Self-adaptive management method and system for three-phase imbalance of transformer area based on scene recognition

Abstract

The invention discloses an area three-phase imbalance self-adaptive management method and system based on scene recognition. The method comprises the following steps: acquiring three-phase current data of users in an area; according to the dynamic fragmentation of the power consumption time period, extracting the power consumption mode characteristics of the user in the time fragmentation; calculating a similarity matrix according to the features, constructing a user scene mapping graph, and obtaining a user scene identifier through graph clustering; determining a three-phase imbalance evaluation weight according to the identifier, establishing a hierarchical evaluation criterion in combination with three-phase current data, generating a time-sharing contribution degree of a user to three-phase imbalance, formulating a regulation and control strategy according to the time-sharing contribution degree, adding differential noise encryption to the regulation and control strategy and power utilization mode characteristics, and performing multi-party fusion on encrypted data through federal learning to obtain a three-phase imbalance evaluation result. And a multi-user balance regulation and control scheme is generated to regulate the three-phase electrical load of the users, the three-phase current unbalance deviation after regulation is calculated, a new round of data acquisition is triggered when the three-phase current unbalance deviation exceeds a dynamic threshold value, and accurate identification and self-adaptive management of the three-phase unbalance of the transformer area are realized.

Description

Technical Field

[0001] This invention relates to the field of three-phase imbalance management and intelligent control technology for transformer substations, and particularly to an adaptive management method and system for three-phase imbalance in transformer substations based on scene recognition. Background Technology

[0002] With the continuous increase in electricity load and the integration of distributed energy resources, the three-phase imbalance problem in distribution network substations is becoming increasingly serious. Three-phase imbalance leads to degraded power quality, increased equipment losses, and reduced power supply reliability, posing safety hazards to power grid operation. Traditional three-phase imbalance management methods mainly rely on static adjustment means, such as phase adjustment or static compensation devices, which cannot adapt to the dynamic changes in user electricity consumption patterns.

[0003] Research on three-phase imbalance in power distribution areas mainly focuses on two aspects: first, load-side balancing regulation technology, which achieves phase balance by adjusting the load distribution of each phase; and second, balancing control technology based on compensation devices, which suppresses three-phase imbalance by installing compensation devices in the power grid. However, these methods typically fail to consider the differences in user electricity consumption scenarios, adopting a "one-size-fits-all" control strategy that makes it difficult to achieve refined management. Current technologies largely rely on manual experience or simple statistical analysis to identify user electricity consumption scenarios, making it difficult to accurately capture the time-varying characteristics of user electricity consumption behavior. Furthermore, the privacy protection of user-side data has not received sufficient attention, easily leading to the leakage of sensitive information during data fusion and analysis.

[0004] With the advancement of smart grid construction, the deployment of a large number of smart meters and monitoring equipment has made it possible to obtain refined electricity consumption data from users. However, how to use this data to achieve accurate identification and adaptive management of three-phase imbalance in distribution areas remains an urgent problem to be solved. Summary of the Invention

[0005] This invention provides a method and system for adaptive management of three-phase imbalance in transformer substations based on scene recognition, so as to achieve accurate identification and adaptive management of three-phase imbalance in transformer substations.

[0006] According to one aspect of the present invention, a scene recognition-based adaptive management method for three-phase imbalance in transformer substations is provided, comprising:

[0007] Collect three-phase current data from users within the transformer area;

[0008] Dynamically segment the user's electricity consumption according to the user's time period, and extract the user's electricity consumption pattern characteristics within the time segment;

[0009] A similarity matrix is ​​calculated based on electricity consumption pattern characteristics. A user scenario mapping map is constructed based on the similarity matrix. User scenario identifiers are obtained through map clustering.

[0010] The user's three-phase imbalance assessment weight is determined based on the scenario identifier. A hierarchical assessment criterion is established by combining the three-phase current data. The user's time-sharing contribution to the three-phase imbalance of the transformer area is generated. Based on the time-sharing contribution, a personalized control strategy for the user is formulated.

[0011] Differential noise is added to the personalized control strategy and electricity consumption pattern characteristics for encryption. The encrypted data is then fused from multiple sources through federated learning to generate a multi-user balanced control scheme.

[0012] Adjusting the user's three-phase power load according to the control plan, calculating the unbalanced deviation of the three-phase current after adjustment, and triggering a new round of data acquisition process when the unbalanced deviation exceeds the dynamic threshold.

[0013] According to another aspect of the present invention, a scene recognition-based adaptive management system for three-phase imbalance in transformer substations is provided, comprising:

[0014] The three-phase current acquisition module is used to collect three-phase current data from users within the transformer area.

[0015] The electricity consumption feature extraction module is used to dynamically segment users' electricity consumption according to their time periods and extract users' electricity consumption pattern features within the time segments.

[0016] The scene identification determination module is used to calculate a similarity matrix based on electricity consumption pattern characteristics, construct a user scene mapping map based on the similarity matrix, and obtain the user's scene identification through map clustering;

[0017] The control strategy formulation module is used to determine the user's three-phase imbalance assessment weight based on the scenario identifier, establish a hierarchical assessment criterion by combining three-phase current data, generate the user's time-sharing contribution to the three-phase imbalance of the transformer area, and formulate the user's personalized control strategy based on the time-sharing contribution.

[0018] The control scheme generation module is used to add differential noise to the personalized control strategies and electricity consumption pattern characteristics for encryption. The encrypted data is then fused with multi-party data through federated learning to generate a multi-user balanced control scheme.

[0019] The power load control module is used to adjust the user's three-phase power load according to the control scheme, calculate the unbalanced deviation of the three-phase current after adjustment, and trigger a new round of data acquisition when the unbalanced deviation exceeds the dynamic threshold.

[0020] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0021] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute the scene recognition-based adaptive management method for three-phase imbalance in transformer areas according to any embodiment of the present invention.

[0022] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores computer instructions for causing a processor to execute and implement the scene recognition-based adaptive management method for three-phase imbalance in transformer substations according to any embodiment of the present invention.

[0023] The technical solution of this invention involves collecting three-phase current data from users within a transformer substation; dynamically segmenting the data according to user electricity consumption periods, and extracting user electricity consumption pattern characteristics within each time segment; calculating a similarity matrix based on the electricity consumption pattern characteristics, constructing a user scenario mapping map based on the similarity matrix, and obtaining user scenario identifiers through map clustering; determining user three-phase imbalance assessment weights based on scenario identifiers, establishing hierarchical assessment criteria in conjunction with three-phase current data, generating the time-sharing contribution degree of each user to the three-phase imbalance in the transformer substation, and formulating personalized control strategies for users based on the time-sharing contribution degree; adding differential noise to the personalized control strategies and electricity consumption pattern characteristics for encryption, and fusing the encrypted data through multi-party data fusion using a federated learning approach to generate a multi-user balance control scheme. This system adjusts the three-phase power load of users according to the control plan, calculates the unbalanced deviation of the three-phase current after adjustment, and triggers a new round of data acquisition when the unbalanced deviation exceeds the dynamic threshold. This solves the prominent problem of three-phase imbalance in transformer areas under the growth of power load and the access of distributed energy. Traditional static adjustment methods cannot adapt to the dynamic changes in power consumption patterns. Existing methods ignore the differences in user scenarios, lack the precision of control, rely on manual experience or simple statistics for scenario identification, resulting in low accuracy, and are prone to leakage of user privacy during data fusion. This system achieves the beneficial effects of accurately reflecting the characteristics of user load changes, establishing a quantitative correlation between user and transformer area imbalance to realize hierarchical assessment and personalized control, completing multi-party data fusion while protecting data privacy, and having real-time adaptive adjustment capabilities.

[0024] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart of an adaptive management method for three-phase imbalance in transformer substations based on scene recognition, provided in an embodiment of the present invention;

[0027] Figure 2 A schematic diagram of a scene recognition-based three-phase imbalance adaptive management system for transformer substations is provided in an embodiment of the present invention.

[0028] Figure 3 A schematic diagram of the structure of an electronic device for implementing a scene recognition-based adaptive management method for three-phase imbalance in a transformer area, according to an embodiment of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] Figure 1The flowchart illustrates a scene recognition-based adaptive management method for three-phase imbalance in transformer substations, which is applicable to the management and intelligent control of three-phase imbalance in transformer substations. This method can be executed by a scene recognition-based adaptive management system for three-phase imbalance in transformer substations. This system can be implemented in hardware and / or software and can be configured in electronic devices.

[0032] like Figure 1 As shown, the method specifically includes the following steps:

[0033] S1. Collect three-phase current data of users within the transformer area.

[0034] S2. Dynamically segment the user's electricity consumption according to the user's electricity consumption time period, and extract the user's electricity consumption pattern characteristics within the time segment.

