A multi-target cooperative charging scheduling optimization method and system

Abstract

The application provides a multi-target cooperative charging scheduling optimization method and system, and relates to the technical field of charging station scheduling management. The technical scheme provided by the application represents the numerical characteristics and coupling relationship of the triple uncertainty of charging load-power-market by constructing a joint probability distribution model, generates random scene samples based on the model, and screens out typical scenes that can represent different operating conditions under a high proportion of renewable energy access scenarios using a clustering effectiveness index. According to the grid peak shaving error threshold corresponding to each typical scene, a dynamic adjustment function is determined, a double-layer cooperative optimization model including a day-ahead pre-scheduling layer and an intra-day rolling correction layer is constructed, a benchmark scheduling plan is generated based on the typical scenes at the day-ahead level, and the charging power instruction is rolling corrected according to the actual deviation at the intra-day level, so that the balance and stability of scheduling accuracy and economic benefit are realized, thereby improving the scheduling effect of the charging station under a high proportion of renewable energy access scenarios.

Description

Technical Field

[0001] This application relates to the field of charging station scheduling and management technology, specifically to a multi-objective collaborative charging scheduling optimization method and system. Background Technology

[0002] With the rapid growth of electric vehicle ownership, charging stations, as key hubs connecting electric vehicles and the power system, have a significant impact on grid security and stability as well as operators' economic benefits. Charging station scheduling, through the rational arrangement of charging power and charging time periods, aims to achieve goals such as grid load regulation, reduced operating costs, and improved equipment utilization while meeting user charging needs. This is a core technical issue in charging station operation and management.

[0003] The most common scheduling method in related technologies is the fixed-price timed charging strategy. Its principle is to schedule charging during low-price periods based on time-of-use pricing, avoiding high-price periods and thus reducing charging costs. However, when applied to scenarios with a high proportion of renewable energy integration, this approach faces challenges. In such scenarios, distributed photovoltaic and wind power output exhibits significant random fluctuations and intermittent characteristics. Power supply capacity is highly time-varying, and charging load is highly uncertain due to random factors such as user arrival time and charging duration. Market electricity prices fluctuate in real-time due to supply and demand, creating a triple uncertainty across charging load, power supply, and the market. The fixed-price timed charging strategy relies solely on historical price patterns for charging plans, failing to adapt to real-time changes in power output and load demand. When actual photovoltaic output falls below expectations, charging power cannot be adjusted in time, leading to scheduling inaccuracies. Furthermore, when concentrated charging overlaps with peak grid periods, it exacerbates grid pressure, causing revenue imbalances and ultimately resulting in poor charging station scheduling performance. Summary of the Invention

[0004] This application provides a multi-objective collaborative charging scheduling optimization method and system, which can improve the scheduling effect of charging stations in scenarios with a high proportion of renewable energy access.

[0005] Firstly, this application provides a multi-objective cooperative charging scheduling optimization method, the method comprising: Historical charging load data, distributed power output data, and market electricity price data of charging stations are collected to construct a joint probability distribution model that characterizes numerical features and coupling relationships. Multiple random scene samples are generated based on the joint probability distribution model, and multiple typical scenes are selected from the multiple random scene samples using a preset clustering effectiveness index. Extract key scene features from each typical scenario and construct a mapping relationship between each typical scenario and the key scene features; Based on the power grid peak-shaving error thresholds corresponding to each typical scenario, the dynamic adjustment function of operator revenue weight and real-time peak-shaving error is determined, and a two-layer collaborative optimization model including a day-ahead pre-scheduling layer and an intraday rolling correction layer is constructed. The system acquires the target key scenario features and actual operating status information of the charging station, inputs these features and operating status information into a two-layer collaborative optimization model, and outputs real-time charging power commands.

[0006] By adopting the above technical solution, historical charging load data, distributed power output data, and market electricity price data of charging stations are collected. A joint probability distribution model is constructed to characterize the numerical features and coupling relationship of the triple uncertainty of charging load, power supply, and market. Based on this model, random scenario samples are generated, and a clustering effectiveness index is used to screen out typical scenarios that can represent different operating conditions under high-proportion renewable energy access scenarios, effectively characterizing the complex uncertainty of the scenario. By extracting key scenario features of each typical scenario and constructing mapping relationships, accurate identification of actual operating conditions is achieved. Based on the grid peak-shaving error threshold corresponding to each typical scenario, a dynamic adjustment function is determined, and a two-layer collaborative optimization model including a day-ahead pre-scheduling layer and an intraday rolling correction layer is constructed. This model dynamically adjusts the operator revenue weight and peak-shaving constraint weight according to the target key scenario features and actual operating status information. At the day-ahead level, a benchmark scheduling plan is generated based on typical scenarios, and at the intraday level, the charging power command is rolled and corrected according to the actual deviation, achieving a balance and stability between scheduling accuracy and economic benefits, thereby improving the scheduling effect of charging stations under high-proportion renewable energy access scenarios.

[0007] Secondly, this application provides a multi-objective cooperative charging scheduling optimization system, the system comprising: The data acquisition module is used to collect historical charging load data, distributed power output data and market electricity price data of the charging station, and to construct a joint probability distribution model that characterizes numerical features and coupling relationships. The scene generation module is used to generate multiple random scene samples based on the joint probability distribution model, and to select multiple typical scenes from the multiple random scene samples using a preset clustering effectiveness index. The feature extraction module is used to extract key scene features for each typical scenario and to construct a mapping relationship between each typical scenario and the key scene features. The model building module is used to determine the dynamic adjustment function of operator revenue weight and real-time peak-shaving error based on the power grid peak-shaving error threshold corresponding to each typical scenario, and to build a two-layer collaborative optimization model including a day-ahead pre-scheduling layer and an intraday rolling correction layer. The processing module is used to acquire the target key scene features and actual operating status information of the charging station, input the target key scene features and actual operating status information into the two-layer collaborative optimization model, and output real-time charging power commands.

[0008] Thirdly, this application provides a computer storage medium that stores multiple instructions adapted for loading by a processor and executing any of the methods described above.

[0009] Fourthly, this application provides an electronic device including a processor, a memory, and a transceiver. The memory is used to store instructions, the transceiver is used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform any of the methods described above.

[0010] In summary, the beneficial effects of the technical solution of this application include: By adopting the above technical solution, historical charging load data, distributed power output data, and market electricity price data of charging stations are collected. A joint probability distribution model is constructed to characterize the numerical features and coupling relationship of the triple uncertainty of charging load, power supply, and market. Based on this model, random scenario samples are generated, and a clustering effectiveness index is used to screen out typical scenarios that can represent different operating conditions under high-proportion renewable energy access scenarios, effectively characterizing the complex uncertainty of the scenario. By extracting key scenario features of each typical scenario and constructing mapping relationships, accurate identification of actual operating conditions is achieved. Based on the grid peak-shaving error threshold corresponding to each typical scenario, a dynamic adjustment function is determined, and a two-layer collaborative optimization model including a day-ahead pre-scheduling layer and an intraday rolling correction layer is constructed. This model dynamically adjusts the operator revenue weight and peak-shaving constraint weight according to the target key scenario features and actual operating status information. At the day-ahead level, a benchmark scheduling plan is generated based on typical scenarios, and at the intraday level, the charging power command is rolled and corrected according to the actual deviation, achieving a balance and stability between scheduling accuracy and economic benefits, thereby improving the scheduling effect of charging stations under high-proportion renewable energy access scenarios. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating a multi-objective cooperative charging scheduling optimization method according to an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a multi-objective cooperative charging scheduling optimization system according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0012] Explanation of reference numerals in the attached drawings: 300, electronic device; 301, processor; 302, communication bus; 303, user interface; 304, network interface; 305, memory. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0014] In the description of the embodiments of this application, words such as "illustrative," "for example," or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "illustrative," "for example," or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Rather, the use of words such as "illustrative," "for example," or "for example" is intended to present the relevant concepts in a specific manner.

