Dual-time-scale regulation method and system considering industrial user regulation potential

Abstract

The application provides a double-time-scale regulation method and system considering industrial user regulation potential, relates to the technical field of industrial park adjustable resource potential evaluation and optimal regulation, establishes an industrial load model for adjustable production equipment through historical industrial load data and equipment parameter information, establishes the association between regulation incentives and equipment, constructs a dynamic industrial user regulation potential time sequence data input model, performs regulation potential analysis on the data to obtain an industrial load potential adjustable range as a constraint, respectively constructs a day-ahead and intraday double-time-scale regulation model with the target of minimizing operation cost, constructs a coupled and interactive day-ahead and intraday decision intelligent agent, solves to generate a day-ahead-intraday collaborative regulation strategy, effectively guarantees the production and energy planning of park users, excavates user regulation potential, realizes efficient day-ahead-intraday collaborative regulation of the industrial park under a dynamic uncertain environment, reduces operation cost, and guarantees stable operation of the power grid.

Description

Technical Field

[0001] This invention relates to the field of adjustable resource potential assessment and optimization control technology in industrial parks, and particularly to a dual-timescale control method and system that considers the control potential of industrial users. Background Technology

[0002] Industrial park users are an important component of the power system, possessing enormous potential for load regulation. This is crucial for alleviating the supply-demand imbalance in the power grid and improving its operational efficiency. Industrial parks encompass a large number of diverse loads; therefore, extrapolating the regulation potential of these diverse load clusters and implementing precise regulation is a key step in achieving friendly grid-load interaction in the new power system.

[0003] However, existing technologies still have many shortcomings in tapping and utilizing the control potential of industrial users, mainly in the following aspects: 1) Inaccurate control potential assessment and analysis: Traditional methods often assess the control potential of industrial users based on simple load models or historical data, failing to fully consider factors such as the actual production situation, equipment characteristics, and control costs of industrial users, resulting in inaccurate control assessment and analysis results, making it difficult to effectively guide actual control operations. 2) Lack of dynamic adaptability in day-ahead control plans: When formulating day-ahead control plans, existing park energy management systems usually rely only on historical data and simple forecasting models, lacking dynamic assessment of the intraday control potential of industrial users, making it difficult for day-ahead plans to adapt to sudden situations and load changes in actual operation. 3) Existing methods mostly use static optimization models, failing to fully consider the dynamic coupling relationship between day-ahead market decisions and intraday real-time control, as well as the problem of incomplete information, resulting in limited effectiveness in highly uncertain dynamic environments. Summary of the Invention

[0004] To address the technical problems of inaccurate assessment of control potential, lack of dynamic adaptability in day-ahead control plans, and inability to cope with dynamic and uncertain environments in existing industrial park control methods, this invention provides a dual-timescale control method and system that considers the control potential of industrial users. It aims to effectively guarantee the energy planning of industrial park users, tap their control potential, and achieve efficient day-ahead and intraday coordinated control of industrial parks in dynamic and uncertain environments by comprehensively considering factors such as the actual production situation, equipment characteristics, and control costs of industrial users through a data-driven approach. This reduces operating costs and ensures stable grid operation.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a dual-timescale control method considering the controllability potential of industrial users in its first aspect, comprising: acquiring historical industrial load data and adjustable production equipment parameter information of industrial park users; establishing an industrial load model for adjustable production equipment based on the acquired data, establishing the correlation between controllability incentives and adjustable production equipment based on the industrial load model, and constructing a time-series data input model for industrial user controllability potential; inputting historical industrial load data into the time-series data input model for industrial user controllability potential, and after obtaining the time-series input data for industrial user controllability potential, performing controllability potential analysis using an industrial load adjustable potential analysis model to obtain the adjustable range of industrial load potential; based on the adjustable production equipment parameter information, under the constraint of satisfying the adjustable range of industrial load potential, constructing a day-ahead production plan model for optimizing the controllability of industrial park production plans with the goal of minimizing the total operating cost of the industrial park, and constructing an intraday control model for optimizing the controllability of flexible adjustable production equipment with the goal of minimizing the total operating cost and power fluctuation of the industrial park; constructing a day-ahead decision agent and an intraday decision agent based on the day-ahead production plan model and the intraday control model respectively, coupling and interacting them and solving them to generate a day-ahead-intraday coordinated control strategy.

[0006] The present invention provides a preferred embodiment in its first aspect, wherein the step of establishing the correlation between control incentives and adjustable production equipment based on an industrial load model specifically utilizes a sliding window method to establish the correlation between control incentives and adjustable production equipment. The sliding window includes three types of sliding input windows: off-peak, peak, and flat periods. The power consumption data of the equipment within the sliding input windows for these three periods serves as the time-series input data for the industrial user's controllability potential. This embodiment, by dividing the period into off-peak, peak, and flat periods and using a sliding window to collect power consumption data, can more precisely capture the load characteristics under different time periods, improve the temporal resolution of the time-series data, and provide a more accurate and structured input for subsequent potential analysis. It can effectively distinguish the response characteristics of user equipment under different electricity price periods, accurately characterize the distribution pattern of the adjustable potential of industrial users in each period, improve the matching degree between the control strategy and actual load changes, and enhance the model's adaptability to dynamic electricity price signals.

[0007] The present invention provides a preferred embodiment in its first aspect, wherein the industrial load adjustable potential analysis model is constructed based on a fully convolutional neural network, a temporal convolutional network, and a long short-term memory network. The step of using the neural network-based industrial load adjustable potential analysis model to perform regulation potential analysis and obtain the adjustable range of industrial load potential includes: selecting time-series input data of the industrial user regulation potential based on a fully convolutional neural network to obtain a preferred time-series dataset; inputting power generation data from the industrial park's self-owned power plant in negative form into the preferred time-series dataset for expansion to obtain a selected time-series dataset; performing fitting analysis on the selected time-series dataset based on a temporal convolutional network to output a preliminary adjustable range of industrial load potential; and correcting the preliminary adjustable range of industrial load potential based on a long short-term memory network to output the adjusted range of industrial load potential. This preferred embodiment integrates a fully convolutional neural network, a temporal convolutional network, and a long short-term memory network, first filtering and expanding the data, and then fitting and correcting it to achieve high-precision and highly robust industrial load potential assessment, improving the accuracy and stability of adjustable range prediction.

[0008] The present invention provides a preferred embodiment in its first aspect, comprising the step of constructing a day-ahead production planning model for optimizing and regulating the production plan of an industrial park, with the objective of minimizing the total operating cost of the industrial park under the constraint of the adjustable range of the industrial load potential, including: constructing a day-ahead objective function using a scenario generation method with the objective of minimizing the total operating cost of the industrial park as the objective, and setting day-ahead constraints for the day-ahead objective function with the adjustable range of the industrial load potential as one of the constraints, thereby obtaining a day-ahead production planning model; the day-ahead objective function consists of the scenario occurrence probability, the regulation interval of the first time scale, the electricity purchase and sale cost, the usage and maintenance cost of adjustable production equipment, and the regulation incentive cost of adjustable production equipment. This preferred embodiment quantifies the impact of uncertainties on operating costs by introducing scenario occurrence probability and regulation incentive cost variables, thereby improving the model's decision robustness under multiple scenarios. Simultaneously, by comprehensively considering fluctuations in electricity purchase and sale prices and equipment operating status, the production load allocation for each time period within the regulation interval is dynamically adjusted to ensure optimal economic efficiency while meeting the production needs of industrial users. Taking into account uncertainties such as photovoltaics, a day-ahead model is constructed with the goal of minimizing total operating costs. Combined with constraints such as the adjustable range of industrial load potential, the model enables the formulation of economically optimal production plans under uncertain environments, thereby enhancing the adaptability and economy of day-ahead scheduling.