[0035] S3. Calculate the similarity matrix based on the electricity consumption pattern characteristics, construct a user scenario mapping map based on the similarity matrix, and obtain the user's scenario identifier through map clustering.

[0036] S4. Determine the user's three-phase imbalance assessment weight based on the scenario identifier, establish a hierarchical assessment criterion by combining three-phase current data, generate the user's time-sharing contribution to the three-phase imbalance of the transformer area, and formulate the user's personalized control strategy based on the time-sharing contribution.

[0037] S5. Add differential noise to the personalized control strategy and electricity consumption pattern characteristics for encryption, and then use federated learning to fuse multi-party data to generate a multi-user balanced control scheme.

[0038] S6. Adjust the user's three-phase power load according to the control plan, calculate the unbalanced deviation of the three-phase current after adjustment, and trigger a new round of data acquisition process when the unbalanced deviation exceeds the dynamic threshold.

[0039] In some embodiments, step S2 involves dynamically segmenting the user's electricity consumption according to the user's electricity consumption time period, and extracting the user's electricity consumption pattern characteristics within the time segment, including:

[0040] S2.1. Calculate the rate of change of three-phase current data collected in S1 at continuous time points, construct the power consumption fluctuation direction sequence according to the rate of change, detect the turning point of the power consumption fluctuation direction sequence, and divide the power consumption data sequence between adjacent turning points into power consumption observation data.

[0041] S2.2. Based on the electricity consumption observation data output by S2.1, perform sliding analysis to calculate the electricity consumption fluctuation amplitude within the sliding window, identify the fluctuation change location according to the growth rate of the electricity consumption fluctuation amplitude, and determine the fluctuation change location as the segmentation point of the electricity consumption period;

[0042] S2.3. Divide the electricity consumption period into segments based on the segmentation points output by S2.2, calculate the dispersion of the electricity consumption pattern within each electricity consumption period, and when the dispersion exceeds the preset threshold, determine the optimal segmentation time based on the fluctuation pattern of the electricity consumption data, and divide the electricity consumption period into sub-electricity consumption periods from the optimal segmentation time.

[0043] S2.4. Perform segmented feature extraction on the electricity consumption data of the sub-electricity consumption period output by S2.3, calculate the similarity between the electricity consumption features and historical electricity consumption features, and normalize the electricity consumption features based on the similarity to obtain the electricity consumption pattern features.

[0044] During the dynamic segmentation of user electricity consumption periods, the rate of change at consecutive time points is first calculated based on the collected three-phase current data. Specifically, for every two adjacent sampling points in the time series, the ratio of their current value difference to the time interval is calculated, forming a rate of change sequence. Based on these rate of change values, a sequence of electricity consumption fluctuation directions is constructed, that is, the positive or negative sign of the rate of change is converted into a corresponding upward or downward direction identifier. The system detects the turning points in this direction sequence, that is, the locations where the direction changes, such as points where the direction changes from upward to downward or from downward to upward. The electricity consumption data sequence between adjacent turning points is divided into a complete electricity consumption observation data unit, and these units reflect the basic fluctuation characteristics of user electricity consumption behavior.

[0045] After acquiring electricity consumption observation data, a sliding window analysis is performed to identify abrupt changes in fluctuation characteristics. The system sets a fixed-width time window, which slides across the electricity consumption observation data in predetermined steps. Within each sliding window, the fluctuation amplitude of the current data is calculated, i.e., the difference between the maximum and minimum values ​​within the window. By comparing the fluctuation amplitudes of adjacent sliding windows, the growth rate of the fluctuation amplitude is calculated, and locations where the growth rate exceeds a preset threshold are marked as fluctuation abrupt change locations. These fluctuation abrupt change locations are determined as segmentation points of electricity consumption periods, which typically correspond to transition points in user electricity consumption patterns, such as the on / off times of home appliances.

[0046] Using identified segmentation points as boundaries, user electricity consumption data is divided into multiple electricity consumption periods. The dispersion of electricity consumption patterns within each period is calculated, assessing the volatility and dispersion of the data. Dispersion can be measured by the standard deviation, coefficient of variation, or distribution characteristics of the fluctuation amplitude of the current data. When the dispersion of a particular electricity consumption period exceeds a preset threshold, it indicates that the period may contain multiple different electricity consumption patterns, requiring further subdivision. Based on the fluctuation patterns of the electricity consumption data within that period, such as the trends in fluctuation frequency and intensity, the system determines the optimal segmentation time. This electricity consumption period is then divided into multiple sub-periods from the optimal segmentation time, with each sub-period exhibiting a more consistent electricity consumption pattern.

[0047] The system extracts segmented features from the subdivided electricity consumption periods, including multi-dimensional indicators such as average electricity consumption, fluctuation amplitude, fluctuation frequency, duration, and peak-valley distribution. These indicators comprehensively reflect the user's electricity consumption behavior characteristics within that period. The system calculates the similarity between the extracted electricity consumption features and feature templates stored in the historical database, using metrics such as cosine similarity or Euclidean distance to evaluate the degree of matching between current and historical features. Based on the calculated similarity values, the electricity consumption features are normalized to ensure comparability between features from different time periods and different users, ultimately resulting in standardized electricity consumption pattern features.

[0048] In the feature normalization process, considering the power differences and operating characteristics of different electrical equipment, a piecewise linear transformation method is used to map the original feature values. For fluctuation amplitude features, they are mapped to the [0, 1] interval; for duration features, multiple reference points are set according to the operating cycle of typical electrical equipment, and piecewise mapping is performed; for fluctuation frequency features, normalization processing is performed in conjunction with the power grid frequency characteristics. This multi-dimensional feature normalization process ensures that the features extracted under different power consumption scenarios have a unified measurement standard, which facilitates subsequent pattern recognition and classification.

[0049] The scene-recognition-based adaptive management technology for three-phase imbalance in transformer substations, implemented using the methods described above, can accurately identify changes in users' electricity consumption patterns, enabling refined analysis of electricity data. This technology, through a dynamic segmentation strategy, adapts to the diverse electricity consumption behaviors of different users, overcoming the limitations of traditional fixed-time-window analysis methods. The system's multi-dimensional extraction and normalization of electricity consumption characteristics improves the accuracy of electricity consumption pattern recognition, providing a precise basis for adjusting the three-phase load balance in transformer substations.

[0050] In some embodiments, step S3 involves calculating a similarity matrix based on electricity consumption pattern characteristics, constructing a user scenario mapping map based on the similarity matrix, and obtaining the user's scenario identifier through map clustering, including:

[0051] S3.1. Divide the electricity consumption pattern features into multiple hierarchical features according to time periods, calculate the temporal correlation strength between different hierarchical features, and adaptively weight the similarity between users according to the temporal correlation strength to obtain the user similarity matrix.

[0052] S3.2. Based on the temporal correlation strength output from S3.1, the hierarchical similarity matrix is ​​fused to obtain the user similarity matrix;

[0053] S3.3. Extract the temporal variation features from the user similarity matrix output in S3.2, and map the temporal variation features to the scene feature space according to the change trend to generate a user scene mapping map;

[0054] S3.4. Calculate the feature propagation intensity between users in the user scene mapping map output by S3.3, and identify the node with the largest feature propagation intensity as the scene cluster center;

[0055] S3.5. Set the scene division threshold based on the propagation path strength calculated in S3.4. Mark user nodes with a strength greater than the scene division threshold as stable scenes and user nodes with a strength less than the scene division threshold as transitional scenes.

[0056] S3.6. Determine the scene identifier based on the evolution trajectory of the user node in the user scene mapping graph output by S3.5.

[0057] Electricity consumption patterns are categorized into multiple hierarchical features based on time periods, including hourly, daytime, and periodic features. Hourly features capture users' electricity consumption behavior over a short period, such as electricity consumption characteristics during the morning wake-up time; daytime features reflect users' electricity consumption patterns throughout the day, such as the difference between weekdays and weekends; and periodic features characterize users' electricity consumption patterns over a longer time span, such as seasonal variations. The temporal correlation strength is calculated for each hierarchical feature, and the trend of feature changes at consecutive time points is analyzed using a sliding time window to calculate the correlation coefficient between features at adjacent time points. If the correlation coefficient is higher than a preset threshold of 0.75, the feature at that hierarchical level is considered to have a strong temporal correlation. Features with high temporal correlation strength are assigned larger weights, while those with low correlation strength are assigned smaller weights, achieving adaptive weighting of user similarity.

