Low-voltage power distribution intelligent regulation method and system based on big data analysis

Abstract

The application discloses a low-voltage power distribution intelligent regulation method and system based on big data analysis, and relates to the field of smart grids, wherein the method comprises the following steps: constructing a multi-dimensional low-voltage power distribution sample set based on big data, identifying a periodical event of regulation and control and corresponding associated factors; constructing a multi-period detection model according to the periodical event and the corresponding associated factors, deploying the multi-period detection model at a low-voltage power distribution regulation and control end, and monitoring a regulation and control target event in real time; when the regulation and control target event is identified, matching regulation and control parameters by using data performance of the associated factors, and performing linkage verification based on the periodical event; generating a regulation and control instruction for the regulation and control parameters that pass the linkage verification, and regulating and adjusting low-voltage power distribution nodes corresponding to the periodical event. The application solves the technical problem that the existing low-voltage power distribution regulation cannot accurately identify and flexibly regulate, and achieves the technical effects of improving the accuracy, flexibility and efficiency of low-voltage power distribution regulation.

Description

Technical Field

[0001] This application relates to the field of smart grids, and in particular to a method and system for intelligent regulation of low-voltage power distribution based on big data analysis. Background Technology

[0002] The stable and efficient operation of low-voltage power distribution systems is crucial for ensuring the orderly integration of new energy-grid-load-storage facilities and meeting users' reliable power needs. Currently, the industry mainly relies on manual inspections, manual control, or simple setpoint-triggered control methods to cope with fluctuations in power distribution system operation. These methods lack in-depth data mining and pattern identification of system operation, making them unsuitable for the complex scenarios brought about by the randomness of new energy sources and load fluctuations. They suffer from drawbacks such as response lag, insufficient control precision, and inability to predict periodic problems.

[0003] At present, low-voltage power distribution regulation suffers from technical problems such as the inability to accurately identify and flexibly control the system. Summary of the Invention

[0004] This application provides a method and system for intelligent regulation of low-voltage power distribution based on big data analysis. By constructing a sample set through big data analysis, periodic events and related factors in the power distribution system are identified. Based on this, a multi-period detection model is trained and deployed to monitor target events in real time. Once an event is identified, regulation parameters are matched and linkage verification is performed. Finally, instructions are generated to intelligently regulate the corresponding power distribution nodes. These technical means solve the technical problems of inaccurate identification and flexible control in existing low-voltage power distribution regulation, and achieve the technical effect of improving the accuracy, flexibility and efficiency of low-voltage power distribution regulation.

[0005] This application provides a low-voltage power distribution intelligent regulation method based on big data analysis, comprising: constructing a multi-dimensional low-voltage power distribution sample set based on big data, identifying periodic events and corresponding correlation factors for regulation; constructing a multi-period detection model based on the periodic events and corresponding correlation factors, deploying the multi-period detection model at the low-voltage power distribution regulation terminal, and monitoring the regulation target events in real time; when a regulation target event is identified, matching regulation parameters using the data performance of correlation factors, and performing linkage verification based on the periodic events; generating regulation commands for the regulation parameters that pass the linkage verification, and adjusting the low-voltage power distribution nodes corresponding to the periodic events.

[0006] In one possible implementation, a multi-dimensional low-voltage power distribution sample set is constructed based on big data, and the following processing is performed: parsing abnormal events in low-voltage power distribution, including at least voltage stability, load fluctuation, and three-phase balance; performing multi-dimensional clustering on the abnormal events to determine dimension description labels; and using the dimension description labels as indexes to collect low-voltage power distribution sample sets for each dimension.

[0007] In a possible implementation, the identification and control of periodic events and corresponding correlation factors involves the following processing: performing flow analysis based on the occurrence sequence of the abnormal events, segmenting the multi-dimensional low-voltage power distribution sample set using the periods before, during, and after the events as periodic nodes; and using the sample sets of each periodic node to perform correlation analysis on electrical signal data, user usage data, and load change data, identifying correlation factors, and establishing a mapping relationship between each periodic event and the correlation factors.

[0008] In a possible implementation, a path analysis is performed based on the occurrence sequence of the abnormal event. Using the events before, during, and after the event as periodic nodes, the following processing is performed: The power distribution network topology is analyzed based on a graph convolutional network to establish an electrical correlation model between nodes; the pre-event, concurrent, and subsequent features of the abnormal event in time sequence are extracted to construct an event propagation path; the electrical correlation model and the event propagation path are fused to identify a chain of events occurring periodically on a specific electrical path, and the chain of events is segmented and labeled according to the relationship between the nodes occurring in time sequence to determine the periodic nodes corresponding to the abnormal events.

[0009] In a possible implementation, a multi-cycle detection model is constructed based on the periodic events and their corresponding correlation factors. This model is then deployed at the low-voltage power distribution control terminal to monitor and control target events in real time. The following processing is performed: Each periodic event is labeled based on historical operating data, and electrical factors related to the event, such as voltage offset, load change rate, phase-to-phase current difference, and reactive power fluctuation amplitude, are extracted to construct a feature matrix for periodic pattern recognition. This feature matrix is ​​then input into the multi-cycle detection model, which consists of a periodic convolutional network, a graph convolutional network, and an attention mechanism layer. This allows the model to simultaneously learn the multi-cycle repetitive patterns of events in the time domain and the spatial correlation of events in the power distribution topology. The multi-cycle detection model is trained, optimized, and validated on a cloud server using a large-scale historical dataset to obtain a multi-cycle detection model with periodic pattern discrimination capabilities suitable for low-voltage power distribution operation scenarios. The trained model parameters are then distributed to edge electrical control nodes, and a lightweight inference engine is deployed in these edge nodes to perform multi-cycle event detection and control target event recognition based on local real-time sampled data.

[0010] In a possible implementation, the multi-period detection model is deployed at the low-voltage power distribution control terminal to monitor control target events in real time. The following processing is also performed: During model inference, edge nodes periodically upload locally identified suspected event sequences and electrical factor changes to the cloud. The cloud performs incremental learning and dynamic calibration of the model based on network-wide data, and reissues the updated model parameters for coordinated optimization between the cloud and edge nodes. When an edge node detects a suspected control target event but lacks sufficient event characteristics, it triggers cross-regional multi-node correlation analysis in the cloud. The cloud then determines whether the event has periodic characteristics that can trigger circuit control and feeds back the event level and control priority to the edge node.

[0011] In a possible implementation, linkage verification is performed based on periodic events, and the following processing is performed: construct an equivalent circuit model of the current region according to the electrical path corresponding to the chain event set where the periodic event is located; inject the control parameters into the equivalent circuit model, and perform multi-scenario simulations of steady-state power flow, transient voltage response, harmonic propagation and protection coordination verification; if the simulation results show that the control causes voltage over-limit, equipment overload, harmonic amplification or protection malfunction, it is determined that the linkage verification fails.