[0009] Furthermore, the day-ahead constraints also include: day-ahead power balance constraints, equipment operating status and power correlation constraints, production task constraints, minimum continuous operating time constraints for equipment, upper limit constraints on the number of equipment start-ups and shutdowns, constraints on the number of equipment starting and stopping simultaneously, and production process coupling constraints. This preferred solution, through multiple constraints such as power balance, equipment operating status, production task, number of start-ups and shutdowns, and process coupling, ensures a reasonable distribution of production plans in time and space, effectively reducing losses caused by frequent equipment start-ups and shutdowns, and lowering operation and maintenance costs. Simultaneously, it balances the rigid requirements of the production process with the flexibility of load adjustment, improving the overall energy efficiency of the system and achieving economic operation goals while ensuring production continuity.

[0010] The present invention provides a preferred embodiment in its first aspect, comprising the step of constructing an intraday control model for optimizing and regulating flexible production equipment with the objective of minimizing the total operating cost and power fluctuation of an industrial park. This step includes: constructing an intraday objective function using a scenario generation method, with the objectives of minimizing the total operating cost and power fluctuation of the industrial park; setting intraday constraints for the intraday objective function; and obtaining the intraday control model. The intraday objective function consists of the scenario occurrence probability, the control interval of the second time scale, the electricity purchase and sale cost, the usage and maintenance cost of the flexible production equipment, the operating cost of the energy storage equipment, and the power fluctuation penalty cost. This preferred embodiment, based on minimizing the total operating cost, incorporates the objective of minimizing power fluctuation and sets constraints such as energy storage operation. By introducing a power fluctuation penalty cost term, it effectively suppresses short-term drastic power fluctuations caused by the randomness of both the source and load sides within the park, achieving a balance between economic dispatch and grid stability, effectively smoothing out power fluctuations in the park, and reducing the impact on the grid.

[0011] Furthermore, the intraday constraints include: intraday power balance constraints, and energy storage device operation constraints including upper and lower limits for charge and discharge power, dynamic updates and upper and lower limits for energy storage capacity, charge and discharge efficiency constraints, and charge and discharge mutual exclusion logic constraints. These ensure the safe and efficient operation of energy storage devices, fully leverage their flexible adjustment capabilities, and support intraday power smoothing and emergency response.

[0012] In its first aspect, this invention provides a preferred embodiment whereby, in the step of constructing a day-ahead decision-making agent and an intraday decision-making agent based on a day-ahead production planning model and an intraday adjustment model, respectively, and then coupling and interacting to solve the problem, the coupling and interaction process involves: mapping the reward function of the intraday decision-making agent onto the day-ahead adjustment time step using an asynchronous transmission mechanism and embedding it into the reward function of the day-ahead decision-making agent, thereby enabling the coupling and interaction between the day-ahead and intraday decision-making agents. This preferred embodiment maps and embeds the reward function of the intraday agent into the day-ahead agent, realizing the linkage and information interaction of dual-timescale strategies, and enhancing the coordination and overall optimization effect between day-ahead planning and intraday adjustment.

[0013] In its first aspect, this invention provides a preferred embodiment in which, in the step of constructing a day-ahead decision-making agent and an intraday decision-making agent based on a day-ahead production planning model and an intraday control model respectively, and then coupling and interacting to solve the problem, the coupled day-ahead decision-making agent and intraday decision-making agent are trained and solved based on an asynchronous dual-interaction flexible actor-critic algorithm architecture. This algorithm is applicable to continuous actions and high-dimensional state spaces, supports parallel learning and collaborative updating of two agents, improves the efficiency and quality of policy learning, and adapts to the decision-making needs in complex dynamic environments.

[0014] The present invention provides a preferred embodiment in its first aspect, wherein the steps of training and solving the coupled interaction between the day-ahead decision agent and the intraday decision agent based on the asynchronous dual-interaction flexible actor-critic algorithm architecture include: initializing the network parameters and target networks of the flexible actor-critic algorithm for the day-ahead and intraday decision agents respectively, and creating independent experience replay pools for each; the day-ahead and intraday decision agents interacting with the environment based on their respective policies, and storing the generated interaction data into their respective experience pools; during the training loop, sampling from the two experience pools respectively, and independently but synchronously updating the critic network, policy network, and target network parameters of the day-ahead and intraday decision agents; after training, the policies obtained by the day-ahead and intraday decision agents are the dual-timescale collaborative control strategies of the industrial user. This preferred scheme establishes experience pools for the day-ahead and intraday agents respectively, and realizes independent sampling and synchronous updating of network parameters. That is, by combining experience playback and target network update mechanisms, it ensures the stability and convergence of dual-timescale policy training, enhances the algorithm's adaptability to complex industrial load control scenarios, ensures that the generated collaborative strategy is both economical and robust, and finally outputs a coordinated dual-scale control strategy.

[0015] The present invention provides a preferred embodiment in its first aspect, wherein the step of constructing a day-ahead production planning model for optimizing and regulating the production plan of an industrial park, with the objective of minimizing the total operating cost of the industrial park under the constraint of satisfying the adjustable range of the industrial load potential, includes: constructing a day-ahead objective function using the partial blob optimization method with the objective of minimizing the total operating cost of the industrial park as the objective, and setting day-ahead constraints for the day-ahead objective function with the adjustable range of the industrial load potential as one of the constraints, thereby obtaining the day-ahead production planning model. This method, by constructing a fuzzy set describing the distribution of uncertainty, optimizes the expected cost in the worst case, exhibiting stronger robustness compared to the scenario generation method, better handling extreme uncertain events, and improving the system's anti-interference capability.

[0016] In a second aspect, this invention provides a dual-timescale control system considering the control potential of industrial users, used to execute the method, comprising: a data acquisition module for acquiring historical industrial load data and adjustable production equipment parameter information of industrial park users; an adjustable production equipment and correlation modeling module for establishing an industrial load model for the adjustable production equipment, establishing a correlation between control incentives and adjustable production equipment based on the industrial load model, and constructing a time-series data input model for industrial user control potential; and a control potential analysis module for inputting historical industrial load data into the industrial user control potential time-series data input model, obtaining industrial user control potential time-series input data, and then using an industrial load adjustable potential analysis module. The system performs a regulation potential analysis to obtain the adjustable range of industrial load potential. A regulation model construction module is used to construct a day-ahead production plan model for optimizing the regulation of the industrial park's production plan, based on the adjustable production equipment parameter information and under the constraint of the adjustable range of industrial load potential, with the goal of minimizing the total operating cost of the industrial park. It also constructs an intraday regulation model for optimizing the regulation of flexible adjustable production equipment, with the goal of minimizing the total operating cost and power fluctuations of the industrial park. A dual-timescale regulation strategy intelligent generation module is used to construct day-ahead and intraday decision-making agents based on the day-ahead production plan model and the intraday regulation model respectively, then couple and solve them to generate a day-ahead-intraday coordinated regulation strategy.

