Energy storage demand analysis method for high-proportion new energy power grid

Abstract

The invention relates to the technical field of electric power energy storage, in particular to an energy storage demand analysis method of a high-proportion new energy power grid, which comprises the following steps of: obtaining a power predicted value and a measured value through a prediction platform to calculate deviation, performing sliding window segmentation, dynamically adjusting length based on volatility to extract variance characteristics, and calculating the energy storage demand of the high-proportion new energy power grid; and the center number is optimized through K-means clustering, a classification result is output, a capacity adjustment value is calculated according to a matching grade adjustment coefficient, time points are extracted and sorted according to a load and output correlation coefficient, and an energy storage capacity space-time distribution table is generated. According to the invention, through prediction of deviation sequence sliding window segmentation and variance feature extraction, dynamic identification of new energy output fluctuation intensity, construction of a deviation fluctuation and energy storage capacity mapping model, quantification of load and output coupling intensity, and construction of a multi-dimensional energy storage correction system, energy storage and source load dynamic matching is realized, and response sensitivity is improved. The risk of resource mismatching is reduced, the robustness of the system is enhanced, and the cooperative efficiency of charging and discharging strategies is optimized.

Description

Technical Field

[0001] The present invention relates to the field of electric energy storage technology, and in particular to a method for analyzing energy storage demand of a high-proportion new energy power grid. Background Art

[0002] The field of electric energy storage technology involves the entire process of storing and releasing electric energy using physical and electrochemical methods. It mainly includes the selection and configuration of energy storage systems, characteristic analysis of electric energy storage media, grid operation status perception and prediction, electric energy charging and discharging scheduling strategies, and dynamic response mechanisms of power systems. This technology is widely used in scenarios such as new energy grid integration, power peak and frequency regulation, power quality management, and grid accident response. It is of great significance in supporting the access of a high proportion of renewable energy and ensuring the safe and stable operation of the power grid. With the rapid development of intermittent new energy sources such as wind power and photovoltaics, the characteristics of the power grid that the source follows the load and the load follows the source are becoming more and more prominent. This has led to increased spatiotemporal uncertainty in the operation of the power system and a multi-dimensional increase in the demand for energy storage. In this context, conducting energy storage configuration demand analysis for power grids with a high proportion of new energy has become an important part of power system planning and regulation.

[0003] Traditional energy storage demand analysis methods assess the deviation between renewable energy output fluctuations and load variations in the power system through methods such as comparing raw grid operation data, overlaying and analyzing typical daily load curves, and matching installed capacity with load gaps based on static indicators. This approach then determines the basic requirements for energy storage capacity and operational strategies. These methods rely on categorizing and statistically analyzing typical daily loads and outputs, and determining energy storage charging and discharging time periods and power requirements through offline data analysis. They lack an assessment mechanism capable of responding to renewable energy fluctuations in real time, and they are unable to fully consider the coupling effects of differentiated renewable energy output forms and load characteristics on different time scales.

[0004] Traditional methods rely on the superposition of typical daily load curves and offline statistics of original data, and use fixed time windows for energy storage demand analysis, ignoring the temporal fluctuation characteristics of renewable energy output forecast deviations and the spatiotemporal heterogeneity of source-load coupling relationships. The static capacity matching mechanism is difficult to match the rapid fluctuations of renewable energy output and the dynamic changes in load demand, resulting in systematic deviations between energy storage capacity planning and the actual grid operation status. When the cumulative effect of forecast deviations is significant or the source-load correlation suddenly changes, it is easy to cause mismatch of energy storage charging and discharging strategies, resulting in frequency regulation response delays and insufficient emergency backup capacity, which restricts the stable operation capability of power grids with a high proportion of renewable energy. Summary of the Invention

[0005] In order to solve the problem that the traditional method relies on the superposition of typical daily load curves and offline statistics of original data, uses a fixed time window for energy storage demand analysis, ignores the temporal fluctuation characteristics of the new energy output forecast deviation and the spatiotemporal heterogeneity of the source-load coupling relationship, and the static capacity matching mechanism is difficult to match the rapid fluctuation of new energy output and the dynamic change of load demand, resulting in a systematic deviation between the energy storage capacity planning and the actual grid operation status. When the cumulative effect of the forecast deviation is significant or the source-load correlation suddenly changes, it is easy to cause the energy storage charging and discharging strategy mismatch, resulting in frequency regulation response delay and insufficient emergency backup capacity, which restricts the stable operation capability of the high-proportion new energy power grid. The embodiment of the present invention provides an energy storage demand analysis method for a high-proportion new energy power grid. The technical solution is as follows:

[0006] On the one hand, a method for analyzing energy storage demand of a high-proportion new energy power grid is provided, the method comprising:

[0007] S1: Obtain the power forecast value sequence at the time point in the current forecast period through the new energy output forecast platform. Based on the measured power value sequence at the corresponding time point in the original database, perform point-by-point offset calculation to generate a forecast deviation sequence.

[0008] S2: Input the predicted deviation sequence into the sliding window algorithm to split the time series window, and dynamically adjust the window length based on the volatility of the absolute value of the deviation, calculate the variance of the absolute value of the deviation in the window, and generate a variance feature set;

[0009] S3: Based on the variance feature set, a K-means clustering algorithm is used to optimize the number of cluster centers by using the silhouette coefficient, clustering is performed using the variance as the feature vector, and a window classification result set is output;

[0010] S4: According to the intensity level labels in the window classification result set, matching the level-capacity adjustment coefficient, performing a proportional adjustment operation on the energy storage capacity segment, and generating a time-divided capacity adjustment value;

[0011] S5: extracting the time points in the time-divided capacity adjustment values, calculating the Pearson correlation coefficients between the load and the new energy output at the corresponding time points, sorting the time-divided capacity adjustment values ​​according to the correlation coefficients, and generating a spatiotemporal distribution table of energy storage capacity.