[0058] The temporal correlation strength between hierarchical features is used for feature fusion to construct a multi-level similarity matrix. For each hierarchical feature, the feature distance between users is calculated, and the distance values ​​are converted into similarity values ​​using a Gaussian kernel function to form the user similarity matrix for that level. The bandwidth parameter of the Gaussian kernel function is adaptively adjusted according to the dispersion of the feature distribution to ensure a reasonable similarity distribution. Fusion weights are set based on the previously calculated temporal correlation strength, and the similarity matrices of each level are weighted and fused to form a unified user similarity matrix. During the fusion process, hierarchical features with higher temporal correlation strength are assigned larger weights, such as 0.4 for hourly features, 0.35 for daytime features, and 0.25 for periodic features, achieving effective integration of features across multiple time scales.

[0059] The user similarity matrix contains rich temporal variation features. By calculating the change in the similarity matrix between adjacent time windows through difference, a similarity change rate sequence is formed. A moving average filter is applied to the change rate sequence to eliminate the interference of random fluctuations and extract stable change trends. Based on the direction and intensity of the change trends, users are mapped to a predefined scene feature space. The scene feature space consists of two dimensions: electricity consumption behavior features and time features. Electricity consumption behavior features include indicators such as electricity consumption magnitude and fluctuation frequency, while time features include attributes such as duration and periodicity. By connecting the user's electricity consumption features in different time windows according to the change trends in the scene feature space, a trajectory is formed, generating a user scene mapping map.

[0060] In a user scenario mapping graph, the feature propagation intensity among users is calculated, drawing inspiration from the information flow propagation model in community detection algorithms. Starting from each user node, the diffusion process of feature information in the graph is simulated, recording the intensity attenuation of feature information propagating from the source node to other nodes. The propagation intensity is directly proportional to the similarity between user nodes and inversely proportional to the path distance. Through multiple random walk experiments, the average propagation efficiency of each node as an information source is statistically analyzed. The node with the highest feature propagation intensity is identified and designated as the scenario cluster center. These center nodes typically represent typical electricity consumption scenarios, such as the characteristic electricity consumption patterns of residential areas, commercial areas, or industrial areas.

[0061] A scene partitioning threshold is set based on the strength of the feature propagation path. The propagation path strength between each user node and the scene cluster center is calculated; the path strength is calculated by multiplying the similarity between nodes. An adaptive threshold is set, determined based on the global propagation strength distribution, typically the average global propagation strength plus or minus the standard deviation. User nodes with values ​​greater than the scene partitioning threshold are marked as stable scenes; these users have relatively fixed power consumption patterns and clear scene affiliations. User nodes with values ​​less than the scene partitioning threshold are marked as transitional scenes; these users exhibit power consumption patterns that overlap across multiple scenes or change frequently, requiring further analysis to determine their scene affiliation.

[0062] The scene identifier is determined based on the evolutionary trajectory of user nodes in the user scene mapping map during transitional scenarios. The positional changes of user nodes within a continuous time window are analyzed to extract the main direction and fluctuation amplitude of their trajectories. For users with small fluctuation amplitudes, scene identifiers are determined based on the stable scene cluster centers their trajectories are close to. For users with large fluctuation amplitudes, the periodic features of their trajectories are extracted, recurring patterns are identified, users are classified into dynamic scene categories, and their primary and secondary activity scenes are labeled. For users who are difficult to classify in special cases, fuzzy scene identifiers are used, and multiple possible scene affiliations and corresponding probability values ​​are recorded to provide a more comprehensive reference for subsequent three-phase balance adjustments.

[0063] The aforementioned adaptive management method for three-phase imbalance in transformer substations based on scene recognition achieves accurate identification and classification of user electricity consumption scenarios. This method overcomes the shortcomings of traditional transformer substation load management in terms of insufficient user scenario identification. Through multi-level feature fusion and time-series correlation analysis, it captures the dynamic changes in user electricity consumption behavior. The graph-based scenario clustering and labeling method effectively distinguishes between users in stable and transitional scenarios, improving the accuracy of scenario classification.

[0064] In some embodiments, step S4 determines the user's three-phase imbalance assessment weight based on the scenario identifier, establishes a hierarchical assessment criterion by combining three-phase current data, generates the user's time-sharing contribution to the three-phase imbalance of the transformer area, and formulates a personalized control strategy for the user based on the time-sharing contribution, including:

[0065] S4.1. Extract the electricity consumption period sequence based on the scene identifier, and calculate the amplitude variation characteristics of the three-phase current in the electricity consumption period sequence;

[0066] S4.2. Divide the amplitude variation characteristics of the output of S4.1 into ground state components and disturbance components according to the duration, and determine the user's three-phase imbalance assessment weight based on the superposition effect of ground state components and disturbance components.

[0067] S4.3. The three-phase current data is weighted using the user three-phase imbalance assessment weights output from S4.2 to obtain the hierarchical assessment benchmark. The assessment hierarchical criteria are generated based on the gradient of the hierarchical assessment benchmark to determine the user three-phase imbalance contribution value.

[0068] S4.4. Perform correlation analysis between the user's three-phase imbalance contribution value output from S4.3 and the three-phase imbalance degree of the transformer area to generate the user's time-sharing contribution to the three-phase imbalance of the transformer area;

[0069] S4.5. Determine the control parameters based on the time-series characteristics of the time-division contribution output in S4.3, and combine the control parameters with the superposition effect of the ground state component and the disturbance component to generate the user's personalized control strategy.

[0070] The system retrieves user scenario tags from a scenario identifier database, such as residential scenarios, commercial office scenarios, or industrial production scenarios. These scenario tags are then used to extract corresponding electricity consumption time-period sequences. These sequences contain the user's electricity consumption characteristics at different times, such as daily living electricity consumption from 6:00 AM to 9:00 AM and work electricity consumption from 9:00 AM to 5:00 PM. For the extracted electricity consumption time-period sequences, the amplitude variation characteristics of the three-phase current are calculated. These characteristics include the mean, peak value, fluctuation range, and rate of change of the three-phase current. By comparing the relative differences in the three-phase currents, key periods of three-phase load imbalance are identified. For example, the current differences between phases A and B, B and C, and C and A are calculated to form a current difference sequence. The changing trends of the current difference sequence are analyzed to determine the time points and durations of the most severe three-phase imbalance.

[0071] The amplitude variation characteristics are divided into a ground-state component and a disturbance component based on their duration. The ground-state component represents the user's long-term stable power consumption pattern, which typically lasts for a long time, such as the continuous operation of basic household lighting loads or industrial infrastructure equipment. The disturbance component represents short-term power consumption fluctuations, such as the start-up and shutdown of air conditioners or the operation of elevators, which involve instantaneous high-power consumption. The distinction is made by applying a low-pass filter to the current data, setting the filter cutoff frequency to the frequency value corresponding to a 1-hour period, and using the filtered result as the ground-state component; subtracting the ground-state component from the original current data yields the disturbance component. The superposition effect of the ground-state component and the disturbance component determines the user's actual contribution to the three-phase imbalance. The user's three-phase imbalance assessment weight is determined based on the superposition effect. The calculation method is as follows: assign a weight coefficient of 0.7 to the ground-state component and a weight coefficient of 0.3 to the disturbance component, and then use the weighted sum of the two as the user's three-phase imbalance assessment weight. For users with frequent and large disturbances, the weight coefficient of the disturbance component can be appropriately increased, up to a maximum of 0.5.

[0072] Three-phase current data is weighted using user three-phase imbalance assessment weights to obtain a tiered assessment benchmark. The three-phase current data is then multiplied by the corresponding assessment weight to generate weighted three-phase current values. A tiered assessment system is constructed based on the weighted current data, classifying users into high, medium, and low levels according to their imbalance contribution. The gradient of the assessment benchmark is used to generate the tiered assessment criteria; the gradient is calculated as the difference between the assessment benchmarks at adjacent time points divided by the time interval. When the gradient value is greater than a preset threshold, it is determined to be a high-level imbalance contribution; when the gradient value is between two thresholds, it is determined to be a medium-level imbalance contribution; and when the gradient value is less than the lower threshold, it is determined to be a low-level imbalance contribution. The user's three-phase imbalance contribution value is determined according to the tiered criteria: the contribution value range for high-level users is 0.8-1.0, for medium-level users it is 0.4-0.7, and for low-level users it is 0.1-0.3. Users may belong to different contribution levels at different times, forming a dynamic contribution sequence.

