Day load characteristic set extraction and application method and system for energy storage planning operation

Abstract

The application relates to the field of power systems, and particularly discloses a daily load characteristic set extraction and application method and system for energy storage planning and operation, which comprises the following steps: obtaining minute-level daily load data, monthly calculating dynamic time warping distances between daily load curves, performing K-medoids clustering to extract typical daily load characteristic curves and curve clusters, dynamically time warping the curve clusters, point-by-point calculating upper and lower envelope lines and preserving characteristic smoothing, generating general mode representative curves and boundary mode representative curves, extracting characteristic values to form a daily load characteristic set, using a general mode subset for energy storage configuration, using a boundary mode subset for production simulation and safety and stability checking, reversely correcting clustering parameters and a boundary coefficient, and iteratively optimizing until the checking is passed; and the system comprises four functionally matched modules. The application realizes deep coupling between daily load characteristic extraction and energy storage planning and operation requirements, and improves the scientificity of an energy storage configuration scheme and the reliability of power grid checking.

Description

Technical Field

[0001] This invention relates to the field of power system technology, specifically to a method and system for extracting and applying daily load characteristic sets for energy storage planning and operation. Background Technology

[0002] Building a new power system with new energy sources as the mainstay requires large-scale development of energy storage to cope with the randomness, intermittency, and volatility of new energy output. Accurate daily load characteristics are the basic data required for independent energy storage planning and grid operation verification, and directly determine the rationality of energy storage technology type selection, power scale, and time-of-use configuration.

[0003] Existing methods for extracting daily load characteristics of power systems can be mainly divided into four categories: First, traditional statistical methods, including typical daily average methods, peak-valley feature extraction methods, and segmented feature extraction methods. These methods are simple and easy to implement, but they excessively smooth the original data, resulting in a significant peak-shaving and valley-filling effect, and fail to retain the extreme characteristics of the load. Second, clustering methods, including K-means clustering, K-medoids clustering, and density clustering. Although these methods can identify typical load patterns, they mostly rely on Euclidean distance to measure curve similarity, ignoring the temporal shape and local detailed features of the load curves, and the clustering is usually directly... Using the mean or center point to represent the curve within the cluster further exacerbates the loss of extreme information; thirdly, deep learning methods, including autoencoders, convolutional neural networks, and Transformer time series models, rely on a large amount of training data, and the black-box nature of the model leads to a lack of interpretability of the extracted load features, making it difficult to meet the stringent requirements of power system planning for physical consistency; fourthly, single feature decomposition methods, including Fourier decomposition, principal component analysis, and polynomial fitting, have limited application scenarios and cannot simultaneously meet the differentiated needs of energy storage planning and configuration and grid operation verification.

[0004] The core shortcomings of the aforementioned existing technologies are as follows: First, there is a fundamental difference in the focus of daily load characteristics required during the power grid planning phase and the operation verification phase. The planning phase focuses on typical load patterns under general conditions to determine energy storage configuration schemes, while the operation verification phase focuses on extreme load conditions under boundary conditions to verify the system's safety margin. However, existing methods lack an integrated framework for extracting daily load characteristics across the entire energy storage planning and operation chain, resulting in inconsistencies and difficulties in coordinating the load characteristics used for planning and verification. Second, existing typical curve extraction often employs a mean synthesis strategy, which severely weakens the representativeness of peak load, valley load, and extreme ramp rate, failing to accurately characterize the most stringent charging and discharging switching scenarios that energy storage needs to cope with. Third, the boundary processing and feature extraction processes of existing methods are disconnected, and the boundary coefficients lack physical constraints and verification feedback mechanisms, making it difficult to construct extreme operating condition boundaries that meet the needs of production simulation and safety and stability analysis while preserving the continuity of curve shapes. Fourth, existing methods lack a systematic approach to extracting energy storage-sensitive indicators such as maximum upward ramp rate, maximum downward ramp rate, and peak-valley duration, and have not established a direct mapping channel from raw measurement data to energy storage configuration parameters.

[0005] Therefore, there is an urgent need for a method for extracting and applying daily load characteristic sets for energy storage planning and operation, so as to unify the model to meet the differentiated needs of planning and operation, and to establish a closed-loop correction mechanism between the extraction results and the verification of energy storage configuration. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for extracting and applying daily load characteristic sets for energy storage planning and operation, in order to solve the technical problems that existing daily load characteristic extraction methods cannot simultaneously meet the differentiated needs of energy storage planning and configuration and grid operation safety verification with a unified dataset, and lack a closed-loop correction mechanism from raw measurement data to energy storage configuration verification.

[0007] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution: A method for extracting and applying daily load characteristic sets for energy storage planning and operation includes the following steps: S1. After preprocessing the minute-level daily load data, calculate the dynamic time regularization distance between every two curves in the monthly daily load curve sequence, and perform K-medoids clustering using the dynamic time regularization distance as a metric to extract typical daily load characteristic curves and curve clusters. S2. Based on the cluster center, the curve cluster is dynamically time-normalized and aligned, the upper and lower envelopes are calculated point by point while maintaining feature smoothness, and a general pattern representative curve is generated through continuity constraints and power correction, and a boundary pattern representative curve is generated through extreme value expansion and ramp enhancement. Feature values ​​are extracted to form a daily load characteristic set. S3. Use the general pattern subset for energy storage configuration and the boundary pattern subset for production simulation and safety and stability verification. Correct the clustering parameters and boundary coefficients in reverse based on the verification deviation, and iterate until the verification is passed.

[0008] As a preferred embodiment of the present invention, S1 specifically includes: S11. Export the 1-minute resolution daily load data of each bus load point from the power system monitoring system, identify and remove abnormal data points that exceed three times the standard deviation of the mean using the 3σ principle, fill in the removed points and the original missing points using the cubic spline interpolation method, and form a daily load curve sequence by slicing the preprocessed daily load data according to the natural day. S12. Calculate the dynamic time-normalized distance between every two curves in the daily load curve sequence month by month, and construct a distance matrix; S13. Using the dynamic time-normalized distance in the distance matrix as a similarity metric, perform K-medoids clustering on the daily load curves, extract the curve with the smallest distance sum as the typical daily load characteristic curve, and retain all the original curves within each cluster.

[0009] As a preferred embodiment of the present invention, S12 specifically includes: S121. Select one curve as the reference sequence and the other curve as the query sequence from the daily load curve sequence of each month, and construct a dynamic time warping cumulative distance matrix. The rows of the matrix correspond to each sampling time of the reference sequence, and the columns correspond to each sampling time of the query sequence. Each element in the matrix is ​​the absolute value of the difference between the load values ​​of the reference sequence and the query sequence at the corresponding time. S122. In the cumulative distance matrix, search for the optimal regularized path from the starting position to the ending position. The path search follows the constraint that the path can only move to the right, down or to the lower right, so as to minimize the sum of the cumulative distances of all elements on the path. The minimum sum of cumulative distances is the dynamic time regularized distance between the two curves. S123. Iterate through all pairwise combinations of daily load curves within each month, repeatedly perform cumulative distance matrix construction and optimal regularization path search, and arrange all dynamic time regularization distance calculation results according to row and column correspondence to construct a symmetric distance matrix.

[0010] As a preferred embodiment of the present invention, S13 specifically includes: S131. The optimal number of clusters is determined by using the silhouette coefficient, which is calculated based on the ratio of the average dynamic time warp distance between daily load curves within the same cluster to the minimum average dynamic time warp distance between curves in different clusters. S132. Under the optimal number of clusters, execute the K-medoids clustering algorithm. After randomly selecting the initial medoid, assign each daily load curve to the cluster where the nearest medoid is located. Update the curve in each cluster with the smallest sum of dynamic time regularization distances to all other curves in the cluster as the new medoid. Repeat the iteration until the medoid no longer changes. S133. Use the final determined medoid curve of each cluster as the cluster center curve, that is, the typical daily load characteristic curve representing different date types. Save the set of all original daily load curves in each cluster for envelope analysis in subsequent steps.

[0011] As a preferred embodiment of the present invention, S2 specifically includes: S21. Using the cluster center curve as the time reference, the original curves within the cluster are dynamically time-normalized and aligned. The maximum and minimum values ​​are taken point by point to form upper and lower envelopes. After feature-preserving smoothing, the initial representative curve is constructed. S22. Perform physical boundary clipping, rate of change constraint, charge correction and time period regularization on the initial representative curve to generate a general mode representative curve; at the same time, perform extreme value expansion, slope enhancement, peak-valley alignment and N-1 upward correction to generate a boundary mode representative curve. S23. Extract load characteristic values, including peak load, valley load, peak-valley difference, peak-valley difference rate, peak load duration, valley load duration, maximum upward ramp rate, maximum downward ramp rate, and daily load curve index, from the general mode representative curve and the boundary mode representative curve, respectively. The load characteristic values ​​of the general mode representative curve constitute a subset of the general mode daily load characteristics, and the load characteristic values ​​of the boundary mode representative curve constitute a subset of the boundary mode daily load characteristics. The two together constitute a daily load characteristic set.

