material flow, accompanied by energy flow. Molten Pool → Superconducting Heat Transfer Plate / Furnace Wall Cooling System: Energy flow. Molten Pool → Dust Collector Fan: Material and energy flow. Superconducting Heat Transfer Plate → Heat Exchanger: Energy flow. Model Training and Inference: The GNN is trained using historical normal operation data to learn how node states influence and propagate through connections. The trained model, based on real-time input node data, uses message passing and node state update mechanisms to ultimately converge into a low-dimensional, dense global state vector. This vector encapsulates the overall system's operational health and energy distribution. The GNN model can quantitatively capture complex coupling relationships that are difficult to describe using traditional methods, such as how adjusting electrode power affects exhaust gas temperature. This provides unprecedented, deep system insights for AI decision-making and forms the foundation for intelligent collaboration.

[0024] S3: AI Collaborative Decision Making: The global state vector is input into a reinforcement learning agent trained based on a proximal policy optimization algorithm. This agent outputs a multi-dimensional collaborative control instruction set. Reinforcement learning is an ideal tool for solving this type of sequential decision-making problem, and the proximal policy optimization algorithm is preferred due to its stability and efficiency. In specific implementation: Agent Training: In a simulation environment or historical data, define the state, i.e., the global state vector output by S2; define the actions, i.e., the multi-dimensional collaborative control instruction set to be output, such as electrode power setpoint, toner injection rate, fan speed, etc.; define the reward function, designed with multiple objectives such as minimum total system energy consumption, most stable temperature, and longest equipment lifespan. The agent obtains rewards through interaction with the environment and continuously optimizes its decision-making strategy through the PPO algorithm, ultimately learning which collaborative control instruction should be executed under which system state to obtain the maximum long-term accumulated reward. Online Decision Making: After training, the agent is deployed to the real-time control system. It receives the real-time global state vector from S2 and instantly outputs the corresponding optimal multi-dimensional collaborative control instruction set. PPO agents can learn complex collaborative strategies that far exceed the experience of human experts. Their decisions are globally optimal, rather than a simple superposition of multiple local optima, thereby fundamentally solving control conflict problems and maximizing energy efficiency.

[0025] S31: Real-time monitoring of the core thermal state parameters of the electric arc furnace system, including the total system heat load H, the comprehensive heat exchange efficiency K of the heat exchanger system at the far end of the heat-conducting plate, and the waste gas heat grade Q; using these core thermal state parameters as the benchmark for linkage control, a unified multi-dimensional collaborative control instruction set is generated and executed through an AI collaborative controller; this step is the concretization and core logic of the AI ​​collaborative decision-making in S3. It clarifies the key physical benchmarks and linkage mechanisms upon which the decision is based. The AI ​​collaborative controller here is the reinforcement learning agent mentioned in S3. Core parameter calculation: Total system heat load H: A comprehensive heat balance index calculated based on the electrical power input to the electrode system, the heat released by the chemical reaction, and the heat lost through the heat exchanger. It reflects the current total heat production and heat dissipation of the system. Comprehensive heat exchange efficiency K: A performance index determined by the heat exported per unit time through the superconducting heat-conducting plate, the effective area of ​​the heat-conducting plate, and the temperature difference between its hot end and the far-end heat exchanger. K = Q_export / (A * ΔT), where Q_export is the heat transfer, A is the area, and ΔT is the temperature difference. It reflects the health and efficiency of the heat removal system. Exhaust gas calorific value Q: This is an assessment value that integrates the sensible heat of the exhaust gas and the chemical energy of its combustible components. Q = Fg * Cp * (Tg - T0) + Fg * Cg * CV, where Fg is the flow rate, Cp is the specific heat capacity, T0 is the ambient temperature, Cg is the concentration of combustible components, and CV is the calorific value of the component. Interlocking control benchmarks: These three parameters constitute the dashboard of the system's energy flow. The decision-making strategy of the AI ​​collaborative controller essentially learns to use these three parameters as benchmarks to perform the following interlocking controls: Based on H control: When H increases, the electrode system is synchronously adjusted to reduce the input power, and the carbon powder injection system is adjusted to change the carbon injection rate, addressing both the increase and decrease sides simultaneously. Based on K control: When K is low, the heat dissipation intensity of the heat exchanger system is dynamically adjusted, and the heat distribution path is optimized. Based on Q-based control: When Q exceeds a threshold, a high-efficiency waste heat recovery mode is activated, simultaneously increasing the dust collector fan's airflow to capture more high-quality exhaust gas and adjusting the waste heat boiler parameters to maximize heat recovery. Through these three core thermal state parameters, the complex multi-device control problem is condensed into a clear and logically rigorous linkage control framework. This ensures that the AI's decisions are not only intelligent but also conform to physical laws and process principles, guaranteeing the reliability and efficiency of control.

[0026] S4: Multi-device Cooperative Execution and Feedback: Execute the multi-dimensional cooperative control instruction set, dynamically adjust the operating parameters of the multiple cooperative devices, and collect system feedback data to form a closed-loop control. This step is crucial for putting intelligent decision-making into practice and completing the learning loop. Specifically: Cooperative Execution: The central control unit sends the multi-dimensional cooperative control instruction set generated in S3 to each cooperative device through the actuators. These instructions are sent simultaneously or in a very short sequence, ensuring the synchronization and coordination of device actions. After the instructions are executed, the sensor network in S1 continuously collects system response data, such as new temperatures, pressures, and flow rates. This data constitutes the system feedback data. The feedback data is then fed back into the cooperative state modeling module in S2 and the AI ​​decision-making module in S3. On one hand, it is used to evaluate the effectiveness of the previous round of control instructions; on the other hand, this accumulated historical data can be used for adaptive model updates and optimization, i.e., periodically retraining the GNN model and reinforcement learning agent, enabling the system to adapt to equipment aging, raw material changes, and other operating condition drifts, achieving continuous self-optimization. This step not only ensures the real-time and accuracy of the control effect, but also endows the entire system with the ability to learn and evolve adaptively throughout its life, enabling the energy-saving effect to remain at a high level for a long time.