[0012] As a further solution of the present invention, the predicted deviation sequence includes a timestamp deviation value, a deviation direction attribute, and a cumulative deviation amount; the variance feature set includes a window variance value, a self-matching window length, and a fluctuation intensity level; the window classification result set includes a cluster label, cluster center coordinates, and an intra-class variance value; the time period capacity adjustment value includes a time period benchmark capacity, a level adjustment coefficient, and a fluctuation intensity-capacity mapping relationship; and the energy storage capacity spatiotemporal distribution table includes a time priority index, a Pearson correlation coefficient value, and spatiotemporal distribution coordinate points;

[0013] The self-matching window length refers to the window time span parameter that is dynamically adjusted according to the deviation volatility;

[0014] The level adjustment coefficient refers to a capacity adjustment ratio parameter based on clustering strength level matching.

[0015] As a further solution of the present invention, the specific steps of S1 include:

[0016] S101: Obtain a power prediction value sequence at a time point in the current prediction period in the new energy output prediction platform, call a measured power value sequence at a corresponding time point in the original database, calculate the numerical offset between the two at the time point, and generate a power offset sequence;

[0017] S102: Based on the power offset sequence, calculate the absolute value of the offset at each time point, use the predicted value as a reference value to perform item-by-item ratio calculation, organize the data sequence according to a unified time index format, and obtain a corresponding offset rate sequence;

[0018] S103: calling the corresponding offset rate sequence, calculating the offset degrees of the offset rates of adjacent time points in time index order, and performing linear sorting processing on the offset degrees to obtain a prediction deviation sequence.

[0019] As a further solution of the present invention, the specific steps of S2 include:

[0020] S201: Based on the predicted deviation sequence, a sliding window algorithm is used to set an initial window length and slide the position in sequence, extract the deviation values ​​in the window and take the absolute values ​​to form a set, thereby generating a deviation absolute value sequence distribution set;

[0021] S202: calling the deviation absolute value sequence distribution set, calculating the difference between the maximum and minimum values ​​in the window as the window fluctuation amplitude, comparing it with the set deviation fluctuation benchmark threshold, and adjusting the window length to obtain a dynamic window length sequence;

[0022] The deviation fluctuation reference threshold is the average value of the absolute value of the deviation plus or minus the empirical coefficient k times the standard deviation;

[0023] S203: Based on the time series window after the dynamic window length sequence is adjusted, a set of absolute values ​​of deviations within the window is called, variance values ​​are calculated item by item, and their fluctuation levels are recorded to generate a variance feature set.

[0024] As a further solution of the present invention, the specific steps of S3 include:

[0025] S301: Based on the variance feature set, each window is fitted into a feature vector including the fluctuation amplitude, the window variance value and the self-matching length, input into the K-means clustering model, set multiple cluster centers to perform clustering operation, calculate the window silhouette coefficient under each group of clusters, and generate a window silhouette coefficient value set;

[0026] The window silhouette coefficient is an indicator that measures the degree of cohesion and separation of each window in the cluster;

[0027] S302: calling the window silhouette coefficient value set, calculating the mean of the silhouette coefficient under the number of cluster centers, extracting the mean sequence and comparing the values ​​thereof, selecting the center number corresponding to the maximum value as the K-means input parameter, re-executing clustering, and obtaining the optimal cluster label number group;

[0028] S303: According to the optimal clustering label number group, the average variance characteristics and fluctuation level of the representative window are extracted, a "window intensity level-variance characteristic" dynamic matching model is constructed, a mapping table of category labels and fluctuation intensity levels is established, and a window classification result set is obtained.

[0029] As a further solution of the present invention, the mean value of the silhouette coefficient is calculated using the formula:

[0030] ;

[0031] in, The number of representative cluster centers is The mean of the window silhouette coefficient when , Represents the current number of cluster centers The total number of windows under Representative The number of windows in the cluster center is The silhouette coefficient value under Representative The number of windows in the cluster center is The corresponding window variance value is Representative The number of windows in the cluster center is The corresponding fluctuation amplitude value is Representative The number of windows in the cluster center is The average of the fluctuation amplitude values ​​in the lower window.

[0032] As a further solution of the present invention, the specific steps of S4 include:

[0033] S401: Obtaining the intensity level label of each window based on the window classification result set, and dynamically extracting the capacity adjustment coefficient that matches the fluctuation intensity through a preset level-capacity adjustment parameter mapping table to generate a label adjustment coefficient;

[0034] The label adjustment coefficient is based on the window fluctuation intensity level and the capacity adjustment factor dynamically extracted through the mapping table to proportionally correct the energy storage capacity data;

[0035] S402: Based on the label adjustment coefficient, perform a multiplication ratio conversion on the original data of the energy storage capacity in the window, and apply the adjustment coefficient to the corresponding capacity value according to the energy storage capacity value in the time point sequence to obtain the time increment number;

[0036] S403: Performing a data weighted operation on the energy storage capacity at a time point according to the moment increment number, calculating an adjustment value based on the original capacity at each moment and its corresponding change, replacing the full-segment capacity data, and generating a time-divided capacity adjustment value.

[0037] As a further solution of the present invention, the time increment number is calculated using the formula:

[0038] ;

[0039] in, Represents at a point in time The capacity increment after adjustment is in MWh. Represents at a point in time The original energy storage capacity value, in MWh, Represents the window The capacity adjustment factor extracted from the intensity level label corresponds to Representative time point The original capacity value, in MWh, Representative Window The average of all original capacity values ​​in MWh, Representative Window The capacity change trend adjustment factor extracted by regression prediction is in MWh. Representative time point The corresponding capacity fluctuation error compensation value is in MWh. Represents the current window The number of time points included in Represents the time point number currently being processed. is a dynamic weight factor, whose value is calculated by the energy storage controller according to the real-time grid frequency deviation.

[0040] As a further solution of the present invention, the specific steps of S5 include:

[0041] S501: extracting time points based on the time-based capacity adjustment values, performing normalization processing according to the time window and calculating linear correlation based on the load and new energy output data, and generating a time point correlation coefficient sequence;

[0042] S502: Calling the time point correlation coefficient sequence, setting a position mapping rule in descending order of correlation coefficient values, calculating the capacity adjustment sequence difference, reorganizing the original time period capacity adjustment values ​​according to the index to generate a new value arrangement, and obtaining the capacity adjustment sequence ranking value;

[0043] The capacity adjustment sequence ranking value is the result of re-ranking the original capacity adjustment data according to the degree of coupling between load and renewable energy output, reflecting the priority and value of the time point in energy storage regulation;

[0044] S503: Matching the geographic location code according to the capacity adjustment sequence ranking value, establishing a ranking value and spatial coordinate mapping table, and calculating the regional mean and dispersion to generate a spatiotemporal distribution table of energy storage capacity.