[0012] As a preferred embodiment of the present invention, S21 specifically includes: S211. Using the cluster center curve as a reference sequence and any original daily load curve within the same cluster as a query sequence, construct a dynamic time warping cumulative distance matrix, and search for the optimal warping path from the starting point to the ending point in the cumulative distance matrix. S212. Based on the optimal regularization path, remap each sampling point of the query sequence to the time axis of the reference sequence, so that the peaks and valleys of each curve and the climbing sections achieve precise point-to-point matching in time sequence; on the aligned curve set, take the maximum and minimum load values ​​of all curves for each time sampling point to form the upper envelope and lower envelope. S213. The upper and lower envelopes are lightly smoothed by combining moving window midpoint filtering and local linear interpolation to preserve features, remove high-frequency spikes while retaining the amplitude of extreme points and the slope characteristics; based on the smoothed upper and lower envelopes, the mean curve of the two or the combination of the two is taken as the initial representative curve.

[0013] As a preferred embodiment of the present invention, S22 specifically includes: S221. Apply reasonable upper and lower limit physical boundary clipping to the initial representative curve to limit the load value to a range that is non-negative and does not exceed 95% of the historical highest load of the load point; apply adjacent time change rate constraint to the clipped curve to force the adjacent time load change rate not to exceed the preset maximum ramp rate, and use piecewise linear interpolation to repair it when it exceeds the limit. S222. Perform power consistency correction on the repaired curve, scaling it according to the ratio of the actual daily total power to the daily total power represented by the curve; perform time period normalization on the power correction curve, normalizing it into a stable platform segment and a linear ramp segment according to the power grid dispatching time period, and generating a general mode representative curve for production simulation verification. S223. Perform an extreme value expansion operation on the initial representative curve, increasing the peak load upward to 1.05 to 1.20 times the original value and decreasing the valley load downward to 0.80 to 0.95 times the original value; perform a slope enhancement operation on the expanded curve, increasing the slope of the rising edge and the slope of the falling edge to 1.1 to 1.5 times the original value, respectively. S224. Perform peak-valley time-shift alignment operation on multiple enhanced curves in the same area to align the peak and valley times of each curve respectively; after alignment, perform N-1 scenario-adaptive load upward adjustment on the loads corresponding to key substations or key lines according to a preset ratio to generate representative boundary mode curves for safety and stability analysis.

[0014] As a preferred embodiment of the present invention, S3 specifically includes: S31. Input the subset of daily load characteristics of the general mode into the independent energy storage configuration model, determine the energy storage technology type based on the maximum upward ramp rate and the maximum downward ramp rate, determine the energy storage power scale based on the peak load, determine the energy storage duration based on the product of the peak load and its duration, and check the curve shape matching degree of the energy storage configuration scheme based on the daily load curve index to generate the energy storage configuration scheme. S32. Input both the general mode daily load characteristic subset and the boundary mode daily load characteristic subset into the production simulation verification model, and input the boundary mode daily load characteristic subset into the safety and stability verification model to verify the adaptability of the energy storage configuration scheme under normal and extreme operating conditions and the grid safety margin. S33. Based on the deviation information between the production simulation and safety verification results, reverse the number of clusters, the width of the dynamic time warping window, the expansion coefficient and the ramp-up enhancement coefficient, iteratively optimize the daily load characteristic set, and output the final energy storage configuration scheme after the two-level verification is passed.

[0015] As a preferred embodiment of the present invention, S33 specifically includes: S331. Obtain the power shortage and power shortage duration output by the production simulation verification model. When the power shortage exceeds the preset threshold and the power shortage period is distributed in multiple discrete time periods, increase the number of clusters in step S1 to subdivide the load pattern. When the power shortage duration exceeds the preset threshold, increase the extreme value expansion coefficient in step S22. S332. Obtain the voltage over-limit amplitude and voltage fluctuation frequency output by the safety and stability verification model. When the voltage fluctuation frequency exceeds the preset threshold, relax the curvature window width constraint of dynamic time curvature in step S1 to retain more local fluctuation features. When the voltage over-limit amplitude exceeds the preset threshold, increase the ramp slope enhancement coefficient in step S22. S333. Using the corrected number of clusters, dynamic time warping window width, extreme value expansion coefficient, and ramp-up enhancement coefficient as new parameters, re-execute the dynamic time warping distance calculation and K-medoids clustering of the daily load data, as well as the time-series aligned dual-mode envelope boundary processing, i.e., repeat steps S1 and S2 to generate the optimized daily load characteristic set. S334. Repeat the iterative process of generating energy storage configuration scheme, production simulation verification and safety and stability verification, and correcting parameters in reverse according to verification deviation. That is, repeat steps S31 to S33 until the verification results of both levels meet their respective preset threshold requirements, and output the final energy storage configuration scheme.

[0016] A system for extracting and applying daily load characteristic sets for energy storage planning and operation, comprising: The data acquisition and preprocessing module is used to acquire daily load data at minute-level resolution, perform anomaly removal, interpolation preprocessing, and generate daily load curve sequences by daily slices; The temporal morphological clustering module is used to measure the temporal morphological similarity of daily load curves based on dynamic time warping distance. It performs K-medoids clustering with dynamic time warping distance as the index, extracts the cluster centers as typical daily load characteristic curves, and retains the original curve set within each cluster. The dual-mode envelope boundaryization module is used to perform time-series alignment of the original curves within the cluster based on the cluster center, calculate the upper and lower envelopes point by point and construct the initial representative curve while preserving the feature smoothness; perform physical boundary clipping, rate of change constraint, power correction, and time period normalization on the initial representative curve to generate the general mode representative curve, and perform extreme value expansion, ramp enhancement, peak-valley alignment, and N-1 upward correction to generate the boundary mode representative curve; extract load feature values ​​to form a subset of daily load characteristics of the general mode and the boundary mode; The collaborative application and closed-loop correction module for energy storage planning and operation is used to input the general mode subset into the energy storage configuration model to determine the energy storage technology type, power scale and energy storage duration. It inputs the dual mode subset into the production simulation verification model and the boundary mode subset into the safety and stability verification model for verification. Based on the verification deviation, it reversely corrects the number of clusters, the width of the dynamic time warping window, the expansion coefficient and the ramp-up enhancement coefficient, and iterates and optimizes until the two-level verification is passed before outputting the final energy storage configuration scheme.

[0017] Compared with the prior art, the present invention has the following advantages: 1. Using dynamic time warping distance as a similarity metric for K-medoids curve clustering, typical load patterns are identified in the time-domain morphological space. This preserves both the overall trend and the details of local fluctuations, fundamentally avoiding the neglect of time-series morphology by traditional Euclidean distance clustering and the peak-shaving and valley-filling problems caused by mean synthesis. This enables high-precision load characteristic modeling based on actual power grid operation data.

[0018] 2. A dynamic time warping time sequence alignment and upper and lower envelope extraction mechanism is introduced. Based on the cluster center, the curves within the cluster are nonlinearly scaled and aligned. The upper and lower envelopes are calculated point by point to represent the running boundary. This avoids the peak-shaving and valley-filling problem of traditional mean synthesis and ensures the extreme representativeness of peak value, valley value and extreme ramp rate.

[0019] 3. A dual-mode coupling architecture of general mode and boundary mode is proposed to unify the dataset and meet the differentiated needs of energy storage planning and configuration and grid security verification. A closed-loop mechanism of verification deviation reverse correction clustering strategy and boundary coefficient is established to realize the continuous evolution of daily load characteristic set. Attached Figure Description

[0020] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0021] Figure 1This is a flowchart illustrating the method described in Embodiment 1 of the present invention. In the figure, DTW is an abbreviation for Dynamic Time Warping.

[0022] Figure 2 This is a framework diagram of the system described in Embodiment 2 of the present invention. Detailed Implementation

[0023] 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.

[0024] The concepts involved in this application will first be described with reference to the accompanying drawings. It should be noted that the following descriptions of various concepts are only for the purpose of making the content of this application easier to understand and do not constitute a limitation on the scope of protection of this application; furthermore, the embodiments and features in the embodiments of this application can be combined with each other unless otherwise specified. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] Example 1

[0026] like Figure 1 As shown, this invention provides a method for extracting and applying daily load characteristic sets for energy storage planning and operation, including the following steps: S1. After preprocessing the daily load data at the minute level, calculate the dynamic time warping (DTW) distance between every two curves in the monthly daily load curve sequence. Using the DTW distance as a metric, perform K-medoids clustering to extract typical daily load characteristic curves and curve clusters; specifically including: S11. Minute-level daily load data acquisition and preprocessing, specifically: S111. Export the 1-minute resolution daily load data of each bus load point from the power system monitoring system to form a raw measurement dataset. Perform statistical characteristic calculations on this raw measurement dataset, iterate through all measurement values ​​to calculate the arithmetic mean, and calculate the deviation of each measurement value from the arithmetic mean. Calculate the standard deviation based on all deviations. Use the 3σ principle to identify outliers. Data points whose measurement values ​​exceed three times the standard deviation of the arithmetic mean are identified as outliers. These data points are usually caused by communication interruptions, sensor failures, or transient electromagnetic interference, and are removed from the raw measurement dataset.