[0073] This study correlates user three-phase imbalance contributions with the three-phase imbalance degree of the transformer area to generate a time-sharing contribution level of users to the three-phase imbalance. The three-phase imbalance degree of the transformer area is calculated using the three-phase current imbalance rate, which is the ratio of the maximum deviation to the average value of the three-phase current. Three-phase current data from all users in the transformer area are collected, and the time-series variation curve of the overall three-phase imbalance degree of the transformer area is calculated. User contribution values ​​and transformer area imbalance degrees are compared synchronously over time, and their correlation coefficient is calculated. A higher correlation coefficient indicates a greater impact of users on the transformer area's imbalance. Based on the correlation analysis results, the contribution ratio of users to the three-phase imbalance of the transformer area in each time period is calculated, forming a time-sharing contribution level table. For example, a user's contribution during morning and evening peak hours may be as high as 30%, while at other times it may only be around 5%.

[0074] Control parameters are determined based on the temporal characteristics of time-of-use contribution. Daily, weekly, and seasonal variation patterns of user contribution are analyzed to extract peak contribution periods and their duration. Differentiated control parameters are set for different characteristic patterns, including control intensity coefficient, control duration coefficient, and control response coefficient. The control intensity coefficient determines the proportion of load transfer, the control duration coefficient determines the duration of control measures, and the control response coefficient reflects the user's acceptance of the control measures. The control parameters are combined with the superposition effect of ground-state and disturbance components to generate personalized control strategies for users. For users where the ground-state component is dominant, a long-term stable phase adjustment strategy is adopted; for users with significant disturbance components, a dynamic load transfer strategy during peak and off-peak periods is adopted. Personalized control strategies include specific control period selection, phase adjustment schemes, and load transfer suggestions, such as adjusting some of the user's single-phase load from the heavily loaded phase to the lightly loaded phase, or suggesting that users use electricity during off-peak hours at specific times.

[0075] The adaptive management method for three-phase imbalance in transformer substations based on scene recognition enables precise assessment and personalized control of the contribution of three-phase imbalance to users in the substation area. This method overcomes the problem of insufficient sensitivity to short-term power fluctuations in traditional assessment methods by distinguishing the contributions of the ground-state component and the disturbance component; it achieves differentiated management for different types of users through a hierarchical assessment system; and based on time-sharing contribution analysis, it accurately identifies the impact patterns of users on substation imbalance, providing a basis for precise control.

[0076] In some embodiments, step S5 involves adding differential noise to encrypt the personalized control strategy and electricity consumption pattern characteristics, and then fusing the encrypted data through federated learning to generate a multi-user balanced control scheme, including:

[0077] S5.1. Extract the temporal change trend of personalized control strategies and electricity consumption pattern characteristics based on scene identifiers, calculate the stability characteristics of the scene, and establish time-segmented sensitivity evaluation criteria.

[0078] S5.2. Combine the sensitivity evaluation criteria output by S5.1 with the scene stability characteristics to generate differential noise, and then superimpose the differential noise onto the personalized control strategy and the power consumption pattern characteristics for encryption.

[0079] S5.3. Extract the fluctuation direction sequence of the encrypted data output by S5.2, associate and map the fluctuation direction sequence with the scene stability features, calculate the gradient information, receive gradient information uploaded by multiple users, calculate the temporal correlation strength of the gradient information, perform weighted aggregation of the gradient information based on the temporal correlation strength, and update the fusion parameters based on the aggregation result;

[0080] S5.4. Based on the scene stability characteristics in S5.1, separate the user ground state component and disturbance component, determine the time-division contribution level, use the time-division contribution level to weight the fusion parameters, perform multi-party data fusion on the encrypted data, and generate a multi-user balanced control scheme.

[0081] A time window analysis was performed on electricity consumption data to calculate the standard deviation, peak factor, and coefficient of variation of current fluctuations within the window. For residential users, electricity consumption fluctuations increased significantly in the morning and evening; for commercial users, electricity consumption remained relatively stable during working hours; and for industrial users, load changes due to production cycles showed obvious regularity. The stability characteristics of the scenario were calculated, using the ratio of the variance to the mean of user electricity consumption data as a stability index; the smaller the value, the more stable the scenario. A time-segmented sensitivity assessment criterion was established, assigning weight coefficients to the sensitivity of different time periods based on electricity consumption patterns. Sensitivity was assessed by the degree of response of electricity consumption characteristics to external disturbances; periods with high sensitivity were assigned higher protection weights, and periods with low sensitivity were assigned lower protection weights. For example, industrial users had high sensitivity during peak production periods, with a weight of 0.8-0.9; residential users had low sensitivity during late-night hours, with a weight of 0.2-0.3.

[0082] Differential noise is generated by combining sensitivity assessment criteria with scenario stability characteristics to protect user data privacy. The differential noise generation employs a Laplace mechanism, with noise amplitude directly proportional to sensitivity and inversely proportional to the privacy budget. Differential privacy budgets are set for different scenarios: smaller budget values ​​(e.g., 0.1) are set for scenarios with high stability, while larger budget values ​​(e.g., 0.5) are set for scenarios with low stability. The generated differential noise is superimposed on personalized control strategies and electricity consumption pattern characteristics to form encrypted data. For control strategies, noise is mainly added to control time points and control intensity parameters; for electricity consumption pattern characteristics, noise is added to load distribution and fluctuation amplitude data. The encryption process ensures that, while meeting differential privacy requirements, the data retains its original statistical characteristics and trends, enabling subsequent analysis to be based on the encrypted data.

[0083] The algorithm extracts the fluctuation direction sequence from the encrypted data to capture the trend characteristics of data changes. The fluctuation direction sequence is determined by comparing data values ​​at adjacent time points: values ​​greater than the previous time point are considered an increase, values ​​less than are considered a decrease, and values ​​equal are considered stable. The fluctuation direction sequence is then mapped to scene stability features to construct a feature vector containing three dimensions: fluctuation direction, fluctuation duration, and scene stability. Gradient information is calculated based on the feature vector. The gradient contains the direction and strength of the influence of the control parameters on the system balance. Gradient calculation uses a federated learning framework, where each user calculates the gradient locally and then uploads only the gradient information, not the original data. The algorithm receives gradient information uploaded by multiple users and calculates the temporal correlation strength of the gradient information. The correlation strength is obtained by calculating the cross-correlation function of the gradient sequences from different users, with a value ranging from -1 to 1; a larger value indicates a stronger correlation. The gradient information is then weighted and aggregated according to the temporal correlation strength, with gradients of high correlation strength assigned higher weights and gradients of low correlation strength assigned lower weights, forming a global gradient. The fusion parameters are updated based on the aggregation results. The fusion parameters include the temporal distribution, intensity allocation, and phase adjustment scheme of the control strategy.

[0084] Based on the stability characteristics of the scene, the user's ground state component and disturbance component are separated, and wavelet transform is used to decompose the electricity consumption data into multiple scales. The low-frequency component corresponds to the ground state component, representing the user's long-term stable electricity consumption behavior; the high-frequency component corresponds to the disturbance component, representing short-term electricity consumption fluctuations. For users with high scene stability, the ground state component is dominant; for users with low scene stability, the disturbance component has a larger proportion. The time-sharing contribution level is determined to assess the impact of each user on the three-phase imbalance of the transformer area at different times. The contribution level is calculated by the correlation between the user's three-phase load imbalance and the overall imbalance of the transformer area, with a value range of 0 to 1, where a larger value indicates a greater contribution. The time-sharing contribution level is used to weight the fusion parameters, with users with high contribution levels corresponding to larger weights and users with low contribution levels corresponding to smaller weights. Multi-party data fusion is performed on the encrypted data, and secure multi-party computation technology is used to achieve data sharing and collaborative analysis. The fusion process ensures that the data privacy of all parties is not leaked, while achieving the globally optimal balance control effect. Based on the fusion results, a multi-user balance control scheme is generated, including load allocation schemes, peak-valley balancing strategies, and phase adjustment suggestions.

[0085] For load sharing schemes, some single-phase loads of high-contribution users are shifted from heavily loaded phases to lightly loaded phases, with the adjustment ratio determined based on the user's contribution to the imbalance. For peak-valley balancing strategies, peak load periods in the distribution area are identified, guiding users to stagger their electricity consumption and reducing the concentration of electricity consumption on the same phase. For phase adjustment recommendations, based on user electricity consumption patterns and scenario stability, an overall phase adjustment scheme is proposed, such as changing the access phase of certain users from phase A to phase B. The control scheme considers user acceptance and implementation difficulty, prioritizing control measures with low adjustment costs and significant effects. The scheme is implemented gradually, first adjusting key users with high contributions, and then gradually expanding to other user groups. During the control process, the three-phase imbalance in the distribution area is continuously monitored, and the scheme parameters are dynamically adjusted based on the actual results to achieve closed-loop feedback control.

[0086] The scene-recognition-based adaptive management method for three-phase imbalance in transformer substations addresses data security and multi-party collaboration issues in traditional control schemes by introducing differential privacy protection and federated learning techniques. This method achieves secure fusion and analysis of multi-user data while protecting user privacy, overcoming the limitations of insufficient adaptability of single control strategies. Through precise calculation of scene stability characteristics and time-sharing contribution, it enables accurate assessment of user impact and the formulation of targeted control schemes.