[0045] As a further solution of the present invention, the capacity adjustment sequence difference is calculated using the formula:

[0046] ;

[0047] in, Representative The capacity adjustment sequence difference corresponding to each position is in MW. Represents the original At the time point The capacity adjustment value of each location, in MW, Representative Load data at each time point, in MW, Representative The new energy output offset value corresponding to each time point, in MW, Indicates the total number of time points, Indicates the The capacity adjustment value corresponding to each position is the mean value linearly normalized to [0, 1]. Represents the weight coefficient, for load data Offset with new energy The dimensions of are normalized.

[0048] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0049] Based on the sliding window segmentation and variance feature extraction mechanism of the prediction deviation sequence, the distribution law of the fluctuation intensity of renewable energy output is dynamically identified. A mapping model between the deviation fluctuation characteristics and the energy storage capacity adjustment coefficient is established through the clustering algorithm. The spatiotemporal coupling intensity of load and renewable energy output is quantified by combining the Pearson correlation coefficient. A multi-dimensional energy storage capacity dynamic correction system is constructed to achieve accurate mapping of the relationship between energy storage configuration and the dynamic matching relationship between grid source and load, improve the response sensitivity of the energy storage system to the diversity of renewable energy output forms and random load fluctuations, reduce the risk of energy storage resource mismatch caused by planning deviations, enhance the robustness of the power system to the cumulative effect of prediction deviations, and optimize the spatiotemporal coordination efficiency of energy storage charging and discharging strategies in frequency regulation scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 Schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0052] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0053] In the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same. The terms "of," "corresponding," and "corresponding" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same.

[0054] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0055] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0056] See also Figure 1 The embodiment of the present invention provides a method for analyzing energy storage demand of a high-proportion new energy power grid. The processing flow of the method may include the following steps:

[0057] S1: Obtain the power forecast value sequence at the time point in the current forecast period through the new energy output forecast platform. Based on the measured power value sequence at the corresponding time point in the original database, perform point-by-point offset calculation to generate a forecast deviation sequence.

[0058] S2: Input the prediction deviation sequence into the sliding window algorithm to split the time series window. Based on the volatility of the absolute value of the deviation, the window length is dynamically adjusted. The variance of the absolute value of the deviation within the window is calculated to generate the variance feature set.

[0059] S3: Based on the variance feature set, the K-means clustering algorithm is used to optimize the number of cluster centers through the silhouette coefficient, clustering is performed with variance as the feature vector, and the window classification result set is output;

[0060] S4: Based on the intensity level labels in the window classification result set, match the level-capacity adjustment coefficient, perform proportional adjustment operations on the energy storage capacity segments, and generate time-based capacity adjustment values;

[0061] S5: Extract the time points in the time-based capacity adjustment values, calculate the Pearson correlation coefficient between the load and the new energy output at the corresponding time points, sort the time-based capacity adjustment values ​​according to the correlation coefficient, and generate a spatiotemporal distribution table of energy storage capacity.

[0062] The prediction deviation sequence includes the timestamp deviation value, deviation direction attribute, and cumulative deviation amount. The variance feature set includes the window variance value, self-matching window length, and fluctuation intensity level. The window classification result set includes the cluster label, cluster center coordinates, and intra-class variance value. The time period capacity adjustment value includes the time period benchmark capacity, level adjustment coefficient, and fluctuation intensity-capacity mapping relationship. The energy storage capacity spatiotemporal distribution table includes the time priority index, Pearson correlation coefficient value, and spatiotemporal distribution coordinate points.

[0063] The self-matching window length refers to the window time span parameter that is dynamically adjusted according to the deviation volatility;

[0064] The level adjustment coefficient refers to the capacity adjustment ratio parameter based on clustering intensity level matching.

[0065] Specifically, the steps of S1 are:

[0066] S101: Obtain a power prediction value sequence at a time point in the current prediction period in the new energy output prediction platform, call a measured power value sequence at a corresponding time point in the original database, calculate the numerical offset between the two at the time point, and generate a power offset sequence;

[0067] In the process of obtaining the power forecast value sequence within the current forecast period in the new energy output forecast platform, it is necessary to first obtain the wind power or photovoltaic output forecast data for every 5 minutes for a period of time in the future (such as 24 hours) through the forecast platform of the dispatching control center. This data is calculated by multiple forecast models and then obtained by weighted average. For example, for photovoltaic power station A, the power sequence output by the forecast system is: , (unit: MW, corresponding to the times 8:00, 8:05, 8:10, 8:15, 8:20, 8:25), then call the measured power data sequence in the original database that is exactly the same as the predicted time point, and obtain the corresponding actual output sequence by calling the data recorded by the SCADA system deployed on site of power station A and accurately matching the timestamps, such as: , (the unit is MW), and then perform the numerical offset calculation operation between the two, specifically: the predicted value and the measured value at the same time point are subtracted correspondingly to obtain the power offset sequence, for example, the offset at the first moment is MW, the offset at the second moment is MW, and so on to obtain the sequence: , where the positive or negative value of each item indicates that the predicted value is higher or lower than the measured value. The complete power offset sequence is constructed in the above way. In this process, if there are missing values ​​in the predicted value or the measured value (such as equipment failure or communication interruption), it is necessary to perform linear interpolation to fill in the missing values ​​with the weighted average of the adjacent time points before and after the moment. The weighted calculation method uses the values ​​of the previous and next time points multiplied by 0.4 and 0.6 respectively and then added to form the filling term. For example, if the predicted value at 8:10 is missing, the two predicted values ​​of 9.1 and 11.0 at 8:05 and 8:15 are used to generate the filling value. MW is substituted into the offset calculation. For scenarios where power offset is abnormal, the system sets the offset threshold to ±2.0MW. If the offset at a certain moment exceeds this range, the moment is marked as a prediction inaccuracy moment.

[0068] Table 1: Deviation between predicted and measured values

[0069] As shown in Table 1, by comparing the predicted values ​​with the measured values, the obtained offsets did not exceed the judgment threshold of ±2.0MW. Therefore, the prediction accuracy for this time period was within an acceptable range. The subsequent steps can continue to analyze the offset rate and its degree of change.