[0017] Compared with the prior art, the present invention has the following advantages: To address the technical problems of inaccurate assessment of control potential, lack of dynamic adaptability in day-ahead control plans, and inability to cope with dynamic and uncertain environments in existing industrial park control methods, this invention first establishes an industrial load model for the adjustable production equipment of industrial park users. Based on the industrial load model, it establishes the correlation between control incentives and adjustable production equipment to construct an industrial load adjustable potential analysis model. This model fully considers the actual production situation and equipment characteristics of industrial users, effectively and accurately simulating the adjustable boundary of industrial equipment under the influence of control incentives, improving the accuracy of control potential analysis and assessment results, and effectively guiding actual control operations. Furthermore, based on the parameter information of adjustable production equipment, a day-ahead production plan model and an intraday control model with dual time scales are constructed with the goal of minimizing the total operating cost of the industrial park. This fully considers control cost factors and effectively ensures the production energy planning of park users and taps into their adjustment potential. Simultaneously, based on the previously established correlation between control incentives and adjustable production equipment, a time-series data input model for industrial user control potential is constructed to obtain time-series input data for industrial user control potential, providing support for dynamic assessment of control potential. Then, the industrial load adjustable potential analysis model is used to analyze the control potential of the time-series input data to obtain the adjustable range of industrial load potential, enabling dynamic assessment of control potential. Then, under the constraint of meeting the adjustable range of industrial load potential, a day-ahead decision agent and an intraday decision agent are constructed based on a dual-time-scale day-ahead production planning model and an intraday control model, and the two are coupled and interacted. The dynamic coupling relationship between day-ahead production planning and intraday real-time control is fully considered, so as to realize the dynamic assessment of the intraday control potential of industrial users. This allows the day-ahead plan to adapt to sudden situations and load changes in actual operation, while realizing efficient day-ahead and intraday coordinated control and precise control of industrial parks in a dynamic and uncertain environment, reducing operating costs and ensuring the stable operation of the power grid. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the steps of a dual-timescale control method considering the control potential of industrial users, provided in a specific embodiment of the present invention. Figure 2 This is a flowchart of a dual-timescale control method that considers the control potential of industrial users, provided as another specific embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: Please refer to Figure 1 The dual-timescale control method considering the control potential of industrial users provided in this embodiment is mainly implemented through steps S0-S4: S0. Industrial Load Data Acquisition: Acquire historical industrial load data and adjustable production equipment parameter information of users in the industrial park.

[0022] Understandably, the historical industrial load data acquired from industrial park users is used in subsequent step S2 to obtain time-series input data on the control potential of industrial users, enabling control potential analysis before control measures are implemented. In an optional implementation, adjustable production equipment (also known as typical equipment) mainly includes the main production equipment and auxiliary production equipment in the industrial park. The main production equipment includes equipment subject to continuous impact loads and equipment subject to indirect impact loads. The parameter information of all these adjustable production equipment is acquired and used in subsequent step S3 for constructing the control model. This implementation method allows for a more refined characterization of the operating characteristics and control capabilities of different types of equipment, thereby improving the overall model's adaptability to complex industrial scenarios. By integrating multi-source data and mechanistic models, dynamic evaluation of equipment-level adjustable potential is achieved, providing precise support for upper-level collaborative control strategies.

[0023] S1. Adjustable production equipment and related modeling in industrial parks: Based on the acquired data, establish an industrial load model for the adjustable production equipment, establish the correlation between control incentives and adjustable production equipment based on the industrial load model, and construct a time series data input model for the control potential of industrial users.

[0024] Specifically, in a preferred embodiment, an industrial load model is established for the adjustable production equipment (also known as typical equipment) of industrial park users. Based on the industrial load model, the correlation between control incentives and adjustable production equipment is established using the sliding window method, and a time-series data input model for the controllability potential of industrial users is constructed. Step S1 provides a model basis for the analysis of the adjustable potential of industrial users and the control of the park by establishing physical models (i.e., load models) of the main production equipment and auxiliary production equipment in the industrial park, as well as a correlation model between adjustable production equipment and control incentives. The main process is: establishing continuous and indirect impact load equipment models, auxiliary production equipment models, and defining control incentives; establishing a correlation model between adjustable production equipment and control incentives based on the sliding window method.

[0025] Please refer to Figure 2 In one optional implementation, the impact load equipment, the indirect impact load equipment, and the auxiliary production equipment are modeled separately to obtain the impact load equipment load model, the indirect impact load equipment load model, and the auxiliary production equipment load model. Therefore, please refer to... Figure 2 Step S1 mainly includes modeling the main production equipment of the industrial park, modeling the auxiliary production equipment, and modeling the correlation between regulatory incentives and the adjustable potential of industrial users. The specific process is as follows: S11. Modeling of major production equipment: Major industrial production equipment mainly includes equipment subjected to continuous impact loads and equipment subjected to indirect impact loads.

[0026] Modeling of equipment subjected to continuous impact loads: ; In the formula: For equipment subjected to continuous impact loads during the period The power; This refers to shutdown periods with continuous impact loads. This is the period during which sustained impact loads begin; The rated power of equipment subjected to continuous impact load; This refers to the time required for a device under continuous impact load to reach its rated power from the start of operation.

[0027] Modeling of equipment subjected to indirect impact loads: ; In the formula: For indirect impact load equipment during the time period The power; This refers to the energizing period for equipment subjected to indirect impact loads. The rated power of the indirect impact load equipment; The time required for an indirect impact load device to reach its rated power after being energized; For indirect impact load equipment operating in steady state during time periods Power fluctuations; This refers to the power outage period for equipment subjected to indirect impact loads. This refers to the time it takes for an indirect impact load device to reduce its power to zero after a power outage.

[0028] S12. Modeling of Auxiliary Production Equipment: ; In the formula: To assist production equipment during the time period The power; The rated power of auxiliary production equipment; , These refer to the start and stop times of auxiliary production equipment; , These are the power reduction and ramp-up time of auxiliary production equipment, respectively. , These refer to the ramp speed for reducing and increasing the power of auxiliary production equipment, respectively. To adjust the power.

[0029] S13. Construction of Time Series Data Input Model for Regulation Potential: Based on the industrial load model, the sliding window method is used to establish the correlation between regulation incentives and adjustable production equipment, and a time series data input model for industrial user regulation potential is constructed.

[0030] Specifically, based on the sliding window, the regulation stimulus and industrial load power are modeled as follows: ; ; ; In the formula: , , These correspond to sliding input windows for off-peak hours, peak hours, and normal hours, respectively. , , These represent the adjustable power consumption of production equipment under three time periods: off-peak hours, peak hours, and normal hours, which are the sum of the power consumption of equipment under continuous impact load, indirect impact load, and auxiliary production equipment. Under off-peak hours... = + + During peak hours: = + + ; During normal periods, = + + . , , These are the corresponding control incentives for off-peak hours, peak hours, and normal hours.

[0031] Through the above process, a time-series data input model for industrial user control potential was constructed, and the resulting time-series input data for industrial user control potential, namely, the power consumption data of equipment within the sliding input window for three time periods, was obtained.

[0032] In a preferred embodiment, the control excitation is further expressed as follows: ; In the formula: The selling price of industrial products per unit of load; The unit production cost of industrial products; Electricity consumption per unit output of industrial products; The electricity purchase price; To adjust the total battery level upwards; This refers to the total power generation generated by the industrial user's self-owned power plant during the corresponding time period. This refers to the cost of generating electricity from a self-owned power plant. It's understandable that in industrial product pricing, "load unit" typically refers to the unit of measurement for electricity load, reflecting the electricity consumption required by the enterprise during production. "Upward adjustment" refers to the total electricity consumption exceeding the power generation capacity of its own power plant, requiring the purchase of electricity from external sources.

