A Multi-Source Data-Driven Method for Power Supply Load Resource Management and Control Decision Generation

Abstract

This invention relates to the field of load management technology, specifically providing a multi-source data-driven method for generating power supply load resource management decisions. The method includes: classifying load resources and accurately quantifying voltage drop based on multi-source electricity consumption data, calculating the total guaranteed voltage drop caused by rigid loads; generating multiple alternative management strategies that meet grid voltage drop constraints during the management preparation phase using a genetic algorithm; automatically selecting the optimal initial execution plan after management is initiated, and continuously evaluating the management completion status based on rolling time-domain prediction; automatically re-optimizing flexible resource start-up and shutdown arrangements and outputting a correction list when the execution performance deviates from a reasonable range, while continuously iterating and optimizing the prediction and optimization process based on the latest measured load. This addresses the technical problems in related technologies, such as difficulty in load resource screening, poor accuracy in management plan formulation, and poor real-time management.

Description

TECHNICAL FIELD The present application relates to the technical field of power grid management and control, in particular to a multi-source data driven power supply load resource management and control decision generation method. BACKGROUND The accurate management and control of load management schemes is a top priority for power supply, and traditional load management rigid measures and flexible load resources need to be quickly regulated and coordinated in an orderly manner. High efficiency interaction, both of which promote the efficient use of load resources and the accurate management and control of power supply measures. At present, the following pain points still exist in power supply.

[0001] First, it is difficult to investigate a large amount of load resources. The base of power supply users is huge, and the traditional method does not deeply mine the user attributes and does not combine the digital attribute classification to formulate control measures. The grassroots need to control each household to ensure the overall scheme to be executed in place, which consumes a lot of control effort.

[0002] Second, the speed and accuracy of pre-war scheme preparation is low. Traditional load management scheme preparation relies on user reporting mode and does not consider the change of user response ability under time-of-use electricity price. The grassroots lack digital auxiliary means for optimal scheduling of load resources, and spend a lot of time on user-side docking work and scheme preparation, which has execution deviation.

[0003] Third, it is difficult to execute and control in wartime. Traditional load control has no prediction analysis module and only reports real-time data. Under the requirement of accurate control, the grassroots lack quick decision-making means to respond to daily gap changes, and it is difficult to achieve dynamic adjustment and accurate control of scheme users.

[0004] In view of the technical problems of difficult load resource investigation, poor accuracy of control scheme preparation, and poor real-time control in the related art, no effective solution has been proposed so far. SUMMARY The multi-source data driven power supply load resource management and control decision generation method provided by the embodiments of the present application at least solves the technical problems of difficult load resource investigation, poor accuracy of control scheme preparation, and poor real-time control in the related art.

[0005] This invention provides a multi-source data-driven method for generating power supply load resource management decisions, including: collecting multi-dimensional electricity consumption data from users through the power business system, generating user load benchmark curves adapted to time-of-use pricing, classifying load resources based on users' historical voltage drop performance using a neural network model, and calculating adjustable response parameters for each type of load; quantifying the voltage drop capacity of a single user through time-series load integration, aggregating the voltage drop values ​​of fixed voltage drop loads and basic adjustable loads to obtain the minimum voltage drop; during a pre-set preparation period before the power supply management cycle, using the constraint that the total voltage drop capacity of all users falls within the upper and lower limits of the supply index, calling a genetic algorithm to solve multivariate constraint inequalities, iteratively calculating the optimal initial start-up time combination for each flexibly adjustable user, and generating multiple load management strategies; and after entering the power supply management cycle, performing an optimal selection process. The system selects a load control strategy as the initial execution benchmark, acquires on-site measured load data at fixed time intervals, and calls a rolling time-domain prediction model to fuse the measured load that has occurred and the predicted load for the remaining control period. It calculates the average completion rate of the initial execution benchmark and derives the adjustment range of the remaining control period. When the calculated average completion rate exceeds the preset qualified range, the flexible resource real-time scheduling model is activated. With the objective function of minimizing the deviation between the total voltage drop and the target index, the optimal combination of start-up and shutdown times for each flexible control user is re-solved, and a list of incremental / reduction times for mobile control is output as a corrective control decision. The system receives the latest load data collected by users after the corrective control decision is executed in real time, updates the load gap and scheme completion rate prediction parameters, and sends the updated parameters into the rolling time-domain prediction model to trigger a new round of control prediction and control decision solving iterations.

[0006] As an optional solution, the power business system includes a marketing system, an electricity consumption data collection system, and a load management system; the user's multi-dimensional electricity consumption data includes the user's static business expansion attributes and 1-minute / 15-minute high-frequency time-series historical load data; before connecting to the power business system to collect the user's multi-dimensional electricity consumption data and generating a user load benchmark curve adapted to time-of-use pricing, the method further includes: extracting the user's high-frequency time-series historical load of the past five days from the user's multi-dimensional electricity consumption data, generating an original baseline based on the user's high-frequency time-series historical load of the past five days using the PJM algorithm; segmenting the daily time series according to the preset time-of-use pricing period division rules, recalculating the average value of all time-series load points under the same price period within the original baseline, reconstructing the time-series curve segment by segment, and converting it to obtain a user load benchmark curve adapted to time-of-use pricing.

[0007] As an optional approach, a neural network model is used to classify load resources based on users' historical voltage drop performance and calculate adjustable response parameters for each type of load. This includes: extracting historical voltage drop features from multi-dimensional user electricity consumption data, inputting them into a pre-trained neural network model, and outputting the corresponding user load resource type; wherein, the user load resource type is divided into stable voltage drop users, ordinary fluctuation users, and flexible control users; matching corresponding adjustable response parameters based on the classification results, wherein the adjustable response parameters include: for stable voltage drop users, the adjustable response parameter is a fixed voltage drop constant; for ordinary fluctuation users, the minimum value of the voltage drop range is taken as the basic voltage drop parameter; for flexible control users, there are self-owned power plants, energy storage, and production lines that can be quickly started and stopped, and the voltage drop varies dynamically with the start-up time.

[0008] As an optional approach, the voltage reduction capacity per user is quantified by integrating time-series loads, and the voltage reduction values ​​of fixed voltage reduction loads and basic adjustable loads are aggregated to obtain the minimum voltage reduction. This includes: dividing the time-series load curve into an unresponsive phase, a flexible resource activation and control phase, and a load recovery phase; calculating the baseline electricity consumption for each phase using the integral area of ​​the time-series load curve; and obtaining the voltage reduction capacity per user by combining the baseline electricity consumption for each phase and performing difference calculations.

[0009] As an optional approach, with the constraint that the total voltage drop of all users falls within the upper and lower limits of the supply guarantee index, a genetic algorithm is used to solve the multivariate constraint inequalities. The optimal initial start-up time combination for each flexible control user is iteratively calculated to generate multiple load management strategies. These strategies include: a pre-set preparation period before the arrival of the power supply guarantee management cycle, where the multivariate constraint inequalities are pre-war multivariate constraint inequalities; combining the pre-war multivariate constraint inequalities with the upper and lower limits of the voltage drop range for ordinary fluctuating users to complete the constraint configuration; and under the constraint that the total voltage drop of all users falls within the upper and lower limits of the supply guarantee index, the start-up time of flexible control users is encoded and iteratively encoded using a genetic algorithm to obtain the optimal initial start-up time combination for each flexible control user, thus generating multiple load management strategies.

[0010] As an optional approach, after entering the power supply guarantee and control cycle, a set of load control strategies is determined as the initial execution benchmark through a selection and screening operation. This includes: prioritizing the selection of load control strategies with a total voltage drop close to the median value of the supply guarantee target as primary candidate strategies; if there is only one primary candidate strategy, it is directly used as the initial execution benchmark; if there are multiple primary candidate strategies, the primary candidate strategy with a smaller flexible resource investment scale is prioritized as a secondary candidate strategy; if there is only one secondary candidate strategy, it is directly used as the initial execution benchmark; if there are still multiple secondary candidate strategies, the secondary strategy with a more balanced distribution of flexible resource start-up and shutdown periods is selected as a tertiary candidate strategy; if there is only one tertiary candidate strategy, it is directly used as the initial execution benchmark; if there are multiple tertiary candidate strategies, one is randomly selected as the initial execution benchmark.

