A method, device, equipment and medium for processing power outages in power security areas

Abstract

The present invention discloses a method, device, equipment and medium for processing power outages in an electric power security area, wherein the method includes: performing time marking at the moment of power outage according to a high-precision time synchronization method, and collecting historical operation data of each node of the power system at the same time point; extracting features of the operation data based on a deep learning algorithm, extracting deep feature vectors that reflect the intrinsic structure and time scale characteristics of the data; establishing a power outage risk assessment model based on the deep feature vector, and optimizing the power outage risk assessment model using a Bayesian network or Markov decision process method; obtaining the current operation data of the power system, inputting the current operation data into the optimized power outage risk assessment model, outputting the assessment result, and constructing a power outage processing strategy executed by a safety warning terminal that matches the assessment result. The method provided by the present invention realizes accurate monitoring and effective control of the power outage risk of the power system.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method, device, equipment and medium for processing power outages in a power security area. Background Art

[0002] In power outage management, there's a technical conflict between timescale accuracy and the accuracy of outage risk assessment. On the one hand, improving timescale precision helps more accurately assess outage risk, especially when distinguishing between short, frequent outages and long, single outages. While short-lived, single outages often occur more frequently, potentially causing greater impact and inconvenience. On the other hand, improving timescale precision requires more complex systems and algorithms, which can increase system complexity and, in turn, the probability of system failures and misjudgments.

[0003] In practical power outage management, a trade-off between time accuracy and system complexity is necessary. Research has found that by constructing a risk function model with time control parameters as variables, the outage risk level at different time accuracies can be quantified. This model can intuitively demonstrate the relationship between time parameter adjustments and risk changes, providing a basis for decision-making. However, in the process of constructing the risk function model, several technical challenges remain. First, how to select appropriate time control parameters that accurately reflect the temporal characteristics of power outage events? Second, the form of the risk function must be carefully designed, considering both the impact of time accuracy and other factors, such as outage frequency and impact range. Third, the computational efficiency of the model is also a key issue. Especially in the context of large-scale power grids and massive amounts of outage data, how to improve computational speed and reduce resource consumption while ensuring model accuracy also requires further research. Summary of the Invention

[0004] The present invention provides a power outage processing method, device, equipment and storage medium for a power security area to solve the technical problem of balancing time accuracy and system complexity in actual power outage processing business, realize accurate monitoring and effective control of power system power outage risks, and improve the safety and stability of the power system.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a method for handling power outages in a power security area, comprising:

[0006] Use high-precision time synchronization methods to time-stamp the moment a power outage occurs and collect historical operating data from each node in the power system at the same time point;

[0007] A deep learning algorithm based on a convolutional neural network performs feature extraction and representation learning on the historical operation data, automatically learning local and global patterns in the historical data through convolutional layers and pooling layers, and extracting deep feature vectors that reflect the intrinsic structure and time scale characteristics of the data;

[0008] Establishing a power outage risk assessment model based on the deep feature vector, and optimizing the power outage risk assessment model using a Bayesian network or a Markov decision process method;

[0009] The current operating data of the power system is obtained, the current operating data is input into the optimized power outage risk assessment model, the assessment result is output, and a power outage processing strategy executed by the safety warning terminal that matches the assessment result is constructed.

[0010] As one of the preferred solutions, time synchronization technology is combined with power outage detection. Through high-speed voltage sampling and edge detection algorithms, the grid voltage waveform is monitored in real time and the exact moment when the power outage occurs is recorded.

[0011] As one of the preferred solutions, after collecting the operating data of each node in the power system at the same time point, it also includes:

[0012] The high-precision time synchronization method is used to obtain synchronization timestamps, and the collected power outage event data is time-calibrated and aligned. The start and end time difference of the power outage event is calculated to obtain the precise power outage duration. Aggregate statistics are performed according to different time scales to form a multi-granularity power outage duration distribution map.

[0013] Using geographic information system technology, the power outage event data is spatially associated with the power grid topology structure. Buffer analysis technology and spatial connection algorithm are used in turn to locate the impact range of the power outage event, and the number of affected users and power load size are calculated to form a visual heat map of the power outage impact range.

[0014] As one preferred solution, after extracting the deep feature vector reflecting the intrinsic structure and time scale characteristics of the data, the following is further included:

[0015] Inputting the processed deep feature vector into a multi-layer perceptron (MLP) network for training to obtain an anomaly recognition model;

[0016] Establish a historical experience database based on a graph database to obtain various abnormal pattern types and corresponding treatment plans that occur during the operation of the storage system;

[0017] The current operating data is obtained and input into the anomaly recognition model. After identifying a new anomaly pattern, the new anomaly pattern is matched based on the graph matching algorithm. If the match fails, the dynamic misjudgment risk assessment mechanism is triggered; otherwise, the corresponding disposal plan is directly called.

[0018] As one of the preferred solutions, establishing a power outage risk assessment model based on the deep feature vector and optimizing the power outage risk assessment model using a Bayesian network or Markov decision process method include:

[0019] Through high-speed data acquisition equipment and intelligent sensors, real-time collection of time series data during power grid operation;

[0020] Processing the time series data based on signal processing technology to extract power outage risk assessment indicators;

[0021] Processing the power outage risk assessment indicators using data mining and machine learning algorithms to construct a power outage risk assessment model;

[0022] According to the power outage risk assessment model, a Bayesian network inference algorithm is used to quantitatively calculate the probability parameters of each risk factor output by the power outage risk assessment model, and the probability parameters are dynamically updated through maximum likelihood estimation and expectation maximization algorithms to achieve real-time reasoning and prediction of power outage risks.

[0023] As one preferred solution, the method of constructing a matching power outage handling strategy executed by the security warning terminal based on the output results of the optimized power outage risk assessment model includes:

[0024] The power outage risk assessment result output by the power outage risk assessment model is processed using an equidistance partitioning method to obtain a corresponding risk level;

[0025] A power outage handling strategy executed by the safety warning terminal is constructed based on the risk level.

[0026] As one preferred solution, the security warning terminal includes at least one of the following:

[0027] Operation and maintenance team terminal, maintenance personnel terminal, or power responsible person terminal.

[0028] Another embodiment of the present invention provides a power outage processing device for a power security area, comprising:

[0029] The acquisition module is used to time-stamp the moment a power outage occurs based on a high-precision time synchronization method and collect historical operating data of each node in the power system at the same time point;

[0030] an extraction module for performing feature extraction and representation learning on the historical operation data using a deep learning algorithm based on a convolutional neural network, automatically learning local and global patterns in the historical operation data through convolutional layers and pooling layers, and extracting deep feature vectors that reflect the intrinsic structure and time scale characteristics of the data;

[0031] An evaluation module, configured to establish a power outage risk assessment model based on the deep feature vector and optimize the power outage risk assessment model using a Bayesian network or a Markov decision process method;

[0032] The output module is used to obtain the current operating data of the power system, input the current operating data into the optimized power outage risk assessment model, output the assessment results, and construct a power outage processing strategy executed by the safety warning terminal that matches the assessment results.

[0033] Another embodiment of the present invention provides a power outage processing device for a power security area, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the power outage processing method for the power security area as described above is implemented.

[0034] Yet another embodiment of the present invention provides a computer-readable storage medium storing a computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the power outage processing method for the power security area as described above is implemented.

[0035] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0036] (1) By installing a high-performance data acquisition system, the status of the present invention is monitored in real time, and operation data such as voltage, current, frequency and other key indicators are automatically collected, and abnormal patterns are identified, and the system operation status and potential risks are evaluated in real time, and the algorithm strategy is optimized to reduce the impact on system performance.

[0037] (2) The data patterns identified by the algorithm are compared with the patterns in the historical experience database, and the time scale parameters are automatically adjusted or the operation and maintenance personnel are prompted to perform manual inspections, thereby improving the accuracy of risk assessment.

[0038] (3) By developing a power outage risk function model, we can quantitatively analyze the power outage risk at different time scales, ensure that the model can be dynamically adjusted over time, and achieve continuous monitoring of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1It is a flowchart of a method for handling power outages in a power security area in one embodiment of the present invention;

[0040] Figure 2 It is a structural schematic diagram of a power outage processing device in a power security area in one embodiment of the present invention;

[0041] Figure 3 It is a schematic diagram of power outage processing equipment in a power security area in one embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0043] In the description of this application, the terms "first," "second," "third," etc. are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first," "second," "third," etc. may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0044] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the two components. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are for illustrative purposes only, and do not indicate or imply that the device or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0045] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this application have the same meanings as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by those skilled in the art in specific circumstances.