[0012] In a possible implementation, the following process is performed: when the linkage verification fails, the control parameters are iteratively adjusted based on the constraint optimization algorithm until the safe operation boundary is met.

[0013] In possible implementations, control commands are generated to adjust the low-voltage distribution nodes corresponding to periodic events, performing the following processes: determining the type of periodic event, selecting the corresponding circuit action path based on the type of periodic event, and performing low-voltage distribution node matching adjustment; wherein, when the periodic event is a daily load peak, the output voltage level is changed by controlling the cascaded H-bridge unit in the power electronic voltage regulation module, and at the same time driving the switch to switch to the standby line, capacitor or SVG module; when the periodic event is a rapid increase in distributed photovoltaic output, the voltage rise is alleviated by adjusting the reactive power support capability of the photovoltaic inverter, switching the voltage regulation direction of the H-bridge unit, or activating the energy storage system to absorb energy; when the periodic event is a three-phase imbalance, the load phase is transferred by driving the intelligent phase-switching switch, and at the same time the three-phase balanced H-bridge compensation branch is activated to achieve synchronous balance of three-phase voltage and current.

[0014] This application also provides a low-voltage power distribution intelligent regulation system based on big data analysis, including: a multi-dimensional low-voltage power distribution sample set construction module, used to construct a multi-dimensional low-voltage power distribution sample set based on big data, and identify the periodic events of regulation and the corresponding correlation factors; a multi-period detection model construction module, used to construct a multi-period detection model based on the periodic events and the corresponding correlation factors, and deploy the multi-period detection model at the low-voltage power distribution regulation terminal to monitor the regulation target events in real time; a regulation parameter matching module, used to match the regulation parameters using the data performance of the correlation factors when the regulation target event is identified, and to perform linkage verification based on the periodic event; and a low-voltage power distribution node regulation module, used to generate regulation commands for the regulation parameters that have passed the linkage verification, and to regulate the low-voltage power distribution nodes corresponding to the periodic events.

[0015] The proposed method and system for intelligent low-voltage power distribution regulation based on big data analysis first constructs a multi-dimensional low-voltage power distribution sample set based on big data to identify periodic events and corresponding correlation factors. Then, based on these periodic events and correlation factors, a multi-period detection model is constructed and deployed at the low-voltage power distribution control terminal to monitor target events in real time. When a target event is identified, control parameters are matched using the data performance of correlation factors, and linkage verification is performed based on the periodic event. Finally, control commands are generated for the control parameters that pass the linkage verification, adjusting the low-voltage power distribution nodes corresponding to the periodic event. Through this process, the proposed method and system achieve the technical effect of improving the accuracy, flexibility, and efficiency of low-voltage power distribution regulation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating the intelligent regulation method for low-voltage power distribution based on big data analysis provided in this application embodiment.

[0018] Figure 2 This is a schematic diagram of the structure of a low-voltage power distribution intelligent regulation system based on big data analysis, provided in an embodiment of this application.

[0019] Figure labeling: Multi-dimensional low-voltage power distribution sample set construction module 10, multi-cycle detection model construction module 20, control parameter matching module 30, low-voltage power distribution node adjustment module 40. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] This application provides a method for intelligent regulation of low-voltage power distribution based on big data analysis, such as... Figure 1 As shown, the method includes: Step S100: Construct a multi-dimensional low-voltage power distribution sample set based on big data to identify periodic events of regulation and their corresponding correlation factors.

[0022] Specifically, at the data acquisition layer, smart sensors, smart meters, and PMU (Phase Quantity Measurement Unit) synchronous phasor measurement devices are deployed to collect three-phase voltage, current, power, frequency, harmonics, equipment status, environmental data, and user electricity consumption information every second or millisecond. The collected data undergoes preprocessing, including removing missing values ​​through linear interpolation or mean filling, detecting and processing outliers using Z-score or IQR methods, standardizing or normalizing data of different dimensions using Min-Max normalization or Z-score normalization, and integrating data from different data sources and formats into a unified data platform. Multi-dimensional feature construction is performed, including electrical features, temporal features, topological features, and environmental and user features. The processed data is sliced ​​according to specific time windows or event triggers to form individual samples, which are then labeled accordingly, such as voltage sag events, three-phase imbalance events, and normal operation, thereby constructing a multi-dimensional low-voltage distribution sample set. Based on this, recurring periodic events are identified through time series analysis and statistical methods. Then, association rule mining, Bayesian networks, or machine learning models are used to mine various factors associated with these periodic events from the sample set, such as specific user behavior, weather conditions, and equipment status changes, to establish a mapping relationship between events and factors.

[0023] In one possible implementation, a multi-dimensional low-voltage power distribution sample set is constructed based on big data. Step S100 further includes step S110, analyzing abnormal events in low-voltage power distribution, including at least voltage stability, load fluctuation, and three-phase balance. Specifically, various abnormal situations in the operation of the power distribution system are automatically or semi-automatically identified and classified from the constructed sample set. Specifically, based on national or industry standards, equipment tolerance, and system operating experience, clear normal operating ranges and abnormal thresholds are set for key indicators such as voltage stability, load fluctuation, and three-phase balance. Event detection algorithms are applied to process real-time or historical data streams. Selectable algorithms include thresholding, sliding windowing, trend analysis, and machine learning-based anomaly detection. Detected abnormal events are classified, such as voltage sags and three-phase imbalances, and their occurrence time, duration, involved nodes / equipment, severity, and electrical quantity changes / waveforms / data before and after the event are recorded.

[0024] Step S120 involves multi-dimensional clustering of the abnormal events to determine dimensional description labels. Specifically, from the event features recorded in step S110, the features most discriminative for event classification are selected. For example, for voltage sag events, features may include sag amplitude, duration, occurrence time, accompanying current changes, and occurrence node. The K-means clustering algorithm is used to divide the data into K predefined clusters. For example, all voltage sag events are clustered based on amplitude and duration, resulting in different types of clusters such as deep and short, shallow and long, and medium-length clusters. After clustering the event feature vectors using the clustering algorithm, each cluster is analyzed in depth to summarize the common characteristics of the events within that cluster, such as specific time periods, specific locations, and specific electrical quantity change patterns. Based on the common characteristics of each cluster, dimensional description labels are assigned. These labels summarize the key attributes of the events in that cluster, such as time dimension labels, event characteristic dimension labels, spatial / location dimension labels, and preliminary correlation factor dimension labels, such as photovoltaic power output fluctuation correlation events and large motor start-up correlation events.

[0025] Step S130: Using the dimensional description tags as indexes, collect low-voltage power distribution sample sets for each dimension. Specifically, in the data storage platform, each original data sample or event record is labeled with one or more dimensional description tags, and an index relationship between the tags and the data is established. Based on specific analytical objectives, data is queried and extracted by combining tags from different dimensions. After the retrieved data is extracted, necessary format conversions and organization are required, such as standardizing data formats, time alignment, and handling missing values, to form a new sample set targeting a specific research objective. This sample set contains complete electrical quantity data, environmental data, and information about preceding and following events when such an event occurs. As new data is continuously collected, new events occur, and the tagging system is optimized, the sample sets for each dimension are updated periodically or in real-time to ensure their timeliness and representativeness.