[0011] As an optional approach, the fixed time interval is 10-30 minutes. On-site measured load data is acquired at fixed time intervals. A rolling time-domain prediction model is used to fuse the measured load that has occurred and the predicted load for the remaining control period. The average completion rate of the initial execution benchmark is calculated, and the regulation power range for the remaining control period is derived. This includes: after the control cycle starts and reaches the first observation node, acquiring all time-series measured load data from the start of control to the current observation node, and solving for the average completion rate of the initial execution benchmark; using the linear regression error term built into the rolling time-domain prediction model, identifying and eliminating regular load fluctuation interference caused by planned start-up and shutdown loads, and correcting the average completion rate of the initially calculated initial execution benchmark; and solving for the regulation power range required for the remaining control period based on the completion rate correction result.

[0012] As an optional approach, the preset acceptable range is a completion rate of 100% to 110%. A flexible resource real-time scheduling model is activated, using the minimization of the deviation between the total voltage drop and the target indicator as the objective function. The optimal combination of start-up and stop times for each flexible control user is re-solved, and a list of incremental / reduced flexible control measures is output as a corrective control decision. This includes: when the average completion rate exceeds the preset acceptable range, after activating the flexible resource real-time scheduling model, in the process of solving for the optimal combination of start-up and stop times for each flexible control user, dual constraints of maximum adjustable power and allowable control time window are configured, and a wartime multivariate constraint inequality is constructed based on the real-time calculated range of adjustable power. Under the combined constraints of the dual constraints and the wartime multivariate constraint inequality, the flexible resource real-time scheduling model completes the solution with the minimization of the deviation between the total voltage drop and the target indicator as the objective function, and automatically outputs a list of incremental / reduced flexible control measures as a corrective control decision.

[0013] Another aspect of the present invention provides an electronic device, comprising: a processor, and a memory storing a program, the program including instructions that, when executed by the processor, cause the processor to perform the method described in any one of the preceding descriptions.

[0014] Another aspect of the present invention provides a non-transitory machine-readable medium storing computer instructions for causing the computer to perform any of the methods described above.

[0015] The multi-source data-driven power supply load resource management and control decision generation method provided in this invention classifies load resources and accurately quantifies voltage reduction based on multi-source electricity consumption data, and calculates the total guaranteed voltage reduction for rigid loads. During the management and control preparation phase, a genetic algorithm is used to generate multiple alternative management and control strategies that meet the grid voltage reduction target constraints. After management and control are initiated, the initial execution plan is automatically selected based on the best option, and the management and control completion status is continuously evaluated based on rolling time-domain prediction. When the execution results deviate from the reasonable range, the flexible resource start-up and shutdown arrangements are automatically re-optimized and a correction list is output. Simultaneously, the prediction and optimization process is continuously iterated based on the latest measured load. The entire solution constructs a two-level time-scale control framework of "pre-war multi-strategy pre-optimization + wartime rolling evaluation + closed-loop dynamic correction," changing the traditional mode of coarse instantaneous power control and one-time execution of static plans. This improves the accuracy of power supply load resource voltage reduction calculation, the adaptability of control plans, and the self-correction capability of the execution process. While ensuring the grid voltage reduction target is met, the scale of flexible load dispatch is optimized, reducing the impact on the normal production order of industry and commerce. Attached Figure Description To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a multi-source data-driven power supply load resource management and control decision generation method according to an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram illustrating a single user performing modeling and analysis according to an embodiment of the present invention.

[0018] Figure 3 This is a schematic diagram of the power supply guarantee and control application process according to an embodiment of the present invention.

[0019] Figure 4 This is a schematic diagram of the average completion rate prediction model according to an embodiment of the present invention.

[0020] Figure 5 This is a schematic diagram of a flexible real-time resource scheduling model according to an embodiment of the present invention.

[0021] Figure 6 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0022] To address the aforementioned technical problems, embodiments of the present invention provide a multi-source data-driven method for generating power supply load resource management and control decisions, such as... Figure 1 As shown, the method includes the following steps.

[0023] Step S101: Connect to the power business system to collect multi-dimensional electricity consumption data of users, generate user load benchmark curves adapted to time-of-use pricing, classify load resources according to users' historical voltage drop performance through a neural network model and calculate the adjustable response parameters of various types of loads; quantify the voltage drop of a single user through time-series load integration, and aggregate the voltage drop values ​​of fixed voltage drop loads and basic adjustable loads to obtain the minimum voltage drop total.

[0024] Step S102: During the preset preparation period before the arrival of the power supply guarantee and control cycle, with the total voltage drop of all users falling within the upper and lower limits of the supply guarantee index as a constraint, the genetic algorithm is called to solve the multivariate constraint inequality, and the optimal initial start-up time combination of each flexible control user is iteratively calculated to generate multiple load control strategies.

[0025] Step S103: After entering the power supply guarantee and control cycle, perform the best selection and screening operation to determine a load control strategy as the initial execution benchmark, obtain the on-site measured load data at fixed time intervals, call the rolling time domain prediction model to integrate the measured load that has occurred and the predicted load of the remaining control period, calculate the average completion rate of the initial execution benchmark, and derive the adjustment power range of the remaining control period.

[0026] Step S104: When the calculated average completion rate exceeds the preset qualified range, the flexible resource real-time scheduling model is started. The objective function is to minimize the deviation between the total voltage drop and the target index. The optimal combination of start and stop times for each flexible control user is re-solved, and the list of mobile control increments / reductions is output as a corrective control decision.

[0027] Step S105: Receive the latest load data collected by users after the execution of the corrective control decision in real time, update the load gap and scheme completion rate prediction parameters, and send the updated parameters into the rolling time domain prediction model to trigger a new round of control prediction and regulation decision solving iteration.

[0028] The multi-source data-driven power supply load resource management and control decision generation method provided in this invention classifies load resources and accurately quantifies voltage drop based on multi-source electricity consumption data, and calculates the total guaranteed voltage drop formed by rigid loads. During the management and control preparation stage, a genetic algorithm is used to generate multiple alternative management and control strategies that meet the grid voltage drop index constraints. After management and control are initiated, the initial execution plan is automatically selected based on the best one, and the management and control completion status is continuously evaluated based on rolling time-domain prediction. When the execution results deviate from the reasonable range, the flexible resource start-up and shutdown arrangements are automatically re-optimized and a correction list is output. At the same time, the prediction and optimization process is continuously iterated based on the latest measured load.

[0029] The entire solution constructs a two-level time-scale control framework of "pre-war multi-strategy pre-optimization + wartime rolling assessment + closed-loop dynamic correction". It changes the traditional mode of extensive control of instantaneous power and one-time execution of static schemes, improves the accuracy of power supply load resource pressure reduction calculation, the adaptability of control schemes and the self-correction capability of the execution process, optimizes the scale of flexible load dispatch on the premise of ensuring that the grid pressure reduction target is achieved, and reduces the impact on the normal production order of industry and commerce.

[0030] The aforementioned power business systems may include marketing systems, electricity consumption data collection systems, load management systems, and other business platforms capable of outputting user electricity consumption profiles and time-series load data. The corresponding multi-dimensional user electricity consumption data may include user static attributes, historical 1 / 15-minute high-frequency time-series load data, historical power supply guarantee voltage drop execution records, and other multi-source information.

[0031] The aforementioned user load baseline curve serves as a time-series electricity consumption curve that acts as a reference standard for voltage drop calculation. It distinguishes different time-of-use pricing periods and is used to calculate the theoretical baseline electricity consumption under control scenarios.

[0032] Time-series load integration method: With time as the horizontal axis and load power as the vertical axis, the area of ​​the time-series load curve is calculated to obtain the electricity consumption for the corresponding time period, thereby realizing the quantification of electricity consumption, which is different from the single-point instantaneous power judgment method.

[0033] Fixed voltage drop loads and adjustable base loads: These correspond to users with stable voltage drop and users with ordinary fluctuations, respectively. Users with stable voltage drop can achieve a fixed voltage drop. Users with ordinary fluctuations have a voltage drop range, and the minimum value of the range is taken as the minimum adjustable capacity.

[0034] Minimum voltage reduction: The total voltage reduction that can be stably achieved by relying solely on stable voltage reduction users and ordinary fluctuating users, without relying on flexible resources such as energy storage and self-owned power plants, serves as the basic foundation of the management and control scheme.

[0035] Connect with the power business system to collect multi-dimensional electricity consumption data from users, and integrate multi-source electricity consumption data across systems to build a load reference baseline.

