Big data-based industrial silicon smelting energy consumption data management platform

Abstract

This invention belongs to the field of industrial data processing and energy efficiency management, specifically involving a big data-based energy consumption data management platform for industrial silicon smelting. The platform includes a data acquisition and sensing device, a working condition identification unit, a production scheduling construction component, an energy consumption model analysis engine, a collaborative optimization execution system, a multi-objective evaluation terminal, and a real-time feedback module. The collaborative optimization execution system integrates a Pareto evolutionary algorithm based on Gaussian learning, using a Gaussian mixture model to fit the elite solution distribution and adaptively adjust the search strategy; the energy consumption analysis engine introduces a non-standard loss correction coefficient to establish an energy efficiency mapping; and the multi-objective evaluation terminal performs hierarchical evaluation of the solutions. This invention achieves deep integration of production planning and energy consumption control, improves the efficiency of finding the optimal production scheduling solution under complex working conditions, and achieves closed-loop scheduling optimization with performance balance among carbon emissions, delay costs, and load balancing, providing decision support for the green transformation of industrial silicon smelting.

Description

Technical Field

[0001] This invention belongs to the field of industrial data processing and energy efficiency management, specifically involving an industrial silicon smelting energy consumption data management platform based on big data. Background Technology

[0002] With the continuous advancement of green manufacturing and digital transformation in industry, energy consumption data management platforms in the industrial silicon smelting sector have become a key infrastructure for improving production efficiency and reducing energy intensity. Traditional smelting management systems mainly focus on equipment status monitoring and basic energy consumption statistics, aiming to achieve real-time collection of operating parameters of submerged arc furnaces through digital means. Under the constraints of dual carbon targets, the industrial silicon smelting process places higher demands on the refined control of energy consumption, requiring not only monitoring single-dimensional electricity consumption but also establishing a deep logical connection between monthly production plans and the entire process carbon footprint.

[0003] The coordinated optimization of production scheduling and energy consumption control is a core function of smelting management platforms. In existing technologies, the production scheduling of industrial silicon is often disconnected from actual energy consumption control, leading to production decisions that primarily focus on meeting output targets and delivery deadlines. Multiple heterogeneous submerged arc furnaces generate non-quota energy losses during start-up, shutdown, or load adjustments. This requires the management platform to perform complex trade-off analyses on multiple conflicting objectives, such as carbon emission constraints, delay penalty costs, and power load balancing, to find the optimal balance point for system operation.

[0004] Traditional scheduling and management models often rely on linear programming or simple heuristic algorithms, making it difficult to analyze the nonlinear coupling relationship between production orders and complex energy consumption fluctuations. This results in poor adaptability of production scheduling schemes under dynamic operating conditions. Existing algorithms, when dealing with high-dimensional multi-objective optimization problems, lack the ability to deeply mine the distribution characteristics of the excellent solution space, making it difficult to adaptively fit the multimodal distribution characteristics present in the smelting process. Due to the lack of a learning mechanism for the implicit characteristics of the production process, the management platform cannot identify special production scheduling patterns that can reduce global energy consumption, often falling into the dilemma of optimal efficiency for a single furnace but uncontrolled global energy consumption, making it difficult to achieve synergistic maximization of carbon emissions and production efficiency on a monthly scale. Summary of the Invention

[0005] The purpose of this invention is to provide an industrial silicon smelting energy consumption data management platform based on big data, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] The industrial silicon smelting energy consumption data management platform based on big data includes data acquisition and sensing devices, production condition identification units, smelting production scheduling components, energy consumption model analysis engines, collaborative optimization execution systems, multi-objective evaluation and processing terminals, and real-time feedback control modules, as follows:

[0008] The data acquisition and sensing device is used to acquire production and operation data of multiple heterogeneous submerged arc furnaces in real time, including but not limited to electrode parameters, furnace temperature, raw material ratio, and current and voltage characteristics, and to standardize the production and operation data to eliminate communication differences between heterogeneous devices.

[0009] The production condition identification unit is used to identify the current smelting stage based on the collected production operation data, analyze the energy flow characteristics of different submerged arc furnaces during start-up and shutdown or load adjustment, and focus on extracting the feature vector of non-quota energy loss.

[0010] The smelting production scheduling component is used to receive monthly production plan instructions and, in conjunction with the real-time status, maintenance plans, and delivery requirements of each heterogeneous submerged arc furnace, establish an initial production task sequence under multiple constraints.

[0011] The energy consumption model analysis engine is used to construct a nonlinear mapping relationship between production scheduling and energy consumption emissions. It uses historical big data to train and obtain the energy efficiency curves of each heterogeneous furnace, and predicts the cumulative energy efficiency and instantaneous carbon emission intensity under different production scheduling strategies.

[0012] The collaborative optimization execution system is the core computing component of the platform. It integrates a Pareto evolution algorithm based on Gaussian learning, which is used to adaptively adjust the search strategy by learning the distribution characteristics of the elite solution set in the search space, and find the globally optimal solution set in the huge scheduling scheme space.

[0013] The multi-objective evaluation and processing terminal is used to evaluate the optimized production scheduling plan from multiple dimensions, balancing multiple conflicting objectives such as carbon emission intensity, delay penalty cost, and power load balance.

[0014] The real-time feedback control module is used to convert the final determined optimal production scheduling plan into specific equipment control commands and send them to the actuators of the electric arc furnace. It performs closed-loop adjustment based on the deviation between the actual operating effect and the predicted value.

[0015] Preferably, during operation, the collaborative optimization execution system uses a Gaussian mixture model to fit the spatial distribution of excellent individuals in the current population. The Gaussian mixture model is composed of multiple components, each of which represents a potential high-quality production scheduling area in the search space. By dynamically adjusting the weights, mean vectors, and covariance matrices of each component, the algorithm can capture the multimodal distribution characteristics in the smelting production scheduling problem.

[0016] Furthermore, the collaborative optimization execution system adopts a probabilistic sampling mechanism to generate a new generation of production scheduling candidate schemes. That is, based on the probability density function determined by the Gaussian mixture model, it focuses on searching in regions with a high success rate tendency, rather than conducting disordered trial and error in the entire space, thereby improving the convergence speed of the algorithm when processing high-dimensional smelting data.

[0017] Furthermore, during the evolutionary search process, the collaborative optimization execution system monitors the convergence and diversity of the solution set in real time by calculating the back-generation distance index. When the back-generation distance index does not change within a preset number of iterations or exceeds a preset threshold, the diversity enhancement mechanism is automatically activated, and a perturbation operator is introduced to prevent the algorithm from getting trapped in local optima.