[0026] In one possible implementation, step S100, which identifies the periodic events and corresponding correlation factors for regulation, further includes step S140, which involves performing a flow analysis based on the occurrence sequence of the abnormal events. The multi-dimensional low-voltage power distribution sample set is segmented using the periods before, during, and after the event as periodic nodes. Specifically, event correlation analysis is performed through time window correlation and causal relationship inference. Time window correlation refers to analyzing whether other abnormal events or significant changes in electrical quantities occur before and after an abnormal event. Causal relationship inference refers to inferring possible causal or triggering relationships between events by analyzing the sequence of events and the logical relationships of changes in electrical quantities. Four key cycle nodes are defined: The period before an event occurs is defined as the time from the detection of a precursor signal indicating the event or from a fixed point in time until the event officially occurs; this stage is used to analyze the causes and warning signals. The period during an event occurs is defined as the time period from when the main characteristics of the abnormal event begin to appear and persist; this stage is used to analyze the peak characteristics and severity of the event. The period during event regulation is defined as the time period from when the system issues a regulation command to when the implementing agency responds and begins to take effect, until the system begins to recover to normal; this stage is used to analyze the response speed and initial effect of the regulation measures. The period after event regulation is defined as the time period from when the system basically returns to normal until after an observation period confirms that the event's impact has been eliminated and the system is operating stably; this stage is used to evaluate the final effect of the regulation and the system's recovery capability. Based on these four cycle nodes, each event record in the multi-dimensional sample set collected in step S130 is divided into four corresponding segments according to its timestamp; each segment represents a stage in the event's lifecycle.

[0027] In one possible implementation, a flow analysis is performed based on the occurrence sequence of the abnormal event, using the periods before, during, and after the event as periodic nodes. Step S140 further includes step S141, analyzing the power distribution network topology based on a graph convolutional network to establish an electrical correlation model between nodes. Specifically, a network topology graph is constructed, abstracting key equipment or connection points in the power distribution network as nodes, each node containing its attribute characteristics; the physical connections between nodes are abstracted as edges, and the edge weights can be initialized with values ​​related to line impedance, length, or admittance, representing the tightness of the physical connection between nodes. Within a specific time segment or time period, the electrical quantity data of each node are assigned as feature vectors to the corresponding nodes in the graph. A graph convolutional network (GCN) model is designed and trained. The model adopts a multi-layer GCN structure, where the convolution operation of each layer of GCN aggregates the feature information of a node and its neighboring nodes, thereby updating the feature representation of that node. The training objective is to enable the GCN model to learn the electrical correlation between nodes. This can be achieved using unsupervised representation learning methods, such as algorithms like DeepWalk and Node2Vec, or GCN-based autoencoders. Without explicit supervision, the model learns low-dimensional vector representations of nodes, ensuring that nodes close to each other in the vector space share similar electrical characteristics or functions. The trained GCN model outputs a weight value for each edge in the graph, representing the electrical correlation between the corresponding two nodes. A high correlation indicates that a change in the electrical state of one node more easily affects the other.

[0028] Step S142: Extract the pre-event, concurrent, and subsequent features of the abnormal event in time sequence to construct the event propagation trajectory. Specifically, for each target abnormal event E, extract all event records within a sufficiently long time window containing E from the time-series database, including normal and abnormal events, and arrange them in chronological order to form an event sequence. Define and extract pre-event, concurrent, and subsequent features. Pre-event features refer to features that appear before event E and may be related to E, including pre-events, trend features, and threshold crossing features; concurrent features refer to features that occur simultaneously or almost simultaneously with event E during the occurrence of event E, including concurrent events, peak features, and abrupt change features; subsequent features refer to features that appear after event E ends and may have a causal relationship with E, including subsequent events, recovery features, and regulatory response features. The extracted antecedent features, concurrent features, and subsequent features are organized according to time sequence and logical relationships to form a complete trajectory describing event E from the emergence of the trigger, to its development, and then to its end and impact. This trajectory not only includes the event itself, but also its related key features and possible causal chains.

[0029] Step S143: The electrical correlation model and event propagation path are integrated to identify a set of chain events that occur periodically on a specific electrical path. The chain event set is then segmented and labeled according to the temporal relationships between nodes to determine the periodic nodes corresponding to abnormal events. Specifically, preliminary event chain identification is performed. Using the electrical correlation model established in step S141, for an identified abnormal event E, a set of nodes with high correlation is identified. These nodes are potential objects affected or influenced by E. Based on the propagation path of event E, the time window is expanded to check whether other abnormal events occur at these highly correlated nodes before and after E. If event E1 occurs at node n1, followed by event E2 at a highly correlated node n2, and then event E3 at a highly correlated node n3, and E1, E2, and E3 have a close temporal sequence, then they can be preliminarily determined to constitute an event chain. Next, periodic detection is performed. For the preliminarily identified event chains, their pattern features are extracted, such as event type sequence, involved node path sequence, and event interval time. Then, over a longer timeframe, a search is conducted to determine if other event chains similar to the current event chain pattern exist. This can be achieved using a sliding window method combined with sequence matching algorithms such as Dynamic Time Warping (DTW) to measure the similarity between event chains. If multiple highly similar event chains are found within a fixed time interval, the event chain pattern is considered periodic. All identified event chains belonging to the same periodic pattern are grouped into a set, called the periodic chain event set. Finally, periodic node segmentation and labeling are performed. For the entire periodic chain event set, four unified periodic nodes are defined: before the chain occurs, during the chain occurs, during the chain adjustment, and after the chain adjustment. For each specific event chain and each independent event within the chain in the chain event set, based on its position in the entire event chain's lifecycle, it is mapped to the corresponding global periodic node, thereby determining the periodic node corresponding to each anomalous event.

[0030] Step S150 involves using the sample sets of each cycle node to perform correlation analysis on electrical signal data, user usage data, and load change data, identifying correlation factors, and establishing a mapping relationship between each cycle event and the correlation factors. Specifically, based on the cycle nodes determined in steps S140 and S143, sample subsets corresponding to different cycle nodes are extracted from the multi-dimensional sample set constructed in step S130. In each sub-node sample set, data from different data sources, including electrical signal data, user usage data, and load change data, are aligned and fused. Deep analysis is performed using association analysis algorithms, which include statistical correlation analysis, causal inference, and machine learning methods. In statistical correlation analysis, the Pearson correlation coefficient measures the linear correlation between two continuous variables, the Spearman rank correlation coefficient measures the monotonic correlation, and the chi-square test examines whether a significant association exists between two categorical variables. In causal inference, the Granger causality test determines whether historical information of one variable helps predict the future value of another based on time series data; mutual information measures the dependency between two variables, and transfer entropy measures the information transfer from one time series to another. Machine learning methods train a predictive model with the occurrence or severity of cyclical events as the target variable, and various electrical signal characteristics, user behavior characteristics, and load characteristics as input features. Feature importance scores are used to identify the features with the greatest impact on the target variable. Through one or more of these association analysis algorithms, factors that are statistically significantly correlated with or potentially causally related to specific cyclical events or event chains are selected; these factors are called association factors. A mapping table or model is then established to show which association factors are associated with which cyclical events, as well as the strength and direction of the association.

