Energy-saving control method and system for high-energy-consumption enterprise based on multiple load cooperation

Abstract

This invention discloses an energy-saving control method and system for high-energy-consuming enterprises based on multi-load coordination, belonging to the field of industrial energy conservation and automatic control technology. The method includes: real-time acquisition of operating data from adjustable motors, interruptible heating systems, and process loads with thermal inertia in high-energy-consuming enterprises; construction of a multi-objective optimization function with the core objective of minimizing overall electricity costs, taking into account time-of-use and demand-based electricity charges, and establishment of a load model; rolling optimization at three time scales: day-ahead, intraday, and real-time, generating scheduling instructions; execution of coordinated control of various loads; and closed-loop optimization and model correction through model predictive control. The system includes modules for data acquisition, optimization decision-making, control execution, and closed-loop feedback. This invention achieves coordinated and economical scheduling of heterogeneous loads in high-energy-consuming enterprises, effectively reducing electricity costs and improving energy utilization efficiency and system operational stability.

Description

Technical Field

[0001] This invention relates to the field of industrial process control and energy management technology, specifically to an energy-saving control method and system for high-energy-consuming enterprises based on the coordination of multiple loads. Background Technology

[0002] High-energy-consuming enterprises typically refer to industrial enterprises whose energy consumption per unit of output is far higher than the social average. Their production processes involve intensive electrical loads from motors, electric heating equipment, and high-temperature process equipment, resulting in energy costs accounting for a very high proportion of their total costs. Currently, energy conservation and consumption reduction efforts for these enterprises mainly focus on partial modifications and management at the equipment level, such as installing frequency converters for motors, modifying the insulation of heating furnaces, or implementing simple time-sharing shutdowns.

[0003] Existing technologies typically optimize individual load types independently, lacking coordinated management from the perspective of the enterprise's overall energy system. For example, motor energy-saving control, optimized combustion and waste heat recovery in heating furnaces, and intelligent management of air conditioning systems are often handled separately by different subsystems or based on human experience. This individual management model has significant drawbacks: First, the rapid response of motor loads and the slow, high-inertia thermal loads cannot achieve dynamic complementarity and coordination, leading to drastic fluctuations in the enterprise's total electricity load curve, making it difficult to effectively avoid peak demand charges from the grid and fully utilize time-of-use pricing policies. Second, existing methods rely heavily on fixed rules or static models, making it difficult to adapt to complex dynamic factors such as production plan adjustments, raw material changes, and market electricity price fluctuations, resulting in limited and delayed optimization effects. Third, the lack of adaptive correction capabilities for changes in load characteristics and model mismatches leads to a gradual decline in energy efficiency over long-term operation.

[0004] Therefore, existing technologies have failed to provide a systematic energy-saving control solution that can deeply coordinate multiple heterogeneous loads within high-energy-consuming enterprises, dynamically optimize across multiple time scales, and possess closed-loop self-learning capabilities, resulting in bottlenecks in improving the overall energy efficiency and economic performance of enterprises. Summary of the Invention

[0005] This application provides an energy-saving control method and system for high-energy-consuming enterprises based on the coordination of multiple loads, which at least addresses the problems existing in the prior art.

[0006] A first aspect of this application provides an energy-saving control method for high-energy-consuming enterprises based on multi-load coordination, comprising the following steps: S1: Real-time acquisition of various load data from high-energy-consuming enterprises, including adjustable motor loads, interruptible heating loads, and process loads with thermal inertia; S2: Based on various load data, construct a multi-objective collaborative optimization function with the minimum comprehensive electricity cost as the core, taking into account time-of-use electricity charges, demand charges, and penalty costs for deviation from production constraints, and establish load models for the power adjustable range and dynamic response characteristics of various loads; S3: Based on the optimization function and load model, rolling optimization and decision coordination are performed on three time scales: day-ahead, intraday, and real-time, to generate scheduling instructions for loads with different response speeds. S4: Send dispatch instructions to the local controllers of each load to perform coordinated control of motor speed, heater power level and process equipment operation; S5: Real-time monitoring of load response status and total energy consumption; dynamic feedback and correction of load model and optimization function through model predictive control algorithm to achieve closed-loop optimization.

[0007] This application addresses the challenge of coordinated control of diverse heterogeneous loads, such as motors, heating systems, and processes with high thermal inertia, in high-energy-consuming enterprises by establishing a load model, constructing a multi-objective optimization function that considers both economic efficiency and production constraints, and employing a combination of multi-timescale rolling optimization and model predictive control closed-loop feedback. By integrating the discrete load resources of high-energy-consuming enterprises into a flexibly controllable virtual whole, it achieves a leap from equipment-level energy saving to system-level optimization. Under the premise of strictly ensuring production safety and continuity, it significantly reduces overall electricity costs and improves energy utilization efficiency and grid interaction capabilities.

[0008] In some embodiments of this application, the real-time acquisition of various load data in step S1 specifically includes: S1.1: Through intelligent sensing units deployed in various load circuits or production equipment, the real-time power, speed and torque of the motor, the power and temperature of the heating equipment, and the process parameters and energy consumption data of the process equipment are collected synchronously. S1.2: Through the energy management system or data interface of high-energy-consuming enterprises, obtain in real time the time-of-use electricity price signal released by the power grid, the total incoming power, and the future preset production plan from the production scheduling system; S1.3: The collected and acquired data are preprocessed by aligning timestamps, cleaning invalid data, and normalizing, forming a unified time-series integrated data queue.

[0009] This application constructs a unified and high-quality real-time data foundation for enterprise energy information through the synchronous acquisition and standardized preprocessing of multi-source heterogeneous data. It effectively overcomes the shortcomings of traditional data acquisition, such as scattered data collection, inconsistent time scales, and uneven quality. It provides reliable and consistent data input for subsequent high-precision load modeling and optimization decision-making, and is the cornerstone for ensuring the stable and effective operation of the entire collaborative control system.

[0010] In some embodiments of this application, step S2, which involves constructing a multi-objective collaborative optimization function and establishing a load model, specifically includes: S2.1: The multi-objective collaborative optimization function is expressed as: Total cost = Time-of-use electricity cost + Maximum demand cost + Deviation penalty cost, where the deviation penalty cost is dynamically calculated based on the deviation between the actual load operation curve and the production plan requirement curve; S2.2: For adjustable motor loads, establish a nonlinear mapping model between motor power, speed, and load rate, and determine the safe speed regulation range and response delay; for interruptible heating loads, establish a thermodynamic dynamic model and determine the allowable interruption duration and safe temperature fluctuation range; for process loads with thermal inertia, establish an inertial hysteresis model between input energy and key state parameters, where the key state parameters are temperature and pressure. S2.3: Integrate various load models into a multi-objective collaborative optimization function in the form of constraints, so that the optimization solution must be carried out under the premise of satisfying the physical laws and operating limits of various loads.

[0011] This application precisely quantifies the physical characteristics and operational constraints of various loads into mathematical models and embeds them into optimization functions, achieving deep coupling between physical constraints and economic optimization. This ensures that the optimized scheduling instructions follow the physical laws of the equipment and the production safety boundaries from the very beginning of their generation, avoiding the risk of damaging equipment lifespan or affecting product quality in pursuit of economy, and making the optimization results both economical and engineering feasible.

[0012] In some embodiments of this application, step S2 further includes processing the coupling with the external excitation signal: S2.4: In the multi-objective collaborative optimization function, a weighting factor linked to the real-time time-of-use electricity price is set so that during periods of high electricity price, the optimization algorithm tends to reduce total power or transfer movable loads; during periods of low electricity price, it tends to start high-energy-consuming processes in advance or carry out heat and cold storage operations. S2.5: Introduce the predicted maximum demand threshold as a hard constraint into the multi-objective collaborative optimization function.

[0013] This application internalizes external market signals and internal management objectives into the core driving force of the optimization algorithm by setting a weighting factor for electricity price linkage and a hard constraint on demand threshold. This enables the system to respond automatically and intelligently to electricity price fluctuations, guide the load to be optimally distributed in the time dimension, and actively suppress power peaks. As a result, demand-side management is upgraded from passive execution to proactive optimization, directly and significantly reducing electricity cost and demand cost.

[0014] In some embodiments of this application, step S3, which involves rolling optimization and decision coordination across three time scales, specifically includes: At the day-ahead scale, based on the production plan and time-of-use electricity price forecast for the next 24 hours, a multi-objective collaborative optimization function is solved at 1-hour intervals to formulate the baseline operation plan for various loads and the charging and discharging plan for loads with energy storage characteristics. At the intraday rolling scale, with a period of 15 minutes or 30 minutes, the day-ahead baseline operation plan and charging and discharging plan are rolled over and revised based on load forecasts, real-time electricity prices, and actual operating status. At the real-time control scale, with a period of seconds or minutes, adjustment commands are generated and directly issued to the local controller based on system deviations and short-term forecasts to achieve second-level coordination of loads with different response speeds.

[0015] This application employs a multi-timescale hierarchical optimization architecture to decompose the complex global optimization problem into sub-problems of different time granularities for collaborative solution. The current plan provides global economic guidance, intraday rolling corrections enhance adaptability to uncertainty, and real-time control ensures precise and rapid execution of instructions, achieving a unification of long-term optimization goals and short-term dynamic adjustments, enabling the system to smoothly respond to various internal and external changes.

[0016] In some embodiments of this application, the multi-timescale decision coordination in step S3 also includes a cross-scale information interaction mechanism: the real-time control scale uploads infeasible instructions or model mismatch information encountered during execution to the intraday rolling scale to trigger strategy adjustments; the intraday rolling scale uploads the plan adjustment needs caused by external conditions to the day-ahead scale to trigger global plan rescheduling.

[0017] This application's cross-scale information interaction mechanism establishes a bottom-up feedback channel and a top-down adjustment mechanism, enabling model errors or constraint conflicts encountered in the real-time execution layer to trigger timely adjustments to the upper-level strategy, thus avoiding the accumulation of erroneous instructions. At the same time, significant external changes can also trigger the reconstruction of the global plan. The dynamic coordination capability significantly enhances the robustness and adaptability of the system, ensuring the continuous effectiveness of the optimization strategy in complex dynamic environments.

[0018] In some embodiments of this application, the generation and issuance of scheduling instructions in step S4 specifically includes: S4.1: Convert the scheduling instructions generated in step S3 into a standardized set of control commands that can be recognized by each local controller, based on the load type and communication protocol; S4.2: Distribute standardized control command sets to the corresponding edge computing units via industrial Ethernet or wireless network; S4.3: Verify the security of instructions in the edge computing unit and execute them within the preset security boundaries, while transmitting the start and end status of execution and actual effect data back in real time.

[0019] This application ensures that optimization decisions can be safely, accurately, and reliably implemented into equipment actions through standardized conversion, reliable transmission, and edge-side security verification.

[0020] In some embodiments of this application, the coordinated control in step S4 also includes strategies to ensure production continuity: when executing load adjustment instructions, priority is given to adjusting the loads of auxiliary equipment or redundant links that have the least impact on the final product quality; for the adjustment of critical process loads, a slow ramp change method is adopted; a graded response strategy is set up so that when an emergency demand response signal from the power grid is received, non-production-essential heating and cooling loads are interrupted first, and then auxiliary production loads are adjusted.

[0021] This application, by setting adjustment priorities, adopting ramp adjustment and graded response rules, minimizes interference with core production processes while pursuing energy conservation and economy, balancing energy conservation goals with production stability requirements, and ensuring that energy conservation control does not come at the expense of product quality, output or equipment safety, thus greatly improving the industrial applicability and acceptability of the method.

[0022] In some embodiments of this application, step S5, which implements closed-loop optimization through model predictive control, specifically includes: S5.1: Continuously monitor and compare the deviation between the actual response of each load and the expected value of the scheduling instruction generated in step S3; S5.2: Input the deviation into the model predictive control algorithm to predict the system behavior in the next few cycles and calculate the feedforward-feedback composite control quantity that enables the system to return to the optimal trajectory the fastest. S5.3: Using feedforward-feedback composite control variables, the load model parameters established in step S2 are automatically identified and updated periodically.