[0018] Furthermore, when establishing the mapping relationship, the energy consumption model analysis engine specifically introduces a non-quota loss correction coefficient. This coefficient is determined by analyzing the instantaneous energy consumption peak of the electric arc furnace at the moment of power step change. It is used to quantify the additional energy waste caused by frequent load adjustments or erroneous start-stop logic, ensuring that the prediction model can truly reflect the hidden energy consumption caused by improper scheduling.

[0019] Furthermore, the multi-objective evaluation processing terminal adopts a ranking strategy based on dominance relationships, dividing all candidate solutions into different non-dominated levels. Solutions in the first level are considered Pareto optimal solutions. Each solution cannot improve other objectives by adjusting parameters without compromising at least one objective, thus providing decision-makers with a set of alternative solutions with balanced performance.

[0020] Furthermore, when extracting feature vectors, the production condition identification unit adopts a combination of time-domain and frequency-domain analysis methods. By analyzing the harmonic distribution of the current signal and the step response of the furnace bottom temperature, it accurately identifies whether the electric arc furnace is in abnormal conditions such as crusting, perforation, or material collapse, and uses these conditions as hard constraints for production scheduling.

[0021] Furthermore, when establishing a task sequence, the smelting production scheduling component will access the local electricity market's time-of-use price data in real time. Through logical calculations, it will allocate the energy-intensive melting stage as much as possible during periods when the electricity price is in a preset low price range, while allocating the heat preservation or refining stages to periods in a preset high price range, thereby achieving synergistic optimization of production efficiency and cost.

[0022] Furthermore, the real-time feedback control module includes compensation logic based on deviation accumulation. When the actual output is lower than the preset monthly progress threshold, the module will automatically trigger an early warning and send a parameter correction signal to the collaborative optimization execution system, prompting the algorithm to increase the weight of the output target in subsequent calculations.

[0023] Furthermore, the data acquisition and sensing device has high-frequency sampling capability, which can capture transient signals in voltage fluctuations at a preset time step, and filter out environmental electromagnetic interference through preset noise reduction logic to ensure that the data input to subsequent units has high accuracy and representativeness.

[0024] Furthermore, the energy consumption model analysis engine is also equipped with a carbon emission calculation plugin. This plugin calculates the carbon footprint of each batch based on real-time power consumption and raw material consumption, according to a preset carbon emission factor, and accumulates them to form a monthly carbon emission report for the entire production line.

[0025] Furthermore, during the search process, the collaborative optimization execution system will identify and prioritize production scheduling patterns with implicit energy-saving characteristics. For example, by identifying that when a specific furnace briefly enters a low-load operation state, although the thermal efficiency of a single furnace decreases, it can avoid large-scale furnace shutdowns of the entire production line to balance the load, thereby reducing systemic carbon pulse emissions.

[0026] Furthermore, when dealing with target conflicts, the multi-target evaluation and processing terminal allows operators to dynamically adjust target weights based on current production priorities. For example, during the delivery guarantee period, the penalty coefficient for delay costs can be increased, while during periods of strict environmental control, the priority of carbon emission targets can be increased.

[0027] Furthermore, the real-time feedback control module interfaces with the automated control system of the submerged arc furnace via industrial Ethernet, and uses standard communication protocols to issue target power settings, feeding ratio parameters, and electrode lifting commands for each furnace.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] 1. This invention achieves deep integration of industrial silicon production planning and energy consumption control by integrating Gaussian learning and Pareto evolution algorithm, breaking the traditional disconnect between production scheduling and energy saving, solving the problem of non-quota energy loss caused by improper scheduling, and reducing unit energy consumption in the smelting process.

[0030] 2. This invention utilizes a Gaussian mixture model to deeply mine the distribution characteristics of the elite solution space, enabling the management platform to possess powerful adaptive fitting capabilities. It can handle complex multimodal distribution data during the smelting process, improving the efficiency and accuracy of searching for the optimal production scheduling scheme. Its back-substitution distance index is optimized compared to traditional algorithms.

[0031] 3. This invention can identify and discover hidden energy-saving production scheduling patterns that are difficult for human experts to detect. By optimizing the overall monthly scale, it avoids the misconception of blindly pursuing the highest efficiency of a single furnace, which leads to an increase in the overall carbon pulse. It finds a better balance between the three conflicting objectives of carbon emissions, delay costs and load balancing, and achieves optimal energy consumption from a global perspective.

[0032] 4. This invention establishes a refined energy consumption prediction and carbon footprint tracking mechanism. By accurately modeling the characteristics of heterogeneous furnaces, it can predict in advance the impact of different scheduling commands on the carbon emissions of the entire production line, providing a scientific decision support tool for the green transformation and implementation of dual-carbon goals in the industrial silicon smelting industry.

[0033] 5. The platform architecture of this invention has good real-time performance and closed-loop adjustment capabilities. Through real-time feedback of production site data and command issuance, it can ensure that the optimized production scheduling plan is executed and can dynamically adjust the control strategy according to changes in working conditions, thereby improving the intelligence level and anti-interference capability of the smelting process. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0035] Figure 2 This is a schematic diagram of the core principle framework of the Pareto evolution algorithm based on Gaussian learning in this invention;

[0036] Figure 3 This is a flowchart illustrating the logical process framework for production condition identification and non-quota energy loss feature extraction in this invention.

[0037] Figure 4 This is a flowchart illustrating the logical flow framework of the energy consumption model analysis engine in this invention for constructing nonlinear mapping relationships and carbon footprint tracking.

[0038] Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow of multi-objective evaluation processing and closed-loop feedback control in this invention. Detailed Implementation

[0039] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0040] The big data-based industrial silicon smelting energy consumption data management platform includes a data acquisition and sensing device, a production condition identification unit, a smelting production scheduling component, an energy consumption model analysis engine, a collaborative optimization execution system, a multi-objective evaluation and processing terminal, and a real-time feedback control module.

[0041] The data acquisition and sensing device is used to acquire real-time production and operation data from multiple heterogeneous submerged arc furnaces and standardize the data to eliminate communication differences between the heterogeneous devices. In this embodiment, the data acquisition and sensing device is deployed at the field equipment layer of the smelting workshop and is connected to the automated control systems of multiple submerged arc furnaces of different models and capacities via an industrial fieldbus. The production and operation data includes multi-dimensional physical quantities, specifically including electrode parameters, furnace temperature, raw material ratio, and current and voltage characteristics.