[0015] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0016] Please see Figure 1 This is a flowchart illustrating a multi-objective cooperative charging scheduling optimization method provided in an embodiment of this application. This method can be implemented using a computer program, a microcontroller, or run on a multi-objective cooperative charging scheduling optimization system based on the von Neumann architecture. The computer program can be integrated into the application or run as a standalone utility application. The specific steps of the multi-objective cooperative charging scheduling optimization method are described in detail below.

[0017] S101: Collect historical charging load data, distributed power output data, and market electricity price data of charging stations to construct a joint probability distribution model that characterizes numerical features and coupling relationships; Among them, historical charging load data refers to the time-series data of electric vehicle charging power demand recorded by charging stations over a period of time, including different load characteristics on weekdays and holidays; distributed power output data represents the actual power generation data of renewable energy power generation equipment such as photovoltaic and wind power connected to charging stations or distribution networks; market electricity price data is used to represent real-time electricity price or time-of-use electricity price information at different times in the electricity market; joint probability distribution model refers to a probability model that can simultaneously characterize multiple random variables and their interdependencies, and can capture the coordinated fluctuation characteristics among charging load, distributed power output, and market electricity price; numerical features represent the statistical quantities of various types of data, such as mean, variance, peak value, volatility, etc.; coupling relationship refers to the correlation or causal relationship between different variables, reflecting the mutual influence mechanism between multiple uncertain factors.

[0018] Specifically, this step is performed during system initialization or model training to lay the data foundation for subsequent scenario generation. First, at least one year of historical data is acquired through the charging station's data acquisition system or energy management system, ensuring data coverage of different seasons and operating conditions. After data acquisition, preprocessing is required, including missing value imputation, outlier removal, and time alignment, ensuring complete synchronization of the three types of data across time. Then, statistical analysis is performed on each type of data to extract marginal distribution characteristics. Kernel density estimation is used to fit the probability density function of the charging load, and Weibull distribution is used to fit the wind power output characteristics. Based on this, a joint probability distribution is constructed using Copula theory or a Gaussian mixture model, and model parameters are calibrated using maximum likelihood estimation or the EM algorithm.

[0019] In some embodiments, the joint probability distribution model can be constructed in various ways. Optionally, a Copula-based modeling method is adopted. First, marginal distributions are fitted to charging load, distributed power output, and market electricity price respectively. The Kolmogorov-Smirnov test is used to select the most suitable distribution type. Then, the rank correlation coefficients between the variables are calculated. Based on the correlation structure, an appropriate family of Copula functions is selected. The Copula parameters are calibrated using the maximum likelihood estimation method. Finally, the marginal distributions are combined with the Copula functions to form a complete joint probability distribution model. This method can flexibly handle nonlinear correlations and has no strict requirements on the form of the marginal distributions. It is understood that other methods can also be used to achieve joint probability distribution modeling, which are not limited here.

[0020] S102: Generate multiple random scene samples based on the joint probability distribution model, and use the preset clustering effectiveness index to select multiple typical scenes from the multiple random scene samples; Here, random scenario samples represent potential future operating scenarios generated through random sampling based on a joint probability distribution model. Each scenario includes time-series curves of complete charging load, distributed power output, and market electricity price. Clustering effectiveness indicators are quantitative metrics used to evaluate the quality of clustering results, measuring the density within sample clusters and the separation between clusters. Typical scenarios refer to scenario samples that represent a certain type of similar operating conditions. It should be understood that the typical scenarios in this application specifically represent different situations under scenarios with a high proportion of renewable energy access.

[0021] Specifically, this step is performed during the scenario library construction phase. Its purpose is to extract a limited number of representative scenarios from a massive amount of uncertain scenarios to balance computational efficiency and accuracy. First, based on the established joint probability distribution model, thousands to tens of thousands of random scenario samples are generated using random sampling techniques. Each sample contains three-dimensional time-series data for a future scheduling cycle. Then, all scenario samples are normalized to eliminate the influence of different physical quantities and statistical feature parameters are extracted to form a clustering feature vector. These features include load peak value, valley value, peak-valley difference, volatility, total power output, power output fluctuation coefficient, weighted average electricity price, and the proportion of high-price periods, forming a high-dimensional feature space. Next, a clustering algorithm is used to group the scenario samples, classifying scenarios with similar operating characteristics into the same cluster. During the clustering process, the optimal number of clusters needs to be determined. By calculating the clustering effectiveness index under different cluster numbers, the number of clusters that optimizes the index is selected. Finally, the scene sample closest to the cluster center is selected from each cluster as the typical scene of that cluster. This scene can represent the operating characteristics of all scenes in the entire cluster with the smallest feature deviation, thus covering the main probability distribution area of ​​the original random scene space with a limited set of typical scenes.

[0022] In some embodiments, typical scenarios can be selected in various ways. Optionally, an improved iterative clustering combined with silhouette coefficient optimization is used. First, a candidate range for the number of clusters is defined. For each candidate cluster, a clustering algorithm is executed, and the mean silhouette coefficient of all samples is calculated. The silhouette coefficient is obtained by calculating the difference between the average distance of a sample to other samples in its own cluster and the average distance to the nearest neighbor cluster, and then dividing by the maximum of the two. The cluster with the largest mean silhouette coefficient is selected as the optimal number of clusters. Then, within each cluster, the distance of all samples to the cluster center is calculated, and the sample with the smallest distance is selected as the typical scenario. At the same time, the probability of this typical scenario appearing in the original random scenario set is recorded as the scenario weight for subsequent weighted optimization calculations. It is understood that other clustering methods can also be used to select typical scenarios, and this is not limited here.

[0023] S103: Extract key scene features for each typical scenario and construct a mapping relationship between each typical scenario and key scene features; Among them, key scenario features refer to the core feature parameters that can significantly distinguish different typical scenarios and affect scheduling decisions, including peak load time, peak power, expected power output, electricity price distribution during the time period, load-power matching degree, etc. These features have lower dimensionality than complete time series curves but retain the main decision-related information. The mapping relationship represents the corresponding function or lookup mechanism from key scenario features to corresponding typical scenarios. By establishing a classification model or constructing a feature-scenario similarity matrix, a fast matching from feature vectors to typical scenario numbers can be achieved.

[0024] Specifically, this step is performed after the typical scenario library is built. Its purpose is to establish a rapid matching mechanism from real-time observed features to historical typical scenarios, supporting the rapid selection of appropriate scheduling strategies during the day-ahead scheduling phase. First, feature engineering is performed on the complete time-series data of each typical scenario to extract key features that represent the essential characteristics of that scenario. Feature extraction can employ dimensionality reduction methods to retain principal components that explain most of the variance, or use domain knowledge to select features with clear physical meaning. Then, a mapping relationship between feature vectors and typical scenarios is constructed. This can be done using supervised learning methods, with typical scenario numbers as labels and key features as input, to train a classification model. After training, the model can predict the most matching typical scenario category based on the input key features. Alternatively, a non-parametric similarity matching method can be used. Key feature vectors for all typical scenarios are pre-calculated and stored. When new observed features are obtained, the similarity index between this feature vector and the feature vectors of all typical scenarios is calculated, and the typical scenario with the highest similarity is selected as the matching result. The establishment of this mapping relationship allows for the rapid location of the historically most similar typical scenario to the current forecast situation during the day-ahead scheduling phase, simply by extracting key features from short-term forecast data.