[0023] This application proactively compensates for uncertainties such as model inaccuracies and unknown disturbances by comparing the deviation between expectations and actual results in real time and continuously optimizing future control sequences. Simultaneously, it automatically updates model parameters using operational data, enabling the system to have self-learning capabilities. This allows it to automatically adjust to changes in equipment performance and operating conditions, thus maintaining excellent optimization performance over the long term and overcoming the problem of traditional static optimization methods' effectiveness decaying over time.

[0024] In some embodiments of this application, step S5 further includes a performance evaluation and self-learning mechanism: periodically calculating and visualizing the total electricity cost, peak demand, and load curve smoothness index before and after implementing collaborative management and control, and evaluating the energy-saving effect; based on historical optimization data and energy-saving effect, training an auxiliary decision-making model to learn the optimal strategy mode under different combinations of external conditions.

[0025] This application demonstrates energy-saving results intuitively through quantitative indicators, providing data support for management decisions. More importantly, by training the auxiliary decision-making model with historical data, it can continuously accumulate and solidify optimal operating experience, enabling the system to respond quickly when dealing with repetitive or similar scenarios, improving optimization efficiency, and gradually reducing reliance on complex online calculations, thus promoting the development of the system towards intelligence and knowledge.

[0026] A second aspect of this application provides an energy-saving control system for high-energy-consuming enterprises based on multi-load coordination, used to implement an energy-saving control method for high-energy-consuming enterprises based on multi-load coordination, including: The data acquisition module is used to execute step S1; The optimization decision module is used to execute step S2; The scheduling instruction module is used to execute step S3; The control execution module is used to execute step S4; The closed-loop feedback module is used to execute step S5.

[0027] In the above embodiments, the system protects the hardware architecture and software module entities that implement the method. Through the collaborative work of each module, efficient, reliable, and scalable physical support is provided for the implementation of the method. Attached Figure Description

[0028] Figure 1 A flowchart illustrating an energy-saving control method for high-energy-consuming enterprises based on multi-load coordination, provided as an embodiment of this application; Figure 2 A schematic diagram of an energy-saving control system for high-energy-consuming enterprises based on multi-load coordination, provided as an embodiment of this application; Figure 3 A schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0029] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0030] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0031] Please refer to Figure 1 , Figure 1This application provides an energy-saving control method for high-energy-consuming enterprises based on multi-load coordination, including the following steps: S1: Real-time acquisition of various load data from high-energy-consuming enterprises, including adjustable motor loads, interruptible heating loads, and process loads with thermal inertia; S2: Based on various load data, construct a multi-objective collaborative optimization function with the minimum comprehensive electricity cost as the core, taking into account time-of-use electricity charges, demand charges, and penalty costs for deviation from production constraints, and establish load models for the power adjustable range and dynamic response characteristics of various loads; S3: Based on the optimization function and load model, rolling optimization and decision coordination are performed on three time scales: day-ahead, intraday, and real-time, to generate scheduling instructions for loads with different response speeds. S4: Send dispatch instructions to the local controllers of each load to perform coordinated control of motor speed, heater power level and process equipment operation; S5: Real-time monitoring of load response status and total energy consumption; dynamic feedback and correction of load model and optimization function through model predictive control algorithm to achieve closed-loop optimization.

[0032] Understandably, high-energy-consuming enterprises refer to industrial enterprises whose energy consumption per unit of output is significantly higher than the social average during the production process. These enterprises typically possess a large and continuously operating group of electrical equipment, with energy costs accounting for a very high proportion of their total production costs. Typical examples include industries such as steel smelting, non-ferrous metal processing, basic chemicals, and cement building materials, all characterized by the extensive use of electric motors, electric heating furnaces, and high-temperature kilns in their production processes. Adjustable motor loads refer to equipment driven by electric motors whose operating speed or torque can be continuously or segmentally adjusted by power electronic devices. For example, fans, pumps, and compressors on production lines can have their motor speed adjusted according to actual process requirements by adding frequency converters, thereby changing their input power and achieving on-demand energy supply. This is the main energy-saving adjustment potential of this type of load. Interruptible heating loads refer to electric heating and space heating equipment that can temporarily suspend power supply or reduce operating power without affecting core production processes or safety. For example, auxiliary electric heating furnaces used for preheating raw materials and HVAC systems in non-production areas can be temporarily shut down or operated at reduced speeds during peak electricity consumption periods to reduce the overall load. Process loads with thermal inertia specifically refer to electric heating equipment whose internal temperature or key state parameters change slowly due to their physical structure or process. For example, glass melting furnaces, ceramic tunnel kilns, or large heat treatment furnaces have enormous internal heat storage capacity, and both heating and cooling processes require considerable time. This thermal inertia gives their power regulation a certain delay and predictability, allowing for power reduction within a certain time frame without causing a drastic change in furnace temperature, thus providing a time window for load adjustment.

[0033] The comprehensive cost of electricity is a composite cost indicator designed to comprehensively measure the economic efficiency of a company's electricity consumption. It includes not only the basic electricity charge calculated based on electricity consumption and time-of-use pricing, but also the demand charge based on the company's maximum power consumption within a billing cycle, and the penalty cost calculated for output loss or quality risks caused by load adjustments that may deviate from the original production plan. This indicator is a core objective for optimizing dispatch and minimizing costs. Time-of-use pricing is a pricing mechanism implemented by electricity suppliers to guide users to use electricity rationally and smooth the grid load curve. It divides the 24 hours of a day into peak, off-peak, and low-peak periods, setting differentiated prices for each, with the highest price during peak hours and the lowest during off-peak hours. Adjusting the user's electricity consumption period directly affects electricity costs. Demand charge is a special fee in electricity billing, charged based on the user's maximum average power consumption within a billing cycle. This fee aims to reflect the cost of the power supply capacity that the grid must reserve to meet the user's instantaneous maximum electricity demand. By actively managing and reducing peak electricity consumption, companies can effectively reduce this cost. The penalty cost for deviating from production constraints is a virtual cost term set in the optimization model to quantify the potential production losses or risks caused by executing load adjustment commands. When the scheduling scheme suggested by the optimization algorithm to reduce electricity costs causes production rhythm, product quality, or equipment safety to deviate from the predetermined allowable range, this penalty cost increases significantly, thus guiding the optimization scheme to automatically tend towards seeking economic optimality while ensuring core production requirements. The multi-objective collaborative optimization function is a mathematical expression that takes minimizing the aforementioned comprehensive electricity cost as its core objective, while simultaneously considering multiple sometimes conflicting demands such as equipment operating boundaries, production process continuity, and grid interaction requirements through constraints or additional objective terms. Its purpose is to find a system operation scheme that achieves the best balance among multiple objectives through mathematical programming methods. A load model is a mathematical or logical abstraction used to describe the electrical and controllable characteristics of various types of electrical loads. It typically includes a static model that quantitatively describes the relationship between load power and its controllable variables, and a dynamic response model that describes its state changes over time. For example, a motor load model may include a curve showing the relationship between its power and speed, as well as the time constant required from receiving a speed adjustment command to the actual power reaching a new steady state.

[0034] Adjustable power range refers to the extent to which the input electrical power of a specific load can be adjusted within its permissible safe and process operating range. For example, the adjustable power range of a variable frequency water pump may be between 30% and 100% of its rated power, defining the boundary of the load's ability to participate in system regulation. Dynamic response characteristics describe the speed and manner in which the load's output power or state changes in response to its control commands. For example, the power response of a resistance heater is almost instantaneous, while the temperature response of a large kiln is a slow inertial process. Understanding these characteristics is crucial for designing coordinated control strategies. Day-ahead scale typically refers to the time range of the next 24 hours or one operating day. Optimization at this scale mainly relies on relatively reliable production plans and electricity price forecasts to develop a baseline plan for load operation throughout the day, focusing on overall economic arrangements. Intra-day scale typically refers to the time range of several hours after the current moment. Rolling optimization at this scale involves fine-tuning and correcting the day-ahead plan based on the latest ultra-short-term load forecasts, real-time electricity price information, and the actual system status to address forecast errors and short-term fluctuations. Real-time scale typically refers to the time range of several minutes or even seconds after the current moment. Control decisions at this scale directly address the equipment, demanding extremely high speed and accuracy. They are used to execute fine-grained power regulation and dynamic balancing to track upper-level scheduling commands and suppress real-time disturbances. Rolling optimization is an iterative optimization strategy. Instead of performing a one-time long-term optimization at a single point in time, it re-executes the optimization calculation within a finite time horizon at each decision point, based on updated information and the latest system state, thereby continuously generating optimal or suboptimal control commands that adapt to new situations. Decision coordination refers to the mechanism that ensures the interconnection and mutual support of optimization objectives, constraints, and control commands across multiple time scales or different subsystems. It guarantees the continuity and consistency from long-term planning to short-term execution, avoiding contradictions between decisions at different levels. Scheduling commands are the output of optimization calculations, commands issued to specific load controllers for execution. Their form depends on the load type and may be a target speed setpoint, a power level command, or a setpoint curve showing the time-varying changes of a key process parameter.

[0035] A local controller is an automated device installed at the production equipment site that directly drives or regulates the operation of the equipment, such as a programmable logic controller (PLC), frequency converter, or dedicated process controller. It is responsible for receiving and executing scheduling instructions from the upper-level system. Model predictive control (MMC) is an advanced process control method that uses a dynamic load model to predict the system's behavior trajectory under different control actions over a future period. By solving an optimization problem online, it calculates a series of control actions that best approximate the predicted trajectory to the desired target. Typically, only the first control action is executed, and this process is repeated in the next cycle. This model-based, rolling optimization, and feedback correction characteristic makes it highly suitable for handling constrained multivariable cooperative control problems.

[0036] Closed-loop optimization refers to the operating mode of the entire control system. It's not simply about generating and executing scheduling commands once and for all, but rather continuously feeding back the actual effects of command execution through a monitoring system and comparing them with the expected results. Using this feedback information, the system can automatically correct its internal model parameters or optimization strategies, leading to more accurate decisions in the future. This cyclical process of perception, decision-making, execution, feedback, and readjustment enables the system to adapt to changes and continuously improve its performance.

[0037] This application addresses the challenge of coordinated control of diverse heterogeneous loads, such as motors, heating systems, and processes with high thermal inertia, in high-energy-consuming enterprises by establishing a load model, constructing a multi-objective optimization function that considers both economic efficiency and production constraints, and employing a combination of multi-timescale rolling optimization and model predictive control closed-loop feedback. By integrating the discrete load resources of high-energy-consuming enterprises into a flexibly controllable virtual whole, it achieves a leap from equipment-level energy saving to system-level optimization. Under the premise of strictly ensuring production safety and continuity, it significantly reduces overall electricity costs and improves energy utilization efficiency and grid interaction capabilities.

[0038] In some embodiments disclosed in this application, the real-time acquisition of various load data in step S1 specifically includes: S1.1: Through intelligent sensing units deployed in various load circuits or production equipment, the real-time power, speed and torque of the motor, the power and temperature of the heating equipment, and the process parameters and energy consumption data of the process equipment are collected synchronously. S1.2: Through the energy management system or data interface of high-energy-consuming enterprises, obtain in real time the time-of-use electricity price signal released by the power grid, the total incoming power, and the future preset production plan from the production scheduling system; S1.3: The collected and acquired data are preprocessed by aligning timestamps, cleaning invalid data, and normalizing, forming a unified time-series integrated data queue.

[0039] In the embodiments disclosed in this application, step S1 is the basic data layer for realizing subsequent optimization and control, the core of which is to build a real-time, unified and high-quality panoramic view of enterprise energy and production information.

[0040] As can be understood, an intelligent sensing unit refers to an intelligent measurement device that integrates sensing elements, a microprocessor, a communication interface, and embedded software. Unlike traditional analog sensors that only output raw electrical signals, intelligent sensing units perform signal conditioning, analog-to-digital conversion, digital filtering, and nonlinear correction locally, directly outputting engineering unit values, and possessing self-diagnosis, self-calibration, and network communication capabilities. For example, an intelligent power sensor deployed in a motor circuit can output real-time fused data of multiple parameters such as active power, reactive power, speed, and torque, while an intelligent temperature sensor can directly output a Celsius value after cold junction compensation and linearization.