[0033] S2. Industrial Load Adjustment Potential Analysis: Historical industrial load data is input into the industrial user adjustment potential time-series data input model. After obtaining the industrial user adjustment potential time-series input data, the adjustment potential analysis is performed using the industrial load adjustable potential analysis model to obtain the adjustable range of industrial load potential. Step S2 provides adjustment boundaries for multi-time-scale control of industrial parks by analyzing the adjustable potential of adjustable production equipment in different time periods. In a more preferred embodiment, the industrial load adjustable potential analysis model is constructed based on multiple neural networks (fully convolutional neural network, temporal convolutional network, and long short-term memory network). The main process is as follows: industrial load data is selected and expanded to establish a selected time-series dataset of industrial load adjustment potential, then fitting analysis is performed to output the preliminary adjustable range of industrial load potential, and finally the preliminary adjustable range of industrial load potential is corrected to output the adjustable potential range of industrial equipment.

[0034] In this preferred embodiment, step S2 is mainly implemented through an industrial load regulation potential analysis model based on a fully convolutional-temporal convolutional-long short-term memory neural network. Please refer to... Figure 2 The specific process is as follows: S21. Selection and expansion of fully convolutional neural network dataset: Based on the fully convolutional neural network, the time-series input data of the industrial user control potential are selected to obtain the preferred time-series dataset. The power generation data of the self-owned power plant of the industrial park is input into the preferred time-series dataset in negative form to expand it, and the selected time-series dataset is obtained.

[0035] First, based on a fully convolutional neural network, time-series input data on the control potential of industrial users are selected to form a time-series dataset for optimal industrial load control potential.

[0036] ; In the formula: The data represents industrial load data optimized through fully convolutional layers. To optimize the total number of data points; The function types corresponding to different layers of a fully convolutional neural network. This is the kernel limit. The sampling factor; For the first The value after convolution of 1 data point , The first , The sampling factor corresponding to each data point , Each convolution operation is applied to the first... , The kernel size used for each data point is the convolution coefficient selected in one convolution operation.

[0037] Secondly, the power generation data of the industrial park's self-owned power plants are input into the selected preferred time-series dataset in negative form to expand the dataset. This avoids ignoring negative values ​​during potential analysis. After introducing the power generation data from the self-owned power plants into the dataset, a refined dataset is obtained, in the following format: ; In the formula: For the first Power generation data from a self-owned power plant; For the corresponding number Fixed parameters for each power generation data point; Let be the i-th industrial load data point. The formula means that for each element in the input vector, if it is positive, it is retained; if it is negative, it is divided by a fixed parameter (greater than 1) to reduce its absolute value, and then the negative sign is retained.

[0038] S22. Preliminary analysis of the potential range of temporal convolutional networks: Based on the fitting analysis of selected time-series datasets using temporal convolutional networks, the preliminary adjustable range of industrial load potential (i.e., the preliminary industrial load regulation potential) is output.

[0039] Furthermore, to improve generalization ability, in a preferred embodiment, the input data is subjected to Gaussian processing, that is, Gaussian linear error units are introduced to enable the neural network to fit the load data under different time periods more efficiently. Specifically, this is achieved through the Gaussian linear error unit function. The calculation is as follows: ; Temporal convolutional networks are built from dilated convolutions, and the formula for calculating dilated convolutions is: ; In the formula: Data points after the output of a fully convolutional neural network The convolution result at the point; This indicates that historical industrial load data is convolved. These are the filter coefficients of the temporal convolutional network. For expansion coefficient, " indicates convolution; The kernel size; The first before dilated convolution There are 1 data point, where d is the expansion coefficient and c is the index of s.

[0040] S23. Long Short-Term Memory Network Potential Range Correction: Based on the Long Short-Term Memory network, the initial adjustable range of industrial load potential is corrected, and the adjustable range of industrial load potential (i.e., industrial load regulation potential) is output. This correction reduces errors. The Long Short-Term Memory network formula is as follows: ; ; In the formula: for Output of the time-lapse forget gate; For the Sigmoid function; The weight of the forgetting gate; for Industrial load signal during specific time periods; for Industrial load power during a given time period; To transform paranoia; for Cell output in a time-segmented long short-term memory network; for Industrial load information in the time period cell; for The time period is a variable that determines the degree to which new information is retained; for The newly input industrial load information is stored in the time unit core. The Long Short-Term Memory (LSTM) network corrects for errors in the potential assessment results by remembering the short-term relationships between industrial loads. The output is the final industrial load regulation potential assessment result, i.e., the adjustable range of typical industrial load, which is input as a constraint into the industrial park's day-ahead production plan generation module.

[0041] For steps S22 and S23, this embodiment constructs the overall architecture of: input data (Gaussian processing) - temporal convolution - long short-term memory network.

[0042] S3. Construction of Industrial Park Control Model: Based on the adjustable production equipment parameter information, and under the constraint of the adjustable range of industrial load potential, a day-ahead production plan model is constructed with the goal of minimizing the total operating cost of the industrial park, and an intraday control model is constructed with the goal of minimizing the total operating cost and power fluctuation of the industrial park, to optimize the control of flexible adjustable production equipment.

[0043] This step primarily involves constructing a daytime production planning model at the first timescale (hourly) and a daytime control model at the second timescale (every 15 minutes). Under the constraint of the adjustable potential range derived in step S2, an adjustable production equipment control decision model is constructed based on the acquired adjustable production equipment parameter information. The goal is to optimize the industrial park's production plan by minimizing the total operating cost. The daytime control strategy aims to smooth power fluctuations in the industrial park by adjusting the speed of equipment such as electric arc furnaces and the charging and discharging power of the energy storage system. Please refer to [reference needed]. Figure 2 The specific process is as follows: S31. Construction of the Day-ahead Production Planning Model and Constraints: A day-ahead objective function is constructed using a scenario generation method, with the goal of minimizing the total operating cost of the industrial park. The adjustable range of the industrial load potential is used as one of the constraints to set day-ahead constraints for the objective function, thus obtaining the day-ahead production planning model. The day-ahead objective function consists of the scenario occurrence probability and the adjustment range of the first time scale (per hour). The costs consist of electricity purchase and sale costs, the operating and maintenance costs of adjustable production equipment, and the control and incentive costs of adjustable production equipment. Day-ahead constraints also include: day-ahead power balance constraints, constraints on the correlation between equipment operating status and power, production task constraints, constraints on the shortest continuous operating time of equipment, constraints on the upper limit of the number of equipment start-ups and shutdowns, constraints on the number of equipment starting and stopping simultaneously, and constraints on production process coupling.

[0044] More specifically, the day-ahead production planning model for industrial parks includes optimizing the day-ahead objective function and setting constraints: First, in the day-ahead production plan of the industrial park, on the basis of ensuring normal production of the industrial park and not exceeding the control potential range, an objective function is constructed with the goal of minimizing the total operating cost of the park. At the same time, taking into account the uncertainty of photovoltaic constraints, a photovoltaic output scenario is generated using the scenario generation method to deal with this uncertainty.

