Intelligent power grid operation and maintenance system based on federated learning and edge calculation

Abstract

The invention discloses an intelligent power grid operation and maintenance system based on federated learning and edge calculation, and the system comprises an intelligent electric meter enhancement module which is disposed at a power grid monitoring node and is used for collecting electric energy quality parameters and environment data in real time; the edge calculation layer is used for carrying out data preprocessing, power quality parameter anomaly detection and building a federated learning model; the secure communication framework is used for establishing a hybrid communication network and multi-level security protection, carrying out routing decision and carrying out encrypted transmission on interactive data among the modules; and the end analysis platform is used for integrating the multi-source heterogeneous data of the power grid, predicting the state of the power grid and generating a maintenance plan. According to the invention, comprehensive monitoring and predictive maintenance of the operation state of the power grid are realized.

Description

Technical Field

[0001] The present invention relates to a smart grid operation and maintenance system based on federated learning and edge computing, and belongs to the technical field of smart grid operation and maintenance. Background Art

[0002] A smart grid is a power system that combines traditional power grids with modern sensing, communication, and computing technologies. It enables intelligent grid management through data collection, analysis, and control. As smart grid construction progresses, power quality monitoring and predictive maintenance have become key technologies for ensuring stable grid operation.

[0003] Power quality is a key indicator of the quality of power supply in an electric power system, primarily encompassing parameters such as voltage sags / swells, harmonic distortion, and phase imbalance. Traditional power quality monitoring relies primarily on fixed-point sampling and manual inspections, assessing the grid's status by measuring a single parameter. Predictive maintenance, on the other hand, analyzes equipment operating data to predict potential failures and proactively conduct maintenance, effectively reducing maintenance costs and power outage losses.

[0004] Edge computing is a distributed computing architecture that deploys data processing capabilities at the edge of the network, close to the data source, reducing data transmission latency and bandwidth usage. Federated learning, a distributed machine learning method, allows multiple participants to jointly train models while protecting data privacy. It is well-suited for intelligent applications in distributed systems like power grids.

[0005] The power quality monitoring systems currently on the market have the following technical limitations:

[0006] First, most monitoring systems focus only on single parameters such as temperature and voltage. For example, the SEL-735 power quality monitor from the US-based SEL company and the MAVOWATT series from Germany's GMC-I company can monitor basic electrical parameters in real time, but they lack the ability to comprehensively analyze multi-dimensional power quality indicators such as harmonics, voltage swells and dips, and phase imbalance.

[0007] Secondly, existing systems generally use fixed communication methods for data acquisition and transmission. For example, Schneider's PowerLogic series products use Modbus communication, and ABB's network analyzers use proprietary communication protocols. This single communication mode is prone to communication bottlenecks in complex power grid environments, affecting the real-time and reliability of data transmission.

[0008] Third, traditional systems mainly rely on simple threshold judgments for data analysis and processing. For example, Siemens SICAMQ100 can only judge power quality anomalies based on preset thresholds. It lacks advanced artificial intelligence analysis methods such as deep learning and federated learning, making it difficult to extract valuable information from massive data, and its predictive maintenance capabilities are insufficient.

[0009] Furthermore, with increasing demands for data security and user privacy, existing systems have relatively weak security measures during data collection, transmission, and analysis. Most systems rely on traditional encryption methods, such as Philips' power quality analyzers, which use basic SSL encryption. However, facing growing cybersecurity threats, particularly those posed by quantum computing, existing security mechanisms may not provide adequate protection.

[0010] Overall, existing power quality monitoring systems lack comprehensive monitoring capabilities, reliable communication solutions, intelligent data analysis, and strict security protection, making it difficult to meet the needs of modern smart grids for high-reliability and high-efficiency operation and maintenance. Summary of the Invention

[0011] The purpose of this invention is to provide a smart grid operation and maintenance system based on federated learning and edge computing. The system consists of four core modules: a smart meter enhancement module, an edge computing layer, a secure communication framework, and a cloud analysis platform. Through layered design and collaborative work, each module can achieve comprehensive monitoring and predictive maintenance of the grid operation status.

[0012] In order to achieve the above object, the present invention adopts the following technical solutions:

[0013] The present invention provides a smart grid operation and maintenance system based on federated learning and edge computing, comprising:

[0014] Smart meter enhancement module, deployed at grid monitoring nodes, for real-time collection of power quality parameters and environmental data;

[0015] The edge computing layer is used to pre-process the data collected by the smart meter enhancement module and perform power quality anomaly detection;

[0016] A secure communication framework for establishing hybrid communication networks and multi-layered security, making routing decisions, and encrypting data transmission;

[0017] A cloud-based analysis platform that integrates multi-source heterogeneous data from power grids to predict grid status and generate maintenance plans.

[0018] Preferably, the smart meter enhancement module includes:

[0019] A non-intrusive load monitoring unit is used to identify the characteristics and power consumption patterns of electrical devices in the power grid; the characteristics of electrical devices refer to the electrical characteristics and operating patterns of various types of electrical devices in the power grid;

[0020] A data acquisition unit, configured to acquire power quality parameters and environmental data; the power quality parameters include grid voltage, current, harmonics, and phase; and the environmental data include temperature and humidity;

[0021] The local storage unit is used to store and back up the data acquired by the non-intrusive load monitoring unit and the data acquisition unit.

[0022] Preferably, the non-invasive load monitoring unit adopts a sensor array, or a load identification method based on sound characteristics, or an electromagnetic field characteristic analysis method.

[0023] Preferably, the local storage unit includes:

[0024] a memory for storing data acquired by the non-intrusive load monitoring unit and the data acquisition unit for at least 30 days;

[0025] A data compression module is used to compress the acquired data;

[0026] The data backup module is used to create redundant copies of data and adopt a hierarchical backup strategy.

