A method for improving the interaction capability of high-energy-consuming industrial users under carbon emission constraints

Abstract

This invention discloses a method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints. The method includes: real-time acquisition of dynamic power output characteristic data from new energy sources, capturing key dynamic features such as instantaneous power fluctuations, ramp rate changes, and spatiotemporal correlations of wind farms and photovoltaic power plants; constructing an adaptive generation model for multimodal risk scenarios, and using cluster analysis and stochastic process theory to summarize the complex stochastic fluctuations of new energy power output into several representative typical risk modes. The robustness-first rolling decision optimization mechanism of this invention prioritizes ensuring system power balance and safety under all generated risk scenarios in its core objective function and constraints. This fundamentally overcomes the decision failure risk caused by prediction bias in traditional methods, thus providing the power grid with excellent resilience and anti-interference capability under complex and uncertain operating conditions.

Description

Technical Field

[0001] This invention relates to the field of new power system operation and control, and in particular to a method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints. Background Technology

[0002] High-proportion renewable energy grid connection has become an inevitable trend in power system development. Against this backdrop, the ability of high-energy-consuming industrial users, who have significant regulation potential in the power grid and are major energy consumers and carbon emitters, to participate in grid interaction (i.e., "grid-load interaction") has become a key technological direction for balancing the volatility of renewable energy and ensuring the safe and stable operation of the power grid.

[0003] CN120016438A discloses a method and system for transferable load regulation applicable to high-energy-consuming industrial enterprises, comprising: acquiring load data of various equipment within the high-energy-consuming industrial enterprise; calculating the optimal load regulation strategy based on the load data and a pre-constructed high-energy-consuming industrial load scheduling model; and distributing the optimal load regulation strategy to various equipment within the high-energy-consuming industrial enterprise for load regulation; wherein the high-energy-consuming industrial load scheduling model is constructed based on the total electricity cost of industrial users as the objective function combined with constraints.

[0004] The above schemes are still static optimizations based on deterministic data. Their models rely solely on internal enterprise load data, completely neglecting the strong random impacts brought by external renewable energy grid connection, and lacking quantitative analysis and proactive response mechanisms for uncertainty. Their "transferable load regulation" only performs time-series optimization on known, fixed load data. Once external disturbances such as sudden changes in renewable energy output occur, the optimization results will immediately become invalid, resulting in weak system resilience. Furthermore, the comparative schemes only include the cost of carbon emission trading rights as an economic item in the total cost objective function. This pursuit of economic efficiency through the use of high-carbon power sources renders the carbon emission reduction target ineffective. Its constraints are limited to traditional physical boundaries such as unit output and ramp rate, failing to embed carbon emission intensity as a hard boundary constraint for real-time operation. Therefore, carbon-electricity synergy lacks practical real-time control guarantees. Summary of the Invention

[0005] The purpose of this invention is to provide a method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints. It can overcome the shortcomings of existing technologies and is a simple, easy-to-implement method for improving the grid-load interaction capability of high-energy-consuming industrial users. It can accurately capture the random fluctuation characteristics of new energy output, and fundamentally improve the grid's anti-interference capability in the face of uncertainty through multi-modal risk scenario coverage and robust rolling decision-making. At the same time, it uses carbon emission intensity as a hard constraint throughout the entire process of multi-timescale regulation.

[0006] The technical solution of this invention: A method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints, characterized by comprising the following steps: Step 1: Real-time collection of dynamic power output characteristic data of new energy sources to capture key dynamic characteristics such as instantaneous power fluctuations, ramp rate changes, and spatiotemporal correlations of wind farms and photovoltaic power stations.

[0007] By capturing key dynamic features such as instantaneous power fluctuations, ramp rate changes, and spatiotemporal correlations of wind farms and photovoltaic power plants, we provide accurate and comprehensive basic data support for the subsequent construction of adaptive generation models for multimodal risk scenarios. Based on these dynamic features, we use cluster analysis to summarize the complex random fluctuations of new energy output into typical risk modes such as power drop, rise, and oscillation. Then, we combine stochastic process theory to construct a multimodal risk scenario library covering various extreme and normal operating conditions.

[0008] The instantaneous power P(t) in step one represents the output power of the wind farm / photovoltaic power station at time t, and its mathematical expression is: (1) Where W(t) is the change of wind speed over time, and G(t) is the change of light intensity over time; The expression for instantaneous power fluctuation is: in, express The instantaneous power fluctuation at any given moment. , indicating that The time is the endpoint and the duration is The average power within the sliding time window, It can be set according to the fluctuation characteristics of new energy power output and the control requirements.

[0009] The ramp rate S(t) represents the rate of increase or decrease of power, and is defined as: (2) Where Δt is the sampling time interval; The expression for the change in the gradient rate is: in, represent The change in the rate of ascent at time t, S(t) is The basic ramp rate at any given time.

[0010] The spatiotemporal correlation is represented by the covariance matrix Cov(P1,P2), where P1 and P2 represent the instantaneous power output at different locations.

[0011] The instantaneous power outputs P1 and P2 can be the instantaneous power outputs at different wind turbine locations within a wind farm, or the instantaneous power outputs at different photovoltaic module array locations within a photovoltaic power station. They do not refer solely to photovoltaic or wind power, but rather are key data covering both types of new energy power stations. They indicate the differences in physical layout locations within the same new energy power station (wind farm or photovoltaic power station). By capturing the power output data at these different locations and combining the sampling time interval, the spatiotemporal correlation is quantified using a covariance matrix, which can comprehensively reflect the spatial distribution characteristics and dynamic correlation patterns of new energy power output.

