Method and System for Generating Power Load Scenarios by Integrating Cross-Seasonal Data

Abstract

This application proposes a method and system for generating power load scenarios by fusing cross-seasonal data, belonging to the field of power load scenario generation technology. The method includes: determining the collection time window for the target area and collecting cross-seasonal time-series data for that area; parsing the cross-seasonal time-series data to perform non-standard seasonal division, obtaining multiple candidate seasonal intervals; performing rapid modeling based on mathematical models for the multiple candidate seasonal intervals to generate multiple candidate load scenarios; calculating the representativeness and coverage metrics of the candidate load scenarios and performing adaptive interval adjustments to output multiple specialized seasonal intervals; forming specialized power load scenarios and merging these specialized power load scenarios to output the target power load scenario. By fusing cross-seasonal data and performing adaptive seasonal interval adjustments, the representativeness and coverage of the power load scenarios are improved, thereby generating more accurate and comprehensive power load scenarios.

Description

Technical Field

[0001] This invention relates to the field of power load scenario generation technology, and in particular to a method and system for generating power load scenarios by integrating cross-seasonal data. Background Technology

[0002] Existing power load scenario generation methods primarily employ a modeling approach based on standard seasonal divisions. This involves dividing time periods according to the traditional four seasons (spring, summer, autumn, and winter) or by month, and then establishing separate load forecasting models for each standard season. While this method considers the impact of seasonal factors on power load to some extent, it suffers from the following shortcomings: First, the standard seasonal division method is too rigid and fails to reflect the differences in climate characteristics and electricity consumption habits across different regions, resulting in inaccurate seasonal boundary delineation. Second, existing methods often limit themselves to data within a single season during the modeling process, failing to fully explore and utilize the load variation patterns and correlation characteristics inherent in cross-seasonal data. These technical deficiencies lead to insufficient representativeness and coverage of existing power load scenario generation methods when facing complex and ever-changing actual load conditions. These methods struggle to comprehensively reflect the true load characteristics of the target area, thus affecting the accuracy and reliability of power system dispatching. Summary of the Invention

[0003] This invention addresses the technical problem in existing technologies where power load scenario generation is based solely on standard seasonal divisions and fails to fully utilize cross-seasonal data features, resulting in insufficient representativeness and coverage of load scenarios. It provides a method and system for generating power load scenarios by integrating cross-seasonal data to solve this problem.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0005] In a first aspect, the present invention provides a method for generating power load scenarios by integrating cross-seasonal data, comprising: determining a collection time window for a target area, and collecting cross-seasonal time-series data of the target area according to the collection time window; parsing the cross-seasonal time-series data to perform non-standard seasonal division and obtain multiple candidate seasonal intervals; performing rapid modeling based on mathematical models for the multiple candidate seasonal intervals according to the cross-seasonal time-series data to generate multiple candidate load scenarios; calculating representativeness metrics and coverage metrics of the candidate load scenarios, and performing adaptive interval adjustments for the candidate load scenarios accordingly, and outputting multiple specialized seasonal intervals according to the adjustment results; combining the multiple specialized seasonal intervals with the cross-seasonal time-series data to perform iterative modeling to form specialized power load scenarios, and merging and outputting the multiple specialized power load scenarios as a target power load scenario.

[0006] Secondly, this invention provides a power load scenario generation system that integrates cross-seasonal data. The system includes a data acquisition module for determining the acquisition time window for a target area and acquiring cross-seasonal time-series data for that target area according to the acquisition time window; a season division module for parsing the cross-seasonal time-series data to perform non-standard season division and obtain multiple candidate season intervals; a rapid modeling module for performing rapid modeling based on mathematical models for the multiple candidate season intervals according to the cross-seasonal time-series data to generate multiple candidate load scenarios; an adaptive adjustment module for calculating representativeness and coverage metrics of the candidate load scenarios and performing adaptive interval adjustments accordingly, outputting multiple specialized season intervals based on the adjustment results; and an iterative fusion module for combining the multiple specialized season intervals with the cross-seasonal time-series data to perform iterative modeling, forming specialized power load scenarios, and merging and outputting the multiple specialized power load scenarios as the target power load scenario.

[0007] The beneficial effects of this invention are:

[0008] The data collection time window for the target area is determined, and cross-seasonal time-series data for the target area is collected according to the corresponding time window to obtain continuous time-series load data covering multiple seasons, laying the data foundation for subsequent cross-seasonal data analysis and modeling. The cross-seasonal time-series data is analyzed to perform non-standard seasonal division, obtaining multiple candidate seasonal intervals. This overcomes the limitations of traditional standard seasonal divisions and allows for flexible seasonal boundary identification based on the characteristics of actual load data, resulting in seasonal interval divisions that better reflect the load variation patterns of the target area. Based on the cross-seasonal time-series data, rapid mathematical modeling is performed on multiple candidate seasonal intervals to generate multiple candidate load scenarios, thereby quickly constructing each candidate seasonal region. The system uses a load forecasting model to initially generate corresponding load scenarios, providing candidate solutions for subsequent optimization and adjustment. It calculates representativeness and coverage metrics for candidate load scenarios and performs adaptive interval adjustments accordingly. Based on the adjustment results, it outputs multiple specialized seasonal intervals. By quantitatively evaluating the quality of load scenarios, it identifies intervals with insufficient representativeness or coverage and performs adaptive adjustments to obtain more accurate specialized seasonal intervals. It iterative modeling is then performed by combining multiple specialized seasonal intervals with cross-seasonal time-series data to form specialized power load scenarios. These specialized power load scenarios are then merged and output to generate a target power load scenario with high representativeness and high coverage.

[0009] The above technical solutions can fully utilize cross-seasonal data characteristics to achieve adaptive seasonal interval optimization, improve the representativeness and coverage of power load scenarios, and thus generate more accurate and comprehensive power load scenarios. Attached Figure Description

[0010] Figure 1 A flowchart illustrating the method for generating power load scenarios by fusing cross-seasonal data provided by the present invention;

[0011] Figure 2 A schematic diagram of the structure of the power load scenario generation system that integrates cross-seasonal data provided by the present invention.

[0012] In the attached diagram, the components represented by each number are as follows:

[0013] Data acquisition module 11, seasonal division module 12, quick modeling module 13, adaptive adjustment module 14, iterative fusion module 15. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] In the description of this invention, 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 number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0016] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0017] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for generating power load scenarios by fusing cross-seasonal data, including:

[0018] S1. Determine the collection time window for the target area, and collect cross-seasonal time series data of the target area according to the collection time window.

[0019] Specifically, the data collection time window for the target area is determined, and corresponding cross-seasonal time-series data is collected based on this time window to provide basic data for generating power load scenarios.

[0020] The target area refers to the geographical region where power load scenario analysis needs to be conducted, which can be a city, industrial park, residential area, or other power supply area with clear boundaries. The data collection time window refers to the time range used for data collection, which is determined based on the periodic variation and seasonal characteristics of power load. In actual implementation, determining the data collection time window requires comprehensive consideration of the following factors: First, ensuring data coverage of the complete seasonal cycle to obtain a full picture of load changes; second, ensuring data continuity and completeness to avoid analytical biases caused by missing data; and third, considering the practical operability and economic efficiency of data collection.

[0021] The data is collected based on cross-seasonal time-series data for the target area within the corresponding time window. Cross-seasonal time-series data refers to time series data spanning multiple seasons, including but not limited to load information, meteorological information, and date-type information. Load information includes actual power load values ​​and load change rates; meteorological information includes environmental parameters such as temperature, humidity, wind speed, and light intensity; and date-type information includes time attribute identifiers such as weekdays, rest days, and holidays.

[0022] By collecting cross-seasonal time-series data, a data foundation was laid for subsequent seasonal division and load scenario modeling, ensuring that the generated power load scenarios have good representativeness and practicality.

[0023] S2. Parse the cross-seasonal time series data to perform non-standard seasonal division and obtain multiple candidate seasonal intervals.

[0024] Specifically, non-standard seasonal division refers to an adaptive division of seasonal intervals based on the actual characteristics and patterns of power load changes, rather than following the traditional four seasons. This method better reflects the true changing patterns of power load and avoids the problem of load characteristic confusion that may arise from traditional fixed seasonal divisions.

[0025] In practical implementation, the first step is to analyze the acquired cross-seasonal time-series data, extracting load information, meteorological information, and date type information to extract load-related features, meteorological features, and time-cycle features, thus constructing a multi-dimensional feature vector. Load-related features include peak load, average load, load fluctuation amplitude, and load change trend; meteorological features include temperature change rate, humidity fluctuation, and wind speed level; and time-cycle features include time-regularity indicators such as daily, weekly, and monthly cycles.