[0041] Synchronous data acquisition refers to acquiring measurement data from multiple dispersed measurement points at the same time or strictly equal time intervals, using a unified time base and trigger signal. Its technical implementation relies on high-precision clock synchronization protocols such as IEEE 1588 or GPS timing to ensure that intelligent sensing units distributed in different geographical locations maintain time consistency within microsecond-level precision. The purpose of synchronous acquisition is to eliminate data phase differences caused by different acquisition times, providing a unified time base prerequisite for subsequent multi-source data fusion and causal analysis.

[0042] An enterprise energy management system is a comprehensive energy monitoring and management software platform deployed within high-energy-consuming enterprises. This system connects to intelligent instruments and control equipment distributed throughout workshops and processes via industrial communication networks, collecting real-time consumption data for various energy media such as electricity, water, gas, and heat. It also possesses basic functions such as data storage, statistical analysis, cost accounting, anomaly alarms, and report generation. This system typically also provides standard data interfaces, allowing upper-level optimization systems to read the required real-time energy consumption and process data.

[0043] A production scheduling system is a business information system used to formulate, issue, and track enterprise production operation plans. Based on sales orders, inventory status, equipment capacity, and process routes, this system generates the production types, outputs, and schedules for each production line and section in the future. In high-energy-consuming enterprises, the production plan output by this system is a key input for predicting load curves and determining shiftable load windows.

[0044] The future pre-set production plan refers to the production operation arrangement generated and issued by the production scheduling system, covering a specific future time window. This plan is usually presented in time series form, including the expected output, equipment start / stop status, and process formulas for each production unit at each moment within, for example, the next 24 hours or a shift. This plan is the core basis for determining the load baseline curve in day-ahead scale optimization scheduling.

[0045] Timestamp alignment refers to the process of mapping data points from different data sources, each with its own independent timestamp, onto the same standard time coordinate axis. Because different acquisition terminals or business systems may have clock deviations or sampling period differences, timestamp alignment requires first calibration using a clock synchronization protocol, and then using interpolation or resampling techniques to adjust all data sequences to a common, equal time interval, such as one data point per second or every fifteen minutes.

[0046] Invalid data cleaning refers to the process of identifying and processing outliers in raw data caused by sensor malfunctions, communication interference, human error, etc. Cleaning targets include outliers exceeding physical measurement ranges, zero or null values ​​caused by communication interruptions, constant dead zones caused by sensor drift, and non-realistic operating data during equipment maintenance. Typical processing methods include threshold truncation, median filtering, and smoothing repair based on neighborhood mean.

[0047] Normalization preprocessing refers to a processing technique that maps various data features with different physical dimensions and numerical ranges to a unified scale range through mathematical transformation. Since electric motor power can reach hundreds of kilowatts, while kiln temperature is only on the order of thousands of degrees Celsius, without normalization, features with large numerical magnitudes will dominate the weights in subsequent optimization algorithms. Common methods include min-max normalization, which linearly maps data to the zero-to-one interval, and Z-score standardization, which normalizes data into a distribution with a mean of zero and a standard deviation of one.

[0048] A comprehensive data queue refers to a multidimensional time-series dataset that has undergone alignment, cleaning, and normalization preprocessing, and is arranged in an ordered manner along a unified time axis. This queue is indexed by time, and each data record at any given moment contains a feature vector containing comprehensive information on all loads at that moment, including power, status, process parameters, electricity price signals, and production plans. This queue serves as standardized data input for subsequent load modeling and optimization solutions.

[0049] Steps S1.1 to S1.3 together constitute a complete closed loop for multi-source heterogeneous data acquisition and preprocessing. Its core principle can be decomposed into three tightly coupled technical layers.

[0050] The first layer is the data mapping of the physical world. By deploying intelligent sensing units in various load circuits and production equipment, the electromagnetic processes of motors, the thermodynamic processes of heating equipment, and the material conversion processes of process equipment are converted into calculable and transmittable digital signals in real time. Taking the load of an adjustable motor as an example, intelligent power sensors use the Hall effect or current transformer principle to obtain voltage and current waveforms, which are then used by a digital signal processor to calculate active power. Simultaneously, the rotor speed is obtained through an encoder or magnetoelectric speed sensor, while the torque sensor obtains mechanical torque based on strain gauges or magnetostriction effect. These multidimensional parameters are synchronously sampled and packaged into data frames, which are then uploaded via fieldbus or industrial Ethernet. The temperature data of heating equipment typically uses PT100 resistance temperature detectors or thermocouples paired with intelligent temperature transmitters. After cold junction compensation and linearization at the equipment end, the Celsius value is directly output.

[0051] The second layer involves the access and integration of information from external systems. The enterprise energy management system has already aggregated the total incoming power and electricity metering data for each branch, serving as the core source for obtaining total power consumption and individual energy consumption data. Time-of-use (TOU) electricity price signals from the power grid are typically issued through the power trading platform's API interface or the dispatch system, containing price codes and corresponding prices for each time period of the day or the next day. Future pre-set production plans from the production dispatch system are obtained through database interaction interfaces or middleware integration, usually stored in relational data tables, including fields such as timestamps, production line numbers, product specifications, and planned output. These three types of external data are inherently asynchronous with the enterprise's internal load data in terms of time. For example, electricity price signals change on an hourly or 15-minute basis, production plans are issued hourly or in batches, while load power data is sampled continuously at the second level. Therefore, a preprocessing stage is necessary.

[0052] The third layer involves data quality governance and engineering representation. First, timestamp alignment uses a master clock server to synchronize clocks across all data sources. Cubic spline interpolation or linear interpolation algorithms are then used to uniformly resample all sequences to the set optimization cycle reference frequency. For example, if the daily rolling optimization cycle is fifteen minutes, all second-level power data must be aggregated and calculated as the average within that time window, and the electricity price change time must be precisely matched to the corresponding optimization period. Second, invalid data cleaning employs composite criteria, including a fixed threshold method based on physical limits (e.g., motor power should not exceed 120% of the rated value), a sliding window method based on statistics (e.g., if the current value deviates from the mean of the previous five sampling points by more than three standard deviations, it is considered an outlier), and a logical rule method based on equipment status (e.g., power data should be zero when the equipment is in a shutdown state). Finally, normalization preprocessing uses min-max normalization to scale all features to the zero-to-one interval. The transformation formula is: the normalized value equals the original value minus the minimum value of the feature divided by the maximum value minus the minimum value of the feature. This transformation preserves the original data distribution and places the weights of each feature's influence on the optimization objective on the same dimension. After these three steps, the resulting time-indexed multidimensional data queue possesses three major characteristics: determinism, consistency, and computability, and can be directly input into a multi-objective collaborative optimization function solver.

[0053] Take a professional manufacturer of high-temperature structural ceramics as an example. This company's main product is high-performance alumina ceramics, widely used in aerospace and semiconductor equipment components. Its production process includes four main stages: powder preparation, pressing and molding, high-temperature sintering, and precision machining after sintering. The high-temperature sintering stage is completed by three large gas-fired and electrically heated tunnel kilns, with sintering temperatures reaching 1600 degrees Celsius and a single kiln power exceeding 800 kilowatts. This is a typical process load with enormous thermal inertia, resulting in extremely high hardness of the sintered ceramic blanks, with Rockwell hardness reaching HRA 90 or higher. Diamond-coated tools are typically used for drilling, cutting, and grinding in subsequent machining processes. These machining equipment mainly consist of dozens of CNC machine tools and machining centers. Their spindle motors and feed servo motors are typical adjustable motor loads. Furthermore, the factory covers a vast area, and large air-source heat pump units are required for heating in the production workshops and office buildings during winter, representing an interruptible heating load.

[0054] In this embodiment, the specific deployment and execution of step S1 are as follows: In the data acquisition phase, the company deploys intelligent platinum-rhodium thermocouple temperature sensors in each temperature zone of each tunnel kiln. These sensors incorporate cold junction compensation and a 24-bit analog-to-digital converter, achieving a measurement accuracy of ±0.25 degrees Celsius. The kiln temperature is reported every two seconds via an RS-485 bus. Optical pyrometers are deployed at the kiln head and tail to monitor the actual surface temperature of the brick blanks. Each burner's gas flow regulating valve and electric auxiliary heating power module is equipped with an intelligent power monitoring instrument, which collects voltage, current, active power, and power factor in real time. In the CNC machine tool area, an intelligent power module is installed at the main power input of each machine tool. This module uses a DSP chip to calculate the instantaneous power, load rate, and spindle speed during machining in real time. Torque data is obtained by reading the internal bus of the machine tool's CNC system. The plant's heating heat pump units have intelligent meters and temperature controllers installed in their distribution cabinets to collect data on the unit's total power, outlet water temperature, and return water temperature.

[0055] In terms of external system integration, the company has established a unified energy management center. The optimization system reads the total incoming power and the power consumption of each branch from the company's energy management system in real time via the OPC UA protocol. The power grid company pushes the next day's time-of-use electricity price schedule through the power trading system at 10:00 AM daily, and the system automatically retrieves and parses this file via a Web Service interface. The production scheduling system is the core module of the enterprise resource planning system. It generates a 24-hour processing task schedule for each machine tool based on sales orders daily, including the quantity of ceramic parts to be processed, the estimated processing time, and the latest allowed completion time. The system reads the updated production plan every 15 minutes through synchronization with an intermediate database table.

[0056] In the data preprocessing stage, the system faces significant challenges due to data heterogeneity. Kiln temperature data is reported every two seconds, machine tool power data fluctuates dramatically with spindle start / stop and cutting load, and is sampled at a frequency of one second. Heat pump unit power data is uploaded every ten seconds, while electricity price signals change only every fifteen minutes, and production plans are issued at the hourly level. The system first synchronizes all data acquisition terminals and servers with millisecond-level clocks using an NTP time server deployed in the enterprise's central computer room. Subsequently, a linear interpolation algorithm is used to resample the kiln temperature sequence to equal intervals of fifteen seconds, and the machine tool power sequence to equal intervals of five seconds. To meet the requirement of a fifteen-minute daily rolling optimization cycle, the system employs a sliding window averaging method to calculate the time-period average power from all second-level power data within fifteen minutes, ensuring that the electricity price code for that time period is precisely aligned with the electricity price period published by the power system. During invalid data cleaning, on a certain day, the temperature measuring thermocouple of a sintering kiln output a constant value of zero due to long-term high-temperature aging. Based on the temperature change trend of the kiln before and after the event and the correlation of adjacent measuring points, the system identified this data anomaly and automatically marked it as invalid. Simultaneously, it used linear interpolation of adjacent temperature zones to fill the invalid data. In the normalization process, the system establishes a feature library, recording the maximum and minimum power values ​​of each load over the past week. It maps various types of real-time collected data online to the zero-to-one interval and continuously updates the normalization parameters to adapt to changes in operating conditions.

[0057] After the above processing, the system generates a new comprehensive data queue for the park every fifteen minutes. This queue uses a timestamp as the primary key, and each data record contains the temperature of each temperature zone of the three kilns, the power of each burner, the spindle power and load rate of the twenty CNC machine tools, the power and water temperature of the six heat pump units, the total incoming power of the enterprise, the real-time electricity price code, and a list of tasks to be processed by each machine tool for the next two hours. This queue serves as standardized input, directly driving the construction of the multi-objective collaborative optimization function and the parameter identification of the load model in subsequent step S2.

