A Method and System for Identifying Privately Installed Energy Storage Loads Based on Electricity Consumption Behavior Patterns

Abstract

This application discloses a method and system for identifying privately installed energy storage loads based on electricity consumption behavior patterns, belonging to the field of privately installed energy storage identification technology. The method includes: obtaining hierarchical load sets by performing hierarchical analysis of historical load data based on differences in identification results; reconstructing typical load curves by morphological clustering of daily load curves for each hierarchical load set; constructing baseline load curves using rigid load confidence weight coefficients and typical load curves of each hierarchical load set; obtaining energy storage behavior curves under different user behavior patterns based on the baseline load curves and typical load curves; obtaining privately installed energy storage load characteristics corresponding to the user behavior patterns based on the energy storage behavior curves; and performing privately installed energy storage load identification based on the privately installed energy storage load characteristics. The beneficial effects of this application are: improving the accuracy and comprehensiveness of privately installed energy storage load identification by performing identification of privately installed energy storage loads with different energy storage behaviors.

Description

Technical Field

[0001] This application relates to the field of privately installed energy storage identification technology, and in particular to a method and system for identifying privately installed energy storage loads based on electricity consumption behavior patterns. Background Technology

[0002] Distributed energy resources (DERs), represented by user-side energy storage, are experiencing rapid development. However, the influx of a large number of undeclared privately installed energy storage systems into the distribution network poses serious analytical challenges to the safe and stable operation of the grid and the management of electricity. The user-side data obtained by the grid company typically comes from the net load readings of smart meters. This net load data is a highly aggregated hybrid signal, simultaneously embedding the user's basic electricity load and the charging and discharging power of privately installed energy storage.

[0003] For power grid management, simply determining whether a user has illegally installed energy storage and knowing the probability of such installation is far from sufficient. To achieve refined management, security verification, anti-theft and illegal activity detection, and even potential resource aggregation, power grid companies urgently need to non-intrusively decouple the actual operating information of energy storage from the net load readings of smart meters. This means accurately predicting and estimating the basic electricity load, actual charging and discharging power, start and end times, and daily charging and discharging volume.

[0004] Currently, for estimating user baseline load and non-intrusive load decomposition, existing technologies mainly rely on statistical baseline estimation, such as the "X of Y days" method or the simple historical mean method; regression analysis-based methods attempt to establish regression models of load with external variables such as temperature, humidity, and date type.

[0005] However, the aforementioned existing technologies have fundamental flaws when applied to predicting baseline loads of users with privately installed energy storage and separating their charging and discharging information. First, existing baseline estimation methods typically assume that the historical data used is normal. However, for users with privately installed energy storage, their historical data includes both normal electricity consumption days and energy storage activity days, causing the model to contaminate the charging and discharging behavior of energy storage and incorrectly learn it as part of the baseline, resulting in severely distorted estimation results. Second, this data contamination problem is further amplified by behavioral similarities; that is, the model struggles to distinguish whether peak shaving and valley filling phenomena stem from economic arbitrage of energy storage or from proactive energy-saving demand response by users, leading to policy misjudgments. Summary of the Invention

[0006] This application addresses the problem of poor accuracy in identifying privately installed energy storage loads in existing technologies by providing a method and system for identifying privately installed energy storage loads based on electricity consumption behavior patterns. The method stratifies load data by analyzing differences in historical identification results to separate load information under different user behaviors. Based on the confidence level of no-energy-storage behavior corresponding to each level and typical load curves, it separates energy storage behavior curves under different levels, thereby obtaining the characteristics of privately installed energy storage loads under different user behavior patterns, enabling targeted identification of privately installed energy storage loads and improving identification accuracy.

[0007] To achieve the above technical objectives, this application provides a technical solution: a method for identifying privately installed energy storage loads based on electricity consumption behavior patterns, comprising the following steps: performing historical load data stratification based on differences in identification results in historical privately installed energy storage identification data to obtain stratified load sets; performing morphological clustering of daily load curves for each stratified load set to obtain core clusters, and reconstructing typical load curves using the core clusters; matching rigid load confidence weight coefficients of the stratified load sets based on the identification results; constructing baseline load curves using the rigid load confidence weight coefficients and typical load curves of each stratified load set; obtaining energy storage behavior curves under different user behavior patterns based on the baseline load curves and typical load curves, and obtaining privately installed energy storage load characteristics corresponding to the user behavior patterns based on the energy storage behavior curves; and performing privately installed energy storage load identification using the privately installed energy storage load characteristics.

[0008] Furthermore, based on the differences in identification results in historical privately installed energy storage identification data, historical load data is stratified to obtain a stratified load set, including: performing privately installed energy storage identification based on the comparison results of user load data, normal load data, and typical energy storage equipment load data, and obtaining the suspicion level; and stratifying historical load data according to the suspicion level in historical privately installed energy storage identification data and the preset suspicion level step size to obtain a stratified load set corresponding to different suspicion level intervals.

[0009] Furthermore, the step of performing morphological clustering of daily load curves for each hierarchical load set to obtain core clusters, and reconstructing typical load curves using core clusters, includes: calculating the similarity of daily load curves in each hierarchical load set using dynamic time warping; and using DBSCAN to perform morphological clustering of daily load curves based on the similarity of daily load curves, filtering out noisy curves and obtaining core clusters.

[0010] Furthermore, the step of using DBSCAN to perform morphological clustering of daily load curves based on their similarity, filtering out noisy curves, and obtaining core clusters includes: sorting all daily load curve similarities for each daily load curve in ascending order, selecting the daily load curve similarity corresponding to a preset minimum number of neighbors as a threshold benchmark; constructing a k-distance curve based on all threshold benchmarks in each hierarchical load set, using the elbow of the k-distance curve as the neighborhood radius of DBSCAN; and performing morphological clustering of daily load curves within each hierarchical load set based on the neighborhood radius of each set.

[0011] Furthermore, the reconstruction of the typical load curve based on the core cluster includes: obtaining the highest density cluster in the core cluster, and reconstructing the typical load curve based on the highest density cluster using a point-by-point mass calculation method.