[0036] By using a neural network model to classify load resources based on users' historical load drop performance and calculating the adjustable response parameters for each type of load, the system leverages the neural network to uncover historical load drop behavior characteristics, automatically distinguishing three types of load resources and assigning their respective adjustable potential. This enables differentiated profiling of heterogeneous load resources, addressing the problem of homogeneous management. The voltage reduction capacity of a single user is quantified by time-series load integration, and the voltage reduction values ​​of fixed voltage reduction loads and basic adjustable loads are aggregated to obtain the total guaranteed voltage reduction. Time-series integration is used to replace single-point power measurement to accurately calculate the voltage reduction capacity that a single user can contribute throughout the entire control cycle; the guaranteed voltage reduction baseline is obtained by summarizing rigid adjustable loads.

[0037] This allows for the quantification of voltage drop capability in terms of power consumption, avoiding the limitation that instantaneous power indicators cannot measure the overall effectiveness over the entire period, and providing precise input boundaries for subsequent strategy optimization.

[0038] The aforementioned neural network model can be a multi-layer feedforward BP neural network (supervised classification neural network); its basic architecture includes an input layer, at least two hidden layers, and an output layer. The input layer is used to input historical voltage drop feature vectors such as the duration of historical voltage drop, the magnitude of historical voltage drop, the load fluctuation coefficient, and production period characteristics. The activation function of the hidden layers is ReLU. The output layer is configured with a three-class output, corresponding to users with stable voltage drops, users with moderate fluctuations, and users with flexible control.

[0039] During training, historical supply guarantee case datasets are used as samples; sample labels are manually labeled based on past control records to determine user types; gradient descent algorithm is used to complete offline training; after training converges, the model has fixed weights, and only forward inference is performed in the online stage to achieve automatic user classification and synchronously output adjustable response parameters corresponding to each type of user.

[0040] The aforementioned power supply guarantee and control period is the window for implementing grid demand response. The pre-set preparation period is before the arrival of the control period (T-1 day pre-battle phase), used to pre-generate multiple alternative load control strategies. The upper and lower limits of the aforementioned supply guarantee indicators represent the allowable range of total voltage reduction that the grid needs to achieve in the designated areas.

[0041] The aforementioned multivariate constraint inequalities may include constraints such as power balance, user-adjustable upper and lower limits, and control time windows; they are divided into pre-war multivariate constraint inequalities in the preparation phase and wartime multivariate constraint inequalities in the control and execution phase.

[0042] The aforementioned flexible load users are a type of user load resource, generally including self-owned power plants, energy storage, and production lines that can be started and stopped quickly. These flexible load resources have adjustable start and stop times and their voltage drop capacity varies with the start-up time, making them one of the main targets for resource regulation.

[0043] The aforementioned standardized load management strategy can include a complete pre-battle management plan that includes various user load reduction arrangements and flexible resource start-up time combinations, generating multiple alternatives at once.

[0044] During the pre-set preparation period before the arrival of the power supply guarantee and control cycle, with the total voltage drop of all users falling within the upper and lower limits of the supply guarantee index as a constraint, a genetic algorithm is called to solve the multivariate constraint inequalities, iteratively calculate the optimal initial start-up time combination for each flexible control user, and generate multiple load control strategies.

[0045] Before the implementation of control measures, pre-optimization is completed; the range of power grid voltage drop index is used as a hard constraint, and the global optimization capability of the genetic algorithm is used to solve the nonlinear constraint optimization problem. Different combinations of flexible resource start-up timings are traversed to generate multiple feasible alternative solutions in parallel.

[0046] This allows for the early completion of pre-war plan reserves, overcoming the limitations of relying solely on a single plan; by using a multi-strategy solution set to cover different load fluctuation scenarios, it provides a pool of alternative plans for the start of the control cycle, thereby improving the robustness of the plans.

[0047] The aforementioned genetic algorithm is an intelligent global optimization algorithm used to solve nonlinear optimization problems with inequality constraints. Specifically, it first encodes the start time of each flexible control user with real numbers; sets an evolutionary mechanism, including selection, crossover, and mutation to iteratively update the population; and constructs a fitness function based on the matching degree between the total voltage drop and the supply guarantee index.

[0048] The goal is to find a flexible resource start-up timing combination that satisfies the pre-war multivariate constraint inequalities and whose total voltage drop falls within the index range; after the algorithm terminates, multiple feasible solutions are retained to form multiple sets of standardized control strategies.

[0049] The above-mentioned selection and screening operation: at the start of the control cycle, a hierarchical judgment logic is automatically selected from multiple pre-generated strategies as the initial execution benchmark.

[0050] Rolling Time-Domain Forecasting Model: This model uses a sliding time window to fuse measured load data from the past period with the predicted load for the remaining time period, continuously updating the load trend forecast results. The above rolling time-domain forecasting model employs the rolling window forecasting framework (RHP).

[0051] Specifically, a sliding time window architecture is adopted, with the window including the executed time period and the future control time period to be executed. It continuously absorbs the latest measured load data to update the model input and embeds a linear regression error correction term inside.

[0052] It does not require online retraining and is an online inference iterative framework; it continuously uses real-time measured load to correct the load forecast results for subsequent periods.

[0053] The above average completion rate is the ratio of the actual cumulative reduction in electricity consumption to the planned reduction in electricity consumption, and is used to evaluate the effectiveness of the control and management implementation.

[0054] When the control system is officially launched, the initial execution benchmark will be automatically selected based on the best performance. The specific method will be explained later. Then, the actual load on site will be continuously collected at fixed intervals. The rolling time-domain model will continuously use the measured data to correct future load forecasts and calculate the voltage drop completion status and the power reserve that needs to be supplemented in the future in real time.

[0055] It enables seamless integration between pre-planned strategies and real-time operational conditions; it continuously senses deviations in on-site execution, anticipates gaps in advance, provides quantitative basis for dynamic correction, and eliminates the shortcoming of static plans failing to keep pace with load changes.

[0056] The above-mentioned preset qualified range is the reasonable range for the completion rate. If it exceeds this range, it is determined that there is a significant deviation in the pressure reduction gap, and dynamic correction needs to be initiated.

[0057] The aforementioned flexible resource real-time scheduling model is a dynamic optimization solution module for the control process, which re-optimizes the flexible resource start-up and shutdown combinations with the objective of minimizing voltage drop deviation. The flexible resource real-time scheduling model can be a constrained continuous optimization model. The optimization objective is to minimize the deviation between the total voltage drop and the supply guarantee target.

[0058] The constraint set includes maximum adjustable power constraints, allowable control time window constraints, and wartime multivariate constraint inequalities. The solver can use sequential quadratic programming or lightweight genetic algorithms for online fast optimization, enabling dynamic rescheduling of the control process.

[0059] The aforementioned list of incremental / reduced flexible resource allocations is a wartime corrective decision-making output document that specifies the arrangements for the addition or withdrawal of flexible resources.

[0060] When the calculated average completion rate exceeds the preset qualified range, the flexible resource real-time scheduling model is activated. With the objective function of minimizing the deviation between the total voltage drop and the target indicator, the optimal combination of start-up and stop times for each flexible control user is re-solved, and the list of incremental / reduction of mobile control is output as a corrective control decision.

[0061] A preset acceptable range is set as the correction trigger condition. Once the control effect deviates from the reasonable range, the optimization problem is reconstructed, and the optimal solution for start-up and shutdown sequence is found within the flexible resource adjustment boundary, and the incremental or reduction control arrangement is output.

[0062] This enables proactive correction of deviations in the control process, dynamically balancing the pressure reduction target requirements with the impact on user production, and avoiding the continuous expansion of the pressure reduction gap or unnecessary over-regulation.

[0063] The system receives the latest load data collected from users in real time after the implementation of corrective control decisions, updates the load gap and plan completion rate prediction parameters, and sends the updated parameters into the rolling time domain prediction model to trigger a new round of control prediction and regulation decision solving iteration.

[0064] This forms a continuous closed-loop system of "issuing corrective commands → collecting the latest load data → updating forecast parameters → re-forecasting and evaluating." This constructs a continuously adaptive iterative control framework that continuously optimizes based on on-site operating conditions, achieving dynamic closed-loop management throughout the entire process.