[0046] An embodiment of the present invention provides a method for handling power outages in a power security area. For details, see Figure 1 , Figure 1 The figure shows a flow chart of a method for handling power outages in a power security area provided in one embodiment of the present invention, which includes steps S1 to S4:

[0047] S1: Time-stamp the moment a power outage occurs using a high-precision time synchronization method, and collect historical operating data from each node in the power system at the same time point;

[0048] In this embodiment, a high-speed data acquisition card, such as a data acquisition card with a PCI-Express bus architecture, is used, equipped with a multi-channel synchronous ADC to achieve synchronous sampling of multiple measurement points. The sampling rate can reach between thousands of Hz and tens of thousands of Hz. Combined with high-speed cache and DMA transmission technology, real-time data acquisition and low-latency transmission are ensured.

[0049] Smart sensors, including voltage transformers, current transformers, and frequency measurement units, are installed at various power system nodes to convert analog signals into digital signals. The collected data is then uploaded to the data center in real time via industrial Ethernet or fiber optic communications. Distributed streaming computing platforms, such as Apache Kafka and Apache Spark Streaming, are deployed to process and analyze the massive amounts of collected power grid operation data in real time. The Apache Kafka Connect component is used to enable real-time data access, aggregating heterogeneous data streams from multiple sources into Kafka in a unified format. Data is partitioned and persistently stored in Kafka to ensure high throughput and fault tolerance.

[0050] Apache Spark SQL is used to perform structured processing on real-time data streams. By defining streaming data tables and continuous queries, real-time data conversion, filtering, and aggregation are achieved. By setting a reasonable data window interval (e.g., 5 seconds), quasi-real-time feature extraction and statistical indicator calculation, such as average voltage, current peak, and frequency deviation, are achieved, providing data support for subsequent anomaly detection and risk assessment. Highly reliable time synchronization mechanisms, such as the IEEE 1588 Precision Time Protocol (PTP), are employed. By deploying PTP master-slave clocks at each node in the power system and using GPS timing signals as the synchronization source, precise synchronization of clocks at all nodes in the entire network is achieved, with synchronization accuracy reaching sub-microseconds. This ensures the consistency and comparability of the collected data timestamps, laying the foundation for subsequent time series data analysis and causal relationship mining.

[0051] A multi-dimensional data quality monitoring and verification mechanism is built for real-time grid operation data. By defining a series of data quality indicators (such as sampling rate, timestamp continuity, and numerical range), the system uses statistical methods to calculate the values ​​of each indicator in real time and compares them with pre-set thresholds to promptly identify and address issues such as missing data and data anomalies. When data missing exceeds a certain percentage (such as 0.5%), the data repair process is automatically triggered, and the missing data is estimated and filled in through interpolation and fitting algorithms to ensure data integrity and accuracy.

[0052] Based on real-time collected and processed grid operation data, combined with historical data and expert knowledge, a multi-model fusion fault diagnosis and risk warning model is constructed. Time series feature extraction methods, such as sliding windows and wavelet transforms, are used to automatically extract key features from the data and construct high-dimensional feature vectors. Machine learning algorithms, such as support vector machines and random forests, are used to train the fault diagnosis model. Model hyperparameters are optimized through cross-validation and grid search to improve fault diagnosis accuracy. For risk warning tasks, deep learning models, such as long short-term memory (LSTM) networks, are employed to learn temporal dependencies and abnormal patterns in the data. Real-time predictions are performed using a sliding window approach, and alerts are automatically generated when the predicted risk probability exceeds a certain threshold. The warning model is continuously optimized by evaluating the model's predictive performance in real time and dynamically adjusting the threshold based on expert feedback. Fault diagnosis and risk warning results are presented in real time on a visual dashboard, and on-duty personnel are promptly notified via text messages and emails, improving the safety and reliability of grid operations.

[0053] For example, to achieve high-performance data acquisition, a PCI-Express data acquisition card with eight simultaneous sampling channels can be selected. Each channel is equipped with a 24-bit ADC with a sampling rate of 20kHz. The collected data is transferred to memory in real time via DMA, with data transmission latency controlled to less than 100 microseconds. Voltage and current transformers are installed on the high-voltage and low-voltage sides of the 220kV substation, respectively, with transformation ratios of 1000:1 and 500:1. Signal conditioning circuits condition the voltage and current signals to a ±10V range, matching the range of the data acquisition card.

[0054] Collected data is uploaded to the data center via industrial Ethernet. The Ethernet switch uses gigabit fiber ports, offering a bandwidth of up to 1 Gbps and a transmission latency of less than 1 millisecond. A Kafka cluster is deployed in the data center, consisting of three broker nodes, each equipped with 32GB of memory and a 1TB SSD drive. These nodes utilize rack-mounted servers and are interconnected via a 10 Gbps network. Kafka's data retention period is set to 7 days, with the number of partitions set to 16. The collected data stream is imported into Kafka via KafkaConnect. Kafka data is consumed using Spark Streaming. The Spark cluster consists of one master node and four worker nodes, each equipped with 64GB of memory and a 2TB hard drive. HDFS is used as the distributed file system. Spark Streaming's batch interval is set to 5 seconds. Streaming data tables are defined using Spark SQL. Real-time data aggregation is achieved through continuous queries, with statistical metrics such as average voltage, current, and frequency calculated every 5 minutes.

[0055] A PTP master clock is deployed within the substation, receiving timing signals via a GPS antenna. Other smart devices act as PTP slave clocks, synchronizing with the master clock over the network. The synchronization period is set to 1 second, and clock deviation can be controlled within 1 microsecond. In terms of data quality monitoring, collected data is checked every minute. Data quality indicators such as mean, variance, maximum, and minimum are calculated and compared with historical data. If data quality anomalies are detected, the abnormal data is automatically marked as invalid, triggering a data repair process. The invalid data is repaired using the Lagrange interpolation algorithm, with an interpolation window size of 10 data points.

[0056] A fault diagnosis model was constructed using a support vector machine algorithm. Input features included real-time indicators such as voltage, current, and frequency, as well as environmental parameters such as temperature and humidity. Sample data was selected from historical data. After data cleaning and feature engineering, the model was trained and evaluated using a 10-fold cross-validation approach, achieving an accuracy rate exceeding 95%. For risk warning, an LSTM model was used, using the past hour's data as input to predict the risk probability within the next hour. When the risk probability exceeds 80%, an alarm message was automatically generated and notified to on-duty personnel via email and text message, improving the safety of power grid operations.

[0057] Furthermore, time synchronization technology is combined with power outage detection to monitor the grid voltage waveform in real time through high-speed voltage sampling and edge detection algorithms, and record the exact moment when the power outage occurs.

[0058] In this embodiment, in order to achieve high-precision time synchronization, the present invention also adopts high-precision GPS time synchronization technology. By receiving GPS satellite signals, the system clock is synchronized in real time to ensure that the time synchronization accuracy reaches the microsecond level, providing a reliable time reference for the precise time mark of the moment the power outage occurs.

[0059] This invention combines time synchronization technology with power outage detection. Using high-speed voltage sampling and edge detection algorithms, it monitors the grid voltage waveform in real time. Once a voltage sag or loss is detected, a power outage event is immediately triggered, generating a timestamp with a microsecond response speed to record the precise moment the outage occurred. The outage's time stamp is recorded in a standardized timestamp format, including year, month, day, hour, minute, second, and millisecond. This ensures readability and manageability, facilitating subsequent data analysis and application. A redundant data storage mechanism is employed to simultaneously write the outage timestamp to multiple independent storage media, such as local Flash and remote server databases. Distributed consensus algorithms, such as Paxos or Raft, are used to ensure data consistency across these multiple storage media, improving data security and reliability and preventing the loss of critical data due to single point failures.

[0060] Using big data analysis techniques such as time series analysis and anomaly detection algorithms, we analyze and mine the temporal distribution, frequency, and duration of outages recorded from outage data to identify underlying patterns and trends. We then integrate outage data with other relevant data (such as weather conditions and load conditions) to construct a multi-dimensional feature dataset, providing a rich data foundation for subsequent machine learning model training.

[0061] By leveraging machine learning algorithms (such as LSTM neural networks), combined with historical outage data and relevant influencing factors, a power outage prediction model is established to estimate the risk of power outages within a certain timeframe. LSTM neural networks effectively capture long-term dependencies in time series data. By learning from historical data, they automatically extract the temporal characteristics of power outages and the complex relationships between influencing factors, thereby predicting future power outage risks. Continuous training and optimization of the model improves the accuracy and reliability of predictions, providing a basis for grid operations and maintenance departments to take preventive measures, reduce the probability of power outages, and improve grid stability and reliability.