[0070] S102: Based on the power offset sequence, calculate the absolute value of the offset at each time point, use the predicted value as a reference value to perform item-by-item ratio calculation, organize the data sequence according to a unified time index format, and obtain the corresponding offset rate sequence;

[0071] In the process of calculating the absolute value of the offset at each time point and generating the corresponding offset rate sequence, it is first necessary to obtain the power offset sequence Taking the absolute value of each item, we get , and then compare the value with the predicted power series Calculate the corresponding offset rate by taking the absolute value of the offset as the numerator and the predicted power at the corresponding moment as the denominator, and perform a division operation. For example, the corresponding offset rate at the first time point is , the corresponding offset rate at the second time point is , and so on, we can finally get the corresponding offset rate sequence , which is a unified time index format. Timestamp information can be added to the sequence to generate an indexed time series. In this process, for moments with a predicted value of zero, since the corresponding offset rate cannot be calculated, a judgment logic should be set to directly mark the moment as an "invalid time point". If this happens, the record of this time point needs to be eliminated in subsequent analysis. In this embodiment, the predicted values ​​of the moments are all non-zero, so no additional processing is required. The final corresponding offset rate sequence will serve as the basic data for subsequent offset rate deviation analysis.

[0072] S103: Call the corresponding offset rate sequence, calculate the offset degree of the offset rate of adjacent time points in time index order, and linearly sort the offset degrees to obtain a prediction deviation sequence;

[0073] In the process of calculating the offset degree of the offset rate at adjacent time points and generating the prediction deviation sequence, the corresponding offset rate sequence generated needs to be Execute adjacent difference calculation item by item in the order of time index, specifically: subtract each two adjacent items and process the result as absolute value. For example: the offset rate of the first adjacent time point is , the offset rate of the second adjacent time point is , and so on, we get the offset rate offset sequence , and then sort the above sequence in ascending order according to the linear sorting logic, and the final sorting result is , this sequence represents the temporal variation of the prediction deviation. The purpose of sorting is to provide a reference for subsequent statistical analysis. For example, if the offset at a certain moment corresponds to the extreme value after sorting, it indicates that the prediction model has a large deviation at this moment, and further evaluation of the associated external factors or model parameters is required.

[0074] Specifically, the steps of S2 are:

[0075] S201: Based on the prediction deviation sequence, a sliding window algorithm is used to set the initial window length and slide the position in sequence, extract the deviation values ​​in the window and take the absolute values ​​to form a set, and generate a deviation absolute value sequence distribution set;

[0076] Based on the forecast deviation series First, we need to determine the initial window length of the sliding window algorithm. Assuming that the sequence is recorded at equal time intervals, each data interval is 5 minutes, and the initial window length is selected as 3, which means that each analysis consists of 3 consecutive time points to form a window. The first window position is the 1st to 3rd data, and the corresponding sequence is , process each data value by absolute value to obtain , record this set and use it as the deviation absolute value set of the first window, then perform the window sliding operation, that is, the starting position is shifted to the right by one time point to form the second window , and perform absolute value processing again to obtain The process continues sliding until the complete window is enumerated, and finally multiple deviation absolute value sets are formed, which are the first window , second window , the third window , a total of 3 sets of data sets, as shown in Table 2,

[0077] Table 2: Distribution of absolute deviation sequence

[0078] As shown in Table 2, each set of deviation absolute value sets corresponds to the time period covered by the sliding window. The set is used for subsequent fluctuation amplitude analysis. By enumerating the deviation absolute value sets, the coverage and statistics of the prediction error in the differentiated time window are achieved.

[0079] S202: Call the deviation absolute value sequence distribution set, calculate the difference between the maximum and minimum values ​​in the window, use it as the window fluctuation amplitude, compare it with the set deviation fluctuation benchmark threshold, and adjust the window length to obtain a dynamic window length sequence;

[0080] The deviation fluctuation benchmark threshold is the average value of the absolute value of the deviation plus or minus the empirical coefficient k times the standard deviation;

[0081] Call the above deviation absolute value set, perform the difference operation between the maximum and minimum values ​​of the data in each set, and calculate the fluctuation range in the current window. The specific operation is: the maximum value in the first window is 0.0273, and the minimum value is 0.0081. Subtracting them, the fluctuation range is , the maximum value in the second window is 0.0402, and the minimum value is 0.0268. Subtracting them, we get , the maximum value in the third window is 0.0637, and the minimum value is 0.0273. Subtracting them, we get Then the fluctuation amplitude is compared with the set deviation fluctuation benchmark threshold, which is calculated by predicting the deviation sequence. Calculate its mean and standard deviation, the mean is:

[0082] ;

[0083] The standard deviation is:

[0084] ;

[0085] Set the experience coefficient , then the deviation fluctuation benchmark threshold is:

[0086] Threshold ;

[0087] The fluctuation range Comparing each item with the threshold of 0.0510, none of them exceeds the threshold. According to the preset logic, when the fluctuation amplitude is lower than the threshold, the current window length is kept unchanged. If it exceeds, the window length is increased from the original value of 3 to 5. In this example, the window fluctuations are all lower than the threshold, so the dynamic window length sequence is , this sequence provides a reference for setting the timing window for subsequent variance calculation.

[0088] S203: Based on the time series window adjusted by the dynamic window length sequence, call the set of absolute values ​​of deviations in the window, calculate the variance value item by item and record its fluctuation degree to generate a variance feature set;

[0089] Based on the dynamic window length sequence obtained above , continue to call the deviation absolute value set generated by the previous sliding 、 、 , perform variance calculation on each data item in each set. The calculation method is to first find the mean of the set, then calculate the square difference between each item and the mean, and then take the average. The specific process is as follows:

[0090] For the first group , the mean is:

[0091] ;

[0092] The variance is:

[0093] ;

[0094] Group 2 The mean is , the variance is:

[0095] ;

[0096] Group 3 The mean is , the variance is:

[0097] ;

[0098] The final generated variance feature set is ,This sequence reflects the discrete degree of prediction deviation within the ,differentiated time window, which serves as the basis for subsequent anomaly ,identification or dynamic correction.