[0012] Step 2: Construct an adaptive generation model for multimodal risk scenarios. This involves using cluster analysis to summarize the complex random fluctuations in new energy output into representative typical risk modes based on the dynamic feature data collected in Step 1, and then constructing multimodal risk scenarios using stochastic process theory. The second step, which involves classifying power fluctuation data using cluster analysis, includes: Suppose that the instantaneous power P(t) at different time points is obtained through step one. By calculating its variability and correlation, the power changes can be summarized into at least two representative risk modes. Using the clustering-based K-means algorithm, a loss function is defined. as follows: (3) in, C represents the number of clusters. i Indicates the first The set of sample points in each cluster, μ i It is the center point of this cluster. Represents instantaneous power P(t) and the center point The Euclidean distance between them; The representative risk modes can be obtained by formula (3), including the power drop mode, the power rise mode and the power oscillation mode.

[0013] The power sag mode, power surge mode, and power oscillation mode mentioned in step two specifically refer to: Using formula (3) as the loss function of the K-means algorithm, the instantaneous power data P(t) at different time points is minimized along with the values ​​of each cluster center point μ. iThe sum of squared Euclidean distances is used to cluster power fluctuation data with similar characteristics into one class, thereby achieving the alignment of the three modes. Specifically, when the loss function J reaches its minimum, P(t) with the characteristic of "rapid and significant decrease" will cluster into the same cluster, denoted as the power drop mode; P(t) with the characteristic of "rapid and significant increase" will form another cluster, denoted as the power rise mode; and P(t) with the characteristic of "repeated fluctuation within a certain range" will be classified into the third class, denoted as the power oscillation mode. The center point μ of each cluster... i This corresponds to the typical power characteristics of the mode. The three modes are a classification and summary of the complex random fluctuations in the output of new energy sources. The power drop mode corresponds to the fluctuation of rapid and significant power decline, the power rise mode corresponds to the fluctuation of rapid and significant power increase, and the power oscillation mode corresponds to the fluctuation of power that changes back and forth within a specific range. Furthermore, the extreme fluctuation scenario mentioned in the text, such as the peak power deviation exceeding 30% in 15 minutes, is also covered within the fluctuation characteristics of these typical modes.

[0014] The specific content of step two, which uses stochastic process theory to construct multimodal risk scenarios, includes: (2-1) Let the current system state be This indicates the current power state and system load state; The system in step (2-1) refers to the grid-load interaction and coordination system, which includes high-energy-consuming industrial users, new energy power plants and the grid load side.

[0015] (2-2) Introduce the state transition matrix T, and use the elements in the matrix to represent the transition probability from one risk mode to another, i.e., T ij This represents the transition probability from mode i to mode j, thereby simulating the transformation from a single mode to a dynamic scenario, in order to construct a multimodal risk scenario library; (2-3) Based on historical data and the frequency of occurrence of risk modes, the state transition matrix T is estimated by maximum likelihood estimation (MLE) to ensure that the transition of each mode conforms to the actual situation. The elements in the transition matrix represent the transition probability T. ij The calculation formula is: (4) Where I(·) is the indicator function, N is the total number of sample data, and C i and C j These represent the power fluctuation categories corresponding to mode i and mode j, respectively.

[0016] Step 3: Establish a robustness-first rolling decision optimization model. Using a rolling time-domain optimization strategy, within each decision cycle, based on the latest system state and the updated risk scenario library, solve for an optimal scheduling instruction that is well-adapted to all possible scenarios in the current and near-future periods. The specific implementation steps of step three include: (3-1) Construct the objective function of the optimization model: When setting the objective function, we must consider not only economic benefits but also ensure the steady state of the system under the worst-case risk scenario. Let P... net (t) represents the network load demand at time t, P gen (t) represents the renewable energy generation power at time t, P load If (t) represents the industrial load regulation power, then the objective function is expressed as follows: (5) Where α1 and α2 are weighting coefficients, P max It is the maximum adjustment range of industrial load. In the objective function, the first part α1(P) gen (t)-P load (t)-P net (t)) 2 It measures power balance, ensuring that the renewable energy generation power P at time t is [value missing]. gen (t) and industrial load regulation power P load (t) Network load demand P as close as possible to time t net (t), to meet the grid load demand; the second part α2max(0,P) load (t)-P max ) 2 The consideration is load safety constraints, when the industrial load regulating power P load (t) Exceeds the maximum adjustment range P max At that time, penalties will be imposed to ensure that industrial load adjustments do not exceed their physical limits; (3-2) Considering the uncertainty of the system, risk constraints are introduced to deal with the worst-case risk scenario. In step two, a risk scenario library based on cluster analysis and stochastic process theory has been constructed. Let S i (t) represents the power fluctuation situation under the i-th risk scenario, S worst This represents the power fluctuation scenario under the worst-case risk condition. The grid-load interaction and coordination system needs to ensure power balance and safety constraints under the worst-case scenario at every time t. The expression for the constraint condition is as follows: (6) Where δ is the tolerance error coefficient, representing the maximum fluctuation ratio that the power balance can tolerate under the worst scenario; In step (3-2), S i (t) represents the power fluctuation situation under the i-th risk scenario, serving as a specific representation of all typical risk modes in the risk scenario library. The power balance fluctuation range in the constraint conditions is limited by δ·P in formula (6). net (t) represents; S worst This represents the power fluctuation scenario under the worst-case risk scenario. As the most extreme risk scenario, the core objective of the constraint is to ensure power balance and safety under the worst-case scenario. This is achieved by using the upper limit control of the power deviation in formula (6) to realize S. worst The corresponding extreme power fluctuation amplitude, i.e., the S value quantified by the tolerance error coefficient δ, is... worst The maximum acceptable fluctuation ratio in the given scenario.