[0065] As an optional embodiment, the power business system includes a marketing system, an electricity consumption data collection system, and a load management system; the user's multi-dimensional electricity consumption data includes the user's static business expansion attributes and 1-minute / 15-minute high-frequency time-series historical load data; before collecting the user's multi-dimensional electricity consumption data from the power business system and generating a user load benchmark curve adapted to time-of-use pricing, the method further includes: extracting the user's high-frequency time-series historical load of the past five days from the user's multi-dimensional electricity consumption data, generating an original baseline based on the user's high-frequency time-series historical load of the past five days using the PJM algorithm; segmenting the daily time series according to the preset time-of-use pricing period division rules, recalculating the average value of all time-series load points under the same price period within the original baseline, reconstructing the time-series curve segment by segment, and converting it to obtain a user load benchmark curve adapted to time-of-use pricing.

[0066] The aforementioned marketing system can serve as a business backend for storing static business expansion files and time-of-use electricity pricing contract information for industrial and commercial users, providing basic user attributes and electricity time period division rules. The electricity consumption data acquisition system collects high-frequency time-series power / energy consumption data from user's on-site meters and terminals, outputting load time series data at 1-minute and 15-minute granularities. The load management system stores historical power supply control records and user's past voltage reduction execution ledgers, providing multi-period historical load samples.

[0067] The high-frequency time-series historical load data for the past five days consists of 15-minute granular load sequences from the five consecutive non-holiday working days prior to the control event, serving as the standard input sample for the PJM baseline. The original baseline is a basic load reference curve based solely on the historical load average without overlaying time-of-use pricing segmentation corrections.

[0068] The above-mentioned time-of-use pricing rules can serve as a fixed-time slicing standard for peak / flat / valley time periods in the power grid's industrial and commercial time-of-use pricing, used for segmenting the 24-hour time sequence of a day.

[0069] Existing technologies only retrieve single-source load data, resulting in incomplete data dimensions and a lack of user profiles and electricity price contract information to support baseline correction. This embodiment integrates three types of heterogeneous business systems, simultaneously acquiring static profiles and two levels of dynamic time-series load data to form a complete multi-source input dataset. It takes into account three key types of information: user identity, electricity consumption time series, and electricity price contract, providing full-dimensional raw data for subsequent baseline reconstruction and neural network feature extraction.

[0070] Eliminate the problem of missing information from a single data source, ensure that load profiles and voltage drop calculations have user-specific attributes, and improve the matching degree between baseline curves and actual power consumption scenarios.

[0071] The system extracts the user's high-frequency time-series historical load data from the past five days within the user's multi-dimensional electricity consumption data. Based on this historical load data, the PJM algorithm is used to generate an original baseline. The PJM baseline is a general multi-similarity daily average baseline algorithm for electricity demand response, employing the statistical logic of "selecting the load of the past five qualified working days and taking the average value at the same time point."

[0072] Select the 15-minute time-series loads of normal working days with no control, no production stoppages, and no maintenance in the past 5 days; calculate the arithmetic mean of the load power at the same time section each day (e.g., 9:00, 9:15). Piece together the daily average series in chronological order to obtain the original baseline curve of the unadapted electricity price.

[0073] Using five consecutive days of stable historical load as a sample, and eliminating abnormal load disturbances caused by single-day production stoppages and maintenance, a standard electricity consumption curve that can represent the user's daily uncontrolled state is constructed as the underlying benchmark for voltage drop comparison.

[0074] The baseline is based on multiple days of historical data to smooth out abnormal fluctuations, avoiding distortion of the benchmark caused by special operating conditions on a single day, and ensuring the accuracy of the subsequent voltage drop power integration calculation.

[0075] The daily time sequence is segmented according to the preset time-of-use pricing rules. The average value of all time-of-use load points under the same price period within the original baseline is recalculated. The time sequence curve is reconstructed segment by segment and transformed to obtain the user load benchmark curve adapted to the time-of-use pricing.

[0076] The process is implemented step by step. Based on the peak / flat / valley time-of-use electricity pricing rules issued by the marketing system, the 24-hour time series is divided into multiple electricity price intervals. All 15-minute time series points of the original baseline are traversed and grouped according to their respective electricity price periods. The average value of all load points within the same electricity price period is calculated twice, and the original fine-grained points are replaced with the segmented average value. The average value sequences of each electricity price segment are spliced ​​together in chronological order to reconstruct a new benchmark curve that is segmented, smooth, and conforms to the electricity price mechanism.

[0077] The original PJM baseline only averaged the usage over the same day, failing to account for the differences in electricity consumption behavior among industrial and commercial users during time-of-use pricing. Users typically reduce production load during peak price periods and increase capacity during off-peak periods, and the original baseline could not differentiate between these price-related behavioral differences. This step involves reconstructing the baseline curve based on a second-order average of electricity prices over different time periods, ensuring the curve aligns with users' time-of-use habits and allowing voltage drop calculations to differentiate the adjustable potential across different price ranges.

[0078] This addresses the incompatibility between the general PJM baseline and the domestic time-of-use pricing mechanism, ensuring that the benchmark curve aligns with the time-of-use electricity consumption characteristics of industrial and commercial users, and significantly reducing the error in the integral calculation of voltage drop over all time periods.

[0079] As an optional embodiment, a neural network model is used to classify load resources based on users' historical voltage drop performance and calculate the adjustable response parameters of various types of loads. This includes: extracting users' historical voltage drop features from multi-dimensional electricity consumption data, inputting them into a pre-trained neural network model, and outputting the corresponding user load resource type; wherein, the user load resource type is divided into stable voltage drop users, ordinary fluctuation users, and flexible control users.

[0080] Based on the classification results, corresponding adjustable response parameters are matched. The adjustable response parameters include: for users with stable voltage drop, the adjustable response parameter is a fixed voltage drop constant; for users with ordinary fluctuations, the minimum value of the voltage drop range is taken as the basic voltage drop parameter; for users with flexible control, including self-owned power plants, energy storage, and production lines that can be quickly started and stopped, the voltage drop power changes dynamically with the start-up time.

[0081] User historical voltage drop characteristics can be feature vectors extracted from collected multi-dimensional electricity data, including quantitative indicators such as voltage drop magnitude during historical control periods, voltage drop duration, load fluctuation rate, peak-valley electricity difference, past control execution compliance rate, and production maintenance cycle.

[0082] Adjustable response parameters are quantitative indicators used to characterize the stable voltage reduction capacity that a single user can contribute during power supply control. The value rules for different types of users are completely different. The fixed voltage reduction constant is the fixed voltage reduction that a stable voltage reduction user can achieve during the control period, unaffected by fluctuations in electricity prices and production cycles.

[0083] The minimum value of the voltage reduction range is the lower limit of the range, which is taken as the guaranteed adjustable base voltage reduction for ordinary fluctuating users whose voltage reduction is affected by production orders and time-of-use electricity prices.

[0084] Existing technologies rely on human experience to classify user types, which is highly subjective and has low efficiency in batch classification. This embodiment is based on objective historical pressure drop quantification features and automatically completes batch classification through a supervised neural network. It uses massive historical regulation data as training basis, unifies classification judgment criteria, and eliminates human judgment bias.

[0085] It enables automated and standardized classification of massive industrial and commercial loads, adapting to large-scale regional supply guarantee and control scenarios; it unifies classification and judgment standards, avoiding the confusion of load classification caused by manual division, and ensuring the consistency of subsequent voltage drop calculation boundaries.

[0086] Match exclusive adjustable response parameters according to user type, and clarify the parameter value rules for three types of users: users with stable pressure drop have a fixed production mode, and the pressure drop is not affected by external conditions during the control period. The adjustable capacity is directly represented by a fixed constant, and can be fully included in the minimum pressure drop.

[0087] For ordinary fluctuating users, electricity consumption fluctuates with orders and electricity prices, and voltage reduction has a fluctuating range. The minimum value of the range is conservatively selected to calculate the minimum voltage reduction to ensure that the lower limit of the control plan is met.

[0088] Self-owned power plants, energy storage, and adjustable production lines are flexible resources, and their start-up and shutdown times can be freely optimized. There is no fixed minimum voltage drop, and the voltage drop value is dynamically calculated by subsequent genetic algorithms and real-time scheduling models to optimize the start-up and shutdown periods.

[0089] This allows for accurate calculation of the aggregated pressure drop, resulting in the guaranteed total pressure drop. A conservative value is chosen for the rigid load pressure drop boundary to ensure the reliability of the lower limit of the pre-war control plan. By distinguishing between rigid and flexible loads, tiered resource management is achieved, reducing the frequency of subsequent flexible resource allocation and minimizing disruption to enterprise production.