[0042] The electrode parameters further include the furnace entry depth, lifting speed, and fluctuations in the electrode operating current for each phase electrode; the furnace body temperature includes the furnace bottom temperature, infrared thermography data of key furnace shell nodes, and the inlet and outlet temperature difference of the circulating cooling water; the raw material ratio involves the instantaneous feeding flow rate and batch interval of reducing agents such as silica, charcoal, coal, and petroleum coke; the current and voltage characteristics cover the phase voltage, phase current, active power, reactive power, and power factor of the secondary side of the submerged arc furnace transformer. The data acquisition and sensing device integrates a high-performance data preprocessing module configured to execute high-frequency sampling logic, capable of capturing transient signals in voltage fluctuations with millisecond-level time steps. To ensure data accuracy, the preprocessing module also includes noise reduction logic based on median filtering and Kalman filtering to filter out random pulse interference generated by the strong electromagnetic environment at the smelting site. After cleaning the raw signals, the device uses a preset unified data structure model to convert raw data from different manufacturers and following different protocols into a standardized industrial IoT message format, providing a consistent data foundation for subsequent logical calculations.

[0043] The production condition identification unit is used to identify the current smelting stage based on the collected production operation data, analyze the energy flow characteristics of different submerged arc furnaces during start-up and shutdown processes or load adjustments, and focus on extracting the feature vector of non-quota energy loss. In this embodiment, the production condition identification unit has deep feature analysis capabilities, which performs secondary processing on the production data through a combination of time-domain and frequency-domain analysis methods. This production condition identification unit is configured to perform a fast Fourier transform on the current signal, extract the energy distribution ratio of its various harmonics, and combine it with the step response curve of the furnace bottom temperature to accurately determine whether the submerged arc furnace is currently in the melting, refining, or tapping stage.

[0044] The unit specifically extracts features for abnormal operating conditions, including but not limited to crusting, perforation, material collapse, and electrode breakage, and uses this status information as core hard constraints for subsequent production scheduling. When analyzing energy flow characteristics, the unit focuses on quantifying non-quota energy losses, i.e., energy consumption exceeding the standard theoretical unit consumption caused by production interruptions, frequent load fluctuations, or improper operation. By constructing a residual analysis model of the energy balance equation, the unit can identify the thermal inertia characteristics of each submerged arc furnace in a specific dynamic process and extract feature vectors to describe the system's heat loss patterns.

[0045] The smelting production scheduling component receives monthly production plan instructions and, in conjunction with the real-time status, maintenance plans, and delivery requirements of each heterogeneous submerged arc furnace, establishes an initial production task sequence under multiple constraints. In this embodiment, the smelting production scheduling component serves as the decision-making entry point for production management, and it maintains a dynamic task constraint library internally. This constraint library includes multi-dimensional boundary conditions, such as the rated production capacity of each submerged arc furnace, the upper limit of allowed continuous operating time, mandatory furnace shutdown maintenance cycles, quality grade indicators for each batch of products, and the final delivery date agreed upon in the customer contract. The smelting production scheduling component accesses dynamic time-of-use electricity price data from the local electricity market in real time through an interface and converts it into economic dispatch constraints. When establishing the initial production task sequence, the smelting production scheduling component uses rule-based heuristic logic to decompose the monthly total output target into weekly and daily plans, and, under the premise of satisfying equipment physical constraints, initially plans the operating time windows of each furnace. The smelting production scheduling component is also configured to dynamically adjust the flexibility of the production plan based on the current raw material inventory level and logistics transportation status, ensuring that the initial sequence has redundancy to cope with unexpected operating conditions.

[0046] The energy consumption model analysis engine is used to construct a nonlinear mapping relationship between production scheduling and energy consumption emissions. It utilizes historical big data for training to obtain energy efficiency curves for various heterogeneous furnaces, predicting cumulative energy efficiency and instantaneous carbon emission intensity under different production scheduling strategies. In this embodiment, the energy consumption model analysis engine employs a modeling technique combining multilayer perceptron neural networks and radial basis function networks to deeply mine massive amounts of historical production records. Its core logic lies in establishing a high-dimensional functional relationship mapping. Input variables include task intensity, switching frequency, and equipment operating condition indicators in the production scheduling sequence, while output variables are the predicted total electricity consumption, raw material consumption rate, and carbon oxide emission values.

[0047] The engine incorporates a non-fixed loss correction coefficient, determined through big data analysis of the instantaneous energy consumption peak of the submerged arc furnace at power step moments. This correction coefficient is used to accurately quantify the hidden energy waste caused by frequent load adjustments due to production scheduling instructions. The engine also includes a built-in carbon emission calculation plugin. Based on preset carbon emission factor standards, this plugin calculates the full lifecycle carbon footprint of each furnace cycle in real time, generating a detailed carbon emission intensity prediction report, providing accurate quantitative basis for subsequent multi-objective optimization.

[0048] The collaborative optimization execution system, the core computing component of the platform, integrates a Pareto evolutionary algorithm based on Gaussian learning. This algorithm adaptively adjusts the search strategy by learning the distribution characteristics of elite solutions in the search space, thereby finding the globally optimal solution set within a vast production scheduling space. In this embodiment, the collaborative optimization execution system is configured to execute complex optimization logic. The system initializes a population containing multiple production scheduling schemes, each scheme representing a specific sequence of production task allocations. During each generation of evolutionary iterations, the system constructs a Gaussian mixture model to fit the spatial distribution of excellent individuals (i.e., those on the non-dominant frontier) in the current population. The Gaussian mixture model consists of multiple Gaussian components, each with a specific mean vector, covariance matrix, and weight coefficients. Through dynamic adjustment of these parameters, the algorithm can accurately capture the multimodal distribution characteristics unique to the smelting production scheduling problem, i.e., identify multiple independent but similarly performing high-quality solution regions.

[0049] Furthermore, the collaborative optimization execution system employs a probabilistic sampling mechanism to generate a new generation of production scheduling candidate solutions. Instead of blindly trying across the entire search space, the system focuses on sampling regions with a high success rate (i.e., potential elite distribution areas) based on the probability density function determined by the Gaussian mixture model. This strategic guessing approach improves the convergence speed of the algorithm when processing high-dimensional smelting data and optimizes the back-generation distance index. During the evolutionary process, the system monitors the convergence and diversity of the solution set in real time. When it detects that the back-generation distance index has not been optimized within a preset number of generations or that the solution set shows a clustering tendency, the system automatically activates a diversity enhancement mechanism. By introducing Gaussian perturbation operators or random recombination strategies, it forcibly guides the search flow towards unexplored blank areas, preventing the algorithm from falling into the trap of local optima.

