Load self-adaptive adjustment method for high-proportion new energy power grid based on double-layer optimization model

Abstract

This invention belongs to the field of power system operation and optimization technology, and particularly relates to an adaptive load adjustment method for high-proportion renewable energy power grids based on a two-layer optimization model. Addressing the increased uncertainty in power grid supply and demand under extreme summer and winter weather conditions, this invention achieves precise load control through a six-step core process: constructing a source-load physical characteristic model to quantify the impact of extreme weather; employing the VMD-LSTM residual correction algorithm and improved clustering technology to complete source-load prediction and multi-scenario quantification; establishing a five-segment dynamic time-of-use pricing and comprehensive compensation equivalent system; combining the Sigmoid function to characterize user response intentions; activating the adjustment mechanism through extreme weather perception; constructing a two-layer optimization model of "system economy-power supply quality"; and finally forming a closed loop through command issuance and feedback calibration. This method effectively improves the accuracy of source-load prediction, stimulates user response potential, reduces system operating costs, improves voltage stability and load fluctuation levels, and significantly enhances the resilience and reliability of high-proportion renewable energy power grids under extreme conditions.

Description

Technical Field

[0001] This invention belongs to the field of power system operation and optimization technology, and particularly relates to a method for adaptive adjustment of high-proportion renewable energy grid load based on a two-level optimization model. Background Technology

[0002] With the accelerated global transition to green and low-carbon energy, the installed capacity of new energy sources such as wind power and photovoltaics has continued to grow. As of April 2025, the total installed capacity of wind and solar power had surpassed that of thermal power, and the power system has entered a new stage of "high proportion of renewable energy + high proportion of power electronic equipment". However, the strong randomness, volatility, and intermittency of new energy power generation, coupled with the frequent occurrence of extreme weather (such as high temperatures and droughts, cold waves and freezing), has led to increased uncertainty on both the supply and demand sides of the power grid, highlighting system security risks. For example, the power shortage caused by high temperatures in Sichuan in 2022 and the large-scale power outage caused by a cold wave in Texas, USA, both exposed the vulnerability of power grids with a high proportion of renewable energy to extreme weather impacts—the sudden drop in renewable energy output and the surge in load created dual pressures that traditional dispatching modes could not cope with.

[0003] Current research has made progress in grid-side risk assessment and operation optimization: a grid security assessment system centered on voltage and power flow exceedance has been constructed; a method for assessing power adequacy under extreme high temperatures has been proposed; and technologies such as deep active learning and conditional generative adversarial networks have been used to achieve rapid assessment and scenario generation of supply-demand imbalance risks. In terms of demand-side management, existing research mainly focuses on electricity price guidance and resource coordination, such as designing load adjustment strategies through real-time electricity pricing, microgrid optimization, and elastic matrix models, and exploring the synergistic mechanism between energy storage and demand response.

[0004] However, existing load adjustment strategies still have significant limitations under extreme weather conditions: First, they lack dynamic modeling of the thermal inertia and saturation effect of temperature-controlled loads (such as air conditioning and electric heating), leading to a large discrepancy between load forecasts and actual response capabilities. Second, incentive mechanisms mostly adopt "one-way command" or fixed compensation models, failing to precisely quantify the revenue losses (such as raw material scrapping and equipment restart costs) incurred by users in different industries due to production interruptions, making it difficult to balance user interests with system adjustment needs, resulting in insufficient response rates of adjustment resources under extreme scenarios. Therefore, there is an urgent need for a load adaptive adjustment method that can dynamically adapt to extreme conditions and accurately incentivize user participation to improve the resilience and reliability of high-proportion renewable energy power grids. Summary of the Invention

[0005] To address the problems of delayed temperature-controlled load forecasting, inaccurate user incentive mechanisms, and difficulty in balancing the interests of the system and users in traditional load adjustment strategies for high-proportion renewable energy power grids under extreme weather conditions, this invention aims to provide an adaptive adjustment scheme. By accurately characterizing the uncertainty of source and load under extreme weather conditions, a scientific incentive and response model is constructed, balancing the minimization of total system operating costs with the optimization of power supply quality. This solves the pain point of insufficient response rate of load adjustment resources under extreme scenarios, realizes dynamic coordination of source, grid, load, and storage resources, enhances the power grid's ability to cope with extreme weather impacts, and ensures the safe, stable, and economical operation of the power system.

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

[0007] A method for adaptive load adjustment of high-proportion renewable energy power grids based on a two-level optimization model includes the following steps:

[0008] S1 Extreme Weather Source-Load Uncertainty Modeling: For extreme weather in summer and winter, a source-load physical characteristic model is constructed, including a wind turbine output model that takes into account high temperature attenuation, a photovoltaic unit output model that takes into account snow shading, and a temperature control load model for summer and winter.

[0009] S2 Source Load Prediction and Multi-Scenario Quantization: Based on the S1 physical model, the baseline value of the source output is determined, and the VMD-LSTM residual correction prediction algorithm is used to obtain the predicted value of the source output. A dynamic load prediction model considering temperature control inertia is constructed to solve the problem of load prediction lag under extreme weather conditions. Then, LHS is used to generate a massive initial scenario set. By improving the Canopy-spectral clustering-K-means hybrid clustering algorithm, scenarios are reduced, typical scenarios and their corresponding occurrence probabilities are extracted, and the uncertainty boundary of source load under extreme weather conditions is quantified.

[0010] S3 Comprehensive Compensation System and User Response Model Construction: Construct a five-segment dynamic time-of-use electricity price model and a comprehensive compensation equivalent system, classify users into continuous production and discrete manufacturing users, construct revenue loss models for each typical industry, and construct a user response willingness model based on the Sigmoid function to provide incentives and response basis for load adjustment;

[0011] S4 Extreme Weather Sensing and Status Assessment: Real-time monitoring of meteorological parameters, activation of adjustment mechanism when thresholds are continuously triggered, and simultaneous real-time push of weather warnings, electricity price signals and system adjustment needs to users to guide users to adjust their electricity consumption in advance;

[0012] S5 Two-Layer Optimization Model Construction: A two-layer optimization model is constructed. The upper-layer model aims to minimize the total system operating cost within the scheduling cycle, while the lower-layer model aims to minimize load fluctuations and voltage deviations and improve power supply quality.

[0013] S6 Command Issuance and Feedback Calibration: The commands obtained from solving the two-layer optimization model are issued to each execution unit, and the deviation between the actual execution data and the target value is compared to calibrate the parameters of each model; a real-time settlement mechanism for evaluating and compensating user response effects is established to improve user response enthusiasm.

[0014] In S1:

[0015] Wind turbine output modeling under high temperature environment: Based on the physical characteristics of reduced air density and easy high temperature protection shutdown of wind turbines caused by high temperature in summer, the model is divided into the range of wind turbine cut-in, rated and cut-out wind speeds. An air density correction coefficient and rated power reduction coefficient related to ambient temperature are introduced into the traditional wind turbine output model. At the same time, temperature cut-out logic is added to quantify the changes in wind turbine output caused by the decrease in wind energy capture efficiency and equipment protection shutdown due to high temperature.

[0016] Modeling the output of photovoltaic units under snow and cold waves: In response to the dual impacts of irradiance attenuation caused by winter cold waves and snow and snow shading of photovoltaic modules, the rated output under standard photovoltaic test conditions is used as the basis. Temperature correction is performed by combining the power temperature characteristics of the modules. At the same time, a snow shading loss coefficient that is non-linearly related to snow thickness and snow melting rate is introduced, and parameters such as inverter efficiency are included to correct the actual photovoltaic output and realize the quantification of photovoltaic output attenuation under the dual impacts.

[0017] Summer and winter temperature control load modeling: Both are based on the basic principles of thermodynamics, and respectively construct summer air conditioning cooling and winter electric heating load models. The models include the deviation between ambient temperature and equipment set temperature, equipment energy efficiency ratio, air and building physical parameters and user-related indicators. By quantifying the temperature deviation and the correlation between various physical and equipment parameters, the models can accurately calculate the sharp increase in cooling or heating load caused by the increase in temperature deviation under extreme weather conditions.

[0018] In S2:

[0019] VMD-LSTM residual correction for source-end power output prediction: First, the theoretical baseline value of power output is calculated using a source-end physical model under extreme weather conditions, and the residual sequence between the theoretical value and the historical measured value is extracted. Then, variational mode decomposition (VMD) is used to decompose the non-stationary, multi-scale residual sequence into multiple intrinsic mode components. The component parameters are iteratively updated by solving variational constraint problems to avoid the mode aliasing problem of traditional decomposition. Subsequently, a long short-term memory (LSTM) network model is constructed for each component, and long-term temporal dependencies are captured by combining extreme weather feature vectors. The predicted components are then superimposed and reconstructed to compensate the residuals to the physical baseline value, thus obtaining the final predicted value of source-end power output.

[0020] Dynamic load prediction considering temperature control inertia: Based on the benchmark load, the temperature control component reflecting the building's cumulative heat effect is coupled to obtain the total load prediction value; the cumulative heat effect temperature is introduced to characterize the building's thermal energy storage state, and the current ambient temperature is coupled with the previous thermal state through a thermal inertia weighting coefficient to quantify the building's sensitivity to changes in ambient temperature; this solves the problem that traditional models cannot capture the hysteresis and inertia of temperature control loads under extreme temperatures, as well as the prediction lag for nonlinear load increases;

[0021] Latin Hypercube Probability Sampling (LHS) generates the initial scene set: To meet the need for capturing extremely low-probability events, the probability distribution of prediction errors is sampled in layers; the probability interval is divided into equally probable sub-intervals, and samples are independently and randomly drawn from each interval and obtained by inverse transformation of the cumulative distribution function to ensure uniform coverage of the samples in the full probability space, and finally a massive initial scene set containing the temporal correlation of wind, light, and load is generated.

[0022] An improved Canopy-spectral clustering-K-means hybrid algorithm for scene reduction: A three-step clustering strategy is employed to achieve efficient and reasonable scene reduction while preserving key features.