[0090] As an optional implementation, the voltage drop of a single user is quantified by integrating the time-series load, and the voltage drop values ​​of fixed voltage drop loads and basic adjustable loads are aggregated to obtain the minimum voltage drop total. This includes: dividing the time-series load curve into an unresponsive stage, a flexible resource activation and control stage, and a load recovery stage; calculating the baseline electricity consumption for each stage by integrating the area of ​​the time-series load curve; and obtaining the voltage drop of a single user by combining the baseline electricity consumption for each stage through difference calculation.

[0091] The above three-stage time sequence division is based on the user's complete electricity consumption time sequence load curve, and the entire demand response process is divided into three segments: the non-response stage, which is the period when the power grid has not issued control instructions and the user's normal baseline electricity consumption period.

[0092] The flexible resource activation and control phase is the core period for implementing pressure reduction and control, equipment production limitation / temperature control adjustment / peak shifting.

[0093] The load recovery phase is the period after the control measures end, equipment restarts, and electricity consumption rebounds (the rebound load will offset the voltage drop, which must be calculated separately).

[0094] The integral calculation of the time-series load curve of the baseline electricity consumption for each stage is a continuous power-time curve. The area enclosed by the curve and the time axis equals the total electricity consumption for that period. Integrating the three curve segments respectively yields the baseline electricity consumption for each segment.

[0095] The difference calculation yields the single-user voltage drop capacity. Voltage drop capacity = Baseline power consumption during the non-response phase. The baseline electricity consumption during the control phase; the rebound electricity consumption during the load recovery phase is deducted and corrected separately. Finally, the fixed voltage drop load (fixed equipment capacity, fixed voltage drop value) + the basic adjustable load (flexible adjustment) are aggregated to obtain the user's minimum voltage drop, which is the minimum electricity reduction limit that the user can achieve, and serves as the global constraint input.

[0096] This embodiment uses time-series integrals to replace the traditional single-point average baseline, dynamically distinguishing rebound loads and significantly improving quantification accuracy; the three-segment split fully covers the entire life cycle of demand response, solving the industry calculation error problem of "rebound offsetting pressure drop after regulation"; it distinguishes between two types of adjustable loads and aggregates them separately, and the guaranteed pressure drop can be used as a reliable lower limit constraint for global optimization.

[0097] As an optional implementation, with the constraint that the total voltage drop of all users falls within the upper and lower limits of the supply guarantee index, a genetic algorithm is used to solve the multivariate constraint inequalities, iteratively calculating the optimal initial start-up time combination for each flexible control user, and generating multiple load control strategies. These strategies include: during a pre-set preparation period before the arrival of the power supply guarantee control cycle, the multivariate constraint inequalities are pre-war multivariate constraint inequalities; the pre-war multivariate constraint inequalities are combined with the upper and lower limits of the voltage drop range for ordinary fluctuating users to complete the constraint configuration; under the constraint that the total voltage drop of all users falls within the upper and lower limits of the supply guarantee index, the start-up time of flexible control users is encoded and iteratively encoded using a genetic algorithm to obtain the optimal initial start-up time combination for each flexible control user, generating multiple load control strategies.

[0098] During the preparation period before the supply guarantee and control period (pre-battle), with the total voltage drop of the entire network at the upper and lower limits of the supply guarantee target as a hard constraint, the initial time for all flexible users to start control is optimized, multiple feasible control plans are generated in batches, and alternative strategies are reserved for dispatching to avoid supply gaps caused by single plan failures or user non-response.

[0099] Construct a pre-war multivariate constraint inequality with the total voltage drop of all users ∈ the lower limit to the upper limit of the supply guarantee index as the core hard constraint; at the same time, superimpose the upper and lower limits of the voltage drop range of each user with ordinary fluctuations, and each user's own power reduction cannot exceed its own maximum adjustable value and cannot be lower than the guaranteed voltage drop, complete all constraint configurations, and form a multivariate inequality equation system.

[0100] The variable is the initial startup time of each user with flexible control (discrete time series variable).

[0101] Pre-war optimization can generate multiple contingency plans in batches before the supply guarantee arrives, rather than calculating a single plan on an ad-hoc basis, thereby improving emergency redundancy; multi-layered constraint coupling takes into account both the total supply guarantee indicators of the entire network and the adjustable upper and lower limits of a single user, avoiding the plan from deviating from the user's actual adjustable capabilities.

[0102] Genetic algorithms are used to solve discrete-time multivariate inequalities, which are suitable for nonlinear optimization scenarios with discrete startup times and multivariate coupling, and are difficult to solve using traditional linear programming.

[0103] As an optional implementation, after entering the power supply guarantee and control cycle, a set of load control strategies is determined as the initial execution benchmark through a selection and screening operation. This includes: prioritizing the selection of load control strategies with a total voltage drop close to the median value of the supply guarantee index as primary candidate strategies; if there is only one primary candidate strategy, it is directly used as the initial execution benchmark; if there are multiple primary candidate strategies, the primary candidate strategy with a smaller flexible resource investment scale is prioritized as the secondary candidate strategy; if there is only one secondary candidate strategy, it is directly used as the initial execution benchmark; if there are still multiple secondary candidate strategies, the secondary strategy with a more balanced distribution of flexible resource start-up and shutdown periods is selected as the tertiary candidate strategy; if there is only one tertiary candidate strategy, it is directly used as the initial execution benchmark; and if there are multiple tertiary candidate strategies, one is randomly selected as the initial execution benchmark.

[0104] After the supply guarantee period officially starts, the single optimal benchmark solution is selected from the multiple load control strategies generated, according to a four-level priority tier. This takes into account supply guarantee compliance, resource input costs, and control balance, forming a standardized selection logic that eliminates the need for subjective human judgment.

[0105] The primary screening strategy prioritizes selecting the strategy with the total voltage reduction closest to the median value of the supply guarantee target. A voltage reduction close to the median value avoids both insufficient reduction (risk of power shortage) and excessive reduction (excessive power loss and compensation costs for enterprises), maximizing the margin for error.

[0106] The secondary screening process prioritizes solutions with fewer users / equipment requiring intervention from the primary candidates. Smaller intervention scales result in lower user compensation costs, less disruption to industrial and commercial production, and lower scheduling execution costs.

[0107] The third-level screening selects strategies from the second-level candidates that allow for flexible resource activation and more evenly distributed shutdown periods. This avoids a concentrated load drop during the same period, prevents sudden drops / rebounds in the power grid during localized periods, and reduces the risk of short-term power grid impacts.

[0108] If there are still multiple sets, then proceed to the final stage to randomly select one set as the initial execution benchmark.

[0109] The four-tiered standardized selection process forms objective decision-making rules that can be automatically screened through a program, eliminating the need for manual judgment by dispatchers. The selection priority aligns with the three core requirements of the power grid: power supply security → economic cost → grid stability, with a logical progression. The combination of multiple contingency plans and standardized screening addresses the shortcomings of emergency response when a single strategy fails or there are no alternative solutions.

[0110] As an optional implementation, the fixed time interval is 10-30 minutes; on-site measured load data is acquired at fixed time intervals, and the rolling time-domain prediction model is called to fuse the measured load that has occurred and the predicted load for the remaining control period, calculate the average completion rate of the initial execution benchmark, and derive the regulation power range for the remaining control period, including: after the start of the control cycle, reaching the first observation node, acquiring all time-series measured load data from the start of control to the current observation node, and solving for the average completion rate of the initial execution benchmark; through the linear regression error term built into the rolling time-domain prediction model, identifying and eliminating regular load fluctuation interference caused by planned start-up and shutdown loads, and correcting the average completion rate of the initial execution benchmark obtained in the initial calculation; based on the completion rate correction result, the regulation power range required for the remaining control period is solved.

[0111] By performing high-frequency rolling verification every 10–30 minutes, execution deviations are corrected in real time, and the amount of electricity that needs to be adjusted up or down is accurately calculated. This solves the problems of inaccurate traditional static scheduling, large interference from planned start-up and shutdown loads, and unclear calculation of remaining adjustable electricity.

[0112] Once the entire power supply guarantee and control cycle is initiated, every 10–30 minutes serves as an observation node. All real measured load data from the start of control to the current moment are captured, compared with the preset control benchmark target, and the average completion rate of the current control plan is calculated to determine whether the current execution progress is too fast, too slow, meets the standard, or exceeds the standard.

[0113] The power grid contains a large number of planned start-stop loads (industrial loads, energy storage, and adjustable equipment that switch on and off at fixed times), which will produce regular fluctuations, causing the completion rate calculated from the actual measured data to be distorted and misjudged.