[0050] More importantly, the collaborative optimization execution system possesses the ability to identify implicit energy-saving production scheduling patterns during the search process. This capability stems from its deep analysis of the distribution characteristics of elite solution sets. For example, the system may identify a special production scheduling combination: allowing a certain submerged arc furnace with specific parameters to operate at low load during certain periods. Although this may not be optimal in terms of single-furnace efficiency, this strategy avoids large-scale furnace shutdowns across the entire production line to balance the total load, reducing the huge systemic carbon pulse emissions caused by frequent start-ups and shutdowns. This holistic optimization capability based on a global perspective breaks the traditional cognitive bias that local optima are global optima, achieving optimal coordination between energy consumption and emissions on a monthly scale.

[0051] The multi-objective evaluation and processing terminal is used to evaluate the optimized production scheduling scheme from multiple dimensions, balancing multiple conflicting objectives such as carbon emission intensity, delay penalty cost, and power load balance. In this embodiment, the multi-objective evaluation and processing terminal adopts a non-dominant ranking strategy based on dominance relationships. The terminal maps all candidate solutions generated by the collaborative optimization execution system to a high-dimensional objective space and divides all schemes into different frontier levels according to their dominance relationships. The set of schemes at the first level is the Pareto optimal solution set. In these solution sets, any further optimization of any objective must come at the expense of at least one other objective. To assist decision-making, the terminal provides a dynamic weight adjustment interface, allowing managers to adjust the priority of each objective in real time according to the current production focus. For example, during periods of urgent delivery deadlines, the terminal can automatically increase the penalty coefficient for delay costs; while during periods of strict environmental control, it automatically increases the evaluation weight of carbon emission targets. The terminal also integrates visualization analysis tools, displaying the comprehensive performance of different schemes in multiple dimensions such as energy efficiency, cost, and environmental protection to managers through radar charts, parallel coordinate graphs, etc.

[0052] The real-time feedback control module is used to convert the finalized optimal production schedule into specific equipment control commands and send them to the actuators of the submerged arc furnace. It performs closed-loop adjustments based on the deviation between the actual operating results and the predicted values. In this embodiment, the real-time feedback control module has powerful communication conversion and logic execution functions. It seamlessly interfaces with the DCS (Distributed Control System) or PLC (Programmable Logic Controller) of the submerged arc furnace via industrial Ethernet, using standard industrial protocols (such as OPC-UA or Profinet) to convert the abstract production schedule into specific parameter commands, including the target power setting for each furnace, the expected feeding time, the feeding ratio parameters, and the target electrode lifting position. This module has a built-in compensation logic unit based on deviation accumulation. During actual production, if the actual output falls below the preset monthly progress threshold due to raw material fluctuations or temporary equipment failures, this unit automatically triggers an early warning and feeds back a parameter correction signal to the collaborative optimization execution system. This feedback mechanism enables the optimization system to automatically increase the weight of the output target in the next iteration calculation, ensuring that production tasks can be completed on time, forming a complete, self-healing closed-loop control system.

[0053] During operation, the collaborative optimization execution system constructs a Gaussian mixture model to fit the spatial distribution of superior individuals in the current population. This Gaussian mixture model consists of multiple components, each representing a potential high-quality production scheduling region in the search space. By dynamically adjusting the weights, mean vectors, and covariance matrices of each component, the algorithm can capture the multimodal distribution characteristics of the smelting production scheduling problem. In this way, the system can not only find a single global optimal solution but also identify multiple superior production modes with different physical backgrounds, providing a rich selection of scheduling options under different operating conditions.

[0054] Furthermore, the collaborative optimization execution system employs a probabilistic sampling mechanism to generate a new generation of production scheduling candidate solutions. Specifically, based on the probability density function determined by the Gaussian mixture model, it focuses its search on regions with a high success rate tendency, rather than engaging in unordered trial and error across the entire space. This improves the convergence speed of the algorithm when processing high-dimensional smelting data. This mechanism utilizes prior statistical distribution information to guide the search direction, avoiding the ineffective waste of computational resources.

[0055] Furthermore, during the evolutionary search process, the collaborative optimization execution system monitors the convergence and diversity of the solution set in real time by calculating the back-generation distance index. When the back-generation distance index does not change within a preset number of iterations or exceeds a preset threshold, a diversity enhancement mechanism is automatically activated, introducing a perturbation operator to prevent the algorithm from getting trapped in local optima. This index quantifies the distance between the current solution set and the true Pareto front, serving as a key parameter for evaluating the quality of the search.

[0056] Furthermore, the energy consumption model analysis engine introduces a non-fixed-rate loss correction coefficient when establishing the mapping relationship. This coefficient is determined by analyzing the instantaneous energy consumption peak of the electric arc furnace at the moment of power step change. It is used to quantify the additional energy waste caused by frequent load adjustments or erroneous start-stop logic, ensuring that the prediction model can truly reflect the hidden energy consumption caused by improper scheduling. This correction mechanism corrects the prediction bias of traditional static energy consumption models in dynamic scheduling environments.

[0057] Furthermore, the multi-objective evaluation and processing terminal employs a dominance-based ranking strategy, dividing all candidate solutions into different non-dominated levels. Solutions in the first level are considered Pareto optimal solutions, where each solution, without compromising at least one objective, cannot improve other objectives by adjusting parameters, thus providing decision-makers with a set of balanced alternatives. This method provides a scientific evaluation criterion for decision-making in multi-objective conflicts.

[0058] Furthermore, the production condition identification unit employs a combination of time-domain and frequency-domain analysis when extracting feature vectors. By analyzing the harmonic distribution of the current signal and the step response of the furnace bottom temperature, it accurately identifies whether the submerged arc furnace is in abnormal conditions such as crusting, perforation, or material collapse, and uses these conditions as hard constraints for production scheduling. This ensures the safety and feasibility of the production scheduling plan at the physical execution level.

[0059] Furthermore, when establishing a task sequence, the smelting production scheduling component accesses real-time time-of-use electricity price data from the local electricity market. Through logical calculations, it allocates the energy-intensive melting stage to periods when electricity prices are in a preset low price range, while allocating the heat preservation or refining stages to periods in a preset high price range, thus achieving synergistic optimization of production efficiency and costs. This feature transforms energy market cost fluctuations into improved enterprise production efficiency.

[0060] Furthermore, the real-time feedback control module includes compensation logic based on cumulative deviation. When the actual output falls below a preset monthly progress threshold, the module automatically triggers an early warning and sends a parameter correction signal to the collaborative optimization execution system, prompting the algorithm to increase the weight of the output target in subsequent calculations. This function ensures that the management platform has dynamic self-correction capabilities.