[0012] Furthermore, the step of matching the rigid load confidence weight coefficient of the hierarchical load set according to the identification result includes: dividing each hierarchical load set into a baseline dominant layer load set and an energy storage activity layer load set according to a preset doubt threshold; and assigning a rigid load confidence weight coefficient to the baseline dominant layer load set according to the exponential decay model.

[0013] Furthermore, the process of matching the rigid load confidence weight coefficients of the hierarchical load sets based on the identification results also includes: using the ratio between the rigid load confidence weight coefficient corresponding to the highest suspicion level and the rigid load confidence weight coefficient corresponding to the lowest suspicion level as the target suppression rate, and obtaining the confidence decay constant based on the preset target suppression rate and the exponential decay model through reverse decoupling; and calculating the rigid load confidence weight coefficients of each hierarchical load set whose suspicion level is less than the preset suspicion level threshold using the confidence decay constant.

[0014] Furthermore, the construction of the baseline load curve using the rigid load confidence weight coefficient and the typical load curves of each stratified load set includes: obtaining sample size correction parameters based on the amount of data in the stratified load set; and constructing the baseline load curve by weighting and fusing the typical load curves of each stratified load set using the sample size correction parameters and the rigid load confidence weight coefficient.

[0015] Furthermore, obtaining energy storage behavior curves under different user behavior modes based on the baseline load curve and the typical load curve includes: calculating the energy storage behavior curves of each energy storage activity layer load set based on the baseline load curve and the typical load curves of the energy storage activity layer load set.

[0016] Another technical solution provided in this application is a privately installed energy storage load identification system based on electricity consumption behavior patterns, used to implement the method described above, and further comprising: a data layering unit, used to perform historical load data layering based on the differences in identification results in historical privately installed energy storage identification data, and obtain layered load sets; a typical load analysis unit, used to perform morphological clustering of daily load curves for each layered load set to obtain core clusters, and reconstruct typical load curves with core clusters; a confidence analysis unit, used to match the rigid load confidence weight coefficients of the layered load sets according to the identification results; a baseline load analysis unit, used to construct baseline load curves with the rigid load confidence weight coefficients and the typical load curves of each layered load set; a feature analysis unit, used to obtain energy storage behavior curves under different user behavior patterns based on the baseline load curves and typical load curves, and obtain privately installed energy storage load characteristics corresponding to the user behavior patterns based on the energy storage behavior curves; and an identification analysis unit, used to perform privately installed energy storage load identification with the privately installed energy storage load characteristics.

[0017] The beneficial effects of this application are as follows: 1. By stratifying historical load data based on differences in identification results in historical privately installed energy storage identification data, user load data with different identification results are separated, avoiding overlap of load characteristics containing different levels of energy storage activity, ensuring that the energy storage characteristics in each layered load set are similar, and improving the accuracy of energy storage load characteristic separation. Simultaneously, morphological clustering is used to remove occasional electricity noise in the layered load sets, ensuring the accuracy of the data used for energy storage load characteristic separation. Furthermore, based on the identification results, the rigid load confidence weight coefficients of the layered load sets are matched, making the typical load curve corresponding to low-probability privately installed energy storage play a more dominant role in constructing the baseline load curve. This ensures that the baseline load curve closely matches the rigid electricity consumption characteristics when users are not engaged in energy storage activity. Thus, by using the baseline load curve and the typical load curve to separate energy storage behavior curves under different identification results, the energy storage load characteristics of users under different behavior modes are displayed. This allows for the identification of privately installed energy storage loads with different energy storage behaviors, improving the accuracy and comprehensiveness of privately installed energy storage load identification.

[0018] 2. Dynamic Time Warping (DTW) is used to calculate the similarity of daily load curves in each hierarchical load set, allowing curves to flexibly align on the time axis. This adapts to slight time drift in electricity consumption behavior, ensuring the accuracy of similarity judgment and improving the robustness of pattern recognition. Furthermore, DBSCAN (Density-Based Spatial Clustering) is used to mark daily load curves with low similarity as noise and remove them to avoid interference from noisy data. Based on curve density, daily load curves with similar patterns are aggregated to form core clusters, reflecting different user electricity consumption behavior patterns within the same hierarchical load set.

[0019] 3. Generate baseline load curves corresponding to pure rigid load changes based on the baseline dominant layer load set, and filter out pure rigid load changes in the energy storage activity layer load set based on the baseline load curves, retaining the load changes brought about by energy storage activities, and construct energy storage behavior curves corresponding to each energy storage activity layer load set to reflect the energy storage load fluctuation characteristics under different user behavior modes. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the method for identifying privately installed energy storage loads based on electricity consumption behavior patterns in this application.

[0021] Figure 2 This is a schematic diagram illustrating the typical load curve construction process of the privately installed energy storage load identification method based on electricity consumption behavior patterns in this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of this application and are only used to explain this application. They do not limit the scope of protection of this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] like Figure 1 As shown in the first embodiment of this application, the method for identifying privately installed energy storage loads based on electricity consumption behavior patterns includes the following steps:

[0024] Based on the differences in identification results in historical privately installed energy storage identification data, historical load data is stratified to obtain a stratified load set;

[0025] For each hierarchical load set, morphological clustering of the daily load curves is performed to obtain core clusters, and typical load curves are reconstructed using the core clusters;

[0026] Based on the identification results, match the rigid load confidence weight coefficients of the hierarchical load set;

[0027] Baseline load curves are constructed using the rigid load confidence weighting coefficients and typical load curves of each stratified load set;

[0028] Based on the baseline load curve and typical load curve, obtain the energy storage behavior curve under different user behavior modes, and obtain the privately installed energy storage load characteristics corresponding to the user behavior modes based on the energy storage behavior curve.

[0029] Identification of privately installed energy storage loads is performed based on their characteristics.