[0027] Preferably, the edge computing layer includes:

[0028] A data preprocessing module is used to obtain the data collected by the smart meter enhancement module, and to remove abnormal values ​​and noise signals, repair missing data, and perform standardization;

[0029] Anomaly detection module, used to detect power quality parameter anomalies in real time based on preset rules and thresholds;

[0030] The federated learning module is used to build a power quality anomaly detection model based on federated learning and perform anomaly detection on power quality parameters.

[0031] Preferably, the anomaly detection module is specifically used to:

[0032] Grid voltage, current and power parameters are continuously sampled, filtered and stored at different time granularities;

[0033] Detect abnormal power quality parameters as follows:

[0034] The sliding window method is used for voltage detection. When the voltage variation exceeds ±10% of the nominal value and lasts for more than half a cycle, it is judged as a voltage sag / swell and the detection flag is triggered.

[0035] For harmonics, the FFT algorithm is used to calculate the 2nd to 50th harmonic content, evaluate the total harmonic distortion rate, and determine the harmonic distortion;

[0036] For each phase, the unbalance of the three-phase voltage / current is calculated based on the symmetrical component method.

[0037] Preferably, the federated learning module performs abnormality detection on power quality parameters, and the specific implementation method is as follows:

[0038] A local anomaly detection model is constructed at each power grid monitoring node based on a neural network structure; the local anomaly detection model includes an input layer, three hidden layers, and an output layer; the input layer receives an n-dimensional power quality feature vector, and the output layer outputs an anomaly probability value;

[0039] The local anomaly detection model is trained using local data to obtain local model parameters; during the training process, binary cross entropy plus a regularization term is used as a loss function; the local data refers to data collected and stored locally at each power grid monitoring node, including power quality parameters and environmental data collected by each node, and load characteristic data obtained by a non-intrusive load monitoring unit;

[0040] Use weighted average method to aggregate local model parameters, obtain global model parameters, and update the global model;

[0041] The global model parameters are distributed to the local anomaly detection model using a dynamic scheduling strategy to perform anomaly detection.

[0042] Preferably, the secure communication framework includes:

[0043] A communication routing module monitors the quality of each communication link, makes routing decisions based on the link quality, data priority, and network load, dynamically selects the optimal path, and automatically switches to a backup communication link in the event of a communication failure. Communication links refer to the communication connections between different components in the system, including those between the smart meter enhancement module and the edge computing layer, between the edge computing layer and the cloud analytics platform, and between different edge computing nodes.

[0044] A hybrid communication network, for establishing a hybrid communication network in a power grid, wherein the hybrid communication network includes any of power line carrier communication network, cellular network, mesh network, wide area network, optical fiber communication and microwave communication;

[0045] An encryption module for end-to-end encryption of data transmitted over the communication link based on a quantum-resistant algorithm;

[0046] Protocol adapter module, used to be compatible with different communication protocols.

[0047] Preferably, the encryption module is specifically used to:

[0048] A key exchange protocol based on the ring learning error problem is used to generate keys;

[0049] Use NewHope post-quantum key encapsulation mechanism to protect session keys;

[0050] And the improved ChaCha20-Poly1305 authentication encryption algorithm is used to transmit data.

[0051] Preferably, the cloud analysis platform includes:

[0052] A data integration module is used to integrate multi-source heterogeneous data from the power grid, process unstructured data, and establish a unified data model; the multi-source heterogeneous data from the power grid includes power grid operation data, environmental data, and historical equipment maintenance records;

[0053] An analysis engine, configured to predict a power grid state based on the multi-source heterogeneous data of the power grid;

[0054] The maintenance scheduling module is used to predict the probability of equipment failure and generate optimized maintenance plans.

[0055] Preferably, the analysis engine includes:

[0056] The data feature extraction layer uses a convolutional neural network to extract the time series features of the power quality waveform;

[0057] The pattern recognition layer uses a recurrent neural network to analyze the extracted power quality waveform timing characteristics to identify potential fault modes;

[0058] The prediction and analysis layer predicts the power grid status based on the attention mechanism.

[0059] Preferably, the maintenance scheduling module is specifically used to:

[0060] The equipment health index is calculated by considering the operating status, service life and environmental factors of power grid equipment, the equipment failure probability is predicted, and an optimized maintenance plan is generated using a heuristic algorithm combined with maintenance resource constraints.

[0061] Preferably, the smart grid operation and maintenance system adopts a three-layer interaction mechanism, including data acquisition layer interaction, model training interaction and maintenance decision interaction;

[0062] In the interaction with the data acquisition layer, the smart meter enhancement module establishes a real-time data channel with the edge computing layer. The edge computing layer obtains the power quality parameters and environmental data collected by the smart meter enhancement module, and performs data preprocessing and power quality anomaly detection.

[0063] During model training interaction, the federated learning module of the edge computing layer maintains periodic model updates with the cloud. Local nodes encrypt and transmit model parameters to the cloud, which then aggregates the model and securely distributes the updated global model to local nodes.

[0064] In the maintenance decision interaction, the edge computing layer is responsible for real-time power quality anomaly detection and early warning, and the cloud analysis platform is responsible for long-term grid status prediction.

[0065] Preferably, the smart grid operation and maintenance system is further configured with a visual display interface, which includes:

[0066] The data display module is used to support data browsing from global to local through multi-level display methods;

[0067] The trend analysis module provides trend display and prediction at multiple time scales, including: the changing trends of power quality parameters such as voltage, current, and harmonic content, the evolution trends of equipment health indicators, changes in load levels and distribution, potential fault risks, and the impact of environmental parameters on grid operation;

[0068] The maintenance recommendation module is used to convert trend analysis results into specific maintenance recommendations and sort them by priority.

[0069] The beneficial effects of the present invention are as follows:

[0070] (1) The present invention uses a multi-dimensional power quality analysis method. The system can monitor key indicators such as voltage sag / swell, harmonic distortion, and phase imbalance. Compared with the single parameter monitoring of the existing technology, the detection rate of abnormal grid status is improved, the monitoring blind spots are greatly reduced, and the coverage and accuracy of grid monitoring are improved.