[0087] In some embodiments, step S5.1 involves extracting the temporal variation trends of personalized control strategies and electricity consumption pattern characteristics based on the scenario identifier, calculating the stability characteristics of the scenario, and establishing time-segmented sensitivity assessment criteria, including:

[0088] S6.1. Extract time-series data of personalized control strategies and power consumption patterns based on scene identifiers, calculate the change in data between adjacent time points to obtain time-series difference sequences, use adaptive asymmetric variational decomposition to obtain fluctuation components and trend components, calculate energy entropy of fluctuation components through wavelet packet transform to obtain contribution weights, and reconstruct the time-series change trend based on contribution weights.

[0089] S6.2. Based on the contribution weight in S6.1, adaptively determine the time window length, calculate the cumulative variance feature vector of the fluctuation component within the window, use the phase correlation of the cumulative variance feature vector as the scene stability feature, and obtain the scene transition probability based on the changing trend of the scene stability feature.

[0090] S6.3. Construct a nonlinear mapping function using the energy distribution of the cumulative variance eigenvector in S6.2, map the scene transition probability to the sensitivity space, adjust the characteristics of the mapping function according to the scene stability characteristics, and generate time-segmented sensitivity evaluation criteria.

[0091] Based on scenario identifiers, time-series data of personalized control strategies and electricity consumption patterns are extracted, and user electricity consumption records and control response data for the corresponding scenarios are retrieved from the database. The extracted time-series data includes electrical parameters such as three-phase current values, phase angles, and power factors, as well as control strategy parameters such as control time points, control intensity, and control duration. The change in data between adjacent time points is calculated to obtain a time-series difference sequence, i.e., subtracting the data value of the previous time point from the data at each time point to form a difference sequence reflecting the rate of data change. An adaptive asymmetric variational decomposition method is used to process the difference sequence. This method iteratively optimizes the time-series signal into fluctuation components and trend components. The fluctuation component represents short-term random fluctuations, while the trend component represents long-term variation patterns. During the decomposition process, an adaptive penalty factor is set, assigning different weights to different frequency components, with lower weights for high-frequency components and higher weights for low-frequency components. For scenarios with significant changes in electricity consumption, such as the start-up and shutdown times of industrial production, the penalty factor is adjusted to reduce the suppression of high-frequency components. The decomposed fluctuation components are subjected to wavelet packet transform using the db4 wavelet basis function, with a decomposition level of 5. The energy entropy value of each decomposition node is calculated. The smaller the energy entropy, the greater the amount of information contained in the node, and its contribution weight is set accordingly higher. Based on the calculated contribution weights, the trend components are reconstructed, retaining the main features while filtering out noise interference, and finally obtaining the time-series change trend that reflects the essential characteristics of electricity consumption behavior.

[0092] The time window length is adaptively determined based on contribution weights, with the window length positively correlated with the weights. For periods with higher weights, such as peak morning electricity consumption, a longer window length (e.g., 60 minutes) is set; for periods with lower weights, such as off-peak hours at night, a shorter window length (e.g., 15 minutes) is set. Within each time window, the cumulative variance eigenvector of the fluctuation component is calculated. Specifically, the variance values ​​of the fluctuation components at each moment within the window are sequentially accumulated to form a cumulative variance curve that increases over time. This curve is then discretized into equally spaced eigenvectors. The correlation coefficient between the cumulative variance eigenvectors of different phases is calculated as a feature of scene stability. The correlation coefficient ranges from -1 to 1. A value closer to 1 indicates a more consistent trend between phases and a more stable scene; a value closer to -1 indicates a more opposite trend between phases and a more severe imbalance; a value close to 0 indicates no significant correlation between phases and the system is in a state of random fluctuation. Based on the changing trend of the scene stability features within continuous time windows, a Markov state transition model is constructed to calculate the probability of the scene transitioning from the current stable state to other stable states. The transition probability reflects the dynamic characteristics of scene changes and provides a basis for sensitivity assessment.

[0093] A nonlinear mapping function is constructed using the energy distribution of the cumulative variance eigenvector to map the scene transition probability to the sensitivity space. The energy distribution is obtained by calculating the sum of squares of each component of the eigenvector, and then normalized to form an energy distribution vector. A piecewise sigmoid function is constructed based on the energy distribution vector. The input of the function is the scene transition probability, and the output is the sensitivity evaluation value. The function design includes three key parameters: the midpoint value, the gain coefficient, and the saturation threshold. The midpoint value is set as the median of the transition probability, representing the inflection point of sensitivity change; the gain coefficient controls the slope of the function, affecting the sensitivity to changes in transition probability; the saturation threshold limits the output range of the function to prevent the sensitivity evaluation from being too high or too low in extreme cases. The characteristics of the mapping function are adjusted according to the scene stability features. For scenes with high stability, the gain coefficient is increased to make the sensitivity more sensitive to changes in transition probability; for scenes with low stability, the gain coefficient is decreased to reduce the volatility of the sensitivity evaluation. A time-segmented sensitivity assessment criterion is generated, dividing the 24 hours of a day into several time periods, such as early morning (5:00-8:00), working hours (8:00-17:00), evening peak (17:00-22:00), and nighttime (22:00-5:00). An average sensitivity value is calculated for each time period, and a time-segment sensitivity comparison table is established as the basis for subsequent differential privacy protection and control strategies.

[0094] In residential electricity consumption scenarios, sensitivity is typically high during early morning and evening peak hours, with sensitivity assessment values ​​reaching 0.8-0.9; sensitivity is lower during midday and late night, with assessment values ​​around 0.2-0.3. In commercial office scenarios, sensitivity is relatively stable during working hours, with assessment values ​​remaining at 0.5-0.7; sensitivity is lower during non-working hours, with assessment values ​​around 0.1-0.3. In industrial production scenarios, sensitivity is highest at production cycle transition points, with assessment values ​​reaching 0.9-1.0; sensitivity is moderate during stable production periods, with assessment values ​​around 0.4-0.6. The assessment criteria also consider seasonal factors, such as significantly higher sensitivity during peak summer air conditioning load periods compared to other seasons. The sensitivity assessment criteria provide a precise privacy budget allocation strategy for subsequent differential privacy protection, allocating more privacy budget to periods with higher sensitivity to ensure that data protection strength matches data value. Simultaneously, the assessment criteria also provide a basis for the refined design of control strategies, employing differentiated control methods and intensities for periods with different levels of sensitivity.

[0095] To achieve dynamic adjustment of the sensitivity assessment criteria, a feedback mechanism is established. The system monitors changes in the three-phase balance in the control backend area, calculates the difference between the actual balance improvement and the expected target, and uses this difference as a feedback signal. Sensitivity assessment parameters, including the midpoint value of the mapping function, gain coefficient, and saturation threshold, are adjusted based on the feedback signal. When the actual effect is lower than expected, the sensitivity assessment value is appropriately reduced, and the control intensity is increased; when the actual effect exceeds expectations, the sensitivity assessment value is appropriately increased, and the control intensity is reduced. Through this feedback adjustment mechanism, the sensitivity assessment criteria are adaptively matched to the actual application scenario, improving the system's stability and reliability. The assessment criteria are stored in the system database in the form of a numerical index table, indexed by time period, scenario type, and user characteristics for easy retrieval and application.

[0096] The time-segmented sensitivity assessment criterion in the adaptive management method for three-phase imbalance in transformer substations based on scene recognition accurately characterizes user electricity consumption behavior and privacy protection needs. By combining adaptive asymmetric variational decomposition and wavelet packet transform, the fluctuation and trend components in electricity consumption data are effectively separated, overcoming the limitations of traditional time-series analysis methods in processing non-stationary signals. Phase correlation analysis based on cumulative variance eigenvectors accurately assesses scene stability characteristics, providing a reliable basis for sensitivity mapping. The design of a nonlinear mapping function achieves precise mapping from scene transition probabilities to the sensitivity space, establishing the theoretical foundation for time-segmented sensitivity assessment.

[0097] In some embodiments, step S6 involves adjusting the user's three-phase electrical load according to the control scheme, calculating the adjusted three-phase current imbalance deviation, and triggering a new round of data acquisition when the imbalance deviation exceeds a dynamic threshold. This includes:

[0098] S7.1. Read the target adjustment value of each phase according to the execution sequence of the control scheme output by S5.4, adjust the load of each phase to the target adjustment value, and obtain the adjusted three-phase power load of the user;

[0099] S7.2. Calculate the difference between the adjusted phase current and the three-phase average current output from S7.1. Divide the maximum value of the difference by the average current to obtain the three-phase current imbalance deviation.