[0027] In step S31, the total system heat load H is a comprehensive index calculated based on the electrode system power and the heat exchanger system. The total system heat load H is a comprehensive index reflecting the instantaneous heat balance of the electric arc furnace system. It characterizes the sum of the total heat generated within the system and the heat that needs to be carried away through cooling and exhaust gas. Its core is to establish a simplified real-time heat balance model. Specific calculation formulas and data sources: Electrode input electrical power P_2: The sum of the three-phase active power directly obtained from the electrical energy metering unit of the electrode system; this is the primary source of heat. Chemical reaction exothermic P_3: Mainly from the reaction between carbon powder and elements in the metal raw materials. This part can be estimated by multiplying the carbon powder injection rate V_c by an empirical exothermic coefficient η_c, i.e., P_3 = V_c * η_c.

[0028] The output item is the heat conduction of the heat exchanger system, Q_cooling: This is calculated by collecting the inlet temperature Tin, outlet temperature Tout, and flow rate F_medium of the cooling medium at the remote heat exchanger, based on the medium's specific heat capacity Cp_medium: Q_cooling = F_medium * Cp_medium * (Tout - Tin). This value represents the heat actively removed by cooling systems such as superconducting heat plates.

[0029] The total system heat load H can be approximated as the sum of the main components on the heat generation and heat dissipation sides, considering a system heat loss coefficient κ (κ, usually close to 1). The formula is: H = κ * (P_electrode + P_chemical + Q_cooling) (κ is usually close to 1). This calculation is performed several times per second, resulting in a dynamically updated H value that reflects the overall thermal stress of the system. The comprehensive heat transfer efficiency K is a performance index determined based on the heat removed per unit time, the effective area of ​​the heat-conducting plate, and the temperature difference between the hot end and the far end of the heat exchanger system. The comprehensive heat transfer efficiency K is used to evaluate the effectiveness of the entire heat removal system composed of the superconducting heat-conducting plate and its far-end heat exchanger. Essentially, it is a heat transfer coefficient; the higher the value, the stronger the system's thermal conductivity per unit temperature difference and unit area.

[0030] Specific calculation formulas and data sources: Heat exported per unit time (Q_export): This is the Q_cooling used in the calculation of the total heat load H above. Effective area of ​​the heat-conducting plate (A): This is a fixed design parameter determined by the specific structure and number of heat-conducting plates. For example, the total effective heat exchange area is calculated based on the size and number of each plate. Temperature difference between the hot end and the far end (ΔT): Hot end temperature (T_hot): Obtained directly by thermocouples embedded in the electrode or furnace wall substrate near the installation point of the superconducting heat-conducting plate. Far end temperature (T_cold): Usually, the inlet temperature (Tin) of the cooling medium of the heat exchanger is taken as the far end reference temperature. Therefore, ΔT = T_hot - T_cold. The overall heat transfer efficiency K is calculated using the following basic heat transfer formula: K = Q_export / (A * ΔT). A decrease in the K value may indicate a problem in the system, such as degradation of the medium performance inside the superconducting heat-conducting plate, pipe scaling, or poor cooling on the heat exchanger side. Therefore, K is not only a control benchmark but also a diagnostic indicator of the system's health status.

[0031] The waste gas sensible heat grade Q is an assessment value that comprehensively considers the physical sensible heat of the waste gas and the chemical energy of its combustible components; the waste gas sensible heat grade Q aims to assess the energy carried in the waste gas. Higher-grade heat is more suitable for recovery and utilization, while lower-grade heat has lower utilization value. Specific calculation formulas and data sources: Physical sensible heat (Q_sensible): The waste gas temperature (Tg) is measured by a thermocouple on the flue gas duct before the dust collector fan, and the waste gas flow rate (Fg) is measured by a flow meter. Q_sensible = Fg * Cp_gas * (Tg - T0), where Cp_gas is the average isobaric specific heat capacity of the waste gas, and T0 is the reference temperature of the waste heat recovery system. Chemical energy (Q_chemical): The waste gas component concentration (Cg) is monitored in real time by an online gas analyzer on the same flue gas duct, with a focus on monitoring the concentration of combustible gases such as carbon monoxide (CO) and hydrogen (H2). For each combustible component i, its chemical energy Q_chem_i = Fg * Cg_i * CV_i, where CV_i is the calorific value per unit of that component, such as the calorific value of CO, which is approximately 3.0 MJ / Nm³. The total chemical energy is the sum of all components: Q_chemical = Σ Q_chem_i. The calorific value Q of the exhaust gas is a weighted sum of sensible heat and chemical energy. Chemical energy is typically given a higher weight due to its higher work-capacity (weighting coefficient ω>1): Q = Q_sensible + ω * Q_chemical.

[0032] Using the three core thermal state parameters as a unified and coordinated control benchmark, an AI collaborative controller generates a unified multi-dimensional collaborative control instruction set. This instruction set synchronously coordinates the operating states of the following devices: This part is the concentrated embodiment of the entire invention's control logic. The strategy network embedded in the AI ​​collaborative controller has been trained and can understand the complex mapping relationship between the three parameters H, K, and Q and the optimal operating points of each device. Its linkage control logic is specifically manifested as follows: based on the total system heat load H, the input power of the electrode system and the carbon powder injection rate of the carbon powder injection system are synchronously adjusted; if the H value is too high, it indicates that the system generates too much heat or dissipates too little heat, posing a risk of overheating; if the H value is too low, the smelting intensity is insufficient. The controller needs to adjust from both ends synchronously. When H is continuously higher than the target upper limit, the controller synchronously outputs instructions: reduce the input power setting value of the electrode system and may appropriately increase the carbon powder injection rate. The purpose of increasing the carbon powder rate is to enhance the formation of foam slag. Foam slag can cover the surface of the molten steel, playing a role in heat preservation, improving thermal efficiency, and protecting the furnace lining. This helps maintain the stability of the molten pool temperature and avoids a sudden temperature drop while reducing the input power. Conversely, the opposite is also true. This synchronous adjustment avoids drastic fluctuations in the molten pool temperature caused by adjusting the electrode power alone, thus maintaining process stability while adjusting the total heat load.