[0026] Subsequently, based on the constructed multi-dimensional feature vectors, a clustering algorithm is used to cluster the cross-seasonal time-series data, yielding the clustering results. Preferably, a community detection algorithm based on graph neural networks can be used for cluster analysis. This algorithm can effectively identify similarity patterns and inherent relationships in time-series data, thereby achieving more accurate seasonal interval division.

[0027] Based on the clustering results, multiple candidate seasonal intervals can be defined, with each interval exhibiting similar characteristics and patterns in power load variation. These candidate seasonal intervals provide a time segmentation basis for subsequent mathematical modeling and scenario generation, thereby effectively improving the accuracy and representativeness of power load scenario generation.

[0028] S3. Based on the cross-seasonal time series data, perform rapid modeling based on mathematical models for multiple candidate seasonal intervals to generate multiple candidate load scenarios.

[0029] Specifically, rapid modeling refers to the use of computationally efficient and fast-converging mathematical modeling methods that can quickly generate preliminary load scenarios while ensuring modeling accuracy. Compared to traditional complex modeling methods, it has the advantages of lower computational cost and stronger real-time performance.

[0030] In practice, the process begins by performing homomorphic extraction on the cross-seasonal time series data based on multiple candidate seasonal intervals, resulting in multiple interval time series data. Homomorphic extraction refers to dividing the complete cross-seasonal time series data into corresponding time periods according to the time boundaries of the candidate seasonal intervals, while maintaining the original structure and characteristics of the data. This ensures that each interval time series data contains complete load information, meteorological information, and date type information.

[0031] Subsequently, a mathematical regression analysis model of time-scenario load indicators was established by traversing multiple time-series data intervals. This model describes the mathematical relationship between time variables and load scenarios. By analyzing the variation patterns of load over time within an interval, it can reflect the changing trends and fluctuation characteristics of load within a specific seasonal interval.

[0032] Next, the fitting performance of mathematical regression analysis models for multiple time-scenario load indicators is verified, and the predictive accuracy and stability of the models are evaluated. The mathematical regression analysis models for time-scenario load indicators that meet the basic performance constraints are output as candidate load scenarios, thereby obtaining multiple preliminary candidate load scenarios, providing basic model support for subsequent adaptive adjustment and optimization.

[0033] By performing rapid modeling based on mathematical models for multiple alternative seasonal intervals, multiple alternative load scenarios are generated, providing a preliminary modeling foundation for subsequent scenario optimization and adjustment.

[0034] S4. Calculate the representative metrics and coverage metrics of the candidate load scenarios, and perform adaptive interval adjustments for the candidate load scenarios accordingly. Output multiple specialized seasonal intervals based on the adjustment results.

[0035] Specifically, the representativeness metric measures the ability of alternative load scenarios to represent actual load data, reflecting the model's fitting accuracy and prediction precision. The coverage metric assesses the extent to which alternative load scenarios cover the range of load variations, ensuring that the generated scenarios can cover the main variation range of the actual load.

[0036] First, the average load restoration accuracy of each candidate load scenario within its corresponding interval is calculated as a representative metric. This representative metric quantifies the model's representational effectiveness by comparing the deviation between the predicted and actual load values. Then, based on a preset bias coefficient and the representative metric, coverage boundary constraints are calculated. Through statistical analysis, the percentage of data satisfying the coverage boundary constraints is calculated, and this percentage is output as a coverage metric. This coverage metric reflects the adequacy of the candidate load scenarios' coverage of load fluctuation ranges.

[0037] Next, based on the calculation results of representative and coverage metrics, adaptive interval adjustments are performed. If either metric is less than the lower limit of the preset advanced performance constraint interval, the corresponding candidate load scenario is narrowed to reduce the seasonal interval range and improve modeling accuracy. If both metrics are greater than the upper limit of the advanced performance constraint interval, the corresponding candidate load scenario is expanded, and rapid modeling based on a mathematical model is iteratively performed until the advanced performance constraint interval requirements are met. Afterward, the interval reduction and expansion results are merged to obtain multiple specialized seasonal intervals. Through adaptive interval adjustment, the optimized configuration of seasonal intervals is achieved, laying a solid foundation for generating high-quality power load scenarios.

[0038] By adaptively adjusting the range of alternative load scenarios, multiple specialized seasonal ranges are output, thus realizing the transformation from preliminary modeling to accurate modeling.

[0039] S5. Iteratively model the data by combining multiple specialized seasonal intervals with the cross-seasonal time series data to form specialized power load scenarios, and merge and output multiple specialized power load scenarios as the target power load scenario.

[0040] Specifically, iterative modeling refers to repeatedly adjusting the load scenario based on the optimized specialized seasonal interval, gradually improving the fitting accuracy and generalization ability. Specialized power load scenarios refer to high-precision load scenarios optimized for specific seasonal intervals, possessing stronger representativeness and accuracy.

[0041] First, based on the obtained multiple specialized seasonal intervals, homomorphic extraction is performed on the cross-seasonal time-series data to form multiple specialized interval time-series data. Compared with the initial interval time-series data, this specialized interval time-series data has more accurate time boundaries and higher data quality. Then, by traversing multiple specialized interval time-series data, regression analysis models of time-scenario load indicators are established, and the fitting performance of these regression analysis models is verified. Through iterative optimization, the parameters of the time-scenario load indicator regression analysis models are continuously adjusted to improve the model's prediction accuracy and stability. The output consists of multiple regression analysis models that meet the target performance constraints, representing the specialized power load scenarios.

[0042] Next, scenario fusion processing is performed on the generated multiple specialized power load scenarios. Specifically, multiple specialized power load scenarios are traversed, and a smooth transition interval is calculated based on a preset smoothing operator. Combining a preset smooth transition model, such as a Bézier curve, an adaptive fusion calculation weight for the smooth transition interval is defined, and adjacent specialized power load scenarios are adaptively fused. Then, the multiple specialized power load scenarios are iteratively and adaptively fused to obtain the target power load scenario.

[0043] Through iterative modeling and scenario fusion, high-precision and highly representative power load scenarios were generated, providing reliable scenario support for power system load forecasting and scheduling optimization.

[0044] Furthermore, the data collection time window for the target area is determined, and cross-seasonal time-series data for the target area is collected according to the data collection time window, including:

[0045] S11. Based on confidence analysis of time-varying similarity, perform stability detection on the load change trend of the target area and identify the time period in which the load fluctuation amplitude in a continuous segment is lower than the preset stability threshold.

[0046] S12. Define the interval in the identification result where the confidence level is higher than a preset threshold as the collection time window, and use the collection time window as a time range constraint to collect the cross-seasonal time series data.

[0047] In a preferred embodiment, firstly, based on confidence analysis oriented towards time-varying similarity, stability detection is performed on the load change trend of the target area to identify time periods where load fluctuation amplitudes within consecutive segments are lower than a preset stability threshold. Here, confidence analysis oriented towards time-varying similarity refers to assessing the stability and reliability of load data by analyzing the degree of similarity in load change patterns across different time periods. Time-varying similarity reflects the consistency of load changes over time, while confidence quantifies the reliability of this time-varying similarity. Specifically, firstly, historical load data for the target area over several years is acquired to construct a cross-year load dataset. A strategy of gradually reducing the time span is adopted for stability analysis. Initially, a relatively long time span is selected, such as 5 years of load data, and year-on-year alignment analysis is performed on load data from different years according to the same time period. For each time period, load data for the corresponding period of different years is extracted, and the inter-year time-varying similarity is calculated. By comparing the magnitude, trend, and fluctuation patterns of the load in the same period of different years, the inter-year consistency of load change patterns is quantified. If the change patterns of load data in the same period over multiple years are similar and the fluctuation amplitudes are small, the time-varying similarity is high. Based on the time-varying similarity calculation results, a confidence level assessment is performed. Statistical analysis is conducted on the degree of variation in load data for the same period in different years. If the inter-year differences are small and the trends are consistent, the confidence level for that period is high, indicating a stable and reliable load change trend. If the confidence level of the data within the current time span does not meet the requirements, the time span is reduced, for example, from 5 years to 3 years, and the time-varying similarity analysis and confidence level assessment are repeated. This step-by-step reduction is used to find the optimal time span that meets the confidence requirements. Based on the determined suitable time span, time periods within continuous segments where load fluctuations are below a preset stability threshold are further identified. The fluctuation range of load data within each time period is calculated, and the degree of load change within that time period is quantified by the ratio of the difference between the maximum and minimum load values ​​to the average value. Continuous time periods with load fluctuations below the preset stability threshold are selected; these time periods exhibit good load stability characteristics.