[0062] For example, to achieve high-precision time synchronization, a GPS time synchronization module, such as the NEO-M8N module, can be used. By receiving the PPS (Pulse Per Second) signal broadcast by GPS satellites, it synchronizes with UTC (Coordinated Universal Time) with a time accuracy of up to 100 nanoseconds. Simultaneously, a voltage sampling module samples the grid voltage in real time, with a sampling frequency set to 10 kHz and 100 voltage samples collected per cycle. The voltage waveform is then analyzed in the frequency domain using a fast Fourier transform (FFT) algorithm to extract characteristic quantities such as the fundamental amplitude and phase angle.

[0063] When a voltage drop exceeding a threshold (such as 50% of the rated voltage) or a complete voltage loss is detected, a power outage is detected, triggering the timestamp module to generate a time stamp for the outage in the format "YYYY-MM-DD HH:mm:ss.SSS" with millisecond accuracy. The time stamp is published to the Kafka message queue via the MQTT protocol and then synchronized with the local SSD and remote MySQL database via the KafkaConnect component. Data consistency across multiple storage nodes is achieved using the Raft protocol, ensuring high data reliability.

[0064] Using the Facebook Prophet time series prediction algorithm, the company leverages the massive amount of stored outage time data to model the temporal distribution of historical outage events, characterizing the periodicity of outages (e.g., daily, weekly, and monthly). External factors such as weather conditions (e.g., temperature, humidity, and wind speed) and load conditions (e.g., current and power) are incorporated as regression variables to construct a multivariate time series prediction model. This model provides a rolling forecast of outage risk over the next seven days and automatically generates an alert when the predicted outage probability exceeds 60%. By comparing the model's predictions with actual outages, the company continuously optimizes model hyperparameters (e.g., seasonality and trend patterns) to improve forecast accuracy and enable precise estimation and early warning of outage risks.

[0065] In this embodiment, preferably, after collecting the operating data of each node of the power system at the same time point, the method further includes:

[0066] The high-precision time synchronization method is used to obtain synchronization timestamps, and the collected power outage event data is time-calibrated and aligned. The start and end time difference of the power outage event is calculated to obtain the precise power outage duration. Aggregate statistics are performed according to different time scales to form a multi-granularity power outage duration distribution map.

[0067] Using geographic information system technology, the power outage event data is spatially associated with the power grid topology structure. Buffer analysis technology and spatial connection algorithm are used in turn to locate the impact range of the power outage event, and the number of affected users and power load size are calculated to form a visual heat map of the power outage impact range.

[0068] In this embodiment, the outage event data output by the data acquisition system is time-calibrated and aligned based on synchronized timestamps obtained using high-precision time synchronization technology. The precise outage duration is calculated by calculating the start and end time difference between the outage events. Aggregated statistics are then generated at different time scales, such as 15 minutes, one hour, and one day, to form a multi-granular distribution map of outage duration. Using Geographic Information System (GIS) technology, the outage event data is spatially correlated with the grid topology. Using algorithms such as buffer zone analysis and spatial joins, the impact of the outage event is quickly located. Key indicators, such as the number of affected users and power load, are calculated to create a visual heat map of the outage's impact.

[0069] Using data mining and machine learning algorithms, such as K-means clustering and Apriori association rule mining, we cluster and analyze the spatiotemporal distribution patterns of power outages, identifying patterns in high-incidence areas and periods. Combined with external factors such as meteorological data and equipment operating data, we construct a power outage risk prediction model to quantitatively assess and categorize future power outage risks. Based on the level of power outage risk, we dynamically adjust the strategies and rules of the risk control system. By setting different thresholds and trigger conditions, we achieve hierarchical control and adaptive response to power outage risks.

[0070] To address the complexity of risk control systems, the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation methods were used to qualitatively and quantitatively assess factors such as the number of risk control strategies, the complexity of rules, and the length of the decision chain. A three-level hierarchical structure consisting of objectives, criteria, and scenarios was established, and a pairwise comparison matrix was constructed. The maximum eigenvalue and consistency ratio of the judgment matrix were calculated to determine the weights of each factor. Then, based on the membership function, the membership of the comments for each factor was calculated to construct a fuzzy evaluation matrix. Finally, a weighted average synthesis was performed to obtain a comprehensive complexity index for the risk control system. This was then compared with industry benchmarks to promptly identify and adjust overly complex or overly simplistic risk control mechanisms. Graph theory and complex network theory were also used to model and analyze the topological structure and relationships of the risk control system. By calculating metrics such as average path length, clustering coefficient, and centrality, the robustness and vulnerability of the risk control system were assessed. Structural optimization and simplification were then implemented to improve the efficiency and reliability of risk control.

[0071] During the quantitative assessment process, Monte Carlo simulation and sensitivity analysis were used to perform multiple random sampling and perturbations on key parameters such as time synchronization accuracy, data acquisition frequency, outage duration, and outage impact range. Latin hypercube sampling was used to generate a large number of random sample points within the parameter range. The simulation model was then used to calculate the outage handling timescale and risk control complexity index corresponding to each sample point, resulting in a set of simulation results. Statistical methods such as variance analysis and correlation analysis were used to assess the impact of each parameter on the simulation results. The Sobol sensitivity index was calculated to identify the key parameters that contributed most to the model output, providing decision support for parameter optimization and resource allocation. The results of the quantitative assessment were fed back to the time synchronization system and data acquisition system in real time, establishing a closed-loop feedback mechanism for outage handling and risk control.

[0072] Based on the evaluation results, parameters such as time synchronization accuracy and data collection frequency are dynamically adjusted to continuously improve data quality and analysis accuracy. At the same time, the effectiveness of risk control is fed back to the risk prediction model and knowledge graph. Through incremental learning and parameter optimization, risk control strategies and rules are continuously improved, forming a self-optimizing and self-improving risk management closed loop, continuously improving the intelligence level and actual effectiveness of power outage handling and risk control. By establishing a knowledge graph and case library for power outage events, the causes, impacts, and handling processes of historical power outage events are structured and semantically associated, forming a complete power outage response knowledge system. When a new power outage event occurs, similar cases are quickly retrieved through knowledge reasoning and case matching algorithms. Based on the handling experience and effectiveness evaluation of the cases, preliminary response plans and decision recommendations are automatically generated to assist operation and maintenance personnel in power outage handling and risk control.

[0073] For example, in the spatial correlation analysis between power outage event data and power grid topology, the power outage event data and power grid GIS data are first imported into the ArcGIS platform. Using the spatial join tool, the power outage event point is used as the target element and the power grid lines and equipment are used as the connection elements. Within a search radius of 100 meters, the two types of elements are spatially matched, and the attribute information of the power grid elements, such as the line to which they belong, the type of equipment, the power supply area, etc., are associated with the power outage event point. Then, through buffer analysis, buffer areas with different radii such as 500 meters, 1 kilometer, and 2 kilometers are generated with the power outage event point as the center, and overlaid with the power user distribution layer to count the number and type of users in each buffer zone. At the same time, using spatial interpolation methods such as Kriging interpolation, a spatial distribution raster layer of the power outage duration is generated according to the power outage duration of the power outage event point, and weighted overlay is performed with the power grid load distribution layer to obtain the spatial distribution of power load loss during the power outage. Use hotspot analysis tools, such as Getis-OrdGi*, to identify areas with high incidence of power outages and power outage losses, generate hotspot distribution maps, and provide a spatial decision-making basis for power outage risk assessment and prevention and control.

[0074] To quantitatively assess the uncertainty of key parameters, the distribution type and range of the parameters were first estimated. For example, time synchronization accuracy follows a normal distribution with a mean of 1 millisecond and a standard deviation of 0.2 milliseconds; data acquisition frequency follows a uniform distribution with a range of 10 to 60 seconds; outage duration follows an exponential distribution with a mean of 30 minutes; and the outage impact area follows a triangular distribution with a minimum of 1 square kilometer, a most likely value of 5 square kilometers, and a maximum of 10 square kilometers. Then, using Latin hypercube sampling, 1,000 sample points were drawn within the parameter distribution range to generate the parameter input matrix. This input matrix was then applied to a simulation model of outage handling timescales and risk control complexity, and 1,000 Monte Carlo simulations were performed to determine the distribution characteristics and uncertainty range of the model output.

[0075] Through sensitivity analysis, we calculated the Sobol first-order index and overall index of each parameter's impact on the model output, identifying the key parameters with the greatest impact on the output. We then analyzed the response surface relationship between these parameters and the output, providing a quantitative basis for parameter optimization. For example, the data acquisition frequency had a first-order index of 0.35 and an overall index of 0.52, indicating that it had the greatest impact on the outage timescale statistics, and ensuring its stability and reliability was a priority.