[0017] (3-3) In response to the dynamic changes in the output of new energy sources and the dynamic changes in the system status, a rolling mechanism that recalculates each decision cycle is used to solve for the optimal scheduling instruction that is well adapted to all possible scenarios. The optimization objective of step (3-1), the risk constraints of step (3-2), the updated multimodal risk scenario library of step two, and the real-time system status are combined to generate the optimal scheduling instruction that is adapted to the latest operating conditions of the grid-load interaction and coordination system. This avoids the failure of a single static decision due to changes in scenario or status. The optimal scheduling instruction is the direct input for step four to transform the robust optimization decision into specific load adjustment actions.

[0018] The optimal scheduling instruction in step (3-3) is expressed as: industrial load regulation power P in the objective function. load (t) represents the power adjustment required for high-energy-consuming industrial loads, such as electric arc furnaces and electrolytic cells, during a specific time period. It includes the specific power increase or decrease required, or the allowable power adjustment range, to ensure that the new energy power generation P at time t is consistent with the power generation at that time. gen (t) and network load demand P net (t) Collaboration to meet the goals of power balance and safety constraints in the worst-case scenario.

[0019] Step four: Implement load elastic control with carbon-electric coupling constraints. The optimal scheduling command generated in step three is sent to the high-energy-consuming industrial load through edge computing devices and control terminals to issue adjustment commands. At the same time, carbon emission intensity is embedded as a hard constraint into the control logic.

[0020] Step four refers to the layered implementation of control across multiple time scales to achieve precise adaptation between carbon-electric coupling constraints and load elastic adjustment. Specific implementation steps include: (4-1) Formulate monthly control strategies: Under the monthly regulation scale, the grid-load interaction and coordination system first needs to plan the overall carbon quota and load regulation strategy based on historical data and long-term system operation trends. The main objective is to ensure that the total regulation of all high-energy-consuming industrial users meets the grid's load demand throughout the month and does not exceed the monthly carbon emission limit. The objective function is set as follows: (7) Among them, P gen (t) represents the renewable energy generation power at time t, P load (t) represents the industrial load regulation power at time t, C load (t) represents the carbon emission intensity during load regulation, λ carbon As a weight for carbon emissions, and It is the adjustment coefficient; The historical data in step (4-1) includes historical data on grid load demand for past months, historical data on output of new energy sources (wind farms, photovoltaic power stations), historical data on load regulation of high-energy-consuming industrial users, and historical data on carbon emission intensity and total amount for past months; the long-term trend of system operation includes the long-term trend of grid load change, the long-term pattern of new energy output fluctuation, the long-term trend of production load of high-energy-consuming industrial users, and the long-term trend of carbon emission intensity control under the carbon emission reduction target.

[0021] (4-2) Optimization of intraday and daytime control measures: During weekday and daytime regulation processes, the distribution network load interaction and coordination system optimizes industrial load regulation strategies based on market forecasts, weather forecasts, and grid load demand. For this time scale, the objective function is rewritten as follows: (8); (4-3) During the intraday adjustment phase: During the intraday control phase, the distribution network-load interaction and coordination system requires refined scheduling optimization down to the hour or even minute level to ensure that load adjustments can meet real-time carbon emission reduction targets. The constraints at this point are: (9) Among them, P net (t) represents the grid demand, i.e., the network load demand at time t, and δ is the tolerance error coefficient; (4-4) Real-time response and edge computing control, namely edge computing-driven second-level real-time response control. When the distribution network load interaction and coordination system enters the real-time control stage, the fluctuations of grid load and new energy output will be more severe. Real-time collection and processing of power data from wind farms, photovoltaic power stations and industrial loads, and dynamic adjustment of load through local calculation; (4-5) Coordination of load elasticity control and carbon-electric coupling constraints, i.e., multi-timescale carbon-electric coupling coordinated control, to achieve interaction between grid load and high-energy-consuming industrial users.

[0022] The coordination of load elasticity control and carbon-electric coupling constraints in steps (4-5) refers to the effective coordination between load elasticity control and carbon-electric coupling constraints during scheduling across all time scales. This is achieved by deploying a two-layer nested real-time control algorithm in the edge computing device. Specifically, the inner algorithm of the edge computing transforms the carbon emission intensity constraint into a real-time permissible range for load regulation. That is, based on the current grid carbon intensity signal and the preset carbon quota, it dynamically calculates the maximum and minimum allowable regulation power for each high-energy-consuming industrial load unit at the current moment, forming the feasible domain boundary under carbon constraints. The outer algorithm then executes the scheduling instructions generated by robust optimization within this boundary. Through the Model Predictive Control (MPC) framework, it solves for the optimal load regulation amount in a rolling optimization manner, ensuring that the regulation action at each moment not only responds to the power balance requirements of the grid but also strictly satisfies the C... load (t)≤C max (t) is a hard carbon constraint, where C max (t) represents the dynamic carbon limit, C load (t) refers to the carbon emission intensity generated during the load adjustment process at time t, ultimately aiming to improve the grid-load interaction capability of high-energy-consuming industrial users.