[0033] Based on the overall heat exchange efficiency K, the heat dissipation intensity and heat distribution path of the heat exchanger system are dynamically adjusted; the K value reflects the system's ability to remove heat. The goal is to maintain K within the high-efficiency range. If the K value is too low, it indicates poor heat dissipation, requiring enhanced heat dissipation; simultaneously, the destination of the removed heat is determined based on the heat quality. When the K value falls below the target range, the controller outputs a command to increase the speed of the cooling fan in the heat exchanger system or increase the flow rate of the cooling medium to improve the heat dissipation intensity and raise the K value.

[0034] Heat distribution path: The controller simultaneously considers the ambient temperature. In winter, heat is prioritized for distribution to the plant's heating system via valve switching; in summer, heat is prioritized for discharge to the atmosphere via cooling towers and other equipment, and heat dissipation may be further enhanced to protect the equipment. This ensures the cooling safety of the electrodes and furnace walls while achieving on-demand distribution and maximum utilization of waste heat resources. Based on the waste gas calorific value Q, the controller coordinates the operation airflow of the dust removal fan system and the waste heat recovery mode; the Q value determines the energy recovery value of the waste gas. The control objective is high quality and high utilization, low quality and low consumption. When the Q value is higher than the preset high-grade threshold (Q_th_high), the controller determines that the waste gas has high recovery value. It simultaneously outputs instructions: increase the operation airflow of the dust removal fan system to capture more high-grade waste gas and prevent its escape; at the same time, it sends instructions to the waste heat recovery system to switch to a high-efficiency recovery mode to maximize energy conversion efficiency. When the Q value is lower than the preset low-grade threshold (Q_th_low), the controller synchronously outputs a command: reduce the dust collector fan's airflow to the minimum level required to maintain environmental protection to save energy, and switch the waste heat recovery system to energy-saving operation mode or bypass mode. This achieves a dynamic optimal balance between waste gas energy recovery efficiency and dust collection system operating energy consumption, avoiding the negative effect of consuming a large amount of fan power to recover a small amount of low-grade heat.

[0035] Step S311: Synchronously adjust the input power of the electrode system according to the total system heat load H, including coordinated temperature control of the superconducting heat-conducting plate inside the electrode. The specific steps are as follows: This step is a further execution of the instruction to adjust the electrode system according to the total system heat load H. Adjusting the electrode power is not only for controlling the molten pool temperature, but also a critical thermal management event, requiring simultaneous management of the resulting electrode heating. By introducing a superconducting heat-conducting plate that is linked to the electrode power adjustment, this invention achieves active thermal management of the electrode body, thereby solving the industry problems of easy overheating and loss of the electrode under high power and insufficient waste heat recovery under low power.

[0036] Real-time monitoring of the electrode body operating temperature Te and the electric arc furnace molten pool temperature Tf; Te and Tf are two independent but related key state variables. Electrode body operating temperature Te: monitored in real time using a high-precision infrared thermometer or a high-temperature resistant embedded thermocouple embedded below the electrode holder and inside the electrode body. This measurement point must reflect the true operating temperature of the electrode under arc heating and furnace radiant heat. Electric arc furnace molten pool temperature Tf: measured periodically or continuously using a secondary lance probe installed on the furnace wall or a radiation pyrometer installed on the furnace top. This provides first-hand data on the electrode's thermal state and the core state of the smelting process, offering direct input for subsequent collaborative decision-making. Based on the deviation between the electric arc furnace molten pool temperature Tf and the target smelting temperature, the required electrode power adjustment ΔP is predicted; this step is the core function of a traditional electrode regulator, aiming to maintain the stability of the molten pool process temperature. A target smelting temperature T_target is set. The real-time deviation ΔT = Tf - T_target is calculated. Based on the magnitude and trend of ΔT, the electrode power adjustment ΔP required to correct this deviation is calculated using a preset incremental PID algorithm. For example, if Tf is low, ΔP is positive, indicating that power needs to be increased; conversely, it is negative. This ensures the process stability of the smelting process and provides an initial power adjustment drive signal for the entire coordinated control. Based on the electrode power adjustment ΔP and the electrode body operating temperature Te, the thermal conductivity adjustment ΔΦ of the superconducting heat-conducting plate inside the electrode is determined. The determination of ΔΦ follows a dual-objective decision-making mechanism: changes in electrode power (ΔP) inevitably cause changes in electrode heat generation. According to the law of conservation of energy, ΔΦ should be positively correlated with ΔP. That is, when the power increases by ΔP, the expected electrode heat generation will increase, so it is necessary to increase the heat dissipation ΔΦ of the heat-conducting plate in advance or simultaneously to offset the additional heat generation and prevent the electrode from overheating. This can be a linear or nonlinear mapping relationship established based on historical data or a physical model: ΔΦ_base = f(ΔP). The real-time monitored Te is the final correction benchmark. The system presets an optimized temperature range for the electrode body [Te_min, Te_max]. This range comprehensively considers the temperature range with the lowest electrode oxidation rate, optimal mechanical strength, and effective heat recovery. When Te is within [Te_min, Te_max], the final ΔΦ is mainly based on ΔΦ_base, i.e., it is adjusted according to power changes. When Te exceeds [Te_min, Te_max], this condition takes priority: regardless of the value of ΔP, the system will prioritize initiating the temperature protection program. It calculates an additional ΔΦ_override coverage amount directed towards enhanced heat dissipation, such that ΔΦ = ΔΦ_base + ΔΦ_override, with the goal of quickly pulling Te back to the safe range. This achieves dynamic decoupling and coordination between power regulation and temperature protection, ensuring both process response and equipment safety, avoiding the traditional contradiction of sacrificing equipment for process optimization.By adjusting the heat dissipation power of the heat exchanger at the far end of the superconducting heat-conducting plate, the thermal conductivity adjustment amount ΔΦ is controlled, maintaining the electrode body operating temperature Te within the preset optimized temperature range [Te_min, Te_max]. ΔΦ is a decision variable that needs to be translated into specific equipment actions. The thermal conductivity Φ is proportional to the temperature difference between the two ends of the superconducting heat-conducting plate. Adjusting the heat dissipation power of the far end heat exchanger is the most effective means to change its cold end temperature, thereby increasing the temperature difference and enhancing thermal conductivity. The heat exchanger is usually a finned tube air cooler or a water cooler. The specific way to adjust the heat dissipation power is to control the speed of the cooling fan through a frequency converter or to control the flow rate of cooling water through a regulating valve. The system establishes a correspondence between ΔΦ and the fan speed increment ΔFanSpeed ​​or the valve opening increment ΔValveOpen. Upon receiving the ΔΦ command, the actuator is driven to make the corresponding adjustment. The control command is accurately translated into physical action, forming a complete closed loop from state perception to decision-making to execution, realizing active and precise control of the electrode temperature. The thermal conductivity adjustment ΔΦ is positively correlated with the electrode power adjustment ΔP, and adjustment is prioritized when Te exceeds the optimized temperature range. This precise control of electrode temperature enhances the overall heat exchange efficiency K, achieving the dual effects of reduced electrode loss and improved energy utilization efficiency. The electrode oxidation loss rate is exponentially related to temperature. Maintaining Te precisely within the optimized range of [Te_min, Te_max] prevents the electrode from entering the high-temperature, high-speed oxidation zone due to overheating. This significantly extends the electrode's lifespan and reduces electrode consumption, which translates to a substantial reduction in production costs for expensive graphite electrodes. Stable electrode temperature also leads to more stable resistivity, reducing additional energy loss caused by temperature fluctuations. Furthermore, preventing overheating itself saves energy. The heat extracted through the superconducting heat plate is high-temperature, high-quality heat. Effectively recovering it is equivalent to directly converting a portion of the input electrical energy into usable heat energy, achieving gradient energy utilization. This step, through precise control of electrode temperature, ensures that the superconducting heat plate system operates under efficient and stable conditions. This is directly reflected in the stability and optimization of the overall heat exchange efficiency K value. A stable and efficient K value means that the heat extraction system is predictable and controllable, which lays a solid foundation for the overall system-level energy synergy optimization. This step is not an isolated temperature control, but a collaborative innovation that deeply integrates process control, equipment protection, and energy recovery. By precisely controlling the key node of electrode temperature, it achieves the dual energy-saving effects of reducing electrode consumption and recovering high-quality waste heat, thus jointly promoting a leap in the overall energy utilization efficiency of the system from two dimensions.