[0058] This embodiment, through the refined data acquisition and preprocessing scheme in step S1, lays a reliable data foundation for energy-saving control methods in high-energy-consuming enterprises. Its beneficial technical effects are reflected in the following aspects: First, the localized digital processing capability of the intelligent sensing unit significantly improves the accuracy and anti-interference ability of the raw data, eliminating accuracy loss during signal transmission and ensuring that key parameters such as motor load rate and kiln temperature gradient truly reflect the physical state of the equipment. Second, the synchronous acquisition of multi-source heterogeneous data and integration with external systems breaks down the silos between energy and production information within the enterprise, enabling optimization decisions to comprehensively consider electricity costs, production progress, and equipment health status within the same time frame. Third, engineered data cleaning and normalization preprocessing eliminates the risk of optimization solution failure due to occasional sensor malfunctions or communication interruptions, and provides a unified mathematical expression for heterogeneous data from equipment from different manufacturers and eras, greatly enhancing the versatility and transferability of the optimization algorithm. Taking this high-temperature ceramics enterprise as an example, after implementing this solution, the system can accurately identify the power curve characteristics of the diamond drill when processing large-size ceramic plates, and use it as a transferable load to coordinate and optimize the heating demand during the empty kiln period, thereby achieving a significant reduction in electricity costs without delaying the delivery of subsequent processes.

[0059] In some embodiments disclosed in this application, step S2, which involves constructing a multi-objective collaborative optimization function and establishing a load model, specifically includes: S2.1: The multi-objective collaborative optimization function is expressed as: Total cost = Time-of-use electricity cost + Maximum demand cost + Deviation penalty cost, where the deviation penalty cost is dynamically calculated based on the deviation between the actual load operation curve and the production plan requirement curve; S2.2: For adjustable motor loads, establish a nonlinear mapping model between motor power, speed, and load rate, and determine the safe speed regulation range and response delay; for interruptible heating loads, establish a thermodynamic dynamic model and determine the allowable interruption duration and safe temperature fluctuation range; for process loads with thermal inertia, establish an inertial hysteresis model between input energy and key state parameters, where the key state parameters are temperature and pressure. S2.3: Integrate various load models into a multi-objective collaborative optimization function in the form of constraints, so that the optimization solution must be carried out under the premise of satisfying the physical laws and operating limits of various loads; S2.4: In the multi-objective collaborative optimization function, a weighting factor linked to the real-time time-of-use electricity price is set so that during periods of high electricity price, the optimization algorithm tends to reduce total power or transfer movable loads; during periods of low electricity price, it tends to start high-energy-consuming processes in advance or carry out heat and cold storage operations. S2.5: Introduce the predicted maximum demand threshold as a hard constraint into the multi-objective collaborative optimization function.

[0060] Understandably, time-of-use electricity cost refers to the electricity expenditure calculated by integrating the purchased power of an enterprise over time according to the time-of-use pricing system published by the power company, multiplying it by the corresponding time-of-use electricity price, and then summing the results. Maximum demand cost refers to the capacity fee charged by the power company based on the maximum sliding window average power of an enterprise within a billing cycle. This power is usually determined by the maximum average active power of the sliding window over a continuous 15-minute or 30-minute period. Deviation from production constraints penalty cost is a virtual economic quantity used to quantify the equivalent economic loss caused by the failure of production tasks to be executed as planned due to load optimization scheduling. Its value is calculated by multiplying the deviation between the planned completion time and the actual completion time by a unit time loss coefficient, or by multiplying the integral deviation between the actual power curve and the planned power curve by a unit power loss coefficient.

[0061] The multi-objective collaborative optimization function is a weighted sum or hierarchical optimization expression that unifies three heterogeneous economic objectives—time-of-use electricity cost, maximum demand cost, and penalty cost for deviating from production constraints—under the same mathematical framework. Its solution space is jointly defined by the physical and technological constraints of various loads, and the solution result is a sequence of optimal power setpoints for each load in each time period.

[0062] Nonlinear mapping models are mathematical functions that describe the non-proportional relationship between the input power of a motor and its speed and load rate. Since the iron loss, copper loss and mechanical loss of the motor change nonlinearly with the speed during variable frequency speed regulation, this model is usually expressed by a quadratic polynomial, a cubic polynomial or a piecewise linear interpolation function based on measured data.

[0063] A thermodynamic dynamic model refers to a differential or difference equation describing the temperature change of a heating device over time and with input power during power on / off or power adjustment. This model simplifies the device to a series structure of a single heat capacity and a single thermal resistance based on the lumped parameter method. Core parameters include heat capacity, thermal resistance, ambient temperature, and initial temperature. An inertial hysteresis model describes the dynamic response mathematical relationship of a process device with large heat capacity as its key internal state parameters slowly approach a new steady state over time after receiving energy input. It is usually expressed as a first-order inertial element plus a pure hysteresis transfer function or a corresponding difference equation. Core parameters include steady-state gain, inertial time constant, and pure hysteresis time.

[0064] The safe speed regulation range refers to the closed interval formed by the lowest and highest speeds of the motor that allow for long-term continuous operation, provided that the load's mechanical characteristics, its own heat dissipation conditions, and control stability are met. The permissible interruption duration refers to the maximum time that an interruptible heating / cooling device can continuously disconnect its power supply while maintaining the temperature above the process requirements or comfort safety threshold. This duration is derived analytically or numerically from a thermodynamic dynamic model under given initial and ambient temperatures. The temperature fluctuation safety range refers to the upper and lower limits of the temperature deviation allowed from the set value during normal operation of the heating / cooling device or thermal inertia process load. This range is determined by process quality specifications or equipment safety regulations; exceeding the range triggers a penalty cost for deviating from production constraints or results in direct scrapping. Constraints refer to a series of mathematical expressions of equality and inequality embedded in a multi-objective collaborative optimization function, used to limit the feasible domain of the optimization variables. These constraints include the physical equations of the load model, equipment operating limits, power balance equations, production plan matching requirements, etc., ensuring that the optimization solution is not only mathematically optimal but also engineering-executable. A weighting factor is an adjustable coefficient introduced into a multi-objective collaborative optimization function to amplify or reduce the influence of a specific objective term on the optimization result. In this step, this factor is linked to the real-time time-of-use electricity price. Hard constraints refer to mathematical constraints that must be strictly satisfied.

[0065] The multi-objective collaborative optimization function decomposes the enterprise's electricity cost into three explicitly computable cost items. Assume the optimization time horizon is divided into T equal-length time periods, each with a length of Δt in hours. Then the multi-objective collaborative optimization function is expressed as:

[0066] in, Time-of-use electricity cost: the sum of the products of the electricity purchased in each time period and the corresponding electricity price; Maximum demand cost: The product of the maximum demand value this month and the unit price of demand; Deviation from production constraints penalty cost: the sum of the product of the actual time of each task exceeding the planned time and the unit delay loss coefficient; Power curve deviation penalty cost: the product of the integral value of the absolute deviation between actual power and planned power in each time period and the penalty coefficient; Pt: Total power purchased by the enterprise during time period t (unit: kilowatts); ct: Time-of-use electricity price for period t (unit: yuan / kWh); Pmax: The maximum demand value (unit: kilowatts) that occurs within the current billing cycle. This value is determined by the power company based on the maximum value of the average power over a continuous 15-minute sliding window and is used as an auxiliary variable to be optimized in the optimization model. λ: Unit price of demand (unit: yuan / kilowatt); J: Total planned production targets for this month; : The actual completion time of the j-th task (in hours); : The planned completion time of task j (in hours); αj: Penalty cost for each hour of delay in the j-th task (unit: yuan / hour); β: Penalty factor for unit power curve deviation (unit: yuan / kilowatt), used for scenarios with large power tracking deviations.

[0067] Considering the real-time tracking deviation of the power curve, an additional item can be added to the above:

[0068] in, Actual power; Planned power.

[0069] For adjustable motor loads, a nonlinear mapping model is established. The relationship between the motor's input power and speed is nonlinear, especially in variable frequency speed control scenarios. Ignoring load rate changes (which are implicitly determined by the production plan), a quadratic polynomial can be used to fit the steady-state power characteristics.

[0070] Wherein, Pm: the input power of the motor, in kilowatts (kW). n: Motor speed, in revolutions per minute (rpm); a, b, and c are the fitting coefficients, which are identified through measured data.

[0071] The safe speed regulation range is constrained as follows:

[0072] nmin: Minimum permissible continuous operating speed, in revolutions per minute (rpm). nmax: Maximum permissible speed, in revolutions per minute (rpm); A thermodynamic dynamic model is established for interruptible heating loads. Its core lies in the boundaries of temperature comfort or basic process requirements, focusing on whether the temperature drop during the interruption process reaches the lower limit. Let the room temperature or medium temperature be Θt in degrees Celsius, and its dynamic changes be described by the thermodynamic model:

[0073] in Θt represents the room temperature in degrees Celsius at time t. Θamb represents the ambient temperature in degrees Celsius. C: Equivalent heat capacity, in joules per degree Celsius (J / °C), indicating the heat storage capacity of the equipment; R: Equivalent thermal resistance, measured in degrees Celsius per watt (°C / W), indicating thermal insulation performance; η: Electrothermal conversion efficiency; Ph,t: Heating power during time period t, in kilowatts (kW). When the equipment is powered off, Ph,t=0.

[0074] The safe temperature fluctuation range is denoted as the lower comfort limit Θmin and the upper comfort limit Θmax, in degrees Celsius. The allowable interruption duration toff is determined by the time required for the current room temperature Θ0 to drop to the lower limit Θmin.

[0075] For process loads with thermal inertia, an inertial hysteresis model is established. Taking temperature as an example, the first-order inertial plus pure hysteresis transfer function of the input power and temperature response is discretized to obtain:

[0076] Tt: Process temperature during time period t, in °C; Pt: Input power at time t, in kilowatts (kW). K: Steady-state gain, in °C / kW, representing the steady-state temperature change caused by a unit power change; τp: Inertial time constant, in seconds (s), characterizing the response rate of temperature to power changes; θ: Pure time delay, in seconds (s), the time delay from power change to the onset of temperature response; The safe range of process temperature is denoted as the lower limit of the process. and process upper limit The unit is degrees Celsius. This interval is used as a strictly embedded optimization problem with hard constraints:

[0077] Additional power change rate constraints were also introduced to prevent thermal shock:

[0078] Let ωt be the weighting factor for the electricity price linkage. This factor is positively correlated with the real-time electricity price ct. A typical setting is as follows:

[0079] where cbase is the reference electricity price, with the unit of yuan per kilowatt-hour. For example, the average electricity price throughout the day is used as the reference electricity price; α is the user-adjustable coefficient, which is used to control the response intensity. After multiplying this weight factor into the time-of-use electricity cost item, the time-of-use electricity cost in the objective function becomes:

[0080] When ct > cbase, ωt > 1, which equivalently amplifies the electricity cost during this period; when ct < cbase, ωt < 1, which equivalently reduces the electricity cost during this period.

[0081] In this application, three types of load models are introduced and transformed into equality constraints or inequality constraints in the optimization problem, and are solved联立 with the multi-objective collaborative optimization function in step S2.1.

[0082] In this application, a complete mathematical mapping mechanism from economic objectives to physical constraints is constructed, which unifies the time-of-use electricity cost, demand charge, and penalty cost for deviating from production constraints into quantifiable optimization objectives. Nonlinear mapping models, thermodynamic dynamic models, and inertial lag models are respectively established for adjustable motors, interruptible heating and heating loads, and process loads with thermal inertia. At the same time, a weight factor linked to the real-time time-of-use electricity price is introduced to dynamically adjust the response intensity, and the predicted maximum demand threshold is embedded as a hard constraint term in the solution process, finally forming a unified control framework with deep coupling of economic optimization and physical feasibility guarantee. On the premise of ensuring the stability of the core production process and the safety of equipment, the systematic optimization of the enterprise's comprehensive electricity cost is achieved.

[0083] In some embodiments disclosed in this application, the rolling optimization and decision coordination in step S3 at three time scales specifically include: On the day-ahead scale, based on the production plan and time-of-use electricity price forecast for the next 24 hours, with an interval of 1 hour, the multi-objective collaborative optimization function is solved to formulate the reference operation plan for various loads and the charge-discharge plan for loads with energy storage characteristics; on the intra-day rolling scale, with a period of 15 minutes or 30 minutes, based on the load forecast, real-time electricity price, and actual operation status, the day-ahead reference operation plan and charge-discharge plan are rolled and corrected; on the real-time control scale, with a period of seconds or minutes, according to the system deviation and short-term forecast, adjustment instructions directly sent to the local controller are generated to achieve the second-level coordination of loads with different response speeds.