[0048] Subsequently, the intervals with confidence levels higher than a preset threshold in the identification results are defined as the collection time window, and the cross-seasonal time series data are collected using this collection time window as a time constraint. Specifically, from the identification results of step S11, the time intervals with confidence levels higher than a preset confidence threshold are determined as the collection time window. This preset threshold is set according to data quality requirements and analysis accuracy needs to ensure that the load change trend within the selected time interval has good interannual stability. Then, within the determined collection time window, cross-seasonal time series data are collected from historical load data, including load information, meteorological information, and date type information. Among them, the load information includes the power load data of the target area within the collection time window, recording the actual load value, change trend, and fluctuation characteristics. The meteorological information includes meteorological environmental data of the same period, such as environmental factors affecting load changes such as temperature, humidity, and wind speed. The date type information includes time attribute identifiers, such as different date types such as weekdays, rest days, and holidays, used to distinguish the differences in load characteristics under different time types.

[0049] By determining the data collection time window based on confidence analysis of time-varying similarity, and employing interannual stability analysis and a gradual reduction strategy, the stability and reliability of load change trends within the data collection time window were ensured, providing high-quality data support for the subsequent generation of cross-seasonal load scenarios.

[0050] Furthermore, the cross-seasonal time-series data is parsed to perform non-standard seasonal division, obtaining multiple candidate seasonal intervals, including:

[0051] S21. Analyze the cross-seasonal time series data, extract load-related features, meteorological features and time cycle features, and construct a multi-dimensional feature vector;

[0052] S22. Based on the multi-dimensional feature vector, the cross-seasonal time series data is clustered and divided, and multiple candidate seasonal intervals are defined based on the clustering results.

[0053] In a preferred embodiment, firstly, seasonal time-series data is analyzed to extract load-related features, meteorological features, and time-cycle features, constructing a multi-dimensional feature vector. Specifically, firstly, the collected cross-seasonal time-series data is analyzed and preprocessed to ensure data integrity and consistency. Three types of key features are extracted from the cross-seasonal time-series data: load-related features, meteorological features, and time-cycle features. The extraction of load-related features includes: calculating statistical features such as peak load, trough load, and average load; analyzing dynamic features such as load change rate, load growth trend, and load decline trend; extracting fluctuation features such as load fluctuation amplitude, load variance, and load standard deviation; and identifying extreme value features such as load peak-to-valley difference and load peak-to-valley ratio. These load-related features can comprehensively describe the numerical characteristics and variation patterns of power load in different time periods. The extraction of meteorological features includes: extracting temperature-related features such as temperature change rate, extreme temperature values, and average temperature; analyzing humidity-related features such as humidity fluctuation and humidity change trend; calculating wind speed-related features such as wind speed level and wind speed change amplitude; and extracting light-related features such as solar radiation intensity and sunshine duration. These meteorological features reflect the degree to which environmental conditions influence changes in electricity load. Extraction of time-cycle features includes: identifying daily cycle features, such as load variation patterns at different times of the day; analyzing weekly cycle features, such as load variation patterns on different days of the week; extracting monthly cycle features, such as load variation trends at different times of the month; and identifying seasonal cycle features, such as load variation characteristics in different seasons. These time-cycle features reveal the temporal regularity of load changes. The extracted load-related features, meteorological features, and time-cycle features are standardized to eliminate the influence of different feature dimensions and numerical ranges. The standardized features are then combined in chronological order to construct a multi-dimensional feature vector. Each time point corresponds to a multi-dimensional feature vector, containing comprehensive information on the load status, meteorological conditions, and time attributes at that time point.

[0054] Then, based on the multi-dimensional feature vectors, the cross-seasonal time series data is clustered, and multiple candidate seasonal intervals are defined based on the clustering results. Specifically, a clustering algorithm is used to perform cluster analysis on the constructed multi-dimensional feature vector sequence. Preferably, a community detection algorithm based on graph neural networks is used for clustering. This algorithm treats each time point in the cross-seasonal time series data as a node in a graph, and the similarity relationship between nodes constitutes the edge weights of the graph. The community detection technique identifies sets of time points with similar characteristics. The specific implementation process of the community detection algorithm based on graph neural networks includes: calculating the similarity measure between time points based on the multi-dimensional feature vectors and constructing a time point similarity graph; using graph neural networks to learn the embedding representation of nodes, capturing high-order relationships and global structural information between nodes; identifying the community structure in the graph through the community detection algorithm, and grouping time points with similar characteristics into the same community; each community corresponds to a set of time points with similar load characteristics, meteorological conditions, and time attributes. In addition to graph neural network methods, other clustering algorithms can also be used, such as K-means clustering, hierarchical clustering, and DBSCAN clustering. The appropriate clustering algorithm is selected based on the distribution characteristics of the multi-dimensional feature vectors. The selection of clustering algorithms should consider factors such as data dimensionality, sample size, and cluster shape to ensure the accuracy and stability of the clustering results. Then, based on the clustering results, multiple candidate seasonal intervals are defined. Time points belonging to the same cluster are arranged chronologically to identify consecutive time periods. These consecutive time periods are defined as candidate seasonal intervals, with time points within each interval exhibiting similar load variation characteristics, meteorological conditions, and temporal attributes. For clusters with time intervals, an interval merging strategy is employed. If the time interval between similar time points is less than a preset merging threshold, they are merged into consecutive candidate seasonal intervals. If the time interval is large, they remain independent candidate seasonal intervals to maintain the rationality of seasonal division.

[0055] Through the above clustering process, multiple candidate seasonal intervals are obtained, and the power load within each interval exhibits similar variation characteristics and patterns. These candidate seasonal intervals overcome the limitations of traditional fixed seasonal divisions, better reflect the actual variation patterns of power load, and provide a more reasonable and accurate basis for time segmentation for subsequent mathematical modeling and scenario generation.

[0056] Furthermore, based on the cross-seasonal time-series data, rapid mathematical modeling is performed for multiple candidate seasonal intervals to generate multiple candidate load scenarios, including:

[0057] S31. Homomorphic extraction is performed on the cross-seasonal time series data according to multiple candidate seasonal intervals to form multiple interval time series data;

[0058] S32. Traverse multiple time-series data intervals and establish a mathematical regression analysis model for time-scenario load indicators;

[0059] S33. Perform fitting performance verification on multiple mathematical regression analysis models, and output the mathematical regression analysis model that satisfies the basic performance constraints as the alternative load scenario.

[0060] In a preferred embodiment, firstly, homomorphic extraction is performed on the cross-seasonal time-series data based on multiple candidate seasonal intervals, forming multiple interval time-series data. Specifically, homomorphic extraction refers to the process of segmenting and extracting cross-seasonal time-series data according to the time boundaries of the candidate seasonal intervals while maintaining the integrity of the original data structure and features, ensuring that the extracted interval time-series data can completely retain the time series features and data correlations of the original data. For each candidate seasonal interval, a data subset of the corresponding time period is extracted from the complete cross-seasonal time-series data based on its start and end times. During the extraction process, the time order and sampling interval of the data are kept unchanged to ensure the temporal continuity of the interval time-series data, resulting in multiple interval time-series data. Each interval time-series data contains complete load information, meteorological information, and date type information within the corresponding candidate seasonal interval. Through homomorphic extraction, the original cross-seasonal time-series data is decomposed into multiple independent interval time-series data, each interval time-series data corresponding to the load change characteristics of a candidate seasonal interval, providing a data foundation for subsequent independent modeling.

[0061] Then, a mathematical regression analysis model of time-scenario load indicators is established by traversing multiple time-series data intervals. Specifically, for each time-series data interval, a regression analysis model describing the mathematical relationship between time variables and load scenarios is established. The mathematical regression analysis model of time-scenario load indicators is used to quantify the regularity and trend of load changes over time within a specific seasonal interval. The modeling process includes: first, determining the input and output variables of the model. Input variables include time variables and auxiliary variables. Time variables reflect the time dependence of load changes, while auxiliary variables include external factors affecting load changes, such as meteorological information and date type information. Output variables are load indicators, including load values, load change trends, and other load characteristic parameters. A suitable regression analysis method is selected to establish the mathematical model. Methods such as linear regression, multinomial regression, and nonlinear regression can be used. The most suitable regression model type is selected based on the distribution characteristics and change patterns of the time-series data intervals. For seasonal intervals with relatively stable load changes, a linear regression model can be used; for seasonal intervals with complex load changes, a multinomial or nonlinear regression model can be used. The parameters of the regression analysis model are determined using the least squares method, maximum likelihood estimation, or other parameter estimation methods. By using time-series data from different time periods, the model parameters are trained and optimized to accurately describe the load variation over time within that seasonal period. The mathematical regression analysis model established for each time-series data period reflects the load variation trend, fluctuation characteristics, and time dependence within the corresponding seasonal period, providing a mathematical basis for load scenario generation, thus yielding multiple mathematical regression analysis models.