[0071] (2) The edge computing layer of the present invention realizes local preprocessing and preliminary analysis of data. Compared with the traditional centralized architecture, the amount of data transmission is reduced, which effectively reduces the pressure on network bandwidth and reduces the system response time from seconds to milliseconds.

[0072] (3) The present invention adopts predictive maintenance technology based on deep learning and multi-scale time series analysis, which can issue early warnings 30 minutes to several hours before a fault occurs. Compared with the traditional post-fault response mode, it can detect most potential problems in advance and effectively avoid equipment damage and power outages.

[0073] (4) The federated learning module of the present invention enables each node to collaboratively train the model without sharing the original data. At the same time, the quantum-resistant encryption algorithm can resist advanced security threats including quantum computing, thereby improving security strength.

[0074] (5) The hybrid communication solution of the present invention is combined with adaptive routing technology. Even if a single communication method fails, the system can still maintain a high data transmission success rate.

[0075] (6) The cloud analysis platform of the present invention integrates data from the entire network, performs deep learning analysis and predictive maintenance, reduces unnecessary maintenance work, reduces emergency repairs caused by sudden failures, reduces comprehensive maintenance costs, and shortens the system investment payback period. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A schematic diagram of the structure of a smart grid operation and maintenance system based on federated learning and edge computing provided in Example 1 of the present invention;

[0077] Figure 2 A schematic diagram of the structure of the smart meter enhancement module provided in Example 1 of the invention;

[0078] Figure 3 Schematic diagram of the edge computing layer structure provided for embodiment 1 of the invention;

[0079] Figure 4 Schematic diagram of the federated learning process of the federated learning module provided in Example 1 of the invention;

[0080] Figure 5 Schematic diagram of the secure communication process provided by embodiment 1 of the invention;

[0081] Figure 6 Schematic diagram of the data processing process of the cloud analysis platform provided in Example 1 of the invention;

[0082] Figure 7 Schematic diagram of the four-layer architecture interaction mechanism process of the smart grid operation and maintenance system provided in Example 1 of the invention. DETAILED DESCRIPTION

[0083] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0084] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, the accompanying drawings only show structures and / or processing steps closely related to the solutions according to the present invention, while other details that are not closely related to the present invention are omitted.

[0085] It should be emphasized that the term "include / comprises" when used herein refers to the existence of features, elements, steps or components, but does not exclude the existence or addition of one or more other features, elements, steps or components.

[0086] It should also be noted that, unless otherwise specified, the term "connection" herein may refer not only to a direct connection but also to an indirect connection involving an intermediate.

[0087] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0088] It should be emphasized here that the step marks mentioned below do not limit the order of the steps, but it should be understood that the steps can be executed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be executed simultaneously.

[0089] Example 1

[0090] This embodiment 1 provides a smart grid operation and maintenance system based on federated learning and edge computing, see Figure 1 The system consists of four core modules: a smart meter enhancement module, an edge computing layer, a secure communication framework, and a cloud-based analytics platform. The system's overall architecture is highly modular, with modules connected via standardized interfaces to enable efficient transfer of data and control information.

[0091] In this embodiment, a smart meter enhancement module is deployed at a grid monitoring node to collect power quality parameters and environmental data in real time. The edge computing layer, directly connected to the smart meter enhancement module, is responsible for data preprocessing, power quality anomaly detection, and federated learning model training. A secure communication framework connects the edge computing layer and the cloud-based analysis platform, employing a hybrid communication solution and multi-layered security protection to ensure data transmission security. The cloud-based analysis platform, serving as the system's central processing unit, integrates network-wide data for deep learning analysis and predictive maintenance decision-making. Each module is connected through a layered design and secure communication channels, forming a complete data collection, analysis, and decision-making chain.

[0092] See also Figure 2 ,The smart meter enhancement module includes three main ,functional units: non-intrusive load monitoring unit, data acquisition unit and ,local storage unit.

[0093] The non-intrusive load monitoring unit utilizes a high-precision sensor array to identify electrical devices on the grid by analyzing current waveform characteristics, enabling non-intrusive monitoring of device characteristics and power usage patterns. Device characteristics refer to the electrical characteristics and operating modes of various types of electrical devices on the grid, including current waveform characteristics, power factor, startup characteristics, and harmonic characteristics. By identifying these characteristics, the non-intrusive load monitoring unit can determine the operating status of different devices and analyze the composition of the grid load.

[0094] The data acquisition unit includes a power quality parameter acquisition channel and an environmental data acquisition channel, which respectively monitor power quality parameters such as grid voltage sag / swell, harmonic distortion, phase imbalance, and environmental parameters such as temperature and humidity.

[0095] The local storage unit consists of a high-speed cache large-capacity memory, a data compression module and a data backup module. It is used to store the monitoring data of the non-intrusive load monitoring unit and the data acquisition unit. The local storage unit can store at least 30 days of historical data, adopts a hierarchical storage strategy to achieve efficient data management, and maintains data integrity when communication is interrupted.

[0096] The data compression module is responsible for compressing the large amount of collected data, using lossless compression algorithms to process key data, and using lossy compression algorithms to process general data, thereby optimizing storage space utilization efficiency.

[0097] Data backup module: responsible for creating redundant copies of data, adopting a hierarchical backup strategy, performing multiple copies of key data, and supporting regular backup and incremental backup to ensure that data can be restored in the event of a storage device failure.

[0098] See also Figure 3 ,The edge computing layer includes three main parts: data preprocessing module, anomaly detection module and federated learning ,module.

[0099] The data preprocessing module is used to preprocess all the raw data collected by the non-intrusive load monitoring unit and the data acquisition unit, including: removing outliers and noise signals through a multi-stage filtering algorithm, repairing data missing using an interpolation algorithm, and applying an adaptive normalization method to standardize different types of data.