[0100] S7.3. Set the product of the rate of change of the three-phase current imbalance deviation output by S7.2 and the historical fluctuation period as the monitoring window length. The three-phase current imbalance deviation within the monitoring window is weighted to obtain a dynamic threshold. When the three-phase current imbalance deviation exceeds the dynamic threshold, a new round of data acquisition is triggered.

[0101] When adjusting a user's three-phase power load according to the control plan, the target adjustment value for each phase must be read according to the preset execution order. The execution order is determined by the user's contribution to the three-phase imbalance, with users contributing more being adjusted first. The control plan includes specific target adjustment values, clearly indicating the values ​​that each phase load should be adjusted to. The target adjustment values ​​are generated through a multi-user balancing control plan, taking into account the load distribution of the entire distribution area. When reading the target adjustment values, a time-segmented strategy is adopted, dividing the day into 24 time periods, with the corresponding target adjustment value read for each time period. For key users, the adjustment accuracy can reach once every 15 minutes; for ordinary users, the adjustment frequency can be reduced to once per hour. During the process of adjusting the load of each phase to the target adjustment value, a gradual adjustment strategy is adopted to avoid grid fluctuations caused by sudden load changes. The adjustment methods include three means: phase switching, load transfer, and intelligent control. Phase switching changes the entire user's access phase and is suitable for single-phase users; load transfer moves part of the user's load from one phase to another and is suitable for multi-phase users; intelligent control automatically adjusts the working status of electrical equipment according to a schedule through intelligent terminal devices and is suitable for users with conditions for intelligent transformation. After the adjustment is completed, the adjusted three-phase power load data of the user is obtained, including parameters such as three-phase current value, power factor and phase angle.

[0102] The difference between the adjusted current of each phase and the average three-phase current is calculated to evaluate the adjustment effect. The calculation method is as follows: First, the average value of the three-phase current is obtained by adding the current values ​​of phases A, B, and C and dividing by 3. Then, the difference between the current of each phase and the average current is calculated, resulting in three difference data points representing the deviation of phase A, phase B, and phase C, respectively. The absolute value of the three difference data points is taken to identify the maximum deviation value and its corresponding phase. The maximum deviation value is divided by the average current value to obtain the three-phase current imbalance deviation, which reflects the degree of imbalance of the three-phase load. The ideal value of the three-phase current imbalance deviation is zero, indicating complete three-phase balance; in practical applications, an imbalance deviation of less than 20% is usually considered acceptable. For important distribution areas, the imbalance deviation control target can be further reduced to less than 10%. The current data collected when calculating the imbalance deviation needs to be filtered to eliminate the influence of random interference on the calculation results. The filtering method uses a combination of median filtering and moving average, with a window length of 5 sampling points, which can effectively suppress pulse interference while retaining the true trend of current changes.

[0103] The monitoring window length is set by multiplying the rate of change of the three-phase current imbalance deviation by the historical fluctuation period. The rate of change is obtained by dividing the difference in imbalance deviation between two consecutive monitoring points by the time interval, reflecting the speed of change in the imbalance condition. The historical fluctuation period refers to the typical cycle of user load changes, obtained through spectral analysis of historical electricity consumption data. For residential users, the typical cycle is usually 24 hours; for commercial users, the typical cycle may be weekly or monthly; for industrial users, the typical cycle may be related to the production cycle. Multiplying the rate of change by the historical fluctuation period yields the adaptive monitoring window length. When the rate of change is large, the monitoring window is shortened to improve system response speed; when the rate of change is small, the monitoring window is extended to reduce unnecessary system operations. The monitoring window length is usually set between 1 hour and 24 hours to ensure that the system can respond promptly to rapid changes without being frequently triggered by short-term fluctuations.

[0104] The three-phase current imbalance deviation within the monitoring window is weighted to generate a dynamic threshold. The weighting method employs an exponentially decaying weighting function, assigning higher weights to recent imbalance deviation data and lower weights to older data. The decay rate of the weighting coefficients is related to the characteristics of the user scenario; a smaller decay rate (e.g., 0.05) is set for stable scenarios, while a larger decay rate (e.g., 0.2) is set for fluctuating scenarios. The weighted imbalance deviation data is then statistically analyzed to determine the dynamic threshold, including statistical characteristics such as mean, standard deviation, and extreme values. The dynamic threshold is typically set as the weighted mean plus 1.5 times the weighted standard deviation, balancing system sensitivity and stability. The dynamic threshold adaptively adjusts with changes in electricity consumption patterns, avoiding oversensitivity or sluggishness caused by fixed thresholds. When the actual three-phase current imbalance deviation exceeds the dynamic threshold, the system determines that the current control scheme is ineffective and requires a reassessment of the user's electricity consumption, triggering a new round of data collection.

[0105] Triggering a new round of data acquisition involves two key steps: determining the data acquisition range and setting the acquisition frequency. The data acquisition range is determined based on the extent of imbalance deviation. When the deviation only slightly exceeds the threshold, it can be limited to users with high contribution rates; when the deviation significantly exceeds the threshold, it is expanded to all users in the entire distribution area. The acquisition frequency is also related to the degree of deviation; the larger the deviation, the higher the acquisition frequency. For deviations significantly exceeding the threshold, the acquisition frequency can be increased to once every 5 minutes; for deviations slightly exceeding the threshold, the acquisition frequency can be set to once every 30 minutes. The new round of data acquisition will re-execute the entire process from steps S1 to S5, including three-phase current data acquisition, power consumption pattern feature extraction, scene identifier generation, three-phase imbalance assessment, and control scheme formulation. The entire cycle constitutes a closed-loop control system, ensuring that the three-phase load in the distribution area remains in a balanced state. A random delay mechanism is introduced during the acquisition process to avoid communication network congestion caused by all acquisition devices working simultaneously. The delay time ranges from 0 to 60 seconds and follows a uniform distribution.

[0106] By implementing a scenario-based adaptive management method for three-phase imbalance in transformer substations, precise control and dynamic monitoring of users' three-phase power load were achieved. This method overcomes the poor adaptability to load changes inherent in traditional fixed-parameter monitoring methods by setting adaptive monitoring windows and dynamic thresholds; it avoids load oscillations during the control process through multi-level control strategies and gradual adjustment mechanisms; and it ensures long-term stable operation of the system through closed-loop feedback control.

[0107] By dynamically segmenting and extracting electricity consumption pattern characteristics, the system accurately reflects user load variation characteristics. This method establishes a quantitative contribution relationship between users and three-phase imbalance in transformer areas, enabling hierarchical assessment and personalized control. By introducing differential privacy and federated learning techniques, multi-party data fusion is achieved while protecting user data privacy. The system possesses adaptive adjustment capabilities, dynamically adjusting strategies based on real-time imbalance conditions.

[0108] Figure 2 This is a schematic diagram of a scene-recognition-based adaptive management system for three-phase imbalance in a transformer substation, provided as an embodiment of the present invention. Figure 2 As shown, the system includes:

[0109] The three-phase current acquisition module 210 is used to acquire three-phase current data of users in the transformer area;

[0110] The electricity consumption feature extraction module 220 is used to dynamically segment the user's electricity consumption according to the user's electricity consumption time period and extract the user's electricity consumption pattern features within the time segment.

[0111] The scene identifier determination module 230 is used to calculate a similarity matrix based on the electricity consumption pattern characteristics, construct a user scene mapping map based on the similarity matrix, and obtain the user's scene identifier through map clustering;

[0112] The control strategy formulation module 240 is used to determine the user's three-phase imbalance assessment weight based on the scenario identifier, establish a hierarchical assessment criterion in combination with the three-phase current data, generate the user's time-sharing contribution to the three-phase imbalance of the transformer area, and formulate the user's personalized control strategy based on the time-sharing contribution.

[0113] The control scheme generation module 250 is used to add differential noise to the personalized control strategy and the electricity consumption pattern characteristics for encryption, and to fuse the encrypted data through multi-party data using federated learning to generate a multi-user balanced control scheme.

[0114] The power load control module 260 is used to adjust the user's three-phase power load according to the control scheme, calculate the unbalanced deviation of the adjusted three-phase current, and trigger a new round of data acquisition process when the unbalanced deviation exceeds the dynamic threshold.

[0115] In some embodiments, the electricity consumption feature extraction module 220 is specifically used for:

[0116] The rate of change at continuous time points is calculated based on the three-phase current data. A power consumption fluctuation direction sequence is constructed based on the rate of change. The turning points of the power consumption fluctuation direction sequence are detected, and the power consumption data sequence between adjacent turning points is divided into power consumption observation data.