[0037] This step aims to solve the problem of thermal equilibrium within the furnace. Step S312 dynamically coordinates the carbon powder injection rate and the heat dissipation intensity of the superconducting heat-conducting plate system based on the total system heat load H and the furnace temperature distribution. Carbon powder injection and superconducting heat-conducting plate heat dissipation are two key means of controlling furnace temperature, but traditionally they are controlled independently. This invention achieves coordinated temperature management with the goal of minimizing total energy consumption by establishing a dynamic coordination relationship between the two. Multiple thermocouples or infrared thermometers installed at different heights and orientations on the furnace wall are used to construct a furnace temperature distribution field and identify local overheating areas in real time. When local overheating is detected, the controller synchronously executes the following instructions: Increase the carbon powder injection rate in that area: Enhanced injection can quickly form a foamed slag layer in the overheated area. The foamed slag can effectively shield the electric arc, reducing the strong radiation of the arc to the furnace lining, and simultaneously consume local excess heat through endothermic reactions. Enhance the heat transfer intensity of the superconducting heat-conducting plate in the corresponding area: By adjusting the cooling power of the heat exchanger at the far end of the heat-conducting plate in that area, excess heat is directly discharged from the furnace wall. The foamy slag formed by the toner provides internal thermal shielding and chemical temperature regulation, while the heat-conducting plate provides external physical heat conduction. These two processes occur simultaneously, rapidly mitigating localized overheating, achieving balanced temperature control, effectively preventing furnace lining burn-off, and improving heating uniformity. A correlation model between toner consumption and heat dissipation is established, optimizing their ratio with the goal of minimizing total system energy consumption. Both toner consumption and heat dissipation system operation are costs. They are interchangeable under different operating conditions. For example, sometimes injecting more toner to form good foamy slag can significantly reduce heat dissipation requirements; while sometimes enhanced heat dissipation can reduce expensive toner consumption. A function describing the relationship between total system energy consumption and toner rate (V) and heat dissipation (Q) is constructed, and an optimization algorithm is used to find the (V, Q) combination that minimizes total energy consumption. The total system heat load H, furnace temperature distribution, current toner rate V, current heat dissipation Q, and corresponding total system power are collected in real time. Based on real-time data, the optimal solution set [V1, Q1] is dynamically solved using an optimization algorithm. V1 is converted into the speed control command of the feeder in the toner injection system; Q1 is converted into the speed or valve opening command of the medium circulation pump in the superconducting heat transfer plate system. By comparing the difference between the actual energy consumption and the model's predicted energy consumption, the model parameters are periodically corrected to achieve model self-calibration and adapt to changes in raw materials or equipment performance degradation.