[0100] The anomaly detection module establishes a normal operating state baseline model based on a clustering algorithm and uses a sliding window to monitor power quality anomalies in real time. It should be noted that the normal operating state baseline model is a statistical model developed through analysis of long-term normal grid operation data, reflecting the distribution range and interrelationships of various power quality parameters under normal grid operation. This model serves as a reference for determining grid operation anomalies and is used by the anomaly detection module to determine whether the current power quality parameter measurements deviate from normal conditions.

[0101] The federated learning module performs power quality anomaly detection based on the trained anomaly detection model. Specifically, the federated learning module uses a hierarchical aggregation approach to implement distributed model training. The federated learning module consists of a local training unit, a model aggregation unit, and a global update unit. The local training unit uses local data to train the power quality anomaly detection model. The model aggregation unit aggregates the anomaly detection model parameters from different grid nodes. The global update unit updates the global model and distributes the global model parameters to each node. It should be noted that local data refers to data collected and stored locally at each grid monitoring node, including power quality parameters (voltage, current, harmonics, and phase) and environmental data (temperature and humidity) collected by each node, as well as load characteristic data obtained by the non-intrusive load monitoring unit.

[0102] In this embodiment, both the federated learning module and the anomaly detection module are used to detect anomalies in power quality parameters. The anomaly detection module is primarily responsible for real-time anomaly detection based on preset rules and thresholds, such as voltage sags, swells, harmonics, and phase imbalance. The federated learning module builds and trains a more complex power quality anomaly detection model, improving anomaly detection capabilities through distributed learning.

[0103] In this embodiment, the specific implementation process of the abnormality detection module for power quality abnormality monitoring is as follows:

[0104] First, high-precision sampling technology is used to continuously sample basic parameters such as voltage, current, and power. The sampling frequency can reach 10kHz, ensuring that rapidly changing electrical characteristics can be captured. After preliminary filtering, the sampled data is stored according to different time granularities.

[0105] Three key power quality indicators are calculated and monitored in real time: voltage sag / swell detection uses a sliding window method, triggering a detection flag when the voltage variation exceeds ±10% of the nominal value and lasts for more than half a cycle; harmonic distortion analysis uses the FFT algorithm to calculate the 2nd-50th harmonic content and evaluate the total harmonic distortion rate; phase imbalance calculation calculates the imbalance of three-phase voltage / current based on the symmetrical component method.

[0106] In this embodiment, the process of the federated learning anomaly detection method is as follows: Figure 4 As shown in Figure 3, the method includes three main stages: local model training, model aggregation, and global update.

[0107] In the local model training phase, the anomaly detection model adopts a neural network structure consisting of an input layer, three hidden layers and an output layer. The input layer receives the n-dimensional power quality feature vector x = (x1, x2, ..., x n ), the number of neurons in the hidden layer is [64, 32, 16], and the output layer outputs the abnormal probability value p∈ [0,1].

[0108] During training, the anomaly detection model uses binary cross entropy plus regularization as the loss function, which is expressed as:

[0109] ,

[0110] in is the true label, is the predicted probability, is the regularization coefficient, is the model weight.

[0111] In the model aggregation stage, the weighted average method is used to aggregate local model parameters. The weight coefficient is determined by the amount of data and model performance, and is expressed as:

[0112] ,

[0113] For nodes local model weights, For nodes Weight coefficient;

[0114] Among them, the weight coefficient Defined as:

[0115] ,

[0116] in, For nodes Local data volume, is the total data volume, For nodes Validation error of the local model.

[0117] In the global update phase, the model aggregation parameters ( ) to update the global model. This embodiment uses a dynamic scheduling strategy to distribute model parameters and implements an incremental update mechanism, transmitting only changed model parameters to reduce communication overhead. Furthermore, the system evaluates model performance using accuracy, precision, recall, and F1 score.

[0118] Dynamic scheduling refers to a mechanism that adaptively determines the priority, frequency, and scope of model parameter distribution based on factors such as node computing power, communication status, and data quality. For example, nodes with sufficient computing resources and high-quality data can be prioritized for updates, while nodes with poor communication conditions may be subject to strategies such as compressed parameter transmission or reduced update frequency.

[0119] In this embodiment, the secure communication framework includes a communication routing module, a hybrid communication network, an encryption module and a protocol adaptation module.

[0120] The communication routing module monitors the quality indicators of each communication link (such as bandwidth utilization, latency, and packet loss rate), makes routing decisions based on communication link quality, data priority, and network load, dynamically selects the optimal path, and automatically switches to an alternative path in the event of a communication failure. It should be noted that communication links refer to the communication connections between different components in the system, primarily including those between the smart meter enhancement module and the edge computing layer, between the edge computing layer and the cloud-based analysis platform, and between different edge computing nodes.

[0121] The hybrid communication network combines power line carrier communication, cellular network and mesh network for data transmission in a stable environment, high bandwidth long distance transmission, and providing redundant paths and extended coverage, respectively.

[0122] The encryption module implements end-to-end encryption based on quantum-resistant algorithms, employing a key exchange protocol based on the ring learning error problem and the NewHope post-quantum key encapsulation mechanism to protect session keys. It also uses a modified ChaCha20-Poly1305 authenticated encryption algorithm for data transmission. The encryption module is responsible for transmitting data exchanged between the edge computing layer and the cloud-based analysis platform, including model parameters, anomaly detection results, power quality data, information shared between different edge computing nodes, and system control instructions and status information. The encryption module ensures the security of this data during transmission.

[0123] The protocol adaptation module is compatible with different communication protocols to achieve wide applicability of the system.