[0027] S112. For data gaps caused by removing outlier data points, and for original missing points in the original measurement dataset due to communication interruption or acquisition device failure, cubic spline interpolation is used to fill in the missing data. This interpolation method uses multiple valid data points adjacent to each other on both sides of the missing data point as interpolation nodes, and constructs a cubic polynomial function in each adjacent node interval. It requires that the function value, first derivative value and second derivative value of each piecewise polynomial at the node remain continuous, so that the filled data curve maintains a smooth transition in local shape and a continuous connection in overall trend, avoiding the introduction of false fluctuations in the interpolation results.

[0028] S113. The daily load data after anomaly removal and interpolation are sliced ​​according to natural days. A complete time unit is defined as the period from 00:00 on each day to 00:00 on the next day. Each time unit contains 1440 minute-level load measurements, forming a daily load curve sequence composed of multiple daily load curves arranged in chronological order. This sequence serves as the basic input for subsequent time-domain morphological clustering based on dynamic time warping distance.

[0029] S12. Construction of the monthly dynamic time-normalized distance matrix, specifically as follows: S121. Selection of reference sequence and query sequence and construction of cumulative distance matrix, specifically: From the monthly daily load curve series, one curve is selected sequentially as the reference series, and another curve is selected sequentially as the query series. Both curves contain all minute-level load measurements within that natural day. A dynamic time-warped cumulative distance matrix is ​​constructed, where the number of rows equals the total number of sampling points in the reference series, and the number of columns equals the total number of sampling points in the query series. For minute-level resolution daily load data, both the number of rows and columns are 1440. The element in the i-th row and j-th column of the matrix is ​​the absolute value of the difference between the load value of the i-th sampling point in the reference series and the load value of the j-th sampling point in the query series. This absolute value characterizes the degree of local difference in load amplitude between the two sampling points.

[0030] By traversing all sampling point combinations of the reference sequence and the query sequence, the absolute values ​​mentioned above are calculated point by point and filled into the corresponding positions of the matrix, thus completing the construction of the dynamic time warping cumulative distance matrix. This matrix provides all local distance information for subsequent search of the optimal warping path.

[0031] When constructing the dynamic time-warped cumulative distance matrix, a warping window width constraint is applied. Local distance calculation and optimal path search are performed only on matrix elements that satisfy the condition that the absolute value of the difference between the sampling time of the reference sequence and the sampling time of the query sequence does not exceed the preset warping window width. The preset warping window width ranges from 10% to 30% of the total sequence length, with a classic value of 20%.

[0032] S122. Optimal regularized path search and dynamic time regularized distance calculation, specifically: In the cumulative distance matrix, the optimal regularized path is searched from the starting position in the upper left corner to the ending position in the lower right corner. The path search adopts a dynamic programming algorithm, which recursively calculates the minimum cumulative distance to each matrix element row by row from the starting position. During the recursion, the local continuity constraint condition is followed, which allows each step to move to the right, down, or lower right, to ensure that the temporal correspondence between the reference sequence and the query sequence remains monotonically continuous.

[0033] For each matrix element, compare the cumulative distances to the element from its left, top, and upper left. Select the minimum of these three distances and add it to the element's local distance to obtain the minimum cumulative distance to that element. Calculate this point by point according to the above recursive rule until the final position. The sum of the minimum cumulative distances at the final position is the dynamic time warping distance between the two curves. This distance quantifies the overall morphological difference between the two daily load curves under conditions that allow for nonlinear time-series scaling.

[0034] S123. Full combinatorial distance calculation and symmetric distance matrix construction, specifically: The algorithm iterates through all pairwise combinations of daily load curves within each month's sequence, repeatedly constructing a cumulative distance matrix and searching for the optimal normalized path for each pair, calculating the dynamic time normalized distance for each pair. All pairwise calculation results are arranged according to the correspondence between the reference and query sequences, constructing a symmetric distance matrix. The rows and columns of this matrix correspond to the numbers of all daily load curves, and the element in the p-th row and q-th column represents the dynamic time normalized distance between the p-th and q-th curves, with the element in the p-th row and q-th column having the same value as the element in the q-th row and p-th column. This symmetric distance matrix fully characterizes the pairwise morphological similarity between all daily load curves within a month, serving as the foundational input for subsequent silhouette coefficient calculation and K-medoids clustering algorithm execution.

[0035] S13. K-medoids clustering and typical curve extraction based on dynamic time warping distance, specifically: S131. Determining the optimal number of clusters guided by the silhouette coefficient, specifically: The optimal number of clusters is determined by the silhouette coefficient method: a candidate range for the number of clusters is pre-defined, and a number is selected as the current number of experiments. Preliminary clustering is performed on all daily load curves within each month.

[0036] For each cluster, calculate the cohesion index, which is the average of the dynamic time-warped distances between all pairs of curves within that cluster; calculate the separation index, which is the minimum of the average of the dynamic time-warped distances between each curve in that cluster and all curves in the nearest other cluster; use the ratio of the separation index to the cohesion index as the silhouette coefficient of that cluster, and calculate the average of all cluster silhouette coefficients to obtain the overall silhouette coefficient for the current number of trials. Iterate through all candidate clusters and select the cluster number corresponding to the largest overall silhouette coefficient as the optimal cluster number, which ensures high similarity among curves within clusters and significant differentiation between curves between clusters.

[0037] S132. The K-medoids clustering algorithm is executed iteratively, specifically: The K-medoids clustering algorithm is executed with a given optimal number of clusters: a curve with the same number of clusters as the optimal number of clusters is randomly selected from all daily load curves within a month as the initial medoid, and each initial medoid represents the center of a cluster.

[0038] For each daily load curve, calculate its dynamic time-warped distance with all initial medoids, and assign the curve to the cluster containing the medoid with the smallest dynamic time-warped distance, completing the initial clustering. Within each partitioned cluster, sequentially treat each curve in the cluster as a candidate medoid, calculate and sum the dynamic time-warped distances between the candidate medoid and all other curves in the cluster, and select the curve with the smallest sum of distances as the new medoid for that cluster. Recalculate the dynamic time-warped distance between each daily load curve and each new medoid, and again assign the curve to the cluster containing the nearest medoid. Repeat the medoid update and curve assignment process, iterating round by round until the medoids of all clusters no longer change, and the clustering results converge and stabilize.

[0039] S133. Cluster center curve extraction and curve cluster storage, specifically: The medoid curve determined after iterative convergence for each cluster is taken as the cluster center curve of that cluster. This cluster center curve is closest in shape to the overall distribution characteristics of all other curves in the cluster and can represent the common pattern of the load curves in the cluster. Therefore, it is determined as the typical daily load characteristic curve representing different date types, such as the typical curve for weekdays, the typical curve for rest days, or the typical curve for seasonality.

[0040] While determining the cluster center curves, a set of all original daily load curves within each cluster is saved. This set, as a curve cluster, completely preserves all historical load measurement information for that date type. It is used in subsequent steps for dynamic time warping and time series alignment based on the cluster center curves, and for calculating the upper and lower envelopes point-by-point on the aligned basis. The cluster center curves of each cluster are numbered and archived by month and date type, forming a typical daily load characteristic curve library for each month and date type, serving as direct input for the subsequent construction of daily load characteristic sets.

[0041] S2. Using the cluster center as a reference, the curve clusters are dynamically time-normalized and aligned. The upper and lower envelopes are calculated point by point while preserving feature smoothness. After continuity constraints and power correction, a general pattern representative curve is generated. The boundary pattern representative curve is generated by extreme value expansion and ramp enhancement. Feature values ​​are extracted to form a daily load characteristic set; specifically including: S21. Temporal flexible alignment and envelope extraction, specifically: S211. Dynamic time warping and timing alignment of curves within a cluster, specifically: Let the set of all original daily load curves within the cluster containing the cluster center curve be denoted as . ,in This represents the Mth original daily load curve within the cluster, where M is the total number of original daily load curves within the cluster; using the cluster center curve as a reference sequence, from the set... Each original daily load curve is selected sequentially as the query sequence.

[0042] Construct a dynamic time-warped cumulative distance matrix. The rows of the matrix correspond to the sampling times of the reference sequence, and the columns correspond to the sampling times of the query sequence. Each element in the matrix is ​​the absolute value of the difference between the load values ​​of the reference sequence and the query sequence at the corresponding time. In the cumulative distance matrix, search for the optimal warped path from the starting position in the upper left corner to the ending position in the lower right corner. The path search follows the constraint that each step can only move to the right, down, or lower right, so as to minimize the sum of the cumulative distances of all elements on the path.