[0045] ; In the formula: the first two items are the electricity purchase and sale costs of the industrial park, and the last item is the sum of the use and maintenance costs and regulation incentive costs of the industrial productive load (i.e., the adjustable production equipment); This represents the total number of typical scenarios generated. Let s be the probability of scenario s occurring. , They are respectively Electricity purchase price and electricity sales price for different time periods; , In scene s respectively Power purchased and sold during specific time periods; For adjustable production equipment The cost of use and maintenance; This represents the total number of scheduling periods within a single complete scheduling cycle. A collection of adjustable production equipment; The power consumption of adjustable production equipment i during time period T; To regulate load changes; The control excitation coefficient (yuan / kWh) for adjustable production equipment i. The day-ahead control interval (time interval) is 1 hour. It is understood that the adjustable production equipment i referred to in this embodiment is a collective term for industrial loads including equipment subject to continuous impact loads, equipment subject to indirect impact loads, and auxiliary production equipment.

[0046] Secondly, the following constraints need to be met during the optimization of the objective function before the date: (1) Day-ahead power balance constraints: ; In the formula: For scene s Photovoltaic unit j's power output during the time period; for The total power consumed by all inflexible electrical loads (adjustable production equipment) in the time-of-use system; Let j be the set of photovoltaic units, and i be the index of the adjustable production equipment.

[0047] (2) Constraints relating equipment operating status and power: use variable Characterizes the operating status of adjustable production equipment, if the equipment exist If it operates during a specific time period, then ,otherwise, ,at this time, The power consumed by the start / stop control equipment during a given time period can be expressed as: ; In the formula: and These represent the industrial load adjustable production equipment in... Power consumed during a given period and the rated power of the equipment.

[0048] (3) Production task constraints (i.e., minimum energy constraints): To ensure the output of industrial products, the power consumed by production equipment throughout the entire scheduling cycle must guarantee the minimum energy required for production. : ; (4) Adjustable range constraint of industrial load potential: To ensure that industrial user load regulation remains within its adjustable potential range, the following constraints must be met:

[0049] ; In the formula, , These are the minimum and maximum adjustable power of equipment i during time period T, as analyzed by the industrial load regulation potential analysis module.

[0050] (5) Minimum continuous operating time constraint of equipment: Meanwhile, considering the continuity of industrial production, the continuous working time of the equipment during a single startup must be greater than the minimum continuous working time. ; In the formula: For adjustable production equipment The lower limit of the duration of a single job. The duration of a single operation of adjustable production equipment. for The working status of the production equipment can be adjusted in real time. for The working status of production equipment can be adjusted in real time.

[0051] (6) Maximum number of equipment start-ups and shutdowns: Furthermore, considering that frequent equipment start-ups and shutdowns may affect product quality and even impact the power grid, it is necessary to constrain the number of start-ups and shutdowns of the production load. ; ; In the formula: , Adjustable production equipment exist Startup status variables for a given time period and the maximum number of times a unit can be started; , Adjustable production equipment exist The shutdown status variables for different time periods and the upper limit for the number of unit shutdowns.

[0052] (7) Constraints on the number of devices that can be started and stopped simultaneously: Further consideration is needed. Maximum number of devices that can be started and stopped simultaneously during a given time period , constraint: To limit the number of devices that can be started simultaneously, Maximum number of devices that can be shut down simultaneously: ; ; (8) Production process coupling constraints: Furthermore, considering the potential coupling relationships between industrial load production processes, it is necessary to further constrain the start-up and shutdown of adjustable production equipment with upstream and downstream process relationships: ; That is, adjustable production equipment Must be in adjustable production equipment Start-stop interval Start and stop. For adjustable production equipment and adjustable production equipment Startup interval, For adjustable production equipment and adjustable production load The shutdown interval.

[0053] S32. Intraday Control Model and Constraint Construction: An intraday objective function is constructed using a scenario generation method, with the objectives of minimizing the total operating cost and power fluctuation of the industrial park. Intraday constraints are then set for the intraday objective function to obtain the intraday control model. The intraday objective function consists of the scenario occurrence probability and the control interval of the second time scale. The cost consists of electricity purchase and sale costs, flexible and adjustable production equipment usage and maintenance costs, energy storage equipment operation costs, and power fluctuation penalty costs; intraday constraints include: intraday power balance constraints, and energy storage equipment operation constraints including upper and lower limit constraints for charging and discharging power, dynamic updates and upper and lower limit constraints for energy storage capacity, charging and discharging efficiency constraints, and charging and discharging mutual exclusion logic constraints.

[0054] More specifically, the intraday control model for industrial parks includes intraday objective function optimization and constraint conditions. The specific process is as follows: First, the intraday control strategy for industrial parks, based on the day-ahead production plan, further optimizes the relatively small but more flexible control resources within the industrial park, such as energy storage charging and discharging power. This smooths out power fluctuations in the industrial park while ensuring optimal operational economy. An intraday control objective function is constructed with the goal of minimizing the total operating cost of the park. Compared to the day-ahead objective function, it additionally considers the penalty cost for power fluctuations. ; In the formula: and The energy storage devices i in the industrial park The charging and discharging power (discharging is positive) and usage and maintenance costs during the time period; A collection of energy storage devices within an industrial park; This refers to a collection of flexible and adjustable production equipment such as electric arc furnaces (which is part of adjustable equipment); the last item represents the penalty imposed by short-term power fluctuations in the industrial park on the impact on the external power grid. This is the penalty coefficient; This represents the total number of scheduling periods within a complete scheduling cycle within a day. , These refer to the electricity purchase price and electricity sales price for time period t, respectively. , These represent the electricity purchased and sold during time period t in scenario s. For adjustable production equipment The cost of use and maintenance; Let t be the power consumption of adjustable production equipment i during time period t; For the power purchased during time period t-1 in scenario s, The electricity sales power during time period t-1 in scenario s; This is a daily control range (time interval), and the previous control time interval was 15 minutes.

[0055] The number of typical scenes generated; Let s be the probability of scenario s occurring. , They are respectively Electricity purchase price and electricity sales price for different time periods; , In scene s respectively Power purchased and sold during specific time periods; For adjustable production equipment The cost of use and maintenance; This represents the total number of scheduling periods within a single complete scheduling cycle. Indicates the total number of scenes; It is a collection of adjustable production equipment. This refers to the collective term for industrial loads, which include equipment subject to continuous impact loads, equipment subject to indirect impact loads, and auxiliary production equipment. To regulate load changes, This is the first control interval of the day, also known as the first time scale, and the day-ahead scheduling interval is 1 hour.

[0056] The constraints that must be met for intraday adjustments are as follows: (1) Intraday power balance constraints: ; In the formula, The production load power generated in the day-ahead production plan during time period t is the result of the day-ahead scheduling solution. Let J be the output power of photovoltaic power generation unit j during time period t in scenario s; The total power consumed by all inflexible electrical loads (adjustable production equipment) in the system during time period t.

[0057] (2) Operational constraints of energy storage equipment: The energy storage equipment configured in the park also needs to meet the following constraints (charge and discharge power upper and lower limit constraints, energy storage capacity dynamic update and upper and lower limit constraints, charge and discharge efficiency constraints, and charge and discharge mutual exclusion logic constraints): ; In the formula: , This indicates the upper and lower limits of the charging and discharging power of the energy storage device. , For energy storage devices i Battery level during the period and initial battery level; , These represent the upper and lower limits of the energy storage capacity of energy storage i, respectively; , The charging and discharging efficiency of energy storage device i. Let k be the energy storage efficiency of energy storage device i, and k be the index of the time period.