[0025] S104: Based on the power grid peak-shaving error thresholds corresponding to each typical scenario, determine the dynamic adjustment function of operator revenue weight and real-time peak-shaving error, and construct a two-layer collaborative optimization model including a day-ahead pre-scheduling layer and an intraday rolling correction layer. Among them, the grid peak-shaving error threshold represents the maximum allowable deviation between the actual peak-shaving effect of charging stations and the grid peak-shaving demand under different typical scenarios, reflecting the grid's tolerance range for peak-shaving accuracy. This threshold is dynamically set according to grid security constraints and dispatch protocols. The operator revenue weight is used to represent the relative importance of the operator's economic revenue objective in the multi-objective optimization function. The larger the weight value, the more the optimization process is biased towards maximizing economic benefits. The real-time peak-shaving error refers to the deviation between the peak-shaving power provided by the charging station during actual operation and the peak-shaving command issued by the grid, reflecting the accuracy of dispatch execution. The dynamic adjustment function represents the dynamic adjustment of the operator's peak-shaving error based on the real-time peak-shaving error. The mathematical mapping relationship of revenue weights allows for prioritizing economic gains when peak-shaving errors are small, while forcibly increasing the priority of peak-shaving performance when peak-shaving errors are large. The day-ahead pre-scheduling layer refers to a long-term optimized scheduling layer based on day-ahead forecast data and typical scenarios, which formulates a baseline scheduling plan for the next day or more. The intraday rolling correction layer refers to a scheduling layer that performs short-term rolling corrections to the day-ahead scheduling plan based on real-time data, responding to actual operational deviations and adjusting charging power commands. The dual-layer collaborative optimization model represents a hierarchical optimization architecture that combines day-ahead pre-scheduling and intraday correction. The upper layer is responsible for long-term planning, and the lower layer is responsible for real-time tracking. The two layers achieve collaborative decision-making through information transmission.

[0026] Specifically, this step is executed during the scheduling model construction phase, aiming to establish a two-layer optimization framework that can adaptively balance economic efficiency and peak-shaving performance. First, grid peak-shaving error data for each typical scenario during historical operation is acquired. Statistical analysis methods are used to calculate the peak-shaving error distribution characteristics for each typical scenario, determining error thresholds at different confidence levels. For high-load scenarios, stricter thresholds are typically set to ensure grid safety, while thresholds can be appropriately relaxed for low-load scenarios to improve economic efficiency. Then, a nonlinear mapping relationship is constructed between operator revenue weights and real-time peak-shaving errors. When the real-time peak-shaving error is less than the set threshold, the operator revenue weight is dynamically increased, shifting the optimization objective towards economic goals such as charge-discharge arbitrage and demand cost management. When the real-time peak-shaving error approaches or exceeds the threshold, the operator revenue weight is rapidly reduced, and the peak-shaving constraint weight is correspondingly increased, shifting the optimization objective towards strictly tracking grid peak-shaving commands. This mapping relationship can be expressed using piecewise functions, continuous smooth functions, or probability functions, forming a dynamic adjustment function. Based on this, a two-layer collaborative optimization model is constructed. The day-ahead pre-scheduling layer takes typical scenario data as input and takes maximizing operator revenue and minimizing grid load fluctuation as multiple objective functions. It considers battery state of charge constraints, charging and discharging power constraints, and power balance constraints to obtain the benchmark charging power sequence for future scheduling cycles. The intraday rolling correction layer takes the actual operating status and day-ahead benchmark instructions as input. It identifies the differences between actual and typical scenarios through feature deviation calculation, uses a dynamic adjustment function to adaptively adjust the target weights, and performs local optimization of the benchmark instructions within a limited fluctuation range to generate real-time charging power instructions, thereby achieving the coordinated cooperation between day-ahead coarse adjustment and intraday fine adjustment.

[0027] S105: Obtain the target key scene features and actual operating status information of the charging station, input the target key scene features and actual operating status information into the two-layer collaborative optimization model, and output real-time charging power command.

[0028] Among them, the target key scenario features represent the scenario feature vectors corresponding to the future time periods that need to be optimized for scheduling. They are extracted based on short-term prediction data and used to match typical historical scenarios. The actual operating status information refers to the real operating parameters of the charging station at the current moment, including real-time monitoring data such as battery state of charge, current charging power, number of connected vehicles, real-time output of distributed power sources, and real-time electricity price of the grid. The real-time charging power command represents the charging power control command that the charging pile should execute at the current moment, which is calculated and output by the two-layer collaborative optimization model. This command takes into account the balance between economy and peak-shaving performance and needs to be sent to the charging pile controller for execution.

[0029] Specifically, this step is executed periodically during the actual scheduling operation phase, aiming to apply the constructed two-layer collaborative optimization model to real-time scheduling decisions. First, before the start of the preset period in the day-ahead pre-scheduling layer, short-term forecast data for that period is acquired, including charging load forecasts, distributed power output forecasts, and market electricity price forecasts. Based on the forecast data, key target scenario features are extracted. This feature vector contains key parameters such as load peak-valley characteristics, expected power output, and electricity price distribution during the forecast period. The extracted key target scenario features are input into the constructed mapping model. Through feature similarity matching or classifier prediction, the target typical scenario most similar to the current predicted operating conditions is selected from the historical typical scenario library. The optimal charging power sequence pre-optimized offline for this typical scenario is retrieved as the baseline scheduling instruction for the day-ahead pre-scheduling layer. In the intraday rolling correction layer, the actual operating status information of the charging stations is acquired at fixed time intervals. The actual status features are extracted and compared with the typical scenario features corresponding to the baseline scheduling instructions. A feature deviation vector is calculated, which quantifies the degree of difference between the actual operating conditions and the predicted scenario. The real-time peak shaving error is calculated based on the characteristic deviation vector. This error value is then input into the dynamic adjustment function to obtain the operator revenue weight and peak shaving constraint weight at the current moment. Within the preset elastic floating threshold range, the updated weight values ​​are used to perform local optimization on the multi-objective optimization function to obtain the corrected real-time charging power command.

[0030] Based on the above embodiments, as an optional implementation method, the method of determining the adjustable charging amount of the first electric vehicle during the scheduled charging period in S105 can be specifically implemented through the following steps S201-S204.

[0031] S201: Obtain the target key scenario features of the charging station within the preset period of the day-ahead pre-scheduling layer, as well as the actual operating status information at the current moment, and input the target key scenario features and the actual operating status information into the two-layer collaborative optimization model; The preset period refers to the time span for scheduling optimization by the pre-scheduling layer, which is usually the next 24 hours or longer. The length of this period is determined according to the charging station's operation mode and the grid's scheduling needs.

[0032] This step is executed at the beginning of each scheduling cycle during the actual scheduling operation phase. The system first obtains the charging load forecast curve, distributed power output forecast curve, and market electricity price forecast curve for the preset period from the short-term forecast module. Features are extracted from the forecast curves, and key parameters such as peak load time, peak-valley power difference, average load level, average power output, output variance, weighted average electricity price, and peak-valley price difference are calculated to form a target key scenario feature vector. Simultaneously, real-time operating status information such as the current battery state of charge, actual output power of charging piles, number of vehicles connected for charging, instantaneous output power of distributed power sources, and real-time grid electricity price are obtained from the charging station monitoring system. The target key scenario feature vector and the actual operating status information are packaged and integrated, then input into a two-layer collaborative optimization model. The target key scenario features are used for scenario matching and baseline instruction generation in the day-ahead pre-scheduling layer, while the actual operating status information is used for deviation calculation and instruction correction in the intraday rolling correction layer. These two types of data jointly drive the collaborative operation of the two-layer optimization model.