[0030] In this embodiment, historical load data is stratified based on the differences in identification results among historical privately installed energy storage identification data. This separates user load data with different identification results, avoiding overlap of load characteristics containing different levels of energy storage activity. This ensures that the energy storage characteristics in each stratified load set are similar, improving the accuracy of energy storage load characteristic separation. Simultaneously, morphological clustering is used to remove occasional electricity noise from the stratified load sets, ensuring the accuracy of the data used for energy storage load characteristic separation. Furthermore, based on the identification results, rigid load confidence weight coefficients are matched to the stratified load sets. This gives the typical load curve corresponding to low-probability privately installed energy storage a greater dominant role in constructing the baseline load curve, ensuring that the baseline load curve closely matches the rigid electricity consumption characteristics when users are not engaged in energy storage activity. Thus, by using the baseline load curve and the typical load curve, energy storage behavior curves under different identification results are obtained, showcasing the energy storage load characteristics of users under different behavior modes. This allows for the identification of privately installed energy storage loads with different energy storage behaviors, improving the accuracy and comprehensiveness of privately installed energy storage load identification.

[0031] In the first scenario, historical load data is stratified based on differences in identification results from historical privately installed energy storage identification data, resulting in a stratified load set including:

[0032] Based on the comparison results of user load data, normal load data, and typical energy storage equipment load data, we perform private energy storage identification and obtain the degree of suspicion.

[0033] Based on the suspiciousness of historical privately installed energy storage identification data and the preset suspiciousness step size, the historical load data is stratified to obtain the stratified load set corresponding to different suspiciousness intervals.

[0034] In this case, the probability of users installing private energy storage is first quantified based on the suspicion index. Suspicion can be quantified by comparing user load data with normal load data under normal production activities and load data under typical energy storage equipment conditions. Then, load data is stratified based on the suspicion in the historical private energy storage identification results, and the most representative typical load curves under different suspicion levels are extracted.

[0035] Specifically, identifying illegally installed energy storage based on comparisons of user load data with normal load data and typical energy storage device load data includes at least the following steps:

[0036] The first identification result is obtained by comparing the current correlation between load and phased electricity price with the normal correlation.

[0037] The daytime load fluctuation pattern is obtained based on historical regional load data, and the second identification result is obtained based on the comparison between the current load fluctuation and the daytime load fluctuation pattern.

[0038] A user behavior baseline is obtained based on historical user electricity consumption data, and a third identification result is obtained based on the comparison between the user's current electricity consumption behavior and the user behavior baseline.

[0039] A pre-defined suspicious assessment strategy is constructed based on the degree of anomaly persistence and anomaly accumulation within a pre-defined observation period;

[0040] Based on the first identification result, the second identification result, the third identification result, and the preset suspicious assessment strategy, the identification result of privately installed energy storage is output.

[0041] The identification results of privately installed energy storage include at least a suspicion level. The identification of privately installed energy storage is carried out daily, and the daily suspicion level is obtained. When the identification result is abnormal, it is recorded as 1, and when the identification result is normal, it is recorded as 0. The sum of the truth values ​​corresponding to all identification results on the day is used as the suspicion level.

[0042] This provides a judgment strategy for the first identification result, the second identification result, and the third identification result. In other cases, the first identification result, the second identification result, and the third identification result can also be output based on the strategy of judging normal and abnormal situations based on user load data, normal load data, and typical energy storage equipment load data.

[0043] The first identification result is obtained based on the comparison between the current correlation and normal correlation between load and phased electricity price, including:

[0044] Based on the time period corresponding to the nighttime off-peak electricity price, obtain user nighttime off-peak load data including user load, duration, and stability, and use user nighttime off-peak load data as the current relevance;

[0045] A normal correlation was constructed using nighttime normal load data of typical energy storage devices. The first correlation between load and duration, the second correlation between load and stability, and the energy storage load benchmark were retrieved from the normal correlation.

[0046] The first identification result is obtained by comparing the user's nighttime off-peak load data with the first correlation, the second correlation, and the energy storage load benchmark.

[0047] Based on historical regional load data, the daytime load fluctuation pattern is obtained. A second identification result is obtained by comparing the current load fluctuation with the daytime load fluctuation pattern, including:

[0048] The midday load fluctuation pattern was obtained based on historical regional load data.

[0049] Obtain the user's average power at noon, average power in the morning, and average power in the afternoon, and calculate the user's average power during the production period based on the user's average power in the morning and average power in the afternoon.

[0050] The current load fluctuation is calculated based on the user's average power during midday and the user's average power during production periods.

[0051] If the current load fluctuation does not conform to the midday load fluctuation pattern, the second identification result is that there is abnormal power consumption behavior at midday.

[0052] User behavior baselines are obtained based on historical user electricity consumption data. Third identification results are obtained by comparing current user electricity consumption behavior with these baselines, including:

[0053] Based on the benchmark period, obtain a set of historical user electricity consumption data for the same type of day, and obtain a user behavior benchmark based on the median curve of the historical user electricity consumption data set;

[0054] The deviation of the user's current electricity consumption behavior from the user behavior benchmark is calculated based on the dynamic time warping algorithm and the time weighting factor corresponding to the phased electricity price.

[0055] If the local morphological deviation is higher than the preset deviation threshold, the third identification result is output as "there is a deviation in electricity consumption behavior".

[0056] Understandably, identifying illegally installed energy storage based on the comparison results of user load data with normal load data and typical energy storage equipment load data precedes identifying illegally installed energy storage load. Since there isn't enough sample data for illegally installed energy storage in the initial identification phase, illegal installation identification is performed first based on the comparison results of user load data with normal load data and typical energy storage equipment load data, outputting the probability of illegally installed energy storage to prompt operators to investigate. Then, after a period of illegal installation identification, the identification of illegally installed energy storage load data is performed to further improve the accuracy of illegal installation identification. Operators can then conduct targeted investigations and management based on the illegally installed energy storage load, improving the efficiency of illegal installation identification.

[0057] In this scenario, if the first identification result indicates the presence of off-peak charging behavior at night, it is considered an abnormal result. If the second identification result indicates abnormal midday power consumption behavior, it is considered an abnormal result. If the third identification result indicates deviation in power consumption behavior, it is considered an abnormal result. Conversely, if the first identification result indicates the absence of off-peak charging behavior at night, it is considered a normal result. If the second identification result indicates the absence of abnormal midday power consumption behavior, it is considered a normal result. If the third identification result indicates the absence of deviation in power consumption behavior, it is considered a normal result. The higher the suspicion level, the more abnormal identification results are corresponding to. Based on the suspicion level, the daily load is stratified so that the load sets at each stratification level correspond to the same suspicion level interval, meaning the number of abnormal identification results is approximately the same. By using load samples with high homogeneity to screen the characteristics of privately installed energy storage loads corresponding to the suspicion level interval, the identification accuracy is improved. In other cases, any judgment strategy that outputs the probability of privately installed energy storage can be used as the suspicion level acquisition method in this embodiment. In this case, the probability of privately installed energy storage is the suspicion level.