[0099] Specifically, the steps of S3 are:

[0100] S301: Based on the variance feature set, each window is fitted into a feature vector including the fluctuation amplitude, the window variance value, and the self-matching length, and the feature vector is input into the K-means clustering model. A clustering operation is performed by setting multiple cluster centers, and the window silhouette coefficient under each cluster group is calculated to generate a window silhouette coefficient value set;

[0101] The window silhouette coefficient is a measure of how cohesive and separate each window is in the cluster;

[0102] Based on the variance feature set, the original time series data needs to be divided into sliding windows. The length of each window is initially set to 30 time steps. For each window, its internal time series data is called, and the fluctuation amplitude value is obtained by calculating the difference between the maximum and minimum values ​​point by point. The result is recorded as the fluctuation amplitude index of each window. For example, the maximum value in the first window is 24.6 and the minimum value is 15.2, then the fluctuation amplitude is 9.4. Then, the variance formula is used to calculate the square average of the deviation between the value in the window and its mean to obtain the variance value of the window. The window length is adjusted from the initial 30 steps to 60, 90, and 120 steps. By comparing the variance volatility under different window lengths, the average variance growth rate does not exceed 15% as the stopping criterion, and the current optimal window length is selected. Finally, a feature vector including three dimensions of fluctuation amplitude, window variance value, and self-matching length is formed. After generating the above features for the window, it is input into the K-means clustering in matrix form. When executing K-means, the number of cluster centers is set separately. The clustering operation is performed one by one from 2 to 8. After each clustering operation, the silhouette coefficient is calculated by calling the intra-cluster tightness and inter-cluster separation degree of each window. , such as the first window in Under this condition, the average distance within the cluster is 1.2, and the average distance of the nearest neighbor cluster is 3.8, so the silhouette coefficient is , place the window in this The silhouette coefficient values ​​under the cluster center are stored in the window silhouette coefficient value set, and the process is repeated to obtain the window silhouette coefficient sequence results under multiple cluster centers.

[0103] S302: Calling the window silhouette coefficient value set, calculating the mean of the silhouette coefficient under the number of cluster centers, extracting the mean sequence and comparing its numerical values, selecting the center number corresponding to the maximum value as the K-means input parameter, re-executing clustering, and obtaining the optimal cluster label number group;

[0104] Call the window silhouette coefficient value set and calculate the number of cluster centers for each cluster The window silhouette coefficient sequence under the following conditions is used to calculate the silhouette coefficient mean, using the formula:

[0105] ;

[0106] in, The number of representative cluster centers is The mean of the window silhouette coefficient when , Represents the current number of cluster centers The total number of windows under Representative The number of windows in the cluster center is The silhouette coefficient value under Representative The number of windows in the cluster center is The corresponding window variance value is Representative The number of windows in the cluster center is The corresponding fluctuation amplitude value is Representative The number of windows in the cluster center is The average of the fluctuation amplitude values ​​in the lower window.

[0107] The specific method is: first call the current The silhouette coefficient value of the window under , corresponding window variance value , Fluctuation amplitude value And the average fluctuation amplitude of the window under clustering , then perform the following operations: one by one arrive Window, calculate the silhouette coefficient value, sum the above values, and then divide by get , for example, setting , the number of windows is 5, and the sampling window data is shown in the following table:

[0108] Table 3: Input table for calculating the mean value of the silhouette coefficient

[0109] As shown in Table 3, substitute the values ​​into the formula and calculate item by item.

[0110] Item 1 is: ;

[0111] The rest of the items are deduced in the same way. The last five items are summed up and the average is obtained to get the mean of the silhouette coefficient. The final result is 1.535+1.657+1.210+1.198+1.386=6.986, and the average is 6.986 / 5=1.397. Next ,choose The maximum corresponding As the optimal number of cluster centers, the clustering process is re-executed to obtain the window label number group under the optimal k value to form the final window label set.

[0112] S303: Extract the average variance characteristics and fluctuation level of the representative window based on the optimal cluster label number group, build a "window intensity level-variance characteristic" dynamic matching model, establish a mapping table between category labels and fluctuation intensity levels, and obtain a window classification result set;

[0113] According to the optimal cluster label number group obtained in the previous step, the window is divided into differentiated categories according to the cluster label. For each cluster class, the variance value of the window is extracted in turn, and its average value is calculated as the representative variance feature of the class. For example, the window variance value in a certain class is [2.4, 3.1, 2.9, 3.0], and its average is (2.4+3.1+2.9+3.0) / 4=2.85. The fluctuation amplitude value in each class window is further extracted and its average value is calculated as the basis for the fluctuation intensity level division. The corresponding "variance-fluctuation level" is generated for each cluster class. " feature pairs, build a dynamic matching table, for example, set the fluctuation level thresholds as: low intensity (0-3), medium intensity (3-7), high intensity (>7), if the average fluctuation amplitude of a cluster is 6.2, it is classified as medium intensity level. By matching the cluster label with the level, establish the final label-intensity level mapping table, which is used to output the final window classification result set. The classification result set includes three dimensions: window number, classification label, and fluctuation level. For example, the classification label of window 5 is 2, corresponding to the medium intensity level, and the final output result is (window 5, category 2, medium intensity).

[0114] Specifically, the steps of S4 are:

[0115] S401: Obtain the intensity level label of each window based on the window classification result set, and dynamically extract the capacity adjustment coefficient that matches the fluctuation intensity through a preset level-capacity adjustment parameter mapping table to generate a label adjustment coefficient;

[0116] The tag adjustment coefficient is based on the window fluctuation intensity level and the capacity adjustment factor dynamically extracted through the mapping table to proportionally correct the energy storage capacity data;

[0117] According to the window classification result set, we first need to read the intensity level label of each window one by one. The label is an integer value generated in the preset classification range, ranging from 1 to 5, corresponding to the fluctuation intensity of "very low", "low", "medium", "high" and "very high" respectively. The specific judgment is based on the standard deviation of the energy storage capacity data in the current window. If the standard deviation is between 01, it is divided into "very low" (level 1), 12 is "low" (level 2), 23 is "medium" (level 3), 34 is "high" (level 4), and above 4 is "very high" (level 5). Based on this classification basis, the original capacity sequence is read window by window, the standard deviation of the sequence is calculated, and it is classified into the corresponding level interval. Taking a window containing 10 time points as an example, assuming that the original capacity sequence is , its standard deviation is about 0.8, so it is classified as level 1; next, according to the extracted level number, the corresponding adjustment coefficient is retrieved from the level-capacity adjustment parameter mapping table. The table is a pre-set parameter mapping structure, as shown in Table 4. The capacity adjustment factor in the table is a real value, which serves as the proportional factor for subsequent capacity correction; according to the level number, the corresponding adjustment coefficient in the mapping table is extracted. If the current window strength level is 1, the corresponding coefficient is 0.85, which is stored as the label adjustment coefficient in the corresponding window data structure to execute the next stage of processing.