[0023] ①Canopy coarse clustering quickly divides scene subsets through dual distance thresholds, objectively determines the optimal number of clusters, and overcomes the subjectivity of traditional algorithms that preset the number of clusters;

[0024] ② Spectral clustering nonlinear dimensionality reduction: By constructing a Laplace matrix, high-dimensional time-series scene data is mapped to a low-dimensional feature space, which can identify complex data structures and retain the nonlinear and time-series correlation characteristics of source load fluctuations under extreme weather conditions.

[0025] ③ K-means fine partitioning: using the low-dimensional feature matrix as input and the number of clusters determined by Canopy as the basis, iterative optimization is performed to minimize the intra-cluster error, and finally a representative set of typical scenarios is extracted and the probability of occurrence of each scenario is determined.

[0026] In S3:

[0027] Five-segment dynamic time-of-use pricing model: A five-segment dynamic time-of-use pricing model is constructed, consisting of peak-peak-flat-valley-deep valley. The electricity price elasticity matrix quantifies the cross-time-shifting effect of electricity prices on load at different times. At the same time, the revenue equivalent of time-of-use pricing is introduced to aggregate the cross-time-shifting electricity price difference revenue generated by users' peak shifting and valley filling, calculate the electricity cost savings per unit of response electricity, and transform the price signal into an intuitive economic incentive per unit of electricity, which serves as the basic value component of the comprehensive incentive system.

[0028] Comprehensive compensation equivalent system:

[0029] Construct a demand response compensation equivalent, which is divided into two dimensions: electrical energy and reserve compensation. Based on parameters such as subsidy standards, effect judgment coefficients, and total monthly reserve costs, the total economic compensation for demand response per unit of electricity is quantified.

[0030] To address the diverse needs of load-side resources participating in system regulation, three types of ancillary service compensation equivalents are constructed: frequency regulation, reserve, and ramping. Based on the core impact parameters of each service, the corresponding service compensation amount per unit of response power is calculated, thereby achieving precise quantification of the value of users participating in different forms of system regulation.

[0031] By organically integrating the time-of-use electricity price revenue equivalent, demand response compensation equivalent, and compensation equivalent of various ancillary services, a comprehensive compensation equivalent on the user side is constructed. This enables a unified quantification and aggregation of the price revenue, demand response value, and system ancillary service value created by users' participation in load adjustment, forming a clear total economic incentive signal for unit electricity adjustment and providing a unified revenue reference for user decision-making.

[0032] Typical industry revenue loss models: Industrial users are divided into continuous production and discrete manufacturing based on their production characteristics. A refined unit power reduction production revenue loss model is constructed for the two types of users based on the differences in production rigidity or elasticity: for continuous production users, the model focuses on the chain losses caused by the rigidity of the production process; for discrete manufacturing users, the model focuses on the efficiency and opportunity losses caused by the disruption of the production rhythm. At the same time, a typical industry loss parameter library is established, and the parameters are continuously iterated and calibrated through actual response data to improve the accuracy of cost quantification.

[0033] User Response Willingness Model: A game-theoretic collaborative participation rate model is constructed using the Sigmoid function. The actual user participation rate is defined as a nonlinear function of "comprehensive compensation equivalent / unit revenue loss". A subsidy benefit sensitivity coefficient and a participation willingness threshold are introduced to quantify the degree of user participation under different benefit-cost ratios. This enables an accurate characterization of the willingness to participate in load adjustment for different industries and user groups, providing a basis for optimizing and adjusting the incentive mechanism.

[0034] In S4:

[0035] By monitoring key meteorological parameters such as wind speed, irradiance, ambient temperature, and snow depth in real time, an adjustment mechanism is activated when thresholds are continuously triggered. This dynamically corrects future wind and solar power output and load forecast curves, accurately identifying risks such as system power deficits, node voltage exceedances, and line overloads. Simultaneously, the power and electricity price elasticity matrix is ​​updated to be applicable to the current scenario. In addition to user response potential assessment parameters, the system simultaneously pushes real-time weather warnings, electricity price signals, and system adjustment needs to users, enabling them to adjust their electricity consumption behavior in advance based on their own electricity consumption flexibility and production plans, thereby improving the initiative and accuracy of response and providing accurate data input for optimized decision-making.

[0036] In S5:

[0037] Upper-level model: The core optimization objective is to minimize the total operating cost of the system within the scheduling cycle. The objective function comprehensively incorporates the four core cost items of system operation under extreme weather conditions and introduces weight coefficients to quantify the impact of environmental protection costs. Specifically, it covers the total cost of various generating units in time period t, the cost of thermal power pollutant treatment, the total demand response compensation cost, and the cost of external power purchase. At the same time, through the weight coefficient of pollutant treatment cost, the weight of thermal power environmental protection treatment cost in the total operating cost is clarified, so as to achieve a balance between the economic operation of the system and environmental protection requirements, and to define the economic constraint boundary of load adjustment at the system level.

[0038] Lower-level model: Within the framework of upper-level system economic optimization, the core optimization objective is to minimize power quality-related indicators such as load fluctuation and voltage deviation, while also considering the accuracy of demand response execution. A multi-indicator comprehensive optimization objective function is constructed: comprehensively considering three core power quality indicators—node net load fluctuation, node voltage deviation, and tie-line power fluctuation—and also incorporating the following error between the expected and actual adjustment amounts in demand response to ensure that the demand response execution effect matches the system scheduling expectations; normalized weight coefficients are assigned to each indicator to achieve balanced optimization of multiple power quality indicators, ensuring power supply stability during load adjustment from the user side and avoiding damage to the user's electricity experience due to system-side economic optimization.

[0039] In S6:

[0040] The instructions from the two-layer optimization model are sent to each execution unit to synchronously track the core operating parameters of key users; the deviations of key indicators such as voltage fluctuation rate, tie line power, and actual user participation rate from the optimized values ​​are calculated, and the parameters of each model are dynamically calibrated based on feedback. A real-time settlement mechanism for evaluating user response effects and compensation is established, and users can easily view their own response volume, compensation benefits, and contribution to system regulation.

[0041] The VMD-LSTM residual correction prediction algorithm is as follows:

[0042] The theoretical benchmark value of source-end output was calculated based on the S1 physical model. And extract its values ​​and historical measured values. residual sequence Given the non-stationary, multi-scale time-frequency characteristics of residual sequences under extreme weather conditions, variational mode decomposition (VMD) is employed to decompose them into... Each intrinsic mode component (IMF) has a different center frequency;

[0043] VMD transforms the signal decomposition process into finding the optimal center frequency of each mode by solving variational constraint problems. The objective function for the bandwidth process is constructed as follows:

[0044]

[0045]

[0046] In the formula, For the first One modal component, Let's define the Dirac function. Introduce Lagrange multipliers. and secondary penalty factor Iterative updates using the alternating direction multiplier method and The process continues until convergence, effectively avoiding the mode aliasing problem in EMD decomposition and accurately extracting high-frequency random components and low-frequency trend components from the residual sequence.

[0047] For each decomposed IMF component, a Long Short-Term Memory (LSTM) prediction model is constructed; the LSTM introduces a forgetting gate. Input gate and output gate The gating mechanism overcomes the gradient vanishing problem of traditional RNNs and effectively captures extreme weather characteristics. The long-term time-series dependency between the predictions and the residual components; after the predicted values ​​of each component are superimposed and reconstructed, the corrected final predicted output is obtained:

[0048]

[0049] In the formula, This is the input extreme weather feature vector.

[0050] The two-layer optimization model includes:

[0051] Upper-level model:

[0052] The objective is to minimize the total system operating cost within the scheduling cycle.

[0053]

[0054] In the formula, , , They represent Total cost of various types of generator units during the period, cost of treating pollutants from thermal power plants, total demand response compensation cost, and cost of purchasing external power; This represents the weighting coefficient for pollutant treatment costs;

[0055] Lower-level model:

[0056] Aimed at minimizing load fluctuations and voltage deviations to improve power quality:

[0057]

[0058] In the formula, , , They represent Time period node i net load fluctuation, node Voltage deviation, tie line Power fluctuations; and They represent Expected and actual adjustments to demand response during different time periods; Let represent the weighting coefficients, and .

[0059] A system for implementing the high-proportion renewable energy grid load adaptive adjustment method based on a two-level optimization model includes:

[0060] Modeling module: Constructs source-load physical characteristic models for extreme summer and winter weather, specifically including a wind turbine output modeling unit that takes into account high temperature attenuation, a photovoltaic unit output modeling unit that takes into account snow shading, and a temperature control load modeling unit for summer and winter seasons, quantifies the physical impact of extreme weather on source-load output, and outputs source-load physical characteristic parameters.

[0061] Predictive clustering module: Based on the source output baseline value output by the modeling module, the source output is predicted through the VMD-LSTM residual correction prediction unit. The dynamic load prediction unit solves the load prediction lag problem under extreme weather conditions. Then, through the scenario generation and reduction unit, LHS sampling and improved Canopy-spectral clustering-K-means hybrid clustering algorithm are used to extract typical scenarios and their occurrence probabilities.

[0062] Incentive Response Module: Constructs a five-segment dynamic time-of-use electricity price calculation unit and a comprehensive compensation equivalent calculation unit, classifies two types of industrial users, establishes revenue loss models for each typical industry, constructs a user response willingness unit through the Sigmoid function, and outputs load adjustment incentive signals and user response benchmark data;

[0063] Perception and Assessment Module: Accesses real-time meteorological monitoring data, sets a threshold judgment unit, activates the adjustment mechanism when the threshold is continuously triggered, and constructs a signal push unit to simultaneously push meteorological warnings, electricity prices and regulation demand signals;

[0064] Optimization Solver Module: Constructs a two-layer optimization solver. The upper layer aims to minimize the total operating cost of the system, while the lower layer aims to minimize load fluctuations and voltage deviations. It embeds corresponding constraints and solution algorithms and outputs load adjustment commands.

[0065] Command feedback module: issues adjustment commands to each execution unit, sets up deviation comparison and calibration unit, corrects model parameters, and constructs response evaluation and compensation settlement unit to improve user response enthusiasm and ensure adjustment accuracy.