[0058] S4. Intelligent Generation of Dual-Time-Scale Control Strategy for Industrial Parks: Based on the day-ahead production planning model and the intraday control model, day-ahead decision-making agents and intraday decision-making agents are constructed respectively, coupled and interacted, and solved to generate a day-ahead-intraday coordinated control strategy. The coupling and interaction process is as follows: based on the asynchronous transmission mechanism, the reward function of the intraday decision-making agent is mapped on the day-ahead control time step and embedded into the reward function of the day-ahead decision-making agent, so that the day-ahead decision-making agent and the intraday decision-making agent are coupled and interact. Furthermore, considering the problem of insufficient intraday control response capability in existing technologies, that is, in the intraday control stage, existing systems often lack efficient real-time data processing and decision support mechanisms, cannot quickly respond to load changes, and cannot fully utilize the control potential of industrial users for real-time adjustments, thereby affecting the energy efficiency and stability of the entire park, this invention provides a more preferred embodiment, specifically based on the asynchronous dual-interaction Flexible Actor-Critic Algorithm (SAC algorithm) architecture to train and solve the coupled and interacted day-ahead decision-making agents and intraday decision-making agents to generate a day-ahead-intraday coordinated control strategy. Based on the objective function and constraints generated in step S3, this step constructs a day-ahead decision-making agent and an intraday decision-making agent (both agents are based on the SAC algorithm framework), and builds an asynchronous dual-interaction SAC algorithm solution architecture to intelligently generate day-ahead and intraday coordinated control strategies. This preferred implementation, supported by a neural network model and the asynchronous dual-interaction SAC algorithm, can achieve real-time data processing and efficient output, rapidly respond to load changes, fully utilize the control potential of industrial users to make real-time adjustments, effectively guarantee the production energy planning of park users, tap into user adjustment potential, and fully realize efficient day-ahead and intraday coordinated control of industrial parks in dynamic and uncertain environments, reducing operating costs and ensuring stable grid operation.

[0059] As is understandable, the SAC algorithm is a deep reinforcement learning algorithm based on the principle of maximum entropy. Its core objective is to encourage exploration, maximizing both the accumulated expected reward and the policy entropy, thereby achieving more robust and efficient offline learning. An SAC agent consists of five main neural networks: a policy network for outputting the probability distribution of actions, two critic networks, and two target critic networks for evaluating the value of actions. Based on the current state, the agent samples and executes actions through the policy network, obtaining rewards from environmental feedback and the new state, continuously updating the network until the reward is maximized.

[0060] Please refer to Figure 2 Step S4. Intelligent generation of dual-timescale control strategies for industrial parks, the specific process is as follows: S41. Construction of the Day-ahead Decision-Making Agent: Based on the day-ahead production planning model, construct the day-ahead decision-making agent. The action space of this agent... for: ; The state space of the intelligent agent for: ; The agent's reward function for: ; In the formula: To regulate the penalty factor for exceeding the user's control potential, To limit the potential for regulation, the reward function is designed so that the agent, while minimizing costs during decision-making, is also guaranteed to regulate within the range of the user's regulatory potential. This involves mapping the reward function of the intraday decision-making agent to the day-ahead control time step, and asynchronously transmitting this reward function to achieve coupling and interaction between the day-ahead decision-making agent and the intraday decision-making agent's policies.

[0061] S42. Construction of Intraday Decision-Making Agent: Based on the intraday regulation model, construct an intraday decision-making agent.

[0062] The agent's action space for: ; In the formula: This refers to the status of flexible equipment such as electric arc furnaces.

[0063] The state space of the intelligent agent for: ; In the formula: , These represent the minimum and maximum adjustable power of device i during time period t, respectively, which is the adjustable range of industrial load potential obtained from the control potential analysis in step S2.

[0064] Therefore, the decisions made in the day-ahead phase will serve as known observations for the intraday phase, thereby achieving coordination between the day-ahead decision-making agent and the intraday dual-timescale strategy. The reward function of this agent... for: ; Inputting the adjustable range constraint of industrial load potential obtained in S31, the day-ahead decision agent first formulates the day-ahead production plan; the intraday decision agent formulates the intraday control plan to correct the load deviation of the day-ahead production plan, and feeds back the control experience to the day-ahead decision agent.

[0065] S43. Asynchronous Dual-Interaction SAC Solution: Based on the SAC algorithm, an asynchronous dual-interaction SAC solution method is constructed to solve the day-ahead and intraday decision-making agents. The main solution process is as follows: Initialize the network parameters and target networks of the flexible actor-critic algorithm for the day-ahead and intraday decision-making agents respectively, and create independent experience replay pools for each; The day-ahead and intraday decision-making agents interact with the environment based on their respective policies, and store the generated interaction data into their respective experience pools; During the training loop, samples are taken from the two experience pools respectively, and the parameters of the critic network, policy network, and target network of the day-ahead and intraday decision-making agents are updated independently but synchronously; After training, the policies obtained by the day-ahead and intraday decision-making agents are the dual-timescale collaborative control policies of the industrial user.

[0066] The more detailed solution process is as follows: S431. Randomly initialize the SAC parameter vectors of the day-ahead decision agent and the intraday decision agent; S432. Randomly initialize the target network of SAC; S433. Create a day-ahead experience replay pool for the day-ahead decision-making agent and an intraday experience replay pool for the intraday decision-making agent; S434. (Start training in the main loop) Obtain the initial state space of the day-ahead decision agent and the state space of the day-intraday decision agent; S435. (Time step loop) The day-ahead decision agent randomly samples an action and executes it based on the current policy and state, receives the day-ahead reward and the next state, and stores the interaction experience in the day-ahead experience replay pool; the day-intraday decision agent randomly samples an action and executes it based on the current policy and state, receives the day-intraday reward and the next state, and stores the interaction experience in the day-intraday experience replay pool. S436. (Network Update) If training has not stopped, perform the following network update steps: Randomly sample experience data from the intraday experience replay pool and calculate the current objective (reward). Update the critic network of the intraday decision agent and update the policy network of the intraday decision agent using the sampled gradients. Update the objective network parameters of the intraday decision agent. Randomly sample experience data from the previous day's experience replay pool and calculate the current objective. Update the critic network of the previous day's decision agent and update the policy network of the previous day's decision agent using the sampled gradients. Update the objective network parameters of the previous day's decision agent; S437. Repeat S435 and S436 until the preset training is completed; S438. After training, the strategies derived by the daytime decision-making agent and the intraday decision-making agent are the dual-timescale collaborative control strategies for the industrial user.

[0067] S436. If training has not stopped, perform the following network update steps: randomly sample experience data from the intraday experience replay pool, calculate the current objective, update the critic network of the intraday decision agent, update the policy network of the intraday decision agent using the sampled gradients, and update the objective network parameters of the intraday decision agent; randomly sample experience data from the previous day's experience replay pool, calculate the current objective, update the critic network of the previous day's decision agent, update the policy network of the previous day's decision agent using the sampled gradients, and update the objective network parameters of the previous day's decision agent. S437. Repeat S435 and S436 until the preset training is completed; S483. After training, the strategies derived by the daytime decision-making agent and the intraday decision-making agent are the dual-timescale collaborative control strategies for the industrial user.