[0058] In this embodiment, the preset suspicion level step size is dynamically adjusted based on the number of identification strategies for the preceding privately installed energy storage. If the number of identification strategies is small, the corresponding number of identification results is also small, so the preset suspicion level step size is small, such as being set to 1, with each suspicion level corresponding to one level. If the number of identification strategies is large, the corresponding number of identification results is also large, so the preset suspicion level step size can be set larger, such as being 5, to balance the data similarity within the layer and the number of samples within the layer, avoiding the problem that an excessively large step size leads to excessively high heterogeneity of data within the layer, making it impossible to extract typical curves, or that an excessively small step size leads to insufficient sample size in some levels.

[0059] In the second scenario, historical load data is stratified based on differences in identification results from historical privately installed energy storage identification data, resulting in a stratified load set including:

[0060] Based on the differences in identification results (correct and abnormal, incorrect and abnormal, correct and normal, incorrect and normal), historical load data is stratified to obtain a stratified load set.

[0061] In this scenario, an "abnormal" identification result indicates the presence of unauthorized energy storage, a "normal" result indicates the absence of unauthorized energy storage, a "correct" result means the result matches the manual investigation result, and an "incorrect" result means the result differs from the manual investigation result. By analyzing the correctness and status of the identification results, historical load data is stratified. The correctness of the identification results helps identify misjudged samples, preventing erroneous data from interfering with subsequent identification of unauthorized energy storage load characteristics.

[0062] In the third scenario, a two-step stratification process can also be performed: stratifying historical load data based on differences in identification results from historical privately installed energy storage identification data, resulting in a stratified load set including:

[0063] Based on the differences in identification results (correct and abnormal, incorrect and abnormal, correct and normal, incorrect and normal), perform one stratification of historical load data to obtain an initial stratified load set;

[0064] The initial stratified load set is stratified a second time based on the suspicion level and the preset suspicion level step size to obtain stratified load sets corresponding to different suspicion level intervals.

[0065] In this scenario, a first stratification is performed based on the correctness of the identification and the status of the identification results to isolate misjudgments. A second stratification is then performed based on the degree of suspicion to reflect the differences in the degree of impact from energy storage, achieving accurate segmentation of homogeneous data. This second stratification further stratifies the initial stratified load sets, separating standard rigid loads with a suspicion level approaching 0 from rigid loads with low suspicion levels exhibiting energy storage-like behavior. This avoids mixing rigid loads with different behavioral patterns, which could lead to the generated typical load curves failing to accurately reflect loads under different behavioral patterns. This ensures that the load data in each stratified load set possesses not only state homogeneity but also behavioral homogeneity, guaranteeing the accuracy and relevance of the generated typical load curves.

[0066] In the fourth case, the secondary stratification of the initial stratified load set based on the suspicion level and a preset suspicion level step size, to obtain stratified load sets corresponding to different suspicion level intervals, further includes:

[0067] The initial stratified load set is stratified a second time based on the suspicion level and the preset suspicion level step size. The failure sample level is obtained based on the result of the second stratification and the preset sample number threshold.

[0068] The load data in the failure sample hierarchy is averaged based on the initial hierarchical load set, and the failure sample hierarchy is screened out to obtain the hierarchical load set.

[0069] After secondary stratification, some suspicious intervals may have insufficient sample sizes. Strata with fewer than a preset sample size threshold after secondary stratification are designated as failed strata, as they are insufficient for subsequent clustering and reconstruction. The load data of these failed strata are evenly distributed to adjacent, similar strata with sufficient sample sizes, based on the same initial stratified load set, and then these failed strata are deleted. The initial stratified load set ensures that the partitioning of the failed strata does not cause heterogeneity in other strata. For example, if the initial stratified load set of a failed stratum corresponds to a correctly identified and normal result, but its adjacent suspicious intervals belong to another stratified load set, then the load data of that failed stratum will not be assigned to that stratum. This ensures data homogeneity, and the even partitioning avoids wasting effective data while preventing excessive data addition that could negatively impact the original stratified load characteristics.

[0070] like Figure 2 As shown, morphological clustering of daily load curves is performed on each stratified load set to obtain core clusters. The reconstructing of typical load curves using these core clusters includes:

[0071] The similarity of daily load curves in each stratified load set is calculated using dynamic time warping.

[0072] DBSCAN is used to perform morphological clustering of daily load curves based on their similarity, filtering out noisy curves and obtaining core clusters.

[0073] In this embodiment, Dynamic Time Warping (DTW) is used to calculate the similarity of daily load curves in each hierarchical load set. This allows the curves to be flexibly aligned on the time axis, thus adapting to slight time drifts in electricity consumption behavior, ensuring the accuracy of similarity judgment, and improving the robustness of pattern recognition. Furthermore, DBSCAN (Density-Based Spatial Clustering) is used to mark daily load curves with low similarity as noise and remove them to avoid interference from noisy data. Based on curve density, daily load curves with similar patterns are aggregated to form core clusters, reflecting different user electricity consumption behavior patterns within the same hierarchical load set. It can be understood that daily load curves are constructed based on daily load data.

[0074] Specifically, DBSCAN is used to perform morphological clustering of daily load curves based on their similarity, filtering out noisy curves and obtaining core clusters, including:

[0075] The similarity of all daily load curves for each daily load curve is sorted in ascending order, and the similarity of the daily load curves corresponding to the preset minimum number of neighbors is selected as the threshold benchmark.

[0076] A k-distance curve is constructed based on all threshold benchmarks in each hierarchical load set, and the elbow of the k-distance curve is used as the neighborhood radius of DBSCAN.

[0077] Morphological clustering of daily load curves within each hierarchical load set is performed based on the neighborhood radius of each set.

[0078] In this embodiment, the k-distance algorithm is used to dynamically change the neighborhood radius corresponding to each hierarchical load set, ensuring that the neighborhood radius adapts to the similarity distribution differences of different hierarchical load sets, thereby improving the accuracy of core cluster acquisition.