[0124] In this embodiment, the specific implementation process of the quantum-resistant encryption algorithm is as follows: lattice cryptography is used as the basis, and multiple encryption mechanisms are combined to ensure the long-term security of data transmission. In terms of key generation, a key exchange protocol based on Ring-LWE is adopted, the dimension n of the polynomial ring is selected to be 1024, and the modulus q is set to 12289 to ensure at least 128 bits of post-quantum security strength. The key generation process includes generating a random element a, each party generates a private key s and random noise e, calculates the public key b = as + e, and establishes a shared key by exchanging public keys. A hybrid encryption mechanism is implemented at the data encryption level, and the NewHope post-quantum key encapsulation mechanism is used to protect the session key. The improved ChaCha20-Poly1305 authentication encryption algorithm is used to transmit actual data. A hierarchical key architecture is also implemented in this embodiment, including a root key protected by a hardware security module, an intermediate key derived from the root key, and a session key for a specific communication session, and a forward security protocol is used to ensure that key leakage does not affect the security of historical data.

[0125] Secure communication methods such as Figure 5 As shown, specifically: monitor the quality indicators of each communication link in real time, judge the status of the communication link based on the quality indicators, and if it is normal, select the optimal communication path based on the network load; if the communication link is abnormal, trigger the backup communication link and then select the optimal communication path; encrypt the communication data before transmission.

[0126] The cloud-based analysis platform includes a data integration module, an analysis engine, and a maintenance scheduling module.

[0127] The data integration module is used to integrate multi-source heterogeneous data such as power grid operation data, environmental data, and equipment historical maintenance records. It uses semantic analysis technology to process unstructured data and establishes a unified data model to support subsequent analysis.

[0128] The analysis engine is used to predict the power grid status based on multi-source heterogeneous data integrated by the data integration module. Specifically, the analysis engine adopts a multi-layer deep learning network structure, including a data feature extraction layer that uses a convolutional neural network to extract the power quality waveform characteristics, a pattern recognition layer that analyzes the time series characteristics of the power quality waveform through a recurrent neural network to identify potential fault modes, and a predictive analysis layer that predicts the power grid status based on the attention mechanism.

[0129] The maintenance scheduling module calculates equipment health indicators by comprehensively considering the operating status, service life, and environmental factors of power grid equipment, predicts the probability of equipment failure, and uses a heuristic algorithm to generate an optimized maintenance plan based on maintenance resource constraints. The main prediction methods include: (1) building an equipment health indicator evaluation model based on historical data and current operating status; (2) utilizing the deep learning network in the analysis engine; and (3) combining equipment life models with environmental factor impact analysis.

[0130] Predictive maintenance methods based on cloud analysis platforms such as Figure 6 shown.

[0131] The interaction mechanism of the smart grid system in this embodiment is as follows: Figure 7 As shown in the figure, it mainly includes three aspects: data acquisition layer interaction, model training interaction and maintenance decision interaction.

[0132] In the interaction of the data acquisition layer, the smart meter enhancement module establishes a real-time data channel with the edge computing layer. The edge computing layer obtains the power quality data and environmental data collected by the smart meter enhancement module, and performs data preprocessing and anomaly detection. The edge computing layer dynamically adjusts the sampling strategy and preprocessing parameters.

[0133] During model training interaction, the federated learning module of the edge computing layer maintains periodic model updates with the cloud. The local node encrypts and transmits model parameters to the cloud. The cloud completes model aggregation and securely distributes the updated global model.

[0134] In the maintenance decision-making interaction, the edge computing layer is responsible for real-time power quality anomaly detection and early warning, and the cloud analysis platform is responsible for in-depth analysis and long-term grid status prediction. The two layers ensure timely response and accurate decision-making through a collaborative decision-making mechanism.

[0135] The smart grid operation and maintenance system in this embodiment utilizes a flexible architecture design, supporting horizontal system expansion and adapting to grid monitoring needs of varying scales. New monitoring nodes can be added to the system via plug-and-play, with edge computing nodes automatically identifying and configuring new devices. The system implements a dynamic load-based balancing mechanism and a distributed storage architecture to optimize resource utilization.

[0136] The smart grid operation and maintenance system in this embodiment employs a multi-layered security architecture, providing comprehensive security protection throughout the entire data collection, transmission, and storage process. Device-level security protection includes a security chip, a hardware-level random number generator, and anti-tampering mechanisms. Data transmission utilizes an end-to-end encryption scheme based on quantum-resistant cryptography and a dynamic key agreement mechanism. Access control implements fine-grained, role-based permission management and detailed operation auditing.

[0137] The smart grid operation and maintenance system in this embodiment implements multiple backups and fault self-recovery capabilities, ensuring stable operation under various abnormal circumstances. The data collection layer implements a data caching and backup mechanism, temporarily storing data locally in the event of communication interruptions. The edge computing layer adopts a master-slave redundancy design with automatic switchover. The model training process implements a fault-tolerant mechanism, enabling global model updates to continue even if a node fails.

[0138] By adopting the above-mentioned technical means, this embodiment uses distributed anomaly detection technology based on federated learning to achieve distributed model training while protecting data privacy, and adopts a weighted average method to aggregate local model parameters to improve the accuracy of anomaly detection; adopts a multi-dimensional power quality analysis method to comprehensively analyze multi-dimensional indicators such as voltage sag / swell, harmonic distortion, and phase imbalance, and combines environmental data to achieve comprehensive monitoring of the power grid operation status; adopts an adaptive hybrid communication solution to flexibly combine power line carrier communication, cellular network and mesh network to achieve dynamic routing selection and fault self-recovery based on link quality; based on deep learning predictive maintenance technology, a multi-layer deep learning network structure is used to analyze power grid operation data, realize equipment health status assessment and fault prediction, and significantly reduce maintenance costs; and a secure communication mechanism that is resistant to quantum encryption and a key exchange protocol based on the ring learning error problem achieve future-oriented communication security protection capabilities.

[0139] It should be noted that in addition to the four-tier architecture used in this embodiment, the system can also adopt a three-tier architecture, combining the smart meter enhancement module and the edge computing layer into "intelligent edge nodes" directly connected to the secure communication framework. This design reduces hardware costs and is suitable for small power grids or resource-constrained scenarios. Another alternative architecture is to adopt a "fog computing model," adding a regional control center between the edge computing layer and the cloud-based analysis platform, forming a five-tier architecture, which is more suitable for the hierarchical management of large regional power grids.