[0114] This embodiment uses the built-in linear regression error term in the rolling time-domain prediction model to automatically identify such regular human-induced fluctuations, eliminate interference, calibrate the initial completion rate, and obtain a true, interference-free corrected completion rate.

[0115] Using the corrected accuracy completion rate, combined with the load predicted by the model for subsequent periods, we can reverse the calculation to determine the range of power regulation that the power grid needs to adjust upwards or downwards during the remaining control period, thus providing accurate data for subsequent corrective scheduling.

[0116] By using high-frequency rolling, the inflexibility of fixed 24-hour plans is avoided, enabling minute-level dynamic verification. Furthermore, it boasts strong anti-interference capabilities: specifically eliminating regular errors and anomalies in planned load starts and stops. It can also accurately output the remaining adjustable power range, providing a safety net for precise decision-making.

[0117] As an optional implementation, the preset acceptable range is a completion rate of 100% to 110%. The flexible resource real-time scheduling model is activated, using the minimization of the deviation between the total voltage drop and the target indicator as the objective function. The optimal combination of start-up and shutdown times for each flexible control user is re-solved, and a list of incremental / reduced flexible control measures is output as a corrective control decision. This includes: when the average completion rate exceeds the preset acceptable range, after activating the flexible resource real-time scheduling model, in the process of solving for the optimal combination of start-up and shutdown times for each flexible control user, dual constraints of maximum adjustable power and allowable control time window are configured, and a wartime multivariate constraint inequality is constructed based on the real-time calculated range of adjustable power. Under the combined constraints of the dual constraints and the wartime multivariate constraint inequality, the flexible resource real-time scheduling model completes the solution with the minimization of the deviation between the total voltage drop and the target indicator as the objective function, and automatically outputs a list of incremental / reduced flexible control measures as a corrective control decision.

[0118] When the real-time completion rate exceeds or fails to meet the target, the intelligent correction model is automatically activated. Under multiple real-world constraints, it calculates the optimal start-stop combination for all adjustable users and outputs a precise list of load increases and decreases, thus achieving closed-loop management.

[0119] A completion rate of 100%–110% is considered acceptable. Below 100% indicates insufficient voltage reduction and inadequate power supply. Above 110% indicates excessive voltage reduction, wasting load resources and impacting user power consumption. Exceeding these ranges will trigger real-time corrective dispatch immediately.

[0120] When the corrected average completion rate falls outside the 100%–110% acceptable range, the flexible resource real-time scheduling model is immediately activated. With the core objective of minimizing the deviation between the total voltage drop and the target voltage, the start and stop times of all adjustable users are re-optimized.

[0121] The flexible resource real-time scheduling model does not perform unlimited optimization during solution processing, but strictly adheres to the actual situation of the power grid and is bound by three layers of constraints: Resource constraints: Maximum adjustable power limit for each flexible user (cannot exceed equipment capacity); Time constraints: Each user is allowed to adjust the time window (cannot start or stop arbitrarily, and must avoid peak production / electricity consumption periods); Wartime constraints: Based on the calculated real-time power adjustment range, a wartime multivariate constraint inequality is constructed to adapt to emergency supply scenarios.

[0122] The flexible resource real-time scheduling model iteratively solves under multiple constraints, allowing the overall power deviation to approach 0 infinitely. Finally, it automatically outputs a list of dynamic control increments / decreases, which clearly identifies which users increase their load, which users decrease their load, and the start and stop times, serving as the final corrective control decision for direct implementation.

[0123] By setting a completion rate threshold, the system automatically corrects deviations when targets are exceeded, preventing control failures. The flexible resource real-time scheduling model employs multiple constraints that closely align with engineering realities, remaining within the bounds of equipment limits, user power consumption rules, and wartime grid requirements. The model directly outputs an executable user control list for convenient subsequent use.

[0124] It should be noted that this embodiment also provides an optional implementation method, which will be described in detail below. From a system construction perspective, this embodiment provides implementation details of a data-driven method for precise control and decision analysis of power supply load resources.

[0125] The primary data source is the marketing system, data collection system, and load management system. Core data information includes: User business expansion application information, such as industry, capacity, and new energy information such as photovoltaic / energy storage, is used to obtain the user's inherent attributes, which are then applied to the tag-based management model to generate static user tags.

[0126] User load data is collected to obtain historical load / power data, analyze and calculate its load characteristics, apply it to the tag-based management model, and generate dynamic user tags; instantaneous load data of users is also collected (collection frequency 15 minutes / 1 minute) for use in the completion rate prediction model.

[0127] User baseline information (average load curve of the five days prior to the user response date) under the PJM algorithm is used to obtain historical electricity consumption data and to develop models for each stage of the scheme.

[0128] The second single-household power supply guarantee model was constructed. With the ultimate goal of achieving precise control and reducing the burden on grassroots levels, the analysis revealed that the failure to effectively activate load resources to reduce load was the main pain point in grassroots power supply guarantee. The first step was a shift in thinking: The first step is to transform the instantaneous power control mode into the interval power control mode. The second step is to convert the user's real-time baseline under the PJM algorithm into the average baseline under time-of-use pricing. The third step is to convert the user's reported load reduction in response into the user's actual average maintenance load.

[0129] In response, this study uses modeling and analysis to examine the pressure reduction effectiveness of individual response users in actual supply guarantee operations, such as... Figure 2 As shown, the voltage drop effect of any user's power consumption can be calculated using a formula, and the area of ​​the shaded region represents its response effect.

[0130] Figure 2 Modeling and analysis for a single user: - (Equation 1) Therefore, it is only necessary to digitally model all users ensuring power supply, classify and manage different types of users, and quickly aggregate similar load reduction resources through tag-based management to achieve precise control. The total power reduction of n users within the plan must meet the upper and lower limits of the plan's indicators.

[0131] (Equation 2) The third aspect involves the construction of a precise full-cycle management and control model. This is to overcome the pain points of power supply security and achieve precise management and control of power load during peak summer periods, such as... Figure 3 As shown, this implementation method establishes a precise control model for the entire power supply cycle, which can realize functions such as tapping user resources, real-time prediction of plan completion rate, flexible resource optimization operation, and rapid intelligent compilation.

[0132] 1. User tagging classification model based on neural network algorithm.

[0133] To accurately identify user load characteristics and fully explore the depth and flexibility of available user load resources, a streamlined, practical, and flexibly configurable user tag library is established. Users are categorized based on their voltage drop performance in the model, enabling refined management of load resources through "layering, partitioning, time-sharing, and classification."

[0134] Stable voltage drop users: These include users that are shut down or scheduled to start / stop. There are many such users. By combining the average baseline for different time periods (peak, high, average, low) and the expected average load for the corresponding time periods, their voltage drop can be calculated. The voltage drop effect is treated as a constant and set as follows: S 稳 .

[0135] Ordinary users: Such as those significantly affected by photovoltaic fluctuations or whose production is unstable. This number is relatively small. By combining the average baseline for different time periods (peak, high, average, and low) and the expected average load for the corresponding time periods, the voltage drop range can be calculated. This is done when the voltage drop effect is constant. S 普通 (min, max) .

[0136] Flexible dispatchable users: Load resources that can be stably dispatched during load management periods and can quickly provide or absorb power to cope with changes in power supply and demand. Examples include self-owned power plants, energy storage, and fast-start / stop production lines; these users are currently few in number. In actual dispatching, their voltage drop varies due to different start-up times. S 灵活1、2、3…n， The mathematical model for load management based on tagging is as follows: (Equation 3) The pressure reduction effect of n flexibly adjustable users is a variable, and different pressure reduction effects exist under different combinations of flexible resource allocation.

[0137] 2. Load gap prediction model based on rolling time domain.

[0138] Using system-collected data and prediction models at the current time (observation point), load adjustment optimization strategies are obtained for the current and future supply guarantee periods. This provides suggestions on whether intervention and control are needed, the average load adjustment demand after the observation point, and the range of load and power to be adjusted. In practical applications, it can be applied to regional-level (province, city, district / county) forecasting and individual-level (power supply station, customer manager) forecasting, achieving different levels of effectiveness.

[0139] (Equation 4) Taking a precise control requirement of 100%-110% as an example: (Equation 5) (Equation 6) like Figure 4 As shown, the predictive model of the real-time decision-making mechanism: (Equation 7) in, This is the error term, indicating that user variables in the model were not executed as planned. (Solution completion rate) It is affected by two main factors: one is the independent variable related to load voltage drop capacity. The other is an independent variable related to the intensity of regulation. By collecting historical data, a linear regression model is used to estimate... , and The value of .