[0079] The preset minimum number of neighbors (min_samples) can be set based on expert experience. In this embodiment, the preset minimum number of neighbors is 4 or 5. In DBSCAN of this embodiment, the minimum number of neighbors required to form a core object requires that a non-occasional power consumption pattern must have appeared at least (min_samples + 1) times in the historical data. Setting it to 4 or 5 can effectively filter out random power consumption interference that only appears once or twice, while ensuring robust identification of mainstream patterns.

[0080] Using the k-distance algorithm, k is set as the preset minimum number of neighbors. The similarity of the kth daily load curve of each daily load curve is obtained as the threshold benchmark. The k-distance curves are constructed by sorting the threshold benchmarks in ascending order. The elbow point with the largest curvature change of the curve is identified, and the threshold benchmark corresponding to the elbow point is used as the neighborhood radius.

[0081] The elbow of the k-distance curve is essentially a natural boundary point for curve similarity. The left side of the elbow corresponds to the majority of curves, and the right side corresponds to the minority of curves. The neighborhood radius is set based on this point to avoid overly coarse or overly fine clustering and to ensure the suitability of the clustering.

[0082] Furthermore, reconstructing typical load curves using the core cluster includes:

[0083] Obtain the highest density cluster in the core cluster, and reconstruct typical load curves based on the highest density cluster using a point-by-point mass point calculation method.

[0084] After morphological clustering, each hierarchical load set may contain multiple core clusters, representing various mainstream electricity consumption patterns. The cluster with the largest number of samples is designated as the maximum density cluster, representing the most frequently occurring stable load pattern among users at the current level. Centroid calculations are performed on all daily load curves within the maximum density cluster to generate typical load curves for that level.

[0085] ;

[0086] Where t represents the sampling time, This represents the typical load curve for the i-th level. This represents the number of intra-cluster curves for the maximum density cluster at the i-th level. This represents the load curve for the d-th historical day. Indicates the highest density cluster. This represents the load curve.

[0087] By performing point-by-point centroid calculations on all daily load curves in the highest density cluster, typical load curves for that level are reconstructed to ensure that the typical load curves reflect the representative load patterns of that level, thereby obtaining the core behavioral characteristics of that level.

[0088] Corresponding to the first case in this embodiment, the rigid load confidence weight coefficients for matching the hierarchical load set based on the identification results include:

[0089] Based on a preset suspicion threshold, each layered load set is divided into a baseline dominant layer load set and an energy storage activity layer load set.

[0090] The baseline dominant layer load set is assigned a rigid load confidence weight coefficient based on the exponential decay model.

[0091] In this scenario, the suspicion threshold for triggering an alarm on unauthorized energy storage installation is used as the preset suspicion threshold in this embodiment. Layered load sets with suspicion levels below the preset threshold are classified as baseline dominant load sets. These sets primarily reflect users' rigid electricity demand and are less affected by unauthorized energy storage installations. Layered load sets with suspicion levels greater than or equal to the preset threshold are classified as energy storage active load sets. These sets contain significant energy storage operation characteristics. Rigid load confidence weight coefficients are assigned to each baseline dominant load set based on an exponential decay model. These rigid load confidence weight coefficients quantify the degree of quantification of the baseline load curve construction by different levels of baseline dominant load sets.

[0092] Because the risk of baseline contamination increases dramatically and non-linearly with a slight increase in the degree of doubt, an exponential decay model is used to assign confidence weights to the rigid loads.

[0093] ;

[0094] in, Let represent the confidence weight coefficient of the rigid load at the i-th level, and j represent the confidence decay constant (j > 0). This represents the representativeness of the i-th level.

[0095] Representativeness doubt can be obtained based on the geometric center of the doubt interval at each level. For example, if the doubt interval is 0-5, the representativeness doubt is 2.5. An exponential decay model ensures that when the representativeness doubt is 0, the rigid load confidence weight coefficient is 1. As the representativeness doubt increases, the rigid load confidence weight coefficient decreases exponentially, thereby quickly suppressing the influence of potential interference data.

[0096] Among them, the rigid load confidence weight coefficients for matching the hierarchical load set based on the identification results also include:

[0097] The ratio between the rigid load confidence weight coefficient corresponding to the highest suspicion level and the rigid load confidence weight coefficient corresponding to the lowest suspicion level is used as the target suppression rate, and the confidence decay constant is obtained by inverse decoupling based on the preset target suppression rate and the exponential decay model.

[0098] The rigid load confidence weight coefficients for each hierarchical load set whose doubt level is less than the preset doubt level threshold are calculated using the confidence level decay constant.

[0099] The target suppression rate is set at the ratio of the rigid load confidence weight coefficient corresponding to the rigid load corresponding to the baseline dominant load set with the highest suspicion level to that corresponding to the rigid load with the lowest suspicion level. This limits the weight difference between the baseline dominant load set most likely to have data interference and the baseline dominant load set least likely to have data interference. The confidence attenuation coefficient is decoupled in reverse according to the preset target suppression rate. This ensures that the confidence attenuation coefficient will not lead to the introduction of potential energy storage pollution data from the high-suspicion baseline dominant load set into the baseline load curve due to insufficient suppression, nor will it lead to the waste of a small amount of effective rigid load information from the high-suspicion baseline dominant load set due to excessive suppression. At the same time, it ensures that the baseline dominant load set with the lowest suspicion level dominates the generation of subsequent baseline load curves.

[0100] Specifically, the target suppression rate is the ratio between the rigid load confidence weight coefficient corresponding to the highest level of suspicion and the rigid load confidence weight coefficient corresponding to the lowest level of suspicion.

[0101] ;

[0102] Where r represents the target inhibition rate, Indicates the highest level of suspicion. Indicates the lowest level of suspicion. This represents the confidence weight coefficient for the rigid load corresponding to the highest degree of doubt. This represents the confidence weight coefficient of the rigid load corresponding to the lowest level of suspicion. It is understandable that the ratio between the confidence weight coefficient of the rigid load corresponding to the highest representative level of suspicion and the confidence weight coefficient of the rigid load corresponding to the lowest representative level of suspicion can also be used as the target suppression rate.