[0033] S202: In the day-ahead pre-scheduling layer, the target typical scenario corresponding to the key scenario features of the target is determined based on the mapping relationship, and the optimal solution of the target typical scenario is used as the benchmark scheduling instruction of the day-ahead pre-scheduling layer. Among them, the target typical scenario represents the historical typical scenario that is most similar to the current predicted operating conditions. The operating characteristics of this scenario have the highest matching degree with the characteristics of the target key scenario. The optimal solution refers to the optimal charging power time series obtained by offline optimization calculation for the target typical scenario. This series achieves the optimal balance between operator revenue and grid peak-shaving performance under various constraints.

[0034] This step, executed at the day-ahead pre-scheduling layer, aims to quickly determine the baseline scheduling strategy. The system inputs the target key scene feature vector obtained in step S201 into the constructed mapping model. This model calculates the similarity or distance between the target feature vector and the feature vectors of each typical scene in the scene library, selecting the typical scene with the highest similarity or smallest distance as the target typical scene. After determining the target typical scene, the optimal charging power sequence corresponding to that scene is retrieved from the optimization result storage area of ​​the scene library. This sequence, calculated offline during the scene library construction phase, includes the charging power setpoint for each time step within a preset period. This optimal charging power sequence is directly used as the baseline scheduling instruction for the day-ahead pre-scheduling layer. This instruction reflects the optimal scheduling strategy under typical operating conditions, providing a baseline reference trajectory for subsequent intraday rolling corrections. This avoids online optimization calculations in the day-ahead phase, significantly reducing computation time and resource consumption.

[0035] Based on the above embodiments, as an optional implementation method, the method of determining the target typical scene corresponding to the target key scene features based on the mapping relationship in S202 can be implemented through the following steps S2021-S2022.

[0036] S2021: Calculate the similarity index between the key scene features of the target and the key scene features of each pre-stored typical scene; The similarity index is a quantitative value used to measure the degree of similarity between the feature vectors of two scenes. The larger the value, the closer the operating characteristics of the two scenes are. Commonly used measurement methods include distance measurement and correlation measurement.

[0037] This step is performed during the scenario matching phase of the day-ahead pre-scheduling layer. The system retrieves key scenario feature vectors for all pre-stored typical scenarios from the scenario library storage area. Each vector contains feature components such as load peak-valley characteristics, power output statistics, and electricity price distribution parameters for that typical scenario. The target key scenario feature vector is compared with the key scenario feature vectors of each typical scenario one by one. During the calculation, the two feature vectors are first aligned in dimension and normalized numerically to eliminate the influence of different feature dimensions. Then, a distance metric is used to calculate the difference between the two vectors. The overall distance is obtained by taking the square root of the sum of the squares of the differences of each feature component; the smaller the distance, the higher the similarity. Alternatively, a correlation metric is used to calculate the ratio of the inner product of the two vectors to the product of their respective magnitudes; the closer this ratio is to 1, the higher the similarity. After completing the similarity calculation for each typical scenario in the scenario library, a set of similarity index values ​​is obtained, with each value corresponding to the degree of matching between a typical scenario and the target scenario.

[0038] S2022: Select typical scenarios that meet the preset matching conditions for similarity index as target typical scenarios.

[0039] Among them, the preset matching conditions represent the judgment criteria used to filter typical target scenarios. These criteria are set according to the threshold or ranking rules of the similarity index to ensure that the selected typical scenarios have sufficient similarity to the target scenarios.

[0040] Specifically, the system evaluates all calculated similarity indices and filters them according to preset matching conditions. When using a threshold-based method, all typical scenarios with similarity indices exceeding the preset threshold are selected. If multiple scenarios meet the conditions, the scenario with the highest similarity index is selected as the target typical scenario. If using a distance metric, the scenario with the smallest distance among those with a distance less than the preset threshold is selected. When using a ranking-based method, all typical scenarios are sorted from highest to lowest similarity index, and the top-ranked typical scenario is directly selected as the target typical scenario. After determining the target typical scenario, the scenario number and corresponding similarity index value are recorded for subsequent use in calling the optimal scheduling strategy for that scenario and evaluating the scenario matching quality. If the similarity indices of all typical scenarios do not meet the preset matching conditions, a scenario library update mechanism is triggered or an emergency scheduling strategy is adopted to ensure the continuous operation of the scheduling system.

[0041] Based on the above embodiments, as an optional implementation method, the optimal solution of the target typical scenario in S202 is used as the baseline scheduling instruction of the day-ahead pre-scheduling layer, which can be specifically implemented through the following steps S2023-S2025.

[0042] S2023: Retrieve historical charging load data, distributed power output data, and market electricity price data corresponding to the target typical scenario as day-ahead optimization boundary conditions; Among them, the day-ahead optimization boundary conditions represent the input parameters and constraint range required when performing optimization calculations at the day-ahead pre-scheduling layer. These conditions define the solution space and feasible region of the optimization problem.

[0043] This step is executed after the target typical scenario is identified. Based on the target typical scenario's identifier, the system retrieves complete time-series data corresponding to that scenario from the historical database, including historical charging load data, distributed power generation output data, and market electricity price data. This historical data records the actual operating conditions of the typical scenario in the past, including load power values, power generation output values, and electricity price values ​​for each time step within a preset period. The retrieved three types of historical data are used as boundary conditions input for day-ahead optimization. Historical charging load data defines the range of changes in charging demand, distributed power generation output data defines the fluctuation characteristics of renewable energy supply, and market electricity price data defines the time-varying pattern of economic costs.

[0044] S2024: Construct a day-ahead multi-objective optimization function with the objectives of maximizing operator revenue and minimizing grid load volatility, and set battery state of charge constraints, charge and discharge power constraints and power balance constraints; Among them, the operator revenue maximization objective means maximizing the economic benefits obtained by the charging station operator through optimizing the charging and discharging strategy, with revenue sources including charging service fees, electricity price arbitrage income, and ancillary service compensation; the grid load volatility minimization objective means maximizing the smoothness of the total load curve observed on the grid side and minimizing the fluctuation amplitude by adjusting the charging power; the day-ahead multi-objective optimization function refers to a mathematical expression that simultaneously contains multiple optimization objectives, and the trade-off relationship between multiple objectives is handled through weighted combination or Pareto optimization methods; the battery state of charge constraint is used to limit the charging state of energy storage batteries within the safe operating range to prevent overcharging or over-discharging; the charging and discharging power constraint is used to limit the charging and discharging power of charging piles or energy storage systems within the rated capacity range of the equipment; and the power balance constraint is used to ensure that the energy conservation relationship is satisfied between the charging load, distributed power output, and grid interaction power within the charging station at any given time.