[0023] The beneficial effects of the technical solution provided by this invention are: This invention replaces the traditional scheduling mode that relies on a single deterministic prediction trajectory by constructing a multimodal risk scenario adaptive generation model based on real-time dynamic characteristics. This model uses cluster analysis to summarize the complex random fluctuations in renewable energy output into typical risk modes such as power drops, surges, and oscillations. It also utilizes stochastic process theory to adaptively generate a scenario library covering various extreme operating conditions in the short term. Based on this, step three designs a robustness-first rolling decision optimization mechanism. Its core objective function and constraints prioritize ensuring system power balance and safety under all generated risk scenarios. This defensive benchmark, combined with a rolling time-domain implementation, allows scheduling commands to proactively adapt to the strong random fluctuations of renewable energy, fundamentally overcoming the decision failure risk caused by prediction bias in traditional methods. This provides the power grid with excellent resilience and anti-interference capabilities under complex and uncertain operating conditions.

[0024] This invention deeply embeds carbon emission intensity as a hard constraint into the entire process of load elastic control. By constructing a multi-timescale carbon-electric coupling control architecture of "monthly-weekly / daytime-intraday-real-time," the constraint or penalty term of load regulation carbon emission intensity is explicitly considered in the optimization objective function of each timescale. Especially in the most refined intraday and real-time control stages, the carbon emission intensity constraint is directly used as the boundary condition for algorithms such as model predictive control, achieving second-level closed-loop control through edge computing devices. This ensures that every regulation command issued to high-energy-consuming industrial loads strictly complies with preset low-carbon requirements while meeting the power balance needs of the power grid. It realizes real-time, dynamic, and precise coordination between the activation of load-side regulation potential and system-level carbon emission reduction targets, providing key technical support for the low-carbon operation of new power systems.

[0025] At the macro level, this invention provides industrial users with a clear low-carbon adjustment baseline through monthly carbon quota coordination. At the micro level, edge real-time control based on robust optimization decision-making enables precise, second-level regulation of high-energy-consuming equipment such as electric arc furnaces and electrolytic cells according to the real-time state of the power grid and carbon constraints. This not only ensures the operational safety of industrial loads during regulation but also directly contributes to the system's low-carbon balance through a carbon-electricity coupling mechanism. Ultimately, this method transforms high-energy-consuming industrial users from relatively passive and inefficient electricity consumers into high-quality, flexible resources that can be precisely called upon by the power grid, respond rapidly, and possess both safety and low-carbon attributes. This achieves a precise, safe, and low-carbon improvement in grid-load interaction capabilities overall. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall framework of a method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints, as described in this invention.

[0027] Figure 2 This is a schematic diagram of the robust priority rolling decision optimization mechanism of a method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints, as described in this invention.

[0028] Figure 3 This is a schematic diagram of real-time response and edge computing control for a method to improve the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints, as described in this invention.

[0029] Figure 4 This is a comparison chart of typical risk mode power fluctuations in a method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints, as described in this invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below.

[0031] To address the problems existing in the background technology, this invention, in the context of a high proportion of new energy integration in the new power system, proposes a systematic method to improve the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints, addressing the grid power balance problem caused by the inherent randomness and intermittency of wind and solar power output. This method aims to fundamentally enhance the robustness and anti-interference capability of grid dispatching decisions in the face of uncertain fluctuations in new energy sources by constructing a multimodal risk scenario adaptive generation model and a robustness-first rolling decision optimization mechanism. Simultaneously, the scheme embeds carbon emission intensity as a hard constraint into the entire process of load elastic control in real time, ensuring that while fully stimulating the adjustment potential of high-energy-consuming industrial loads and assisting the grid in absorbing new energy sources, the preset carbon emission reduction target is strictly achieved, ultimately realizing a precise, safe, and low-carbon improvement in grid-load interaction capability.

[0032] Example 1 like Figures 1-4 As shown, a method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints fundamentally enhances the grid's ability to withstand uncertainties by covering multi-modal risk scenarios and robust rolling decision-making. Simultaneously, carbon emission intensity is used as a hard constraint throughout the entire multi-timescale control process. Specifically, the method includes the following steps: Step 1: Real-time collection of dynamic power output characteristic data of new energy sources to capture key dynamic characteristics such as instantaneous power fluctuations, ramp rate changes, and spatiotemporal correlations of wind farms and photovoltaic power stations.

[0033] ① Instantaneous power P(t) represents the output power of a wind farm / photovoltaic power station at time t, and its mathematical expression is: (1) Where W(t) is the change of wind speed over time, and G(t) is the change of light intensity over time; The expression for instantaneous power fluctuation is: in, express The instantaneous power fluctuation at any given moment. , indicating that The time is the endpoint and the duration is The average power within the sliding time window, It can be set according to the fluctuation characteristics of new energy power output and the control requirements.

[0034] ② The rate of increase or decrease of power, S(t), represents the rate at which power increases or decreases, and is defined as: (2) Where Δt is the sampling time interval; The expression for the change in the gradient rate is: in, represent The change in the rate of ascent at time t, S(t) is The basic ramp rate at any given time.

[0035] ③ The spatiotemporal correlation is represented by the covariance matrix Cov(P1,P2), where P1 and P2 represent the instantaneous power output at different locations.