[0031] Step S200: Based on the periodic events and corresponding correlation factors, construct a multi-period detection model, and deploy the multi-period detection model at the low-voltage power distribution control terminal to monitor the control target events in real time.

[0032] Specifically, model requirements analysis and architecture design are conducted, clarifying that the core task of the model is to receive streaming data from the power distribution system in real time and determine whether a defined periodic event is currently occurring or about to occur. Performance requirements include high accuracy, low latency, and a moderate model size. The model architecture selects network structures that can effectively capture time dependencies, such as Temporal Convolutional Networks (TCNs), combined with networks capable of handling graph-structured data, such as Graph Attention Networks (GATs). An attention mechanism is introduced to highlight the importance of key features and different periodic patterns. The final multi-period detection model can be a fusion model, for example, by concatenating or weighting the temporal features extracted by TCNs and the spatial features extracted by GATs, and then feeding them into a fully connected layer for classification or regression prediction. Training data preparation and model training are performed. The sample set with periodic event labels and correlation factor features constructed in steps S130 and S140 is divided into a training set, a validation set, and a test set. The features of the input model are reconfirmed and optimized. Then, the designed model is trained on a high-performance cloud server or GPU cluster using the training set. The model performance is monitored using the validation set, and hyperparameters are tuned using methods such as grid search, random search, or Bayesian optimization to achieve optimal performance. Next, the model is deployed and integrated with the inference engine. If the deployment environment has limited resources, the trained model can be lightweighted and converted into a format suitable for running on edge devices. A lightweight AI inference engine is deployed at the low-voltage power distribution control terminal. This engine is responsible for loading the model and performing fast inference computation. Once deployed, the model receives data streams from sensors and smart devices in real time through the interface of the edge nodes. The inference engine preprocesses the real-time data and then inputs it into the model for inference. The model outputs the probability or category of a certain periodic event occurring at the current moment. If the probability exceeds the set threshold, it is determined that the target event for regulation has been detected, and subsequent regulation processes are triggered.

[0033] In one possible implementation, a multi-cycle detection model is constructed based on the periodic events and their corresponding correlation factors. This model is then deployed at the low-voltage power distribution control terminal to monitor and control target events in real time. Step S200 further includes step S210, which involves labeling each periodic event based on historical operating data and extracting voltage offset, load change rate, phase-to-phase current difference, and reactive power fluctuation amplitude electrical factors related to the event to construct a feature matrix for periodic pattern recognition. Specifically, event labeling is performed by defining an event category list based on the periodic event types identified in step S100. For example, category 0 represents normal operation, category 1 represents peak daily load voltage sag, category 2 represents… Then, the historical operating data sample set used for model training is traversed. For each sample, if it is determined to belong to a specific periodic event, the corresponding event category label is assigned to the sample, forming the target variable for model training. Electrical factor extraction is performed. For each tagged sample, whether it is an event sample or a normal sample, key electrical parameters that characterize the sample's properties are calculated and extracted from the raw electrical quantity data it contains. These parameters include voltage offset, load change rate, phase-to-phase current difference, and reactive power fluctuation amplitude. Specifically, voltage offset = (measured voltage - rated voltage) / rated voltage × 100%, from which maximum offset, minimum offset, average offset, and standard deviation of offset can be extracted; load change rate = (current load - previous load) / previous load × 100% / time interval, from which active power change rate, reactive power change rate, maximum change rate, and average change rate can be extracted; phase-to-phase current difference can be obtained by taking the maximum value, average value, or calculating the three-phase current imbalance; reactive power fluctuation amplitude can be calculated by taking the difference between the maximum and minimum reactive power values ​​or the standard deviation of reactive power within the sample time window.

[0034] Finally, a feature matrix is ​​constructed. For each sample, all extracted electrical factors are arranged in a predetermined order to form a one-dimensional numerical vector, which constitutes the feature vector of that sample. The feature vectors of all samples in the dataset are stacked row by row to form a two-dimensional feature matrix. Each row of the feature matrix represents a sample, and each column represents a specific electrical factor feature.

[0035] Step S220 involves inputting the feature matrix into a multi-period detection model composed of a periodic convolutional network, a graph convolutional network, and an attention mechanism layer. This enables the model to simultaneously learn the multi-period repetitive patterns of events in the time domain and the spatial correlations of events in the power distribution topology. Specifically, the periodic convolutional network (TCN) learns patterns in time series data, particularly long-range dependencies and multi-scale periodicities. The TCN employs causal convolution and dilated convolution, combined with residual connections. The input to the TCN is the time series feature segment corresponding to each sample in the feature matrix, and the output is a high-dimensional temporal feature representation at each time step or the end of the sequence after TCN processing, capturing the multi-period patterns in the time series. The graph convolutional network (GCN) learns spatial correlation patterns in graph-structured data. In a power distribution system, the GCN can learn the electrical relationships between nodes based on the network topology. Its core idea is message passing, where each node's new features are updated by aggregating the feature information of its neighboring nodes. This process allows each node to perceive the state of its neighbors, thereby learning the global spatial structure features. The input to the GCN is the graph structure of the entire power distribution network and the feature vectors of all nodes in the graph at a specific time step. The output is a high-dimensional spatial feature representation of each node after GCN processing. This representation integrates its own information and the information of its neighboring nodes, reflecting the node's position and role in the network topology. The attention mechanism acts as an intelligent filter, automatically learning the importance weights of the input features, allowing the model to focus on key information and suppress irrelevant or secondary information when making decisions. The input to the attention mechanism layer can be the temporal feature sequence output by the TCN, the set of node spatial features output by the GCN, or a joint feature obtained by concatenating the two. Its output is a feature representation weighted by attention weights, highlighting key temporal patterns or key nodes / regions. Finally, the model fuses the temporal features learned by the TCN and the spatial features learned by the GCN through feature concatenation or summation, and feeds it into the classification head to output a probability distribution, representing the probability that the current input belongs to each periodic event category.

[0036] Step S230: The multi-cycle detection model is trained, optimized, and validated on a cloud server using a large-scale historical dataset to obtain a multi-cycle detection model with periodic pattern discrimination capability adaptable to low-voltage power distribution operation scenarios. Specifically, in the cloud, the dataset is divided into training, validation, and test sets using large-scale historical data. The model is trained using the training set, and the loss function is minimized through backpropagation and an optimizer. During training, the model performance is monitored using the validation set, and overfitting is prevented and the model's generalization ability is optimized by adjusting hyperparameters and applying regularization techniques. Finally, the trained model is evaluated using an independent test set to ensure that it can accurately identify periodic events even on unseen data.