[0043] S212. Temporal remapping and upper and lower envelope calculation based on optimal regularized paths, specifically: Establish a time mapping relationship between the query sequence and the reference sequence based on the optimal regularization path. Remap each sampling point of the query sequence to the time axis of the reference sequence according to the correspondence of the optimal regularization path, so that the peak points, valley points and ramps of the query sequence are precisely matched with the reference sequence in time, and the time alignment of a single curve is completed. Perform the above alignment operation on all original daily load curves in the cluster one by one, so that the peak and valley characteristics of all curves in the cluster are aligned with the ramps on a unified time reference.

[0044] After all curves are aligned, at each time sampling point t, the maximum and minimum load values ​​of all aligned curves at that moment are taken. The upper envelope is formed by arranging the sequence of maximum load values ​​from all times in chronological order. The upper envelope represents the upper limit boundary of this type of load during operation, and its mathematical expression is:

[0045] The lower envelope is formed by arranging the load minimum values ​​at all times in chronological order. The lower envelope represents the lower limit boundary of this type of load during operation, and its mathematical expression is:

[0046] in Indicates the first The curve at time Load value after dynamic time-normalized alignment.

[0047] The upper and lower envelopes correspond point by point, jointly defining the complete boundary range of load fluctuations within the cluster, providing a boundary benchmark for the subsequent construction of the initial representative curve.

[0048] S213. Envelope feature-preserving smoothing and initial representative curve construction, specifically: The upper and lower envelopes are subjected to feature-preserving light smoothing. A moving window midpoint filter is used to remove high-frequency spikes. The window width is determined based on the minute-level sampling resolution of the daily load curve. The median of the load values ​​of all sampling points within the window is taken as the filtering result of the center point, which effectively suppresses isolated noise points and small jitters, while keeping the amplitude of extreme points unchanged, without peak clipping or altering the ramp slope.

[0049] Local linear interpolation is performed on the filtered envelope to establish a linear transition relationship between adjacent sampling points, correcting the small step discontinuities that may be introduced by median filtering, so that the overall envelope remains smooth and continuous and conforms to the physical trend of load change.

[0050] Based on the smoothed upper and lower envelopes, the arithmetic mean of the two is calculated point by point to form an initial representative curve. This curve comprehensively reflects the average trend and boundary range of load operation within the cluster, serving as the common basic input for subsequent general mode boundary processing for production simulation verification and boundary mode boundary processing for safety and stability analysis.

[0051] S22. Dual-mode envelope boundary delimitation and representative curve generation, specifically: S221. Physical boundary clipping and rate of change constraints, specifically: The initial representative curve is subjected to reasonable upper and lower limit physical boundary pruning. The load value is read point by point through all sampling times of the curve. The load value is restricted to a non-negative range as the lower limit constraint, that is, all load values ​​less than 0 are corrected to 0 to ensure that the load does not reach the physically impossible state of negative value. The upper limit threshold is adaptively determined based on the historical operating data of the bus load point. 95% of the historical highest load of the load point is taken as the upper limit threshold. Sampling points exceeding the upper limit threshold are subjected to asymptotic saturation processing, so that the over-limit load value gradually converges to the upper limit threshold with a decreasing gradient, avoiding curve shape distortion and derivative abrupt change caused by direct truncation.

[0052] After physical boundary trimming is completed, a continuous boundary constraint is applied. The first-order difference between every two adjacent sampling points is calculated as the load change, and the ratio of the load change to the sampling interval is used as the rate of change. This rate of change is forced not to exceed the preset maximum allowable ramp rate. When the load change rate at adjacent time points is detected to exceed the maximum allowable ramp rate, three methods are used in sequence for correction: first-order difference constraint, moving window slope constraint, and piecewise linear repair. The first-order difference constraint redistributes the total change between the first and last points of the over-limit section to adjacent sampling intervals. The moving window slope constraint scans the curve with a fixed-width window and smooths abnormal slopes within the window. The piecewise linear repair establishes a linear transition relationship within the over-limit section and re-interpolates the load value, so that the rate of change at adjacent time points is evenly distributed and meets the ramp constraint, ensuring that the curve has no instantaneous abrupt changes, no negative loads, and no unreasonable sudden increases or decreases, thus maintaining physical continuity and engineering feasibility.

[0053] S222. Power Consistency Correction and Time Period Normalization, specifically: After the curve is repaired by the rate of change constraint, a power consistency correction is performed. Numerical integration is performed on the curve along the time axis to calculate the total power of the curve integral for the day represented by the curve, which is taken as the original power. The actual total power of the statistical day is obtained from the actual operation record of the day corresponding to the load point, which is taken as the true power. Based on the true power, the load values ​​of all sampling points of the curve are scaled proportionally according to the ratio of the true power to the original power to ensure that the simulated power is consistent with the statistical power. The scaling formula is as follows:

[0054] in The load value at time t before the power correction. This is the original power level. This is the actual battery level. This represents the load value at time t after the power correction.

[0055] The power-corrected curve undergoes time-period normalization. Based on the grid dispatch time-period division rules, the curve slope variation characteristics are scanned segment by segment. Continuous segments with absolute slope values ​​less than a preset threshold are identified as stable platform segments, while continuous segments with absolute slope values ​​greater than the preset threshold and in the same direction are identified as linear ramp segments. The preset threshold ranges from 0.1% to 2.0% per minute of the historical peak load of the bus, with a classic value of 0.5% per minute. Minor fluctuations within the stable platform segments are smoothed by mean value processing to eliminate sawtooth jitter. Nonlinear sawtooth patterns within the linear ramp segments are linearly fitted to achieve a monotonically smooth linear transition, eliminating sawtooth and spike effects. This generates a smooth, continuous, and representative curve for production simulation verification that conforms to the characteristics of the dispatch time period.

[0056] S223. Extremum expansion and slope enhancement, specifically: An extreme value expansion operation is performed on the initial representative curve. The curve is scanned point by point to identify local maxima as peak load points and local minima as valley load points. While preserving the overall shape of the curve and the monotonicity of adjacent segments, the load value at the peak load point is increased upward to 1.05 to 1.20 times the original value, and the load value at the valley load point is reduced downward to 0.80 to 0.95 times the original value, so that the peak-valley difference is significantly expanded and the load boundary is stretched in the unfavorable direction.

[0057] The expanded curve undergoes maximum ramp reinforcement boundary processing. Rising and falling edge segments are identified segment by segment. Rising edge segments are defined as segments where the load value of consecutive sampling points monotonically increases, and falling edge segments are defined as segments where the load value of consecutive sampling points monotonically decreases. The ramp slope of the rising edge segments is amplified to 1.1 to 1.5 times the original value, and the ramp slope of the falling edge segments is simultaneously amplified to 1.1 to 1.5 times the original value, constructing the steepest ramp and steepest descent scenarios. During the reinforcement process, the continuity of slope between adjacent segments and piecewise linear interpolation smoothing transition techniques are used to ensure that the curve remains continuous and does not jump at peak-valley inflection points and segment junctions, avoiding unreasonable load abrupt changes. This ensures that the expanded curve still conforms to the physical laws of load change, generating an intermediate boundary curve with extreme representativeness.

[0058] S224. Peak-valley time-shift alignment and N-1 scene adaptation, specifically: For multiple intermediate boundary curves under the same region or voltage level after ramp-up enhancement, a peak-valley time-shift alignment operation is performed. The peak and valley times of each curve are identified individually. The peak times of all curves are uniformly adjusted to the same preset target time, so that the entire network load reaches an extreme heavy load state at the same time. Similarly, the valley times of all curves are uniformly adjusted to the same preset target time, so that the entire network load reaches an extreme light load state at the same time. During the time-shift adjustment process, the rising edge shape before the peak time and the falling edge shape after the peak time of each curve remain unchanged. Only the time axis is shifted overall or locally stretched, constructing extreme operating condition boundaries where the entire network is simultaneously under heavy load and simultaneously under light load.

[0059] After alignment, the boundary mode representative curves corresponding to key substations or key lines undergo N-1 scenario adaptation boundary processing. Based on the power grid topology and the N-1 fault set, the power transfer path after the fault is identified, and the corresponding load value is adjusted upward according to a preset transfer ratio. This ensures that the load increment reflects the power transfer effect after the N-1 fault, forming a verification boundary load where the lines and main transformers are still not overloaded after the N-1 fault. The curves after peak-valley time-shift alignment and N-1 scenario adaptation serve as the boundary mode representative curves for safety and stability analysis. These curves possess the most stringent extreme operating condition characteristics and are used to verify the safety and stability margin of the power grid under scenarios of multiple faults and extreme loads.