[0140] It should be noted that, in addition to the high-precision sensor array, the non-invasive load monitoring unit of this embodiment can also employ load identification technology based on sound signatures or electromagnetic field signature analysis methods. The data acquisition unit can use a distributed microsensor network instead of a centralized data acquisition method, with each microsensor focusing on monitoring specific parameters, which helps improve data acquisition accuracy and system reliability. The local storage unit can use a selective storage mechanism based on importance scoring, maintaining high-fidelity storage for important data and using compression or downsampling for common data to optimize storage resource utilization.

[0141] It should be noted that, in addition to traditional filtering and interpolation methods, the data preprocessing module of the edge computing layer of this embodiment can use an autoencoder network for data reconstruction and anomaly filtering, or use a multi-resolution analysis method based on wavelet transform to extract useful signals. The anomaly detection module can also use a topology-aware anomaly detection method based on a graph neural network to better capture the dependencies between power grid devices, or use an unsupervised anomaly detection method based on a variational autoencoder to reduce reliance on labeled data. In addition to parameter aggregation, the federated learning module can also use a knowledge distillation method to achieve model sharing, or use a decentralized federated learning architecture to avoid single-point failure problems at the central aggregation point.

[0142] It should be noted that, in terms of model structure, the federated learning module of this embodiment can use, in addition to multi-layer perceptrons, convolutional neural networks to capture local features of power quality waveforms, or long-short-term memory networks to analyze time series data. For model aggregation, a trust-based aggregation method can be used to dynamically adjust weights based on the node's historical contribution, or a secure aggregation protocol can be used to prevent privacy leaks. For global updates, a decentralized update mechanism based on the Gossip protocol can be used, or a joint optimization method can be used to simultaneously consider communication efficiency and model performance. Furthermore, an adaptive training strategy based on reinforcement learning can be used to dynamically adjust the participation and training frequency of each node.

[0143] It should be noted that, in addition to the three methods described above, the secure communication framework of this embodiment can also utilize low-power wide area network technologies (such as LoRa and NB-IoT), fiber optic communications, or microwave communications, with flexible selection based on the deployment environment and requirements. The communication routing module can employ an adaptive routing algorithm based on reinforcement learning to optimize routing decisions by continuously learning network status, or a predictive routing method to proactively adjust routing based on historical data to predict link status changes. The encryption module can also utilize other algorithms based on lattice cryptography (such as NTRU and CRYSTALS-Kyber) or homomorphic encryption techniques to perform specific computational operations while maintaining encryption.

[0144] It should be noted that the data integration module in the cloud-based analysis platform of this embodiment can adopt a semantic integration method based on knowledge graphs to better express and utilize the relationships between data, or use federated data virtualization technology to avoid physical data migration. The analysis engine can use a fault diagnosis method based on causal reasoning to improve the accuracy of fault cause analysis, or use transfer learning technology to quickly adapt to new equipment types using existing knowledge. The maintenance scheduling module can use a multi-objective optimization algorithm to simultaneously consider maintenance costs, system reliability, and resource constraints, or use a simulation-assisted decision-making method based on digital twins to evaluate the effectiveness of different maintenance strategies through a virtual environment.

[0145] It should be noted that the system operation interaction mechanism of this embodiment achieves efficient coordination between the various modules of the system through multi-level information interaction and collaborative decision-making. In terms of interaction protocols, a message queue system based on a publish-subscribe model (such as MQTT, Kafka) can be used instead of a direct communication method to improve the loose coupling and scalability of the system. In terms of decision coordination, a distributed decision-making framework based on a multi-agent system can be implemented, in which each node can make decisions independently and work together, or a decision-making mechanism based on a consensus algorithm can be adopted to ensure the consistency of the system's decisions in a distributed environment. In terms of early warning mechanisms, an adaptive early warning strategy based on situational awareness can be implemented to dynamically adjust the early warning threshold and response level according to the operating status of the power grid and the external environment.

[0146] It should be noted that this embodiment adopts an elastic architecture design to support the horizontal expansion of the system. In terms of expansion mechanism, a microservice architecture based on container technology can be adopted to improve the modularity and deployment flexibility of the system, or a serverless computing model can be used to allocate computing resources on demand. In terms of load balancing, a global optimization scheduling strategy based on genetic algorithms can be implemented, or a resource allocation method based on market mechanisms can be used to achieve efficient resource allocation through virtual pricing and bidding mechanisms. In terms of storage architecture, a distributed storage system based on blockchain can be adopted to improve data traceability and tamper-proof capabilities, or a multi-level caching strategy can be used to optimize the storage hierarchy according to data access patterns.

[0147] It should be noted that this embodiment adopts a multi-level security architecture. In terms of identity authentication, a multi-factor authentication mechanism based on biometrics can be adopted, or a zero-knowledge proof protocol based on trust proof can be used to complete identity authentication without leaking key information. In terms of data protection, data desensitization technology based on differential privacy can be implemented to protect the privacy of sensitive data during the analysis process, or trusted execution environment technology can be used to process sensitive data in a hardware-isolated security area. In terms of security management, an artificial intelligence-based security situation awareness system can be used to automatically detect and respond to potential security threats, or a blockchain-based security event tracing mechanism can be implemented to ensure the non-repudiation and traceability of security incidents.

[0148] It should be noted that this embodiment implements multiple backups and fault self-recovery capabilities. In terms of fault detection, an abnormal behavior detection algorithm based on deep learning can be used to actively identify abnormal system states, or a distributed health monitoring protocol can be used to detect faults through mutual supervision between nodes. In terms of data recovery, a distributed redundant storage mechanism based on erasure codes can be implemented to improve data recovery capabilities and storage efficiency, or a data consistency assurance mechanism based on version control can be used to ensure the consistency of the system state. In terms of system recovery, a transaction-based state rollback mechanism can be used to roll back to the most recent normal state when a fault occurs, or a fault isolation strategy based on adaptive control can be implemented to dynamically adjust the system topology to isolate faulty nodes.