[0037] Step S240 involves distributing the trained model parameters to the edge electrical control nodes and deploying a lightweight inference engine on each edge node to perform multi-cycle event detection and target event identification based on local real-time sampling data. Specifically, the model parameters trained in the cloud are distributed to each edge electrical control node via a secure communication protocol. At the edge nodes, a lightweight AI inference engine is deployed and the model is loaded. The edge nodes collect local electrical quantity data in real time, perform preprocessing consistent with that used during training, and then input the data into the model for rapid inference. The inference results include event categories and confidence levels, used to determine whether a target event requiring regulation has occurred.

[0038] In one possible implementation, the multi-period detection model is deployed at the low-voltage power distribution control terminal to monitor control target events in real time. Step S200 further includes step S250, where, during model inference, edge nodes periodically upload locally identified suspected event sequences and electrical factor changes to the cloud. The cloud performs incremental learning and dynamic calibration of the model based on network-wide data, and redistributes the updated model parameters for coordinated optimization between the cloud and edge nodes. Specifically, edge nodes periodically upload locally detected suspected event data, related electrical factor changes, and model inference logs to the cloud. The cloud aggregates data from multiple edge nodes and analyzes it in conjunction with historical data to determine whether new event patterns have emerged or whether the performance of the existing model has deteriorated. If optimization is needed, the model is incrementally trained or fine-tuned using this new data to adapt to new system operating characteristics and avoid catastrophic amnesia. The updated model parameters are redistributed to the edge nodes to achieve continuous model optimization and self-adaptation.

[0039] Step S260: When an edge node detects a suspected control target event but lacks sufficient event characteristics, a cross-regional multi-node correlation analysis is triggered in the cloud. The cloud determines whether the event has periodic characteristics that can trigger circuit control and feeds back the event level and control priority to the edge node. Specifically, when an edge node detects a suspected event but its characteristics are unclear or its confidence level is low, it reports the relevant information to the cloud for assistance. The cloud utilizes its global data perspective to retrieve the operating data and event records of other relevant edge nodes during the event's occurrence period, such as the same feeder and adjacent areas, to perform cross-regional, multi-node correlation analysis and more complex pattern matching. The cloud comprehensively determines whether the event is a truly periodic event requiring control, assesses its severity and control priority, and feeds back the results to the requesting edge node to guide its subsequent actions, such as triggering control, continuing observation, or ignoring it.

[0040] Step S300: When a target event for regulation is identified, the regulation parameters are matched using the data performance of the correlation factors, and linkage verification is performed based on the periodic event.

[0041] Specifically, once a clear target event for regulation is identified, the system, based on the event type and the specific data performance of the associated factors that triggered the event, matches a preliminary set of regulation parameters from a pre-set regulation strategy library or through a rule engine. Examples include adjusting transformer tap positions, switching capacitor banks, and controlling SVG reactive power output. To ensure safe and effective regulation, linkage verification is required. The preliminary regulation parameters are input into the equivalent circuit model of the power distribution system for multi-scenario simulations, including steady-state power flow, transient response, harmonic analysis, and protection coordination verification. This assesses whether the system meets safety operation constraints after regulation, such as voltage within acceptable ranges, equipment not overloaded, harmonics not exceeding limits, and protection not malfunctioning. If not, the regulation parameters are iteratively adjusted based on a constraint optimization algorithm until the simulation results meet the safety boundary conditions.

[0042] In one possible implementation, linkage verification is performed based on periodic events. Step S300 further includes step S310, which constructs an equivalent circuit model of the current area based on the electrical path corresponding to the chain event set in which the periodic event is located. Specifically, based on the specific electrical nodes and lines involved in the identified periodic chain events, relevant information, such as line impedance, transformer parameters, and load characteristics, is extracted from the network topology database and equipment parameter database. Using this information, an equivalent circuit model that accurately reflects the current operating state and electrical connection relationships of the power distribution system in the area is constructed, providing a foundation for simulation verification.

[0043] Step S320: The control parameters are injected into the equivalent circuit model to perform multi-scenario simulations of steady-state power flow, transient voltage response, harmonic propagation, and protection coordination verification. Specifically, the control parameters to be verified are applied to the equivalent circuit model to simulate the execution of control measures. Subsequently, multi-dimensional simulation analysis is performed on the model. Among these, steady-state power flow calculation verifies whether the voltage, branch current, and power distribution of each node are reasonable after control, and whether there are any exceeding limits. Transient voltage response simulation evaluates the dynamic changes in voltage during and shortly after control, such as whether overvoltage or oscillation occurs. Harmonic propagation analysis checks whether control will lead to harmonic amplification or generate new harmonic problems, ensuring that the harmonic distortion rate meets the standard. Protection coordination verification simulates whether the relay protection device can operate correctly and in coordination when a fault occurs in the system after control, avoiding failure to operate or false operation.

[0044] Step S330: If the simulation results show that the control causes voltage exceeding limits, equipment overload, harmonic amplification, or protection malfunction, the linkage verification is deemed unsuccessful. When the linkage verification fails, the control parameters are iteratively adjusted based on a constraint optimization algorithm until the safe operating boundary is met. Specifically, after the simulation is completed, the simulation results are compared with the preset safe operating constraints. If voltage exceeding limits, equipment overload, harmonic exceedance, or protection coordination problems are found, the control scheme is deemed unsuccessful. At this time, the constraint optimization algorithm is activated, using the control objective, such as voltage recovery or minimum network loss, as a function, and the safety boundary as a constraint, to automatically adjust and optimize the control parameters. After generating a new control parameter scheme, simulation verification is performed again. This iterative process continues until a set of control parameters that can satisfy all safety constraints and effectively achieve the control objective is found.

[0045] Step S400: Generate control commands for the control parameters that have passed the linkage verification, and adjust the low-voltage power distribution nodes corresponding to the periodic events.

[0046] Specifically, the final control parameter scheme that has passed the linkage verification is converted into specific, executable control instructions. These instructions clearly specify the equipment to be operated, the type of operation, and the specific parameter values. Through a reliable communication network, these instructions are sent to the corresponding field actuators or smart terminals, which drive the physical equipment to complete the adjustment operation, thereby eliminating or mitigating the adverse effects of detected periodic events.

[0047] In one possible implementation, a control command is generated to adjust the low-voltage distribution node corresponding to the periodic event. Step S400 further includes step S410, which determines the type of periodic event and selects the corresponding circuit action path according to the type of periodic event to perform low-voltage distribution node matching adjustment. Specifically, when the periodic event is a peak daily load, the output voltage level is changed by controlling the cascaded H-bridge unit in the power electronic voltage regulation module, and at the same time, the switch is driven to connect to the standby line, capacitor, or SVG module. When the periodic event is a rapid increase in distributed photovoltaic output, the voltage rise is alleviated by adjusting the reactive power support capability of the photovoltaic inverter, switching the voltage regulation direction of the H-bridge unit, or activating the energy storage system to absorb energy. When the periodic event is a three-phase imbalance, the load phase is transferred by driving the intelligent phase-switching switch, and at the same time, the three-phase balanced H-bridge compensation branch is activated to achieve synchronous balance of three-phase voltage and current.