[0076] S2: A deep learning algorithm based on a convolutional neural network performs feature extraction and representation learning on the historical operation data. The algorithm automatically learns local and global patterns in the historical operation data through convolutional layers and pooling layers, and extracts deep feature vectors that reflect the intrinsic structure and time scale characteristics of the data.

[0077] In this embodiment, a deep learning algorithm based on a convolutional neural network (CNN) is used to perform feature extraction and representation learning on the collected time series data. The local and global patterns in the data are automatically learned through convolutional layers and pooling layers, and deep feature vectors reflecting the intrinsic structure and time scale characteristics of the data are extracted.

[0078] The extracted deep feature vectors are fed into a multi-layer perceptron (MLP) network. Through fully connected layers and activation functions, the features are nonlinearly transformed and classified. The MLP network is trained to identify and classify different types of abnormal patterns, determining the possible types and confidence levels of abnormal patterns in the data. Based on domain knowledge and historical data, a mapping relationship between abnormal patterns and risk levels is established, building a risk assessment rule library.

[0079] The identified anomaly pattern type and confidence level are matched against the risk assessment rule base to infer the current system operating status in real time. Based on the severity and duration of the anomaly pattern, the system's potential risk level is dynamically assessed. A rule-based reasoning engine automatically generates risk warnings and recommended response measures. A visual dashboard displays the system's operating status and risk level in real time, allowing operations personnel to quickly understand the system's health.

[0080] To address massive amounts of time series data and complex anomaly pattern recognition tasks, we employ distributed computing frameworks such as Apache Spark and TensorFlow Distributed Edition to distribute data processing and model training tasks across multiple computing nodes for parallel execution, fully utilizing the cluster's computing resources and improving data processing and model training efficiency. Leveraging Spark's RDD (Resilient Distributed Dataset) and DAG (Directed Acyclic Graph) scheduling mechanisms, we design appropriate data sharding and task scheduling strategies to optimize data transmission between nodes and load balancing of computing tasks, minimizing task execution latency and resource consumption. We employ incremental learning and online learning strategies to continuously optimize and update the anomaly pattern recognition model.

[0081] Newly collected data is streamed into the model in real time, and the model parameters are fine-tuned using the backpropagation algorithm, allowing the model to adapt to dynamic changes in the data and improve the accuracy of abnormal pattern recognition. At the same time, active learning is used to select the most representative and informative sample data through uncertainty sampling. High-quality training data is obtained through manual labeling, and the model is regularly trained and optimized offline to improve its generalization performance.

[0082] To address algorithmic resource consumption, model compression and acceleration techniques are employed to minimize the algorithm's computational and storage overhead while ensuring anomaly pattern recognition performance. Knowledge distillation techniques are used to transform complex deep learning models into more lightweight student models. Soft labels and temperature parameters are used to control knowledge transfer, reducing the model's parameter count and computational complexity. Model quantization techniques are employed to convert floating-point operations in the model into fixed-point operations. Through quantized mapping of weights and activations, the model's numerical precision is reduced, improving computational efficiency and deployment speed. Regarding algorithm performance evaluation and optimization, anomaly pattern recognition algorithms are comprehensively evaluated from the perspectives of time complexity, space complexity, precision, and recall. Big-O notation is used to analyze the algorithm's asymptotic time complexity, and the algorithm's computational efficiency and scalability are evaluated by varying data scale and batch size. Memory analysis tools are used to assess the algorithm's memory usage and data read and write overhead, optimizing data memory layout and access patterns. Cross-validation and leave-one-out methods are used to evaluate the algorithm's precision, recall, F1 score, and other metrics on various datasets to analyze its generalization and robustness.

[0083] In response to performance bottlenecks and optimization points, measures such as feature selection, model pruning, and parameter tuning are taken to continuously improve the performance and effectiveness of the algorithm. A comprehensive model monitoring and evaluation mechanism is established to continuously monitor and evaluate the performance of the abnormal pattern recognition algorithm. By designing a reasonable evaluation index system and benchmarking scheme, the performance of the model is regularly evaluated and compared, problems in the model are promptly discovered and located, and the performance and effectiveness of the abnormal pattern recognition algorithm are continuously improved through model optimization and algorithm improvement. At the same time, a model version management and rollback mechanism is established to manage and track iterative updates of the model, ensuring that when problems arise in the new version of the model, it can be quickly rolled back to the stable version to ensure the continued stable operation of the power system.

[0084] For example, the following method can be used to implement the abnormal pattern recognition algorithm. A CNN model containing three convolutional layers and two pooling layers is used. The first convolutional layer contains 16 3x3 convolution kernels, the second convolutional layer contains 32 3x3 convolution kernels, and the third convolutional layer contains 64 3x3 convolution kernels. The ReLU activation function is used between the convolutional layers, and the pooling layer uses maximum pooling with a stride of 2. The output of the CNN model is transformed through a fully connected layer containing 128 neurons, and then classified through a Softmax layer containing N neurons, where N is the number of abnormal pattern categories. The Adam optimizer is used, the learning rate is set to 0.001, the batch size is set to 128, and training is performed for 100 epochs.

[0085] During the construction of the risk assessment rule base, we analyzed anomaly patterns in historical data and identified five common anomaly patterns: data mutation, data stagnation, data loss, abnormal data fluctuation, and abnormal data trend. These patterns correspond to five risk levels, ranging from 1 to 5, from low to high. Risk assessment rules were designed based on the confidence level and duration of the anomaly pattern. For example, if the confidence level of an anomaly pattern exceeds 0.8 and the duration exceeds 10 minutes, the risk level increases by one level.

[0086] In the design of the distributed computing framework, Spark's RDD programming model is used to partition time series data by timestamp, with each partition containing 100,000 data items. Data sharding is performed using Spark's HashPartitioner to ensure that data from the same device is allocated to the same partition. In the implementation of incremental learning and active learning strategies, the model is incrementally trained for every 10,000 new data items received, with a learning rate of 0.0001 and a batch size of 256, for 10 epochs. A maximum entropy-based uncertainty sampling strategy is employed, selecting the 1,000 lowest-confidence data items, obtaining labels through manual annotation, and adding them to the training set for offline training. Offline training is conducted weekly with a learning rate of 0.0005 and a batch size of 512, for 50 epochs.

[0087] To compress and accelerate the model, knowledge distillation technology was used. A lightweight CNN model consisting of two convolutional layers and one pooling layer was selected as the student model. The temperature parameter of the teacher model's softmax layer was set to 10, and the temperature parameter of the student model was set to 1. Training was performed for 100 epochs. 8-bit fixed-point quantization was used to quantize the model's weights and activation values. The quantized model size was reduced by 4 times, while the inference speed was increased by 3 times.

[0088] When evaluating algorithm performance, a 10-fold cross-validation approach was used to assess the model's precision, recall, and F1 score on the anomaly pattern recognition task. The average precision reached 95%, the average recall reached 93%, and the average F1 score reached 94%. Through feature selection and model pruning techniques, the number of CNN model parameters was reduced by 30%, and inference latency was lowered by 20%, further improving the algorithm's real-time performance and resource efficiency.

[0089] In this embodiment, preferably, after extracting the depth feature vector reflecting the intrinsic structure and time scale characteristics of the data, the method further includes:

[0090] Inputting the processed deep feature vector into a multi-layer perceptron (MLP) network for training to obtain an anomaly recognition model;

[0091] Establish a historical experience database based on a graph database to obtain various abnormal pattern types and corresponding treatment plans that occur during the operation of the storage system;

[0092] The current operating data is obtained and input into the anomaly recognition model. After identifying a new anomaly pattern, the new anomaly pattern is matched based on the graph matching algorithm. If the match fails, the dynamic misjudgment risk assessment mechanism is triggered; otherwise, the corresponding disposal plan is directly called.

[0093] In this embodiment, a time series anomaly detection algorithm based on deep learning is used to adaptively model and identify abnormal patterns of system operation data collected in real time. The time dependency of time series data is modeled through a long short-term memory network (LSTM), and the local and global features of the data are extracted in combination with a convolutional neural network (CNN), thereby achieving accurate identification of abnormal patterns.

[0094] During the model training process, a semi-supervised learning method is adopted. A large amount of unlabeled historical data is first used for pre-training. The normal pattern of the data is preliminarily learned by reconstructing the loss function that minimizes the error. Then a small amount of labeled abnormal data is used for fine-tuning. The cross-entropy loss function and Adam optimization algorithm are used to improve the model's ability to recognize abnormal patterns. The generalization performance of the model is evaluated through cross-validation and early stopping strategies.