[0060] S23. Dual-mode load feature extraction and daily load characteristic set construction, specifically: S231. Scan all sampling times point by point on the representative curve of the general mode, and identify local maxima as peak loads. Identify local minima as valley loads The difference between peak load and valley load is calculated as the peak-valley difference, and the ratio of the peak-valley difference to the peak load is calculated as the peak-valley difference rate.

[0061] The duration of peak load exceeding a preset threshold is defined as the peak load duration. The duration of the valley load is defined as the length of time that the valley load remains below a preset threshold. The preset judgment threshold is set based on the baseline fluctuation characteristics of the load curve. The judgment threshold for the duration of peak load is 85% to 98% of the peak load, and the classic value is 95% of the peak load. The judgment threshold for the duration of valley load is 102% to 115% of the valley load, and the classic value is 110% of the valley load.

[0062] The maximum load increase between adjacent minute-level sampling points is calculated as the maximum upward ramp rate. The maximum load decrease between adjacent minute-level sampling points is calculated as the maximum downhill ramp rate. The daily load curve index is obtained by performing numerical integration along the time axis on the representative curve of the general model. It is used to comprehensively characterize the overall morphological features of the daily load curve.

[0063] The above-mentioned peak load, valley load, peak-valley difference, peak-valley difference rate, peak load duration, valley load duration, maximum upward ramp rate, maximum downward ramp rate, and daily load curve indicators constitute a subset of the daily load characteristics of the general model. Its mathematical expression is: .

[0064] S232. Perform the same eigenvalue extraction process on the boundary pattern representative curve, scanning point by point to identify the peak load after extreme value expansion. Valley load The peak duration after time period normalization was statistically analyzed. Duration of valley Calculate the maximum uphill climbing rate after climbing enhancement. With maximum downhill climbing rate Extract peak-to-valley difference, peak-to-valley difference rate, and daily load curve indicators after boundary-based processing. Since the boundary mode representative curve has undergone extreme value expansion, ramp-up enhancement, and peak-valley time shift alignment, the extracted eigenvalues ​​reflect the load boundary characteristics under extreme operating conditions. These eigenvalues ​​constitute a subset of the daily load characteristics of the boundary mode. Its mathematical expression is: .

[0065] S233. The daily load characteristic subset of the general mode and the daily load characteristic subset of the boundary mode are merged to form a daily load characteristic set, the mathematical expression of which is: ; This daily load characteristic set covers both typical load patterns under normal conditions and extreme operating conditions under boundary conditions, providing consistent basic data on daily load characteristics for independent energy storage planning and configuration and grid operation safety verification.

[0066] S3. The general pattern subset is used for energy storage configuration, and the boundary pattern subset is used for production simulation and safety and stability verification. Clustering parameters and boundary coefficients are then adjusted in reverse based on the verification deviation, and iterative optimization is performed until the verification is passed. Specifically, this includes: S31. Generation of energy storage configuration schemes based on a subset of daily load characteristics under general conditions, specifically: A subset of daily load characteristics under normal conditions is input into an independent energy storage configuration model, which includes four functional units: energy storage technology selection module, power scale calculation module, duration configuration module, and form matching verification module.

[0067] S311. The energy storage technology selection module reads the maximum upward ramp rate and the maximum downward ramp rate from the daily load characteristic subset of the general mode, compares their absolute values, and takes the larger one as the benchmark for energy storage power response demand. When this benchmark exceeds a preset high threshold, it is determined that the load fluctuation is severe, and a power-type energy storage technology is selected to meet the minute-level rapid charging and discharging demand; when the benchmark is within the preset low threshold range, it is determined that the load change is gradual, and an energy-type energy storage technology is selected to meet the long-term energy throughput demand; when the benchmark is between the high and low thresholds, a power-energy balanced energy storage technology is selected. The preset high threshold ranges from 8% to 15% of the peak load of the bus per minute, with a classic value of 10% per minute; the preset low threshold ranges from 2% to 5% of the peak load of the bus per minute, with a classic value of 3% per minute.

[0068] The S312. Power Scale Calculation Module reads the peak load from the daily load characteristic subset of the general mode as the maximum load gap. It introduces the energy storage converter conversion efficiency parameter, with a value ranging from 0.88 to 0.97 and a classic value of 0.93. Dividing the maximum load gap by this conversion efficiency compensates for power losses during the AC / DC bidirectional conversion process, yielding the loss compensation power. A power redundancy coefficient, ranging from 1.1 to 1.3 and a classic value of 1.2, is introduced. Multiplying the loss compensation power by this coefficient allows for a comprehensive safety margin to account for unexpected load increases, equipment aging, and environmental deviations, resulting in the energy storage rated power configuration benchmark value. When this benchmark value exceeds the upper limit of the rated power of a single energy storage converter, the total demand power is divided by the rated power of the single unit and rounded up to determine the number of units to be connected in parallel, forming an energy storage system power scale configuration scheme that includes the total rated power, the power of a single unit, and the number of units connected in parallel.

[0069] S313. The duration configuration module reads the product of the peak load and its duration as the cumulative energy demand, introduces the depth of discharge parameter (range 0.80 to 0.95, classic value 0.90), divides the cumulative energy demand by this parameter to compensate for unused capacity reserved to avoid over-discharge, and introduces the charge-discharge round-trip efficiency parameter (range 0.80 to 0.92, classic value 0.85), divides the corrected energy demand by this parameter to compensate for cycle energy loss, and obtains the energy storage rated capacity baseline value. Dividing this baseline value by the energy storage rated power yields the energy storage duration configuration. When this duration exceeds the upper limit of the rated duration of a single battery cluster, the total required duration is divided by the rated duration of a single cluster and rounded up to determine the number of clusters, forming an energy storage system duration configuration scheme that includes the total rated capacity, duration, and number of clusters.

[0070] S314. The morphology matching verification module reads the daily load curve indicators from the subset of daily load characteristics in the general mode, and simultaneously reads the output curves of the candidate energy storage configuration schemes jointly generated by the energy storage technology selection module, power scale calculation module, and duration configuration module. The Pearson curve correlation coefficient is used to quantify the linear correlation between the two curves in terms of peak-valley correspondence, ramp-up period distribution, and power balance characteristics. Specifically, the ratio of the product of the covariance and standard deviation of the two curves is calculated to measure the degree of synchronization of fluctuation trends; the closer the coefficient is to 1, the higher the morphology matching degree. When the matching degree indicator meets the preset threshold, the energy storage configuration scheme is solidified and the final result is output; when it does not meet the threshold, a parameter fine-tuning command is triggered and fed back to the power scale calculation module and duration configuration module, iteratively optimizing the rated power and duration until the matching degree meets the standard. The preset threshold ranges from 0.80 to 0.95, with a classic value of 0.90.

[0071] S32. Dual-mode verification and grid adaptability validation, specifically: S321. Obtain the subset of daily load characteristics of the general mode and the subset of daily load characteristics of the boundary mode output in step S2, and input both into the production simulation verification model. The model includes three functional units: time series simulation engine module, energy storage output tracking module and power balance verification module.

[0072] The time-series simulation engine module simulates the entire daily operation process point by point at minute-level resolution. It inputs the daily load characteristic subsets of the general mode and boundary mode as load boundary conditions into the power grid simulation model, and simultaneously embeds the new energy output fluctuation scenario and the dynamic constraints of energy storage charge state to simulate the real-time balance process of power generation and consumption during the day. It monitors the system frequency deviation, spinning reserve capacity occupancy status and power flow distribution of key sections in each time period, and outputs the system operation status sequence minute by minute during the day, providing time-series simulation basic data for energy storage output tracking and power balance verification.

[0073] Based on the generated energy storage configuration scheme and the operating state sequence output by the time-series simulation engine, the energy storage output tracking module simulates the energy storage system performing charging and discharging operations according to load fluctuations at various time periods. It records the deviation between the actual energy storage output and the target output, the frequency of charging and discharging state switching, and the distribution of continuous power deficit periods every minute. When the actual output is lower than the target output and the deviation persists, it marks the power deficit event and counts the duration of a single power deficit and the cumulative duration of power deficit within the day, and outputs an energy storage output tracking performance evaluation report.

[0074] The power balance verification module calculates the total daily power generation, total load, and total energy storage charge / discharge based on time-series simulation results. It then calculates the system's net power deficit and power deficit duration. The preset range for power deficit is 0% to 5% of the energy storage's rated capacity, with a classic value of 3%. The preset range for power deficit duration is 15 minutes for a single continuous deficit and no more than 30 minutes for the cumulative daily deficit, with a classic value of 10 minutes for a single deficit and 20 minutes for the cumulative deficit. When both the power deficit and power deficit duration are within their respective preset ranges, the energy storage configuration scheme is deemed to meet the requirements for power adaptability and energy adaptability under normal and extreme operating conditions.

[0075] S322. Input the daily load characteristic subset of the boundary mode into the safety and stability verification model. The model includes five functional units: power flow calculation module, N-1 fault scanning module, transient stability assessment module, voltage safety margin analysis module, and thermal stability verification module.