[0036] Step two involves constructing an adaptive generation model for multimodal risk scenarios. This involves using cluster analysis to summarize the complex random fluctuations in new energy output into representative typical risk modes based on the dynamic feature data collected in Step one. Then, through stochastic process theory, a multimodal risk scenario is constructed. Specifically, this includes: Suppose that the instantaneous power P(t) at different time points is obtained through step one. By calculating its variability and correlation, the power changes can be summarized into at least two representative risk modes. Using the clustering-based K-means algorithm, a loss function is defined. as follows: (3) in, C represents the number of clusters. i Indicates the first The set of sample points in each cluster, μ i It is the center point of this cluster. Represents instantaneous power P(t) and the center point The Euclidean distance between them; The representative risk modes can be obtained by formula (3), including the power drop mode, the power rise mode and the power oscillation mode.

[0037] Using formula (3) as the loss function of the K-means algorithm, the instantaneous power data P(t) at different time points is minimized along with the values ​​of each cluster center point μ. i The sum of squared Euclidean distances is used to cluster power fluctuation data with similar characteristics into one class, thereby achieving the alignment of the three modes. Specifically, when the loss function J reaches its minimum, P(t) with the characteristic of "rapid and significant decrease" will cluster into the same cluster, denoted as the power drop mode; P(t) with the characteristic of "rapid and significant increase" will form another cluster, denoted as the power rise mode; and P(t) with the characteristic of "repeated fluctuation within a certain range" will be classified into the third class, denoted as the power oscillation mode. The center point μ of each cluster... i This is a typical power characteristic representation of the corresponding mode.

[0038] The specific content of constructing multimodal risk scenarios using stochastic process theory includes: (2-1) Let the current system state be This indicates the current power state and system load state; The system in step (2-1) refers to the grid-load interaction and coordination system, which includes high-energy-consuming industrial users, new energy power plants and the grid load side.

[0039] (2-2) Introduce the state transition matrix T, and use the elements in the matrix to represent the transition probability from one risk mode to another, i.e., T ij This represents the transition probability from mode i to mode j, thereby simulating the transformation from a single mode to a dynamic scenario, in order to construct a multimodal risk scenario library; (2-3) Based on historical data and the frequency of occurrence of risk modes, the state transition matrix T is estimated by maximum likelihood estimation (MLE) to ensure that the transition of each mode conforms to the actual situation. The elements in the transition matrix represent the transition probability T. ij The calculation formula is: (4) Where I(·) is the indicator function, N is the total number of sample data, and C i and C j These represent the power fluctuation categories corresponding to mode i and mode j, respectively. like Figure 4 As shown, curve 1 corresponds to the power surge-gradual decline mode, such as when the power deviation peak exceeds 30% at 15 minutes, which is an extreme scenario where the output of new energy sources is significantly over-produced; curve 2 corresponds to the small oscillation mode; curve 3 corresponds to the medium fluctuation mode; and curve 4 corresponds to the deep drop mode. These curves cover power fluctuation risks of different intensities and intuitively present the differences in power fluctuations under different risk modes. Through the above steps, the model can adaptively generate a library of risk scenarios that may occur in the short term in the future based on the real-time system state. This library covers a variety of typical scenarios such as power drop, power surge, and power oscillation. This adaptive generation method based on multi-modal risk scenarios avoids the inherent limitations of accurately predicting a single future trajectory, provides a comprehensive risk reference for robust optimization, effectively improves the system's anti-interference ability and flexible response ability under complex and uncertain operating conditions, and provides sufficient data support for rolling decision optimization in subsequent steps. Step 3: Establish a robustness-first rolling decision optimization model. Using a rolling time-domain optimization strategy, within each decision cycle, based on the latest system state and the updated risk scenario library, solve for an optimal scheduling instruction that is well-adapted to all possible scenarios in the current and near-future periods. The specific implementation steps of step three include: (3-1) Construct the objective function of the optimization model: When setting the objective function, we must consider not only economic benefits but also ensure the steady state of the system under the worst-case risk scenario. Let P... net (t) represents the network load demand at time t, P gen (t) represents the renewable energy generation power at time t, P load If (t) represents the industrial load regulation power, then the objective function is expressed as follows: (5) Where α1 and α2 are weighting coefficients, P max It is the maximum adjustment range of industrial load. In the objective function, the first part α1(P) gen (t)-P load (t)-P net (t)) 2 It measures power balance, ensuring that the renewable energy generation power P at time t is [value missing]. gen (t) and industrial load regulation power P load (t) Network load demand P as close as possible to time t net (t), to meet the grid load demand; the second part α2max(0,P) load (t)-P max ) 2 The consideration is load safety constraints, when the industrial load regulating power P load (t) Exceeds the maximum adjustment range P max At that time, penalties will be imposed to ensure that industrial load adjustments do not exceed their physical limits; (3-2) Considering the uncertainty of the system, risk constraints are introduced to deal with the worst-case risk scenario. In step two, a risk scenario library based on cluster analysis and stochastic process theory has been constructed. Let S i (t) represents the power fluctuation situation under the i-th risk scenario, S worst This represents the power fluctuation scenario under the worst-case risk condition. The grid-load interaction and coordination system needs to ensure power balance and safety constraints under the worst-case scenario at every time t. The expression for the constraint condition is as follows: (6) Where δ is the tolerance error coefficient, representing the maximum fluctuation ratio that the power balance can tolerate under the worst scenario; In step (3-2), S i (t) represents the power fluctuation situation under the i-th risk scenario, serving as a specific representation of all typical risk modes in the risk scenario library. The power balance fluctuation range in the constraint conditions is limited by δ·P in formula (6). net (t) represents; S worstThis represents the power fluctuation scenario under the worst-case risk scenario. As the most extreme risk scenario, the core objective of the constraint is to ensure power balance and safety under the worst-case scenario. This is achieved by using the upper limit control of the power deviation in formula (6) to realize S. worst The corresponding extreme power fluctuation amplitude, i.e., the S value quantified by the tolerance error coefficient δ, is... worst The maximum acceptable fluctuation ratio in the given scenario; (3-3) In response to the fluctuations in new energy output and the dynamic changes in system status, a rolling mechanism that recalculates each decision cycle is used to combine the optimization objective of step (3-1), the risk constraints of step (3-2), the updated multimodal risk scenario library of step two, and the real-time system status to generate the optimal scheduling instruction that adapts to the latest operating conditions of the grid-load interaction and coordination system. This avoids the failure of a single static decision due to changes in scenario or status. The optimal scheduling instruction is the direct input for step four to transform the robust optimization decision into specific load adjustment actions.