[0095] Establish a historical experience database based on a graph database to store various abnormal patterns and corresponding disposal plans that appear during the operation of the system. Use a graph structure to represent the correlation and evolution laws between abnormal patterns to form an abnormal pattern knowledge graph. When the algorithm identifies a new abnormal pattern, it uses a graph matching algorithm to retrieve similar or related abnormal patterns in the historical experience database, and uses metrics such as cosine similarity or edit distance to calculate the similarity score between the new pattern and the known pattern. If the highest score is lower than the preset threshold (such as 0.8), it is judged as a match failure, triggering a dynamic misjudgment risk assessment mechanism; otherwise, the corresponding disposal plan is directly called.

[0096] The state-based misjudgment risk assessment mechanism uses Bayesian reasoning and decision tree algorithms to assess the credibility and conduct risk analysis on anomaly patterns identified by the algorithm. Based on the anomaly pattern's characteristic attributes and frequency of occurrence, the naive Bayesian algorithm is used to calculate the posterior probability of its category, generating a confidence score. Based on expert knowledge and empirical data, appropriate prior probabilities and conditional probability distributions are set. Then, taking into account the current system operating status and environmental factors, the decision tree algorithm is used to assess the anomaly pattern's potential impact and degree of harm to the system. A risk level is generated using the Gini index or information gain ratio as a classification criterion. Pruning and optimization algorithms are used to control the tree's complexity and overfitting risk. If the confidence score falls below a threshold (e.g., 0.6) and the risk level rises above a threshold (e.g., level 3), a misjudgment risk is identified, triggering the adaptive time-scale adjustment mechanism. This mechanism dynamically adjusts the data sampling frequency and time window size to adapt to the time-scale characteristics of different anomaly patterns.

[0097] When the risk of misjudgment occurs, the time window is first expanded by a factor of 2 and the data sampling frequency is reduced to half to obtain more contextual information and capture the long-term trends and evolution of the anomaly pattern. If the risk of misjudgment persists after adjustments, the time window is further expanded by a factor of 3 and the data sampling frequency is increased to a factor of 2 to capture the fine-grained characteristics and instantaneous changes of the anomaly pattern. If the misjudgment rate decreases by no more than 5% after three consecutive adjustments to the time scale parameters, a manual intervention mechanism is triggered, automatically sending an alert to operations personnel, indicating the suspected anomaly pattern and related data fragments, for manual review and judgment. Operations personnel can use visualization tools to view the anomaly pattern's detailed information and contextual data, and, based on their experience and domain knowledge, determine whether it is a true anomaly or a misjudgment. If confirmed as an anomaly, it is marked as a new anomaly pattern, added to the historical experience database, and a corresponding remediation plan is designed. If it is determined to be a misjudgment, the algorithm model and rule thresholds are further optimized to improve recognition accuracy.

[0098] During the dynamic misjudgment risk assessment process, a closed-loop linkage mechanism between abnormal pattern recognition and risk assessment is established. The results of abnormal pattern recognition are fed back to the risk assessment module in real time, dynamically updating the prior probabilities and conditional probabilities in the risk assessment model. The risk assessment results are also fed back to the abnormal pattern recognition module, dynamically adjusting the similarity threshold and feature weights to achieve real-time assessment and dynamic control of system operation risks. To quantitatively evaluate the effectiveness of the dynamic misjudgment risk assessment mechanism, a set of risk assessment accuracy metrics is designed, including key indicators such as abnormal pattern recognition rate, false positive rate, missed positive rate, and average response time. Regular offline analysis and backtesting of system operation data is performed, using evaluation tools such as confusion matrices and ROC curves to calculate the values ​​of various indicators and compare them with benchmark values ​​to promptly identify and optimize deficiencies in the algorithm model and evaluation mechanism.

[0099] For example, the following method can be used to implement an abnormal pattern recognition algorithm. First, a historical dataset containing 10,000 unlabeled samples and 1,000 labeled samples is used, with a ratio of 9:1 between normal and abnormal samples. During the pre-training phase, an LSTM autoencoder is used to reconstruct the unlabeled data. The normal patterns of the data are learned by minimizing the reconstruction error. Training is performed for 100 epochs with a learning rate of 0.001.

[0100] During the fine-tuning phase, a CNN model was stacked on top of the LSTM model. Supervised learning was performed using labeled data. The model's anomaly recognition ability was improved by minimizing cross-entropy loss. Training was performed for 50 epochs with a learning rate of 0.0001. The model's generalization performance was evaluated using 5-fold cross-validation, achieving an average accuracy of over 95%. To construct the historical experience database, the Neo4j graph database was used, modeling anomaly patterns and resolution plans as nodes and relationships, respectively. Graph traversal and matching were implemented using the Cypher query language.

[0101] When matching anomaly patterns, the cosine similarity algorithm is used to calculate the similarity score between new and known patterns, with a threshold set to 0.8. In dynamic misjudgment risk assessment, the naive Bayes algorithm is used to calculate the posterior probability of the anomaly pattern. The prior probability is statistically derived from historical data, and the conditional probability assumes independence of each feature attribute. The CART decision tree algorithm is used to assess the risk level of the anomaly pattern. The feature space is partitioned using the Gini index, and a decision tree with a maximum depth of 5 is generated. The tree is optimized using a cost-complexity pruning algorithm. The confidence score and risk level thresholds are set to 0.6 and 3, respectively. In adaptive time scaling, the initial time window is 10 minutes, and the data sampling frequency is 1 second. When the risk of misjudgment occurs, the time window is expanded to 20 minutes, and the sampling frequency is reduced to 2 seconds. If the risk persists, the time window is expanded to 30 minutes, and the sampling frequency is increased to 0.5 seconds. After three consecutive adjustments, if the misjudgment rate decreases by no more than 5%, manual intervention is triggered. In the closed-loop linkage mechanism, online learning algorithms such as AdaBoost and random forest are used to update the risk assessment model and abnormal pattern recognition model in real time, and the system's adaptability is continuously improved through incremental training and parameter tuning.

[0102] The evaluation metric system uses common metrics such as accuracy, precision, recall, and F1 value, calculated using a confusion matrix. The ROC curve and AUC values ​​are used to evaluate the model's classification performance. Through regular offline analysis and manual feedback, threshold settings are optimized, raising the abnormal pattern recognition rate to over 99%, keeping the false positive rate below 1%, and reducing the average response time to less than 10 seconds, achieving efficient and accurate dynamic false positive risk assessment.

[0103] S3: Establishing a power outage risk assessment model based on the deep feature vector, and optimizing the power outage risk assessment model using a Bayesian network or a Markov decision process method;

[0104] In this embodiment, preferably, establishing a power outage risk assessment model based on the processed operating data and optimizing the power outage risk assessment model using a Bayesian network or Markov decision process method include:

[0105] Through high-speed data acquisition equipment and intelligent sensors, real-time collection of time series data during power grid operation;

[0106] Processing the time series data based on signal processing technology to extract power outage risk assessment indicators;

[0107] Processing the power outage risk assessment indicators using data mining and machine learning algorithms to construct a power outage risk assessment model;

[0108] According to the power outage risk assessment model, a Bayesian network inference algorithm is used to quantitatively calculate the probability parameters of each risk factor output by the power outage risk assessment model, and the probability parameters are dynamically updated through maximum likelihood estimation and expectation maximization algorithms to achieve real-time reasoning and prediction of power outage risks.

[0109] In this embodiment, high-speed data acquisition equipment and intelligent sensors are used to collect key parameters such as voltage, current, and power during grid operation in real time. Time series data is stored at millisecond-level granularity, and preprocessing operations such as cleaning, normalization, and standardization are performed on the data to improve data quality and usability. Signal processing techniques such as wavelet transform and Fourier transform are used to extract features and reduce dimensionality of the collected time series data. Key indicators that reflect the characteristics of power outage risks, such as voltage sag frequency, current distortion rate, and harmonic content, are extracted, and a multi-dimensional power outage risk assessment indicator system is constructed.

[0110] Data mining and machine learning algorithms, such as association rule mining and time series anomaly detection, are used to analyze the spatiotemporal relationships and evolutionary patterns between indicators for power outage risk assessment. This allows for the identification of precursory characteristics and influencing factors for power outages. A causal network model of power outage risk is constructed to reveal the underlying mechanisms of risk propagation. Based on this causal network, a Bayesian network inference algorithm is employed to quantify the prior probabilities and conditional probabilities of each risk factor. Maximum likelihood estimation and expectation maximization algorithms are then used to dynamically update probability parameters, enabling real-time reasoning and prediction of power outage risk.