[0140] 3. A flexible resource coordination and real-time scheduling model based on rolling optimization.

[0141] To achieve precise control of large-scale flexible load resources, the control strategy is optimized based on real-time load demand and the responsiveness of flexible resources, so as to realize the stable operation of the power system and the efficient utilization of load resources, and help the power supply personnel make rapid decisions.

[0142] like Figure 5 As shown, in the actual implementation of load management, the data is collected and refreshed at one point every 15 minutes, and 1 hour can be converted into 4 points for consideration.

[0143] During the scheme development phase: The adjustable power capacity of flexible load resources needs to cover the voltage drop fluctuation range of unstable users. The voltage drop of n flexibly controlled users is related to the response capability and the time of their deployment of the flexible control resources. The response capability is constant, and its voltage drop effect is as follows: (Equation 7) Where A, B, C...N represent the responsiveness of the corresponding flexible resource adjustment, which are constants and are attributes of the flexible resource adjustment. , ... To accommodate the startup time of flexible load resources, during the scheme development phase:

[0144] During the wartime execution phase: Prioritize adjusting the operations of users who are not performing well, and develop a wartime flexible load resource activation plan based on real-time forecast gaps.

[0145] (Equation 8) Establish a mathematical analysis model. To predict the real-time results of the application execution plan completion rate, through transformation analysis, the complex load management problem is ultimately reduced to an n-variable inequality problem of finding the integer solution set, where the solution set is the start time point corresponding to each of the n mobile users.

[0146] The core objective of the flexible resource coordination real-time scheduling model is to minimize potential deviations during load regulation, ensuring that the total voltage drop accurately reaches the predetermined target. Under the premise of meeting power supply guarantee requirements and user responsiveness, it optimizes the start-up and shutdown times and responsiveness of flexible load resources. The objective function can be expressed as: (Equation 9) in, This represents the target value for the total required voltage reduction. For the i-th flexible load resource at startup time The voltage drop of the power.

[0147] (1) Resource response capability constraints The responsiveness of each flexible load resource is limited by its own characteristics: (Equation 10) in, Let be the response power of the i-th resource, a constant. This is the end time of the plan's execution.

[0148] (2) Startup time constraints The resource startup time must be within a feasible time window: (Equation 11) By leveraging the global search capability of genetic algorithms, the encoding method is selected as either real number encoding or integer encoding, and the fitness function is designed as the reciprocal of the objective function, so as to transform the minimization problem into a fitness maximization problem.

[0149] (iv) Analytical approach and innovative points 1. Innovate load management and control thinking: shift the focus of control from the past focus on the execution rate of all small, scattered and wide users to the precise application of large and flexible load resources, reduce the pressure on grassroots control, and transform the instantaneous power control mode into the interval power control mode to reduce the difficulty of control.

[0150] 2. Innovative User Tagging Management Model: Focusing on usability, this model streamlines invalid and inefficient tags, promoting their practical application. It innovatively categorizes users based on their actual load reduction performance, achieving refined management of load resources through "layering, partitioning, time-based allocation, and classification." Furthermore, based on users' historical supply guarantee response data and behavioral characteristics, a deep learning-based user tagging model is established to automatically categorize user loads. This transforms the traditional, repetitive, and lengthy load resource survey into digital tagging, laying a solid foundation for the construction of a demand-side response resource pool.

[0151] 3. Innovate an intelligent and precise model based on rolling optimization strategy: First, construct a day-ahead load scheduling optimization model that considers multi-resource coordination, combine the effectiveness of user tagging management, classify users and implement control measures, and calculate the start and stop times of flexible load resources within the plan through the model.

[0152] Second, construct a load forecasting model based on the rolling time domain: combine control constraints and objective functions, apply historical / instantaneous system data collection and forecasting models to obtain load adjustment and optimization strategies during the current and future supply guarantee periods, and realize real-time forecasting of supply guarantee effectiveness.

[0153] Third, create a flexible resource coordination and real-time scheduling model: By analyzing and mining a large amount of historical data and combining it with the user priority response scheduling mechanism, we can carry out flexible real-time scheduling and resource coordination of load resources. We innovatively introduce a feedback mechanism and, based on the execution results, perform rolling optimization through model linkage to adjust the scheduling strategy in real time, so as to achieve rapid correction and precise control of wartime supply.

[0154] 4. Optimize the execution interface of the traditional load management system: Focus on the pain points of grassroots business management and control, embed the precise management and control model of this implementation method into the management and control platform, and realize digital guidance and burden reduction for grassroots work.

[0155] In a specific implementation case, a power supply company in a prefecture-level city in a developed region of southern China established a flexibly configurable user tag library and promoted the use of a precise control model in the province's load management system. This enabled refined management of load resources by "layering, zoning, time, and classification," improving the speed and accuracy of load management plan preparation and assisting in real-time analysis and decision-making during power supply guarantee periods.

[0156] With precise control as the goal, the plan improves its accuracy by advancing and coordinating efforts across three time dimensions: "peacetime, pre-war, and wartime." Taking load management measures during a certain period in summer as an example, the demand response plan is implemented from 18:00 to 22:00. A district in a prefecture-level city applied the precise control model, and the final average deviation rate of the plan was less than 1%, achieving precise control.

[0157] Currently, load management remains largely based on an "administrative-led, market-assisted" business model. However, with technological advancements, demand-side flexibility resources will play an increasingly important role in ensuring power supply and stabilizing prices. It is foreseeable that, with the new market mechanism and the increasing allocation of supply guarantee subsidies to flexible load resources, the advantages of this implementation model in power supply assurance will be further demonstrated. In the future, the company will continue to focus its efforts in the following areas to facilitate precise control of power supply assurance.

[0158] First, we will further explore flexible resources such as virtual power plants, energy storage, and self-owned power plants, and cultivate potential market users through new market mechanisms and preferential subsidies, so as to increase their effectiveness in supply guarantee schemes and "efficiently adapt" them to the precise control model.

[0159] Second, we will deepen the application of "load labeling management" to realize the digital application of load resources in "classification, zoning, and stratification".

[0160] Third, optimize the prediction model to achieve "minute-level effectiveness prediction".

[0161] Fourth, optimize the plan preparation model to achieve "one-click preparation" of supply guarantee plans and "one-click correction" during wartime execution.

[0162] The project's findings can provide new management and control strategies and system support for subsequent peak summer (winter) load management, and improve the effectiveness of load management at the provincial, municipal, and county levels.

[0163] Embodiments of the present invention also provide a non-transitory machine-readable medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of the present invention.

[0164] Embodiments of the present invention also provide a computer program product, including a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform the method of an embodiment of the present invention.

[0165] Embodiments of the present invention also provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the method of the embodiments of the present invention.

[0166] refer to Figure 6 The present invention will now be described in the form of a structural block diagram of an electronic device that can serve as an embodiment of the present invention, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0167] like Figure 6 As shown, the electronic device includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. The RAM 603 may also store various programs and data required for the operation of the electronic device. The computing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0168] Multiple components in the electronic device are connected to I / O interface 605, including: input unit 606, output unit 607, storage unit 608, and communication unit 609. Input unit 606 can be any type of device capable of inputting information into the electronic device. Input unit 606 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of the electronic device. Output unit 607 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 608 may include, but is not limited to, disks and optical discs. Communication unit 609 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, and / or wireless communication transceivers, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0169] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, CPUs, graphics processing units (GPUs), various special-purpose artificial intelligence (AI) computing units, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above. For example, in some embodiments, the method embodiments of the present invention can be implemented as a computer program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on an electronic device via ROM 602 and / or communication unit 609. In some embodiments, the computing unit 601 can be configured to perform the methods described above by any other suitable means (e.g., by means of firmware).

[0170] Computer programs for implementing the methods of embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0171] In the context of embodiments of the present invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, or infrared systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0172] It should be noted that the term "comprising" and its variations used in the embodiments of the present invention are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The modifications of "one" and "a plurality" mentioned in the embodiments of the present invention are illustrative and not restrictive, and those skilled in the art should understand that unless explicitly indicated otherwise in the context, they should be understood as "one or more".