[0045] This step is executed after obtaining the day-ahead optimization boundary conditions. The system first constructs an operator revenue objective function, which calculates the charging service revenue minus the electricity purchase cost plus ancillary service revenue for each time step within a preset period, summing the total revenue over all time steps. Then, it constructs a grid load volatility objective function, calculating the sum of squares of the differences in total grid load between adjacent time steps; a smaller value indicates a smoother load curve. The two objective functions are combined into a comprehensive objective function using weighting coefficients, which reflect the operator's preference for economic efficiency and peak-shaving performance. When constructing constraints, upper and lower limits are set for battery state of charge (SOC) at each time step to ensure it remains within acceptable limits; upper and lower limits are set for charging and discharging power at each time step to ensure power does not exceed equipment capacity; and a power balance equation constraint is set, requiring the sum of charging load, distributed power output, energy storage charging and discharging power, and grid interaction power at each time step to be zero. After completing the construction of the objective functions and constraints, a complete day-ahead multi-objective optimization function expression is formed.

[0046] S2025: Solve the day-ahead multi-objective optimization function to obtain the optimal charging power sequence for each time step within the future preset period, and use the optimal charging power sequence as the benchmark scheduling instruction for the day-ahead pre-scheduling layer.

[0047] The time step indicates that the preset period is divided into multiple discrete time periods, and the length of each time period is called the time step, which is usually set to fifteen minutes or one hour.

[0048] This step is executed after the multi-objective optimization function is constructed. The system uses a numerical optimization algorithm to solve the optimization function. The algorithm is selected based on the characteristics of the objective function: a convex optimization solver is used when the objective function is convex and the constraints are linear; a heuristic optimization algorithm or gradient descent algorithm is used when the objective function is non-convex. During the solution process, the algorithm searches for the optimal solution within the feasible region that satisfies all constraints, and continuously adjusts the value of the charging power variable through iterative calculations to gradually bring the comprehensive objective function value to its optimum. After the solution is completed, the optimal charging power value corresponding to each time step within the preset future period is obtained. These values ​​are arranged in chronological order to form the optimal charging power sequence.

[0049] S203: In the intraday rolling correction layer, extract the actual state features from the actual operating state information, and calculate the feature deviation vector between the actual state features and the typical scenario features corresponding to the baseline scheduling instructions. Among them, actual state features refer to the feature parameters extracted from real-time operating status information that can reflect the current operating conditions, including current load power, real-time power output, real-time electricity price, etc.; typical scenario features refer to the key feature parameters of the target typical scenario corresponding to the baseline dispatch instruction; feature deviation vector is used to represent the difference between actual state features and typical scenario features in each feature dimension, reflecting the degree of deviation between actual operation and predicted scenario.

[0050] This step is executed periodically within the intraday rolling correction layer to quantify the deviation between actual operation and predicted scenarios. The system extracts actual state features from the actual operating status information, including parameters such as the actual charging load power, the actual output power of distributed power sources, and the actual grid electricity price for the current period. The statistical characteristics of these parameters are calculated to form an actual state feature vector. Simultaneously, the feature vector of the target typical scenario corresponding to the baseline dispatch instruction is retrieved, which has been determined in step S202. The differences between the actual state feature vector and the typical scenario feature vector in each dimension are calculated to form a feature deviation vector. Each component of this vector represents the magnitude and direction of the deviation in the corresponding feature dimension.

[0051] S204: Using a dynamic adjustment function, the baseline scheduling command is adjusted according to the characteristic deviation vector to generate a real-time charging power command.

[0052] The system inputs the characteristic deviation vector calculated in step S203 into the dynamic adjustment function. This function calculates the real-time peak-shaving error based on the magnitude or key component value of the deviation vector. It then compares the real-time peak-shaving error with a preset grid peak-shaving error threshold and determines the current values ​​of the operator's revenue weight and peak-shaving constraint weight based on the comparison result. When the real-time peak-shaving error is less than the threshold, the dynamic adjustment function outputs a larger revenue weight and a smaller peak-shaving constraint weight; when the real-time peak-shaving error exceeds the threshold, the function outputs a smaller revenue weight and a larger peak-shaving constraint weight. The updated weight values ​​are used to reconstruct the intraday optimization objective function. Within the power adjustment range limited by a preset elastic floating threshold, local optimization is performed using the benchmark dispatch command as the initial value to obtain the corrected charging power command. This real-time charging power command comprehensively considers actual operating deviations, grid peak-shaving demand, and operator economic benefits. It is then sent to the charging pile execution unit via the communication interface to achieve refined control of the charging power.

[0053] Based on the above embodiments, as an optional implementation method, the method of determining the target typical scene corresponding to the target key scene features based on the mapping relationship in S202 can be implemented through the following steps S2041-S2043.

[0054] S2041: Obtain source-load fluctuation characteristics under high-proportion renewable energy access scenarios and determine multiple elastic floating thresholds for baseline dispatch instructions; Among them, the high proportion of renewable energy access scenario refers to the operation scenario in which the installed capacity of renewable energy such as distributed photovoltaic and wind power accounts for a high proportion of the total capacity of charging stations or distribution networks. In this scenario, both power output and load demand exhibit significant random fluctuations. Source-load fluctuation characteristics refer to the changing patterns of renewable energy output and charging load over time, including statistical characteristics such as fluctuation amplitude, fluctuation frequency, and fluctuation trend. The flexible floating threshold represents the upper and lower floating limits that allow real-time charging power commands to be adjusted based on the baseline dispatch command. This threshold is dynamically set according to the source-load fluctuation characteristics, and has a certain degree of flexibility and adaptability.

[0055] This step is performed before the intraday rolling correction layer is initiated. First, time-series data of distributed generation output and charging load under high-proportion renewable energy access conditions are extracted from the historical operating database. Statistical analysis is performed on this data to calculate parameters such as the standard deviation, coefficient of variation, and maximum fluctuation amplitude of output and load, quantifying the source-load fluctuation characteristics. Then, based on the source-load fluctuation characteristics, a strategy for setting the flexible floating threshold is determined. When the source-load fluctuation amplitude is large, a wider floating threshold is set to allow for greater power adjustment space; when the source-load fluctuation amplitude is small, a narrower floating threshold is set to maintain the stability of the dispatch plan. Multiple differentiated flexible floating thresholds are set for different time periods and power levels of the baseline dispatch command. Stricter thresholds are set during high-load periods to ensure grid security, while more lenient thresholds are set during low-load periods to improve economic efficiency. After determining the flexible floating thresholds, these thresholds are stored as constraint parameters for intraday correction, used for subsequent power range division and boundary constraints for local optimization, ensuring that real-time adjustments respond to actual deviations without deviating too far from the day-ahead plan.

[0056] S2042: Calculate the power fluctuation corresponding to the characteristic deviation vector, and determine the power range in which the power fluctuation rate is located based on the elastic floating threshold. Among them, the power fluctuation amount represents the magnitude of the charging power adjustment required due to the deviation between the actual operation and the predicted scenario. This value reflects the power difference between the actual state and the baseline scheduling. The power range refers to the range of allowable power adjustment divided according to the magnitude and direction of the power fluctuation amount and the elastic floating threshold. Different ranges correspond to different adjustment strategies and optimization target weights.

[0057] This step is executed after obtaining the characteristic deviation vector. First, each characteristic component in the characteristic deviation vector is converted into a quantitative impact on power adjustment requirements. When the actual charging load is higher than the typical scenario load, charging power needs to be increased or discharging power needs to be reduced. When the actual output of distributed power sources is lower than the typical scenario output, charging power needs to be reduced or discharging power needs to be increased. When the real-time electricity price is higher than the typical scenario electricity price, charging power needs to be reduced to lower costs. Combining the impact of each characteristic component, the overall power fluctuation is calculated. This value represents the power amplitude and direction that needs to be adjusted relative to the baseline dispatch command. Then, the calculated power fluctuation is compared with multiple elastic floating thresholds determined in step S2041. The power range is determined based on the absolute value of the fluctuation. When the fluctuation is less than the first threshold, it is classified as a small deviation range. When the fluctuation is between the first and second thresholds, it is classified as a medium deviation range. When the fluctuation exceeds the second threshold, it is classified as a large deviation range.