[0111] To address the temporal nature and uncertainty of power outage risk, a Markov decision process model is employed to categorize power outage risk states into multiple levels, including normal, warning, and severe. A transition probability matrix and immediate reward function are defined between risk states. Through policy iteration and value iteration algorithms, the optimal risk decision strategy is solved, enabling dynamic assessment and early warning of power outage risks at different time scales. Based on the optimal strategy of the Markov decision process, risk control measures and emergency response plans are designed, resulting in a multi-scenario, multi-level emergency response plan.

[0112] Through reinforcement learning algorithms such as Q-learning and the Sarsa algorithm, risk control strategies are continuously optimized and improved, enhancing the effectiveness and reliability of risk prevention and control. To adapt to the dynamic changes and uncertainties of the power grid operating environment, methods such as adaptive dynamic Bayesian networks and non-stationary Markov decision processes are used to perform online learning and updates on the power outage risk assessment model. Adaptive dynamic Bayesian networks dynamically adjust the network structure and conditional probability distribution to adapt to changes in data distribution by introducing time-varying parameters and adaptive mechanisms. Non-stationary Markov decision processes dynamically adjust the optimal decision strategy to adapt to environmental uncertainty by considering the time-varying characteristics of state transition probabilities and reward functions. Based on newly collected data and expert feedback, the model is optimized and incrementally updated in real time using techniques such as online gradient descent and incremental decision trees, improving its dynamic adaptability.

[0113] Establish a closed-loop feedback mechanism for power outage risk assessment and regularly evaluate and assess the effectiveness of risk assessment models and methods. Use local and global sensitivity analysis methods to assess the model's sensitivity to different parameters and factors, and quantify the model's uncertainty and reliability through techniques such as Monte Carlo simulation. Introduce model stability indicators, such as confidence intervals for model parameters and volatility of prediction errors, to assess the model's stability and reliability. Analyze and summarize actual power outage events to verify and calibrate the model. Use feedback to drive parameter adjustments and structural optimization, continuously improving and refining the risk function model and enhancing the accuracy and practicality of risk prediction.

[0114] Establish a visual simulation and early warning platform for power outage risks. Using data visualization technologies such as heat maps and radar charts, intuitively display the distribution characteristics and evolution trends of power outage risks across time and space. Set risk warning thresholds and trigger rules. When the risk level exceeds the preset threshold, automatically generate risk warning information and alerts, which are pushed to relevant responsible personnel in real time via text messages, emails, and other means, improving the timeliness of risk perception and response. Compare and analyze risk assessment results with actual operating data to evaluate the relevance and effectiveness of risk control measures, and make timely adjustments and optimizations. This creates a closed-loop risk management system with continuous improvement and dynamic optimization, providing reliable guarantees for the safe and stable operation of the power system. By continuously iterating and improving the power outage risk function model, multi-timescale, all-round risk perception and early warning for the power system can be achieved, enhancing the safety, reliability, and resilience of power grid operations.

[0115] For example, during the data acquisition and preprocessing phase, a high-speed data acquisition card, such as the NIPXIe-6368, is used to sample grid operating parameters. The sampling rate is set to 10kHz, and the data bit width of each sampling channel is 16 bits. A data acquisition program is developed using LabVIEW to achieve real-time storage and management of massive time series data. Wavelet denoising is performed on the collected data using the Daubechies4 wavelet basis function with a decomposition level of 5. Soft thresholding is performed on the wavelet coefficients using the thresholding method to remove high-frequency noise. The data is then normalized using the Z-score, mapping the data to the range of 0 to 1 to eliminate dimensionality effects. Principal component analysis (PCA) is then used to extract features from the data. The top k principal components with a cumulative variance contribution greater than 95% are selected as key features to achieve data dimensionality reduction and compression. During the causal relationship mining phase, the Apriori association rule mining algorithm was used, with a minimum support of 0.05 and a minimum confidence of 0.8. Association rules between power outage risk indicators were discovered, such as "a high frequency of voltage sags and a current distortion rate greater than 5% indicates a severe power outage risk." Furthermore, a Gaussian anomaly detection algorithm was used to detect anomalies in time series data, with an anomaly threshold set at 3 standard deviations, to identify precursory features of power outages. During the Bayesian network construction phase, the TabuSearch algorithm was used to optimize the network structure, with a taboo table length of 10 and a number of iterations of 100. The minimum description length (MDL) score was used to assess network quality. The expectation maximization (EM) algorithm was used to iteratively calculate the conditional probability table, with a convergence threshold of 0.001.

[0116] During the Markov decision process optimization phase, a value iteration algorithm was used with a discount factor of 0.9 and a convergence threshold of 0.0001 to solve for the optimal state value function and decision strategy. During the adaptive dynamic adjustment phase, a particle swarm optimization (PSO) algorithm was used with a particle number of 50, an acceleration factor of 1.5, and an inertia weight of 0.7 to perform online optimization of the Bayesian network and Markov decision process parameters to adapt to dynamic changes in the data distribution.

[0117] In the sensitivity analysis stage, the Sobol global sensitivity analysis method was used to generate 5000 sets of Sobol sequences, calculate the first-order and total sensitivity index of each input parameter, evaluate the importance and interaction of the parameters, and perform 1000 random samplings through Monte Carlo simulation to calculate the mean, variance and quantile of the model output to quantify the uncertainty of the model.

[0118] In terms of visual early warning, Grafana is used to build a real-time monitoring panel, and the risk level threshold is set to 0.6. When the risk assessment value exceeds the threshold, an alarm is automatically triggered, and the warning information is pushed to the Web front end in real time through WebSocket, realizing real-time perception and visualization of risks.

[0119] S4: Acquire current operating data of the power system, input the current operating data into an optimized power outage risk assessment model, output an assessment result, and construct a power outage handling strategy that matches the assessment result and is executed by a security warning terminal.

[0120] In this embodiment, preferably, the step of constructing a matching power outage handling strategy executed by the security warning terminal based on the output result of the optimized power outage risk assessment model includes:

[0121] The power outage risk assessment result output by the power outage risk assessment model is processed using an equidistance partitioning method to obtain a corresponding risk level;

[0122] A power outage handling strategy is constructed based on the risk level and executed by a security warning terminal, wherein the security warning terminal includes at least one of the following:

[0123] Operation and maintenance team terminal, maintenance personnel terminal, or power responsible person terminal.

[0124] In this embodiment, based on the outage risk assessment results calculated using a risk function model, an equal-distance partitioning method is used to categorize the risk range into multiple levels: low, medium, high, and extremely high. Risk values ​​range from 0 to 1, with 0-0.3 representing low risk, 0.3-0.6 representing medium risk, 0.6-0.8 representing high risk, and 0.8-1 representing extremely high risk. This classification standard comprehensively considers factors such as the safety margin of power grid operation and the distribution characteristics of historical risk events, and can reasonably reflect the level of risk. Through power system vulnerability analysis, weak links and key nodes in the power grid are identified. Indicators such as the power flow distribution and static safety margin of each node and line are calculated to assess their impact on the outage risk. Based on this information, outage handling resources, such as backup power sources and emergency generators, are optimally allocated to improve the power grid's risk mitigation capabilities. The output of the risk function model is monitored in real time to dynamically track the changing trends of the risk value. When the risk value exceeds a preset threshold, an early warning mechanism is automatically triggered. Different early warning strategies and notification methods are employed depending on the risk level.

[0125] For high-risk levels, an audible and visual alarm and SMS notification are triggered, and the emergency plan is activated. For extremely high-risk levels, in addition to the audible and visual alarm and SMS notification, a phone notification is also triggered, and the emergency plan is elevated to the highest priority. Based on the risk level and scope of impact, the corresponding operation and maintenance team and professionals are automatically matched, and a detailed risk analysis report and disposal recommendations are delivered to guide operation and maintenance personnel in risk prevention and control and emergency response. Based on risk assessment results at different time scales, the time window and granularity of power outage processing are flexibly adjusted to enable multi-time-scale risk analysis. If the short-term risk value within one hour exceeds the threshold, the raw data is aggregated from minute to hourly granularity to improve the sensitivity and real-time nature of risk perception. If the long-term risk value exceeds the threshold for more than one day, the hourly granularity data is aggregated to daily granularity to capture the long-term trends and cumulative effects of the risk. Through dynamic adjustment of the time scale, in-depth risk prevention and control and panoramic perception are achieved.