[0173] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this invention are subject to strict compliance with relevant laws, regulations, and regulatory requirements in their collection, storage, use, processing, transmission, provision, and disclosure, and adhere to the principles of legality, legitimacy, necessity, and good faith. The acquisition of relevant information and data is premised on the user's explicit consent or other legitimate reasons, and a clear and convenient authorization management approach is provided to the user, allowing the user to independently choose to consent, withdraw consent, or refuse to provide relevant information. For functions that rely on user information, if the user does not authorize or withdraws authorization, the corresponding technical function cannot be implemented, and the technical solution of this invention is not applicable in this scenario.

[0174] The steps described in the method embodiments provided by the present invention can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of protection of the present invention is not limited in this respect.

[0175] The term "embodiment" in this specification refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of the invention. The appearance of this phrase in various places in the specification does not necessarily imply the same embodiment, nor does it imply independence or alternativeity from other embodiments. The various embodiments in this specification are described in a related manner, with reference to each other for similar or identical parts. In particular, for apparatus, device, and system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, and relevant details are referred to in the description of the method embodiments.

[0176] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A multi-source data-driven method for generating power supply load resource management and control decisions, characterized in that, include: By connecting with the power business system to collect multi-dimensional electricity consumption data from users, a user load benchmark curve adapted to time-of-use pricing is generated. The load resources are classified according to the user's historical voltage drop performance through a neural network model, and the adjustable response parameters of various types of loads are calculated. The voltage drop of a single user is quantified by time-series load integration, and the voltage drop values ​​of fixed voltage drop loads and basic adjustable loads are aggregated to obtain the minimum voltage drop total. During the pre-set preparation period before the arrival of the power supply guarantee and control cycle, with the total voltage drop of all users falling within the upper and lower limits of the supply guarantee index as a constraint, the genetic algorithm is called to solve the multivariate constraint inequalities, and the optimal initial start-up time combination of each flexible control user is iteratively calculated to generate multiple load control strategies. After entering the power supply guarantee and control cycle, a set of load control strategies is determined as the initial execution benchmark through a selection and screening operation. On-site measured load data is obtained at fixed time intervals. The rolling time domain prediction model is called to integrate the measured load that has occurred and the predicted load for the remaining control period. The average completion rate of the initial execution benchmark is calculated, and the adjustment power range for the remaining control period is derived. When the calculated average completion rate exceeds the preset qualified range, the flexible resource real-time scheduling model is activated. Taking the minimization of the deviation between the total voltage drop and the target index as the objective function, the optimal combination of start and stop times for each flexible control user is re-solved, and the list of mobile control increments / reductions is output as a corrective control decision. The system receives the latest load data collected from users in real time after the implementation of corrective control decisions, updates the load gap and plan completion rate prediction parameters, and sends the updated parameters into the rolling time domain prediction model to trigger a new round of control prediction and regulation decision solving iteration.

2. The method according to claim 1, characterized in that, The power business system includes a marketing system, an electricity consumption data collection system, and a load management system; the user's multi-dimensional electricity consumption data includes the user's static business expansion attributes and 1-minute / 15-minute high-frequency time-series historical load data; Before collecting multi-dimensional electricity consumption data from users through the power business system and generating a user load benchmark curve adapted to time-of-use pricing, the method further includes: Extract the user's high-frequency time-series historical load of the past five days from the user's multi-dimensional electricity consumption data, and generate the original baseline based on the user's high-frequency time-series historical load of the past five days using the PJM algorithm; The daily time sequence is segmented according to the preset time-of-use pricing rules. The average value of all time-of-use load points under the same price period within the original baseline is recalculated. The time sequence curve is reconstructed segment by segment and converted to obtain the user load benchmark curve adapted to the time-of-use pricing.

3. The method according to claim 2, characterized in that, The load resources are classified based on the historical voltage drop performance of users using a neural network model, and the adjustable response parameters of each type of load are calculated, including: Extract historical voltage drop features from multi-dimensional electricity consumption data of users, input them into a pre-trained neural network model, and output the corresponding user load resource type; wherein, the user load resource type is divided into stable voltage drop users, ordinary fluctuation users, and flexible control users; Based on the classification results, corresponding adjustable response parameters are matched. The adjustable response parameters include: for users with stable voltage drop, the adjustable response parameter is a fixed voltage drop constant; for users with ordinary fluctuations, the minimum value of the voltage drop range is taken as the basic voltage drop parameter; and for users with flexible control, there are self-owned power plants, energy storage, and production lines that can be quickly started and stopped, and the voltage drop power changes dynamically with the start-up time.

4. The method according to claim 3, characterized in that, The voltage drop per user is quantified by time-series load integration, and the voltage drop values ​​of fixed voltage drop loads and basic adjustable loads are aggregated to obtain the minimum voltage drop total, including: The time-series load curve is divided into three phases: non-response phase, flexible resource activation and control phase, and load recovery phase. The baseline electricity consumption for each stage is calculated by integrating the area of ​​the time-series load curve. By combining the baseline electricity consumption at each stage, the voltage drop of a single user is obtained through differential calculation.

5. The method according to claim 1, characterized in that, With the constraint that the total voltage drop of all users falls within the upper and lower limits of the supply guarantee target, a genetic algorithm is used to solve the multivariate constraint inequalities, iteratively calculating the optimal initial start-up time combination for each flexibly controlled user, and generating multiple load control strategies, including: During the pre-set preparation period before the arrival of the power supply guarantee and control cycle, the multivariate constraint inequality is the pre-war multivariate constraint inequality. The pre-war multivariate constraint inequalities are combined with the upper and lower limits of the ordinary fluctuation user pressure drop range to complete the constraint configuration. Under the constraint that the total voltage drop of all users falls within the upper and lower limits of the supply guarantee index, the start-up time of flexible control users is encoded and iterated through a genetic algorithm to obtain the optimal combination of initial start-up times for each flexible control user, thereby generating multiple load control strategies.

6. The method according to claim 1, characterized in that, After entering the power supply guarantee and control cycle, a set of load control strategies is determined as the initial execution benchmark through a selection and screening operation, including: The load management strategy with a total voltage drop close to the median value of the supply guarantee index is selected as the first-level candidate strategy; if there is only one first-level candidate strategy, it is directly used as the initial execution benchmark. When there are multiple primary candidate strategies, the primary candidate strategy with the smaller scale of flexible resource investment is selected as the secondary candidate strategy; when there is only one secondary candidate strategy, it is directly used as the initial execution benchmark. If there are multiple secondary candidate strategies, the secondary strategy with a more balanced distribution of flexible resource start and stop times is selected as the tertiary candidate strategy; if there is only one tertiary candidate strategy, it is directly used as the initial execution benchmark. When there are multiple candidate strategies at the three levels, one is randomly selected as the initial execution benchmark.

7. The method according to claim 6, characterized in that, The fixed time interval is 10-30 minutes; on-site measured load data is acquired according to the fixed time interval, and the rolling time-domain prediction model is called to fuse the measured load that has occurred and the predicted load for the remaining control period, calculate the average completion rate of the initial execution benchmark, and derive the adjustment power range for the remaining control period, including: After the control cycle starts, the first observation node is reached to obtain all time-series measured load data from the start of control to the current observation node, and the average completion rate of the initial execution benchmark is calculated. By using the linear regression error term built into the rolling time-domain prediction model, the regular load fluctuation interference caused by planned start-up and shutdown loads is identified and eliminated, and the average completion rate of the initial execution benchmark obtained from the initial calculation is corrected. The required adjustment power range for the remaining control period is obtained by solving the correction results based on the completion rate.

8. The method according to claim 1, characterized in that, The preset qualified range is a plan completion rate of 100% to 110%; a flexible resource real-time scheduling model is activated, with the objective function being the minimization of the deviation between the total voltage drop and the target indicator. The optimal combination of start-up and stop times for each flexible control user is re-solved, and a list of incremental / reduced flexible control measures is output as a corrective control decision, including: When the average completion rate exceeds the preset qualified range, the flexible resource real-time scheduling model is activated. In the process of solving the optimal combination of start and stop times for each flexible control user, the maximum adjustable power of the resource and the allowed control time window are configured as dual constraints. Based on the real-time measured range of adjustable power, a wartime multi-constraint inequality is constructed. The flexible resource real-time scheduling model, under the combined constraints of the dual constraints and the wartime multi-constraint inequality, solves the problem by minimizing the deviation between the total voltage drop and the target index, and automatically outputs a list of mobile control increments / reductions as a corrective control decision.

9. An electronic device, comprising: A processor and a memory storing a program, characterized in that the program includes instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 8.

10. A non-transitory machine-readable medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 8.

Images