[0062] Next, the fitting performance of multiple mathematical regression analysis models is validated, and the mathematical regression analysis models that meet the basic performance constraints are output as the candidate load scenarios. Specifically, the fitting performance validation is used to evaluate the accuracy and predictive ability of the established mathematical regression analysis models in describing actual load data. The validation process uses multiple performance validation metrics to comprehensively evaluate model quality. These performance validation metrics include: goodness-of-fit metrics, such as the coefficient of determination R², which measures the model's explanatory power for data variation; prediction accuracy metrics, such as mean squared error (MSE) and mean absolute error (MAE), which quantify the deviation between the model's predicted values ​​and actual values; and stability metrics, such as the confidence interval of model parameters and residual distribution characteristics, which assess the model's reliability and stability. For each mathematical regression analysis model, its various performance validation metrics are calculated on the corresponding time series data. The generalization ability and predictive stability of the model are tested using methods such as cross-validation or leave-one-out validation. Basic performance constraints are set, including the minimum goodness-of-fit threshold, the maximum prediction error threshold, and stability requirements. The mathematical regression analysis models that meet the basic performance constraints are output as candidate load scenarios. Basic performance constraints are determined based on the accuracy requirements of load forecasting and the application scenario, ensuring that the output candidate load scenarios have sufficient modeling accuracy and reliability. For models that do not meet the basic performance constraints, improvements can be made by adjusting the model type, adding variables, optimizing parameters, or by excluding them from the candidate load scenario set.

[0063] Through the rapid modeling process described above, multiple alternative load scenarios that meet the basic performance requirements are obtained. Each alternative load scenario corresponds to the load variation pattern of a candidate seasonal interval. These alternative load scenarios provide a preliminary model foundation for subsequent adaptive adjustment and optimization, realizing an effective transformation from seasonal division to load scenario modeling.

[0064] Furthermore, representative metrics and coverage metrics for candidate load scenarios are calculated, and adaptive interval adjustments are performed on the candidate load scenarios accordingly. Based on the adjustment results, multiple specialized seasonal intervals are output, including:

[0065] S41. Calculate the average load restoration accuracy over the interval as the representative metric.

[0066] S42. Calculate the coverage boundary constraints based on the preset bias coefficient and the representative metric.

[0067] S43. Perform statistical analysis based on the coverage boundary constraints, calculate the percentage of data that satisfies the coverage boundary constraints, and output the coverage metric.

[0068] S44. If either the representative metric or the coverage metric is less than the lower limit of the preset advanced performance constraint interval, then the interval is reduced for the corresponding candidate load scenario until the advanced performance constraint interval is met.

[0069] S45. If either the representative metric or the coverage metric is greater than the upper limit of the preset advanced performance constraint interval, then the interval is expanded for the corresponding candidate load scenario, and iterative modeling based on mathematical models is performed until the advanced performance constraint interval is met.

[0070] S46. Merge the output interval reduction result and the interval expansion result to obtain multiple specialized seasonal intervals.

[0071] In a preferred embodiment, firstly, the average load restoration accuracy over the interval is calculated as a representative metric. Specifically, the average load restoration accuracy is used to quantify the ability of candidate load scenarios to restore the original load data. For each candidate load scenario, the load data within the interval is predicted or reconstructed using its corresponding mathematical regression analysis model, resulting in a sequence of load prediction values ​​output by the model. The load values ​​predicted by the model are compared with the actual load values ​​at the corresponding time points, and the prediction error is calculated. The single-point prediction accuracy is quantified using relative error or absolute error, and then the average prediction error for all time points within the entire seasonal interval is calculated to obtain the average load restoration accuracy index. The smaller this average load restoration accuracy index, the stronger the ability of the candidate load scenario to restore the actual load data, and the better its representativeness.

[0072] Then, based on the preset bias coefficient and representative metrics, coverage boundary constraints are calculated. Specifically, coverage boundary constraints are used to determine the boundary of the load variation range that the candidate load scenario should cover. The bias coefficient is a pre-set adjustment parameter used to control the width of the coverage range, determined according to the fluctuation characteristics of the load data and the risk preference of the application scenario. Based on the model accuracy level reflected by the representative metrics, combined with the preset bias coefficient, the upper and lower boundaries of the load variation range are calculated. The specific calculation method includes: using the statistical characteristics of the actual load data as a benchmark, such as the load mean and standard deviation; adjusting the width of the confidence interval according to the representative metrics; the better the representative metrics, the narrower the confidence interval can be; further adjusting the confidence interval in combination with the bias coefficient to form the final coverage boundary constraints. The coverage boundary constraints define the range of load values ​​that the candidate load scenario should cover, providing a judgment standard for subsequent coverage evaluation. This coverage boundary constraint considers both the actual performance of the model and the preset risk control requirements.

[0073] Subsequently, statistical analysis is performed based on coverage boundary constraints to calculate the percentage of data that meets these constraints, outputting a coverage metric. Specifically, the statistical analysis assesses the extent to which actual load data falls within the coverage boundary constraints. For each seasonal interval corresponding to a candidate load scenario, the distribution of actual load data points within that interval is statistically analyzed. The ratio of the number of load data points falling within the coverage boundary constraints to the total number of load data points within that seasonal interval is calculated to obtain the coverage percentage, which serves as the coverage metric. This coverage percentage reflects the sufficiency of the candidate load scenario's coverage of the load variation range; a higher percentage indicates better coverage.

[0074] If either the representativeness metric or the coverage metric falls below the lower limit of a preset advanced performance constraint interval, interval reduction is performed on the corresponding candidate load scenario until the advanced performance constraint interval is met. Specifically, the advanced performance constraint interval is a pre-defined performance requirement range, including the minimum and maximum standards that the representativeness metric and coverage metric should meet. When either metric falls below the lower limit, it indicates that the modeling quality of the current candidate load scenario is insufficient, and performance needs to be improved through interval reduction. The interval reduction process includes: identifying the time periods causing performance deficiencies; locating performance weaknesses by analyzing the modeling accuracy and coverage quality of different time periods within the seasonal interval; employing an iterative reduction strategy to gradually narrow the time range of the seasonal interval, prioritizing the elimination of time periods with lower modeling accuracy or poorer data quality; recalculating the representativeness metric and coverage metric after each reduction to verify the performance improvement effect; and continuously performing interval reduction until both metrics meet the requirements of the advanced performance constraint interval. Interval reduction improves data consistency and modeling accuracy by reducing the time span of the seasonal interval, thus achieving the goal of performance optimization.

[0075] If both the representativeness metric and the coverage metric exceed the preset upper limit of the advanced performance constraint interval, the interval is expanded for the corresponding candidate load scenario, and rapid modeling based on mathematical models is iteratively performed until the advanced performance constraint interval is met. Specifically, when both metrics exceed the upper limit of the interval, it indicates that the current candidate load scenario has high modeling quality and has the potential to expand its coverage. Through interval expansion, the temporal coverage of the load scenario can be increased while maintaining modeling quality. The interval expansion process includes: identifying the expandable time direction, analyzing the data characteristics and modeling feasibility of adjacent time periods before and after the seasonal interval; adopting a stepwise expansion strategy to expand the time boundary of the seasonal interval forward or backward to increase the temporal coverage; re-performing rapid modeling based on mathematical models on the expanded new seasonal interval to establish a new mathematical regression analysis model; calculating the expanded representativeness metric and coverage metric to verify the expansion effect; if the metrics are still within a reasonable range, interval expansion continues; if the metrics exceed the upper limit, expansion stops and the optimal interval boundary is determined. Interval expansion improves the coverage and representativeness of the load scenario by increasing the time span of the seasonal interval, thereby optimizing modeling efficiency.