[0103] Therefore, the confidence decay constant obtained by reverse decoupling is:

[0104] ;

[0105] In this embodiment, r is 0.1, meaning the rigid load confidence weight corresponding to the lowest suspicion level is 10% of the rigid load confidence weight coefficient corresponding to the highest suspicion level. In other embodiments, a preset target suppression rate can be set according to the grid accuracy requirements. For example, for users with a small amount of historical data, a larger preset target suppression rate can be set to ensure that the influence of the baseline dominant load set with a high suspicion level is moderately suppressed while retaining a small amount of valid data to supplement the baseline load data, avoiding excessive suppression that leads to insufficient data volume for baseline fusion. For users with a large amount of historical data, a smaller preset target suppression rate can be set to strengthen the suppression of the baseline dominant load set that has a high probability of causing data interference, thereby improving the purity of the baseline load data.

[0106] Corresponding to the second case in this embodiment, the rigid load confidence weight coefficients for matching the hierarchical load set based on the identification results include:

[0107] The first hierarchical load set with correct and normal identification results, and the second hierarchical load set with incorrect and abnormal identification results are used as the baseline dominant layer load set;

[0108] The third-tier load set with correct identification results and abnormal identification results, and the fourth-tier load set with incorrect identification results and normal identification results are used as the energy storage activity layer load set;

[0109] The rigid load confidence weight coefficients for the first and second tiered load sets are assigned based on the amount of data in the first and second tiered load sets.

[0110] In this scenario, the rigid load confidence weight coefficients for the first and second tiered load sets are determined based on the amount of data in each set. Since the second tiered load set represents a misjudgment set without privately installed energy storage, the amount of data in the second tiered load set is necessarily less than that in the first tiered load set. Consequently, the rigid load confidence weight coefficient for the second tiered load set is also necessarily less than that for the first tiered load set, thus ensuring that the first tiered load set, with the lowest probability of interference, dominates the generation of the baseline load curve.

[0111] Corresponding to the third case in this embodiment, the rigid load confidence weight coefficients for matching the hierarchical load set based on the identification results include:

[0112] The first hierarchical load set is defined as the set of hierarchical loads corresponding to those with correct and normal identification results.

[0113] The hierarchical load set corresponding to the identification results being incorrect and abnormal is used as the second hierarchical load set;

[0114] The hierarchical load set corresponding to both correct and abnormal identification results is used as the third hierarchical load set;

[0115] The stratified load set corresponding to both incorrect and normal identification results is used as the fourth stratified load set;

[0116] The first and second tiered load sets are used as the baseline dominant load sets.

[0117] The third and fourth load tiers are used as the load tiers for energy storage activities.

[0118] Based on the linear decay model, rigid load confidence weight coefficients are assigned to the first stratified load set according to the degree of suspicion. Based on the exponential decay model, rigid load confidence weight coefficients are assigned to the second stratified load set according to the degree of suspicion.

[0119] In this scenario, the first and second tiered load sets are used as the baseline dominant load sets. Both sets are rigid load sets without any installed energy storage. A linear decay model assigns a linearly decreasing confidence weight coefficient to the first tiered load set with increasing suspicion, while an exponential decay model assigns an exponentially decreasing confidence weight coefficient to the second tiered load set with increasing suspicion. The load data in the first tiered load set consists entirely of pure rigid load data, with only slight gradient differences in suspicion. The linear decay model allows the confidence weight coefficient of the rigid load to decrease smoothly as suspicion increases, avoiding the waste of pure rigid load data. While the second tiered load set also contains pure rigid loads, they are marked as abnormal due to misidentification. Their electricity consumption behavior may exhibit occasional characteristics similar to energy storage, resulting in weak interference. The exponential decay model allows the confidence weight coefficient of the rigid load in the second tiered load set to decrease rapidly as suspicion increases, effectively suppressing weak interference, improving the rationality of weight allocation, and balancing data utilization and baseline purity.

[0120] The baseline load curve is constructed using the rigid load confidence weighting factor and the typical load curves of each stratified load set, including:

[0121] Logarithmic transformation is used to obtain sample size correction parameters based on the amount of data in the stratified load set;

[0122] Baseline load curves are constructed by weighting and fusing typical load curves of each stratified load set using sample size correction parameters and rigid load confidence weight coefficients.

[0123] In this embodiment, a sample size correction parameter is constructed based on the amount of data in each stratified load set. The sample size correction parameter is larger for stratified load sets with more data and smaller for stratified load sets with less data. This avoids the deviation of the final baseline load curve due to random bias caused by insufficient sample size. At the same time, the logarithmic transformation is used to prevent the sample size correction parameter of stratified load sets with more data from growing explosively. This ensures that the fused baseline load curve not only fits the pure rigid load characteristics of the low suspicion layer to the greatest extent, but also avoids the problems of small sample bias and excessive dominance of large sample. This ensures the accuracy and robustness of the baseline load curve, which reflects the basic load situation of users without energy storage.

[0124] The baseline load curve is constructed by weighting and fusing the typical load curves of each stratified load set using the sample size correction parameter and the rigid load confidence weight coefficient:

[0125] ;

[0126] in, Represents the baseline load curve. Represents the baseline dominant layer load set. This represents the number of curves in the hierarchical load set of the i-th level.

[0127] Based on the baseline load curve and typical load curve, energy storage behavior curves under different user behavior patterns are obtained, including:

[0128] The energy storage behavior curves for each energy storage active layer load set are calculated based on the baseline load curve and the typical load curves of the energy storage active layer load set.

[0129] In this embodiment, a baseline load curve corresponding to the pure rigid load change is generated based on the baseline dominant layer load set. The pure rigid load change in the energy storage activity layer load set is then filtered out based on the baseline load curve, while the load change brought about by energy storage activities is retained. This constructs an energy storage behavior curve corresponding to each energy storage activity layer load set, so as to reflect the energy storage load fluctuation characteristics under different user behavior modes.

[0130] Specifically, based on the baseline load curve aligned with the sampling time and the typical load curve of the energy storage active layer load set, the power difference at each sampling time is calculated to construct the energy storage behavior curve.