[0084] It can be understood that the day-ahead scale refers to the decision-making level with a 24-hour optimization horizon, typically with an optimization cycle of one hour. Based on relatively reliable production plans and electricity price forecasts, it formulates a baseline plan for load operation in each time period of the following day. The intraday rolling scale refers to the optimization and correction level that is repeatedly performed at 15-minute or 30-minute intervals during the day's operation. It dynamically adjusts the day-ahead plan based on the latest ultra-short-term load forecasts, real-time electricity price information, and the actual operating status of the system. The real-time control scale refers to the rapid response level with a second- or minute-level cycle. Based on the current system deviation and ultra-short-term forecasts, it generates precise adjustment commands directly issued to the equipment controllers. Rolling optimization refers to the optimization strategy of resolving an optimization problem within a finite time horizon based on the latest information at each decision point, executing only the decision made at the current moment, and repeating this process at the next moment. Decision coordination refers to establishing a mechanism for target transfer, constraint linkage, and information feedback between different time scales to ensure consistency between long-term plans and short-term controls. The baseline operating plan refers to the sequence of load power setpoints obtained from day-ahead scale optimization, serving as the benchmark for subsequent rolling corrections. A charge / discharge plan refers to the scheduling arrangement of charging and discharging periods and corresponding power for devices with energy storage characteristics, such as thermal storage systems, on a day-ahead scale. System deviation refers to the difference between the actual load response value and the expected value of the dispatch command, including power deviation, temperature deviation, and completion time deviation. Short-term forecasting refers to the prediction of variables such as load, electricity price, and weather within a relatively short future time window based on historical data and real-time information; its accuracy is higher than that of day-ahead forecasting.

[0085] The current scale addresses global economic issues, optimizing the time horizon to cover the next 24 hours, dividing the day into 24 time periods with a one-hour duration. Inputs to this level include: planned output and expected operating times for each production line and equipment within the next 24 hours as per the production scheduling system; the next-day time-of-use electricity price curve published or predicted by the power grid company; physical model parameters and constraints for various loads; and the initial state of charge and capacity parameters of energy storage devices.

[0086] The day-ahead optimizer, based on the multi-objective collaborative optimization function constructed in step S2, aims to minimize the overall electricity cost over the next 24 hours and solves for the power setpoints of each load in each time period. The solution process must satisfy power balance constraints, physical constraints of various load types, energy balance constraints of energy storage devices, and time constraints of the production plan. The output of the day-ahead optimizer includes: the baseline operating plan for various load types, i.e., the expected power of each load per hour over the next 24 hours; the charging and discharging plan for energy storage devices, i.e., whether energy storage is charging, discharging, or idle in each time period; and the predicted total daily electricity cost and maximum demand.

[0087] The core function of the day-ahead plan is to provide a globally optimal baseline trajectory, laying the foundation for intraday rolling adjustments. Due to the uncertainty of day-ahead forecasts, this plan is not a command that must be strictly followed, but rather serves as a reference target and boundary condition for intraday optimization.

[0088] Intraday rolling scales address the issues of forecast errors and short-term fluctuations. Their optimization time horizon is typically 2 to 4 hours into the future, executed in 15- or 30-minute cycles. At each optimization moment, the system re-solves a finite-time optimization problem based on the latest state information.

[0089] The daily rolling inputs include: the current actual load power, equipment status, energy storage charge status, real-time temperature, etc., obtained from the data acquisition module; real-time electricity price signals obtained from external interfaces if they deviate from the day-ahead forecast; load forecast curves for each period in the next 2 to 4 hours provided by the ultra-short-term load forecast module; and the target values ​​for the corresponding periods in the day-ahead plan.

[0090] The objective function of the intraday optimization solver is the same as that of the day-ahead optimization, but a tracking term for the day-ahead plan is added to the constraints. This is usually expressed as a soft constraint, allowing a certain degree of deviation but incurring a penalty cost. After optimization, only the decision for the current moment is executed, i.e., updating the load power setpoints for the next 15 or 30 minutes, and using these setpoints as the target for the real-time control layer. In the next optimization cycle, the system repeats the above process, achieving dynamic optimization that is continuously updated and progresses forward.

[0091] The core advantage of intraday rolling correction lies in its ability to respond promptly to uncertainties such as real-time electricity price changes, production plan fine-tuning, and sudden equipment malfunctions, thus avoiding suboptimal operation caused by inaccurate day-ahead forecasts.

[0092] Real-time control addresses the issues of instantaneous balance and precise tracking, with control cycles on the order of seconds or minutes, and interacts directly with the device controller. This level does not perform complex global optimization; instead, it uses model predictive control algorithms to calculate control variables that enable the system to quickly return to its optimal trajectory based on current system deviations and very short-term predictions.

[0093] The inputs for real-time control include: real-time data such as actual power, speed, and temperature fed back from the local controller; the target setpoint for the current period issued by the intraday rolling layer; and the load change trend for the next few minutes provided by the short-term forecasting module.

[0094] Real-time control algorithms typically employ a model predictive control framework. The core steps are as follows: based on the current system state and load dynamic model, predict the system's behavior trajectory under different control actions within a finite time domain in the future; with the optimization objective of tracking the intraday target and minimizing the deviation, solve a simplified optimization problem to obtain the control sequence for several future control cycles; execute only the instructions for the first control cycle, sending the adjustment amount to the local controller; in the next control cycle, repeat the above process based on the latest feedback state.

[0095] For loads with different response speeds, the real-time control layer adopts a differentiated processing strategy: for fast-response loads such as motors, precise speed or power setpoints are directly issued; for medium-response loads such as heaters, power level or temperature setpoints are issued, and the local controller completes closed-loop regulation; for process loads with large inertia, power change rate commands are issued to ensure smooth temperature transition.

[0096] The three time scales form an organic whole through target transmission and information feedback. The day-ahead plan serves as the global optimal benchmark, providing target references and boundary constraints for intraday rolling; the intraday rolling plan serves as a local optimization correction, providing precise target values ​​for the current time period for real-time control; the execution results of real-time control are uploaded to the intraday rolling layer through the feedback channel. If an instruction is found to be infeasible or the model is mismatched, the intraday layer is triggered to adjust its strategy; if the intraday layer needs to significantly adjust its plan due to sudden changes in external conditions, the information is uploaded to the day-ahead layer, triggering a global plan rescheduling. This hierarchical coordination mechanism ensures both long-term economic efficiency and short-term dynamic response, achieving multi-time scale collaboration from the hourly to the second-level.

[0097] This application constructs a rolling optimization and decision-making coordination mechanism across three time scales: day-ahead, intraday, and real-time. This organically unifies global economic optimization with local dynamic response, enabling the system to handle production plan matching, forecast error correction, and instantaneous disturbance suppression at different time granularities. The day-ahead scale establishes a globally optimal baseline plan on an hourly cycle, providing economic guidance for subsequent optimizations. The intraday scale performs rolling corrections on a 15-minute cycle, responding promptly to changes in electricity prices and fluctuations in actual conditions. The real-time scale adjusts rapidly on a second-level cycle, ensuring precise execution of instructions and stable system operation. These three layers form a closed loop through target transmission and information feedback, avoiding the shortcomings of single-scale optimization and overcoming the disconnect between different levels in traditional hierarchical control. Ultimately, this achieves refined, adaptive, and energy-saving operation for high-energy-consuming enterprises in complex dynamic environments.

[0098] In some embodiments disclosed in this application, the multi-timescale decision coordination in step S3 also includes a cross-scale information interaction mechanism: the real-time control scale uploads infeasible instructions or model mismatch information encountered during execution to the intraday rolling scale to trigger strategy adjustment; the intraday rolling scale uploads the plan adjustment needs caused by external conditions to the day-ahead scale to trigger global plan rescheduling.

[0099] It is understandable that infeasible instructions refer to scheduling instructions that the real-time control layer discovers during execution that cannot be actually executed due to changes in equipment status or conflicts in physical constraints. Examples include situations where sudden equipment failure prevents speed reduction or where interruption cannot continue due to temperature reaching the lower limit. Model mismatch information refers to information identified by the real-time control layer based on the deviation between actual response data and model predictions, indicating that load model parameters do not match actual characteristics. Examples include systematic deviations between the actual power-speed curve and the modeled curve of a motor, and changes in the thermal inertia time constant due to scaling. External conditions refer to significant environmental changes that prevent the intraday rolling plan from continuing execution, including emergency demand response instructions issued by the power grid, load forecast failures due to extreme weather changes, temporary major adjustments to production plans, and shutdowns due to critical equipment failures. Plan adjustment requests refer to requests from the intraday rolling layer to perform global optimization again when local corrections cannot meet the optimization objectives. Global plan rescheduling refers to, upon receiving a plan adjustment request, resolving the optimization problem for the remaining future periods at the day-ahead scale using the current state as the initial condition to generate a new baseline operating plan.

[0100] The core of cross-scale information interaction is to promptly feed back any anomalies or changes encountered during the underlying execution process to the upper-level decision-making hierarchy through standardized information channels, triggering corresponding strategy adjustments. This uplink mechanism breaks the limitation of unidirectional information flow in traditional hierarchical control, forming a bottom-up adaptive closed loop.

[0101] The real-time control layer continuously monitors the actual response status of each load during the execution of scheduling instructions. When either of the following two conditions occurs, the real-time control layer generates feedback information and uploads it to the intraday rolling scale: The first scenario is infeasibility command detection. When the real-time controller attempts to execute a command, if it finds that the target value exceeds the current allowable range of the device, the device does not respond or the response times out, or executing the command will trigger a safety protection mechanism, then it determines that the command is infeasible. For example, a motor may receive a speed reduction command but be unable to reduce its speed due to an inverter malfunction; or a heat pump may be interrupted and the room temperature may have approached the lower limit, making further interruption impossible. The real-time controller immediately stops execution, records the fault type and current status, and encapsulates the infeasibility command information into a feedback data packet for uploading.

[0102] The second scenario is model mismatch detection. The real-time controller continuously compares the actual response data with the model's predicted values, using a sliding window method to calculate the deviation statistics. If the deviation consistently exceeds a set threshold, a model mismatch is determined. For example, the actual power of a motor might be more than 5% higher than the model's predicted value, or the temperature response time constant of a kiln might become significantly longer. The real-time controller packages and uploads the deviation sequence and the identified model parameter change trends for the upper layer to correct the model.

[0103] After receiving feedback from the real-time layer, the intraday rolling scale first performs a priority assessment. For infeasible instructions, the intraday layer immediately pauses the current optimization cycle, uses the current actual state as the initial condition, re-solves the optimization problem for the remaining time period, and adjusts subsequent instructions. For model mismatch information, the intraday layer initiates an online model parameter identification program, uses the uploaded deviation data to correct the load model parameters, and uses the corrected model for the optimization calculation in the next cycle.

[0104] If the intraday rolling scale encounters significant changes that it cannot handle during execution, it will upload a plan adjustment request to the day-ahead scale. Triggering conditions include: the power grid issuing an emergency demand response command requiring immediate load reduction exceeding the intraday adjustment range; sudden weather forecast changes causing load forecast deviations exceeding a threshold for the next few hours; temporary adjustments to production plans by the production scheduling system; and critical equipment failure causing some loads to permanently shut down. The intraday layer will upload the current system status, remaining time period information, and changed external conditions, requesting the day-ahead scale to perform a global plan rescheduling.

[0105] Upon receiving the request for plan adjustment, the system immediately initiated a global rescheduling procedure. The rescheduling starts from the current time, uses the remaining time as the optimization horizon, and the current actual state as the initial condition to resolve the multi-objective collaborative optimization function, generating a new baseline operating plan. The new plan covers the entire remaining time period from the current time to the end of the original plan and is distributed to the intraday rolling layer as a new baseline for subsequent optimization. This rescheduling mechanism enables the system to cope with significant changes and avoids long-term suboptimal operation caused by local adjustments failing to achieve global optimality.