[0048] Specifically, based on the specific type of the identified periodic event, such as intraday load peak, sudden increase in photovoltaic output, or three-phase imbalance, targeted regulation strategies and execution equipment are selected. When the periodic event is an intraday load peak, its main characteristic is a significant increase in the total system load, leading to a drop in line voltage, an increase in transformer and line load rates, and even potential voltage exceedances and equipment overloads. In this situation, the core of the regulation strategy is to improve the system's power supply capacity and voltage level. This includes controlling the power electronic voltage regulation module, driving switches to connect to backup lines, and connecting capacitors or SVG modules. The power electronic voltage regulation module uses a cascaded H-bridge topology converter. By controlling the output voltage and phase of each H-bridge unit, the output voltage level of the entire regulation module can be precisely changed. The control system calculates the amount of voltage compensation required based on the current voltage deviation and load conditions, and then generates a drive signal using pulse width modulation technology to control the on and off of each switching device in the cascaded H-bridge unit, thereby synthesizing the required output voltage and achieving rapid and continuous regulation of the load-side voltage, maintaining the voltage near the rated value.

[0049] If a backup power supply line or sectionalizing switch exists in the system, the control system will issue a command to close the corresponding circuit breaker or load switch, putting the backup line into operation. This increases the power supply path, distributes the load current, reduces the load rate of the original line, thereby reducing line losses and voltage drops, and improving the system's power supply reliability and voltage quality.

[0050] Connecting parallel capacitor banks can increase the reactive power supply of the system, improve the power factor, thereby reducing the reactive current component in the line and reducing active power losses and voltage drops. SVG (Static Var Generator) is a reactive power compensation device that can quickly and continuously adjust the output reactive power. It can provide not only capacitive reactive power but also absorb inductive reactive power, and is more effective in suppressing voltage fluctuations, flicker, and improving voltage stability. The control system determines the number of capacitor banks to connect or sets the reactive power output command value of the SVG based on the load's reactive power demand and voltage conditions.

[0051] When a periodic event occurs where the output of distributed photovoltaic (PV) power increases rapidly, its main characteristic is a sharp increase in the output power of the PV power plant within a short period, leading to a rapid rise in the grid connection voltage, sometimes exceeding the allowable limit and affecting the safety of other users' electrical equipment. To address this, the core regulation strategy is to absorb excess active power or provide inductive reactive power to suppress the voltage rise. Specifically, this includes adjusting the reactive power support capability of the PV inverter, switching the voltage regulation direction of the H-bridge unit, and activating the energy storage system for energy absorption.

[0052] Photovoltaic inverters have four-quadrant operation capability. In addition to outputting active power, they can also provide or absorb reactive power according to grid demand. The control system sends reactive power control commands to the photovoltaic inverter, enabling it to appropriately absorb inductive reactive power from the grid or reduce the output of capacitive reactive power while outputting active power, thereby lowering the voltage level at the grid connection point and alleviating voltage rise.

[0053] The system is equipped with a cascaded H-bridge voltage regulator module with bidirectional power flow, which can switch its operating mode to absorb active power or provide inductive reactive power. By changing the control strategy of the H-bridge unit, its output voltage is made to have a certain phase difference with the grid voltage, thereby absorbing active power from the grid or injecting inductive reactive power into the grid to achieve voltage stabilization.

[0054] Energy storage systems can quickly respond to and absorb excess active power during rapid increases in photovoltaic output. The control system issues commands to put the energy storage converter into charging mode, storing the excess electrical energy. This not only effectively suppresses voltage rise but also achieves peak shaving and valley filling, improving energy utilization efficiency.

[0055] When the periodic event is three-phase imbalance, its main characteristic is that the current magnitudes of each phase in the three-phase system are inconsistent, leading to increased neutral current, increased line losses, reduced transformer utilization, and potentially causing problems such as motor vibration and relay protection malfunctions. To address this, the core of the regulation strategy is to redistribute the load across each phase or inject compensating current to achieve balance in three-phase current and voltage. Specifically, this includes driving intelligent phase-switching switches to transfer load phases and activating the three-phase balanced H-bridge compensation branch.

[0056] A smart phase-switching switch is an intelligent device that can switch a single-phase load from one phase to another without power interruption. The control system calculates the load imbalance of each phase based on real-time monitored three-phase current data, then determines the load to be transferred and the target phase. Next, it sends a switching command to the corresponding smart phase-switching switch, switching part of the single-phase load from the heavily loaded phase to the lightly loaded phase, thereby balancing the load distribution across phases and reducing the three-phase imbalance.

[0057] The three-phase balanced H-bridge compensation branch is also based on the cascaded H-bridge topology, but it has the ability to independently adjust the current of each phase and can be regarded as a type of chain-type static synchronous compensator. The control system calculates the compensation current to be injected into each phase based on the detected three-phase current imbalance. Then, by controlling the switching action of each unit in the H-bridge compensation branch, a compensation current equal in magnitude and opposite in direction to the load imbalance current is generated and injected into the power grid, thereby offsetting the unbalanced component generated by the load and achieving synchronous balance of three-phase current and voltage.

[0058] When performing the above adjustment operations, the system sends specific control commands to the corresponding execution device controllers via the communication network, such as the controllers of power electronic voltage regulation modules, smart switches, photovoltaic inverter control units, and energy storage PCS controllers. After receiving the commands, these controllers drive the corresponding power electronic devices or mechanical switches to complete the predetermined adjustment task.

[0059] This application's embodiments construct a sample set through big data analysis, identify periodic events and related factors in the power distribution system, train and deploy a multi-period detection model based on this, monitor target events in real time, and once an event is identified, match control parameters and perform linkage verification, ultimately generating instructions to intelligently adjust the corresponding power distribution nodes. These technical means solve the technical problems of inaccurate identification and flexible control in existing low-voltage power distribution regulation, achieving the technical effect of improving the accuracy, flexibility and efficiency of low-voltage power distribution regulation.

[0060] In the above text, refer to Figure 1 This paper describes in detail a low-voltage power distribution intelligent regulation method based on big data analysis according to embodiments of the present invention. Next, reference will be made to... Figure 2 This invention describes a low-voltage power distribution intelligent regulation system based on big data analysis according to an embodiment of the present invention.

[0061] The low-voltage power distribution intelligent regulation system based on big data analysis according to embodiments of the present invention is used to solve the technical problems of inaccurate identification and flexible control in existing low-voltage power distribution regulation, thereby improving the accuracy, flexibility and efficiency of low-voltage power distribution regulation. The low-voltage power distribution intelligent regulation system based on big data analysis includes: a multi-dimensional low-voltage power distribution sample set construction module 10, a multi-cycle detection model construction module 20, a regulation parameter matching module 30, and a low-voltage power distribution node regulation module 40.