[0131] At a given sampling time, if the power difference between the power of the typical load curve and the power of the baseline load curve is greater than 0, it indicates that the power supply from the grid has increased at that sampling time, corresponding to the user engaging in energy storage charging behavior. If the power difference between the power of the typical load curve and the power of the baseline load curve is less than 0, it indicates that the power supply from the grid has been offset at that sampling time, corresponding to the user engaging in energy storage discharging behavior. If the power difference between the power of the typical load curve and the power of the baseline load curve is equal to 0, it corresponds to the user not engaging in energy storage charging or discharging behavior.

[0132] Corresponding to the first scenario in this embodiment, each layered load set is divided into a baseline dominant layer load set and an energy storage activity layer load set based on a preset suspicion threshold. In this case, when the suspicion level of the energy storage activity layer load set is close to the preset suspicion threshold, it is considered to reflect the user's load situation when using energy storage at low frequencies or on a trial basis; that is, its corresponding energy storage behavior curve reflects the energy storage power characteristics under the user's light-load behavior mode. When the energy storage activity layer load set has the maximum suspicion level, it is considered to reflect the user's maximized energy storage operation profile under full-power operation; that is, its corresponding energy storage behavior curve reflects the energy storage power characteristics under the user's full-load or typical energy storage mode.

[0133] Corresponding to the second scenario in this embodiment, the third-tier load set with correct identification results and abnormal identification results, and the fourth-tier load set with incorrect identification results and normal identification results are used as the energy storage activity layer load set. In this case, when the energy storage activity layer load set belongs to the fourth-tier load set, since the fourth-tier load set belongs to loads misidentified as normal behavior, it is considered to reflect the user's load situation when using energy storage at low frequency or on a trial basis; that is, its corresponding energy storage behavior curve reflects the energy storage power characteristics under the user's light-load behavior mode. When the energy storage activity layer load set belongs to the third-tier load set, its energy storage behavior is considered relatively typical or its load fluctuation is relatively large; therefore, its corresponding energy storage behavior curve reflects the energy storage power characteristics under the user's full-load or typical energy storage mode.

[0134] Different levels of curves are used to specifically reflect the energy storage operation characteristics of users under different behaviors: lower-level curves reveal users' light-load, trial or intermittent usage habits, while higher-level curves accurately reproduce users' extreme capabilities and typical operating strategies under full load conditions. This makes it easier for operators to quantitatively grasp the actual regulation capacity, power boundary and operating strategy evolution of the energy storage system, and provides rich information far exceeding the dimension of a single curve for grid-side resource survey and dispatch potential assessment.

[0135] In this embodiment, to eliminate spurious power fluctuations caused by smart meter measurement errors and baseline estimation residuals, the present invention introduces an adaptive noise dead zone threshold. Adaptive noise dead zone threshold The setting is based on the normal distribution in statistics. law:

[0136] ;

[0137] in, This represents the standard deviation of the active power measurement of the smart meter. Under the assumption that the measurement noise follows a Gaussian distribution, coefficient 3 corresponds to a 99.73% confidence interval. This means that any amplitude falling within... Power fluctuations within the specified range are highly probable (>99.7%) caused by measurement noise and should be filtered out. Conversely, fluctuations exceeding the adaptive noise dead zone threshold are considered statistically significant energy storage charging and discharging behaviors. This threshold ensures that Gaussian white noise is effectively filtered out.

[0138] At this point, based on the baseline load curve aligned with the sampling time and the typical load curve of the energy storage active layer load set, the power difference at each sampling time is calculated. If the power difference is greater than the maximum adaptive noise dead zone threshold, it is determined to be effective charging power. If the power difference is less than the minimum adaptive noise dead zone threshold, it is determined to be effective discharging power. The energy storage behavior curve is constructed using the effective charging power and the effective discharging power.

[0139] The power difference between the maximum adaptive noise dead zone threshold and the minimum adaptive noise dead zone threshold is considered to be caused by device standby or noise fluctuations and is corrected to 0 to ensure the accuracy of the energy storage behavior curve.

[0140] The characteristics of privately installed energy storage loads corresponding to user behavior patterns, obtained from energy storage behavior curves, include:

[0141] The power characteristics, energy characteristics, and time-domain characteristics of energy storage behavior curves are extracted using a discretization extraction method to obtain the load characteristics of privately installed energy storage.

[0142] The power characteristics include at least the maximum charging power and the maximum discharging power. The energy characteristics include at least the daily charging amount and the daily discharging amount. The time-domain characteristics include at least the charging period and the discharging period.

[0143] In this embodiment, a DBSCAN density clustering algorithm based on Dynamic Time Warping (DTW) is introduced. This algorithm can automatically identify and remove discrete statistical noise points and non-mainstream occasional behaviors in the dataset during the preliminary stage of extracting typical load curves. Through a pipelined process of denoising, clustering, and extraction, it is ensured that the typical load curves participating in subsequent core calculations are generated based on high-density and highly consistent mainstream electricity consumption patterns. This guarantees the purity and stability of the baseline estimation from the data source and significantly reduces the negative impact of outlier data on the algorithm's accuracy.

[0144] As a second embodiment of this application, a privately installed energy storage load identification system based on electricity consumption behavior patterns includes:

[0145] The data stratification unit is used to perform historical load data stratification based on the differences in identification results in historical privately installed energy storage identification data, and obtain a stratified load set.

[0146] Typical load analysis unit is used to perform morphological clustering of daily load curves for each hierarchical load set to obtain core clusters, and then reconstruct typical load curves using core clusters;

[0147] The confidence analysis unit is used to match the rigid load confidence weight coefficients of the hierarchical load set based on the identification results;

[0148] The baseline load analysis unit is used to construct baseline load curves using rigid load confidence weighting coefficients and typical load curves of each stratified load set.

[0149] The feature analysis unit is used to obtain energy storage behavior curves under different user behavior modes based on the baseline load curve and typical load curve, and to obtain the privately installed energy storage load characteristics corresponding to the user behavior mode based on the energy storage behavior curve.

[0150] The identification and analysis unit is used to identify privately installed energy storage loads based on their characteristics.