[0068] Corresponding to the method of this embodiment, a dual-timescale control system that considers the control potential of industrial users is provided, which mainly consists of the following parts: The data acquisition module is used to acquire historical industrial load data and adjustable production equipment parameter information of users in the industrial park; The adjustable production equipment and related modeling module is used to establish an industrial load model for adjustable production equipment, establish the correlation between control incentives and adjustable production equipment based on the industrial load model, and construct a time series data input model for industrial user control potential. The regulation potential analysis module is used to input historical industrial load data into the industrial user regulation potential time series data input model. After obtaining the industrial user regulation potential time series input data, the regulation potential analysis is performed using the industrial load adjustable potential analysis model to obtain the adjustable range of industrial load potential. The control model construction module is used to construct a day-ahead production plan model for optimizing the control of the industrial park's production plan based on the adjustable production equipment parameter information, under the constraint of satisfying the adjustable range of the industrial load potential, with the goal of minimizing the total operating cost of the industrial park, and to construct an intraday control model for optimizing the control of the flexible adjustable production equipment with the goal of minimizing the total operating cost and power fluctuation of the industrial park. The dual-timescale control strategy intelligent generation module is used to construct day-ahead decision agents and intraday decision agents based on the day-ahead production planning model and the intraday control model, respectively, and then couple and interact to solve them to generate day-ahead-intraday coordinated control strategies.

[0069] Example 2: Considering that although the day-ahead model uses the scenario method to handle photovoltaic uncertainties, it does not adequately consider high-risk, low-probability uncertainties such as extreme weather events, sudden equipment failures, sharp fluctuations in market electricity prices, and temporary changes in orders, a regulation method suitable for high uncertainty environments is proposed. That is, for the construction of the day-ahead production planning model, the objective function is constructed by combining the partial blue bar optimization method (DRO) with the goal of minimizing the total operating cost of the industrial park, replacing the scenario generation method in the above implementation method.

[0070] S31'. Construction of the day-ahead production planning model and constraints: The day-ahead objective function is constructed by combining the partial blob optimization method with the goal of minimizing the total operating cost of the industrial park. The adjustable range of the industrial load potential is used as one of the constraints to set the day-ahead constraint conditions for the day-ahead objective function, thus obtaining the day-ahead production planning model.

[0071] The robust optimization method presented in this embodiment is a methodology between stochastic programming (where the exact distribution is known) and robust optimization (where only the set is known). It assumes that the true probability distribution P of the uncertainty parameter (such as photovoltaic output) belongs to a predefined fuzzy set. (A confidence sphere is typically centered on a reference distribution and has a radius equal to the Wasserstein distance or moment information). The optimization objective is to minimize the expected cost under the worst-case distribution. The specific implementation path is as follows: (1) Define fuzzy sets (Ambiguity Set): Constructed based on historical photovoltaic prediction error data, where P is a random variable representing photovoltaic power output. The true probability distribution belongs to fuzzy sets. This set is based on the empirical distribution of historical photovoltaic power output data from industrial parks. Constructed using a metric (such as Wasserstein distance), the formula is: .in, Let ε be the Wasserstein distance and ε be the radius. The magnitude of ε controls the conservatism of the model.

[0072] (2) Reconstruct the objective function: Rewrite the objective function in the following form: ; In the formula: x is a decision variable vector, which includes all optimizable variables (such as: power purchased and sold, power consumed by equipment, etc.); The feasible region of the decision variables is the set defined by all constraints (power balance, equipment operation, control potential, etc.) of the original model. Let P be an uncertain random variable (i.e., a photovoltaic power output random variable), representing the uncertain factors affecting the system, mainly referring to photovoltaic power output here; The true probability distribution; Let P be the expected value of the distribution. This represents the total number of scheduling periods within a single complete scheduling cycle prior to the current day. The current control range (time interval); , They are respectively Electricity purchase price and electricity sales price for different time periods; , These are respectively dependent on photovoltaic random variables hour Power purchased and sold during specific time periods; A collection of adjustable production equipment; For adjustable production equipment The cost of use and maintenance; The power consumption of adjustable production equipment i during time period T; The control excitation coefficient (yuan / kWh) for adjustable production equipment i. To regulate load changes.

[0073] (3) The constraints of the reconstructed objective function are the same as those in S32.

[0074] This specific implementation method constructs the day-ahead objective function by combining the sub-Bruker optimization method. Compared with the scenario method, the sub-Bruker optimization method constructs a fuzzy set describing the range of uncertainty fluctuations and seeks the optimal solution under the worst possible distribution. This effectively prevents errors in the prediction model itself, enhances the system's robustness to unknown disturbances, and is more suitable for the control needs under high uncertainty environments.

[0075] Based on the above embodiments, the present invention can achieve the following beneficial technical effects: This invention first establishes an industrial load model for the adjustable production equipment of industrial park users, fully considering the actual production situation and equipment characteristics of industrial users. Based on the industrial load model, a sliding window method is used to establish the correlation between control incentives and adjustable production equipment. Then, an industrial load adjustable potential analysis model is constructed to effectively simulate the adjustable boundary of industrial equipment under the influence of control incentives. Second, based on the acquired adjustable production equipment parameter information, and under the constraint of satisfying the adjustable range of industrial load potential, a dual-timescale day-ahead production planning model and an intraday control model are constructed with the goal of minimizing the total operating cost of the industrial park. This fully considers control cost factors, effectively improving the accuracy of control potential assessment and providing effective guidance for actual control operations. Third, based on the day-ahead production planning model and the intraday control model, a day-ahead decision-making agent and an intraday decision-making agent are constructed and coupled and interacted with each other. This fully considers the dynamic coupling relationship between day-ahead market decisions and intraday real-time control, enabling precise control in dynamic and uncertain environments. Finally, the solution intelligently generates coordinated control strategies for both day-ahead and intraday phases, achieving efficient day-ahead planning and flexible intraday control of energy in the industrial park. This coordinates various methods, including production planning for adjustable industrial equipment and flexible control of production equipment (such as charging and discharging power control of energy storage devices), under dynamic and uncertain environments. It fully leverages the adjustable potential of various resources within the industrial park, improving economic efficiency. Furthermore, supported by a neural network model and an asynchronous double-interaction SAC algorithm, this application enables real-time data processing and efficient output, rapid response to load changes, and full utilization of industrial users' control potential for real-time adjustments. This effectively ensures the energy planning of park users, taps into their control potential, achieves efficient day-ahead and intraday coordinated control of the industrial park under dynamic and uncertain environments, reduces operating costs, and ensures stable grid operation.

[0076] This invention fully considers the time-dependent nature of industrial production processes and the response characteristics of equipment. Through an asynchronous dual-interaction mechanism, it effectively connects day-ahead planning with intraday adjustments, enhancing the adaptability and robustness of control strategies in dynamic environments. The constructed model can accurately quantify the adjustable potential of each user at different time scales, effectively balancing grid dispatching needs and user energy preferences for production. This reduces the overall operating costs of industrial parks while enhancing grid interaction capabilities, providing an efficient and feasible technical path for industrial parks to participate in demand response.

[0077] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0078] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. Furthermore, the above embodiments only illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. For those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A dual-timescale control method considering the control potential of industrial users, characterized in that, include: Obtain historical industrial load data and adjustable production equipment parameter information from users in the industrial park; Based on the acquired data, an industrial load model is established for adjustable production equipment. Based on the industrial load model, the correlation between control incentives and adjustable production equipment is established, and a time series data input model for industrial user control potential is constructed. Historical industrial load data is input into the industrial user control potential time series data input model. After obtaining the industrial user control potential time series input data, the control potential analysis is performed using the industrial load adjustable potential analysis model to obtain the industrial load potential adjustable range. Based on the adjustable production equipment parameter information, and under the constraint of satisfying the adjustable range of industrial load potential, a day-ahead production planning model is constructed with the goal of minimizing the total operating cost of the industrial park, and an intraday control model is constructed with the goal of minimizing the total operating cost and power fluctuation of the industrial park, to optimize the control of flexible adjustable production equipment. Based on the day-ahead production planning model and the intraday control model, respectively, the day-ahead decision-making agent and the intraday decision-making agent are constructed, coupled and interacted, and solved to generate a day-ahead-intraday coordinated control strategy.