[0126] Based on risk assessment results and power system vulnerability analysis, multi-scenario, multi-level emergency response plans and response plans are automatically generated. Using case-based reasoning and knowledge graph technology, similar historical risk scenarios and response experiences are matched. Combined with the current grid operating status and load characteristics, targeted emergency response plans are dynamically generated, including load shedding, equipment isolation, and dispatch strategies. Simulation and evaluation are then used to optimize the feasibility and effectiveness of the emergency response plans. During emergency response plan execution, the grid operating status and risk level are monitored in real time. The effectiveness and impact of emergency measures are evaluated through real-time analysis of data such as fault information, alarm events, and operation logs. Based on the feedback, emergency response strategies are dynamically optimized and adjusted, such as adjusting load shedding ratios and modifying dispatch plans, to ensure the accuracy and adaptability of emergency response measures. Closed-loop control is implemented for risk assessment and emergency response, integrating the risk function model with the emergency command and dispatch system to automate the entire process of risk perception, assessment, early warning, decision-making, execution, and feedback.

[0127] The reinforcement learning algorithm is used to adaptively adjust the risk threshold, the Bayesian optimization algorithm is used to optimize the hyperparameters of the risk function model, and the incremental learning algorithm is used to dynamically update the emergency knowledge base, etc., to continuously improve the intelligence level and real-time performance of risk management.

[0128] Regarding the dynamic adjustment of risk thresholds, an adaptive mechanism is introduced to dynamically adjust risk thresholds based on historical data and expert experience to address changes in grid operating conditions and iterative updates to the risk assessment model. Methods such as sliding windows and weighted averages are used to calculate dynamic thresholds for risk values ​​in real time. Through dynamic adjustment of the thresholds, adaptive optimization of risk assessment criteria is achieved, improving the flexibility and robustness of threshold settings. The automated response process for exceeding risk thresholds is further refined, including the generation, transmission, escalation, and release of warning information; the task allocation, scheduling, and coordination of operations and maintenance personnel; and the initiation, execution, and tracking of emergency measures. A workflow engine and task scheduling system are used to automate the orchestration and execution of the response process. Real-time monitoring and exception handling mechanisms are implemented to ensure the reliability and continuity of the response process. Furthermore, a contingency plan knowledge base and expert decision support system are established to provide real-time assistance and knowledge empowerment to operations and maintenance personnel, improving the efficiency of human-machine collaboration and emergency response capabilities.

[0129] A risk management visualization platform for operations and maintenance personnel will be built. Using dashboards, geographic information systems, and other technologies, it will intuitively display information such as the grid's risk distribution, emergency resource allocation, and the execution of disposal tasks. It will also provide interactive decision support and expert assistance, strengthening human-machine collaboration and improving the efficiency and accuracy of risk management. At the same time, it will accumulate empirical data and case studies related to risk management to form a reusable and transferable risk management knowledge base, providing data support and knowledge accumulation for subsequent model optimization and decision-making improvements.

[0130] For example, in the process of risk level classification, the risk value output by the risk function model is normalized, and the Min-Max normalization method is used to map the risk value to the 0-1 interval. Then, according to the equal distance division principle, the 0-1 interval is divided into 4 sub-intervals, corresponding to low risk, medium risk, high risk and extremely high risk levels, respectively. The low risk interval is [0, 0.3], the medium risk interval is (0.3, 0.6], the high risk interval is (0.6, 0.8], and the extremely high risk interval is (0.8, 1]. In terms of real-time monitoring and early warning, a sliding window algorithm is used, with the window size set to 1 hour and the step size set to 1 minute. The average risk value in each window is calculated in real time. When the average risk value exceeds the preset threshold (such as 0.7), the early warning mechanism is triggered.

[0131] For high-risk levels, warning notifications are sent every five minutes via text message and email. For extremely high-risk levels, warning notifications are sent every minute via voice broadcast over the phone in addition to text message and email. Warning information is also pushed to the operations and maintenance personnel's mobile terminal app. The app uses push notifications and pop-up notifications to ensure the timeliness and reach of warning information. For emergency plan generation and optimization, a multi-objective optimization model based on a genetic algorithm is used. Minimizing the scope of power outages and power restoration time is the optimization goal, with load shedding ratios, equipment isolation status, and scheduling strategies as decision variables. Through genetic algorithm operations such as selection, crossover, and mutation, the optimal emergency decision combination is searched. In each iteration, the fitness value of the emergency decision combination is calculated through network outage simulation and risk assessment. Based on the fitness value, the genetic code of the individuals in the population is updated, and the search direction is continuously optimized until the global optimal solution that meets the constraints is found, forming the final emergency plan.

[0132] In terms of knowledge base construction and learning, ontology-based knowledge representation and reasoning technology is used to construct a risk management domain ontology, formally represent and semantically organize knowledge such as risk scenarios, assessment models, and emergency plans, use graph-based databases (such as Neo4j) to store ontology knowledge, and implement knowledge retrieval and reasoning through the SPARQL query language.

[0133] In the expert system, a rule-based reasoning engine is used to encode expert experience and judgment rules, forming a decision-making rule base for risk management. Through forward and reverse reasoning, functions such as risk assessment, early warning decision-making, and emergency assistance are implemented. At the same time, machine learning technologies such as incremental learning and active learning are used to continuously optimize and expand the knowledge base and rule base through expert feedback and case accumulation, improving the system's self-learning and evolutionary capabilities.

[0134] like Figure 2 As shown, another embodiment of the present invention provides a power outage processing device for a power security area, comprising a collection module 11, an extraction module 12, an evaluation module 13, and an output module 14, wherein:

[0135] The acquisition module 11 is used to time-mark the moment when the power outage occurs according to the high-precision time synchronization method, and collect the historical operation data of each node in the power system at the same time point;

[0136] An extraction module 12 is configured to perform feature extraction and representation learning on the historical operation data using a deep learning algorithm based on a convolutional neural network, automatically learning local and global patterns in the historical operation data through convolutional layers and pooling layers, and extracting deep feature vectors that reflect the intrinsic structure and time scale characteristics of the data;

[0137] An evaluation module 13 is configured to establish a power outage risk assessment model based on the deep feature vector and optimize the power outage risk assessment model using a Bayesian network or a Markov decision process method;

[0138] The output module 14 is used to obtain the current operating data of the power system, input the current operating data into the optimized power outage risk assessment model, output the assessment results, and construct a power outage processing strategy executed by the safety warning terminal that matches the assessment results.

[0139] like Figure 3 As shown, another embodiment of the present invention provides a power outage processing device for a power security area, including a processor 21, a memory 22, and a computer program stored in the memory 22 and configured to be executed by the processor 21. When the processor 21 executes the computer program, the steps in the power outage processing method embodiment of the power security area are implemented, for example Figure 1 or, when the processor 21 executes the computer program, the functions of the modules in the above-mentioned device embodiments are realized, such as the acquisition module 11, the extraction module 12, the evaluation module 13, and the output module 14.

[0140] Exemplarily, the computer program may be divided into one or more modules, which are stored in the memory 22 and executed by the processor 21 to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of the computer program in the power outage processing equipment in the power security area.

[0141] The power outage processing equipment in the power supply area may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art will appreciate that the schematic diagram is merely an example of the power outage processing equipment in the power supply area and does not limit the power outage processing equipment in the power supply area. The power outage processing equipment may include more or fewer components than shown in the diagram, or a combination of certain components, or different components. For example, the power outage processing equipment in the power supply area may also include input and output devices, network access devices, buses, and the like.

[0142] The processor 21 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor 21 is the control center for power outage processing in the power supply area, and utilizes various interfaces and lines to connect various parts of the power outage processing equipment in the entire power supply area.

[0143] The memory 22 can be used to store the computer programs and / or modules. The processor 21 implements various functions for handling power outages in the power supply area by running or executing the computer programs and / or modules stored in the memory 22 and accessing the data stored in the memory 22. The memory 22 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as sound playback or image playback); the data storage area may store data generated based on the use of the mobile phone (such as audio data and a phone book). Furthermore, the memory 22 may include high-speed random access memory (RAM) and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0144] If the integrated power outage handling module for the power supply area is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can also implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, computer memory, read-only memory (ROM), random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium.

[0145] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0146] Accordingly, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the steps of the power outage processing method in the power protection area of ​​the above embodiment, for example Figure 1 Steps S1 to S4 described in .

[0147] The beneficial effects of the embodiments of the present invention are at least one of the following:

[0148] (1) By installing a high-performance data acquisition system, the status of the present invention is monitored in real time, and operation data such as voltage, current, frequency and other key indicators are automatically collected, and abnormal patterns are identified, and the system operation status and potential risks are evaluated in real time, and the algorithm strategy is optimized to reduce the impact on system performance.

[0149] (2) The data patterns identified by the algorithm are compared with the patterns in the historical experience database, and the time scale parameters are automatically adjusted or the operation and maintenance personnel are prompted to perform manual inspections, thereby improving the accuracy of risk assessment.