[0151] In this embodiment, the data layering unit is connected to the power grid system, the typical load analysis unit is connected to the data layering unit, the confidence analysis unit is connected to both the data layering unit and the typical load analysis unit, the baseline load analysis unit is connected to both the typical load analysis unit and the confidence analysis unit, the feature analysis unit is connected to both the baseline load analysis unit and the typical load analysis unit, and the identification analysis unit is connected to the feature analysis unit.

[0152] In one embodiment, the data layering unit is connected to a privately installed energy storage identification system that performs identification based on the comparison results of user load data, normal load data, and typical energy storage device load data.

[0153] The specific embodiments described above are preferred embodiments of the privately installed energy storage load identification method and system based on electricity consumption behavior patterns in this application, and are not intended to limit the specific scope of this application. The scope of this application includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape and structure of this application are within the protection scope of this application.

Claims

1. A method for identifying concealed loads of privately installed energy storage based on electricity consumption behavior patterns, characterized in that: Includes the following steps: Based on the differences in identification results in historical privately installed energy storage identification data, historical load data is stratified to obtain a stratified load set; For each hierarchical load set, morphological clustering of the daily load curves is performed to obtain core clusters, and typical load curves are reconstructed using the core clusters; Based on the identification results, match the rigid load confidence weight coefficients of the hierarchical load set; Baseline load curves are constructed using the rigid load confidence weighting coefficients and typical load curves of each stratified load set; Based on the baseline load curve and typical load curve, obtain the energy storage behavior curve under different user behavior modes, and obtain the privately installed energy storage load characteristics corresponding to the user behavior modes based on the energy storage behavior curve. Identify concealed loads of privately installed energy storage based on the characteristics of the installed energy storage load; Based on the differences in identification results from historical privately installed energy storage identification data, historical load data is stratified to obtain a stratified load set, including: Based on the comparison results of user load data, normal load data, and typical energy storage equipment load data, we perform private energy storage identification and obtain the degree of suspicion. Based on the suspicion level in historical privately installed energy storage identification data and the preset suspicion level step size, the historical load data is stratified to obtain the stratified load set corresponding to different suspicion level intervals; The rigid load confidence weight coefficients for matching the hierarchical load set based on the identification results include: Based on a preset suspicion threshold, each layered load set is divided into a baseline dominant layer load set and an energy storage activity layer load set. The baseline dominant layer load set is assigned a rigid load confidence weight coefficient based on the exponential decay model.

2. The method for identifying concealed loads of privately installed energy storage based on electricity consumption behavior patterns as described in claim 1, characterized in that: The step of performing morphological clustering of daily load curves for each hierarchical load set to obtain core clusters, and then reconstructing typical load curves using these core clusters, includes: The similarity of daily load curves in each stratified load set is calculated using dynamic time warping. DBSCAN is used to perform morphological clustering of daily load curves based on their similarity, filtering out noisy curves and obtaining core clusters.

3. The method for identifying concealed loads of privately installed energy storage based on electricity consumption behavior patterns as described in claim 2, characterized in that: The step of using DBSCAN to perform morphological clustering of daily load curves based on their similarity, filtering out noisy curves, and obtaining core clusters includes: The similarity of all daily load curves for each daily load curve is sorted in ascending order, and the similarity of the daily load curves corresponding to the preset minimum number of neighbors is selected as the threshold benchmark. A k-distance curve is constructed based on all threshold benchmarks in each hierarchical load set, and the elbow of the k-distance curve is used as the neighborhood radius of DBSCAN. Morphological clustering of daily load curves within each hierarchical load set is performed based on the neighborhood radius of each set.

4. The method for identifying concealed loads of privately installed energy storage based on electricity consumption behavior patterns as described in claim 1, characterized in that: The typical load curve reconstructed using the core cluster includes: The maximum density cluster in the core cluster is obtained, and a typical load curve is reconstructed based on the maximum density cluster using the point-by-point mass calculation method.

5. The method for identifying concealed loads of privately installed energy storage based on electricity consumption behavior patterns as described in claim 1, characterized in that: The rigid load confidence weight coefficient for matching the hierarchical load set based on the identification results also includes: The ratio between the rigid load confidence weight coefficient corresponding to the highest suspicion level and the rigid load confidence weight coefficient corresponding to the lowest suspicion level is used as the target suppression rate, and the confidence decay constant is obtained by inverse decoupling based on the preset target suppression rate and the exponential decay model. The rigid load confidence weight coefficients for each hierarchical load set whose doubt level is less than the preset doubt level threshold are calculated using the confidence level decay constant.

6. The method for identifying concealed loads of privately installed energy storage based on electricity consumption behavior patterns as described in claim 5, characterized in that: The construction of the baseline load curve using rigid load confidence weighting coefficients and typical load curves of each stratified load set includes: The sample size correction parameter is obtained based on the amount of data in the stratified load set; Baseline load curves are constructed by weighting and fusing typical load curves of each stratified load set using sample size correction parameters and rigid load confidence weight coefficients.

7. The method for identifying concealed loads of privately installed energy storage based on electricity consumption behavior patterns as described in claim 1, characterized in that: The process of obtaining energy storage behavior curves under different user behavior modes based on baseline load curves and typical load curves includes: The energy storage behavior curves for each energy storage active layer load set are calculated based on the baseline load curve and the typical load curves of the energy storage active layer load set.

8. A system for identifying concealed loads of privately installed energy storage based on electricity consumption behavior patterns, used to implement the method described in any one of claims 1 to 7, characterized in that: Also includes: The data stratification unit is used to perform historical load data stratification based on the differences in identification results in historical privately installed energy storage identification data, and obtain a stratified load set. Typical load analysis unit is used to perform morphological clustering of daily load curves for each hierarchical load set to obtain core clusters, and then reconstruct typical load curves using core clusters; The confidence analysis unit is used to match the rigid load confidence weight coefficients of the hierarchical load set based on the identification results; The baseline load analysis unit is used to construct baseline load curves using rigid load confidence weighting coefficients and typical load curves of each stratified load set. The feature analysis unit is used to obtain energy storage behavior curves under different user behavior modes based on the baseline load curve and typical load curve, and to obtain the privately installed energy storage load characteristics corresponding to the user behavior mode based on the energy storage behavior curve. The identification and analysis unit is used to identify hidden loads of privately installed energy storage based on the characteristics of the privately installed energy storage load.

Images