[0106] The downlink information flow also has a clear coordination logic. After the day-ahead scale generates a new baseline plan, it sends information such as the plan curve, key target values, and constraint boundaries to the intraday rolling scale. In subsequent rolling optimizations, the intraday rolling scale uses this information as optimization targets and reference benchmarks, but allows for local adjustments within a set range.

[0107] After generating the correction plan for each optimization cycle, the intraday rolling scale sends information such as the precise target value, allowable adjustment range, and predicted trajectory for the current period to the real-time control scale. The real-time control scale uses these target values ​​as tracking benchmarks and generates specific control commands through model predictive control algorithms.

[0108] The three layers exchange information through standardized data interfaces and protocols, ensuring accurate transmission and timely response. All uplink information and downlink commands are timestamped and priority-labeled to facilitate appropriate scheduling by the receiver.

[0109] This application's cross-scale information interaction mechanism establishes a bottom-up feedback channel and a top-down adjustment mechanism, enabling model errors or constraint conflicts encountered in the real-time execution layer to trigger timely adjustments to the upper-level strategy, thus avoiding the accumulation of erroneous instructions. At the same time, significant external changes can also trigger the reconstruction of the global plan. The dynamic coordination capability significantly enhances the robustness and adaptability of the system, ensuring the continuous effectiveness of the optimization strategy in complex dynamic environments.

[0110] In some embodiments disclosed in this application, the generation and issuance of scheduling instructions in step S4 specifically includes: S4.1: Convert the scheduling instructions generated in step S3 into a standardized set of control commands that can be recognized by each local controller, based on the load type and communication protocol; S4.2: Distribute standardized control command sets to the corresponding edge computing units via industrial Ethernet or wireless network; S4.3: Verify the security of instructions in the edge computing unit and execute them within the preset security boundaries, while transmitting the start and end status of execution and actual effect data back in real time.

[0111] Standardized control command sets refer to the conversion of abstract scheduling instructions generated by the upper-level optimization system into a set of control commands that are recognizable by specific equipment and have a unified format, based on the equipment type, communication protocol, and control interface specifications of each load. Examples include motor speed setpoints, inverter frequency commands, and heater power levels or start / stop signals. Edge computing units are computing and communication devices deployed on the production floor, close to the controlled equipment. They possess functions such as data acquisition, local control, protocol conversion, command verification, and edge intelligence, and can maintain basic control functions even when the cloud network is interrupted. Safety verification refers to the process by which the edge computing unit verifies the legality of commands before execution, based on preset equipment protection rules and process safety boundaries, to prevent equipment damage or safety accidents caused by command errors or communication interference. Preset safety boundaries refer to the pre-set operating limits and safety protection parameters for each type of load, including the maximum and minimum speeds of motors, the maximum allowable temperature of heating equipment, and the maximum power change rate of the process load, serving as the basis for command verification. Execution status feedback refers to the process by which the edge computing unit feeds back information such as the start time of instruction execution, the end time of execution, the actual response data during the execution process, and the success or failure of the execution result to the upper-layer optimization system in real time.

[0112] The scheduling instructions generated by the optimization decision layer are usually abstract expressions oriented towards optimization objectives. For example, the target power of a certain motor is 11 kilowatts in a certain period, the target state of a certain heat pump is interruption in a certain period, and the target power of a certain kiln is 700 kilowatts in a certain period. These abstract instructions cannot be directly sent to the equipment controller because different manufacturers and different models of equipment have different communication protocols, data formats, and control interfaces.

[0113] The instruction conversion module first queries the equipment asset database for the equipment type, communication protocol, controller model, and register address mapping table of each controlled device. For motor-type loads, the target power needs to be converted into the target speed, calculated based on the inverse function of the motor nonlinear mapping model established in step S2.2. During the solution process, it must be ensured that the speed is within the safe speed regulation range. If the target power exceeds the adjustable range, the boundary value is taken and an over-limit alarm is recorded.

[0114] For heating loads, the target status may be interrupted or at a specific power level. If it is an interruption command, it is converted into a power-off command for the corresponding controller; if it is a power level adjustment, the target power is mapped to the closest available level according to the equipment's power level table.

[0115] For process loads, the target power is directly issued as the setpoint, but a rate-of-change limit must be added, converting it into a ramp command sequence. For example, if the kiln power is reduced from 800 kW to 700 kW, the target value is issued if the local controller supports direct setting of the target value; otherwise, a continuous command sequence of reducing the power by 5 kW every 15 seconds needs to be generated until the target is reached.

[0116] The instruction conversion module is also responsible for protocol adaptation. For devices using the Modbus RTU protocol, it encapsulates instructions into data frames conforming to the protocol, including device address, function code, register address, data value, and checksum. For devices using the OPCUA protocol, it calls the corresponding service interface to write the variable node. For devices using the Profinet protocol, it directly writes the instructions to the process data area. The converted standardized control command set includes the target device identifier, command type, command value, execution timestamp, and verification information.

[0117] The converted standardized control command set is distributed to edge computing units deployed in the field via the industrial network. The distribution process must ensure the real-time performance, reliability, and security of the commands. After distribution, the system records the distribution time, command content, target device, and other information, storing it in the operation log for subsequent auditing and traceability.

[0118] After receiving the instruction, the edge computing unit first performs local security verification. The verification includes: whether the target value of the instruction is within the preset safety boundary, such as whether the motor speed exceeds nmax, whether the heat pump interruption duration will cause the room temperature to be lower than Θmin, and whether the kiln power change rate exceeds ΔPmax; whether the instruction conflicts with the current state of the equipment, such as receiving a start instruction for a running device; and whether the instruction contradicts the local interlocking logic, such as prohibiting the start of the device when the safety door is open.

[0119] If the verification is successful, the edge computing unit converts the command into an electrical signal or communication message that the local controller can recognize, and sends it to the actuator via a fieldbus such as RS-485, CANopen, or a direct digital output interface. During execution, the edge computing unit continuously monitors the actual status of the equipment, including power, speed, temperature, and position, and compares it with the target value of the command.

[0120] Upon completion, the edge computing unit generates an execution status report containing the following information: device identifier, command reception time, command start time, command completion time, execution result (success or failure), actual response trajectory data such as power change curve, and failure reason (if applicable). This report is transmitted back to the upper-layer optimization system in real time for closed-loop feedback and model correction.

[0121] The edge computing unit also possesses autonomous operation capabilities during network outages. When the network connection to the cloud is interrupted, the edge unit can maintain basic device operation and control based on the latest cached instructions and locally preset emergency strategies, and synchronously execute data once the network is restored. This edge intelligence design ensures the reliability and security of the system in the event of communication failures.

[0122] This application ensures that optimization decisions can be safely, accurately, and reliably implemented into equipment actions through standardized conversion, reliable transmission, and edge-side security verification.

[0123] In some embodiments disclosed in this application, the coordinated control in step S4 also includes a strategy to ensure production continuity: when executing load adjustment instructions, priority is given to adjusting the loads of auxiliary equipment or redundant links that have the least impact on the final product quality; for the adjustment of critical process loads, a slow ramp change method is adopted; a graded response strategy is set up so that when an emergency demand response signal from the power grid is received, non-production-essential heating and cooling loads are interrupted first, and then auxiliary production loads are adjusted.

[0124] It is understandable that priority adjustment refers to prioritizing the adjustment of loads with the least impact on core production processes and final product quality when multiple adjustable loads have adjustment capabilities simultaneously, according to a pre-set priority order, to maximize production stability and product quality. Critical process loads refer to key equipment that directly participates in core production processes, and whose operating status directly affects product output and quality, such as ceramic tunnel kilns, reactors, and heat treatment furnaces. Abnormal fluctuations in these loads may lead to scrap or equipment damage. Slow ramp change means that the adjustment command for critical process loads does not use abrupt changes, but rather adjusts the power or setpoint gradually according to a set rate of change, allowing process parameters to transition smoothly and avoiding thermal or mechanical shocks. Tiered response strategy refers to classifying various loads within the enterprise into several levels based on their importance and response speed. When receiving an emergency demand response signal from the power grid, load adjustment capabilities are activated tier by tier in order of least importance to ensure that power grid requirements are met while minimizing the impact on production. Non-production-essential loads refer to loads that are not directly related to the core production process and whose short-term interruptions do not affect the output of major products, such as office area heating, living area cooling, and non-production lighting. Auxiliary production loads refer to equipment that serves the production process but does not directly change the properties of materials, such as cooling water pumps, fans, material conveying systems, and dust removal equipment. Their short-term power reduction or interruption may affect production efficiency but will not directly lead to scrap.

[0125] During the optimization of scheduling instruction generation and execution, the system incorporates a load importance classification database, assigning a priority coefficient to each type of load and even each piece of equipment. The priority coefficient is determined comprehensively based on the load's impact on final product quality, its importance to production continuity, and the recovery cost after adjustment. For example, tunnel kilns, which directly affect ceramic sintering quality, have the highest priority, followed by auxiliary cooling water pumps, and office heating systems have the lowest priority.

[0126] When generating scheduling instructions in step S3, the optimization solver prioritizes loads with lower priority coefficients for adjustment, while satisfying power balance and demand constraints. This mechanism is achieved by introducing a penalty term related to the priority coefficient into the objective function.

[0127] During the instruction execution phase, the edge computing unit also stores a local priority table. When multiple adjustment instructions are received simultaneously, if some instructions cannot be executed due to changes in device status, the edge unit can select to execute the higher-priority instructions based on its local priority and report the failure information of the lower-priority instructions.

[0128] For critical process loads such as tunnel kilns and reactors, power regulation must adhere to strict rate-of-change constraints to prevent thermal stress from damaging equipment or temperature fluctuations from affecting product quality. This constraint has been introduced into the load model in step S2.2 as an optimization problem in the form of a power change rate constraint |Pt-Pt-1|≤ΔPmax, ensuring that the optimization result itself meets the rate requirement.

[0129] At the instruction execution level, the edge computing unit further converts the target power value into a ramp instruction sequence. Let the current power be Pcur, the target power be Pset, and the maximum allowable rate of change be ΔPmax, in kilowatts per minute or kilowatts per time period. If... Therefore, the adjustment process needs to be broken down into multiple steps. The edge unit calculates the target value for each step using a fixed control period, such as 15 seconds.

[0130] Here, Δtstep represents the control step size, measured in hours or minutes. A sub-instruction is issued for each step until the final target value is reached. Simultaneously, the edge unit continuously monitors actual power and temperature; if the actual response lags behind the instruction, it automatically adjusts subsequent step sizes to ensure a smooth transition of process parameters.

[0131] The tiered response strategy is a pre-defined control logic for responding to emergency demand events in the power grid. The system pre-classifies all loads into three levels: Level 1 is non-production-essential loads, Level 2 is auxiliary production loads, and Level 3 is core process loads. Each level can be further subdivided into sub-levels.

[0132] When the enterprise energy management system receives an emergency demand response signal issued by the power grid, such as a peak shaving instruction or an emergency power restriction notice, the system immediately triggers a hierarchical response procedure. The procedure first determines the amount of power to be cut, Pcut. Then, in the order from level one to level three, it calculates the total capacity that can be cut at each level step by step. If the cut capacity of level one load, Plevel1, is greater than or equal to Pcut, only specific equipment is selected from the level one load to execute the interruption, and the selection principle is also based on the priority coefficient. If Plevel1 < Pcut, all level one loads are interrupted, and the insufficient part is adjusted from the level two load in the priority order, and so on. The hierarchical response logic can be preset as a fixed strategy or calculated in real time by an optimizer. In the real-time calculation mode, the system converts the emergency demand response signal into new constraint conditions, such as the total power must be lower than a certain threshold during a specific future period, and then resolves the optimization problem again to automatically generate a scheduling plan that conforms to the hierarchical principle. In either way, the core goal is to minimize the impact on core production while meeting the requirements of the power grid.