[0150] (3) By developing a power outage risk function model, we can quantitatively analyze the power outage risk at different time scales, ensure that the model can be dynamically adjusted over time, and achieve continuous monitoring of the power system.

[0151] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for handling power outages in a power security area, characterized in that: The following steps are involved: Performing time stamping at the moment of power outage according to a high-precision time synchronization method, and collecting historical operation data of each node of the power system at the same time point; wherein, performing time stamping at the moment of power outage according to a high-precision time synchronization method includes: combining time synchronization technology with power outage detection, monitoring the grid voltage waveform in real time through high-speed voltage sampling and edge detection algorithms, and recording the precise moment when the power outage occurs; Based on a deep learning algorithm of a convolutional neural network, feature extraction and representation learning are performed on the historical operation data, local patterns and global patterns in the historical operation data are automatically learned through convolutional layers and pooling layers, and deep feature vectors reflecting the intrinsic structure and time scale characteristics of the data are extracted; after extracting the deep feature vectors reflecting the intrinsic structure and time scale characteristics of the data, the method further includes: Inputting the processed deep feature vector into a multi-layer perceptron (MLP) network for training to obtain an anomaly recognition model; Establish a historical experience database based on the graph database to obtain various abnormal pattern types and corresponding disposal plans that occur during the operation of the storage system; the abnormal pattern types include data mutation, data stagnation, data loss, abnormal data fluctuation and abnormal data trend; The current operation data is obtained and input into the abnormality recognition model. After a new abnormal pattern is identified, the new abnormal pattern is matched based on the graph matching algorithm. If the match fails, the dynamic misjudgment risk assessment mechanism is triggered; otherwise, the corresponding disposal plan is directly called; the matching process of the new abnormal pattern based on the graph matching algorithm includes: Retrieve similar or related abnormal patterns in the historical experience database through a graph matching algorithm, calculate the similarity score between the new abnormal pattern and each abnormal pattern type, if the highest score is lower than a first preset threshold, the match fails, and the dynamic misjudgment risk assessment mechanism is triggered; otherwise, directly call the corresponding disposal plan; the dynamic misjudgment risk assessment mechanism includes: calculating the confidence and risk level of the new abnormal pattern, if the confidence is lower than a second preset threshold and the risk level is higher than the preset threshold, it is a misjudgment risk, and triggers an adaptive time scale adjustment mechanism; the adaptive time scale adjustment mechanism includes: by dynamically adjusting the data sampling frequency and the time window size, determine whether the new abnormal pattern is a true abnormality, if it is a true abnormality, design a disposal plan corresponding to the new abnormal pattern; Establishing a power outage risk assessment model, and optimizing the power outage risk assessment model using a Bayesian network or a Markov decision process method; wherein establishing a power outage risk assessment model, and optimizing the power outage risk assessment model using a Bayesian network or a Markov decision process method, comprises: Through high-speed data acquisition equipment and intelligent sensors, real-time collection of time series data during power grid operation; Processing the time series data based on signal processing technology to extract power outage risk assessment indicators; Processing the power outage risk assessment indicators using data mining and machine learning algorithms to build a power outage risk assessment model; According to the power outage risk assessment model, a Bayesian network inference algorithm is used to quantify and calculate the probability parameters of each risk factor output by the power outage risk assessment model, and the probability parameters are dynamically updated through maximum likelihood estimation and expectation maximization algorithms to achieve real-time reasoning and prediction of power outage risks; The current operating data of the power system is obtained, the current operating data is input into an optimized power outage risk assessment model, an assessment result is output, and a power outage processing strategy executed by a safety warning terminal that matches the assessment result is constructed; after outputting the assessment result, it also includes: for the assessment results of different time scales, the time window and granularity of the power outage processing are flexibly adjusted to realize multi-time scale risk analysis.

2. The method for handling power outages in a power supply area according to claim 1, characterized in that: After collecting the historical operation data of each node in the power system at the same time point, it also includes: Acquire a synchronization timestamp according to the high-precision time synchronization method, perform time calibration and alignment on the collected power outage event data, calculate the start and end time difference of the power outage event, obtain the accurate power outage duration, and perform aggregate statistics according to different time scales to form a multi-granularity power outage duration distribution map; By using geographic information system technology, the power outage event data is spatially associated with the power grid topology structure. The impact range of the power outage event is located through buffer analysis technology and spatial connection algorithm, and the number of affected users and power load size are calculated to form a visual heat map of the power outage impact range.

3. The method for handling power outages in a power supply area according to claim 1, characterized in that: The step of constructing a power outage processing strategy to be executed by the security warning terminal and matching the evaluation result includes: The power outage risk assessment result output by the power outage risk assessment model is processed using an equidistance division method to obtain a corresponding risk level; A power outage handling strategy executed by the safety warning terminal is constructed based on the risk level.

4. The method for handling power outages in a power supply area according to claim 3, characterized in that: The security warning terminal includes at least one of the following terminals: Terminal for the operation and maintenance team, terminal for maintenance personnel, or terminal for the person responsible for power supply.

5. A power outage processing device in a power security area, characterized in that: include: The acquisition module is used to perform time marking at the moment of power outage according to the high-precision time synchronization method, and collect the historical operation data of each node of the power system at the same time point; wherein, the time marking at the moment of power outage according to the high-precision time synchronization method includes: combining time synchronization technology with power outage detection, monitoring the grid voltage waveform in real time through high-speed voltage sampling and edge detection algorithm, and recording the precise moment when the power outage occurs; An extraction module is used to perform feature extraction and representation learning on the historical operation data based on a deep learning algorithm of a convolutional neural network, automatically learn local patterns and global patterns in the historical operation data through a convolution layer and a pooling layer, and extract a deep feature vector reflecting the intrinsic structure and time scale characteristics of the data; after extracting the deep feature vector reflecting the intrinsic structure and time scale characteristics of the data, it also includes: Inputting the processed deep feature vector into a multi-layer perceptron (MLP) network for training to obtain an anomaly recognition model; Establish a historical experience database based on the graph database to obtain various abnormal pattern types and corresponding disposal plans that occur during the operation of the storage system; the abnormal pattern types include data mutation, data stagnation, data loss, abnormal data fluctuation and abnormal data trend; The current operation data is obtained and input into the abnormality recognition model. After a new abnormal pattern is identified, the new abnormal pattern is matched based on the graph matching algorithm. If the match fails, the dynamic misjudgment risk assessment mechanism is triggered; otherwise, the corresponding disposal plan is directly called; the matching process of the new abnormal pattern based on the graph matching algorithm includes: Retrieve similar or related abnormal patterns in the historical experience database through a graph matching algorithm, calculate the similarity score between the new abnormal pattern and each abnormal pattern type, if the highest score is lower than a first preset threshold, the match fails, and the dynamic misjudgment risk assessment mechanism is triggered; otherwise, directly call the corresponding disposal plan; the dynamic misjudgment risk assessment mechanism includes: calculating the confidence and risk level of the new abnormal pattern, if the confidence is lower than a second preset threshold and the risk level is higher than the preset threshold, it is a misjudgment risk, and triggers an adaptive time scale adjustment mechanism; the adaptive time scale adjustment mechanism includes: by dynamically adjusting the data sampling frequency and the time window size, determine whether the new abnormal pattern is a true abnormality, if it is a true abnormality, design a disposal plan corresponding to the new abnormal pattern; An evaluation module is used to establish a power outage risk assessment model, and optimize the power outage risk assessment model using a Bayesian network or a Markov decision process method; wherein the establishment of the power outage risk assessment model, and the optimization of the power outage risk assessment model using a Bayesian network or a Markov decision process method, includes: Through high-speed data acquisition equipment and intelligent sensors, real-time collection of time series data during power grid operation; Processing the time series data based on signal processing technology to extract power outage risk assessment indicators; Processing the power outage risk assessment indicators using data mining and machine learning algorithms to build a power outage risk assessment model; According to the power outage risk assessment model, a Bayesian network inference algorithm is used to quantify and calculate the probability parameters of each risk factor output by the power outage risk assessment model, and the probability parameters are dynamically updated through maximum likelihood estimation and expectation maximization algorithms to achieve real-time reasoning and prediction of power outage risks; An output module is used to obtain the current operating data of the power system, input the current operating data into an optimized power outage risk assessment model, output the assessment results, and construct a power outage processing strategy executed by a safety warning terminal that matches the assessment results; after outputting the assessment results, it also includes: for the assessment results of different time scales, flexibly adjusting the time window and granularity of power outage processing to achieve multi-time scale risk analysis.

6. A power outage processing device in a power security area, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for handling power outages in the power security area as described in any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the power outage processing method for the power security area as described in any one of claims 1 to 4 is implemented.

Images