[0058] S2043: Within the power range, the objective function is locally optimized using a dynamic adjustment function to generate a real-time charging power command.

[0059] Specifically, firstly, based on the determined power range, a dynamic adjustment function is invoked to calculate the operator revenue weight and peak-shaving constraint weight for the current moment. When the deviation range is small, the dynamic adjustment function outputs a larger revenue weight value and a smaller peak-shaving weight value; when the deviation range is large, it outputs a smaller revenue weight value and a larger peak-shaving weight value; and when the deviation range is medium, it outputs a weight value between the two. The updated weight values ​​are then used to reconstruct the intraday optimization objective function, which dynamically adjusts the balance point between economic benefits and peak-shaving performance based on the actual deviation. Next, local optimization is performed within the power adjustment range defined by the power range. Using the charging power corresponding to the baseline scheduling command as the initial value, the function searches within the upper and lower boundaries of the range to achieve the optimal charging power value for the objective function. The local optimization process employs a fast optimization algorithm, which significantly reduces computation time compared to global optimization due to the significantly smaller search space, meeting the timeliness requirements of intraday real-time scheduling. After optimization, a corrected charging power value is obtained. This value responds to the deviation between actual operation and predicted scenarios, maintains a reasonable deviation range from the baseline scheduling command under the constraint of an elastic floating threshold, and achieves an adaptive balance between economy and peak-shaving performance through dynamic weight adjustment.

[0060] Based on the above embodiments, as an optional implementation method, the method of generating multiple random scene samples based on the joint probability distribution model in S102 and selecting multiple typical scenes from the multiple random scene samples using a preset clustering effectiveness index can be specifically implemented through the following steps S301-S304.

[0061] S301: Based on the joint probability distribution model, a preset number of random scenario samples are generated. Each random scenario sample contains time-series data of charging load, distributed power output, and market electricity price. Among them, the preset quantity refers to the total number of random scenario samples predetermined based on scenario generation requirements and computing resource constraints. This quantity needs to be large enough to cover the main distribution areas of the uncertainty space, while also being controlled within a manageable range. Time series data refers to a data sequence arranged in chronological order, recording the numerical changes of charging load, distributed power output, and market electricity price at each time step within a preset period.

[0062] This step is performed after the joint probability distribution model is constructed. First, the number of random scenario samples to be generated is determined. This number is determined comprehensively based on the scheduling cycle length, time step accuracy, and uncertainty complexity, typically set to several thousand to tens of thousands of samples to ensure sufficient scenario coverage. Then, based on the constructed joint probability distribution model, random sampling techniques are used to extract samples from the model. Each sampling generates a set of numerical triplets containing charging load, distributed power generation output, and market electricity price. This triplet reflects the joint probability distribution characteristics of the three variables at a given time. The sampling process is repeated for each time step within the preset cycle, generating three-dimensional values ​​for that time step. The sampling results from all time steps are combined chronologically to form a complete random scenario sample. This sample contains the time-series curves of charging load, distributed power generation output, and market electricity price within the preset cycle. The above sampling and combination process is repeated until the number of generated random scenario samples reaches the preset number.

[0063] S302: Normalize the generated random scene samples and extract the statistical feature parameters of each random scene sample as clustering feature vectors. Normalization refers to a data preprocessing method that converts data with different dimensions and numerical ranges into a unified standard interval, eliminating the impact of dimensional differences between different physical quantities on subsequent analysis; statistical characteristic parameters refer to statistical quantities that can describe the distribution characteristics and variation patterns of time series data, including central trend parameters, dispersion parameters, extreme value parameters, and morphological parameters.

[0064] This step is performed after the random scenario samples are generated. First, all generated random scenario samples are normalized. For each scenario sample, the maximum and minimum values ​​of the charging load time-series data, distributed power generation output time-series data, and market electricity price time-series data are calculated among all samples. Then, the minimum value is subtracted from the original value of each data point, and the result is divided by the difference between the maximum and minimum values ​​to obtain the normalized value, which falls within the range of zero to one. After normalization, statistical feature parameters are extracted for each scenario sample. For the charging load time-series data, the mean is calculated to reflect the average load level, the standard deviation is calculated to reflect the load fluctuation, the maximum and minimum values ​​are extracted to reflect load extreme characteristics, and the peak-valley difference is calculated to reflect the load variation amplitude. A similar feature extraction process is performed on the distributed power generation output time-series data and the market electricity price time-series data, calculating their respective mean, standard deviation, extreme values, and variation amplitude. Furthermore, cross-variable correlation features are extracted: the correlation coefficient between charging load and power generation output is calculated to reflect the source-load matching degree, and the load proportion during high electricity price periods is calculated to reflect economic pressure characteristics. All extracted statistical feature parameters are arranged in a fixed order to form the clustering feature vector of the scene sample. The dimension of each vector is determined by the number of feature parameters.

[0065] S303: Perform cluster analysis on random scene samples and group scene samples with similar features into the same cluster; Cluster analysis is an unsupervised learning method that divides a dataset into several subsets according to similarity criteria, so that data within the same subset have high similarity while data between different subsets have low similarity. A cluster is a subset of data formed through cluster analysis, and each cluster contains a group of scene samples with similar characteristics.

[0066] This step is performed after the clustering feature vectors are extracted. First, a suitable clustering algorithm is selected. The algorithm type is determined based on the distribution characteristics of the clustering feature vectors and the required number of clusters. When the number of clusters is known in advance, a partition-based clustering method is used; when the number of clusters is unknown and needs to be automatically determined, a density-based or hierarchical clustering method is used. When executing the clustering algorithm, all clustering feature vectors are used as input data. The algorithm calculates the distance or similarity between any two feature vectors. The smaller the distance or the higher the similarity, the closer the operating characteristics of the corresponding scene samples are. Based on the distance or similarity metric, the clustering algorithm iteratively adjusts the cluster affiliation of each scene sample, minimizing the sum of distances between samples within a cluster or maximizing the sum of distances between samples between clusters. The iterative process continues until the cluster affiliation no longer changes or the preset maximum number of iterations is reached. After clustering, several clusters are obtained, each containing a group of scene samples with similar clustering feature vectors. Scene samples belonging to the same cluster have similar load peak-valley characteristics, power output patterns, and electricity price distribution patterns, representing a typical operating condition. The scene samples from different clusters show significant differences, each corresponding to a different category of operating scenario.

[0067] S304: Calculate the clustering effectiveness index corresponding to each cluster, and select the scene sample closest to the cluster center from each cluster as the typical scene of that cluster.

[0068] The cluster center represents the average position or center point of the cluster feature vectors of all scene samples in the cluster. This point best represents the overall characteristics of the cluster in the feature space.