[0040] The optimal scheduling instruction in step (3-3) is expressed as: industrial load regulation power P in the objective function. load (t) represents the power adjustment required for high-energy-consuming industrial loads, such as electric arc furnaces and electrolytic cells, during a specific time period. It includes the specific power increase or decrease required, or the allowable power adjustment range, to ensure that the new energy power generation P at time t is consistent with the power generation at that time. gen (t) and network load demand P net (t) Collaboration to meet the goals of power balance and safety constraints in the worst-case scenario.

[0041] Step four involves implementing load elastic control with carbon-electric coupling constraints. The robust optimization decision generated in step three is transformed into specific control actions. Adjustment commands are issued to high-energy-consuming industrial loads through edge computing devices and control terminals, while carbon emission intensity is embedded as a hard constraint into the control logic.

[0042] To achieve precise adaptation between carbon-electric coupling constraints and load elastic regulation, control is advanced in a layered manner across multiple time scales. Specific implementation steps include: (4-1) Formulate monthly control strategies: Under the monthly regulation scale, the grid-load interaction and coordination system first needs to plan the overall carbon quota and load regulation strategy based on historical data (including historical data of grid load demand in previous months, historical data of output from new energy sources (wind farms and photovoltaic power plants), historical data of load regulation by high-energy-consuming industrial users, and historical data of carbon emission intensity and total amount in previous months) and long-term trends in system operation (long-term trends in grid load, long-term patterns of new energy output fluctuations, long-term trends in production load of high-energy-consuming industrial users, and long-term trends in carbon emission intensity control under carbon reduction targets). The main objective is to ensure that the total regulation amount of all high-energy-consuming industrial users meets the grid load demand throughout the month and does not exceed the total monthly carbon emissions. The objective function is set as follows: (7) Among them, P gen (t) represents the renewable energy generation power at time t, P load (t) represents the industrial load regulation power at time t, C load (t) represents the carbon emission intensity during load regulation, λ carbon As a weight for carbon emissions, and It is the adjustment coefficient; (4-2) Optimization of intraday and daytime control measures: During weekday and daytime regulation processes, the distribution network load interaction and coordination system optimizes industrial load regulation strategies based on market forecasts, weather forecasts, and grid load demand. For this time scale, the objective function is rewritten as follows: (8); (4-3) During the intraday adjustment phase: During the intraday control phase, the distribution network-load interaction and coordination system requires refined scheduling optimization down to the hour or even minute level to ensure that load adjustments can meet real-time carbon emission reduction targets. The constraints at this point are: (9) Among them, P net (t) represents the grid demand, i.e., the network load demand at time t, and δ is the tolerance error coefficient; (4-4) Real-time response and edge computing control, namely edge computing-driven second-level real-time response control. When the distribution network load interaction and coordination system enters the real-time control stage, the fluctuations of grid load and new energy output will be more severe. Real-time collection and processing of power data from wind farms, photovoltaic power stations and industrial loads, and dynamic adjustment of load through local calculation; (4-5) Coordination of load elasticity control and carbon-electric coupling constraints, i.e., multi-timescale carbon-electric coupling coordinated control, to achieve interaction between grid load and high-energy-consuming industrial users.

[0043] The coordination between load resilience control and carbon-electric coupling constraints refers to the effective coordination between load resilience control and carbon-electric coupling constraints during scheduling across all time scales. This is achieved by deploying a two-layer nested real-time control algorithm in edge computing devices. Specifically, the inner algorithm of the edge computing transforms carbon emission intensity constraints into real-time permissible ranges for load regulation. Based on the current grid carbon intensity signal and preset carbon quotas, it dynamically calculates the maximum and minimum allowable regulation power for each high-energy-consuming industrial load unit at the current moment, forming the feasible domain boundary under carbon constraints. The outer algorithm then executes the scheduling instructions generated by robust optimization within this boundary. Using a Model Predictive Control (MPC) framework, it solves for the optimal load regulation amount through rolling optimization, ensuring that the regulation action at each moment not only responds to the grid's power balance requirements but also strictly satisfies the C... load (t)≤C max (t) is a hard carbon constraint, where C max (t) represents the dynamic carbon limit, C load (t) refers to the carbon emission intensity generated during the load adjustment process at time t, ultimately aiming to improve the grid-load interaction capability of high-energy-consuming industrial users.