[0061] Furthermore, the data acquisition and sensing device possesses high-frequency sampling capabilities, enabling it to capture transient signals in voltage fluctuations at a preset time step. It also filters out environmental electromagnetic interference through preset noise reduction logic, ensuring that the data input to subsequent units is highly accurate and representative. High-quality data input is a prerequisite for all subsequent big data analysis and optimization calculations.

[0062] Furthermore, the energy consumption model analysis engine is also equipped with a carbon emission calculation plugin. This plugin calculates the carbon footprint of each batch based on real-time power consumption and raw material consumption, according to preset carbon emission factors, and accumulates them to form a monthly carbon emission report for the entire production line. This provides enterprises with a precise digital tool to achieve their dual carbon targets.

[0063] Furthermore, during the search process, the collaborative optimization execution system identifies and prioritizes production scheduling patterns with implicit energy-saving characteristics. For example, by identifying that when a specific furnace briefly enters a low-load operation state, although the thermal efficiency of a single furnace decreases, it can avoid large-scale furnace shutdowns of the entire production line to balance the load, thereby reducing systemic carbon pulse emissions. This characteristic represents a significant technological leap forward for this invention in the field of energy conservation and emission reduction.

[0064] Furthermore, when handling conflicting objectives, the multi-objective evaluation and processing terminal allows operators to dynamically adjust objective weights based on current production priorities. For example, it can increase the penalty coefficient for delay costs during the delivery guarantee period and increase the priority of carbon emission targets during periods of strict environmental control. This human-machine collaboration mechanism enhances the system's flexibility.

[0065] Furthermore, the real-time feedback control module interfaces with the automated control system of the submerged arc furnace via industrial Ethernet, using standard communication protocols to issue target power settings, feeding ratio parameters, and electrode lifting commands for each furnace. This deep system integration ensures that optimization decisions can be applied to the production line.

[0066] Example 2: Based on the industrial silicon smelting energy consumption data management platform constructed in Example 1 above, this example provides a system implementation method based on cloud-edge collaborative architecture to meet the management needs of large-scale, cross-regional submerged arc furnace clusters.

[0067] In this architecture, the big data-based industrial silicon smelting energy consumption data management platform includes an edge-side data processing subsystem, a cloud-based centralized optimization management platform, and a distributed feedback control network. The edge-side data processing subsystem consists of multiple edge computing nodes deployed in various smelting plants, responsible for real-time data acquisition and high-frequency operating condition identification. The cloud-based centralized optimization management platform centralizes the core computing resources of the energy consumption model analysis engine and the collaborative optimization execution system, used to handle complex optimization problems on a global scale.

[0068] The edge-side data processing subsystem undertakes most of the functions of the data acquisition and sensing device. Each edge node is equipped with a dedicated digital signal processor, configured to perform high-speed preprocessing of current, voltage, vibration, and thermal imaging signals generated by the submerged arc furnace within its area. Due to the extremely large volume of data generated at the smelting site, which contains a significant amount of redundant information, the edge nodes utilize a pre-defined feature extraction algorithm to transform the massive amount of raw sampling points into a compact feature matrix. These feature matrices can characterize the real-time energy efficiency status of the submerged arc furnace, fluctuations in key thermal parameters, and the current operating condition classification (e.g., whether it is in a normal smelting cycle). By performing preliminary abnormal efficiency identification at the edge side, the system can reduce the data bandwidth pressure for transmission to the cloud and improve the response speed to emergency abnormal conditions (such as sudden material collapse).

[0069] The cloud-based centralized optimization management platform receives feature data aggregated from various edge nodes and combines it with global monthly plans and macroeconomic information on the electricity market. In this embodiment, the cloud platform deploys a high-performance graphics processing unit (GPU) cluster specifically for running a Pareto evolutionary algorithm based on Gaussian learning. Due to the cloud's more powerful storage and computing capabilities, its energy consumption model analysis engine can access larger-scale historical datasets (e.g., nationwide smelting data spanning several years) to train more accurate nonlinear mapping models. The cloud platform utilizes Gaussian mixture models to holographically model the collaborative working characteristics of multiple heterogeneous furnaces, capable of handling hundreds or thousands of decision variables covering joint production combinations of submerged arc furnaces in different regions and with different capacities. The Pareto optimal solution set generated by the cloud platform is further pushed to a multi-objective evaluation and processing terminal for global strategy decision-making by production scheduling experts at the enterprise headquarters.

[0070] The distributed feedback control network is built on a low-latency dedicated industrial network. Once the optimal production scheduling plan is determined in the cloud, it is broken down into micro-instruction sets for each branch plant and each piece of equipment. These instruction sets are distributed to each edge node through a distributed gateway. The real-time feedback control module then exhibits a hierarchical structure: the edge nodes are responsible for executing specific loop control and short-term deviation compensation, while the cloud platform is responsible for long-term strategy calibration. For example, when the power load of a branch plant suddenly changes, the edge subsystem immediately makes fine adjustments based on its local energy consumption model to ensure that the power factor is within a safe range, and feeds back the adjustment information to the cloud, triggering the cloud to perform the next round of global balance calculation. This cloud-edge collaborative architecture not only ensures the global optimization depth of the algorithm but also takes into account the real-time and reliability requirements of industrial sites.

[0071] At the data flow level, the platform employs an encrypted industrial messaging protocol. Data acquisition and sensing devices perform data encryption and digest generation at the edge, ensuring the integrity and privacy of production data during transmission. The feature vectors calculated by the production condition identification unit at the edge are timestamped, and the cloud-based energy consumption model analysis engine reconstructs the production process timeline based on these timestamped information, accurately identifying the hysteresis effect and dynamic response characteristics of energy flow. During operation in the cloud, the collaborative optimization execution system dynamically adjusts the participation weight of each furnace based on the actual equipment health status of each branch plant (assessed by the condition identification unit). For example, for furnaces with high heat loss and at the critical point of maintenance, the optimization system automatically reduces their production load weight and assigns lower energy efficiency expectations to avoid high energy consumption risks.

[0072] In this embodiment, the multi-objective evaluation and processing terminal is configured as a hierarchical authorization interface. Frontline operators can view the optimized operation instructions and real-time carbon intensity of their furnace through portable terminals; workshop supervisors can monitor the load balance and production progress of the entire workshop; and senior executives can view the cumulative energy efficiency improvement rate, total carbon emissions, and achievement of predetermined emission reduction targets across the entire group through a large-screen display board. This hierarchical management model complements the cloud-edge collaborative architecture.

[0073] Example 3: This example provides an implementation of an industrial silicon smelting energy consumption data management platform based on digital twin, aiming to further improve the system's prediction accuracy and adaptive adjustment capabilities.