[0149] It should be noted that the power quality monitoring method of this embodiment has the following alternatives:

[0150] In terms of sampling technology, in addition to high-frequency unified sampling, an adaptive sampling strategy can be adopted to dynamically adjust the sampling frequency according to the signal change rate, increase the sampling density at critical moments, reduce the sampling frequency during stable periods, and optimize storage and computing resources. In terms of voltage anomaly detection, a transient detection method based on wavelet transform can be used to improve the sensitivity to short-term voltage disturbances, or a waveform feature extraction technology based on morphological analysis can be used to more accurately capture voltage waveform distortion. In terms of harmonic analysis, short-time Fourier transform can be used instead of traditional FFT to better reflect the time-varying characteristics of harmonic components, or a sparse decomposition algorithm can be used to improve the accuracy of harmonic detection. In terms of phase imbalance analysis, a real-time monitoring method based on dynamic estimation of sequence components can be implemented, or a graphical analysis technology based on ellipse fitting can be used to intuitively present the degree of three-phase imbalance.

[0151] It should be noted that the quantum-resistant encryption algorithm of this embodiment has the following alternatives:

[0152] Regarding post-quantum algorithms, in addition to the Ring-LWE scheme, the SIKE algorithm based on supersingular elliptic curve homology can be used, offering the advantage of smaller key sizes. Alternatively, the SPHINCS+ signature scheme, based on hash functions, can be used to avoid reliance on specific mathematical problems. Regarding key management, a distributed key management system based on threshold cryptography can be implemented to prevent key leakage due to single points of failure. Alternatively, a physical encryption scheme based on quantum key distribution can be employed, leveraging the principles of quantum mechanics to ensure absolute security of key distribution. Regarding authentication mechanisms, zero-knowledge proof-based authentication protocols can be used to complete authentication without revealing identity information, or multi-factor authentication based on biometrics can be employed to improve authentication reliability. Furthermore, dynamic key agreement mechanisms can be considered, adaptively selecting the most appropriate encryption algorithm based on channel characteristics and security requirements, or the use of quantum-safe pseudo-random function generators to enhance the randomness and security of key generation.

[0153] Through the comprehensive application of the above technical solutions, this system has significantly improved the intelligence level and efficiency of power grid operation and maintenance, providing a strong guarantee for the safe and stable operation of the power grid.

[0154] Example 2

[0155] Based on the above-mentioned Example 1, this second embodiment further adds a visual display interface, including a data display module, a trend analysis module, and a maintenance recommendation module. The data display module supports global and local data browsing through a multi-level display method. The trend analysis module provides trend display and prediction functions at multiple time scales. The maintenance recommendation module converts analysis results into specific maintenance recommendations (such as equipment preventive maintenance time window recommendations, load adjustment optimization plans, energy efficiency improvement measures, etc.) and sorts them by priority.

[0156] In this embodiment, the trend analysis module displays trends including: changing trends of power quality parameters such as voltage, current, and harmonic content; evolution trends of equipment health indicators; changes in load levels and distribution; potential failure risks; and trends in the impact of environmental parameters on grid operation.

[0157] The trend analysis module implements trend display in the following way:

[0158] The trend analysis module utilizes a multi-dimensional analysis framework and implements a multi-scale trend decomposition algorithm for time series analysis. First, wavelet transforms are used to decompose raw data to extract trend components at different time scales. Raw data includes power quality parameters such as voltage, current, and harmonic content, as well as environmental data, equipment operating status data, load test data, and historical maintenance records. Long-term trends are modeled using an LSTM network with a prediction window of up to 30 days. For cyclical fluctuations, Fourier transform analysis is used to extract characteristic frequency components. The system also automatically identifies outliers and annotates possible trend breaks with confidence intervals. For data visualization, a rich set of interactive analysis tools are provided, including multi-parameter linkage analysis that supports simultaneous comparison of up to eight indicators, intelligent zooming that automatically adjusts the display scale based on importance scores, and an overlay function that directly marks predicted potential risk points on trend lines with warning information. The module also implements intelligent report generation, automatically summarizing key trend characteristics (such as growth rates and their fluctuations, the strength of cyclical patterns, and the frequency of abnormal events).

[0159] It should be noted that the visual display interface can use virtual reality or augmented reality technology to create an immersive monitoring environment, or use a large-screen splicing system to build a power grid operation situation awareness center. In terms of interaction methods, an intelligent query system based on natural language processing can be implemented to support operations and maintenance personnel to obtain the required information through spoken descriptions, or a multimodal interaction method based on gestures and voice can be used to improve operational efficiency. In terms of decision support, personalized information push based on recommendation systems can be implemented to provide customized content based on different user roles and concerns, or scenario simulation functions can be used to help decision makers assess the potential impact of different maintenance decisions.