[0117] Based on the electricity consumption observation data, a sliding analysis is performed to calculate the electricity consumption fluctuation amplitude within the sliding window. The location of the fluctuation change is identified according to the growth rate of the electricity consumption fluctuation amplitude, and the fluctuation change location is determined as the segmentation point of the electricity consumption period.

[0118] Divide the electricity consumption period into the segmentation points, calculate the dispersion of the electricity consumption pattern in each electricity consumption period, and when the dispersion exceeds a preset threshold, determine the optimal segmentation time based on the fluctuation pattern of the electricity consumption data, and divide the electricity consumption period into sub-electricity consumption periods from the optimal segmentation time.

[0119] The electricity consumption data of the sub-electricity consumption period is segmented and feature extracted. The extracted electricity consumption features are compared with historical electricity consumption features. The electricity consumption features are then normalized based on the similarity to obtain the electricity consumption pattern features.

[0120] In some embodiments, the scene identification determination module 230 is specifically used for:

[0121] The electricity consumption pattern features are divided into multiple hierarchical features according to time periods. The temporal correlation strength between different hierarchical features is calculated. The user similarity matrix is ​​obtained by adaptively weighting the similarity between users based on the temporal correlation strength.

[0122] The user similarity matrix is ​​obtained by fusing the hierarchical similarity matrix based on the temporal correlation strength.

[0123] Extract the temporal change features from the user similarity matrix, and map the temporal change features to the scene feature space according to the change trend to generate a user scene mapping map.

[0124] In the user scene mapping map, the feature propagation intensity between users is calculated, and the node with the highest feature propagation intensity is identified as the scene cluster center.

[0125] A scene division threshold is set based on the calculated propagation path strength. User nodes with a strength greater than the scene division threshold are marked as stable scenes, and user nodes with a strength less than the scene division threshold are marked as transitional scenes.

[0126] The scene identifier is determined based on the evolution trajectory of the user node in the transition scene in the user scene mapping graph.

[0127] In some embodiments, the regulation strategy formulation module 240 is specifically used for:

[0128] Based on the scene identifier, extract the electricity consumption period sequence and calculate the amplitude variation characteristics of the three-phase current in the electricity consumption period sequence;

[0129] The amplitude variation characteristics are divided into ground state components and disturbance components according to their duration, and the user's three-phase imbalance assessment weight is determined based on the superposition effect of the ground state components and the disturbance components.

[0130] The three-phase current data are weighted using the user three-phase imbalance assessment weight to obtain a hierarchical assessment benchmark. The assessment hierarchical criterion is generated based on the change gradient of the hierarchical assessment benchmark to determine the user three-phase imbalance contribution value.

[0131] The user's three-phase imbalance contribution value is correlated with the three-phase imbalance degree of the transformer area to generate the user's time-sharing contribution to the three-phase imbalance of the transformer area.

[0132] The control parameters are determined based on the time-series characteristics of the time-division contribution, and the control parameters are combined with the superposition effect of the ground state component and the disturbance component to generate the personalized control strategy.

[0133] In some embodiments, the control scheme generation module 250 includes:

[0134] The time-segment sensitivity assessment criterion construction sub-module is used to extract the time-series change trends of the personalized control strategy and the electricity consumption pattern characteristics based on the scenario identifier, calculate the scenario stability characteristics, and establish the time-segment sensitivity assessment criterion.

[0135] The power consumption pattern feature encryption submodule is used to combine the sensitivity assessment criteria with the scene stability features to generate differential noise, and then superimpose the differential noise onto the personalized control strategy and the power consumption pattern features for encryption.

[0136] The fusion parameter update submodule is used to extract the fluctuation direction sequence of the encrypted data, associate and map the fluctuation direction sequence with the scene stability features, calculate gradient information, receive the gradient information uploaded by multiple users, calculate the temporal correlation strength of the gradient information, perform weighted aggregation on the gradient information according to the temporal correlation strength, and update the fusion parameters based on the aggregation result.

[0137] The control scheme generation submodule is used to separate the user ground state component and the user disturbance component according to the stability characteristics of the scenario, determine the time-division contribution degree, use the time-division contribution degree to weight the fusion parameters, perform multi-party data fusion on the encrypted data, and generate the control scheme.

[0138] In some embodiments, the time-segmented sensitivity assessment criterion construction submodule is specifically used for:

[0139] The time-series data of the personalized control strategy and power consumption mode are extracted based on the scene identifier. The change of the data at adjacent time points is calculated to obtain the time-series difference sequence. Adaptive asymmetric variational decomposition is used to obtain the fluctuation component and the trend component. The energy entropy of the fluctuation component is calculated by wavelet packet transform to obtain the contribution weight. The time-series change trend is reconstructed based on the contribution weight.

[0140] The time window length is adaptively determined based on the contribution weight, the cumulative variance feature vector of the fluctuation component within the window is calculated, the phase correlation of the cumulative variance feature vector is used as the scene stability feature, and the scene transition probability is obtained based on the changing trend of the scene stability feature.

[0141] A nonlinear mapping function is constructed using the energy distribution of the cumulative variance feature vector to map the scene transition probability to the sensitivity space. The characteristics of the mapping function are adjusted according to the scene stability characteristics to generate a time-segmented sensitivity evaluation criterion.

[0142] In some embodiments, the power load regulation module 260 is specifically used for:

[0143] According to the execution sequence of the control scheme, the target adjustment value of each phase is read, and the load of each phase is adjusted to the target adjustment value to obtain the adjusted user three-phase power load;

[0144] Calculate the difference between the adjusted current of each phase and the average current of the three phases, and divide the maximum value of the difference by the average current to obtain the three-phase current imbalance deviation.

[0145] The product of the rate of change of the three-phase current imbalance deviation and the historical fluctuation period is set as the monitoring window length. The three-phase current imbalance deviation within the monitoring window is weighted to obtain a dynamic threshold. When the three-phase current imbalance deviation exceeds the dynamic threshold, a new round of data acquisition is triggered.

[0146] The system provided in this embodiment of the invention can execute the adaptive management method for three-phase imbalance in transformer areas based on scene recognition provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0147] Figure 3 This is a schematic diagram of an electronic device for implementing the scene recognition-based adaptive management method for three-phase imbalance in transformer substations, as described in this embodiment of the invention. The electronic device is intended to represent various forms of digital computers, such as laptops, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0148] like Figure 3 As shown, the electronic device 310 includes at least one processor 311 and a memory, such as a read-only memory (ROM) 312 or a random access memory (RAM) 313, communicatively connected to the at least one processor 311. The memory stores computer programs executable by the at least one processor. The processor 311 can perform various appropriate actions and processes based on the computer program stored in the ROM 312 or loaded from storage unit 318 into the RAM 313. The RAM 313 can also store various programs and data required for the operation of the electronic device 310. The processor 311, ROM 312, and RAM 313 are interconnected via a bus 314. An input / output (I / O) interface 315 is also connected to the bus 314.

[0149] Multiple components in electronic device 310 are connected to I / O interface 315, including: input unit 316, such as keyboard, mouse, etc.; output unit 317, such as various types of displays, speakers, etc.; storage unit 318, such as disk, optical disk, etc.; and communication unit 319, such as network card, modem, wireless transceiver, etc. Communication unit 319 allows electronic device 310 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0150] Processor 311 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 311 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 311 performs the various methods and processes described above, such as the scene recognition-based three-phase imbalance adaptive management method for transformer substations.

[0151] In some embodiments, the scene-recognition-based adaptive management method for three-phase imbalance in transformer substations can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 318. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 310 via ROM 312 and / or communication unit 319. When the computer program is loaded into RAM 313 and executed by processor 311, one or more steps of the scene-recognition-based adaptive management method for three-phase imbalance in transformer substations described above can be performed. Alternatively, in other embodiments, processor 311 can be configured to perform the scene-recognition-based adaptive management method for three-phase imbalance in transformer substations by any other suitable means (e.g., by means of firmware).

[0152] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0153] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0154] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0155] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0156] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0157] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0158] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0159] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A scene-recognition-based adaptive management method for three-phase imbalance in transformer substations, characterized in that, include: Collect three-phase current data from users within the transformer area; Dynamically segment the user's electricity consumption according to the user's time period, and extract the user's electricity consumption pattern characteristics within the time segment; A similarity matrix is ​​calculated based on the electricity consumption pattern characteristics. A user scenario mapping map is constructed based on the similarity matrix. The user's scenario identifier is obtained through map clustering. The user's three-phase imbalance assessment weight is determined based on the scenario identifier. A hierarchical assessment criterion is established in combination with the three-phase current data to generate the user's time-sharing contribution to the three-phase imbalance of the transformer area. Based on the time-sharing contribution, a personalized control strategy for the user is formulated. Differential noise is added to the personalized control strategy and the electricity consumption pattern characteristics for encryption. The encrypted data is then fused from multiple sources using federated learning to generate a multi-user balanced control scheme. Adjust the user's three-phase power load according to the control scheme, calculate the unbalanced deviation of the adjusted three-phase current, and trigger a new round of data acquisition process when the unbalanced deviation exceeds the dynamic threshold.