[0066] The present invention has the following beneficial effects:

[0067] (1) In terms of source-load uncertainty perception and scenario generation, the joint prediction framework based on physical mechanism correction and VMD-LSTM residual compensation, combined with Latin hypercube sampling and improved Canopy-spectral clustering-K-means scenario reduction method, can effectively characterize the fluctuation characteristics of new energy output and load under extreme weather conditions, generate a typical risk scenario set covering normal and extreme working conditions, and provide reliable uncertainty input for optimization decision-making.

[0068] (2) In terms of user response mechanism design, the sigmoid function-based willingness-driven model, combined with a multi-type auxiliary service compensation equivalent system and a refined industry revenue loss function, achieves accurate quantification of industrial users' response costs and incentive willingness. This mechanism can significantly improve users' response participation rate and adjustment effect under extreme weather conditions, realizing a dual incentive of "price signal + willingness-driven".

[0069] (3) In terms of system optimization and adaptive adjustment, the constructed two-layer optimization model that takes into account both economic operation and power supply quality, combined with the "perception-evaluation-optimization-feedback" closed-loop mechanism, can dynamically coordinate source, grid, load and storage resources under extreme weather conditions. This strategy can effectively reduce system operating costs and external power dependence, while significantly improving net load fluctuations, voltage quality and tie-line power stability, thus enhancing grid resilience. Attached Figure Description

[0070] Figure 1 This is a flowchart of the adaptive adjustment method for high-proportion renewable energy grid load based on a two-layer optimization model proposed in this invention.

[0071] Figure 2 This is a flowchart of the residual correction prediction process based on VMD-LSTM for the high-proportion renewable energy grid load adaptive adjustment method based on a two-layer optimization model proposed in this invention.

[0072] Figure 3 This is a flowchart illustrating the wind and solar scenario analysis based on improved clustering reduction of the adaptive adjustment method for high-proportion renewable energy grid load proposed in this invention, which is based on a two-layer optimization model.

[0073] Figure 4 This is a flowchart illustrating the adaptive adjustment closed-loop execution and feedback process of the high-proportion renewable energy grid load adaptive adjustment method based on a two-layer optimization model proposed in this invention.

[0074] Figure 5 This paper compares the system operating cost indicators of the high-proportion renewable energy grid load adaptive adjustment method based on the two-layer optimization model proposed in this invention in the summer scenario.

[0075] Figure 6 This invention presents a comparison of user-side power quality in summer scenarios using the adaptive adjustment method for high-proportion renewable energy grid load based on a two-layer optimization model.

[0076] Figure 7 This paper compares the performance of the user-side load adjustment method for the high-proportion renewable energy power grid load adaptive adjustment method based on the two-layer optimization model proposed in this invention in the summer scenario.

[0077] Figure 8 This paper compares the system operating cost indicators of the high-proportion renewable energy grid load adaptive adjustment method based on the two-layer optimization model proposed in this invention in the winter scenario.

[0078] Figure 9 This paper compares the user-side power supply quality indicators of the high-proportion renewable energy grid load adaptive adjustment method based on the two-layer optimization model proposed in this invention in winter scenarios.

[0079] Figure 10 This paper compares the performance of the user-side load adjustment method for the high-proportion renewable energy power grid load adaptive adjustment method based on the two-layer optimization model proposed in this invention in the winter scenario. Detailed Implementation

[0080] The present invention will be further described in detail below with reference to specific embodiments. 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.

[0081] Example 1

[0082] A method for adaptive load adjustment of high-proportion renewable energy power grids based on a two-level optimization model includes the following steps:

[0083] S1 Extreme Weather Source-Load Uncertainty Modeling: For extreme weather in summer and winter, a source-load physical characteristic model is constructed, including a wind turbine output model that takes into account high temperature attenuation, a photovoltaic unit output model that takes into account snow shading, and a temperature control load model for summer and winter.

[0084] S2 Source Load Prediction and Multi-Scenario Quantization: Based on the S1 physical model, the baseline value of the source output is determined, and the VMD-LSTM residual correction prediction algorithm is used to obtain the predicted value of the source output. A dynamic load prediction model considering temperature control inertia is constructed to solve the problem of load prediction lag under extreme weather conditions. Then, LHS is used to generate a massive initial scenario set. By improving the Canopy-spectral clustering-K-means hybrid clustering algorithm, scenarios are reduced, typical scenarios and their corresponding occurrence probabilities are extracted, and the uncertainty boundary of source load under extreme weather conditions is quantified.

[0085] S3 Comprehensive Compensation System and User Response Model Construction: Construct a five-segment dynamic time-of-use electricity price model and a comprehensive compensation equivalent system, classify users into continuous production and discrete manufacturing users, construct revenue loss models for each typical industry, and construct a user response willingness model based on the Sigmoid function to provide incentives and response basis for load adjustment;

[0086] S4 Extreme Weather Sensing and Status Assessment: Real-time monitoring of meteorological parameters, activation of adjustment mechanism when thresholds are continuously triggered, and simultaneous real-time push of weather warnings, electricity price signals and system adjustment needs to users to guide users to adjust their electricity consumption in advance;

[0087] S5 Two-Layer Optimization Model Construction: A two-layer optimization model is constructed. The upper-layer model aims to minimize the total system operating cost within the scheduling cycle, while the lower-layer model aims to minimize load fluctuations and voltage deviations and improve power supply quality.

[0088] S6 Command Issuance and Feedback Calibration: The commands obtained from solving the two-layer optimization model are issued to each execution unit, and the deviation between the actual execution data and the target value is compared to calibrate the parameters of each model; a real-time settlement mechanism for evaluating and compensating user response effects is established to improve user response enthusiasm.

[0089] In S1:

[0090] Wind turbine output modeling under high temperature environment: Based on the physical characteristics of reduced air density and easy high temperature protection shutdown of wind turbines caused by high temperature in summer, the model is divided into the range of wind turbine cut-in, rated and cut-out wind speeds. An air density correction coefficient and rated power reduction coefficient related to ambient temperature are introduced into the traditional wind turbine output model. At the same time, temperature cut-out logic is added to quantify the changes in wind turbine output caused by the decrease in wind energy capture efficiency and equipment protection shutdown due to high temperature.

[0091] Modeling the output of photovoltaic units under snow and cold waves: In response to the dual impacts of irradiance attenuation caused by winter cold waves and snow and snow shading of photovoltaic modules, the rated output under standard photovoltaic test conditions is used as the basis. Temperature correction is performed by combining the power temperature characteristics of the modules. At the same time, a snow shading loss coefficient that is non-linearly related to snow thickness and snow melting rate is introduced, and parameters such as inverter efficiency are included to correct the actual photovoltaic output and realize the quantification of photovoltaic output attenuation under the dual impacts.

[0092] Summer and winter temperature control load modeling: Both are based on the basic principles of thermodynamics, and respectively construct summer air conditioning cooling and winter electric heating load models. The models include the deviation between ambient temperature and equipment set temperature, equipment energy efficiency ratio, air and building physical parameters and user-related indicators. By quantifying the temperature deviation and the correlation between various physical and equipment parameters, the models can accurately calculate the sharp increase in cooling or heating load caused by the increase in temperature deviation under extreme weather conditions.

[0093] In S2:

[0094] VMD-LSTM residual correction for source-end power output prediction: First, the theoretical baseline value of power output is calculated using a source-end physical model under extreme weather conditions, and the residual sequence between the theoretical value and the historical measured value is extracted. Then, variational mode decomposition (VMD) is used to decompose the non-stationary, multi-scale residual sequence into multiple intrinsic mode components. The component parameters are iteratively updated by solving variational constraint problems to avoid the mode aliasing problem of traditional decomposition. Subsequently, a long short-term memory (LSTM) network model is constructed for each component, and long-term temporal dependencies are captured by combining extreme weather feature vectors. The predicted components are then superimposed and reconstructed to compensate the residuals to the physical baseline value, thus obtaining the final predicted value of source-end power output.

[0095] Dynamic load prediction considering temperature control inertia: Based on the benchmark load, the temperature control component reflecting the building's cumulative heat effect is coupled to obtain the total load prediction value; the cumulative heat effect temperature is introduced to characterize the building's thermal energy storage state, and the current ambient temperature is coupled with the previous thermal state through a thermal inertia weighting coefficient to quantify the building's sensitivity to changes in ambient temperature; this solves the problem that traditional models cannot capture the hysteresis and inertia of temperature control loads under extreme temperatures, as well as the prediction lag for nonlinear load increases;

[0096] Latin Hypercube Probability Sampling (LHS) generates the initial scene set: To meet the need for capturing extremely low-probability events, the probability distribution of prediction errors is sampled in layers; the probability interval is divided into equally probable sub-intervals, and samples are independently and randomly drawn from each interval and obtained by inverse transformation of the cumulative distribution function to ensure uniform coverage of the samples in the full probability space, and finally a massive initial scene set containing the temporal correlation of wind, light, and load is generated.

[0097] An improved Canopy-spectral clustering-K-means hybrid algorithm for scene reduction: A three-step clustering strategy is employed to achieve efficient and reasonable scene reduction while preserving key features.

[0098] ①Canopy coarse clustering quickly divides scene subsets through dual distance thresholds, objectively determines the optimal number of clusters, and overcomes the subjectivity of traditional algorithms that preset the number of clusters;

[0099] ② Spectral clustering nonlinear dimensionality reduction: By constructing a Laplace matrix, high-dimensional time-series scene data is mapped to a low-dimensional feature space, which can identify complex data structures and retain the nonlinear and time-series correlation characteristics of source load fluctuations under extreme weather conditions.

[0100] ③ K-means fine partitioning: using the low-dimensional feature matrix as input and the number of clusters determined by Canopy as the basis, iterative optimization is performed to minimize the intra-cluster error, and finally a representative set of typical scenarios is extracted and the probability of occurrence of each scenario is determined.