[0133] In this application, by introducing production continuity guarantee strategies such as priority adjustment, slow ramp change, and hierarchical response, the energy-saving control is deeply integrated with production safety, enabling the system to always prioritize the stability of the core process and product quality while pursuing economic optimization. The priority adjustment mechanism ensures that auxiliary equipment and redundant links first undertake the adjustment tasks, protecting the key processes to the greatest extent; the slow ramp change effectively suppresses the risk of thermal shock to thermally inertial equipment caused by power mutations, extends the equipment life, and ensures product quality; the hierarchical response strategy provides an orderly and controllable plan for responding to the emergency demands of the power grid, which not only meets the requirements of the power grid but also avoids production chaos caused by disorderly adjustment.

[0134] In some embodiments disclosed in this application, the implementation of closed-loop optimization through model predictive control in step S5 specifically includes: S5.1: Continuously monitor and compare the deviation between the actual response of each load and the expected value of the scheduling instruction generated in step S3. S5.2: Input the deviation into the model predictive control algorithm to predict the system behavior in the next few cycles and calculate the feedforward-feedback composite control quantity that can make the system return to the optimal trajectory fastest. S5.3: Use the feedforward-feedback composite control quantity to automatically identify and update the load model parameters established in step S2 regularly.

[0135] It is understood that in this embodiment, deviation refers to the difference between the actual response value of each load and the expected value of the scheduling instruction generated in step S3, including various forms such as power deviation, speed deviation, temperature deviation, and completion time deviation, and is a core indicator for measuring control accuracy. Model predictive control algorithm is an advanced process control method based on a system dynamic model. Its core idea is to solve a finite-time domain optimization problem in each control cycle based on the current system state and the predictive model, obtain the optimal control sequence, and execute only the first control action. Feedforward control quantity refers to the compensation control quantity calculated in advance based on known disturbance information or target trajectory changes, used to improve the system's response speed and tracking accuracy. Feedback control quantity refers to the correction control quantity calculated by the feedback controller based on the deviation between the actual output and the expected value, used to suppress unmodeled disturbances and model errors. Feedforward-feedback composite control quantity is a comprehensive control instruction that combines feedforward control quantity and feedback control quantity, possessing both the speed of feedforward and the robustness of feedback. Automatic parameter identification and updating refers to the process of using actual operating data to estimate the parameters of the load model online through system identification algorithms, and replacing the old parameters in the original model with new parameters so that the model always remains consistent with the actual characteristics of the equipment.

[0136] During real-time control, the system continuously collects actual operating data for each load, including power, speed, temperature, and pressure, and compares it point-by-point with the expected values ​​of the scheduling instructions issued in step S3. Deviation monitoring is performed in parallel at multiple time scales: at the real-time control scale, the difference between instantaneous values ​​and command values ​​is compared at second or millisecond intervals; at the intraday rolling scale, the difference between the average power of the period and the planned value is compared at minute intervals; at the day-ahead scale, the difference between the cumulative power and the predicted value is compared at hourly intervals.

[0137] The method for calculating deviation varies depending on the load type and control objective. For motor power tracking, the deviation is defined as follows: The unit is kilowatts. For temperature tracking, the deviation is defined as... The unit is degrees Celsius. For production task completion time, the deviation is defined as... The unit is hours.

[0138] The system establishes a deviation database, storing all deviation records by device and time, and calculates statistical characteristics such as mean, variance, maximum deviation, and deviation duration. When the deviation continuously exceeds a set threshold, an alarm is triggered and recorded as an abnormal event. The deviation sequence serves as input data for model predictive control and parameter identification, providing a foundation for subsequent steps.

[0139] The model predictive control algorithm performs the following steps in each control cycle: The first step is state estimation and prediction. Based on the current actual state xt and the load dynamic model established in step S2, the system behavior over the next N control cycles is predicted. The prediction model uses the difference equations established in steps S2.2 and S2.3. Taking a motor as an example, its prediction model is as follows:

[0140] This represents the power predicted at time t+k at time t, where ut+k is the control variable to be solved, i.e., the speed command, and Tm is the time constant.

[0141] The second step is optimization. A finite-time optimization problem is constructed, with the objective function typically including terms such as minimizing tracking error and minimizing control variable changes. This is used to track the target power issued by the intraday rolling layer. For example, the objective function is:

[0142] in and The weighted norm is used, and Q and R are weight matrices used to balance tracking accuracy and the severity of control actions. The optimization problem also needs to satisfy equipment constraints such as upper and lower limits of rotational speed and limits on the rate of change.

[0143] The third step is to calculate the feedforward-feedback composite control quantity. After solving the above optimization problem, the optimal control sequence is obtained. The first control variable This is a feedforward-feedback composite control quantity. It includes feedforward information based on the future target trajectory and feedback information based on the deviation from the current state, since the current state xt has already been introduced into the optimization problem through the initial conditions. This control quantity is issued to the local controller for execution.

[0144] The fourth step is rolling updates. In the next control cycle, based on the new actual state xt+1, the above steps are repeated to achieve rolling optimization.

[0145] As equipment operates over a long period, its physical characteristics may change. For example, bearing wear in electric motors can lead to decreased efficiency, scale buildup on heating equipment surfaces can increase thermal resistance, and aging of kiln insulation materials can alter thermal inertia. These changes cause deviations between the initially established load model and the actual equipment characteristics. If the old model is continued to be used for optimization and control, it will result in inaccurate predictions and reduced control effectiveness.

[0146] To address this issue, the system established an automatic model parameter identification and update mechanism. The core idea of ​​this mechanism is to utilize the massive amounts of real-world data generated during the daily operation of the equipment, analyze the correspondence between inputs and outputs, deduce the current actual model parameters, and replace the old parameters with the new ones.

[0147] The basic principle of parameter identification is to compare the difference between the actual response and the model prediction. Taking an electric motor as an example, the system continuously records the actual power data at different speeds. After collecting enough data points, the system uses statistical analysis to find the power-speed relationship curve that best fits the current data points, thus obtaining new model coefficients. This process is similar to redrawing a curve that best fits a set of scattered data points, except that it is done automatically by the system. For thermodynamic models, the system focuses on the temperature change pattern of the equipment during power-on and power-off processes. By analyzing the rate of temperature rise and fall in multiple power-on and power-off cycles, the actual heat capacity and thermal resistance values ​​can be calculated. For example, if the temperature drops more slowly after power is cut off, it indicates improved insulation performance and increased thermal resistance; if the temperature rises more slowly, it indicates decreased heating efficiency or increased heat capacity.

[0148] Parameter updates are not performed constantly, but rather follow a reasonable triggering strategy. One approach is periodic updates, such as automatically performing parameter identification weekly or monthly to ensure the model remains up-to-date. Another approach is event-triggered updates, where parameter identification is immediately initiated when the system detects persistent control deviations or model mismatch alarms, quickly correcting model errors. The updated model parameters take effect immediately and are used for subsequent optimization calculations and real-time control, ensuring the system always maintains an accurate description of the equipment's true characteristics.

[0149] This application utilizes an automatic model parameter identification and update mechanism to enable the load model to adaptively adjust to equipment aging and environmental changes, thereby addressing the pain point of long-term performance degradation in traditional fixed-model control schemes. Parameter identification is based on daily equipment operating data, requiring no additional testing and not affecting normal production. The strategy combining periodic updates with event triggering ensures both the model's timeliness and avoids resource consumption from frequent calculations. The updated model ensures that optimized control is always based on the actual characteristics of the equipment, guaranteeing optimization effectiveness throughout the entire lifecycle from the initial commissioning stage to the end of the equipment's lifespan.

[0150] In some embodiments disclosed in this application, step S5 further includes a performance evaluation and self-learning mechanism: periodically calculating and visualizing the total electricity cost, peak demand, and load curve smoothness index before and after implementing collaborative management and control, and evaluating the energy-saving effect; based on historical optimization data and energy-saving effect, training an auxiliary decision-making model to learn the optimal strategy mode under different combinations of external conditions.

[0151] In this embodiment, performance evaluation refers to the process of quantifying the actual effect of energy-saving control methods by comparing changes in key indicators before and after the implementation of collaborative management and control. This typically involves calculations and analyses across multiple dimensions, such as total electricity cost, peak demand, and load curve smoothness. Visualization refers to presenting the performance evaluation results intuitively on the user interface in the form of charts, dashboards, and trend curves, enabling managers to quickly understand the energy-saving effectiveness and system operating status. The load curve smoothness index is a statistic used to quantify the degree of fluctuation in a company's total electricity load. It is usually characterized by calculating the variance, peak-to-valley difference, or power change rate per unit time of the load curve. Higher smoothness indicates smaller load fluctuations and more stable operation. The auxiliary decision-making model refers to an empirical model trained using machine learning algorithms based on historical optimization data and actual energy-saving effects. It can quickly recommend the optimal strategy mode when similar combinations of external conditions occur, without having to perform complex online optimization calculations each time. The optimal strategy mode refers to the load dispatching scheme template with the lowest comprehensive electricity cost under specific combinations of external conditions, such as specific electricity price curves, specific production plans, and specific weather conditions, verified through historical testing or optimization calculations.

[0152] The core of the performance evaluation mechanism is to compare the operational effects before and after implementing this method, using data to demonstrate the actual value of energy-saving control. The system automatically extracts data from the historical database for two time periods at fixed time intervals, such as daily, monthly, or quarterly: one is the control period, typically the same period before the implementation of this method, such as the same period last year or last month; the other is the evaluation period, i.e., the period during which this method is currently implemented. The system performs comparative analysis from three key dimensions.

[0153] The first dimension is total electricity cost. The system calculates the actual electricity expenses incurred by the enterprise during the comparison period and the evaluation period, including electricity charges calculated based on time-of-use pricing and demand charges calculated based on maximum demand. By directly comparing the total electricity costs in the two periods, the amount of money saved and the percentage of savings can be clearly seen.

[0154] The second dimension is peak demand. The system identifies the maximum power values ​​from the electricity consumption data of the control and evaluation periods. This value directly relates to the electricity cost based on demand and is also an important indicator of whether a company's electricity consumption is stable. Comparing the peak demand in the two periods allows for the quantification of the method's effectiveness in peak shaving.

[0155] The third dimension is the smoothness of the load curve. The system observes the curve of the enterprise's total daily power consumption over time and compares the curve shapes of the control period and the evaluation period. If the curve peaks are lower, the troughs are higher, and the overall fluctuations are smoother during the evaluation period, it indicates that the method effectively smooths the load and reduces the impact of drastic fluctuations on the power grid and equipment.

[0156] After data comparison, the visualization module presents the analysis results to enterprise managers in an intuitive graphical format. The system automatically generates various charts, such as line graphs showing monthly cost savings trends, bar charts comparing peak demand changes, and comparison charts that overlay typical daily load curves from the control and evaluation periods. Managers can view these charts anytime via computer screens or mobile devices to clearly understand the effectiveness of energy-saving efforts and provide a basis for decision-making for continuous improvement.

[0157] As the system has been running for a long time, a large number of historical optimization records have accumulated in the database. Each day's operation corresponds to a specific set of external conditions, such as the day's time-of-use electricity price curve, the output and equipment utilization rate of the production plan, and the temperature data from the weather forecast. It also records the load scheduling scheme generated by the optimizer that day, as well as the final electricity cost and energy-saving effect achieved.

[0158] The core idea of ​​decision support models is to learn patterns from historical experience. When a certain combination of external conditions occurs repeatedly, the system can analyze historical data to identify the scheduling scheme that has achieved the best results under those conditions. This process is similar to the accumulation of human experience: careful consideration is needed when encountering a certain situation for the first time, but a quick decision can be made after encountering similar situations multiple times.

[0159] Training the decision support model does not require complex mathematical derivations; instead, it is based on the fundamental principles of data statistics and pattern recognition. The system first characterizes daily external conditions, such as classifying electricity price curves as having a high or low peak-to-valley ratio, production plans as being busy or relaxed, and weather as being cold or warm. Then, the system performs correlation analysis between these features and the day's optimization plan and its actual effects. After accumulating a sufficient number of samples, the system can establish a correspondence between external conditions and the optimal strategy.

[0160] When new external conditions arise, the system first matches their characteristics with historical data to identify the most similar historical days. Then, it analyzes the best-performing optimization schemes from these historical days, extracting common strategy patterns, such as the typical time period for heat storage under similar electricity price structures and the acceptable interruption time for heat pumps under similar temperatures. Based on these empirical patterns, the system quickly generates a preliminary scheduling scheme suggestion.