[0076] Next, the results of interval reduction and interval expansion are merged to obtain multiple specialized seasonal intervals. Specifically, all seasonal intervals that have undergone interval reduction and expansion are integrated to form the final multiple specialized seasonal intervals. These specialized seasonal intervals are performance-optimized seasonal intervals with higher modeling accuracy and better load representativeness. The merging process includes: collecting all adaptively adjusted seasonal intervals, including both reduced and expanded intervals; examining the temporal relationships between adjusted intervals to identify potential overlaps or gaps; for adjacent intervals with slight overlap, using boundary optimization methods to determine the optimal boundary point; for adjacent intervals with small gaps, evaluating the feasibility of merging, and merging them into continuous intervals if performance requirements are still met after merging; and outputting the final multiple specialized seasonal intervals, each satisfying the advanced performance constraints.

[0077] Through adaptive interval adjustment and merging optimization, the obtained specialized seasonal intervals have good load representativeness and modeling quality, providing an optimized time segmentation basis for subsequent iterative modeling and load scenario generation, and realizing the transformation and improvement from preliminary seasonal division to accurate seasonal intervals.

[0078] Furthermore, adaptive interval adjustment is performed on the candidate load scenarios, and multiple specialized seasonal intervals are output based on the adjustment results, including:

[0079] S471. Construct a similarity matrix among the multiple candidate load scenarios;

[0080] S472. Threshold discrimination based on the similarity matrix:

[0081] S473. If the dynamic time warping distance between any two candidate load scenarios is less than a preset fusion threshold, then the two candidate load scenarios are subjected to linear fusion driven by weighted coefficients to generate a fused scenario.

[0082] S474. Based on the fusion scenario, backtrack the mapping according to the time position, update the seasonal interval boundaries, and output the updated multiple seasonal intervals as the specialized seasonal intervals.

[0083] In a preferred embodiment, firstly, a similarity matrix is ​​constructed among multiple candidate load scenarios. Specifically, the similarity matrix is ​​used to quantify the similarity between any two candidate load scenarios, providing a numerical basis for subsequent scenario fusion decisions. For a scenario containing multiple candidate load scenarios, an n×n-dimensional similarity matrix is ​​constructed, where each element represents a similarity metric between the corresponding two candidate load scenarios. The similarity calculation process includes: for any two candidate load scenarios, extracting their corresponding load change feature vectors, including features such as load numerical sequences, load change trends, and load fluctuation patterns; using a dynamic time warping distance algorithm to calculate the similarity metric between the two load scenarios. The dynamic time warping distance algorithm can handle the problem of inconsistent time series lengths and identify the optimal alignment between time series; converting the dynamic time warping distance into a similarity value, where a smaller distance indicates higher similarity; and performing similarity calculations pairwise for all candidate load scenarios to form a complete similarity matrix. The similarity matrix exhibits symmetry, meaning that the element in the i-th row and j-th column has the same value as the element in the j-th row and i-th column. The diagonal elements represent the similarity between each scenario and itself, typically set to the maximum similarity value. This similarity matrix comprehensively reflects the internal similarity relationships within the set of candidate load scenarios.

[0084] Then, threshold discrimination is performed based on the similarity matrix. Specifically, threshold discrimination is used to identify candidate load scene pairs in the similarity matrix whose similarity meets the fusion requirements. By setting a preset fusion threshold, scene pairs with similarity higher than the threshold are selected as candidate fusion objects. The threshold discrimination process includes: setting a fusion threshold standard, which is determined based on the similarity distribution characteristics of the load scenes and the fusion quality requirements, usually selecting the high quantile of the similarity distribution as the threshold; traversing the upper triangular part of the similarity matrix to avoid duplicate discrimination; identifying scene pairs with matrix elements whose values ​​are higher than the fusion threshold and marking them as high similarity scene pairs; further verifying the identified high similarity scene pairs to ensure that their corresponding seasonal intervals have reasonable temporal proximity or overlap; and outputting a list of candidate load scene pairs that meet the fusion threshold, providing target objects for subsequent fusion processing. Threshold discrimination ensures that only candidate load scenes with truly high similarity are included in the fusion process, avoiding the negative impact of unreasonable fusion on scene quality.

[0085] If the dynamic time warping distance between any two candidate load scenarios is less than a preset fusion threshold, then a weighted coefficient-driven linear fusion is performed on the two candidate load scenarios to generate a fused scenario. Specifically, a dynamic time warping distance less than the fusion threshold indicates that the two candidate load scenarios have highly similar load change patterns and are suitable for fusion processing. Weighted coefficient-driven linear fusion achieves effective integration of scenario information by assigning appropriate weight coefficients to different scenarios. The linear fusion process includes: determining the weight coefficients of the two candidate load scenarios participating in the fusion, where the weight coefficients reflect the degree of contribution of each scenario to the fusion result; preferably, an optimization method based on Kullback-Leibler divergence is used to determine the weight coefficients, and the optimal weight allocation is obtained by minimizing the information divergence between the fused scenario and the original scenario; Kullback-Leibler divergence can quantify the degree of difference between two probability distributions, and by optimizing this divergence index, it is ensured that the fused scenario can retain the information features of the original scenario to the greatest extent; based on the optimized weight coefficients, the two candidate load scenarios are linearly weighted and combined to generate the fused scenario. The load characteristics of the fused scene are a weighted average of the load characteristics of the two original scenes. The fused scene retains the main characteristics of the original scene and eliminates the subtle differences between the scenes through the fusion process, thereby improving the representativeness and stability of the scene.

[0086] Next, based on the fused scenario, a backtracking mapping is performed according to time location, and the seasonal interval boundaries are updated, outputting multiple updated seasonal intervals as specialized seasonal intervals. Specifically, the backtracking mapping is used to reflect the fused scenario information into the corresponding seasonal intervals, and to adjust and optimize the boundary settings of the seasonal intervals according to the fusion results, realizing the transformation from scenario fusion to interval optimization. The backtracking mapping process includes: identifying the seasonal intervals corresponding to the two original candidate load scenarios participating in the fusion, and analyzing their time range and boundary characteristics; determining the time range to be covered after fusion based on the feature distribution of the fused scenario, which usually covers the union of the two original seasonal intervals or the optimized continuous intervals; mapping the load characteristics of the fused scenario back to the corresponding time period according to time location, establishing the correspondence between the fused scenario and the time location; and updating the boundary settings of the seasonal intervals based on the mapping results, including adjusting the start time and end time of the intervals, and optimizing the feature distribution within the intervals. The seasonal interval boundary update strategy includes: merging temporally adjacent or overlapping original seasonal intervals into a unified specialized seasonal interval; evaluating the load characteristics of the interval portion of the original seasonal interval to determine whether to include the interval in the fused specialized seasonal interval; and optimizing the boundary position of the specialized seasonal interval to ensure maximum consistency and continuity of load characteristics within the interval. The updated seasonal intervals are then output as specialized seasonal intervals. These specialized seasonal intervals are the final seasonal division results after similarity analysis, scene fusion, and boundary optimization, exhibiting higher load representativeness, better internal consistency, and a more reasonable time division.

[0087] By using adaptive interval adjustment based on similarity analysis, not only was the redundant scenarios effectively integrated, but the quality and accuracy of seasonal interval division were also improved through intelligent fusion technology, providing a more optimized time segmentation basis for subsequent iterative modeling and load scenario generation.

[0088] Furthermore, by combining multiple specialized seasonal intervals with the cross-seasonal time-series data for iterative modeling, a specialized power load scenario is formed, including:

[0089] S51. Homomorphic extraction is performed on the cross-seasonal time series data according to multiple specialized seasonal intervals to form multiple specialized interval time series data;

[0090] S52. Traverse multiple specialized time-series data, establish a regression analysis model of time-scenario load indicators, and verify the fitting performance of multiple regression analysis models.

[0091] S53. Output multiple regression analysis models that satisfy the target performance constraints for the specialized power load scenario.

[0092] In a preferred embodiment, firstly, homomorphic extraction is performed on cross-seasonal time-series data based on multiple specialized seasonal intervals to form multiple specialized interval time-series data. Specifically, based on the obtained multiple specialized seasonal intervals, precise homomorphic extraction is performed on the cross-seasonal time-series data. Compared with the initial interval time-series data, the specialized interval time-series data has more accurate time boundaries, higher data quality, and better internal consistency. The homomorphic extraction process includes: extracting corresponding time period data from the complete cross-seasonal time-series data according to the optimization boundaries of each specialized seasonal interval, including precise start and end times; strictly maintaining the original structural characteristics of the data during the extraction process to ensure the continuity of the time series, the consistency of sampling frequency, and the integrity of data fields; performing quality verification on the extracted specialized interval time-series data, including data integrity checks, time continuity verification, and numerical rationality checks; and performing refined preprocessing on the extracted data based on the optimization characteristics of the specialized seasonal intervals, including outlier identification and handling, missing value imputation, and data smoothing. Each specialized time-series interval contains complete load, meteorological, and date type information for that specialized seasonal interval, resulting in a significant improvement in data quality compared to the initial extraction stage. This specialized time-series data provides a high-quality data foundation for subsequent high-precision modeling.