[0101] In S3:

[0102] Five-segment dynamic time-of-use pricing model: A five-segment dynamic time-of-use pricing model is constructed, consisting of peak-peak-flat-valley-deep valley. The electricity price elasticity matrix quantifies the cross-time-shifting effect of electricity prices on load at different times. At the same time, the revenue equivalent of time-of-use pricing is introduced to aggregate the cross-time-shifting electricity price difference revenue generated by users' peak shifting and valley filling, calculate the electricity cost savings per unit of response electricity, and transform the price signal into an intuitive economic incentive per unit of electricity, which serves as the basic value component of the comprehensive incentive system.

[0103] Comprehensive compensation equivalent system:

[0104] Construct a demand response compensation equivalent, which is divided into two dimensions: electrical energy and reserve compensation. Based on parameters such as subsidy standards, effect judgment coefficients, and total monthly reserve costs, the total economic compensation for demand response per unit of electricity is quantified.

[0105] To address the diverse needs of load-side resources participating in system regulation, three types of ancillary service compensation equivalents are constructed: frequency regulation, reserve, and ramping. Based on the core impact parameters of each service, the corresponding service compensation amount per unit of response power is calculated, thereby achieving precise quantification of the value of users participating in different forms of system regulation.

[0106] By organically integrating the time-of-use electricity price revenue equivalent, demand response compensation equivalent, and compensation equivalent of various ancillary services, a comprehensive compensation equivalent on the user side is constructed. This enables a unified quantification and aggregation of the price revenue, demand response value, and system ancillary service value created by users' participation in load adjustment, forming a clear total economic incentive signal for unit electricity adjustment and providing a unified revenue reference for user decision-making.

[0107] Typical industry revenue loss models: Industrial users are divided into continuous production and discrete manufacturing based on their production characteristics. A refined unit power reduction production revenue loss model is constructed for the two types of users based on the differences in production rigidity or elasticity: for continuous production users, the model focuses on the chain losses caused by the rigidity of the production process; for discrete manufacturing users, the model focuses on the efficiency and opportunity losses caused by the disruption of the production rhythm. At the same time, a typical industry loss parameter library is established, and the parameters are continuously iterated and calibrated through actual response data to improve the accuracy of cost quantification.

[0108] User Response Willingness Model: A game-theoretic collaborative participation rate model is constructed using the Sigmoid function. The actual user participation rate is defined as a nonlinear function of "comprehensive compensation equivalent / unit revenue loss". A subsidy benefit sensitivity coefficient and a participation willingness threshold are introduced to quantify the degree of user participation under different benefit-cost ratios. This enables an accurate characterization of the willingness to participate in load adjustment for different industries and user groups, providing a basis for optimizing and adjusting the incentive mechanism.

[0109] In S4:

[0110] By monitoring key meteorological parameters such as wind speed, irradiance, ambient temperature, and snow depth in real time, an adjustment mechanism is activated when thresholds are continuously triggered. This dynamically corrects future wind and solar power output and load forecast curves, accurately identifying risks such as system power deficits, node voltage exceedances, and line overloads. Simultaneously, the power and electricity price elasticity matrix is ​​updated to be applicable to the current scenario. In addition to user response potential assessment parameters, the system simultaneously pushes real-time weather warnings, electricity price signals, and system adjustment needs to users, enabling them to adjust their electricity consumption behavior in advance based on their own electricity consumption flexibility and production plans, thereby improving the initiative and accuracy of response and providing accurate data input for optimized decision-making.

[0111] In S5:

[0112] Upper-level model: The core optimization objective is to minimize the total operating cost of the system within the scheduling cycle. The objective function comprehensively incorporates the four core cost items of system operation under extreme weather conditions and introduces weight coefficients to quantify the impact of environmental protection costs. Specifically, it covers the total cost of various generating units in time period t, the cost of thermal power pollutant treatment, the total demand response compensation cost, and the cost of external power purchase. At the same time, through the weight coefficient of pollutant treatment cost, the weight of thermal power environmental protection treatment cost in the total operating cost is clarified, so as to achieve a balance between the economic operation of the system and environmental protection requirements, and to define the economic constraint boundary of load adjustment at the system level.

[0113] Lower-level model: Within the framework of upper-level system economic optimization, the core optimization objective is to minimize power quality-related indicators such as load fluctuation and voltage deviation, while also considering the accuracy of demand response execution. A multi-indicator comprehensive optimization objective function is constructed: comprehensively considering three core power quality indicators—node net load fluctuation, node voltage deviation, and tie-line power fluctuation—and also incorporating the following error between the expected and actual adjustment amounts in demand response to ensure that the demand response execution effect matches the system scheduling expectations; normalized weight coefficients are assigned to each indicator to achieve balanced optimization of multiple power quality indicators, ensuring power supply stability during load adjustment from the user side and avoiding damage to the user's electricity experience due to system-side economic optimization.

[0114] In S6:

[0115] The instructions from the two-layer optimization model are sent to each execution unit to synchronously track the core operating parameters of key users; the deviations of key indicators such as voltage fluctuation rate, tie line power, and actual user participation rate from the optimized values ​​are calculated, and the parameters of each model are dynamically calibrated based on feedback. A real-time settlement mechanism for evaluating user response effects and compensation is established, and users can easily view their own response volume, compensation benefits, and contribution to system regulation.

[0116] The VMD-LSTM residual correction prediction algorithm is as follows:

[0117] The theoretical benchmark value of source-end output was calculated based on the S1 physical model. And extract its values ​​and historical measured values. residual sequence Given the non-stationary, multi-scale time-frequency characteristics of residual sequences under extreme weather conditions, variational mode decomposition (VMD) is employed to decompose them into... Each intrinsic mode component (IMF) has a different center frequency;

[0118] VMD transforms the signal decomposition process into finding the optimal center frequency of each mode by solving variational constraint problems. The objective function for the bandwidth process is constructed as follows:

[0119]

[0120]

[0121] In the formula, For the first One modal component, Let's define the Dirac function. Introduce Lagrange multipliers. and secondary penalty factor Iterative updates using the alternating direction multiplier method and The process continues until convergence, effectively avoiding the mode aliasing problem in EMD decomposition and accurately extracting high-frequency random components and low-frequency trend components from the residual sequence.

[0122] For each decomposed IMF component, a Long Short-Term Memory (LSTM) prediction model is constructed; the LSTM introduces a forgetting gate. Input gate and output gate The gating mechanism overcomes the gradient vanishing problem of traditional RNNs and effectively captures extreme weather characteristics. The long-term time-series dependency between the predictions and the residual components; after the predicted values ​​of each component are superimposed and reconstructed, the corrected final predicted output is obtained:

[0123]

[0124] In the formula, This is the input extreme weather feature vector.

[0125] The two-layer optimization model includes:

[0126] Upper-level model:

[0127] The objective is to minimize the total system operating cost within the scheduling cycle.

[0128]

[0129] In the formula, , , They represent Total cost of various types of generator units during the period, cost of treating pollutants from thermal power plants, total demand response compensation cost, and cost of purchasing external power; This represents the weighting coefficient for pollutant treatment costs;

[0130] Lower-level model:

[0131] Aimed at minimizing load fluctuations and voltage deviations to improve power quality:

[0132]

[0133] In the formula, , , They represent Time period node i net load fluctuation, node Voltage deviation, tie line Power fluctuations; and They represent Expected and actual adjustments to demand response during different time periods; Let represent the weighting coefficients, and .

[0134] A system for implementing the high-proportion renewable energy grid load adaptive adjustment method based on a two-level optimization model includes:

[0135] Modeling module: Constructs source-load physical characteristic models for extreme summer and winter weather, specifically including a wind turbine output modeling unit that takes into account high temperature attenuation, a photovoltaic unit output modeling unit that takes into account snow shading, and a temperature control load modeling unit for summer and winter seasons, quantifies the physical impact of extreme weather on source-load output, and outputs source-load physical characteristic parameters.

[0136] Predictive clustering module: Based on the source output baseline value output by the modeling module, the source output is predicted through the VMD-LSTM residual correction prediction unit. The dynamic load prediction unit solves the load prediction lag problem under extreme weather conditions. Then, through the scenario generation and reduction unit, LHS sampling and improved Canopy-spectral clustering-K-means hybrid clustering algorithm are used to extract typical scenarios and their occurrence probabilities.

[0137] Incentive Response Module: Constructs a five-segment dynamic time-of-use electricity price calculation unit and a comprehensive compensation equivalent calculation unit, classifies two types of industrial users, establishes revenue loss models for each typical industry, constructs a user response willingness unit through the Sigmoid function, and outputs load adjustment incentive signals and user response benchmark data;

[0138] Perception and Assessment Module: Accesses real-time meteorological monitoring data, sets a threshold judgment unit, activates the adjustment mechanism when the threshold is continuously triggered, and constructs a signal push unit to simultaneously push meteorological warnings, electricity prices and regulation demand signals;

[0139] Optimization Solver Module: Constructs a two-layer optimization solver. The upper layer aims to minimize the total operating cost of the system, while the lower layer aims to minimize load fluctuations and voltage deviations. It embeds corresponding constraints and solution algorithms and outputs load adjustment commands.

[0140] Command feedback module: issues adjustment commands to each execution unit, sets up deviation comparison and calibration unit, corrects model parameters, and constructs response evaluation and compensation settlement unit to improve user response enthusiasm and ensure adjustment accuracy.

[0141] Example 2

[0142] A method for adaptive load adjustment of high-proportion renewable energy power grids based on a two-level optimization model includes the following steps:

[0143] I. Uncertainty Perception and Quantification of Source and Load under Extreme Weather Conditions

[0144] (I) Physical characteristics modeling of the impact of extreme weather on both ends of the source-load axis

[0145] 1. Wind turbine output model considering high-temperature attenuation

[0146] In scenarios with sustained high temperatures during summer, reduced air density leads to decreased wind energy capture efficiency, and the turbine may trigger a protection shutdown due to excessively high ambient temperatures. By introducing an air density correction coefficient and temperature cut-off logic, the wind turbine output model is improved as follows:

[0147]

[0148] In the formula, for The output power of the fan at all times; Wind speed at wheel hub height; , , These are the cut-in, rated, and cut-out wind velocities, respectively. Ambient temperature; The wind energy utilization coefficient; Sweeped area; This is the rated power. The air density varies with temperature. This is the rated power reduction factor caused by high temperature.