[0069] This step is performed after cluster analysis is complete. First, a clustering effectiveness index is calculated for each cluster, which is used to evaluate the quality of the clustering results. When calculating the intra-cluster compactness index, for all scene samples within the cluster, the distance from the cluster feature vector of each sample to the cluster center is calculated. All distances are averaged to obtain the intra-cluster average distance; the smaller the distance, the more compact the samples within the cluster. When calculating the inter-cluster separation index, the distance between the cluster center of the current cluster and the cluster centers of all other clusters is calculated. The minimum distance is taken as the inter-cluster separation; the larger the distance, the higher the discriminative power between different clusters. Combining intra-cluster compactness and inter-cluster separation, a clustering effectiveness index is formed, which reflects the reasonableness of the clustering results. After calculating the clustering effectiveness index, a typical scene is selected within each cluster, and the location of the cluster center is calculated. The values ​​of each dimension of the center location are the average values ​​of the corresponding dimensions of all scene samples within the cluster. Then, the distance from the cluster feature vector of each scene sample within the cluster to the cluster center is calculated using the square root of the sum of the squared differences in each dimension. Compare the distances of all samples to the cluster centers, and select the scene sample with the smallest distance as the typical scene of that cluster. This scene is closest to the cluster center in the feature space and best represents the average operating characteristics of the cluster. Repeat the above selection process for all clusters to obtain a set of typical scenes equal to the number of clusters. Each typical scene corresponds to a type of operating condition.

[0070] Based on the above embodiments, as an optional implementation method, the method of determining the dynamic adjustment function of operator revenue weight and real-time peak shaving error according to the power grid peak shaving error threshold corresponding to each typical scenario in S104 can be implemented through the following steps S401-S403.

[0071] S401: Obtain power grid peak-shaving error data for each typical scenario during historical operation, and statistically analyze the power grid peak-shaving error threshold corresponding to each typical scenario; Among them, the historical operation period represents the operating period when each typical scenario was actually applied to the scheduling of charging stations in the past. During this period, real scheduling execution data and grid response data were accumulated. The grid peak shaving error data refers to the deviation record between the peak shaving power actually provided by the charging station and the peak shaving command power issued by the grid. This data reflects the accuracy of scheduling execution and the degree to which grid demand is met under different scenarios.

[0072] This step is performed after the typical scenario library is built. First, peak-shaving execution records for each typical scenario are retrieved from the historical database. The grid peak-shaving command value and the actual peak-shaving power value executed by the charging station are extracted for each time step. The difference between the two is calculated to obtain the peak-shaving error at that moment. A positive value indicates insufficient actual peak-shaving power, while a negative value indicates that the actual peak-shaving power exceeds demand. Peak-shaving error data for all historical operating periods of each typical scenario are collected to form the error time series for that scenario. Then, statistical analysis is performed on the error data for each typical scenario. The mean error reflects systematic deviation, the standard deviation reflects the degree of fluctuation, and the maximum and minimum error values ​​reflect extreme deviations. Based on the statistical distribution characteristics of the error data, the corresponding grid peak-shaving error threshold for each typical scenario is determined. This threshold setting needs to comprehensively consider the grid safety margin requirements and the actual dispatch execution capability. For typical scenarios with high load peaks or strict grid safety requirements, a smaller error threshold is set to ensure peak-shaving accuracy; for typical scenarios with stable loads or large grid margins, a larger error threshold is set to provide more optimization flexibility.

[0073] S402: Construct a nonlinear mapping relationship between operator revenue weight and real-time peak shaving error. When the real-time peak shaving error is less than a preset threshold, increase the operator revenue weight; when the real-time peak shaving error exceeds the preset threshold, decrease the operator revenue weight and increase the peak shaving constraint weight. Among them, the nonlinear mapping relationship represents a function mapping between input variables and output variables that is not linearly proportional. This mapping can produce a non-uniformly changing output response according to different value ranges of the input variables. The peak-shaving constraint weight is used to represent the relative importance of the peak-shaving performance constraint objective in the multi-objective optimization function. The larger the weight value, the more the optimization process focuses on meeting the peak-shaving needs of the power grid.

[0074] This step is executed after the grid peak-shaving error threshold is determined. First, the reasonable relationship between operator revenue weight and real-time peak-shaving error is analyzed. When the real-time peak-shaving error is small, it indicates that the charging station's peak-shaving execution is highly aligned with grid demand. At this time, grid security pressure is low, and scheduling optimization has more room to pursue economic benefits. Therefore, the operator revenue weight needs to be increased to make the optimization objective more biased towards economic goals such as charge-discharge arbitrage and peak-valley electricity price utilization. When the real-time peak-shaving error exceeds the preset threshold, it indicates that the charging station's peak-shaving execution deviates significantly from grid demand, and the grid faces security risks or frequency stability pressure. At this time, the operator revenue weight needs to be reduced to decrease the pursuit of economic benefits, while the peak-shaving constraint weight needs to be increased to strengthen the tracking capability of grid peak-shaving commands and ensure that peak-shaving performance is prioritized. Based on the above analysis, a nonlinear mapping relationship between operator revenue weight and real-time peak-shaving error is constructed. This mapping relationship maintains a high value of revenue weight within the safe range where the error is small, smoothly decreases the revenue weight within the warning range where the error approaches the threshold, and rapidly reduces the revenue weight to a low level within the danger range where the error exceeds the threshold. Correspondingly, the trend of the peak-shaving constraint weight is opposite to that of the return weight: it remains at a low value in the safe range, rises smoothly in the warning range, and rises rapidly to a high level in the danger range.

[0075] S403: Establish a dynamic adjustment function based on nonlinear mapping relationship.

[0076] Specifically, the dynamic adjustment function uses real-time peak-shaving error as the input variable and operator revenue weight and peak-shaving constraint weight as the output variables. The function contains piecewise or continuously smoothed mathematical expressions to achieve a non-linear transformation from input to output. When establishing the function, first, the domain of the function is determined, i.e., the range of values ​​for the real-time peak-shaving error, covering the interval from zero error to the maximum historical error. Then, different sub-intervals are divided within the domain, each corresponding to a different error level. Within the safe interval, the function outputs high-revenue weights and low-peak-shaving weights; within the warning interval, the function outputs weights with a gradient change; and within the danger interval, the function outputs low-revenue weights and high-peak-shaving weights.

[0077] The following are system embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the system embodiments of this application, please refer to the method embodiments of the application.

[0078] Please see Figure 2 This illustration shows a schematic diagram of a multi-objective cooperative charging scheduling optimization system provided in an exemplary embodiment of this application. The system can be implemented through software, hardware, or a combination of both, forming all or part of a larger system. The multi-objective cooperative charging scheduling optimization system includes: The data acquisition module is used to collect historical charging load data, distributed power output data and market electricity price data of the charging station, and to construct a joint probability distribution model that characterizes numerical features and coupling relationships. The scene generation module is used to generate multiple random scene samples based on the joint probability distribution model, and to select multiple typical scenes from the multiple random scene samples using a preset clustering effectiveness index. The feature extraction module is used to extract key scene features for each typical scenario and to construct a mapping relationship between each typical scenario and the key scene features. The model building module is used to determine the dynamic adjustment function of operator revenue weight and real-time peak-shaving error based on the power grid peak-shaving error threshold corresponding to each typical scenario, and to build a two-layer collaborative optimization model including a day-ahead pre-scheduling layer and an intraday rolling correction layer. The processing module is used to acquire the target key scene features and actual operating status information of the charging station, input the target key scene features and actual operating status information into the two-layer collaborative optimization model, and output real-time charging power commands.

[0079] This application also provides a computer storage medium that can store multiple instructions. The instructions are adapted to be loaded and executed by a processor using the multi-objective cooperative charging scheduling optimization method as described above. For details of the execution process, please refer to the specific description of the embodiments, which will not be repeated here.

[0080] Please see Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 3 As shown, the electronic device 300 may include: at least one processor 301, at least one network interface 304, user interface 303, memory 305, and at least one communication bus 302.