[0074] In this implementation, the platform incorporates a digital twin mapping module. This module is deeply integrated with data acquisition and sensing devices and an energy consumption model analysis engine. A digital twin is a virtual model constructed within computing space that perfectly corresponds to the physical submerged arc furnace. These models not only include the geometric features and structural parameters of the physical equipment, but more importantly, they contain a hybrid dynamics model based on physical mechanisms and driven by data.

[0075] The feature vectors extracted by the production condition identification unit are input into the digital twin model in real time to drive the status update of the virtual furnace. When the current of the physical submerged arc furnace fluctuates significantly, the digital twin model can synchronously simulate the changes in its internal thermal, electromagnetic, and flow fields. Through this real-time mapping, the energy consumption model analysis engine no longer relies solely on historical experience data but can combine advanced simulations performed by the virtual model in digital space. For example, before issuing physical commands, the engine can run thousands of power adjustment experiments of different intensities in the digital twin space to more accurately predict the value of non-standard energy loss and correct its non-standard loss correction coefficient.

[0076] In this embodiment, the smelting production scheduling component possesses predictive capabilities. When establishing the initial task sequence, it invokes a digital twin module to perform a virtual rehearsal of the suggested sequence. If the virtual rehearsal indicates that a certain production scheduling sequence poses a risk of localized overheating or excessive electrode wear in a certain submerged arc furnace, the component automatically discards the sequence and regenerates the constraint boundaries. This rehearsal-based scheduling mechanism improves the safety and reliability of the production plan.

[0077] The collaborative optimization execution system, when processing the Pareto evolution algorithm based on Gaussian learning, utilizes massive simulated samples provided by digital twin models to enhance the learning effect of the Gaussian mixture model. In situations where physical samples are scarce (e.g., certain extreme production combinations have never occurred in reality), the system can generate high-quality virtual samples through digital twins to fill the blank areas of the search space. This allows the Gaussian distribution to completely fit potential high-quality production scheduling regions, enabling the algorithm to maintain convergence speed and solution accuracy even when dealing with unprecedentedly complex production demands.

[0078] The real-time feedback control module integrates predictive control logic based on digital twins. This module not only monitors current operational deviations but also uses the digital twin model to predict future state trends within the control cycle. If the prediction indicates that future actual output will deviate from the optimization target, the module will send a feedforward compensation signal to the actuator in advance. This prediction-based closed-loop regulation can smooth out fluctuations in the production process, reduce additional energy efficiency losses caused by over-adjustment, and achieve optimal energy consumption control at a finer time scale.

[0079] In this embodiment, the multi-objective evaluation and processing terminal adds a scenario simulation function. Managers can set different virtual scenarios on the terminal (such as soaring electricity prices, raw material shortages, and adjustments to carbon tax policies). The system will quickly assess the impact of these scenarios on the current production scheduling plan through a digital twin model and display the dynamic migration trajectory of the Pareto front. This provides technical support for enterprises to make forward-looking decisions in complex market environments.

[0080] Example 4: This example focuses on describing the specific internal logic implementation of the big data-based industrial silicon smelting energy consumption data management platform when processing heterogeneous data fusion and multimodal distribution fitting.

[0081] The data acquisition and sensing device faces challenges in acquiring electrode parameters, furnace temperature, raw material ratios, and current and voltage characteristics due to significant differences in sampling frequency (from millisecond-level electrical parameters to hourly-level raw material analysis data) and inconsistent data accuracy. The device's internal standardized processing logic employs a multi-rate data fusion technology based on timestamp alignment and adaptive resampling. This technology smoothly maps low-frequency laboratory analysis data (such as silica oxygen content and carbon source reactivity) to high-frequency production operation sequences. By constructing a confidence evaluation model based on multi-sensor information entropy, the device can identify data distortion caused by sensor aging or link interference and automatically switch to backup data sources or use neighborhood averages for data repair, ensuring the robustness of subsequent energy consumption model analysis engine inputs.

[0082] The production condition identification unit pays particular attention to the thermal delay effect during the smelting process when analyzing energy flow characteristics. Because the submerged arc furnace is a large thermally inertial system, there is a time lag between electrical energy input and thermal energy output (reflected in the rate of silicon metal formation). This unit is configured to use a cross-correlation function analysis method to calculate the dynamic response time difference between the electrical energy input sequence and the furnace bottom temperature and flue gas temperature sequences. These time difference parameters are integrated into the feature vector of non-standard energy loss, enabling the platform to distinguish which energy fluctuations are caused by normal process lags and which are caused by ineffective heat dissipation.

[0083] The energy consumption model analysis engine incorporates a deep residual network architecture when constructing nonlinear mapping relationships. This architecture decomposes energy consumption prediction into static basic energy consumption items and dynamic disturbance energy consumption items. The static basic energy consumption items are calculated based on physical thermal balance theory, while the dynamic disturbance energy consumption items are obtained by learning the random fluctuation characteristics of historical big data. By introducing a non-fixed loss correction coefficient, the model can dynamically correct for energy efficiency degradation caused by long-term factors such as equipment wear and tear and furnace lining thinning. The carbon emission calculation plugin goes a step further, estimating the direct carbon emissions generated by the process in real time based on the consumption rate and oxidation reaction kinetics of each batch of electrodes, and combining them with the indirect carbon emissions generated by electricity consumption. This comprehensive carbon footprint tracking gives the monthly carbon emission reports generated by the management platform audit value.

[0084] The collaborative optimization execution system performs extreme optimization on the update logic of the Gaussian mixture model when running the GLPEA algorithm. The system is configured to use an iterative strategy based on the expectation-maximization algorithm to estimate the parameters of the Gaussian components in real time. To prevent overfitting, the system introduces a Bayesian information criterion to dynamically determine the optimal number of Gaussian components. This adaptive component adjustment mechanism ensures that the algorithm can accurately capture every small, energy-efficient local region in the solution space. In the probabilistic sampling stage, the system introduces a directional guidance factor, enabling the sampling process to not only follow the current probability distribution but also accelerate the search along the descent direction of the non-dominated level, thus significantly improving convergence efficiency.

[0085] The collaborative optimization execution system employs a pattern mining algorithm based on cluster analysis to identify implicit energy-saving production scheduling patterns. This algorithm performs a deep scan of the elite solutions generated in each evolutionary iteration, searching for anomalous solutions that violate traditional experience but possess exceptionally high performance metrics. For example, the algorithm might detect a specific load allocation scheme: among three heterogeneous furnaces, the worst-performing furnace is intentionally kept in a long-term low-temperature standby state, while the other two high-efficiency furnaces alternate between full-load and overload operation. Although this strategy is controversial in terms of single-furnace lifespan, within specific time-of-use electricity price ranges, this model can reduce the instantaneous carbon emission intensity of the entire plant and maximize the use of off-peak electricity prices, achieving a balance between economic benefits and low-carbon goals.