[0093] Then, by traversing multiple specialized time-series data intervals, regression analysis models of time-scenario load indicators were established, and the fitting performance of multiple regression analysis models was verified. Specifically, compared with the preliminary modeling stage, the regression analysis models established based on specialized time-series data intervals adopted more refined modeling strategies and higher performance standards to achieve better modeling results. The regression analysis model establishment process includes: for each specialized time-series data interval, analyzing the complexity of its load changes and data distribution characteristics, and selecting the most suitable regression analysis method; selecting different types of regression models such as linear regression, multinomial regression, and nonparametric regression according to data characteristics, or using an integrated modeling method combining multiple regression techniques; optimizing the configuration of input variables for the model, making full use of meteorological information and date type information as auxiliary variables in addition to time variables to improve the explanatory power of the model; and using iterative optimization algorithms to finely adjust the model parameters, including gradient descent, genetic algorithms, particle swarm optimization, and other optimization methods to ensure that the model parameters reach the optimal configuration. The fitting performance verification adopts more stringent verification standards and a more comprehensive performance index system. The validation process includes: calculating model fitting accuracy metrics such as the coefficient of determination (R²), adjusted coefficient of determination (R²), and mean squared error (MSE) to evaluate the model's fit to the training data; performing cross-validation or time series split validation to test the model's generalization ability and predictive stability; analyzing the distribution characteristics of the model residuals, including their normality, independence, and homoscedasticity, to verify the rationality of the model assumptions; and evaluating the model's robustness by verifying its stability under data fluctuations through perturbation testing and sensitivity analysis. Through iterative optimization and rigorous validation, the established regression analysis model is ensured to have excellent fitting performance and reliable predictive ability.

[0094] Subsequently, multiple regression analysis models that satisfy the target performance constraints are output as the specialized power load scenario. Specifically, the target performance constraints are pre-set high-standard performance requirements with stricter indicator requirements and more comprehensive evaluation dimensions to ensure that the output specialized power load scenario has excellent modeling quality. The target performance constraints include: accuracy constraints, requiring the model's prediction accuracy to reach a preset high-accuracy standard, such as R² greater than a set threshold and prediction error less than a limited range; stability constraints, requiring the model to maintain stable performance under different data conditions, including parameter stability and prediction consistency; robustness constraints, requiring the model to have good resistance to data disturbances and external interference; interpretability constraints, requiring the model to have clear physical meaning and reasonable parameter interpretation; and computational efficiency constraints, requiring the model to have reasonable computational complexity while ensuring accuracy. The performance constraint verification process includes: checking each regression analysis model item by item to see if it meets all the requirements of the target performance constraints; using a comprehensive scoring mechanism to weight and integrate multiple performance indicators to calculate the model's comprehensive performance score; setting a pass threshold for the target performance constraints, only models whose comprehensive performance score reaches this threshold can pass the verification; for models that fail the verification, analyzing the specific reasons for their performance deficiencies, and adopting corresponding improvement strategies for optimization and adjustment. The output regression analysis models that meet the target performance constraints are specialized power load scenarios. Each specialized power load scenario corresponds to a high-precision load variation pattern for a specialized seasonal interval, possessing excellent modeling quality and reliable predictive capabilities. These specialized power load scenarios provide a high-quality model foundation for subsequent scenario fusion and target load scenario generation.

[0095] By combining iterative modeling with specialized seasonal intervals, a quality leap from preliminary to accurate load scenario modeling was achieved, establishing a set of specialized power load scenarios with high representativeness, high accuracy, and high stability, laying the foundation for the generation of the final target power load scenario.

[0096] Furthermore, multiple specialized power load scenarios are merged and output as a target power load scenario, including:

[0097] S54. Traverse multiple specialized power load scenarios and calculate the smooth transition interval based on a preset smoothing operator;

[0098] S55. Based on the preset smooth transition model, define the adaptive fusion calculation weight of the smooth transition interval, and perform adaptive fusion of the adjacent specialized power load scenarios under the smooth transition interval accordingly.

[0099] S56. Iteratively and adaptively fuse multiple specialized power load scenarios to obtain the target power load scenario.

[0100] In a preferred embodiment, firstly, multiple specialized power load scenarios are traversed, and a smooth transition interval is calculated based on a preset smoothing operator. Specifically, the smoothing operator is a mathematical tool used to identify and define transition regions between scenarios. By analyzing the load characteristic differences between adjacent specialized power load scenarios at boundary positions, the time interval range requiring smoothing is determined. The smooth transition interval calculation process includes: arranging all specialized power load scenarios in chronological order and identifying the time boundary positions between adjacent scenarios; for each boundary position, analyzing the load values, trends, and fluctuation characteristics of two adjacent specialized power load scenarios near the boundary; calculating the degree of load difference at the boundary position using a preset smoothing operator, which can be a gradient operator, a difference operator, or other numerical analysis operators; determining the time span of the smooth transition interval based on the magnitude and severity of the load difference, with larger differences requiring longer transition intervals for smooth connection; and recording and marking all identified smooth transition intervals to provide precise interval definitions for subsequent fusion processing. The smooth transition interval typically covers a certain time range before and after the boundary of adjacent specialized power load scenarios. The load data within this interval will be recalculated and optimized through smooth fusion technology to eliminate the discontinuity of scenario transition.

[0101] Then, combining the preset smooth transition model, adaptive fusion calculation weights for the smooth transition intervals are defined, and adaptive fusion of adjacent specialized power load scenarios within the smooth transition intervals is performed accordingly. Specifically, the smooth transition model provides a mathematical framework and calculation method for scenario fusion, achieving smooth connection between adjacent scenarios by establishing a suitable transition function. Preferably, a Bézier curve is used as the smooth transition model, as it possesses good smoothing properties and flexible shape control capabilities. The adaptive fusion calculation weight definition process includes: within each smooth transition interval, determining the fusion weight allocation for two adjacent specialized power load scenarios based on their time position; defining the weight variation law over time using a Bézier curve or other smoothing function: at the beginning of the transition interval, the weight of the preceding scenario is higher, and the weight of the following scenario is lower; as time progresses, the weight of the preceding scenario gradually decreases, and the weight of the following scenario gradually increases; at the end of the transition interval, the weight allocation completely favors the following scenario; the specific shape of the weight change curve is determined by the control point parameters of the Bézier curve and can be adjusted according to scenario characteristics and smoothing requirements. The adaptive fusion process includes: for each time point within the smooth transition interval, calculating the weighted average of the load values ​​of two adjacent specialized power load scenarios according to the corresponding fusion weight; the fused load values ​​retain the load characteristics of the original scenario while eliminating abrupt changes in scenario boundaries through smooth transition; the fusion results are quality checked to ensure that the load changes within the transition interval have good continuity and rationality; the fused load data replaces the original data within the smooth transition interval to achieve smooth scenario connection.

[0102] Subsequently, iterative adaptive fusion is performed on multiple specialized power load scenarios to obtain the target power load scenario. Specifically, iterative adaptive fusion refers to the smooth fusion process performed sequentially on all adjacent specialized power load scenarios. Through multiple rounds of iterative optimization, the target power load scenario is finally generated. The iterative adaptive fusion process includes: sorting all specialized power load scenarios in chronological order to establish a scenario fusion processing queue; starting from the earliest scenario, performing smooth transition fusion on adjacent scenarios sequentially, processing one pair of adjacent scenarios at a time; evaluating the quality of the fusion result after each fusion, including the smoothness of the transition interval, the continuity of the overall scenario, and the rationality of the load values; adjusting the weight allocation strategy and transition model parameters in subsequent fusion processes based on the quality evaluation results to achieve adaptive optimization; repeating the fusion process on adjacent scenarios until all specialized power load scenarios are fused to obtain the generated target power load scenario.

[0103] Through iterative adaptive fusion, multiple independent specialized power load scenarios are integrated into the target power load scenario, realizing the generation of power load scenarios that integrate cross-seasonal data, and providing high-quality load scenario support for power system load forecasting, scheduling optimization and operation management.

[0104] Furthermore, embodiments of this application also include:

[0105] S571. Obtain the residuals of multiple specialized power load scenarios to form a scenario residual distribution set;

[0106] S572. Based on the scenario residual distribution set and the preset confidence constraints, calculate the scenario fluctuation domain and scenario fluctuation probability function of multiple specialized power load scenarios.