[0081] The communication bus 302 is used to enable communication between these components.

[0082] The user interface 303 may include a display screen and a camera.

[0083] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0084] The processor 301 may include one or more processing cores. The processor 301 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 305, and by calling data stored in memory 305. Optionally, the processor 301 may be implemented using at least one hardware form of digital signal processing, field-programmable gate array, or programmable logic array. The processor 301 may integrate one or more of the following: a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.

[0085] The memory 305 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 305 may include a non-transitory computer-readable medium. The memory 305 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch functionality, sound playback functionality, image playback functionality, etc.), instructions for implementing the various method embodiments described above, etc.; the data storage area may store data involved in the various method embodiments described above, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned processor 301. Figure 3 As shown, the memory 305, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a multi-objective cooperative charging scheduling optimization method.

[0086] exist Figure 3 In the electronic device 300 shown, the user interface 303 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 301 can be used to call the application program of a multi-objective cooperative charging scheduling optimization method stored in the memory 305. When executed by one or more processors, the electronic device executes one or more methods as described in the above embodiments.

[0087] An electronic device readable storage medium stores instructions that, when executed by one or more processors, cause the electronic device to perform one or more methods as described in the above embodiments.

[0088] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0089] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0090] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0091] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0092] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0093] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0094] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and practical application disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure.

Claims

1. A multi-objective cooperative charging scheduling optimization method, characterized in that, The method includes: Historical charging load data, distributed power output data, and market electricity price data of charging stations are collected to construct a joint probability distribution model that characterizes numerical features and coupling relationships. Multiple random scene samples are generated based on the joint probability distribution model, and multiple typical scenes are selected from the multiple random scene samples using a preset clustering effectiveness index. Extract key scene features from each of the typical scenarios and construct a mapping relationship between each of the typical scenarios and the key scene features; Based on the power grid peak-shaving error threshold corresponding to each typical scenario, the dynamic adjustment function of operator revenue weight and real-time peak-shaving error is determined, and a two-layer collaborative optimization model including a day-ahead pre-scheduling layer and an intraday rolling correction layer is constructed. The system acquires the target key scene features and actual operating status information of the charging station, inputs the target key scene features and actual operating status information into a two-layer collaborative optimization model, and outputs real-time charging power commands.

2. The method according to claim 1, characterized in that, The process of acquiring the target key scene features and actual operating status information of the charging station, inputting the target key scene features and actual operating status information into a two-layer collaborative optimization model, and outputting real-time charging power commands includes: The target key scene features of the charging station within a preset period of the pre-scheduling layer and the actual operating status information at the current moment are obtained, and the target key scene features and the actual operating status information are input into the two-layer collaborative optimization model. In the day-ahead pre-scheduling layer, the target typical scenario corresponding to the target key scenario feature is determined based on the mapping relationship, and the optimal solution of the target typical scenario is used as the benchmark scheduling instruction of the day-ahead pre-scheduling layer. In the intraday rolling correction layer, the actual state features in the actual operating state information are extracted, and the feature deviation vector between the actual state features and the typical scenario features corresponding to the baseline scheduling instruction is calculated. Using the dynamic adjustment function, the baseline scheduling command is adjusted according to the characteristic deviation vector to generate a real-time charging power command.

3. The method according to claim 2, characterized in that, The step of determining the typical target scene corresponding to the key target scene features based on the mapping relationship includes: Calculate the similarity index between the target key scene features and the key scene features of each of the pre-stored typical scenes; Typical scenarios that meet the preset matching conditions based on the similarity index are selected as the target typical scenarios.

4. The method according to claim 2, characterized in that, The step of using the optimal solution of the target typical scenario as the baseline scheduling instruction of the day-ahead pre-scheduling layer includes: Historical charging load data, distributed power output data, and market electricity price data corresponding to the target typical scenario are retrieved as day-ahead optimization boundary conditions. A day-ahead multi-objective optimization function is constructed with the objectives of maximizing operator revenue and minimizing grid load volatility, and battery state of charge constraints, charge and discharge power constraints, and power balance constraints are set. The day-ahead multi-objective optimization function is solved to obtain the optimal charging power sequence for each time step within a future preset period, and the optimal charging power sequence is used as the base scheduling instruction for the day-ahead pre-scheduling layer.

5. The method according to claim 2, characterized in that, The step of adjusting the baseline scheduling command based on the feature deviation vector using the dynamic adjustment function to generate a real-time charging power command includes: To obtain the source-load fluctuation characteristics under a high proportion of renewable energy access scenario, and to determine multiple elastic floating thresholds for the benchmark scheduling command; Calculate the power fluctuation corresponding to the feature deviation vector, and determine the power range in which the power fluctuation rate is located based on the elastic floating threshold; Within the power range, the objective function is locally optimized using the dynamic adjustment function to generate a real-time charging power command.

6. The method according to claim 1, characterized in that, The process involves generating multiple random scene samples based on the joint probability distribution model, and then using a preset clustering effectiveness index to select multiple typical scenes from these random scene samples, including: Based on the joint probability distribution model, a preset number of random scenario samples are generated, and each random scenario sample contains time-series data of charging load, distributed power output, and market electricity price. The generated random scene samples are normalized, and the statistical feature parameters of each random scene sample are extracted as clustering feature vectors. Cluster analysis is performed on the random scene samples to group scene samples with similar features into the same cluster; Calculate the clustering effectiveness index corresponding to each cluster, and select the scene sample closest to the cluster center from each cluster as the typical scene of that cluster.

7. The method according to claim 1, characterized in that, The step of determining the dynamic adjustment function of operator revenue weight and real-time peak-shaving error based on the power grid peak-shaving error threshold corresponding to each typical scenario includes: Obtain power grid peak-shaving error data for each of the typical scenarios during historical operation, and statistically analyze the power grid peak-shaving error thresholds corresponding to each of the typical scenarios; A nonlinear mapping relationship is established between operator revenue weight and real-time peak shaving error. When the real-time peak shaving error is less than a preset threshold, the operator revenue weight is increased; when the real-time peak shaving error exceeds the preset threshold, the operator revenue weight is decreased and the peak shaving constraint weight is increased. Based on the aforementioned nonlinear mapping relationship, a dynamic adjustment function is established.

8. A multi-objective cooperative charging scheduling optimization system, characterized in that, The system includes: The data acquisition module is used to collect historical charging load data, distributed power output data and market electricity price data of the charging station, and to construct a joint probability distribution model that characterizes numerical features and coupling relationships. The scene generation module is used to generate multiple random scene samples based on the joint probability distribution model, and to select multiple typical scenes from the multiple random scene samples using a preset clustering effectiveness index. The feature extraction module is used to extract key scene features of each typical scene and construct a mapping relationship between each typical scene and the key scene features; The model building module is used to determine the dynamic adjustment function of operator revenue weight and real-time peak-shaving error based on the power grid peak-shaving error threshold corresponding to each typical scenario, and to build a two-layer collaborative optimization model including a day-ahead pre-scheduling layer and an intraday rolling correction layer. The processing module is used to acquire the target key scene features and actual operating status information of the charging station, input the target key scene features and actual operating status information into the two-layer collaborative optimization model, and output real-time charging power commands.

9. A computer storage medium, characterized in that, The computer storage medium stores a plurality of instructions, which are adapted to be loaded by a processor and executed as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The device includes a processor, a memory, and a transceiver, wherein the memory is used to store instructions, the transceiver is used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

Images