[0086] The multi-objective evaluation and processing terminal introduces a decision-making support mechanism based on fuzzy logic when handling objective conflicts. Since managers' preferences regarding carbon emissions and delivery penalties in different contexts are often difficult to express with precise numbers, the terminal allows input of fuzzy preference language (e.g., highly valuing low carbon emissions, moderately concerned about costs). The terminal's internal fuzzy inference engine transforms these linguistic variables into selection operators on the Pareto front, filtering out the production scheduling scheme that best aligns with the current management intent.

[0087] The real-time feedback control module focuses on addressing the smooth switching of command issuance during execution. To avoid excessive impact on the power grid during production scheduling changes, the module integrates a power gradient control-based logic. When the set power of a certain submerged arc furnace needs to be adjusted, the module breaks down the target command into a series of stepped sub-commands, which are executed smoothly within a preset time window. The module monitors the execution status feedback from the submerged arc furnace's automated control system in real time. Once it detects a slow response or obstruction in the actuator (such as the electrode lifting motor), it immediately triggers the local rescheduling logic of the optimization system through an anomaly reporting mechanism, ensuring the closed-loop implementation of the optimization results at the physical level.

[0088] Example 5: This example describes an enhanced industrial silicon smelting energy consumption data management platform with self-learning capabilities.

[0089] In this embodiment, the platform introduces a deep reinforcement learning module, which works in parallel with the collaborative optimization execution system. The deep reinforcement learning module models the smelting environment as a complex Markov decision process. Its state space consists of real-time feature vectors provided by the production condition identification unit, its action space consists of production scheduling instructions, and its reward function is determined by the comprehensive score given by the multi-objective evaluation processing terminal.

[0090] The advantage of this architecture lies in its ability to automatically learn from past decision feedback as runtime increases. Whenever the collaborative optimization execution system generates an excellent scheduling plan and it is physically executed, the reinforcement learning module records the plan's actual performance under different dynamic operating conditions. If the actual energy-saving effect exceeds the model's prediction, the reinforcement learning module will increase the weight preference for that type of scheduling pattern. This algorithm-learning mechanism allows the Gaussian mixture model to obtain a more forward-looking search bias.

[0091] The energy consumption model analysis engine in this embodiment possesses incremental learning capabilities. It does not need to retrain the entire neural network every time new data is received; instead, it absorbs the latest operating conditions by updating subtle changes in the weight parameters. This means that when the smelting process is fine-tuned (e.g., a new reducing agent is used), the platform can automatically detect changes in the energy efficiency curve within a short time and quickly adjust its prediction logic.

[0092] This self-learning management platform can automatically identify and compensate for long-term system deviations caused by sensor drift, equipment aging, or changes in the external environment. It is not merely an optimization tool, but has become a smart brain capable of evolving in sync with the physical production process, providing the ultimate technological path for the intelligent upgrading of the industrial silicon smelting industry.

[0093] Those skilled in the art should understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention. All descriptions involving mathematical formulas, logical relationships, or parameter comparisons have been fully and sufficiently disclosed through textual description, aiming to enable those skilled in the art to implement all functional modules of the present invention and their mutual synergistic optimization. The hardware structures, algorithm logic, data flows, and communication protocols described in the various embodiments can be flexibly combined and transformed according to actual industrial application scenarios, and such combinations and transformations also fall within the scope of the claims of the present invention.

Claims

1. A big data-based industrial silicon smelting energy consumption data management platform, characterized in that, include: A data acquisition and sensing device is used to acquire production and operation data of multiple heterogeneous submerged arc furnaces in real time, and to standardize the production and operation data to eliminate communication differences between heterogeneous devices. The production condition identification unit is used to identify the current smelting stage based on the collected production operation data, analyze the energy flow characteristics of the submerged arc furnace in the dynamic process, and extract the feature vector of non-quota energy loss. The smelting production scheduling component is used to receive monthly production plan instructions and, in combination with the real-time status, maintenance plans and delivery requirements of each heterogeneous submerged arc furnace, establish an initial production task sequence under multiple constraints. An energy consumption model analysis engine is used to construct a nonlinear mapping relationship between production scheduling and energy consumption emissions, and to predict the cumulative energy efficiency and instantaneous carbon emission intensity under different production scheduling strategies. The collaborative optimization execution system is used to run the Pareto evolution algorithm based on Gaussian learning. By learning the distribution characteristics of the elite solution set in the search space, it adaptively adjusts the search strategy to find the globally optimal solution set in the scheduling scheme space. A multi-objective evaluation and processing terminal is used to evaluate the optimal production scheduling scheme from multiple dimensions, weighing carbon emission intensity, delay penalty cost and power load balance. The real-time feedback control module is used to convert the determined optimal production scheduling plan into equipment control commands and issue them, while performing closed-loop adjustment based on the deviation between the actual operating effect and the predicted value.

2. The industrial silicon smelting energy consumption data management platform based on big data according to claim 1, characterized in that, The data acquisition and sensing device includes a data preprocessing module, which is configured to execute high-frequency sampling logic to capture transient signals of current and voltage characteristics in the production operation data with a time step of milliseconds. The data preprocessing module has a built-in noise reduction logic unit, which uses median filtering and Kalman filtering algorithms to filter electromagnetic pulse interference at the smelting site. The data acquisition and sensing device also includes a protocol conversion unit, which, based on a preset unified data structure model, converts raw data following different communication protocols into a standardized industrial Internet of Things message format. The production operation data includes: electrode parameters, covering the furnace entry depth, lifting speed, and fluctuation value of electrode operating current for each phase electrode; Furnace body temperature, including furnace bottom temperature, infrared temperature measurement data of key nodes in the furnace shell, and inlet and outlet temperature difference of circulating cooling water; Raw material proportions, including instantaneous flow rates of silica and reducing agent, as well as batch feeding intervals.