[0118] Table 4: Level-capacity adjustment parameter table

[0119] As shown in Table 4, the capacity adjustment coefficient increases with the increase of the fluctuation level, reflecting the actual need to amplify and correct the capacity fluctuation amplitude when the fluctuation is high.

[0120] S402: Based on the label adjustment coefficient, perform a multiplication ratio conversion on the original data of the energy storage capacity in the window, and apply the adjustment coefficient to the corresponding capacity value according to the energy storage capacity value in the time point sequence to obtain the time increment number;

[0121] Based on the label adjustment coefficient obtained in the previous stage , the original data of energy storage capacity corresponding to the time point in the window is required Perform multiplication conversion to obtain the preliminary correction value. To do this, it is necessary to combine the current time point sequence, such as the time point set is , the original capacity data is , where the capacity value at each time point is in MWh. The increment at each time point is calculated using the formula:

[0122] ;

[0123] in, Represents at a point in time The capacity increment after adjustment is in MWh. Represents at a point in time The original energy storage capacity value, in MWh, Represents the window The capacity adjustment factor extracted from the intensity level label corresponds to Representative time point The original capacity value, in MWh, Representative Window The average of all original capacity values ​​in MWh, Representative Window The capacity change trend adjustment factor extracted by regression prediction is in MWh. Representative time point The corresponding capacity fluctuation error compensation value is in MWh. Represents the current window The number of time points included in Represents the time point number currently being processed. is a dynamic weight factor, whose value is calculated by the energy storage controller according to the real-time grid frequency deviation.

[0124] First calculate the window mean , then calculate the square of the difference between the capacity and the mean at each time point in the window, sum them up and take the square root to obtain the normalized term. The specific steps are as follows:

[0125] ;

[0126] ;

[0127] If the adjustment coefficient extracted from the current window is , and set the window capacity change trend adjustment factor (obtained by the approximate linear regression slope), and the capacity fluctuation error compensation value at each time point is set to , substitute the above parameters into the formula to calculate the capacity increment at the first time point:

[0128] ;

[0129] in, is the dynamic weight factor. Assuming the grid frequency deviation is 0.012Hz, the corresponding ,but:

[0130] ;

[0131] Similarly, repeat the process for the remaining time points to obtain the corresponding incremental set , as the final incremental output.

[0132] S403: Performing a data weighting operation on the energy storage capacity at a given point in time based on the moment increments, calculating an adjustment value based on the original capacity at each moment and its corresponding change, replacing the full-segment capacity data to generate a time-divided capacity adjustment value;

[0133] Based on the capacity increments at each time point calculated above , it is necessary to perform the capacity data replacement operation point by point in the time series to achieve time-based capacity adjustment. The operation process takes the original capacity at each moment in turn. The corresponding increment value , using a simple linear weighted model, the updated value is calculated as , if the original capacity data is:

[0134] ;

[0135] The first value of the corresponding incremental value is 31.35MWh, so the corrected capacity data is The remaining data items are calculated in sequence according to the above method to complete the data update of the entire segment. The larger the capacity increment, the stronger the fluctuation or the more significant the trend change at the current time point, so the adjustment amplitude also increases accordingly. If the original capacity at point t is 53.5MWh and the corresponding increment is 28.5MWh, then the adjusted capacity is 82.0MWh. The adjusted capacity sequence of the entire segment is generated in sequence as the basis for subsequent prediction or scheduling.

[0136] Specifically, the steps of S5 are:

[0137] S501: Extracting time points based on the time-based capacity adjustment values, performing normalization processing on the load and renewable energy output data according to the time window, and calculating linear correlation to generate a time point correlation coefficient sequence;

[0138] Based on the time point extraction of capacity adjustment value by time period, in actual operation, typical load cycles such as 0:00, 6:00, 12:00, 18:00 and other key control time points can be selected at the hourly level. The corresponding load value and new energy output value are collected for each time point and recorded as vector sequences. The sampling period is 1 hour. If the total number of time points is 8, then there is a capacity adjustment value array , load data array , new energy output value , further by performing difference processing on the new energy output value sequence, we can obtain the new energy offset array ,in is the average renewable energy output value in this period, and then the load and renewable energy output data are standardized respectively, using linear normalization processing method, for any data series , after normalization:

[0139] ;

[0140] For example, when the load data is [240, 250, 260, 270, 280, 275, 265, 255] MW, the maximum value is 280 MW, and the minimum value is 240 MW, then the normalized value at the first time point is , the sixth time point value is After repeating the above normalization process for the new energy data, the Pearson correlation coefficient is used to calculate the linear correlation. That is, for each pair of time points (i, j), the correlation coefficient is calculated based on the normalized load and the new energy output value:

[0141] ;

[0142] Finally, a time point correlation coefficient sequence is formed. The absolute value of the correlation coefficient is set to be greater than 0.75 as a strong correlation interval, 0.5 to 0.75 as a moderate correlation interval, and less than 0.5 as a weak correlation interval. The sequence can be illustrated as follows: [0.82, 0.65, 0.78, 0.49, 0.34, 0.72, 0.85, 0.41], indicating that time points 1, 3, and 7 are strongly correlated. The ranking value will be used for subsequent capacity adjustment and reconstruction processing.