[0044] Unless otherwise specified, the model numbers of the various devices in this embodiment of the invention are not limited, and any device that can perform the above functions is acceptable.

[0045] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints, characterized in that... It includes the following steps: Step 1: Real-time collection of dynamic power output characteristic data of new energy sources to capture key dynamic characteristics such as instantaneous power fluctuations, ramp rate changes, and spatiotemporal correlations of wind farms and photovoltaic power stations; Step 2: Construct an adaptive generation model for multimodal risk scenarios. This involves using cluster analysis to summarize the complex random fluctuations in new energy output into representative typical risk modes based on the dynamic feature data collected in Step 1, and then constructing multimodal risk scenarios using stochastic process theory. Step 3: Establish a robustness-first rolling decision optimization model. Using a rolling time-domain optimization strategy, within each decision cycle, based on the latest system state and the updated risk scenario library, solve for an optimal scheduling instruction that is well-adapted to all possible scenarios in the current and near-future periods. Step four: Implement load elastic control with carbon-electric coupling constraints. The optimal scheduling command generated in step three is sent to the high-energy-consuming industrial load through edge computing devices and control terminals to issue adjustment commands. At the same time, carbon emission intensity is embedded as a hard constraint into the control logic.

2. The method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints according to claim 1, characterized in that... The instantaneous power P(t) in step one represents the output power of the wind farm / photovoltaic power station at time t, and its mathematical expression is: （1） Where W(t) is the change of wind speed over time, and G(t) is the change of light intensity over time; The expression for instantaneous power fluctuation is: in, express The instantaneous power fluctuation at any given moment. , indicating that The time is the endpoint and the duration is The average power within the sliding time window, It can be set according to the fluctuation characteristics of new energy power output and the control requirements; The ramp rate S(t) represents the rate of increase or decrease of power, and is defined as: （2） Where Δt is the sampling time interval; The expression for the change in the gradient rate is: in, represent The change in the rate of ascent at time t, S(t) is The base ramp rate at any given moment; The spatiotemporal correlation is represented by the covariance matrix Cov(P1,P2), where P1 and P2 represent the instantaneous power output at different locations.

3. The method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints as described in claim 1, characterized in that... The second step, which involves classifying power fluctuation data using cluster analysis, includes: Suppose that the instantaneous power P(t) at different time points is obtained through step one. By calculating its variability and correlation, the power changes can be summarized into at least two representative risk modes. Using the clustering-based K-means algorithm, a loss function is defined. as follows: （3） in, C represents the number of clusters. i Indicates the first The set of sample points in each cluster, μ i It is the center point of this cluster. Represents instantaneous power P(t) and the center point The Euclidean distance between them; The representative risk modes can be obtained by formula (3), including the power drop mode, the power rise mode and the power oscillation mode.

4. The method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints according to claim 3, characterized in that... The power sag mode, power surge mode, and power oscillation mode mentioned in step two specifically refer to: Using formula (3) as the loss function of the K-means algorithm, the instantaneous power data P(t) at different time points is minimized along with the values ​​of each cluster center point μ. i The sum of squared Euclidean distances is used to cluster power fluctuation data with similar characteristics into one class, thereby achieving the alignment of the three modes. That is, when the loss function J reaches its minimum value, P(t) with the characteristic of "rapid and significant decrease" will cluster into the same cluster, denoted as the power drop mode; P(t) with the characteristic of "rapid and significant increase" will form another cluster, denoted as the power rise mode; and P(t) with the characteristic of "repeated fluctuation within a certain range" will be classified into the third class, denoted as the power oscillation mode. The center point μ of each cluster i This is a typical power characteristic representation of the corresponding mode.

5. The method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints according to claim 3, characterized in that... The specific content of step two, which uses stochastic process theory to construct multimodal risk scenarios, includes: (2-1) Let the current system state be This indicates the current power state and system load state; The system in step (2-1) refers to the grid-load interaction and coordination system, which includes high-energy-consuming industrial users, new energy power plants and the grid load side. (2-2) Introduce the state transition matrix T, and use the elements in the matrix to represent the transition probability from one risk mode to another, i.e., T ij This represents the transition probability from mode i to mode j, thereby simulating the transformation from a single mode to a dynamic scenario, in order to construct a multimodal risk scenario library; (2-3) Based on historical data and the frequency of occurrence of risk modes, estimate the state transition matrix T using maximum likelihood estimation, so that the transition of each mode conforms to the actual situation. The elements in the transition matrix represent the transition probability T. ij The calculation formula is: （4） Where I(·) is the indicator function, N is the total number of sample data, and C i and C j These represent the power fluctuation categories corresponding to mode i and mode j, respectively.