[0149] 2. Photovoltaic unit output model considering snow shading

[0150] During winter cold waves and snowfall, photovoltaic (PV) output is affected by both irradiance degradation and snow cover on the modules. A PV output model considering snow shading and temperature characteristics is established:

[0151]

[0152] In the formula, To contribute to the actual development of photovoltaics; Rated power under standard test conditions (STC); and These are the actual irradiance and the standard irradiance, respectively. The power temperature coefficient; The operating temperature of the component; For reference temperature; For inverter efficiency; The shading loss coefficient caused by snow cover is non-linearly related to snow thickness and snow melting rate.

[0153] 3. Temperature control load model for summer and winter seasons

[0154] Based on thermodynamic principles, the summer air conditioning cooling load can be expressed as:

[0155]

[0156] In the formula, Indicates the number of users. Indicates air density ( ), This indicates specific heat capacity (kJ / (kg·K)). Indicates the cooling energy efficiency ratio. This indicates the air conditioner's set temperature (°C). Indicates the area per capita ( ), It represents the building heat transfer coefficient (W / (m²·K)).

[0157] The winter electric heating load model is as follows:

[0158]

[0159] In the formula, Indicates room volume ( ), Indicates the heat loss coefficient. This indicates the heating efficiency ratio.

[0160] High temperatures in summer and cold waves in winter cause a large deviation between the ambient temperature and the set temperature of the equipment, resulting in a sharp increase in the load of electric heating / cooling.

[0161] (II) Joint prediction model of physical mechanism correction and data-driven residual compensation

[0162] Single physical models or data-driven models have limitations in terms of accuracy under extreme conditions. This step proposes a hybrid prediction strategy of "physical benchmark-data correction," which uses a physical model to determine the output benchmark and leverages deep learning to accurately compensate for elusive random residuals.

[0163] 1. Residual Correction Prediction Algorithm Based on VMD-LSTM

[0164] First, the theoretical baseline value of the source-end output force is calculated based on the physical characteristic model. And extract its values ​​and historical measured values. residual sequence Given the non-stationary, multi-scale time-frequency characteristics of residual sequences under extreme weather conditions, variational mode decomposition (VMD) is employed to decompose them into... Each intrinsic mode component (IMF) has a different center frequency.

[0165] VMD transforms the signal decomposition process into finding the optimal center frequency of each mode by solving variational constraint problems. The objective function for the bandwidth process is constructed as follows:

[0166]

[0167]

[0168] In the formula, For the first One modal component, Let's define the Dirac function. Introduce Lagrange multipliers. and secondary penalty factor Iterative updates using the Alternating Direction Multiplier Method (ADMM) and This process continues until convergence, effectively avoiding the mode aliasing problem in EMD decomposition and accurately extracting high-frequency random components and low-frequency trend components from the residual sequence.

[0169] Subsequently, a Long Short-Term Memory (LSTM) prediction model was constructed for each decomposed IMF component. LSTM introduces a forgetting gate. Input gate and output gate The gating mechanism overcomes the gradient vanishing problem of traditional RNNs and can effectively capture extreme weather characteristics. (e.g., temperature abrupt change rate, snow depth variation) and long-term time-series dependencies between the residual components. After superposition and reconstruction of the predicted values ​​of each component, the corrected final predicted output is obtained:

[0170]

[0171] In the formula, This is the input extreme weather feature vector.

[0172] 2. Dynamic load prediction considering temperature control inertia

[0173] Under extreme temperatures, temperature-controlled loads (TCLs) such as air conditioners and electric heating exhibit significant hysteresis and inertia. Load forecasting models need to be based on baseline loads. Based on this, the temperature control component reflecting the cumulative heat effect is coupled. :

[0174]

[0175] Among them, the cumulative thermal effect temperature This reflects the thermal energy storage state inside the building, and is related to the current ambient temperature and the thermal state at the previous moment:

[0176]

[0177] In the formula, This is the total load forecast value; This is the temperature-controlled load response function; To effectively accumulate temperature; This is the thermal inertia weighting coefficient; the larger the value, the stronger the building's sensitivity to changes in external temperature, and the smaller its thermal inertia. This model effectively solves the problem of predictive lag in traditional models for nonlinear load increases under conditions of persistent heat waves or cold snaps.

[0178] (III) Multi-scenario generation method based on probability sampling and improved clustering

[0179] Given the strong spatiotemporal coupling of the impacts of extreme weather on both sides of the source load, a single point prediction cannot cover the system's operational boundary. To introduce quantifiable uncertainty into stochastic programming, this step employs scenario analysis to transform uncertainty into a set of typical risk scenarios. A massive number of initial scenarios are generated through probability sampling, and then reduced using an improved clustering algorithm.

[0180] 1. Latin Hypercube Probability Sampling (LHS)

[0181] To capture extremely low-probability events, Latin hypercube sampling (LHS) is used to perform stratified sampling of the prediction error probability distribution. The probability intervals are then... Divided into The system generates three equally probable sub-intervals, and independently randomly samples are drawn from each sub-interval. The samples are then subjected to an inverse cumulative distribution function transformation to obtain the sampled values. This ensures uniform coverage of the sample across the entire probability space (especially in extreme boundary regions), generating... A total scheduling cycle Initial scene set of temporal correlations of internal wind, light, and load .

[0182]

[0183] In the formula, For the first Each random scenario contains random time-series data of wind, solar and load data within the entire scheduling cycle T.

[0184] 2. Scenario Reduction Based on Improved Canopy-Spectral Clustering-K-means

[0185] To address the curse of dimensionality caused by massive data scenarios, an improved Canopy-spectral clustering-K-means hybrid clustering reduction strategy is proposed, as follows:

[0186] (1) Canopy coarse clustering to determine the optimal number of scenes K: Preprocessing is performed using the Canopy algorithm's fast computation speed and the fact that it does not require a preset number of clusters. A relaxed distance threshold is set. and compact distance threshold Calculate the Euclidean distance between scenes to quickly divide the initial scene into several overlapping subsets, thereby objectively determining the optimal number of cluster centers. This overcomes the limitation of the traditional K-means algorithm relying on specified parameters. The subjective nature of the value.

[0187] (2) Spectral clustering nonlinear dimensionality reduction mapping: constructing the similarity matrix of scene samples and metric matrix Calculate the Laplace matrix By solving The eigenvalues ​​and eigenvectors are used to map high-dimensional time-series scene data to a low-dimensional feature space, generating a dimensionality-reduced matrix. Spectral clustering can identify complex data structures with non-convex distributions and effectively preserve the nonlinear and temporal correlation characteristics of power output fluctuations under extreme weather conditions.

[0188] (3) K-means low-dimensional space fine partitioning: using low-dimensional feature matrix As input, determined by Canopy The K-means algorithm is executed based on the number of clusters. Through iterative optimization, the sum of squared errors within each cluster is minimized, ultimately extracting the cluster size. One of the most representative typical scenarios and their corresponding probability of occurrence .

[0189] This method combines the advantages of three clustering algorithms to generate a set of typical scenarios. It not only covers high-probability normal fluctuation scenarios, but also accurately retains key boundary risk scenarios that pose a serious threat to the safe and economical operation of the system, such as high temperatures and no wind in summer and cold waves and rain and snow in winter, laying a solid foundation for the subsequent two-layer adaptive optimization model.

[0190] II. Coordinated Incentive Mechanism between Time-of-Use Pricing and Multi-Type Integrated Compensation Equivalent System

[0191] (a) Price signal guidance: Time-of-use electricity pricing model

[0192] To strengthen the guiding role of price signals, a five-stage dynamic time-of-use electricity pricing model based on Shandong Province's current policies is constructed: Peak-Peak-Side-Valley-Deep Valley.

[0193]

[0194] To quantify the cross-period impact of electricity prices at different times on load distribution, an electricity price elasticity matrix is ​​established. The load distribution shift after the implementation of time-of-use pricing is analyzed, and its expression is as follows:

[0195]

[0196] The diagonal elements of this matrix The self-elasticity coefficient (always negative), off-diagonal element The cross-elasticity coefficient (always positive) characterizes the load transfer effect between different time periods.

[0197] To incorporate the electricity cost savings achieved by users under the time-of-use pricing policy into a unified incentive system, a time-of-use pricing revenue equivalent is introduced. The aim is to quantify a specific time period. This refers to the average electricity cost savings that users can achieve per unit of electricity consumed by adjusting their electricity usage behavior. The core calculation involves aggregating the cross-time-period electricity price difference benefits generated by users' participation in the adjustment process.

[0198]

[0199] in, The time-of-use electricity price is the price for the actual period during which the user's load occurs. This represents the selected reference benchmark electricity price. This equivalent directly reflects the direct economic benefits that users can obtain by actively shifting peak loads and filling valleys under the guidance of time-of-use pricing signals, providing a basic price signal value component for the subsequent comprehensive compensation and incentive system.

[0200] (ii) Demand Response Model

[0201] To further quantify the intensity of the equivalent economic incentives for demand response, a demand response compensation equivalent is constructed. It is defined as the total economic compensation that can be obtained per unit of response electricity, and the expression is as follows:

[0202]

[0203] In the formula, , They represent time periods respectively. Domestic demand response equivalent electrical energy and reserve compensation equivalent; This is the effect evaluation coefficient; Electricity subsidy standard (yuan / kWh); The total monthly reserve compensation cost (in yuan).

[0204] (III) Precise economic compensation: a multi-type ancillary service compensation equivalent system

[0205] With the deepening of electricity market development, the forms in which load-side resources participate in system regulation have expanded to include various ancillary services such as frequency regulation, reserve, and ramp-up. To accurately depict users' willingness to participate in these services, it is necessary to construct corresponding ancillary service compensation equivalents.