[0038] The correlation model between toner consumption and heat dissipation is as follows: An energy consumption optimization function E is constructed with toner injection rate V and the equivalent heat dissipation Q of the superconducting heat-conducting plate system as input variables. The calculation expression of the energy consumption optimization function E is: E = a·V + b·Q + c·(T1 - T2)², where a, b, and c are weighting coefficients, T1 is the target furnace temperature, and T2 is the actual furnace temperature; a·V: Toner consumption cost item. a is the toner cost coefficient, converting toner consumption into equivalent energy consumption or economic cost. b·Q: Heat dissipation system operating cost item. b is the heat dissipation system energy consumption coefficient, representing the electrical energy consumed per unit of heat output. c·(T1 - T2)²: Process quality penalty item. c is the penalty coefficient. This item ensures that the optimization result will not deviate significantly from the target temperature T1 required by the process requirements in order to unilaterally pursue low energy consumption. The larger the temperature deviation, the penalty value increases quadratically, forcing the optimization solution to approach the process requirements. Based on real-time collected data of the system's total heat load H and furnace temperature distribution, the optimal solution set [V1, Q1] that minimizes the value of energy E is dynamically calculated using the gradient descent method. The currently collected T2 is substituted into the function E. The gradient descent method searches along the direction that minimizes the value of E within the domains of V and Q. After several iterations, it finds the set of V and Q values ​​that minimizes E, i.e., [V1, Q1]. Since H and T2 change in real time, [V1, Q1] is also a dynamically updated optimal setpoint, achieving real-time optimization of control. An optimization effect feedback mechanism is established. By comparing the difference between actual energy consumption and predicted energy consumption, the weight coefficients a, b, and c are periodically adjusted to achieve self-calibration of the model parameters. The initial a, b, and c are set based on experience or historical data. After the system is running, the actual total energy consumption E_actual can be collected. E_actual is then compared with the predicted value E_predicted calculated by the model at that time. If systematic biases exist, a parameter identification algorithm is used to refit the a, b, and c values ​​using operational data over a period of time, better reflecting the current system characteristics. This gives the model adaptive capabilities, enabling it to maintain optimization accuracy and ensure long-term energy-saving effects by adapting to slowly changing factors such as equipment aging, changes in media performance, and fluctuations in toner quality.

[0039] The heat-conducting plate is a sealed vacuum cavity structure, with the shell made of nickel-based high-temperature alloy material and a wall thickness of 1.0-1.5 mm. The vacuum cavity is filled with a FUD inorganic superconducting medium, composed of a Li₂CO₃-K₂CO₃-Cs₂CO₃ ternary eutectic salt matrix, 0.6 wt% graphene thermal conductivity enhancer, and 0.4 wt% CeO₂ thermal stabilizer. The inner surface of the shell is formed with a microcapillary channel structure through laser cladding, with a channel width of 50-80 μm and a depth of 30-50 μm. This eliminates heat loss caused by internal gas convection, allowing heat transfer to rely primarily on the efficient phase change conduction of the medium, thereby achieving extremely high heat transfer efficiency at the superconducting level. The nickel-based high-temperature alloy is selected from grades such as Inconel 600 / 601, ensuring excellent creep resistance and oxidation resistance in the high-temperature, oxidizing, and corrosive environment of the electric arc furnace. Wall thickness 1.0-1.5mm: While ensuring structural strength and pressure resistance, the wall thickness is minimized to reduce thermal resistance. The superconducting heat-conducting plate is bonded to the electrodes and furnace wall substrate via high-temperature welding, forming a closed-loop circulation with the remote heat exchanger. The remote heat exchanger adopts a finned tube structure, and the heat dissipation intensity is continuously controlled by adjusting the cooling fan speed. Li2CO3-K2CO3-Cs2CO3 ternary eutectic salt: As a phase change working medium, it has high latent heat of vaporization and a suitable operating temperature range, enabling efficient operation under electric arc furnace conditions. 0.6wt% graphene: As a thermal conductivity enhancer, it significantly improves the equivalent thermal conductivity of the composite medium. 0.4wt% CeO2: As a thermal stabilizer, it effectively inhibits the thermal decomposition of the eutectic salt at long-term high temperatures, extending the medium's lifespan. A fine channel structure is formed on the inner wall using laser cladding technology. These channels generate powerful capillary pumping force, enabling the condensed liquid working fluid to be rapidly and evenly returned to the hot end, greatly enhancing the circulation capacity of the working fluid, preventing localized drying, and improving the heat transfer limit and reliability of the heat-conducting plate. The heat-conducting plate is metallurgically bonded to the substrate through high-temperature brazing or electron beam welding, ensuring extremely low thermal contact resistance. It forms a loop with the distal finned tube heat exchanger, and the speed of the axial cooling fan is adjusted via a frequency converter to achieve precise and continuous control of the heat dissipation power.

[0040] Step S313: Based on the calorific value Q of the exhaust gas, coordinate the operation airflow of the dust removal fan system and the waste heat recovery mode. The specific steps are as follows: Monitor Tg, Cg, and Fg in real time using a flue gas analyzer and thermocouples. Calculate the recoverable amount Er and calorific value Q, as described above. When the calorific value Q of the exhaust gas is higher than the preset threshold Q_th, activate the high-efficiency waste heat recovery mode, simultaneously increase the operating airflow of the dust removal fan system to enhance the exhaust gas capture efficiency, and adjust the heat exchange rate of the heat exchanger system to maximize waste heat utilization; High-grade strategy: When Q>Q_th, it indicates that the exhaust gas has high energy quality and high value. Simultaneously increase the dust removal fan airflow: by increasing the frequency of the fan inverter, the suction force is increased, with the aim of capturing as much high-grade exhaust gas as possible and transporting it to the waste heat recovery device to prevent its dissipation and waste. At the same time, send instructions to the waste heat boiler to increase the feedwater flow rate, adjust the pressure and temperature setpoints, and make it operate at the optimal operating point to maximize heat conversion efficiency. When the calorific value Q of the exhaust gas is lower than the preset threshold Q_th, the system switches to energy-saving operation mode, reducing the operating airflow of the dust removal fan system to decrease energy consumption while maintaining basic exhaust gas treatment requirements. Low-grade strategy: When Q ≤ Q_th, it indicates low energy value of the exhaust gas, and the recovery benefit may not offset the fan's power consumption. The fan speed is reduced to the minimum airflow required to meet environmental emission standards, significantly reducing the fan's own power consumption. The waste heat recovery system can switch to low-load operation or bypass mode, prioritizing dust removal efficiency. A correlation model between the exhaust gas calorific value Q and the total system heat load H is established. The AI ​​collaborative controller dynamically optimizes the dust removal fan airflow setpoint and waste heat recovery parameters to achieve a balance between exhaust gas energy recovery and system energy consumption. Through training with historical data, a strong correlation between Q and H is found. This model can be used to predict the short-term trend of Q. The AI ​​controller uses this model for predictive control. For example, if it is predicted that Q will rise, the fan air volume can be increased in advance and gradually, rather than suddenly adjusting after Q exceeds the standard, so that the control is more stable and efficient, and ultimately achieves a dynamic global optimal balance between the energy recovery benefits of waste gas and the operating cost of the dust removal system.