[0076] The power flow calculation module reads minute-level load data from the daily load characteristic subset of the power grid topology parameters and boundary modes. It uses the Newton-Raphson method to iteratively solve for the voltage amplitude and phase angle of each node, the active and reactive power distribution of each line and transformer, and calculates the ratio of the transmission power to the limit transmission capacity of each section as the section load rate. It identifies the power flow heavily loaded sections and voltage-weak nodes, and outputs the benchmark results of the power flow distribution of the entire network, providing the initial values ​​of electrical quantities under normal operating conditions for subsequent N-1 fault scanning and transient stability assessment.

[0077] The N-1 fault scanning module performs single-component fault simulations sequentially based on the list of key power grid components. It disconnects key substation main transformers or key transmission lines, triggering topology reconfiguration and power transfer processes. It then calls the power flow calculation module again to solve the steady-state power flow distribution after the fault, records the power increment, node voltage change, and equipment load rate change at each transfer section, identifies weak components with excessive load rate or voltage after the fault, and forms a scanning result set covering all N-1 fault scenarios.

[0078] Based on the N-1 fault scan, the transient stability assessment module performs transient process simulation on the system after the fault, monitors the swing curve of the rotor power angle of each synchronous generator over time, calculates the maximum relative power angle difference between generators and its rate of change, and determines that the system is transiently stable when the maximum relative power angle difference converges to a stable value within a preset time. When the power angle difference continues to diverge or exceeds the step loss threshold, it determines that non-periodic step loss has occurred and outputs the transient stability margin assessment conclusion. The preset time range is 3s to 10s after fault clearance, with a classic value of 5s, and the step loss threshold range is 120° to 180° of the generator relative power angle difference, with a classic value of 180°.

[0079] The voltage safety margin analysis module, based on power flow calculations and N-1 fault scanning results, monitors the deviation between the voltage amplitude and the rated voltage at each bus node. It calculates the voltage over-limit amplitude, i.e., the absolute value of the difference between the actual voltage and the upper and lower voltage limit boundary values, and counts the number of voltage fluctuations per unit time as the voltage fluctuation frequency. When both the voltage over-limit amplitude and the voltage fluctuation frequency are within the safe range, the voltage safety margin is deemed to meet the requirements; otherwise, a voltage weak area is marked. The safe range for voltage deviation under normal operating conditions is -5% to +5% of the rated voltage; the safe range for voltage deviation under extreme operating conditions after an N-1 fault is -10% to +10% of the rated voltage; and the safe range for voltage fluctuation frequency is no more than 3 voltage fluctuations per minute.

[0080] The thermal stability verification module monitors the steady-state load rate of each transmission line and main transformer under normal operation and after N-1 fault. The ratio of the actual transmission power of the equipment to the long-term allowable current carrying capacity of the equipment is used as the thermal stability load rate. When the load rate exceeds the preset thermal stability threshold, it is determined that the equipment is continuously overloaded and there is a risk of thermal damage. When the thermal stability load rate of all key equipment under normal operation and after N-1 fault does not exceed the threshold, it is determined that the thermal stability margin meets the requirements of the power system operation regulations.

[0081] Based on the output results of the above modules, we verify whether the transient stability margin, voltage safety margin, and thermal stability margin of the power grid under the extreme load superimposed N-1 fault scenario meet the requirements of the power system operation regulations, and form a set of verification results that includes the verification conclusions of normal operating condition adaptability, extreme operating condition adaptability, and multi-dimensional safety margin of the power grid.

[0082] S33. Verification of deviation feedback and iterative optimization of daily load characteristic set, specifically: S331. Production simulation deviation feedback and clustering edge expansion correction, specifically: Obtain the power shortage and power shortage duration output by the production simulation verification model. When the power shortage exceeds the preset threshold and the power shortage period is distributed in multiple discrete periods, it is determined that the existing number of clusters is insufficient to cover the complex and variable load patterns. The number of clusters in step S1 is increased to subdivide the load pattern types. Each time, the number of clusters is increased by 1-3 to improve the ability of the typical curve to distinguish discrete load characteristics. The preset threshold for power shortage is 3% to 10% of the rated capacity of energy storage, with a classic value of 5%.

[0083] When the duration of power shortage exceeds a preset threshold, it is determined that the existing boundary does not fully cover the extreme power demand. The extreme value expansion coefficient in step S22 is increased, and the peak load expansion ratio is increased by 0.05-0.1 on the original basis. The coefficient is gradually increased in each iteration until the duration of power shortage meets the requirements. The preset threshold for the duration of power shortage is in the range of 20-40 minutes for a single continuous shortage, with a classic value of 30 minutes.

[0084] S332. Safety and stability deviation feedback and window width and ramp correction, specifically: The voltage over-limit amplitude and voltage fluctuation frequency output by the safety and stability verification model are obtained. When the voltage fluctuation frequency exceeds the preset threshold, it is determined that the warping window width of the dynamic time warping in step S1 is too narrow, resulting in excessive smoothing of local fluctuation features. The constraint of the dynamic time warping window width in step S12 is relaxed to retain more local time-series fluctuation features. The window width is increased by 5% to 10% of the total sequence length each time, so that the daily load curve retains more load detail changes. The preset threshold value of voltage fluctuation frequency is 3 to 5 times per minute, with the classic value being 3 times per minute.

[0085] When the voltage exceedance exceeds a preset threshold, it is determined that the existing ramp rate boundary is insufficient to support voltage recovery after a fault. The ramp slope enhancement coefficient in step S22 is increased, raising the enhancement ratio of the rising and falling edges by 0.1-0.2 from the original level. This increase is repeated in each iteration until the voltage exceedance meets the requirements. The preset threshold for voltage exceedance ranges from 5% to 10% of the rated voltage, with a classic value of 10%.

[0086] S333. Parameter correction and daily load characteristic set reconstruction, specifically: Using the revised cluster number, dynamic time warping window width, extreme value expansion coefficient, and ramp enhancement coefficient (as modified in S331 and S332) as new parameters, the dynamic time warping distance calculation and K-medoids clustering of daily load data are re-executed, along with time-series aligned dual-mode envelope boundary processing. Specifically, K-medoids clustering in step S1 is re-executed with the new cluster number to extract new cluster center curves and curve clusters; the dynamic time warping distance matrix construction in step S12 is re-executed with the new warping window width to obtain a distance matrix that better reflects actual fluctuation characteristics in terms of time-series alignment sensitivity; and the boundary mode representative curve generation in step S22 is re-executed with the new expansion coefficient and ramp coefficient to obtain boundary load characteristics that cover more extreme operating conditions.

[0087] Repeat all sub-steps of steps S1 and S2 to generate an optimized daily load characteristic set, making the characteristic set more in line with the actual needs of energy storage planning and operation in terms of load pattern differentiation, fluctuation detail retention and extreme condition coverage.

[0088] S334. Iterative convergence and final solution output, specifically: Based on the optimized daily load characteristic set, the energy storage configuration scheme generation and two-level verification are re-executed, that is, the energy storage configuration scheme generation in step S31, the production simulation verification and safety and stability verification in step S32 are repeated, and the parameter correction process based on the verification deviation in step S33 is performed in reverse. After each iteration, the production simulation verification result and the safety and stability verification result are independently verified to see if they meet their respective preset threshold requirements. The preset threshold is consistent with the threshold in S331 and S332.

[0089] When both levels of verification results meet the requirements, the iteration is considered to have converged, the iteration loop is immediately terminated, and the final energy storage configuration scheme, including the energy storage technology type, power scale, duration configuration, and form matching verification conclusions, is output. If any level of verification fails, the next round of parameter correction and characteristic set reconstruction is performed, and steps S331 to S334 are repeated until both levels of verification pass, at which point the final scheme is output.

[0090] Example 2

[0091] like Figure 2 As shown, a system for extracting and applying daily load characteristic sets for energy storage planning and operation is used to implement a method for extracting and applying daily load characteristic sets for energy storage planning and operation, including: A. Data acquisition and preprocessing module, used to acquire daily load data at minute-level resolution from each bus load point in the power system, perform anomaly removal and interpolation preprocessing on the daily load data, and form a daily load curve sequence by daily slices.

[0092] B. The temporal morphological clustering module is used to measure the temporal morphological similarity of daily load curves based on dynamic time warping distance. It performs K-medoids clustering using dynamic time warping distance as the indicator, extracts cluster centers as typical daily load characteristic curves, and retains the original curve set within each cluster. Specifically, it includes: The distance matrix construction unit is used to calculate the dynamic time-normalized distance between each two daily load curves and construct the distance matrix. The clustering execution unit is used to determine the optimal number of clusters using the silhouette coefficient, perform K-medoids clustering based on the distance matrix, take the curve with the smallest sum of dynamic time-warped distances with all other curves in each cluster as the cluster center curve, and save the set of all original daily load curves in each cluster.