[0206] (1) Frequency modulation compensation equivalent

[0207] Frequency regulation service refers to the active power output adjustment service provided through technologies such as automatic power control to reduce system frequency deviation. Its compensation equivalent... Calculated based on FM mileage, performance coefficient, and clearing price:

[0208]

[0209] In the formula, Frequency regulation mileage (MW); For frequency modulation performance coefficient; Frequency regulation clearing price (RMB / MW); Frequency modulation response energy (kWh); This refers to the frequency modulation participation factor on the user side.

[0210] (2) Reserve compensation equivalent

[0211] Backup service refers to services that reserve adjustment capacity and adjust active power output within a specified time to meet the needs of safe system operation. Its compensation equivalent... The calculation is as follows:

[0212]

[0213] In the formula, Reserve capacity (MW); Price for clearing standby capacity (RMB / MW); Backup call time (h); Backup response capacity (kWh); The standard for backup energy compensation is (yuan / kWh).

[0214] (3) Climbing compensation equivalent

[0215] Ramp-up service refers to a service that rapidly adjusts power output according to dispatch instructions to maintain system power balance in response to short-term, significant changes in system net load caused by fluctuations in renewable energy generation. Its compensation equivalent... The calculation is as follows:

[0216]

[0217] In the formula, Ramp capacity (MW); The ramp-up and clearing price (RMB / MW); Duration of the hill climb service (h); Ramp-up response power (kWh); This is the climbing rate adjustment coefficient.

[0218] (iv) Construction of a comprehensive compensation equivalent system

[0219] To create a unified, transparent, and accurate economic incentive signal, and to guide users to make rational decisions and respond proactively, the time-of-use electricity price revenue equivalent... Demand response compensation equivalent and comprehensive compensation equivalent for multiple types of ancillary services By organically integrating these elements, a comprehensive compensation equivalent system for the user side was constructed. :

[0220]

[0221] This system achieves a unified quantification and aggregated expression of the multi-dimensional value created by users' participation in load adjustment, which together constitutes the total economic incentive for unit power adjustment that is clearly presented to users.

[0222] (V) Quantification of Response Costs: Analysis of Revenue Loss Mechanisms for Industrial Users

[0223] Loss of production revenue due to industrial users reducing unit electricity consumption ( The threshold value (TV) is a key economic decision parameter for whether or not a company participates in demand response. This value is not a fixed constant, but a function dynamically determined by core factors such as industry attributes, production processes, techno-economic characteristics, and market environment. To achieve accurate quantification of user response willingness, this section categorizes industrial users into two main types based on the actual load composition of the regional power grid: continuous production enterprises and discrete manufacturing enterprises. For each of these two typical industries, a refined revenue-loss model is constructed for in-depth analysis.

[0224] 1. Revenue and Loss Model for Continuous Production Firms

[0225] Continuous production enterprises (such as chemical and metallurgical companies) require their production facilities to operate stably and uninterrupted throughout the year. Their production processes exhibit significant rigidity, time lag, and high coupling characteristics. Unplanned load reductions or production interruptions not only result in current output losses but also trigger a series of chain reactions of physical and economic losses.

[0226] (1) Chemical industry

[0227] Chemical production is characterized by rigid processes, high temperature and pressure, and continuous reactions. Losses in the industry primarily stem from material waste caused by process interruptions, high restart costs, and safety and environmental risks. This constitutes a comprehensive loss of revenue. The model is:

[0228]

[0229] In the formula, Energy loss due to production line restart; This is due to the loss of raw materials and catalysts due to deactivation. For equipment maintenance and lifespan reduction; Fines for environmental compliance and safety violations; For supply chain default penalties; The cumulative reduction in electricity production (kWh).

[0230] in:

[0231]

[0232] In the formula, Power output under normal operating conditions (kW); This represents the sensitivity coefficient of the production process to power outages. , These are the restart time and normal startup time (in hours), respectively. Electricity price during the restart period (RMB / kWh); The quantity (in tons) of the i-th type of raw material that is scrapped; The unit price of the i-th raw material (yuan / ton); , The costs for catalyst replacement and equipment maintenance are respectively (in yuan). , These are the original value of the equipment and the accumulated depreciation (in yuan); , These are the equipment's design life and the lifespan reduction (in years) caused by abnormal interruptions. Fines imposed by the environmental protection department (in yuan); Environmental remediation costs (RMB); The number of defaulted orders (in tons); Price per ton (RMB). This refers to the percentage of penalties for breach of contract.

[0233] (2) Metallurgical industry

[0234] The metallurgical industry is characterized by extremely high energy consumption, the need to maintain a molten state, and sensitivity to equipment lifespan. The core of revenue losses in this industry lies in the massive quality degradation, physical damage to equipment, and extremely high restart energy consumption caused by drastic changes in operating conditions. Its overall revenue loss... The model is:

[0235]

[0236] In the formula, This refers to the loss of raw material quality, specifically the loss caused by raw material waste and decreased yield due to production interruption. This represents energy consumption and carbon emission losses; This indicates the losses incurred during the start-up and shutdown of smelting equipment.

[0237] in:

[0238]

[0239] In the formula, Unit raw material cost (RMB / ton); The percentage decrease in yield (%). Cost of rework or downgrade processing (RMB / ton); The amount of waste or defective products generated (tons); and The unit product power consumption (kWh / ton) is for normal and abnormal operating conditions, respectively. Electricity price (RMB / kWh); The affected output (tons); Additional carbon emissions (tons) , These are carbon emissions and refractory material replacement costs (RMB / ton), respectively. Additional energy costs (RMB) for restarting; Loss due to solidification of molten metal (yuan); Number of unplanned shutdowns; These are the original value of the equipment and the accumulated depreciation (in yuan); , These are the equipment's design life and the equipment's lifespan depletion (in years).

[0240] 2. Revenue and Loss Model for Discrete Manufacturing

[0241] Discrete manufacturing (such as machining and home appliance manufacturing) consists of multiple separable processes, possessing a certain degree of adjustability and flexibility. Losses primarily stem from decreased production efficiency, order delivery risks, and lost market opportunities caused by disruptions to the production rhythm.

[0242] (1) Machinery processing industry

[0243] The industry's losses are concentrated in areas of compromised production precision, abnormal tool wear and tear, and delivery delays. Its overall revenue loss... The model is:

[0244]

[0245] In the formula, , , , , These respectively represent idle capacity losses, gross profit losses, abnormal tool wear, equipment restart and calibration costs, and delivery delay penalties.

[0246] in:

[0247]

[0248] In the formula, Monthly fixed costs (in yuan), including depreciation, management fees, etc. Total monthly production time (hours); Downtime (in hours) caused by load reduction; Price per unit (RMB / piece); Unit variable cost (RMB / unit); The reduction in production quantity (pieces); The additional tooling cost (in yuan) due to a single abnormal shutdown; The labor cost per unit time (yuan / hour); Cost of calibration consumables per unit time (RMB / hour); Time required for calibration (in hours); The penalty standard (in yuan) is as stipulated in the contract. This is the delay coefficient, which is positively correlated with the number of delay days.

[0249] (2) Home appliance manufacturing

[0250] The industry's losses are closely linked to seasonal market demand, raw material cost fluctuations, and brand reputation, especially during peak sales seasons when the opportunity cost of load adjustments is substantial. This results in overall revenue loss. The model is:

[0251]

[0252] In the formula, , , The losses were quantified separately for raw material price difference losses, idle capacity and peak season opportunity losses, and brand and channel reputation losses.

[0253] in:

[0254]

[0255] In the formula, , These are the spot and contract prices for raw materials (RMB / ton); This is the raw material cost sensitivity coefficient; The sales volume (units) lost during the peak sales season due to production stoppage; The brand value assessment (in RMB); This represents the goodwill loss coefficient.

[0256] By establishing a database of typical industry production revenue loss parameters that covers the aforementioned industry characteristics, differentiated services can be provided to users in different industries. Benchmark reference value. As demand response is continuously implemented, actual response data can be used to... The model continuously iterates and dynamically calibrates its constituent parameters to improve the accuracy and adaptability of the user intention quantification model.

[0257] (vi) Characterizing Response Willingness: Sigmoid Function and Game-Theoretic Collaborative Participation Rate Model

[0258] The percentage of users actually participating in load adjustment This function reflects the economic comparison between subsidy policies and user revenue losses. The Sigmoid function is used to deeply analyze the revenue losses caused by different types of industrial users reducing their electricity consumption per unit:

[0259]

[0260] In the formula: Indicates time period The actual user participation rate ranges from [0,1]. The coefficient representing the sensitivity of the user group to subsidy benefits reflects the user group's... The steepness of the ratio response, The higher the value, the faster the user makes a decision. This represents the threshold for willingness to participate, when the compensation equals the loss ( When ), user participation should be 50% (i.e. ). This indicates that users have lower compensation expectations, meaning they are more easily incentivized. This indicates that users have high compensation expectations, meaning they are difficult to incentivize. When That is, when the compensation is far less than the loss. , , Users basically do not participate; when That is, when the compensation far exceeds the loss. , , Users actively participated.

[0261] III. Construction of a Load Adaptive Adjustment Method Based on a Two-Level Optimization Model

[0262] To dynamically respond to changes in system status under extreme weather conditions, demand-side resources are guided to spontaneously participate in adjustment through price signals and user willingness, while taking into account both the overall economic efficiency of the system and the power supply quality on the user side. The specific process is as follows:

[0263] (I) Extreme Weather Perception and Status Assessment

[0264] By monitoring key meteorological parameters such as wind speed, irradiance, ambient temperature, and snow depth in real time, an adjustment mechanism is activated when thresholds are continuously triggered. This dynamically corrects future wind and solar power output and load forecast curves, accurately identifying system power deficits, node voltage exceedances, and line overload risks. Simultaneously, the electricity and price elasticity matrix is ​​updated to be applicable to the current scenario. and user response potential assessment parameters (such as) , , (etc.) Simultaneously push weather warnings, electricity price signals and system adjustment needs to users in real time, enabling users to adjust their electricity consumption behavior in advance based on their own electricity consumption flexibility and production plans, improve the initiative and accuracy of response, and provide accurate data input for optimized decision-making.