[0107] S573. Associate and store multiple specialized power load scenarios, scenario fluctuation domains and scenario fluctuation probability functions, and output the target power load scenario.

[0108] In a preferred embodiment, firstly, residuals for multiple specialized power load scenarios are acquired to form a scenario residual distribution set. Specifically, residual refers to the difference between the model prediction value and the actual load value at the corresponding time point for a specialized power load scenario. Residual analysis can reveal the prediction bias characteristics and uncertainty distribution patterns of the model. The residual acquisition process includes: for each specialized power load scenario, collecting the model prediction load values ​​for all time points within its corresponding seasonal interval; acquiring the actual load observation values ​​at the same time point and establishing the correspondence between the prediction value and the observation value; calculating the residual value for each time point, where the residual value equals the actual load value minus the model prediction value; collecting all residual values ​​within the corresponding interval of the specialized power load scenario to form a residual sequence for that scenario; performing statistical analysis on the residual sequence, calculating the mean, variance, skewness, kurtosis, and other statistical characteristics of the residuals, and analyzing the distribution pattern of the residuals. The process of constructing the scenario residual distribution set includes: integrating residual data from all specialized power load scenarios and classifying them according to dimensions such as scenario type, time period, and load level; analyzing the differences in residual distribution characteristics under different scenario types to identify the regularity and specificity of the residual distribution; using probability distribution fitting methods, such as normal distribution, t-distribution, and skewed distribution, to fit the residual data to determine the distribution type that best matches the data characteristics; and establishing the data structure of the scenario residual distribution set, including information such as residual statistical characteristics, distribution parameters, and fitting quality indicators for each scenario. The scenario residual distribution set provides a data foundation and statistical basis for subsequent fluctuation domain calculations and probability function establishment.

[0109] Then, based on the scenario residual distribution set and preset confidence constraints, the scenario fluctuation domain and scenario fluctuation probability function for multiple specialized power load scenarios are calculated. Specifically, the confidence constraints are pre-set confidence level requirements used to determine the coverage of the scenario fluctuation domain and the accuracy requirements of the probability function. By combining residual distribution characteristics and confidence constraints, the uncertainty range and probability characteristics of the load scenario are quantified. The scenario fluctuation domain calculation process includes: determining the residual distribution parameters of each specialized power load scenario based on statistical information in the scenario residual distribution set; calculating the upper and lower boundaries of the corresponding confidence intervals according to preset confidence constraints, such as 95% confidence level, 99% confidence level, etc.; combining the confidence interval boundaries with the model prediction values ​​of the scenario to calculate the upper and lower bounds of the scenario fluctuation domain; the upper bound of the scenario fluctuation domain is equal to the model prediction value plus the residual value corresponding to the upper bound of the confidence interval, and the lower bound of the scenario fluctuation domain is equal to the model prediction value plus the residual value corresponding to the lower bound of the confidence interval; calculating the scenario fluctuation domain for different time points and different load levels to form a complete fluctuation domain description. The process of establishing the scenario fluctuation probability function includes: constructing a probability distribution function for the load value based on the probability density function of the residual distribution and the model prediction value of the scenario; the scenario fluctuation probability function describes the probability that the load value falls within each interval at a given time point; calculating the cumulative probability function using numerical integration or analytical methods to provide probability information that the load value is less than a certain threshold; establishing a parameterized representation of the probability function to facilitate subsequent probability calculations and risk assessment applications; and verifying the rationality and accuracy of the probability function to ensure that the probability distribution satisfies basic probability axioms and actual physical constraints. The scenario fluctuation domain provides a description of the uncertainty range of load changes, and the scenario fluctuation probability function provides probability distribution information for the load value; together, they constitute a quantitative system for the uncertainty of the load scenario.

[0110] Subsequently, multiple specialized power load scenarios, scenario fluctuation domains, and scenario fluctuation probability functions are associated and stored, outputting the target power load scenario. Specifically, associated storage refers to the organic integration of deterministic and uncertain information of specialized power load scenarios to form a target power load scenario data structure containing complete probabilistic features. This data structure not only includes the expected load value prediction but also the prediction confidence interval and probability distribution information. The associated storage process includes: establishing a unified data storage structure, including fields such as time index, expected load value, fluctuation domain boundary, and probability function parameters; for each time point, associating and storing the corresponding specialized power load scenario prediction value, the upper and lower bounds of the scenario fluctuation domain, and the distribution parameters of the scenario fluctuation probability function; establishing logical relationships between data to ensure the consistency and completeness of deterministic prediction and uncertainty description; optimizing the data storage format and access interface to facilitate rapid querying and calculation in subsequent applications; and establishing a data quality control mechanism to verify the integrity, consistency, and accuracy of the stored data. The output target power load scenario has the following characteristics: deterministic prediction capability, providing load expectation prediction based on specialized power load scenarios; uncertainty quantification capability, describing the uncertainty range of the prediction through the scenario fluctuation domain; probabilistic assessment capability, providing probability distribution information of load values ​​through the scenario fluctuation probability function; risk assessment support, providing probabilistic load scenario information for risk analysis and decision-making in power systems; and application adaptability, meeting the needs of different application scenarios for deterministic prediction and probabilistic analysis.

[0111] By associating deterministic scenario information and uncertainty quantification information, the final generated target power load scenario not only has high-precision load forecasting capabilities, but also provides complete uncertainty quantification and probability assessment functions, providing more comprehensive and reliable scenario support for power system applications such as load forecasting, risk management, and dispatch optimization.

[0112] Example 2, as Figure 2 As shown, based on the same inventive concept as the power load scenario generation method that integrates cross-seasonal data provided in Embodiment 1, this embodiment of the invention also provides a power load scenario generation system that integrates cross-seasonal data, including:

[0113] The data acquisition module 11 is used to determine the acquisition time window of the target area and acquire cross-seasonal time series data of the target area according to the acquisition time window.

[0114] Season segmentation module 12 is used to parse the cross-seasonal time series data to perform non-standard seasonal segmentation and obtain multiple candidate seasonal intervals;

[0115] The quick modeling module 13 is used to perform quick modeling based on mathematical models for multiple candidate seasonal intervals according to the cross-seasonal time series data, and generate multiple candidate load scenarios.

[0116] The adaptive adjustment module 14 is used to calculate the representative metrics and coverage metrics of the candidate load scenarios, and to perform adaptive interval adjustment for the candidate load scenarios accordingly, and output multiple specialized seasonal intervals based on the adjustment results.

[0117] The iterative fusion module 15 is used to combine multiple specialized seasonal intervals with the cross-seasonal time series data for iterative modeling to form specialized power load scenarios, and to merge and output multiple specialized power load scenarios as a target power load scenario.

[0118] Furthermore, the execution steps of the data acquisition module 11 include:

[0119] Based on confidence analysis oriented towards time-varying similarity, stability detection is performed on the load change trend of the target area to identify time periods in which the load fluctuation amplitude in continuous segments is lower than the preset stability threshold.

[0120] The intervals in the identification results with confidence levels higher than a preset threshold are defined as the collection time window, and the cross-seasonal time series data are collected using the collection time window as a time range constraint.

[0121] Furthermore, the execution steps of the season division module 12 include:

[0122] The cross-seasonal time series data is analyzed to extract load-related features, meteorological features, and time cycle features, and a multi-dimensional feature vector is constructed.

[0123] Based on the multi-dimensional feature vectors, the cross-seasonal time series data are clustered and divided, and multiple candidate seasonal intervals are defined based on the clustering results.

[0124] Furthermore, the execution steps of the quick modeling module 13 include:

[0125] Homomorphic extraction is performed on the cross-seasonal time series data based on multiple candidate seasonal intervals to form multiple interval time series data;

[0126] By traversing multiple time-series data intervals, a mathematical regression analysis model of time-scenario load indicators is established.

[0127] The fitting performance of multiple mathematical regression analysis models is verified, and the mathematical regression analysis model that satisfies the basic performance constraints is output as the alternative load scenario.

[0128] Furthermore, the execution steps of the adaptive adjustment module 14 include:

[0129] Calculate the average load restoration accuracy over the interval, as the representative metric.

[0130] Based on the preset bias coefficient and the representative metric, the coverage boundary constraint is calculated.

[0131] Statistical analysis is performed based on the coverage boundary constraints to calculate the percentage of data that satisfies the coverage boundary constraints, and the output is the coverage metric.

[0132] If either the representative metric or the coverage metric is less than the lower limit of the preset advanced performance constraint interval, then the interval will be reduced for the corresponding alternative load scenario until the advanced performance constraint interval is met.