[0093] C. Dual-mode envelope boundaryization module, used to perform time-series alignment of the original curves within clusters based on cluster centers, calculate upper and lower envelopes point by point while preserving feature smoothness to construct initial representative curves; physical boundary trimming, rate of change constraint, energy correction, and time period normalization are performed on the initial representative curves to generate general mode representative curves, and extreme value expansion, ramp enhancement, peak-valley alignment, and N-1 upward adjustment are performed to generate boundary mode representative curves; load feature values ​​are extracted to form subsets of daily load characteristics for general and boundary modes; specifically including: The time alignment unit is used to construct a dynamic time warping cumulative distance matrix for the original daily load curves within the cluster, using the cluster center curve as a reference sequence, and to search for the optimal warping path, remapping each sampling point of the query sequence onto the time axis of the reference sequence. The envelope generation unit is used to calculate the maximum and minimum values ​​point by point on the aligned curve set, forming the upper and lower envelopes respectively, and performing feature-preserving light smoothing. The general mode processing unit is used to sequentially perform physical boundary clipping, adjacent time change rate constraint, electric quantity consistency correction and time period normalization on the initial representative curve to generate the general mode representative curve. The boundary mode processing unit is used to sequentially perform extreme value expansion, slope enhancement, peak-valley time shift alignment, and N-1 scenario adaptation load upward processing on the initial representative curve to generate the boundary mode representative curve. The feature extraction unit is used to extract peak load, valley load, peak-valley difference, peak-valley difference rate, peak load duration, valley load duration, maximum upward ramp rate, maximum downward ramp rate, and daily load curve indicators from the general mode representative curve and the boundary mode representative curve.

[0094] D. The collaborative application and closed-loop correction module for energy storage planning and operation is used to input a general mode subset into the energy storage configuration model to determine the energy storage technology type, power scale, and energy storage duration. It also inputs a dual-mode subset into the production simulation verification model and a boundary mode subset into the safety and stability verification model for validation. Based on the verification deviation, it reverse-corrects the cluster number, dynamic time warping window width, edge expansion coefficient, and ramp-up enhancement coefficient, iteratively optimizing until both levels of verification pass, and then outputs the final energy storage configuration scheme. Specifically, it includes: An energy storage configuration unit is used to generate an energy storage configuration scheme based on characteristic parameters in a subset of daily load characteristics under a general mode. The verification unit is used to input the dual-mode subset into the production simulation verification model and the safety and stability verification model to verify the adaptability of the energy storage configuration scheme and the grid safety margin. The deviation correction unit is used to reverse the cluster number, dynamic time warping window width, extreme value expansion coefficient and ramp-up enhancement coefficient based on the deviation information of power shortage, duration of power shortage, voltage over-limit amplitude and voltage fluctuation frequency. The iterative control unit is used to repeatedly call the above modules until both levels of verification pass, and output the final energy storage configuration scheme.

[0095] As can be seen from the above description, the embodiments of the present invention achieve the following technical effects: This invention employs dynamic time-warped distance as a similarity metric for K-medoids curve clustering, identifying typical load patterns within the time-domain morphological space. This preserves both the overall trend characteristics of the daily load curve and morphological details such as peak-valley switching and local fluctuations, fundamentally avoiding the neglect of temporal morphology by traditional Euclidean distance clustering. By modeling actual power grid operation data exported from the monitoring system, daily load characteristics highly reflect the actual power grid situation. Furthermore, historical load data can be used for modeling as needed based on the load point settings in the simulation software, significantly improving the accuracy and engineering adaptability of load characteristic extraction.

[0096] This invention introduces a dynamic time-warping time-series alignment and upper / lower envelope extraction mechanism based on clustering. Using the cluster center as a reference, it performs nonlinear scaling and alignment on the curves within the cluster, ensuring precise matching of peak and valley points with ramp segments. It also calculates the upper and lower envelopes point-by-point to characterize the load operating boundary. This avoids the peak-shaving and valley-filling problems associated with traditional mean-based aggregation while ensuring the extreme nature and representativeness of key characteristics such as peak load, valley load, and extreme ramp rate. This method effectively overcomes the weakening effect of mean-based strategies on energy storage-sensitive indicators, providing a load boundary basis that more closely reflects the most demanding actual operating conditions for energy storage configuration.

[0097] This invention proposes a dual-mode coupled daily load characteristic set architecture, combining general and boundary modes. This unified dataset simultaneously covers the differentiated needs of independent energy storage planning and configuration, as well as grid operation safety verification, achieving seamless integration between typical load patterns in the planning phase and extreme operating conditions in the operation phase. By establishing a closed-loop mechanism that uses energy storage configuration schemes and verification results to inversely correct daily load characteristic extraction parameters, and dynamically adjusting clustering strategies and boundary coefficients based on deviations in production simulation and safety and stability verification, the daily load characteristic set continuously evolves with the actual grid operation, effectively solving the technical bottleneck of existing methods that involve one-time extraction and the separation of planning and operation.

[0098] The embodiments and / or implementation methods described above are merely preferred embodiments and / or implementation methods for implementing the technology of the present invention, and are not intended to limit the implementation methods of the technology of the present invention in any way. Any person skilled in the art may make some modifications or alterations to other equivalent embodiments without departing from the scope of the technical means disclosed in the present invention, but these should still be regarded as the technology or embodiments that are substantially the same as the present invention.

[0099] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for extracting and applying daily load characteristic sets for energy storage planning and operation, characterized in that, include: After acquiring minute-level daily load data and preprocessing it, the dynamic time regularization distance between every two curves in the monthly daily load curve sequence is calculated. K-medoids clustering is performed using the dynamic time regularization distance as a metric to extract typical daily load characteristic curves and curve clusters. Based on the cluster center, the curve clusters are dynamically time-normalized and aligned. The upper and lower envelopes are calculated point by point while maintaining feature smoothness. After continuity constraints and power correction, a general pattern representative curve is generated. The extreme value expansion and ramp enhancement generate a boundary pattern representative curve. Feature values ​​are extracted to form a daily load characteristic set. The general pattern subset is used for energy storage configuration, and the boundary pattern subset is used for production simulation and safety and stability verification. The clustering parameters and boundary coefficients are corrected in reverse based on the verification deviation, and iterative optimization is performed until the verification is passed. Using the cluster center curve as the time-series benchmark, the original curves within the cluster are dynamically time-normalized and aligned. The maximum and minimum values ​​are taken point by point to form upper and lower envelopes. After feature-preserving smoothing, the initial representative curve is constructed. Physical boundary clipping, rate of change constraint, charge correction and time period regularization are performed on the initial representative curve to generate a general mode representative curve; at the same time, extreme value expansion, ramp enhancement, peak-valley alignment and N-1 upward correction are performed to generate a boundary mode representative curve. Load characteristic values ​​are extracted from the general mode representative curve and the boundary mode representative curve, respectively. The load characteristic values ​​of the general mode representative curve constitute a subset of the daily load characteristics of the general mode, and the load characteristic values ​​of the boundary mode representative curve constitute a subset of the daily load characteristics of the boundary mode. Together, they constitute a set of daily load characteristics.

2. The method for extracting and applying daily load characteristic sets for energy storage planning and operation according to claim 1, characterized in that, After preprocessing the acquired minute-level daily load data, the dynamic time regularization distance between every two curves in the monthly daily load curve sequence is calculated. K-medoids clustering is then performed using this dynamic time regularization distance as a metric to extract typical daily load characteristic curves and curve clusters. Specifically, this includes: The daily load data of each bus load point with a resolution of 1 minute is exported from the power system monitoring system. The 3σ principle is used to identify and remove abnormal data points that exceed three times the standard deviation of the mean. The cubic spline interpolation method is used to fill in the removed points and the original missing points. The preprocessed daily load data is sliced ​​according to the natural day to form a daily load curve sequence. Calculate the dynamic time-normalized distance between every two curves in the daily load curve sequence each month, and construct a distance matrix; Using the dynamic time-normalized distance in the distance matrix as a similarity metric, K-medoids clustering is performed on the daily load curves. The curve with the smallest distance sum is extracted as the typical daily load characteristic curve, and all original curves within each cluster are retained.

3. The method for extracting and applying daily load characteristic sets for energy storage planning and operation according to claim 2, characterized in that, Calculate the dynamic time-normalized distance between every two curves in the daily load curve sequence each month, and construct a distance matrix, specifically including: From the monthly daily load curve sequence, one curve is selected as the reference sequence and the other curve is selected as the query sequence. A dynamic time-warped cumulative distance matrix is ​​constructed. The rows of the matrix correspond to the sampling times of the reference sequence and the columns correspond to the sampling times of the query sequence. Each element in the matrix is ​​the absolute value of the difference between the load values ​​of the reference sequence and the query sequence at the corresponding time. In the cumulative distance matrix, the optimal regularized path is searched from the starting position to the ending position. The path search follows the constraint that the path can only move to the right, down, or right and down, so as to minimize the sum of the cumulative distances of all elements on the path. The minimum sum of cumulative distances is the dynamic time regularized distance between the two curves. Iterate through all pairwise combinations of daily load curves within each month, repeatedly perform cumulative distance matrix construction and optimal regularization path search, and arrange all dynamic time regularization distance calculation results according to row and column correspondence to construct a symmetric distance matrix.