[0041] Step S314: Dynamically adjust the heat dissipation intensity and heat distribution path of the heat exchanger system according to the comprehensive heat exchange efficiency K and the ambient temperature Ta. The specific steps are as follows: Monitor Tin, Tout, Fm, and calculate the actual K value. Monitor Ta through an ambient temperature and humidity sensor. Compare the actual comprehensive heat exchange efficiency K with the target heat exchange efficiency K_target to generate a heat dissipation intensity adjustment amount ΔK; if K < K_target, it means that the heat dissipation capacity is insufficient, generate a positive ΔK, and the instruction is to strengthen heat dissipation; otherwise, generate a negative ΔK, and the instruction is to weaken heat dissipation to save energy. At the same time, monitor the ambient temperature Ta and correct the priority of the heat distribution path according to the ambient temperature Ta; when Ta is relatively low: it means that there is a heating demand in the factory area. At this time, preferentially distribute the heat to the factory area hot water system or the heating pipe network to achieve high-value utilization of energy. When Ta is relatively high: there is no heating demand, and it is difficult for the equipment to dissipate heat. At this time, preferentially strengthen heat dissipation, and discharge the heat to the atmosphere through a cooling tower, etc., to protect the heat exchanger system itself from overheating and maintain its efficient operation. Control the heat dissipation intensity adjustment amount ΔK by adjusting the speed of the cooling fan of the heat exchanger system to keep the actual comprehensive heat exchange efficiency K within the target range; and dynamically switch the heat output path using a heat distribution valve. ΔK is converted into a frequency command for the cooling fan inverter to achieve continuous adjustment of the heat dissipation intensity. Switch the pipeline through a controllable three-way valve or a series of electric valves to achieve dynamic switching of the heat between paths such as factory area heating and environmental heat dissipation.

[0042] Step S5: Adaptive model update and optimization: Based on the system feedback data collected in Step S4, regularly update the dynamic coupling model based on the graph neural network and the reinforcement learning agent trained based on the proximal policy optimization algorithm. The specific steps are as follows: Endow the system with lifelong learning ability to overcome the performance degradation of the model and controller caused by equipment aging, raw material changes, etc. Collect historical operation data, including equipment operation parameters, control instruction sets, and corresponding system performance indicators; long-term store all the data collected in S1, all the instructions issued in S3, and the calculated energy efficiency indicators in the central database. Calculate the deviation between the actual energy efficiency indicator and the predicted energy efficiency indicator to evaluate the model performance; when the deviation exceeds the preset tolerance, trigger the model update process; regularly calculate the deviation between the recent average actual energy consumption and the predicted values of the GNN model and RL agent every week. Set a tolerance. If the deviation continuously exceeds this tolerance, automatically trigger the model retraining process.

[0043] The graph neural network (GNN) dynamic coupling model was retrained using historical data to optimize the connection weights of energy and material flows between nodes. The latest, larger-scale historical dataset was used to retrain the GNN model, updating the edge weights to more accurately characterize the dynamic coupling relationships between devices in the current system. Simultaneously, a reinforcement learning agent interacted with the updated dynamic coupling model, adjusting policy parameters to adapt to system changes. The updated GNN model was used as a simulation environment, allowing the PPO agent to perform millions of simulations, exploring and learning, adjusting its neural network policy parameters to learn how to make better collaborative decisions under new system characteristics. The accuracy and stability of the updated model were verified, and it was deployed to the real-time control system after simulation testing. A model version management mechanism was established to ensure the continuity and reliability of the control system. Testing and Deployment: The updated model underwent rigorous testing in an offline simulation environment to confirm its performance improvement and stability before being smoothly switched to the online system to replace the old model. Version records and backups were maintained for each model update. If unexpected problems occurred after the new model was deployed, it could be quickly rolled back to the previous stable version, greatly improving the engineering reliability and availability of the entire AI control system.

[0044] Example 2 This embodiment provides an industrial energy-saving AI intelligent control system for implementing the aforementioned method, and the claims protect the hardware system entity implementing the method. The system includes: a global data sensing module, comprising a sensor network deployed on multiple coordinating devices within the electric arc furnace system, for real-time acquisition of operational data from the coordinating devices; the coordinating devices include: an electrode system, a furnace wall cooling system, a carbon powder injection system, a dust removal fan system, a heat exchanger system, and heat-conducting plates integrated into the electrodes and furnace wall; the hardware components include all the sensors described in step S1 and their signal conditioning circuits, a data acquisition card, and an industrial network switch, wherein the sensors include thermocouples, flow meters, power transmitters, gas analyzers, etc. The hardware is responsible for converting physical world signals into digital signals and converging them to the central processing unit.

[0045] The collaborative state modeling and decision-making center, communicatively connected to the global data sensing module, includes: a dynamic coupling model processing unit based on a graph neural network, configured to map the multiple collaborative devices as nodes in a graph, establish energy and material flow connections between nodes, and output a global state vector characterizing the overall system operating state; and an AI collaborative controller trained based on a near-end policy optimization algorithm, configured to receive the global state vector and output a unified multi-dimensional collaborative control instruction set. The AI ​​collaborative controller uses the system's total heat load H, the comprehensive heat exchange efficiency K of the heat exchanger system at the far end of the superconducting heat plate, and the waste gas heat grade Q as core thermal state parameters, serving as the benchmark for coordinated control. The hardware is typically carried by a high-performance industrial server or industrial control computer. The software consists of a GNN processing unit: a software module deployed on a server, loaded with a trained GNN model, responsible for executing step S2. The AI ​​collaborative controller is another software module, encapsulating a trained PPO agent, responsible for executing steps S3 and S31. Communication is achieved through industrial communication protocols such as OPC UA and MQTT for data interaction with the sensing module and execution network.