2. The method according to claim 1, characterized in that, The dynamic segmentation based on user electricity consumption time periods, and the extraction of user electricity consumption pattern characteristics within each time segment, includes: The rate of change at continuous time points is calculated based on the three-phase current data. A power consumption fluctuation direction sequence is constructed based on the rate of change. The turning points of the power consumption fluctuation direction sequence are detected, and the power consumption data sequence between adjacent turning points is divided into power consumption observation data. Based on the electricity consumption observation data, a sliding analysis is performed to calculate the electricity consumption fluctuation amplitude within the sliding window. The location of the fluctuation change is identified according to the growth rate of the fluctuation amplitude, and the fluctuation change location is determined as the segmentation point of the electricity consumption period. Divide the electricity consumption period into the segmentation points, calculate the dispersion of the electricity consumption pattern in each electricity consumption period, and when the dispersion exceeds a preset threshold, determine the optimal segmentation time based on the fluctuation pattern of the electricity consumption data, and divide the electricity consumption period into sub-electricity consumption periods from the optimal segmentation time. The electricity consumption data of the sub-electricity consumption period is segmented and feature extracted. The extracted electricity consumption features are compared with historical electricity consumption features. Based on the similarity, the electricity consumption features are normalized to obtain the electricity consumption pattern features.

3. The method according to claim 1, characterized in that, The step of calculating a similarity matrix based on the electricity consumption pattern characteristics, constructing a user scenario mapping map based on the similarity matrix, and obtaining the user's scenario identifier through map clustering includes: The electricity consumption pattern features are divided into multiple hierarchical features according to time periods. The temporal correlation strength between different hierarchical features is calculated. The user similarity matrix is ​​obtained by adaptively weighting the similarity between users based on the temporal correlation strength. The user similarity matrix is ​​obtained by fusing the hierarchical similarity matrix based on the temporal correlation strength. Extract the temporal change features from the user similarity matrix, and map the temporal change features to the scene feature space according to the change trend to generate a user scene mapping map. In the user scene mapping map, the feature propagation intensity between users is calculated, and the node with the highest feature propagation intensity is identified as the scene cluster center. A scene division threshold is set based on the calculated propagation path strength. User nodes with a strength greater than the scene division threshold are marked as stable scenes, and user nodes with a strength less than the scene division threshold are marked as transitional scenes. The scene identifier is determined based on the evolution trajectory of the user node in the transition scene in the user scene mapping graph.

4. The method according to claim 1, characterized in that, The process involves determining the user's three-phase imbalance assessment weight based on the scenario identifier, establishing a hierarchical assessment criterion by combining the three-phase current data, generating the user's time-sharing contribution to the three-phase imbalance in the transformer area, and formulating a personalized control strategy for the user based on the time-sharing contribution. Based on the scene identifier, extract the electricity consumption period sequence and calculate the amplitude variation characteristics of the three-phase current in the electricity consumption period sequence; The amplitude variation characteristics are divided into ground state components and disturbance components according to their duration, and the user's three-phase imbalance assessment weight is determined based on the superposition effect of the ground state components and the disturbance components. The three-phase current data are weighted using the user three-phase imbalance assessment weight to obtain a hierarchical assessment benchmark. The assessment hierarchical criterion is generated based on the change gradient of the hierarchical assessment benchmark to determine the user three-phase imbalance contribution value. The user's three-phase imbalance contribution value is correlated with the three-phase imbalance degree of the transformer area to generate the user's time-sharing contribution to the three-phase imbalance of the transformer area. The control parameters are determined based on the time-series characteristics of the time-division contribution, and the control parameters are combined with the superposition effect of the ground state component and the disturbance component to generate the personalized control strategy.

5. The method according to claim 1, characterized in that, The step of adding differential noise to encrypt the personalized control strategy and the electricity consumption pattern characteristics, and then fusing the encrypted data through federated learning to generate a multi-user balanced control scheme, includes: Based on the scenario identifier, extract the temporal change trend of the personalized control strategy and the electricity consumption pattern characteristics, calculate the scenario stability characteristics, and establish a time-segmented sensitivity evaluation criterion; The sensitivity assessment criteria are combined with the scene stability characteristics to generate differential noise, and the differential noise is superimposed on the personalized control strategy and the power consumption pattern characteristics for encryption. Extract the fluctuation direction sequence of the encrypted data, associate and map the fluctuation direction sequence with the scene stability features, calculate gradient information, receive the gradient information uploaded by multiple users, calculate the temporal correlation strength of the gradient information, perform weighted aggregation on the gradient information according to the temporal correlation strength, and update the fusion parameters based on the aggregation result; Based on the stability characteristics of the scenario, the user ground state component and the user disturbance component are separated, the time-division contribution degree is determined, the time-division contribution degree is used to weight the fusion parameters, and the encrypted data is fused from multiple sources to generate the control scheme.

6. The method according to claim 5, characterized in that, The step of extracting the time-series change trends of the personalized control strategy and the electricity consumption pattern characteristics based on the scenario identifier, calculating the scenario stability characteristics, and establishing a time-segmented sensitivity evaluation criterion includes: The time-series data of the personalized control strategy and power consumption mode are extracted based on the scene identifier. The change of the data at adjacent time points is calculated to obtain the time-series difference sequence. Adaptive asymmetric variational decomposition is used to obtain the fluctuation component and the trend component. The energy entropy of the fluctuation component is calculated by wavelet packet transform to obtain the contribution weight. The time-series change trend is reconstructed based on the contribution weight. The time window length is adaptively determined based on the contribution weight, the cumulative variance feature vector of the fluctuation component within the window is calculated, the phase correlation of the cumulative variance feature vector is used as the scene stability feature, and the scene transition probability is obtained based on the changing trend of the scene stability feature. A nonlinear mapping function is constructed using the energy distribution of the cumulative variance feature vector to map the scene transition probability to the sensitivity space. The characteristics of the mapping function are adjusted according to the scene stability characteristics to generate a time-segmented sensitivity evaluation criterion.

7. The method according to claim 5, characterized in that, The process involves adjusting the user's three-phase power load according to the control scheme, calculating the adjusted three-phase current imbalance deviation, and triggering a new round of data acquisition when the imbalance deviation exceeds a dynamic threshold. This includes: According to the execution sequence of the control scheme, the target adjustment value of each phase is read, and the load of each phase is adjusted to the target adjustment value to obtain the adjusted user three-phase power load; Calculate the difference between the adjusted current of each phase and the average current of the three phases, and divide the maximum value of the difference by the average current to obtain the three-phase current imbalance deviation. The product of the rate of change of the three-phase current imbalance deviation and the historical fluctuation period is set as the monitoring window length. The three-phase current imbalance deviation within the monitoring window is weighted to obtain a dynamic threshold. When the three-phase current imbalance deviation exceeds the dynamic threshold, a new round of data acquisition is triggered.

8. A scene-recognition-based adaptive management system for three-phase imbalance in transformer substations, characterized in that, include: The three-phase current acquisition module is used to collect three-phase current data from users within the transformer area. The electricity consumption feature extraction module is used to dynamically segment users' electricity consumption according to their time periods and extract users' electricity consumption pattern features within the time segments. The scene identifier determination module is used to calculate a similarity matrix based on the electricity consumption pattern characteristics, construct a user scene mapping map based on the similarity matrix, and obtain the user's scene identifier through map clustering; The control strategy formulation module is used to determine the user's three-phase imbalance assessment weight based on the scenario identifier, establish a hierarchical assessment criterion in combination with the three-phase current data, generate the user's time-sharing contribution to the three-phase imbalance of the transformer area, and formulate the user's personalized control strategy based on the time-sharing contribution. The control scheme generation module is used to add differential noise to the personalized control strategy and the electricity consumption pattern characteristics for encryption, and to fuse the encrypted data through multi-party data using federated learning to generate a multi-user balanced control scheme. The power load control module is used to adjust the user's three-phase power load according to the control scheme, calculate the unbalanced deviation of the adjusted three-phase current, and trigger a new round of data acquisition process when the unbalanced deviation exceeds the dynamic threshold.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the adaptive management method for three-phase imbalance of transformer areas based on scene recognition as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that are used to cause a processor to execute the adaptive management method for three-phase imbalance of transformer areas based on scene recognition as described in any one of claims 1-7.

Images