4. The method for extracting and applying daily load characteristic sets for energy storage planning and operation according to claim 3, characterized in that, The method uses the dynamic time-normalized distance in the distance matrix as a similarity metric, performs K-medoids clustering on the daily load curves, extracts the curve with the smallest distance sum as the typical daily load characteristic curve, and retains all original curves within each cluster. Specifically, this includes: The optimal number of clusters is determined by the silhouette coefficient, which is calculated based on the ratio of the average dynamic time warp distance between daily load curves within the same cluster to the minimum average dynamic time warp distance between curves in different clusters. Under the optimal number of clusters, the K-medoids clustering algorithm is executed. After randomly selecting the initial medoid, each daily load curve is assigned to the cluster where the nearest medoid is located. The curve with the smallest sum of dynamic time warp distances with all other curves in the cluster is updated as the new medoid. This process is repeated until the medoid no longer changes. The final determined medoid curve of each cluster is used as the cluster center curve, which represents the typical daily load characteristics curve of different date types. The set of all original daily load curves in each cluster is saved.

5. The method for extracting and applying daily load characteristic sets for energy storage planning and operation according to claim 4, characterized in that, Using the cluster center curve as a time-series reference, the original curves within the cluster are dynamically time-normalized and aligned. The maximum and minimum values ​​are taken point-by-point to form upper and lower envelopes. After feature-preserving smoothing, an initial representative curve is constructed, specifically including: Using the cluster center curve as a reference sequence and any original daily load curve within the same cluster as a query sequence, a dynamic time warping cumulative distance matrix is ​​constructed, and the optimal warping path from the starting point to the ending point is searched in the cumulative distance matrix. Based on the optimal regularization path, each sampling point of the query sequence is remapped onto the time axis of the reference sequence, so that the peaks and valleys of each curve and the climbing sections are precisely matched point-to-point in time. On the aligned curve set, the maximum and minimum load values ​​of all curves are taken for each time sampling point to form the upper envelope and the lower envelope. The upper and lower envelopes are lightly smoothed by combining moving window midpoint filtering and local linear interpolation to preserve features, remove high-frequency spikes while retaining the amplitude of extreme points and the slope characteristics; based on the smoothed upper and lower envelopes, the mean curve of the two or the combination of the two is taken as the initial representative curve.

6. The method for extracting and applying daily load characteristic sets for energy storage planning and operation according to claim 5, characterized in that, Physical boundary clipping, rate of change constraint, charge correction and time period regularization are performed on the initial representative curve to generate a general mode representative curve; Simultaneously, extreme value expansion, ramp enhancement, peak-valley alignment, and N-1 upward correction are performed to generate a boundary mode representative curve, specifically including: The initial representative curve is subjected to reasonable upper and lower limit physical boundary clipping to limit the load value to a range that is non-negative and does not exceed 95% of the historical highest load of the load point; the clipped curve is subjected to adjacent time change rate constraint to force the load change rate of adjacent time to not exceed the preset maximum ramp rate, and piecewise linear interpolation is used to repair it when it exceeds the limit. The repaired curve is corrected for power consistency by scaling it according to the ratio of the actual daily total power consumption to the daily total power consumption represented by the curve; the power-corrected curve is then normalized by time period, which is normalized into a stable platform segment and a linear ramp segment according to the power grid dispatching time period, generating a general mode representative curve for production simulation verification. An extreme value expansion operation is performed on the initial representative curve, increasing the peak load upward to 1.05 to 1.20 times the original value and decreasing the valley load downward to 0.80 to 0.95 times the original value; a ramp slope enhancement operation is performed on the expanded curve, increasing the ramp slope along the rising edge and the ramp slope along the falling edge to 1.1 to 1.5 times the original value, respectively. For multiple enhanced curves in the same area, a peak-valley time-shift alignment operation is performed to align the peak and valley times of each curve. After alignment, the loads corresponding to key substations or key lines are adjusted upwards according to a preset ratio for N-1 scenario adaptation, generating a representative curve of the boundary mode for safety and stability analysis.

7. The method for extracting and applying daily load characteristic sets for energy storage planning and operation according to claim 6, characterized in that, The process involves using a subset of general patterns for energy storage configuration and a subset of boundary patterns for production simulation and safety and stability verification. Clustering parameters and boundary coefficients are then corrected in reverse based on the verification deviation, and iterative optimization continues until the verification passes. Specifically, this includes: The subset of daily load characteristics of the general mode is input into the independent energy storage configuration model. The energy storage technology type is determined based on the maximum upslope rate and the maximum downslope rate. The energy storage power scale is determined based on the peak load. The energy storage duration is determined based on the product of the peak load and its duration. The curve shape matching degree of the energy storage configuration scheme is checked based on the daily load curve index. An energy storage configuration scheme is generated. Both the general mode daily load characteristic subset and the boundary mode daily load characteristic subset are input into the production simulation verification model, and the boundary mode daily load characteristic subset is input into the safety and stability verification model to verify the adaptability of the energy storage configuration scheme under normal and extreme operating conditions and the grid safety margin. Based on the deviation information between the production simulation and safety verification results, the number of clusters, the width of the dynamic time warping window, the expansion coefficient, and the ramp-up enhancement coefficient are corrected in reverse. The daily load characteristic set is iteratively optimized until the two-level verification is passed, and then the final energy storage configuration scheme is output.

8. The method for extracting and applying daily load characteristic sets for energy storage planning and operation according to claim 7, characterized in that, Based on the deviation information between the production simulation and safety verification results, the number of clusters, the width of the dynamic time warping window, the expansion coefficient, and the ramp-up enhancement coefficient are corrected in reverse. The daily load characteristic set is iteratively optimized until the two-level verification is passed, and then the final energy storage configuration scheme is output. Specifically, this includes: Obtain the power shortage and power shortage duration output by the production simulation verification model. When the power shortage exceeds the preset threshold and the power shortage period is distributed across multiple discrete time periods, increase the number of clusters. When the power shortage duration exceeds the preset threshold, increase the extreme value expansion coefficient. Obtain the voltage over-limit amplitude and voltage fluctuation frequency output from the safety and stability verification model. When the voltage fluctuation frequency exceeds the preset threshold, relax the dynamic time warping window width constraint; when the voltage over-limit amplitude exceeds the preset threshold, increase the ramp slope enhancement coefficient. The modified number of clusters, dynamic time warping window width, extreme value expansion coefficient, and ramp-up enhancement coefficient are used as new parameters to re-execute the dynamic time warping distance calculation and K-medoids clustering of daily load data, as well as the time-aligned dual-mode envelope boundary processing, to generate an optimized daily load characteristic set. Repeatedly execute the energy storage configuration scheme generation, production simulation verification and safety and stability verification, as well as the iterative process of correcting parameters in reverse according to the verification deviation, until the verification results of both levels meet their respective preset threshold requirements, and output the final energy storage configuration scheme.

9. A system for extracting and applying daily load characteristic sets for energy storage planning and operation, characterized in that, The method for extracting and applying daily load characteristic sets for energy storage planning and operation as described in any one of claims 1-8 includes: The data acquisition and preprocessing module is used to acquire daily load data at minute-level resolution, perform anomaly removal, interpolation preprocessing, and generate daily load curve sequences by daily slices; The temporal morphological clustering module is used to measure the temporal morphological similarity of daily load curves based on dynamic time warping distance. It performs K-medoids clustering with dynamic time warping distance as the index, extracts the cluster centers as typical daily load characteristic curves, and retains the original curve set within each cluster. The dual-mode envelope boundaryization module is used to perform time-series alignment of the original curves within the cluster based on the cluster center, calculate the upper and lower envelopes point by point and construct the initial representative curve while preserving the feature smoothness; perform physical boundary clipping, rate of change constraint, power correction, and time period normalization on the initial representative curve to generate the general mode representative curve, and perform extreme value expansion, ramp enhancement, peak-valley alignment, and N-1 upward correction to generate the boundary mode representative curve; extract load feature values ​​to form a subset of daily load characteristics of the general mode and the boundary mode; The collaborative application and closed-loop correction module for energy storage planning and operation is used to input the general mode subset into the energy storage configuration model to determine the energy storage technology type, power scale and energy storage duration. It inputs the dual mode subset into the production simulation verification model and the boundary mode subset into the safety and stability verification model for verification. Based on the verification deviation, it reversely corrects the number of clusters, the width of the dynamic time warping window, the expansion coefficient and the ramp-up enhancement coefficient, and iterates and optimizes until the two-level verification is passed before outputting the final energy storage configuration scheme.

Images