3. The industrial silicon smelting energy consumption data management platform based on big data according to claim 2, characterized in that, The production condition identification unit includes a feature parsing module, which is configured to process the production operation data using a combination of time-domain and frequency-domain analysis methods. The feature analysis module is configured to perform a fast Fourier transform on the current signal, extract the energy distribution ratio of each harmonic, and combine it with the step response curve of the furnace bottom temperature to identify any stage of the ferroelectric furnace currently in the melting period, refining period, or tapping period. The production condition identification unit is also configured to identify abnormal conditions such as crusting, perforation, material collapse, and electrode breakage, and to use these abnormal conditions as hard constraints for production scheduling. The production condition identification unit quantifies the energy consumption beyond the standard theoretical unit consumption caused by production interruptions or frequent load fluctuations by constructing a residual analysis model of the energy balance equation, thereby extracting the feature vector describing the heat loss law of the system.

4. The industrial silicon smelting energy consumption data management platform based on big data according to claim 3, characterized in that, The smelting production scheduling component maintains a dynamic task constraint library, which includes the rated production capacity of each submerged arc furnace, the upper limit of allowed continuous operating time, the mandatory shutdown maintenance cycle, the quality grade indicators of each batch of products, and the final delivery date. The smelting production scheduling component is equipped with an economic dispatch interface, which is connected to the dynamic time-of-use electricity price data of the electricity market in real time. The smelting production scheduling component is configured to use rule-based heuristic logic to decompose the monthly total output target into weekly and daily plans. Under the premise of meeting the physical constraints of the equipment, it uses logical calculations to allocate the energy-intensive melting stage to the time period when the electricity price is in the preset low price range, and allocate the heat preservation stage to the time period in the preset high price range, thereby achieving synergistic optimization of production efficiency and electricity costs.

5. The industrial silicon smelting energy consumption data management platform based on big data according to claim 4, characterized in that, The energy consumption model analysis engine is configured to use a modeling approach that combines multilayer perceptron neural networks and radial basis function networks to construct a functional relationship mapping with task intensity, switching frequency, and equipment operating condition indicators in the production schedule as input variables, and predicted total electricity consumption, raw material consumption rate, and carbon oxide emissions as output variables. The energy consumption model analysis engine is equipped with a loss correction unit, which introduces a non-fixed loss correction coefficient. The non-fixed loss correction coefficient is determined by analyzing the instantaneous energy consumption peak of the electric arc furnace at the moment of power step, and is used to quantify the energy waste caused by frequent load adjustments. The energy consumption model analysis engine is also equipped with a carbon emission calculation plugin, which calculates the full life cycle carbon footprint of each furnace in real time according to the preset carbon emission factor standard.

6. The industrial silicon smelting energy consumption data management platform based on big data according to claim 5, characterized in that, The collaborative optimization execution system is configured to fit the spatial distribution of superior individuals at the non-dominant front in the current population by constructing a Gaussian mixture model during each generation of evolutionary iteration. The Gaussian mixture model is composed of multiple Gaussian components, each of which has a mean vector, a covariance matrix, and weight coefficients. The collaborative optimization execution system is configured to capture the multimodal distribution characteristics in the smelting production scheduling problem by dynamically adjusting the parameters of the Gaussian mixture model, thereby identifying multiple independent and similar high-quality solution regions. The collaborative optimization execution system also incorporates Bayesian information criteria to adaptively determine the optimal number of Gaussian components in the Gaussian mixture model during operation, in order to prevent overfitting of the model.

7. The industrial silicon smelting energy consumption data management platform based on big data according to claim 6, characterized in that, The collaborative optimization execution system is equipped with a probability sampling module, which generates a new generation of production scheduling candidate solutions in regions with a high success rate tendency based on the probability density function determined by the Gaussian mixture model. The collaborative optimization execution system is also equipped with a convergence monitoring module, which monitors the quality of the solution set in real time by calculating the backpropagation distance index. The reverse algebraic distance metric is used to quantify the distance between the current solution set and the true Paleto front; When the backward distance index does not decrease significantly or exceed a preset threshold within a preset number of consecutive iterations, the collaborative optimization execution system automatically activates the diversity enhancement mechanism. By introducing Gaussian perturbation operators or random recombination strategies, the search flow is guided to unexplored regions to avoid the algorithm getting trapped in local optima.

8. The industrial silicon smelting energy consumption data management platform based on big data according to claim 7, characterized in that, The multi-objective evaluation and processing terminal adopts a non-dominated ranking strategy based on dominance relationship, maps all candidate solutions to a high-dimensional objective space, and divides all solutions into different non-dominated levels according to their mutual dominance relationship. The set of solutions in the first level is defined as the Pareto optimal solution set. The multi-objective evaluation and processing terminal is equipped with a dynamic weight adjustment interface, which allows the priority weights among carbon emission intensity, delay penalty cost and power load balance to be adjusted in real time according to the center of production focus. The multi-objective evaluation and processing terminal also integrates visualization analysis tools, which are configured to use radar charts or parallel coordinate graphs to show managers the comprehensive performance of different solutions in terms of energy efficiency, cost, and environmental protection.

9. The industrial silicon smelting energy consumption data management platform based on big data according to claim 8, characterized in that, The real-time feedback control module is connected to the distributed control system or programmable logic controller of the submerged arc furnace via industrial Ethernet, and uses industrial communication protocols to issue control commands including the target set power of each submerged arc furnace, the expected feeding time, the feeding ratio parameters, and the target lifting position of the electrode. The real-time feedback control module has a built-in compensation logic unit based on deviation accumulation. The compensation logic unit is configured to monitor the relationship between the actual output and the preset monthly progress threshold. When the actual output is lower than the preset monthly progress threshold, the compensation logic unit triggers an early warning and sends a parameter correction signal to the collaborative optimization execution system, prompting the optimization algorithm to increase the weight of the output target in subsequent calculations, thus constructing a closed-loop control system with self-healing capabilities.

10. The industrial silicon smelting energy consumption data management platform based on big data according to claim 9, characterized in that, The collaborative optimization execution system is equipped with a latent pattern mining algorithm, which is used to identify and prioritize the retention of production scheduling patterns with energy-saving characteristics during the evolutionary search process. The production scheduling mode includes: identifying when the submerged arc furnace enters a preset low-load operating state, and reducing systemic carbon pulse emissions by avoiding large-scale furnace shutdowns of the entire production line to balance the load; The collaborative optimization execution system is configured to perform global overall optimization of multiple heterogeneous submerged arc furnaces on a monthly scale. Under the premise of meeting the thermal efficiency constraints of individual furnaces, it obtains the optimal cumulative energy efficiency value of the entire production line by sacrificing local instantaneous efficiency. The management platform is also configured with a cloud-edge collaborative architecture, in which the edge-side data processing subsystem deployed at the smelting site performs real-time data acquisition and operating condition identification, and the cloud-based centralized optimization management platform executes the core algorithm of the collaborative optimization execution system.

Images