[0161] This suggestion can serve as the initial solution for subsequent formal optimization calculations, significantly reducing the time required for optimization. In special circumstances, such as when communication interruptions prevent online optimization, this suggested solution can also be directly issued as a scheduling command for execution, ensuring that the system still maintains basic energy-saving effects in emergency situations. As the system's uptime increases and the empirical samples become richer, the recommendations of the auxiliary decision-making model will become more accurate, forming a virtuous cycle of continuous improvement.

[0162] Taking a company that manufactures high-temperature structural ceramics as an example, the system has been running continuously for a year and has accumulated a wealth of historical data.

[0163] At the beginning of each month, the system automatically generates an energy-saving performance assessment report for the previous month. Taking January 2026 as an example, the system selects January 2025 of the same period last year as a comparison period. Regarding total electricity costs, the system calculation shows that the company's actual electricity bill in January 2025 was 1.865 million yuan, while the electricity bill in January 2026 was 1.692 million yuan. Subtracting the two, the system shows a saving of 173,000 yuan this month, a saving rate of approximately 9.3%. This figure directly reflects the economic benefits of energy-saving control. Regarding peak demand, the system identifies the maximum power output in January 2025 as 5120 kW, while the maximum power output in January 2026 is 4850 kW. Peak demand decreased by 270 kW, a reduction rate of approximately 5.3%. This means that the company's electricity bill for the month will decrease accordingly, while also reducing the pressure on transformer capacity. Regarding load curve smoothness, the system extracts load curves from two typical working days in January 2025 and January 2026 for comparison. The curve for 2025 showed distinct morning and evening peaks with a large peak-to-valley difference; while the curve for 2026 was generally flatter, with peak periods effectively reduced and trough periods filled, indicating more stable load operation.

[0164] The visualization module presents these comparative results in chart form. When managers open the system interface, they see a clean dashboard: on the left is a cost-saving trend chart, using a line graph to show the monthly savings increasing over the past 12 months; in the middle is a demand change bar chart, with the peak demand bar height decreasing month by month; on the right is a load curve comparison chart, with the two curves drawn in different colors, clearly showing that the optimized curve is smoother. These intuitive charts allow management to see the effectiveness of energy-saving efforts at a glance and provide direction for continuous improvement.

[0165] After a year of operation, the system database stores 365 days of historical records. Each day's record includes external conditions such as electricity price curves, production plans, weather temperature, the day's optimized scheduling scheme, and the final energy-saving effect.

[0166] As data accumulates, the system begins to develop the ability to learn from experience. For example, the system discovered that in the past few months, whenever similar time-of-use electricity pricing structures resulted in particularly high peak daytime electricity prices and low temperatures, the historically most effective solution was to run the heat pump at full capacity during the early morning off-peak hours to store heat, and then completely shut it down during the daytime peak hours, with the heat storage tank providing heating. The system also found that when there were many high-value precision machining tasks in the production plan, the historically optimal solution was to minimize the use of these precision machine tools and instead prioritize adjusting ordinary machine tools and auxiliary equipment.

[0167] Based on these historical experiences, the system has built an experience knowledge base. When new external conditions arise, the system can quickly match similar historical scenarios. For example, on a certain day in February 2026, the weather forecast showed an average temperature of 3°C, the production plan included 120 tons of output and multiple high-value tasks, and the electricity price curve was highly similar to that of a certain day in December of the previous year. Through matching, the system discovered that on that day in December of the previous year, under similar conditions, a scheme of storing heat during off-peak hours, discharging during peak hours, and interrupting the heat pump for 1.5 hours achieved excellent energy-saving results. Therefore, the system provides this scheme as a recommendation to the optimizer, which then performs rapid calculations based on it, ultimately reducing the optimization solution time from 15 minutes to 3 minutes.

[0168] In another emergency due to a communication line failure, the system was unable to connect to the cloud optimizer for online computation. At this point, the auxiliary decision-making model came into play, providing an empirically recommended solution based on the day's external conditions. The edge computing units then executed the scheduling according to this solution. Although this solution may not be the optimal one, its historical experience ensured that the day's electricity costs did not deviate significantly from the optimal value, avoiding losses from completely uncontrolled operation.

[0169] This application demonstrates energy-saving results intuitively through quantitative indicators, providing data support for management decisions. More importantly, by training the auxiliary decision-making model with historical data, it can continuously accumulate and solidify optimal operating experience, enabling the system to respond quickly when dealing with repetitive or similar scenarios, improving optimization efficiency, and gradually reducing reliance on complex online calculations, thus promoting the development of the system towards intelligence and knowledge.

[0170] See Figure 2 A second aspect of this application provides an energy-saving control system for high-energy-consuming enterprises based on multi-load coordination, used to implement an energy-saving control method for high-energy-consuming enterprises based on multi-load coordination, including: Data acquisition module 21 is used to execute step S1; Optimization decision module 22 is used to execute step S2; The scheduling instruction module 23 is used to execute step S3; Control execution module 24 is used to execute step S4; The closed-loop feedback module 25 is used to execute step S5.

[0171] Please see Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to execute the aforementioned energy-saving control method for high-energy-consuming enterprises based on multi-load coordination.

[0172] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0173] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0174] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0175] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the energy-saving control method for high-energy-consuming enterprises based on multiple load coordination provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0176] In another embodiment of this application, an electronic device is provided. The electronic device stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the energy-saving control method for high-energy-consuming enterprises based on multiple load coordination described in the above embodiments. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in an electronic device, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0177] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0178] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0179] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0180] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0181] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0182] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0183] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for energy-saving control of high-energy-consuming enterprises based on multi-load coordination, characterized in that, Includes the following steps: S1: Real-time acquisition of various load data from high-energy-consuming enterprises, including adjustable motor loads, interruptible heating loads, and process loads with thermal inertia; S2: Based on the aforementioned load data, construct a multi-objective collaborative optimization function with the minimum comprehensive electricity cost as the core, taking into account time-of-use electricity fees, demand electricity fees, and penalty costs for deviation from production constraints, and establish load models for the power adjustable range and dynamic response characteristics of various types of loads; S3: Based on the optimization function and load model, rolling optimization and decision coordination are performed on three time scales: day-ahead, intraday, and real-time, to generate scheduling instructions for loads with different response speeds. S4: Send the scheduling command to the local controller of each load to perform coordinated control of motor speed, heater power level and process equipment operation; S5: Monitor load response status and total energy consumption in real time, and dynamically feed back and correct the load model and optimization function through model predictive control algorithm to achieve closed-loop optimization.

2. The method according to claim 1, characterized in that, The real-time acquisition of various load data in step S1 specifically includes: S1.1: Through intelligent sensing units deployed in various load circuits or production equipment, the real-time power, speed and torque of the motor, the power and temperature of the heating equipment, and the process parameters and energy consumption data of the process equipment are collected synchronously. S1.2: Real-time acquisition of time-of-use electricity price signals, total incoming power, and future preset production plans from the production scheduling system through the energy management system or data interface of high-energy-consuming enterprises; S1.3: The collected and acquired data are preprocessed by timestamp alignment, invalid data cleaning, and normalization to form a unified time-series integrated data queue.

3. The method according to claim 1, characterized in that, The construction of the multi-objective collaborative optimization function and the establishment of the load model in step S2 specifically include: S2.1: The multi-objective collaborative optimization function is expressed as: Total cost = Time-of-use electricity cost + Maximum demand cost + Deviation penalty cost, wherein the deviation penalty cost is dynamically calculated based on the deviation between the actual load operation curve and the production plan requirement curve; S2.2: For adjustable motor loads, establish a nonlinear mapping model between motor power, speed, and load rate, and determine the safe speed regulation range and response delay; for interruptible heating loads, establish a thermodynamic dynamic model and determine the allowable interruption duration and safe temperature fluctuation range; for process loads with thermal inertia, establish an inertial hysteresis model between input energy and key state parameters, where the key state parameters are temperature and pressure. S2.3: Integrate various load models into the multi-objective collaborative optimization function in the form of constraints, so that the optimization solution must be carried out under the premise of satisfying the physical laws and operating limits of various loads.

4. The method according to claim 3, characterized in that, Step S2 further includes processing the coupling with the external excitation signal: S2.4: In the multi-objective collaborative optimization function, a weighting factor linked to the real-time time-of-use electricity price is set, so that during periods of high electricity price, the optimization algorithm tends to reduce total power or transfer movable loads; during periods of low electricity price, it tends to start high-energy-consuming processes in advance or carry out heat and cold storage operations. S2.5: The predicted maximum demand threshold is introduced as a hard constraint into the multi-objective collaborative optimization function.

5. The method according to claim 1, characterized in that, The rolling optimization and decision coordination at three time scales in step S3 specifically includes: At the day-ahead scale, based on the production plan and time-of-use electricity price forecast for the next 24 hours, the multi-objective collaborative optimization function is solved at 1-hour intervals to formulate the baseline operation plan for various types of loads and the charging and discharging plan for loads with energy storage characteristics. At the intraday rolling scale, with a period of 15 minutes or 30 minutes, the day-ahead baseline operation plan and charging and discharging plan are rolled over and revised based on load forecasts, real-time electricity prices, and actual operating status. At the real-time control scale, with a period of seconds or minutes, adjustment commands are generated and directly issued to the local controller based on system deviations and short-term forecasts to achieve second-level coordination of loads with different response speeds.

6. The method according to claim 5, characterized in that, The multi-timescale decision coordination in step S3 also includes a cross-scale information interaction mechanism: the real-time control scale will upload infeasible instructions or model mismatch information encountered during the execution process to the intraday rolling scale to trigger strategy adjustment; The intraday rolling scale will upload the plan adjustment needs caused by external conditions to the day-ahead scale, triggering it to perform a global plan rescheduling.

7. The method according to claim 1, characterized in that, The specific steps in step S4, including generating and issuing scheduling instructions, include: S4.1: Convert the scheduling instructions generated in step S3 into a standardized set of control commands that can be recognized by each local controller, according to the load type and communication protocol; S4.2: The standardized control command set is sent to the corresponding edge computing unit via industrial Ethernet or wireless network; S4.3: Verify the security of instructions in the edge computing unit and execute them within the preset security boundaries, while transmitting the start and end status of execution and actual effect data back in real time.

8. The method according to claim 7, characterized in that, The coordinated control in step S4 also includes strategies to ensure production continuity: when executing load adjustment commands, priority is given to adjusting the loads of auxiliary equipment or redundant links that have the least impact on the final product quality; for the adjustment of critical process loads, a slow ramp change method is adopted; a graded response strategy is set up so that when an emergency demand response signal from the power grid is received, non-production-essential heating and cooling loads are interrupted first, and then auxiliary production loads are adjusted.

9. The method according to claim 1, characterized in that, The closed-loop optimization achieved through model predictive control in step S5 specifically includes: S5.1: Continuously monitor and compare the deviation between the actual response of each load and the expected value of the scheduling instruction generated in step S3; S5.2: Input the deviation into the model predictive control algorithm to predict the system behavior in the next few cycles, and calculate the feedforward-feedback composite control quantity that enables the system to return to the optimal trajectory as quickly as possible; S5.3: Using the feedforward-feedback composite control quantity, the load model parameters established in step S2 are automatically identified and updated periodically.

10. The method according to claim 9, characterized in that, Step S5 also includes a performance evaluation and self-learning mechanism: periodically calculate and visualize the total electricity cost, peak demand, and load curve smoothness index before and after the implementation of collaborative management and control to evaluate the energy-saving effect; based on historical optimization data and energy-saving effect, train an auxiliary decision-making model to learn the optimal strategy mode under different combinations of external conditions.

11. A high-energy-consuming enterprise energy-saving control system based on multi-load coordination, used to implement the method as described in any one of claims 1 to 10, characterized in that, include: The data acquisition module is used to perform step S1; An optimization decision-making module is used to execute step S2; The scheduling instruction module is used to execute step S3; A control execution module is used to execute step S4; A closed-loop feedback module is used to execute step S5.

Images