[0143] S502: Calling the time point correlation coefficient sequence, setting the position mapping rule in descending order of the correlation coefficient values, calculating the capacity adjustment sequence difference, reorganizing the original time period capacity adjustment values ​​according to the index to generate a new value arrangement, and obtaining the capacity adjustment sequence ranking value;

[0144] The capacity adjustment sequence ranking value is the result of re-sorting the original capacity adjustment data according to the degree of coupling between load and renewable energy output, reflecting the priority and value of the time point in energy storage regulation;

[0145] Call the time point correlation coefficient sequence, sort the correlation coefficient values ​​of each time point in descending order, and use this to set the mapping rule of the original capacity adjustment value in the new position. For example, bind each time point in the correlation coefficient sequence [0.85, 0.82, 0.78, 0.72, 0.65, 0.49, 0.41, 0.34] to its original capacity adjustment value, establish the position mapping index [7, 1, 3, 6, 2, 4, 8, 5], and the original capacity The adjusted values ​​are reassembled according to the index. Suppose the original capacity value array is [C1=10, C2=12, C3=14, C4=11, C5=13, C6=15, C7=9, C8=16], and after mapping, it is the new array [C7=9, C1=10, C3=14, C6=15, C2=12, C4=11, C8=16, C5=13]. Then calculate the difference of the capacity adjustment sequence at each new position using the following formula:

[0146] ;

[0147] in, Representative The capacity adjustment sequence difference corresponding to each position is in MW. Represents the original At the time point The capacity adjustment value of each location, in MW, Representative Load data at each time point, in MW, Representative The new energy output offset value corresponding to each time point, in MW, Indicates the total number of time points, Indicates the The capacity adjustment value corresponding to each position is the mean value linearly normalized to [0, 1]. Represents the weight coefficient, for load data Offset with new energy The dimensions of are normalized.

[0148] In the formula is the total number of time points, set to 8, Represents the weight coefficient, which is used to adjust the corresponding criticality of the load and the output offset of new energy. It can be set to 0.6 (load dominates) based on differentiated application scenarios. is the capacity adjustment value of the original i-th time point at the new location j, and They are respectively derived from the normalized load value and new energy offset value in the previous paragraph. is the normalized average value of the capacity adjustment value at the new position j. The normalization processing method is the same as the previous paragraph. Now take the j=1 position as an example. If C is 9MW, the normalized capacity adjustment average value is , load value W=240MW, normalized to 0, new energy offset , after normalization it is 0.25, then:

[0149] ;

[0150] ;

[0151] ;

[0152] Repeat the calculation for each position , we can get the complete capacity adjustment difference array, and the sorting value is the order of the new values ​​corresponding to the original data in the sequence, which represents its response to the coupling degree between load and renewable energy output.

[0153] S503: Matching geographic location codes according to the capacity adjustment sequence ranking values, establishing a ranking value and spatial coordinate mapping table, and calculating regional mean and dispersion to generate a spatiotemporal distribution table of energy storage capacity;

[0154] According to the capacity adjustment sequence ranking value matching geographic location code, first obtain the geographic coordinates of the energy storage node through the dispatch center. For example, sites 1 to 8 correspond to locations (116.38, 39.9), (117.2, 40.0)... and so on. By mapping the ranking value sequence such as [7, 1, 3, 6, 2, 4, 8, 5] to the energy storage site number in sequence according to the location, a mapping table is generated. Then, the arithmetic mean of the capacity adjustment ranking values ​​of the sites in each area (such as 5 longitude and latitude as a type of area) is taken, and the dispersion is calculated by standard deviation. For example, area A contains sites 1, 2, 3, and the ranking value is [7, 1, 3]. The regional average is , the discreteness is:

[0155] ;

[0156] Finally, a spatiotemporal distribution table of energy storage capacity in geographical space is formed.

[0157] Table 5: Energy storage node capacity adjustment ranking values ​​and spatial coordinates

[0158] As shown in Table 5, regions A, B, and C have formed differentiated capacity adjustment sorting distribution characteristics, and subsequently hierarchical regulation and configuration can be carried out based on regional differences.

[0159] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for analyzing energy storage demand in a high-proportion new energy power grid, characterized in that: The following steps are involved: S1: Obtain the power forecast value sequence at the time point in the current forecast period through the new energy output forecast platform. Based on the measured power value sequence at the corresponding time point in the original database, perform point-by-point offset calculation to generate a forecast deviation sequence. S2: Input the predicted deviation sequence into the sliding window algorithm to split the time series window, and dynamically adjust the window length based on the volatility of the absolute value of the deviation, calculate the variance of the absolute value of the deviation in the window, and generate a variance feature set; S3: Based on the variance feature set, a K-means clustering algorithm is used to optimize the number of cluster centers by using the silhouette coefficient, the variance is used as a feature vector for clustering, and a window classification result set is output; S4: According to the intensity level labels in the window classification result set, matching the level-capacity adjustment coefficient, performing a proportional adjustment operation on the energy storage capacity segment, and generating a time-divided capacity adjustment value; S5: extracting the time points in the time-divided capacity adjustment values, calculating the Pearson correlation coefficients between the load and the new energy output at the corresponding time points, sorting the time-divided capacity adjustment values ​​according to the correlation coefficients, and generating a spatiotemporal distribution table of energy storage capacity.

2. The energy storage demand analysis method for a high-proportion new energy power grid according to claim 1 is characterized in that: The predicted deviation sequence includes a timestamp deviation value, a deviation direction attribute, and a cumulative deviation amount; the variance feature set includes a window variance value, a self-matching window length, and a fluctuation intensity level; the window classification result set includes a cluster label, cluster center coordinates, and intra-class variance value; the time-based capacity adjustment value includes a time-based benchmark capacity, a level adjustment coefficient, and a fluctuation intensity-capacity mapping relationship; and the energy storage capacity spatiotemporal distribution table includes a time priority index, a Pearson correlation coefficient value, and spatiotemporal distribution coordinate points; The self-matching window length refers to the window time span parameter that is dynamically adjusted according to the deviation volatility; The level adjustment coefficient refers to a capacity adjustment ratio parameter based on clustering strength level matching.

3. The energy storage demand analysis method for a high-proportion new energy power grid according to claim 1 is characterized in that: The specific steps of S1 include: S101: Obtain a power prediction value sequence at a time point in the current prediction period in the new energy output prediction platform, call a measured power value sequence at a corresponding time point in the original database, calculate the numerical offset between the two at the time point, and generate a power offset sequence; S102: Based on the power offset sequence, calculate the absolute value of the offset at each time point, use the predicted value as a reference value to perform item-by-item ratio calculation, organize the data sequence according to a unified time index format, and obtain a corresponding offset rate sequence; S103: calling the corresponding offset rate sequence, calculating the offset degrees of the offset rates of adjacent time points in time index order, and performing linear sorting processing on the offset degrees to obtain a prediction deviation sequence.