[0062] The multi-dimensional low-voltage distribution sample set construction module 10 is used to construct a multi-dimensional low-voltage distribution sample set based on big data, and identify the periodic events of regulation and the corresponding correlation factors; the multi-period detection model construction module 20 is used to construct a multi-period detection model based on the periodic events and the corresponding correlation factors, and deploy the multi-period detection model at the low-voltage distribution control terminal to monitor the control target events in real time; the control parameter matching module 30 is used to match the control parameters using the data performance of the correlation factors when the control target event is identified, and to perform linkage verification based on the periodic event; the low-voltage distribution node adjustment module 40 is used to generate control commands for the control parameters that have passed the linkage verification, and to adjust the low-voltage distribution nodes corresponding to the periodic events.

[0063] The detailed description of the specific configuration of the multi-dimensional low-voltage power distribution sample set construction module 10 is explained as follows: As mentioned above, the multi-dimensional low-voltage power distribution sample set is constructed based on big data. The multi-dimensional low-voltage power distribution sample set construction module 10 may further include: an abnormal event analysis unit for analyzing abnormal events of low-voltage power distribution, including at least voltage stability, load fluctuation, and three-phase balance; a multi-dimensional clustering unit for performing multi-dimensional clustering on the abnormal events to determine dimension description labels; and a low-voltage power distribution sample set acquisition unit for acquiring low-voltage power distribution sample sets of each dimension using the dimension description labels as indexes.

[0064] The multi-dimensional low-voltage power distribution sample set construction module 10, which identifies and regulates periodic events and corresponding correlation factors, may further include: a sample set segmentation unit for performing flow analysis based on the occurrence sequence of the abnormal events, using the period before the event, during the event, during the event regulation, and after the event regulation as periodic nodes to segment the multi-dimensional low-voltage power distribution sample set; and a correlation factor identification unit for using the sample sets of each periodic node to perform correlation analysis of electrical signal data, user usage data, and load change data, identify correlation factors, and establish a mapping relationship between each periodic event and the correlation factors.

[0065] The process involves analyzing the movement of the abnormal events based on their occurrence sequence, using the periods before, during, and after the events as periodic nodes. The sample set segmentation unit can further include: an electrical correlation model establishment subunit for analyzing the power distribution network topology based on graph convolutional networks and establishing an electrical correlation model between nodes; an event propagation movement line construction subunit for extracting the pre-event, concurrent, and subsequent features of the abnormal events in time sequence and constructing an event propagation movement line; and a periodic node segmentation and annotation subunit for fusing the electrical correlation model and the event propagation movement line to identify a chain of events occurring periodically on a specific electrical path, and segmenting and annotating the chain of events according to the node relationships occurring in time sequence to determine the periodic nodes corresponding to the abnormal events.

[0066] The detailed description of the specific configuration of the multi-cycle detection model construction module 20 is explained as follows: As mentioned above, a multi-cycle detection model is constructed based on the cycle events and corresponding correlation factors. The multi-cycle detection model is deployed at the low-voltage power distribution control terminal to monitor the control target events in real time. The multi-cycle detection model construction module 20 may further include: an event labeling processing unit for labeling each cycle event based on historical operating data, and extracting voltage offset, load change rate, phase-to-phase current difference, reactive power fluctuation amplitude electrical factors related to the event to construct a feature matrix for cycle pattern recognition; and a multi-cycle detection unit for inputting the feature matrix into the cycle... The multi-period detection model, composed of periodic convolutional networks, graph convolutional networks, and attention mechanism layers, enables the model to simultaneously learn the multi-period repetitive patterns of events in the time domain and the spatial correlation of events in the power distribution topology. The model training unit is used to train, optimize, and validate the multi-period detection model on a cloud server using a large-scale historical dataset to obtain a multi-period detection model with periodic pattern discrimination capability that can be adapted to low-voltage power distribution operation scenarios. The multi-period event detection unit is used to distribute the trained model parameters to the edge electrical control nodes and deploy a lightweight inference engine in the edge nodes to perform multi-period event detection and control target event recognition based on local real-time sampling data.

[0067] The multi-period detection model is deployed at the low-voltage power distribution control terminal to monitor control target events in real time. The multi-period detection model construction module 20 may further include: an incremental learning unit, which is used to periodically upload locally identified suspected event sequences and electrical factor changes to the cloud during the model inference process. The cloud performs incremental learning and dynamic calibration on the model based on the data of the entire network, and re-issues the updated model parameters for linkage optimization between the cloud and the edge nodes; and a cross-regional multi-node correlation analysis unit, which is used to trigger cross-regional multi-node correlation analysis in the cloud when the edge node detects a suspected control target event but the event characteristics are insufficient. The cloud performs a judgment on whether the event has periodic characteristics that can trigger circuit control, and feeds back the event level and control priority to the edge node.

[0068] The detailed description of the specific configuration of the control parameter matching module 30 is explained as follows: As mentioned above, based on the periodic event-based linkage verification, the control parameter matching module 30 may further include: an equivalent circuit model construction unit for constructing an equivalent circuit model of the current region according to the electrical path corresponding to the chain event set where the periodic event is located; a simulation unit for injecting the control parameters into the equivalent circuit model to perform multi-scenario simulations of steady-state power flow, transient voltage response, harmonic propagation, and protection coordination verification; and a judgment unit for determining that the linkage verification fails if the simulation results show that the control causes voltage over-limit, equipment overload, harmonic amplification, or protection malfunction.

[0069] The control parameter matching module 30 may further include: when the linkage verification fails, iteratively adjusting the control parameters based on the constraint optimization algorithm until the safe operation boundary is met.

[0070] The detailed description of the specific configuration of the low-voltage distribution node adjustment module 40 is explained as follows: As mentioned above, it generates control commands to adjust the low-voltage distribution nodes corresponding to periodic events. The low-voltage distribution node adjustment module 40 may further include: a low-voltage distribution node matching adjustment unit for determining the type of periodic event, selecting the corresponding circuit action path according to the type of periodic event, and performing low-voltage distribution node matching adjustment; wherein, when the periodic event is the intraday load peak, the output voltage level is changed by controlling the cascaded H-bridge unit in the power electronic voltage regulation module, and at the same time driving the switch to the standby line, capacitor or SVG module; when the periodic event is the rapid increase of distributed photovoltaic output, the voltage rise is alleviated by adjusting the reactive power support capability of the photovoltaic inverter, switching the voltage regulation direction of the H-bridge unit or activating the energy storage system to absorb energy; when the periodic event is three-phase imbalance, the load phase is transferred by driving the intelligent phase switching, and the three-phase balanced H-bridge compensation branch is activated to achieve synchronous balance of three-phase voltage and current.