[0046] A collaborative execution and feedback network, communicatively connected to the collaborative state modeling and decision-making center, includes actuators connected to the multiple collaborative devices. These actuators receive and execute the multi-dimensional collaborative control command set, dynamically adjusting the operating parameters of each device. The sensor network further collects system feedback data after command execution and sends this feedback data back to the collaborative state modeling and decision-making center to form a closed-loop control. This includes all actuators such as frequency converters controlling fans, water pumps, servo drives controlling feeders, and intelligent regulating valves. The network receives digital commands from the decision-making center and converts them into analog or switching signals that can drive the physical devices. After execution, the sensor network collects data again and sends it back to the decision-making center through the sensing module, thus completing a complete perception-decision-execution-feedback control closed loop.

[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the article or apparatus that includes that element.

[0048] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. 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 be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An industrial energy-saving AI intelligent control method for multi-device collaboration, characterized in that, The method includes the following steps: S1: Global Data Awareness: Real-time acquisition of operating data of multiple coordinating devices in the electric arc furnace system, including: electrode system, carbon powder injection system, dust removal fan system, heat exchanger system, and heat-conducting plate integrated into the electrode and furnace wall; S2: Cooperative state modeling: The data collected in S1 is input into a dynamic coupling model based on graph neural network. This model maps the multiple cooperative devices as nodes in the graph, establishes energy flow and material flow connection edges between nodes, and outputs a global state vector that characterizes the overall operating state of the system. S3: AI Collaborative Decision Making: The global state vector is input into a reinforcement learning agent trained based on a proximal policy optimization algorithm, and the agent outputs a multi-dimensional collaborative control instruction set. S31: Real-time monitoring of the core thermal state parameters of the electric arc furnace system, including the total system heat load H, the comprehensive heat exchange efficiency K of the heat exchanger system at the far end of the heat conduction plate, and the heat grade Q of the exhaust gas; using the core thermal state parameters as the benchmark for linkage control, a unified multi-dimensional collaborative control instruction set is generated and executed through an AI collaborative controller. S4: Multi-device collaborative execution and feedback: Execute the multi-dimensional collaborative control instruction set, dynamically adjust the operating parameters of the multiple collaborative devices, and collect system feedback data to form closed-loop control.

2. The industrial energy-saving AI intelligent control method for multi-device collaboration according to claim 1, characterized in that, In step S31, the total system heat load H is a comprehensive index calculated based on the power of the electrode system and the heat exchanger system; the comprehensive heat exchange efficiency K is a performance index determined based on the heat output per unit time, the effective area of ​​the heat conduction plate, and the temperature difference between the heat exchanger system at the hot end and the far end; the exhaust gas heat grade Q is a heat grade evaluation value that integrates the physical sensible heat of the exhaust gas and the chemical energy of its combustible components. Using the three core thermal state parameters as a unified control benchmark, an AI collaborative controller generates a unified multi-dimensional collaborative control instruction set. This instruction set synchronously coordinates the operating status of the following devices: based on the total system heat load H, synchronously adjusts the input power of the electrode system and the carbon powder injection rate of the carbon powder injection system; based on the comprehensive heat exchange efficiency K, dynamically adjusts the heat dissipation intensity and heat distribution path of the heat exchanger system; and based on the waste gas heat grade Q, collaboratively controls the operating air volume and waste heat recovery mode of the dust removal fan system.

3. The industrial energy-saving AI intelligent control method for multi-device collaboration according to claim 2, characterized in that, Step S311: Synchronously adjust the input electrical power of the electrode system according to the total system heat load H, including the coordinated temperature control of the internal heat-conducting plate of the electrode. The specific steps are as follows: The electrode body operating temperature Te and the electric arc furnace molten pool temperature Tf are monitored in real time. Based on the deviation between the electric arc furnace molten pool temperature Tf and the target smelting temperature, the required electrode power adjustment ΔP is predicted. According to the electrode power adjustment ΔP and the electrode body operating temperature Te, the thermal conductivity intensity adjustment ΔΦ of the heat-conducting plate inside the electrode is determined. By adjusting the heat dissipation power of the heat exchanger at the far end of the heat-conducting plate, the thermal conductivity intensity adjustment ΔΦ is controlled so that the electrode body operating temperature Te is maintained within the preset optimized temperature range [Te_min, Te_max]. The thermal conductivity intensity adjustment ΔΦ is positively correlated with the electrode power adjustment ΔP, and when Te exceeds the optimized temperature range, priority is given to adjustment. In this way, the comprehensive heat exchange efficiency K is enhanced through precise control of the electrode temperature, achieving the dual effect of reducing electrode loss and improving energy utilization efficiency.

4. The industrial energy-saving AI intelligent control method for multi-device collaboration according to claim 2, characterized in that, Step S312: Based on the total system heat load H and the furnace temperature distribution, dynamically coordinate the carbon powder injection rate with the heat dissipation intensity of the heat-conducting plate system. When local overheating is detected in the furnace, the carbon powder injection rate and the heat exchange intensity of the corresponding heat-conducting plate are increased simultaneously. Temperature balance control is achieved by strengthening the formation of the foam slag layer and enhancing heat dissipation. Establish a correlation model between toner consumption and heat dissipation, and optimize the ratio between the two with the goal of minimizing the total system energy consumption.

5. The industrial energy-saving AI intelligent control method for multi-device collaboration according to claim 4, characterized in that, The correlation model between carbon powder consumption and heat dissipation is as follows: an energy consumption optimization function E is constructed with carbon powder injection rate V and equivalent heat dissipation Q of the heat conduction plate system as input variables; the calculation expression of the energy consumption optimization function E is: E = a·V + b·Q + c·(T1 - T2)², where a, b, and c are weighting coefficients, T1 is the target furnace temperature, and T2 is the actual furnace temperature; Based on real-time collected data of total system heat load H and furnace temperature distribution, the optimal solution set [V1, Q1] that minimizes E is dynamically solved by gradient descent method; the optimal solution set is then converted into control commands to synchronously adjust the feeder speed of the carbon powder injection system and the medium circulation rate of the heat-conducting plate system. Establish an optimization effect feedback mechanism, and periodically adjust the weight coefficients a, b, and c by comparing the difference between actual energy consumption and predicted energy consumption to achieve self-calibration of model parameters.