6. The method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints according to claim 1, characterized in that... The specific implementation steps of step three include: (3-1) Construct the objective function of the optimization model: Let P net (t) represents the network load demand at time t, P gen (t) represents the renewable energy generation power at time t, P load If (t) represents the industrial load regulation power, then the objective function is expressed as follows: （5） Where α1 and α2 are weighting coefficients, P max It is the maximum adjustment range of industrial load. In the objective function, the first part α1(P) gen (t)-P load (t)-P net (t)) 2 It measures power balance, ensuring that the renewable energy generation power P at time t is [value missing]. gen (t) and industrial load regulation power P load (t) Network load demand P as close as possible to time t net (t), to meet the grid load demand; the second part α2max(0,P) load (t)-P max ) 2 The consideration is load safety constraints, when the industrial load regulating power P load (t) Exceeds the maximum adjustment range P max At that time, penalties will be imposed to ensure that industrial load adjustments do not exceed their physical limits; (3-2) Let S i (t) represents the power fluctuation situation under the i-th risk scenario, S worst This represents the power fluctuation scenario under the worst-case risk condition. The grid-load interaction and coordination system needs to ensure power balance and safety constraints under the worst-case scenario at every time t. The expression for the constraint condition is as follows: （6） Where δ is the tolerance error coefficient, representing the maximum fluctuation ratio that the power balance can tolerate under the worst scenario; (3-3) In view of the dynamic changes in the output of new energy and the dynamic changes in the system state, a rolling mechanism of recalculation in each decision cycle is used to solve the optimal scheduling instruction that has good adaptability to all possible scenarios. The optimization objective of step (3-1), the risk constraints of step (3-2), the updated multimodal risk scenario library and real-time system state in step 2 are combined to generate the optimal scheduling instruction that adapts to the latest working conditions of the grid-load interaction and coordination system, so as to avoid the failure of a single static decision due to changes in scenario or state.

7. The method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints as described in claim 6, characterized in that... The optimal scheduling instruction in step (3-3) is expressed as: industrial load regulation power P in the objective function. load (t) represents the power adjustment required for high-energy-consuming industrial loads, such as electric arc furnaces and electrolytic cells, during a specific time period. It includes the specific power increase or decrease required, or the allowable power adjustment range, to ensure that the new energy power generation P at time t is consistent with the power generation at that time. gen (t) and network load demand P net (t) Collaboration to meet the goals of power balance and safety constraints in the worst-case scenario.

8. The method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints according to claim 1, characterized in that... Step four refers to the layered implementation of control across multiple time scales to achieve precise adaptation between carbon-electric coupling constraints and load elastic adjustment. Specific implementation steps include: (4-1) Formulate monthly control strategies: Under the monthly regulation scale, the grid-load interaction and coordination system first needs to plan the overall carbon quota and load regulation strategy based on historical data and long-term system operation trends. The main objective is to ensure that the total regulation of all high-energy-consuming industrial users meets the grid's load demand throughout the month and does not exceed the monthly carbon emission limit. The objective function is set as follows: （7） Among them, P gen (t) represents the renewable energy generation power at time t, P load (t) represents the industrial load regulation power at time t, C load (t) represents the carbon emission intensity during load regulation, λ carbon As a weight for carbon emissions, and It is the adjustment coefficient; (4-2) Optimization of intraday and daytime control measures: During weekday and daytime regulation processes, the distribution network load interaction and coordination system optimizes industrial load regulation strategies based on market forecasts, weather forecasts, and grid load demand. For this time scale, the objective function is rewritten as follows: （8）； (4-3) During the intraday adjustment phase: During the intraday control phase, the distribution network-load interaction and coordination system requires refined scheduling optimization down to the hour or even minute level to ensure that load adjustments can meet real-time carbon emission reduction targets. The constraints at this point are: （9） Among them, P net (t) represents the grid demand, i.e., the network load demand at time t, and δ is the tolerance error coefficient; (4-4) Real-time response and edge computing control, namely edge computing-driven second-level real-time response control. When the distribution network load interaction and coordination system enters the real-time control stage, the fluctuations of grid load and new energy output will be more severe. Real-time collection and processing of power data from wind farms, photovoltaic power stations and industrial loads, and dynamic adjustment of load through local calculation; (4-5) Coordination of load elasticity control and carbon-electric coupling constraints, i.e., multi-timescale carbon-electric coupling coordinated control, to achieve interaction between grid load and high-energy-consuming industrial users.

9. The method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints as described in claim 8, characterized in that... The historical data in step (4-1) includes historical data on grid load demand, renewable energy output, load regulation of high-energy-consuming industrial users, and carbon emission intensity and total amount for each month. The long-term trend of system operation includes the long-term trend of grid load, the long-term pattern of renewable energy output fluctuations, the long-term trend of production load of high-energy-consuming industrial users, and the long-term trend of carbon emission intensity control under the carbon emission reduction target.

10. The method for improving the grid-load interaction capability of high-energy-consuming industrial users under carbon emission constraints as described in claim 8, characterized in that... The coordination of load elasticity control and carbon-electric coupling constraints in steps (4-5) refers to the effective coordination between load elasticity control and carbon-electric coupling constraints during scheduling across all time scales. This is achieved by deploying a two-layer nested real-time control algorithm in edge computing devices. Specifically, the inner algorithm of the edge computing transforms the carbon emission intensity constraint into a real-time permissible range for load regulation. That is, based on the current grid carbon intensity signal and the preset carbon quota, it dynamically calculates the maximum and minimum allowable regulation power for each high-energy-consuming industrial load unit at the current moment, forming the feasible domain boundary under carbon constraints. The outer algorithm then executes the scheduling instructions generated by robust optimization within this boundary. Through a model predictive control framework, it solves for the optimal load regulation amount using rolling optimization, ensuring that the regulation action at each moment not only responds to the power balance requirements of the grid but also strictly satisfies the C... load (t)≤C max (t) is a hard carbon constraint, where C max (t) represents the dynamic carbon limit, C load (t) refers to the carbon emission intensity generated during the load adjustment process at time t, ultimately aiming to improve the grid-load interaction capability of high-energy-consuming industrial users.

Images