[0071] The low-voltage power distribution intelligent regulation system based on big data analysis provided in the embodiments of the present invention can execute the low-voltage power distribution intelligent regulation method based on big data analysis provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0072] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A low-voltage power distribution intelligent regulation method based on big data analysis, characterized in that, include: A multi-dimensional low-voltage power distribution sample set is constructed based on big data to identify periodic events in regulation and control and their corresponding correlation factors. Based on the periodic events and their corresponding correlation factors, a multi-period detection model is constructed, and the multi-period detection model is deployed at the low-voltage power distribution control terminal to monitor the control target events in real time. When a target event for regulation is identified, the data performance of the correlation factors is used to match the regulation parameters, and linkage verification is performed based on the periodic event. For the control parameters that have passed the linkage verification, control commands are generated to adjust the low-voltage power distribution nodes corresponding to the periodic events; Specifically, based on the periodic events and corresponding correlation factors, a multi-period detection model is constructed and deployed at the low-voltage power distribution control terminal to monitor control target events in real time, including: Based on historical operating data, each periodic event is tagged, and voltage offset, load change rate, phase-to-phase current difference, reactive power fluctuation amplitude electrical factor related to the event are extracted to construct a feature matrix for periodic pattern recognition. The feature matrix is ​​input into a multi-period detection model consisting of a periodic convolutional network, a graph convolutional network, and an attention mechanism layer, so that the model can simultaneously learn the multi-period repetitive pattern of events in the time domain and the spatial correlation of events in the power distribution topology. The multi-cycle detection model is trained, optimized, and validated using a large-scale historical dataset on a cloud server to obtain a multi-cycle detection model with cycle pattern discrimination capability that can be adapted to low-voltage power distribution operation scenarios. The trained model parameters are sent to the edge electrical control nodes, and a lightweight inference engine is deployed in the edge nodes to perform multi-cycle event detection and control target event identification based on local real-time sampling data. Among them, the linkage verification based on periodic events includes: Construct an equivalent circuit model for the current region based on the electrical path corresponding to the chain of events in which the periodic event is located; The control parameters are injected into the equivalent circuit model to perform multi-scenario simulations of steady-state power flow, transient voltage response, harmonic propagation and protection coordination verification. If the simulation results show that the regulation causes voltage over-limit, equipment overload, harmonic amplification, or protection malfunction, it is determined that the linkage verification fails.

2. The intelligent low-voltage power distribution regulation method based on big data analysis according to claim 1, characterized in that, A multi-dimensional low-voltage power distribution sample set was constructed based on big data, including: Analyze abnormal events in low-voltage power distribution, including at least voltage stability, load fluctuations, and three-phase balance; Perform multi-dimensional clustering on the abnormal events to determine dimensional description labels; Using the dimensional description labels as indexes, low-voltage power distribution sample sets are collected for each dimension.

3. The intelligent low-voltage power distribution regulation method based on big data analysis according to claim 2, characterized in that, The identification of cyclical events and corresponding correlation factors includes: Based on the occurrence sequence of the abnormal events, a flow analysis is performed, using the periodic nodes of before the event, during the event, during the event adjustment, and after the event adjustment to segment the multi-dimensional low-voltage power distribution sample set. By using the sample sets of each cycle node, we can perform correlation analysis on electrical signal data, user usage data, and load change data, identify correlation factors, and establish a mapping relationship between each cycle event and the correlation factors.

4. The intelligent low-voltage power distribution regulation method based on big data analysis according to claim 3, characterized in that, Based on the occurrence sequence of the aforementioned abnormal events, a flow analysis is performed, using the periods before the event, during the event, during the event adjustment, and after the event adjustment as periodic nodes, including: Based on graph convolutional network analysis, a power distribution network topology model is established to build an electrical correlation model between nodes. Extract the temporal features of the abnormal events, including their preceding, concurrent, and subsequent occurrences, and construct the event propagation trajectory. By integrating the electrical correlation model with the event propagation path, a chain of events that occur periodically on a specific electrical path is identified. The chain of events is then segmented and labeled according to the temporal relationship between the nodes to determine the periodic nodes corresponding to the abnormal events.

5. The intelligent low-voltage power distribution regulation method based on big data analysis according to claim 1, characterized in that, Deploying the multi-cycle detection model at the low-voltage power distribution control terminal to monitor control target events in real time also includes: During the model inference process, edge nodes periodically upload locally identified suspected event sequences and electrical factor changes to the cloud. The cloud performs incremental learning and dynamic calibration of the model based on the data from the entire network, and reissues the updated model parameters for coordinated optimization between the cloud and edge nodes. When an edge node detects a suspected target event but lacks sufficient event characteristics, it triggers a cross-regional multi-node correlation analysis in the cloud. The cloud then determines whether the event has periodic characteristics that can trigger circuit regulation and feeds back the event level and regulation priority to the edge node.

6. The intelligent low-voltage power distribution regulation method based on big data analysis according to claim 1, characterized in that, If the linkage verification fails, the control parameters are iteratively adjusted based on the constraint optimization algorithm until the safe operation boundary is met.

7. The intelligent low-voltage power distribution regulation method based on big data analysis according to claim 1, characterized in that, Generate control commands to adjust the low-voltage distribution nodes corresponding to periodic events, including: Determine the type of periodic event, select the corresponding circuit action path according to the type of periodic event, and perform low-voltage power distribution node matching adjustment; When the periodic event is the daily load peak, the output voltage level is changed by controlling the cascaded H-bridge unit in the power electronic voltage regulation module, and at the same time the switch is driven to switch to the backup line, capacitor or SVG module. When the periodic event is a rapid increase in the output of distributed photovoltaic power, the voltage rise can be alleviated by adjusting the reactive power support capability of the photovoltaic inverter, switching the voltage regulation direction of the H-bridge unit, or activating the energy storage system to absorb energy. When the periodic event is three-phase imbalance, the load phase is transferred by driving the intelligent phase switching, and the three-phase balanced H-bridge compensation branch is activated to achieve synchronous balance of three-phase voltage and current.

8. A low-voltage power distribution intelligent regulation system based on big data analysis, characterized in that: The system is used to implement the low-voltage power distribution intelligent regulation method based on big data analysis as described in any one of claims 1-7, and the system comprises: A multi-dimensional low-voltage power distribution sample set construction module is used to construct a multi-dimensional low-voltage power distribution sample set based on big data, and to identify the periodic events of regulation and the corresponding correlation factors. The multi-cycle detection model construction module is used to construct a multi-cycle detection model based on the cycle events and corresponding correlation factors, and deploy the multi-cycle detection model at the low-voltage power distribution control terminal to monitor the control target events in real time. The regulation parameter matching module is used to match regulation parameters using the data performance of correlation factors when a regulation target event is identified, and to perform linkage verification based on periodic events. The low-voltage distribution node regulation module is used to generate regulation commands based on the regulation parameters that have passed the linkage verification, and to regulate the low-voltage distribution nodes corresponding to periodic events.

Images