2. The dual-timescale control method considering the control potential of industrial users according to claim 1, characterized in that, In the step of establishing the correlation between control incentives and adjustable production equipment based on the industrial load model, the sliding window method is specifically used to establish the correlation between control incentives and adjustable production equipment. The sliding window includes three types of sliding input windows: off-peak period, peak period, and normal period. The power consumption data of the equipment in the sliding input windows of the three periods are the time-series input data of the industrial user's control potential.

3. The dual-timescale control method considering the control potential of industrial users according to claim 1, characterized in that, The industrial load adjustable potential analysis model is constructed based on a fully convolutional neural network, a temporal convolutional network, and a long short-term memory network. The step of using a neural network-based industrial load adjustable potential analysis model to perform regulation potential analysis and obtain the adjustable range of industrial load potential includes: The time-series input data of the industrial user control potential are selected based on a fully convolutional neural network to obtain an optimal time-series dataset. The power generation data of the self-owned power plant in the industrial park is then input into the optimal time-series dataset in negative form to expand it, resulting in a refined time-series dataset. Based on the temporal convolutional network, a fitting analysis of a selected time-series dataset is performed to output the preliminary adjustable range of industrial load potential. The initial adjustable range of industrial load potential is corrected based on a long short-term memory network, and the adjustable range of industrial load potential is output.

4. The dual-timescale control method considering the control potential of industrial users according to claim 1, characterized in that, The step of constructing a day-ahead production planning model for optimizing and regulating the industrial park's production plan, under the constraint of satisfying the adjustable range of the industrial load potential, with the goal of minimizing the total operating cost of the industrial park, includes: constructing a day-ahead objective function using a scenario generation method with the goal of minimizing the total operating cost of the industrial park, and setting day-ahead constraints for the day-ahead objective function with the adjustable range of the industrial load potential as one of the constraints, to obtain the day-ahead production planning model; the day-ahead objective function consists of the scenario occurrence probability, the regulation interval of the first time scale, the cost of purchasing and selling electricity, the usage and maintenance cost of adjustable production equipment, and the regulation incentive cost of adjustable production equipment.

5. The dual-timescale control method considering the control potential of industrial users according to claim 4, characterized in that, The day-ahead constraints also include: day-ahead power balance constraints, equipment operating status and power correlation constraints, production task constraints, equipment minimum continuous operating time constraints, upper limit constraints on the number of equipment start-ups and shutdowns, constraints on the number of equipment start-ups and shutdowns during the same period, and production process coupling constraints.

6. The dual-timescale control method considering the control potential of industrial users according to claim 1, characterized in that, The steps for constructing an intraday control model for optimizing and regulating flexible production equipment with the goal of minimizing the total operating cost and power fluctuation of the industrial park include: constructing an intraday objective function using a scenario generation method with the goals of minimizing the total operating cost and power fluctuation of the industrial park; setting intraday constraints for the intraday objective function to obtain the intraday control model; the intraday objective function consists of the scenario occurrence probability, the control interval of the second time scale, the cost of purchasing and selling electricity, the usage and maintenance cost of flexible production equipment, the operating cost of energy storage equipment, and the power fluctuation penalty cost.

7. The dual-timescale control method considering the control potential of industrial users according to claim 6, characterized in that, The intraday constraints include: intraday power balance constraints, and energy storage device operation constraints including upper and lower limit constraints for charging and discharging power, upper and lower limit constraints for dynamic updates of energy storage capacity, charging and discharging efficiency constraints, and charging and discharging mutual exclusion logic constraints.

8. The dual-timescale control method considering the control potential of industrial users according to claim 1, characterized in that, In the step of constructing the day-ahead decision agent and the intraday decision agent based on the day-ahead production planning model and the intraday control model respectively, and then coupling and interacting to solve the problem, the coupling and interaction process is as follows: the reward function of the intraday decision agent is mapped on the day-ahead control time step based on the asynchronous transmission mechanism and then embedded into the reward function of the day-ahead decision agent, so that the day-ahead decision agent and the intraday decision agent can couple and interact.

9. The dual-timescale control method considering the control potential of industrial users according to claim 1, characterized in that, In the step of constructing the day-ahead decision agent and the intraday decision agent based on the day-ahead production planning model and the intraday control model respectively, and then coupling and interacting to solve them, the day-ahead decision agent and the intraday decision agent after coupling and interacting are trained and solved based on the asynchronous dual-interaction flexible actor critic algorithm architecture.

10. The dual-timescale control method considering the control potential of industrial users according to claim 9, characterized in that, The steps of training and solving the coupled interaction-based day-ahead and day-intraday decision agents using the flexible actor-critic algorithm architecture based on asynchronous dual interaction include: Initialize the network parameters and target network of the flexible actor-critic algorithm for the day-ahead decision agent and the intraday decision agent respectively, and create independent experience replay pools for each; The decision-making agent and the intraday decision-making agent interact with the environment based on their respective strategies and store the resulting interaction data into their respective experience pools. During the training loop, samples are taken from two experience pools to independently but synchronously update the parameters of the critic network, policy network, and target network of the day-ahead decision agent and the intraday decision agent. After training, the strategies derived by the daytime decision agent and the intraday decision agent are the dual-timescale collaborative control strategies for the industrial user.

11. The dual-timescale control method considering the control potential of industrial users according to claim 1, characterized in that, The step of constructing a day-ahead production planning model for optimizing and regulating the production plan of an industrial park, with the goal of minimizing the total operating cost of the industrial park under the constraint of satisfying the adjustable range of the industrial load potential, includes: constructing a day-ahead objective function by combining the partial blue bar optimization method with the goal of minimizing the total operating cost of the industrial park, and setting day-ahead constraints for the day-ahead objective function with the adjustable range of the industrial load potential as one of the constraints, thereby obtaining the day-ahead production planning model.

12. A dual-timescale control system considering the control potential of industrial users, used to execute the method according to any one of claims 1 to 11, characterized in that, include: The data acquisition module is used to acquire historical industrial load data and adjustable production equipment parameter information of users in the industrial park; The adjustable production equipment and related modeling module is used to establish an industrial load model for adjustable production equipment, establish the correlation between control incentives and adjustable production equipment based on the industrial load model, and construct a time series data input model for industrial user control potential. The regulation potential analysis module is used to input historical industrial load data into the industrial user regulation potential time series data input model. After obtaining the industrial user regulation potential time series input data, the regulation potential analysis is performed using the industrial load adjustable potential analysis model to obtain the adjustable range of industrial load potential. The control model construction module is used to construct a day-ahead production plan model for optimizing the control of the industrial park's production plan based on the adjustable production equipment parameter information, under the constraint of satisfying the adjustable range of the industrial load potential, with the goal of minimizing the total operating cost of the industrial park, and to construct an intraday control model for optimizing the control of the flexible adjustable production equipment with the goal of minimizing the total operating cost and power fluctuation of the industrial park. The dual-timescale control strategy intelligent generation module is used to construct day-ahead decision agents and intraday decision agents based on the day-ahead production planning model and the intraday control model, respectively, and then couple and interact to solve them to generate day-ahead-intraday coordinated control strategies.

Images