[0265] (II) Construction and Solution of the Two-Level Optimization Model

[0266] 1. Upper-level model (system economic optimization)

[0267] The upper-level model aims to minimize the total system operating cost within the scheduling cycle.

[0268]

[0269] In the formula, , , They represent Total cost of various types of generator units during the period, cost of treating pollutants from thermal power plants, total demand response compensation cost, and cost of purchasing external power; This represents the weighting coefficient for pollutant treatment costs.

[0270] 2. Lower-level model (user-side power supply quality optimization)

[0271] The lower-level model aims to minimize load fluctuations and voltage deviations, thereby improving power quality.

[0272]

[0273] In the formula, , , They represent Time period node i net load fluctuation, node Voltage deviation, tie line Power fluctuations; and They represent Expected and actual adjustments to demand response during different time periods. Let represent the weighting coefficients, and .

[0274] (III) Adaptive Closed-Loop Execution and Feedback

[0275] The instructions from the two-layer optimization model are sent to each execution unit, and key user core operating parameters are tracked synchronously. The deviation between key indicators (such as voltage fluctuation rate, tie line power, user actual participation rate, etc.) and the optimized values ​​are calculated, and the user response model parameters and target weight coefficients are dynamically calibrated. A real-time settlement mechanism for user response effect evaluation and compensation is established. Users can easily view their own response volume, compensation benefits and contribution to system regulation, forming a complete dynamic closed-loop regulation mechanism of "meteorological perception → status assessment → two-layer optimization → instruction execution → status monitoring → parameter feedback".

[0276] Application example:

[0277] Taking a local power grid as the research object, this study analyzes two extreme weather scenarios lasting 72 hours: high temperature and no wind in summer, and cold wave and snow in winter. The relevant data before and after adaptive adjustment are compared, and the results are as follows:

[0278] Analysis of prolonged periods of high temperatures and no wind during summer:

[0279] Economic analysis: by Figure 5 It can be seen that, compared with the previous strategy, the proposed strategy reduced the total operating cost of the system by RMB 2.2865 million, a decrease of 16.18%. Among them, by suppressing peak loads, the cost of purchasing high-premium external power was significantly reduced by 39.8%.

[0280] Power quality analysis: by Figure 6 It can be seen that after adaptive adjustment, the average net load volatility of the system decreased from 41.59% to 21.28%, and the average tie-line power volatility decreased from 35.84% to 16.73%, with improvements of 48.83% and 53.32%, respectively. At the same time, the voltage deviation at each node was significantly reduced, with the maximum deviation controlled within 3.5%, and the voltage stability was significantly enhanced.

[0281] In terms of user response, the load adaptive adjustment model effectively stimulated users' adjustment potential, with an average participation rate of 68.3% and an actual response load ratio of 35.9%, resulting in a significant increase in net user benefits.

[0282] Analysis of multiple cold waves, rain, and snow in winter:

[0283] Economic analysis: by Figure 8 It can be seen that, compared with the previous strategy, the proposed strategy reduced the total operating cost of the system by RMB 1.5371 million, a decrease of 11.36%. Among them, the cost of purchasing high-premium external power was significantly reduced by 46.25%. Although the load adjustment compensation cost increased accordingly, the overall economic operation level of the system was effectively improved.

[0284] Power quality analysis: by Figure 9 It can be seen that after adaptive adjustment, the average net load volatility decreased from 33.16% to 15.74%, and the average tie-line power volatility decreased from 26.98% to 11.73%, representing improvements of 52.53% and 56.52%, respectively. At the same time, the voltage deviation at each node was effectively controlled, with the maximum deviation strictly limited to within 4.6%, and the system voltage stability and power supply reliability were significantly enhanced.

[0285] In terms of user response, the average participation rate in load adjustment increased to 72.6%, the actual response load ratio reached 43.1%, and the net user benefit increased significantly.

[0286] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. 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 high-proportion new energy power grid load self-adaptive adjustment method based on a double-layer optimization model, characterized in that: Includes the following steps: S1 Extreme Weather Source-Load Uncertainty Modeling: For extreme weather in summer and winter, a source-load physical characteristic model is constructed, including a wind turbine output model that takes into account high temperature attenuation, a photovoltaic unit output model that takes into account snow shading, and a temperature control load model for summer and winter. S2 Source Load Prediction and Multi-Scenario Quantization: Based on the S1 physical model, the baseline value of the source output is determined, and the VMD-LSTM residual correction prediction algorithm is used to obtain the predicted value of the source output. A dynamic load prediction model considering temperature control inertia is constructed to solve the problem of load prediction lag under extreme weather conditions. Then, LHS is used to generate a massive initial scenario set. By improving the Canopy-spectral clustering-K-means hybrid clustering algorithm, scenarios are reduced, typical scenarios and their corresponding occurrence probabilities are extracted, and the uncertainty boundary of source load under extreme weather conditions is quantified. S3 Comprehensive Compensation System and User Response Model Construction: Construct a five-segment dynamic time-of-use electricity price model and a comprehensive compensation equivalent system, classify users into continuous production and discrete manufacturing users, construct revenue loss models for each typical industry, and construct a user response willingness model based on the Sigmoid function to provide incentives and response basis for load adjustment; S4 Extreme Weather Sensing and Status Assessment: Real-time monitoring of meteorological parameters, activation of adjustment mechanism when thresholds are continuously triggered, and simultaneous real-time push of weather warnings, electricity price signals and system adjustment needs to users to guide users to adjust their electricity consumption in advance; S5 Two-Layer Optimization Model Construction: A two-layer optimization model is constructed. The upper-layer model aims to minimize the total system operating cost within the scheduling cycle, while the lower-layer model aims to minimize load fluctuations and voltage deviations and improve power supply quality. S6 Command Issuance and Feedback Calibration: The commands obtained from solving the two-layer optimization model are issued to each execution unit, and the deviation between the actual execution data and the target value is compared to calibrate the parameters of each model; a real-time settlement mechanism for evaluating and compensating user response effects is established to improve user response enthusiasm. 2.The high-proportion new energy power grid load self-adaptive adjustment method based on a double-layer optimization model according to claim 1, characterized in that: In S1: Wind turbine output modeling under high temperature environment: Based on the physical characteristics of reduced air density and easy high temperature protection shutdown of wind turbines caused by high temperature in summer, the model is divided into the range of wind turbine cut-in, rated and cut-out wind speeds. An air density correction coefficient and rated power reduction coefficient related to ambient temperature are introduced into the traditional wind turbine output model. At the same time, temperature cut-out logic is added to quantify the changes in wind turbine output caused by the decrease in wind energy capture efficiency and equipment protection shutdown due to high temperature. Modeling the output of photovoltaic units under snow and cold waves: In response to the dual impacts of irradiance attenuation caused by winter cold waves and snow and snow shading of photovoltaic modules, the rated output under standard photovoltaic test conditions is used as the basis. Temperature correction is performed by combining the power temperature characteristics of the modules. At the same time, a snow shading loss coefficient that is non-linearly related to snow thickness and snow melting rate is introduced, and parameters such as inverter efficiency are included to correct the actual photovoltaic output and realize the quantification of photovoltaic output attenuation under the dual impacts. Summer and winter temperature control load modeling: Both are based on the basic principles of thermodynamics, and respectively construct summer air conditioning cooling and winter electric heating load models. The models include the deviation between ambient temperature and equipment set temperature, equipment energy efficiency ratio, air and building physical parameters and user-related indicators. By quantifying the temperature deviation and the correlation between various physical and equipment parameters, the models can accurately calculate the sharp increase in cooling or heating load caused by the increase in temperature deviation under extreme weather conditions. 3.The high-proportion new energy power grid load self-adaptive adjustment method based on a double-layer optimization model according to claim 1, characterized in that: In S2: VMD-LSTM residual correction for source-end power output prediction: First, the theoretical baseline value of power output is calculated using a source-end physical model under extreme weather conditions, and the residual sequence between the theoretical value and the historical measured value is extracted. Then, variational mode decomposition (VMD) is used to decompose the non-stationary, multi-scale residual sequence into multiple intrinsic mode components. The component parameters are iteratively updated by solving variational constraint problems to avoid the mode aliasing problem of traditional decomposition. Subsequently, a long short-term memory (LSTM) network model is constructed for each component, and long-term temporal dependencies are captured by combining extreme weather feature vectors. The predicted components are then superimposed and reconstructed to compensate the residuals to the physical baseline value, thus obtaining the final predicted value of source-end power output. Dynamic load prediction considering temperature control inertia: Based on the benchmark load, the temperature control component reflecting the building's cumulative heat effect is coupled to obtain the total load prediction value; the cumulative heat effect temperature is introduced to characterize the building's thermal energy storage state, and the current ambient temperature is coupled with the previous thermal state through a thermal inertia weighting coefficient to quantify the building's sensitivity to changes in ambient temperature; this solves the problem that traditional models cannot capture the hysteresis and inertia of temperature control loads under extreme temperatures, as well as the prediction lag for nonlinear load increases; Latin Hypercube Probability Sampling (LHS) generates the initial scene set: To meet the need for capturing extremely low-probability events, the probability distribution of prediction errors is sampled in a stratified manner. The probability interval is divided into equally probable sub-intervals. Samples are randomly drawn independently in each interval and obtained by inverse transformation of the cumulative distribution function to ensure uniform coverage of the samples in the full probability space. Finally, a massive initial scene set containing the temporal correlation of wind, light, and load is generated. An improved Canopy-spectral clustering-K-means hybrid algorithm for scene reduction: A three-step clustering strategy is employed to achieve efficient and reasonable scene reduction while preserving key features. ①Canopy coarse clustering quickly divides scene subsets through dual distance thresholds, objectively determines the optimal number of clusters, and overcomes the subjectivity of traditional algorithms that preset the number of clusters; ② Spectral clustering nonlinear dimensionality reduction: By constructing a Laplace matrix, high-dimensional time-series scene data is mapped to a low-dimensional feature space, which can identify complex data structures and retain the nonlinear and time-series correlation characteristics of source load fluctuations under extreme weather conditions. ③ K-means fine partitioning: using the low-dimensional feature matrix as input and the number of clusters determined by Canopy as the basis, iterative optimization is performed to minimize the intra-cluster error, and finally a representative set of typical scenarios is extracted and the probability of occurrence of each scenario is determined.