[0133] If either the representative metric or the coverage metric is greater than the preset upper limit of the advanced performance constraint interval, then the interval is expanded for the corresponding candidate load scenario, and rapid modeling based on mathematical models is iterated until the advanced performance constraint interval is met.

[0134] The results of interval reduction and interval expansion are merged to obtain multiple specialized seasonal intervals.

[0135] Furthermore, the execution steps of the adaptive adjustment module 14 also include:

[0136] Construct a similarity matrix among the multiple candidate load scenarios;

[0137] Threshold discrimination is performed based on the similarity matrix:

[0138] If the dynamic time warp distance between any two candidate load scenarios is less than a preset fusion threshold, then the two candidate load scenarios are subjected to linear fusion driven by weighted coefficients to generate a fused scenario.

[0139] Based on the fusion scenario, a backtracking mapping is performed according to the time location, and the seasonal interval boundaries are updated. The updated seasonal intervals are then output as the specialized seasonal intervals.

[0140] Furthermore, the execution steps of the iterative fusion module 15 also include:

[0141] Homomorphic extraction is performed on the cross-seasonal time series data based on multiple specialized seasonal intervals to form multiple specialized interval time series data;

[0142] By traversing multiple specialized time-series data intervals, a regression analysis model of time-scenario load indicators is established, and the fitting performance of multiple regression analysis models is verified.

[0143] The output consists of multiple regression analysis models that satisfy the target performance constraints for the specialized power load scenario.

[0144] Furthermore, the execution steps of the iterative fusion module 15 also include:

[0145] The process iterates through multiple specialized power load scenarios and calculates the smooth transition interval based on a preset smoothing operator.

[0146] Based on the preset smooth transition model, the adaptive fusion calculation weight of the smooth transition interval is defined, and the adaptive fusion of the adjacent specialized power load scenarios under the smooth transition interval is performed accordingly.

[0147] The target power load scenario is obtained by iteratively and adaptively fusing multiple specialized power load scenarios.

[0148] Furthermore, embodiments of this application also include a scene optimization module, the execution steps of which include:

[0149] Obtain the residuals of multiple specialized power load scenarios to form a scenario residual distribution set;

[0150] Based on the scenario residual distribution set and the preset confidence constraints, the scenario fluctuation domain and scenario fluctuation probability function of multiple specialized power load scenarios are calculated accordingly.

[0151] Multiple specialized power load scenarios, scenario fluctuation domains, and scenario fluctuation probability functions are associated and stored, and the output is the target power load scenario.

[0152] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0153] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0154] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0155] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0156] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0157] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0158] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for generating power load scenarios by integrating cross-seasonal data, characterized in that, include: Determine the collection time window for the target area, and collect cross-seasonal time series data for the target area according to the collection time window; The cross-seasonal time series data is parsed to perform non-standard seasonal division, and multiple candidate seasonal intervals are obtained. Based on the cross-seasonal time series data, a rapid modeling based on mathematical models is performed for multiple candidate seasonal intervals to generate multiple candidate load scenarios; The average load restoration accuracy of the candidate load scenarios is calculated as a representative metric. Based on the preset bias coefficient and the representative metric, the coverage boundary constraint is calculated. The percentage of data that meets the coverage boundary constraint is calculated and output as a coverage metric. If either the representative metric or the coverage metric does not meet the preset advanced performance constraint interval, adaptive interval adjustment is performed for the corresponding candidate load scenario, and rapid modeling based on mathematical models is iterated until the advanced performance constraint interval is met. Multiple specialized seasonal intervals are output based on the adaptive interval adjustment results. The adaptive interval adjustment includes interval reduction and interval expansion. Homomorphic extraction is performed on the cross-seasonal time-series data based on multiple specialized seasonal intervals to form multiple specialized interval time-series data. The time-scenario load index regression analysis model is established by traversing the multiple specialized interval time-series data. The fitting performance of the multiple regression analysis models is verified, and the multiple regression analysis models that meet the target performance constraints are output as specialized power load scenarios. The multiple specialized power load scenarios are then merged and output as the target power load scenario.

2. The method for generating power load scenarios by fusing cross-seasonal data as described in claim 1, characterized in that, Determine the collection time window for the target area, and collect cross-seasonal time-series data for the target area corresponding to the collection time window, including: Based on confidence analysis oriented towards time-varying similarity, stability detection is performed on the load change trend of the target area to identify time periods in which the load fluctuation amplitude in continuous segments is lower than the preset stability threshold. The intervals in the identification results with confidence levels higher than a preset threshold are defined as the collection time window, and the cross-seasonal time series data are collected using the collection time window as a time range constraint.

3. The method for generating power load scenarios by fusing cross-seasonal data as described in claim 2, characterized in that, The cross-seasonal time-series data is parsed to perform non-standard seasonal segmentation, obtaining multiple candidate seasonal intervals, including: The cross-seasonal time series data is analyzed to extract load-related features, meteorological features, and time cycle features, and a multi-dimensional feature vector is constructed. Based on the multi-dimensional feature vectors, the cross-seasonal time series data are clustered and divided, and multiple candidate seasonal intervals are defined based on the clustering results.

4. The method for generating power load scenarios by fusing cross-seasonal data as described in claim 3, characterized in that, Based on the cross-seasonal time-series data, rapid mathematical modeling is performed for multiple candidate seasonal intervals to generate multiple candidate load scenarios, including: Homomorphic extraction is performed on the cross-seasonal time series data based on multiple candidate seasonal intervals to form multiple interval time series data; By traversing multiple time-series data intervals, a mathematical regression analysis model of time-scenario load indicators is established. The fitting performance of multiple mathematical regression analysis models is verified, and the mathematical regression analysis model that satisfies the basic performance constraints is output as the alternative load scenario.

5. The method for generating power load scenarios by fusing cross-seasonal data as described in claim 4, characterized in that, The alternative load scenarios are adaptively adjusted, and multiple specialized seasonal intervals are output based on the adjustment results. This also includes: Construct a similarity matrix among the multiple candidate load scenarios; Threshold discrimination is performed based on the similarity matrix: If the dynamic time warp distance between any two candidate load scenarios is less than a preset fusion threshold, then the two candidate load scenarios are subjected to linear fusion driven by weighted coefficients to generate a fused scenario. Based on the fusion scenario, a backtracking mapping is performed according to the time location, and the seasonal interval boundaries are updated. The updated seasonal intervals are then output as the specialized seasonal intervals.

6. The method for generating power load scenarios by fusing cross-seasonal data as described in claim 1, characterized in that, The multiple specialized power load scenarios are merged and output into a target power load scenario, including: The process iterates through multiple specialized power load scenarios and calculates the smooth transition interval based on a preset smoothing operator. Based on the preset smooth transition model, the adaptive fusion calculation weight of the smooth transition interval is defined, and the adaptive fusion of the adjacent specialized power load scenarios under the smooth transition interval is performed accordingly. The target power load scenario is obtained by iteratively and adaptively fusing multiple specialized power load scenarios.

7. The method for generating power load scenarios by fusing cross-seasonal data as described in claim 1, characterized in that, Also includes: Obtain the residuals of multiple specialized power load scenarios to form a scenario residual distribution set; Based on the scenario residual distribution set and the preset confidence constraints, the scenario fluctuation domain and scenario fluctuation probability function of multiple specialized power load scenarios are calculated accordingly. Multiple specialized power load scenarios, scenario fluctuation domains, and scenario fluctuation probability functions are associated and stored, and the output is the target power load scenario.

8. A power load scenario generation system integrating cross-seasonal data, characterized in that, The system is used to implement the method for generating power load scenarios by fusing cross-seasonal data as described in any one of claims 1 to 7, the system comprising: The data acquisition module is used to determine the acquisition time window of the target area and acquire cross-seasonal time series data of the target area according to the acquisition time window. The seasonal segmentation module is used to parse the cross-seasonal time series data to perform non-standard seasonal segmentation and obtain multiple candidate seasonal intervals. The quick modeling module is used to perform quick modeling based on mathematical models for multiple candidate seasonal intervals based on the cross-seasonal time series data, and generate multiple candidate load scenarios. The adaptive adjustment module is used to calculate the representative metrics and coverage metrics of the candidate load scenarios, and to perform adaptive interval adjustments for the candidate load scenarios accordingly, and output multiple specialized seasonal intervals based on the adjustment results. The iterative fusion module is used to combine multiple specialized seasonal intervals with the cross-seasonal time series data for iterative modeling to form specialized power load scenarios, and to merge and output multiple specialized power load scenarios as the target power load scenario.

Images