4. The method for analyzing energy storage demand of a high-proportion new energy power grid according to claim 3 is characterized in that: The specific steps of S2 include: S201: Based on the predicted deviation sequence, a sliding window algorithm is used to set an initial window length and slide the position in sequence, extract the deviation values ​​in the window and take the absolute values ​​to form a set, thereby generating a deviation absolute value sequence distribution set; S202: calling the deviation absolute value sequence distribution set, calculating the difference between the maximum and minimum values ​​in the window as the window fluctuation amplitude, comparing it with the set deviation fluctuation benchmark threshold, and adjusting the window length to obtain a dynamic window length sequence; The deviation fluctuation reference threshold is the average value of the absolute value of the deviation plus or minus the empirical coefficient k times the standard deviation; S203: Based on the time series window after the dynamic window length sequence is adjusted, a set of absolute values ​​of deviations within the window is called, variance values ​​are calculated item by item, and their fluctuation levels are recorded to generate a variance feature set.

5. The method for analyzing energy storage demand of a high-proportion new energy power grid according to claim 4 is characterized in that: The specific steps of S3 include: S301: Based on the variance feature set, each window is fitted into a feature vector including the fluctuation amplitude, the window variance value and the self-matching length, input into the K-means clustering model, set multiple cluster centers to perform clustering operation, calculate the window silhouette coefficient under each group of clusters, and generate a window silhouette coefficient value set; The window silhouette coefficient is an indicator that measures the degree of cohesion and separation of each window in the cluster; S302: calling the window silhouette coefficient value set, calculating the mean of the silhouette coefficient under the number of cluster centers, extracting the mean sequence and comparing the values ​​thereof, selecting the center number corresponding to the maximum value as the K-means input parameter, re-executing clustering, and obtaining the optimal cluster label number group; S303: Based on the optimal clustering label number group, extract the average variance characteristics and fluctuation level of the representative window, build a "window intensity level-variance characteristic" dynamic matching model, establish a mapping table between category labels and fluctuation intensity levels, and obtain a window classification result set.

6. The method for analyzing energy storage demand of a high-proportion new energy power grid according to claim 5, characterized in that: The mean of the silhouette coefficient is calculated using the formula: ； in, The number of representative cluster centers is The mean of the window silhouette coefficient when , Represents the current number of cluster centers The total number of windows under Representative The number of windows in the cluster center is The silhouette coefficient value under Representative The number of windows in the cluster center is The corresponding window variance value is Representative The number of windows in the cluster center is The corresponding fluctuation amplitude value is Representative The number of windows in the cluster center is The average of the fluctuation amplitude values ​​in the lower window.

7. The method for analyzing energy storage demand of a high-proportion new energy power grid according to claim 5, characterized in that: The specific steps of S4 include: S401: Obtaining the intensity level label of each window based on the window classification result set, and dynamically extracting the capacity adjustment coefficient that matches the fluctuation intensity through a preset level-capacity adjustment parameter mapping table to generate a label adjustment coefficient; The label adjustment coefficient is based on the window fluctuation intensity level and the capacity adjustment factor dynamically extracted through the mapping table to proportionally correct the energy storage capacity data; S402: Based on the label adjustment coefficient, perform a multiplication ratio conversion on the original data of the energy storage capacity in the window, and apply the adjustment coefficient to the corresponding capacity value according to the energy storage capacity value in the time point sequence to obtain the time increment number; S403: Performing a data weighted operation on the energy storage capacity at a time point according to the moment increment number, calculating an adjustment value based on the original capacity at each moment and its corresponding change, replacing the full-segment capacity data, and generating a time-divided capacity adjustment value.

8. The method for analyzing energy storage demand of a high-proportion new energy power grid according to claim 7, characterized in that: To calculate the increment number of the time, the formula is used: ； in, Represents at a point in time The capacity increment after adjustment is in MWh. Represents at a point in time The original energy storage capacity value, in MWh, Represents the window The capacity adjustment factor extracted from the intensity level label corresponds to Representative time point The original capacity value, in MWh, Representative Window The average of all original capacity values ​​in MWh, Representative Window The capacity change trend adjustment factor extracted by regression prediction is in MWh. Representative time point The corresponding capacity fluctuation error compensation value is in MWh. Represents the current window The number of time points included in Represents the time point number currently being processed. is a dynamic weight factor, whose value is calculated by the energy storage controller according to the real-time grid frequency deviation.

9. The method for analyzing energy storage demand of a high-proportion new energy power grid according to claim 7, characterized in that: The specific steps of S5 include: S501: extracting time points based on the time-based capacity adjustment values, performing normalization processing according to the time window and calculating linear correlation based on the load and new energy output data, and generating a time point correlation coefficient sequence; S502: Calling the time point correlation coefficient sequence, setting a position mapping rule in descending order of correlation coefficient values, calculating the capacity adjustment sequence difference, reorganizing the original time period capacity adjustment values ​​according to the index to generate a new value arrangement, and obtaining the capacity adjustment sequence ranking value; The capacity adjustment sequence ranking value is the result of re-ranking the original capacity adjustment data according to the degree of coupling between load and renewable energy output, reflecting the priority and value of the time point in energy storage regulation; S503: Matching the geographic location code according to the capacity adjustment sequence ranking value, establishing a ranking value and spatial coordinate mapping table, and calculating the regional mean and dispersion to generate a spatiotemporal distribution table of energy storage capacity.

10. The method for analyzing energy storage demand of a high-proportion new energy power grid according to claim 9, characterized in that: The capacity adjustment sequence difference is calculated using the formula: ； in, Representative The capacity adjustment sequence difference corresponding to each position is in MW. Represents the original At the time point The capacity adjustment value of each location, in MW, Representative Load data at each time point, in MW, Representative The new energy output offset value corresponding to each time point, in MW, Indicates the total number of time points, Indicates the The capacity adjustment value corresponding to each position is the mean value linearly normalized to [0, 1]. Represents the weight coefficient, for load data Offset with new energy The dimensions of are normalized.

Images