6. The industrial energy-saving AI intelligent control method for multi-device collaboration according to claim 1, characterized in that, The heat-conducting plate is a sealed vacuum cavity structure, and the shell is made of nickel-based high-temperature alloy material with a wall thickness of 1.0-1.5mm. The vacuum cavity is filled with FUD inorganic superconducting medium, which consists of a Li2CO3-K2CO3-Cs2CO3 ternary eutectic salt matrix, 0.6wt% graphene thermal conductivity enhancer, and 0.4wt% CeO2 thermal stabilizer. The inner surface of the shell is formed with a microcapillary channel structure through laser cladding process, with a channel width of 50-80μm and a depth of 30-50μm. The heat-conducting plate is bonded to the electrode and furnace wall substrate through a high-temperature welding process, and forms a closed loop with the remote heat exchanger; the remote heat exchanger adopts a finned tube structure, and the heat dissipation intensity is continuously controlled by adjusting the speed of the cooling fan.

7. The industrial energy-saving AI intelligent control method for multi-device collaboration according to claim 4, characterized in that, Step S313: Based on the waste gas calorific value Q, coordinate the operating air volume and waste heat recovery mode of the dust removal fan system. The specific steps are as follows: The system monitors the exhaust gas temperature (Tg), exhaust gas component concentration (Cg), and exhaust gas flow rate (Fg) in real time. Based on these parameters, it calculates the real-time values ​​of the recoverable waste heat (Er) and the exhaust gas calorific value (Q). When the exhaust gas calorific value (Q) is higher than a preset threshold (Q_th), it activates a high-efficiency waste heat recovery mode, simultaneously increasing the operating airflow of the dust removal fan system to enhance exhaust gas capture efficiency and adjusting the heat exchange rate of the heat exchanger system to maximize waste heat utilization. When the exhaust gas calorific value (Q) is lower than the preset threshold (Q_th), it switches to an energy-saving operation mode, reducing the operating airflow of the dust removal fan system to decrease energy consumption while maintaining basic exhaust gas treatment requirements. A correlation model is established between the exhaust gas calorific value (Q) and the total system heat load (H). An AI collaborative controller dynamically optimizes the dust removal fan airflow setpoint and waste heat recovery parameters to achieve a balance between exhaust gas energy recovery and system energy consumption.

8. The industrial energy-saving AI intelligent control method for multi-device collaboration according to claim 2, characterized in that, Step S314: Based on the overall heat exchange efficiency K and the ambient temperature Ta, dynamically adjust the heat dissipation intensity and heat distribution path of the heat exchanger system. The specific steps are as follows: The system monitors the inlet temperature Tin, outlet temperature Tout, and vaporization coefficient Fm of the liquid medium in real time. Based on these parameters, the actual overall heat exchange efficiency K is calculated. The actual overall heat exchange efficiency K is compared with the target heat exchange efficiency K_target to generate a heat dissipation intensity adjustment ΔK. Simultaneously, the ambient temperature Ta is monitored, and the priority of the heat distribution path is adjusted according to the ambient temperature Ta. When the ambient temperature Ta is low, heat is preferentially distributed to the plant's hot water system. When the ambient temperature Ta is high, heat dissipation is prioritized to protect the equipment; the actual comprehensive heat exchange efficiency K is maintained within the target range by adjusting the cooling fan speed of the heat exchanger system and controlling the heat dissipation intensity adjustment amount ΔK; and the heat distribution valve is used to dynamically switch the heat output path.

9. The industrial energy-saving AI intelligent control method for multi-device collaboration according to claim 1, characterized in that, It also includes step S5: Adaptive model update and optimization: Based on the system feedback data collected in step S4, the dynamically coupled model based on graph neural networks and the reinforcement learning agent trained based on the proximal policy optimization algorithm are updated periodically. The specific steps are as follows: Collect historical operating data, including equipment operating parameters, control command sets, and corresponding system performance indicators; calculate the deviation between actual energy efficiency indicators and predicted energy efficiency indicators, and evaluate model performance; When the deviation exceeds the preset tolerance, the model update process is triggered; The graph neural network dynamic coupling model is retrained using historical data to optimize the connection weights of energy flow and material flow between nodes. At the same time, a reinforcement learning agent interacts with the updated dynamic coupling model to adjust policy parameters to adapt to system changes. The accuracy and stability of the updated model are verified, and it is deployed to the real-time control system after simulation testing. A model version management mechanism is established to ensure the continuity and reliability of the control system.

10. An industrial energy-saving AI intelligent control system for implementing the method of any one of claims 1-9, characterized in that, The system includes: a global data sensing module, comprising a sensor network deployed on multiple coordinating devices in the electric arc furnace system, for real-time acquisition of the operating data of the coordinating devices; the coordinating devices include: an electrode system, a carbon powder injection system, a dust removal fan system, a heat exchanger system, and a heat-conducting plate integrated into the electrode and the furnace wall; The collaborative state modeling and decision-making center, communicatively connected to the global data perception module, includes: a dynamic coupling model processing unit based on a graph neural network, configured to map the multiple collaborative devices as nodes in a graph, establish energy flow and material flow connection edges between nodes, and output a global state vector characterizing the overall operating state of the system; and an AI collaborative controller trained based on a near-end strategy optimization algorithm, configured to receive the global state vector and output a unified multi-dimensional collaborative control instruction set; the AI ​​collaborative controller uses the total system heat load H, the comprehensive heat exchange efficiency K of the heat exchanger system at the far end of the heat conduction plate, and the waste gas heat grade Q as core thermal state parameters as the benchmark for linkage regulation; The collaborative execution and feedback network is communicatively connected to the collaborative state modeling and decision-making center. It includes actuators connected to the multiple collaborative devices for receiving and executing the multi-dimensional collaborative control instruction set and dynamically adjusting the operating parameters of each device. The sensor network is further used to collect system feedback data after instruction execution and send the feedback data back to the collaborative state modeling and decision-making center to form closed-loop control.

Images