4. The high-proportion new energy power grid load self-adaptive adjustment method based on a double-layer optimization model according to claim 1, characterized in that: In S3: Five-segment dynamic time-of-use pricing model: Construct a five-segment dynamic time-of-use pricing model with peak-peak-flat-valley-deep valley, and quantify the cross-time-shifting effect of electricity prices on load at different times through the electricity price elasticity matrix; At the same time, time-of-use electricity price revenue equivalent is introduced to aggregate the cross-time period electricity price difference revenue generated by users' peak shifting and valley filling, calculate the electricity cost savings per unit of response electricity, and transform the price signal into an intuitive economic incentive per unit of electricity, which serves as the basic value component of the comprehensive incentive system. Comprehensive compensation equivalent system: Construct a demand response compensation equivalent, which is divided into two dimensions: electrical energy and reserve compensation. Based on parameters such as subsidy standards, effect judgment coefficients, and total monthly reserve costs, the total economic compensation for demand response per unit of electricity is quantified. To address the diverse needs of load-side resources participating in system regulation, three types of ancillary service compensation equivalents are constructed: frequency regulation, reserve, and ramping. Based on the core impact parameters of each service, the corresponding service compensation amount per unit of response power is calculated, thereby achieving precise quantification of the value of users participating in different forms of system regulation. By organically integrating the time-of-use electricity price revenue equivalent, demand response compensation equivalent, and compensation equivalent of various ancillary services, a comprehensive compensation equivalent on the user side is constructed. This enables a unified quantification and aggregation of the price revenue, demand response value, and system ancillary service value created by users' participation in load adjustment, forming a clear total economic incentive signal for unit electricity adjustment and providing a unified revenue reference for user decision-making. Typical industry revenue loss models: Industrial users are divided into continuous production and discrete manufacturing based on their production characteristics. A refined unit power reduction production revenue loss model is constructed for the two types of users based on the differences in production rigidity or elasticity: for continuous production users, the model focuses on the chain losses caused by the rigidity of the production process; for discrete manufacturing users, the model focuses on the efficiency and opportunity losses caused by the disruption of the production rhythm. At the same time, a typical industry loss parameter library is established, and the parameters are continuously iterated and calibrated through actual response data to improve the accuracy of cost quantification. User Response Willingness Model: A game-theoretic collaborative participation rate model is constructed using the Sigmoid function. The actual user participation rate is defined as a nonlinear function of "comprehensive compensation equivalent / unit revenue loss". A subsidy benefit sensitivity coefficient and a participation willingness threshold are introduced to quantify the degree of user participation under different benefit-cost ratios. This enables an accurate characterization of the willingness to participate in load adjustment for different industries and user groups, providing a basis for optimizing and adjusting the incentive mechanism.

5. The method for adaptive adjustment of high-proportion renewable energy grid load based on a two-layer optimization model according to claim 1, characterized in that: In S4: By monitoring key meteorological parameters such as wind speed, irradiance, ambient temperature, and snow depth in real time, an adjustment mechanism is activated when thresholds are continuously triggered. This dynamically corrects future wind and solar power output and load forecast curves, accurately identifying risks such as system power deficits, node voltage exceedances, and line overloads. Simultaneously, the power and electricity price elasticity matrix is ​​updated to be applicable to the current scenario. In addition to user response potential assessment parameters, the system simultaneously pushes real-time weather warnings, electricity price signals, and system adjustment needs to users, enabling them to adjust their electricity consumption behavior in advance based on their own electricity consumption flexibility and production plans, thereby improving the initiative and accuracy of response and providing accurate data input for optimized decision-making.

6. The method for adaptive adjustment of high-proportion renewable energy grid load based on a two-layer optimization model according to claim 1, characterized in that: In S5: Upper-level model: The core optimization objective is to minimize the total operating cost of the system within the scheduling cycle. The objective function comprehensively incorporates the four core cost items of system operation under extreme weather conditions and introduces weight coefficients to quantify the impact of environmental protection costs. Specifically, it covers the total cost of various generating units in time period t, the cost of thermal power pollutant treatment, the total demand response compensation cost, and the cost of external power purchase. At the same time, through the weight coefficient of pollutant treatment cost, the weight of thermal power environmental protection treatment cost in the total operating cost is clarified, so as to achieve a balance between the economic operation of the system and environmental protection requirements, and to define the economic constraint boundary of load adjustment at the system level. Lower-level model: Under the framework of economic optimization of the upper-level system, the core optimization objective is to minimize power quality-related indicators such as load fluctuation and voltage deviation, while also taking into account the accuracy of demand response execution. A multi-indicator comprehensive optimization objective function is constructed: comprehensively considering three core power quality indicators, namely node net load fluctuation, node voltage deviation, and tie-line power fluctuation, and also incorporating the following error between the expected adjustment amount and the actual adjustment amount of demand response, to ensure that the demand response execution effect matches the system scheduling expectation; By configuring normalized weight coefficients for each indicator, a balanced optimization of multiple power supply quality indicators can be achieved, ensuring power supply stability during load adjustment from the user side and avoiding damage to the user's power experience due to economic optimization on the system side.

7. The method for adaptive adjustment of high-proportion renewable energy grid load based on a two-layer optimization model according to claim 1, characterized in that: In S6: The instructions from the two-layer optimization model are sent to each execution unit to synchronously track the core operating parameters of key users; the deviations of key indicators such as voltage fluctuation rate, tie line power, and actual user participation rate from the optimized values ​​are calculated, and the parameters of each model are dynamically calibrated based on feedback. A real-time settlement mechanism for evaluating user response effects and compensation is established, and users can easily view their own response volume, compensation benefits, and contribution to system regulation.

8. The method for adaptive adjustment of high-proportion renewable energy grid load based on a two-layer optimization model according to claim 1, characterized in that: The VMD-LSTM residual correction prediction algorithm is as follows: The theoretical benchmark value of source-end output was calculated based on the S1 physical model. And extract its values ​​and historical measured values. residual sequence Given the non-stationary, multi-scale time-frequency characteristics of residual sequences under extreme weather conditions, variational mode decomposition (VMD) is employed to decompose them into... Each intrinsic mode component (IMF) has a different center frequency; VMD transforms the signal decomposition process into finding the optimal center frequency of each mode by solving variational constraint problems. The objective function for the bandwidth process is constructed as follows: In the formula, For the first One modal component, Let's define the Dirac function. Introduce Lagrange multipliers. and secondary penalty factor Iterative updates using the alternating direction multiplier method and The process continues until convergence, effectively avoiding the mode aliasing problem in EMD decomposition and accurately extracting high-frequency random components and low-frequency trend components from the residual sequence. For each decomposed IMF component, a Long Short-Term Memory (LSTM) prediction model is constructed; the LSTM introduces a forgetting gate. Input gate and output gate The gating mechanism overcomes the gradient vanishing problem of traditional RNNs and effectively captures extreme weather characteristics. Long-term temporal dependencies between the residual components; After the predicted values ​​of each component are superimposed and reconstructed, the corrected final predicted output is obtained: In the formula, This is the input extreme weather feature vector.

9. The method for adaptive adjustment of high-proportion renewable energy grid load based on a two-layer optimization model according to claim 1, characterized in that: The two-layer optimization model includes: Upper-level model: The objective is to minimize the total system operating cost within the scheduling cycle. In the formula, , , They represent Total cost of various types of generator units during the period, cost of treating pollutants from thermal power plants, total demand response compensation cost, and cost of purchasing external power; This represents the weighting coefficient for pollutant treatment costs; Lower-level model: Aimed at minimizing load fluctuations and voltage deviations to improve power quality: In the formula, , , They represent Time period node i net load fluctuation, node Voltage deviation, tie line Power fluctuations; and They represent Expected and actual adjustments to demand response during different time periods; Let represent the weighting coefficients, and .

10. A system for implementing the adaptive load adjustment method for high-proportion renewable energy power grids based on a two-level optimization model as described in any one of claims 1 to 9, characterized in that: include: Modeling module: Constructs source-load physical characteristic models for extreme summer and winter weather, specifically including a wind turbine output modeling unit that takes into account high temperature attenuation, a photovoltaic unit output modeling unit that takes into account snow shading, and a temperature control load modeling unit for summer and winter seasons, quantifies the physical impact of extreme weather on source-load output, and outputs source-load physical characteristic parameters. Predictive clustering module: Based on the source output baseline value output by the modeling module, the source output is predicted through the VMD-LSTM residual correction prediction unit. The dynamic load prediction unit solves the load prediction lag problem under extreme weather conditions. Then, through the scenario generation and reduction unit, LHS sampling and improved Canopy-spectral clustering-K-means hybrid clustering algorithm are used to extract typical scenarios and their occurrence probabilities. Incentive Response Module: Constructs a five-segment dynamic time-of-use electricity price calculation unit and a comprehensive compensation equivalent calculation unit, classifies two types of industrial users, establishes revenue loss models for each typical industry, constructs a user response willingness unit through the Sigmoid function, and outputs load adjustment incentive signals and user response benchmark data; Perception and Assessment Module: Accesses real-time meteorological monitoring data, sets a threshold judgment unit, activates the adjustment mechanism when the threshold is continuously triggered, and constructs a signal push unit to simultaneously push meteorological warnings, electricity prices and regulation demand signals; Optimization Solver Module: Constructs a two-layer optimization solver. The upper layer aims to minimize the total operating cost of the system, while the lower layer aims to minimize load fluctuations and voltage deviations. It embeds corresponding constraints and solution algorithms and outputs load adjustment commands. Command feedback module: issues adjustment commands to each execution unit, sets up deviation comparison and calibration unit, corrects model parameters, and constructs response evaluation and compensation settlement unit to improve user response enthusiasm and ensure adjustment accuracy.

Images