[0160] It should be noted that the trend analysis module has the following alternatives:

[0161] For time series analysis, in addition to LSTM networks, the Transformer architecture can be used for sequence modeling to better capture long-range dependencies. Alternatively, specialized time series forecasting frameworks such as Prophet can be used to accommodate seasonal and holiday effects. For data decomposition, adaptive signal processing methods based on empirical mode decomposition can be implemented without pre-defined decomposition basis functions. Alternatively, singular spectrum analysis can be used to more effectively separate trends, cycles, and noise components. For visualization, high-dimensional data visualization methods based on dimensionality reduction (such as t-SNE and UMAP) can be used to visualize multidimensional feature relationships on a two-dimensional plane. Alternatively, multidimensional data exploration tools based on parallel coordinates can be implemented to intuitively display correlations between multiple variables. For anomaly detection, probabilistic anomaly detection frameworks based on deep generative models can be used to more accurately estimate data distributions and anomaly probabilities. Alternatively, ensemble learning methods can be used to combine the strengths of multiple anomaly detection algorithms. Furthermore, for trend forecasting, multi-scenario forecasting capabilities can be considered to generate multiple possible development paths based on different assumptions, helping decision makers develop robust operational strategies. Causal inference methods can be used to analyze the key factors influencing trend changes and provide more interpretable forecast results.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A smart grid operation and maintenance system based on federated learning and edge computing, characterized by: include: Smart meter enhancement module, deployed at grid monitoring nodes, for real-time collection of power quality parameters and environmental data; The edge computing layer is used to pre-process the data collected by the smart meter enhancement module and perform power quality anomaly detection; A secure communication framework for establishing hybrid communication networks and multi-layered security, making routing decisions, and encrypting data transmission; A cloud-based analysis platform that integrates multi-source heterogeneous data from power grids to predict grid status and generate maintenance plans.

2. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 1, characterized in that: The smart meter enhancement module includes: A non-intrusive load monitoring unit is used to identify the characteristics and power consumption patterns of electrical devices in the power grid; the characteristics of electrical devices refer to the electrical characteristics and operating patterns of various types of electrical devices in the power grid; A data acquisition unit, configured to acquire power quality parameters and environmental data; the power quality parameters include grid voltage, current, harmonics, and phase; and the environmental data include temperature and humidity; The local storage unit is used to store and back up the data acquired by the non-intrusive load monitoring unit and the data acquisition unit.

3. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 2, characterized in that: The non-intrusive load monitoring unit adopts a sensor array, or a load identification method based on sound characteristics, or an electromagnetic field characteristic analysis method.

4. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 2, characterized in that: The local storage unit includes: a memory for storing data acquired by the non-intrusive load monitoring unit and the data acquisition unit for at least 30 days; A data compression module is used to compress the acquired data; The data backup module is used to create redundant copies of data and adopt a hierarchical backup strategy.

5. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 2, characterized in that: The edge computing layer includes: A data preprocessing module is used to obtain the data collected by the smart meter enhancement module, and to remove abnormal values ​​and noise signals, repair missing data, and perform standardization; Anomaly detection module, used to detect power quality parameter anomalies in real time based on preset rules and thresholds; The federated learning module is used to build a power quality anomaly detection model based on federated learning and perform anomaly detection on power quality parameters.

6. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 5, characterized in that: The anomaly detection module is specifically used to: Grid voltage, current and power parameters are continuously sampled, filtered and stored at different time granularities; Detect abnormal power quality parameters as follows: The sliding window method is used for voltage detection. When the voltage variation exceeds ±10% of the nominal value and lasts for more than half a cycle, it is judged as a voltage sag / swell and the detection flag is triggered. For harmonics, the FFT algorithm is used to calculate the 2nd to 50th harmonic content, evaluate the total harmonic distortion rate, and determine the harmonic distortion; For each phase, the unbalance of the three-phase voltage / current is calculated based on the symmetrical component method.

7. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 5, characterized in that: The federated learning module performs anomaly detection on power quality parameters, and the specific implementation method is as follows: A local anomaly detection model is constructed at each power grid monitoring node based on a neural network structure; the local anomaly detection model includes an input layer, three hidden layers, and an output layer; the input layer receives an n-dimensional power quality feature vector, and the output layer outputs an anomaly probability value; The local anomaly detection model is trained using local data to obtain local model parameters; during the training process, binary cross entropy plus a regularization term is used as a loss function; the local data refers to data collected and stored locally at each power grid monitoring node, including power quality parameters and environmental data collected by each node, and load characteristic data obtained by a non-intrusive load monitoring unit; Use weighted average method to aggregate local model parameters, obtain global model parameters, and update the global model; The global model parameters are distributed to the local anomaly detection model using a dynamic scheduling strategy to perform anomaly detection.

8. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 5, characterized in that: The secure communication framework includes: A communication routing module monitors the quality of each communication link, makes routing decisions based on the link quality, data priority, and network load, dynamically selects the optimal path, and automatically switches to a backup communication link in the event of a communication failure. Communication links refer to the communication connections between different components in the system, including those between the smart meter enhancement module and the edge computing layer, between the edge computing layer and the cloud analytics platform, and between different edge computing nodes. A hybrid communication network, for establishing a hybrid communication network in a power grid, wherein the hybrid communication network includes any of power line carrier communication network, cellular network, mesh network, wide area network, optical fiber communication and microwave communication; An encryption module for end-to-end encryption of data transmitted over the communication link based on a quantum-resistant algorithm; Protocol adapter module, used to be compatible with different communication protocols.

9. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 8, characterized in that: The encryption module is specifically used to: A key exchange protocol based on the ring learning error problem is used to generate keys; Use NewHope post-quantum key encapsulation mechanism to protect session keys; And the improved ChaCha20-Poly1305 authentication encryption algorithm is used to transmit data.

10. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 8, characterized in that: The cloud analysis platform includes: A data integration module is used to integrate multi-source heterogeneous data from the power grid, process unstructured data, and establish a unified data model; the multi-source heterogeneous data from the power grid includes power grid operation data, environmental data, and historical equipment maintenance records; An analysis engine, configured to predict a power grid state based on the multi-source heterogeneous data of the power grid; The maintenance scheduling module is used to predict the probability of equipment failure and generate optimized maintenance plans.

11. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 10, characterized in that: The analysis engine includes: The data feature extraction layer uses a convolutional neural network to extract the time series features of the power quality waveform; The pattern recognition layer uses a recurrent neural network to analyze the extracted power quality waveform timing characteristics to identify potential fault modes; The prediction and analysis layer predicts the power grid status based on the attention mechanism.

12. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 10, characterized in that: The maintenance scheduling module is specifically used to: The equipment health index is calculated by considering the operating status, service life and environmental factors of power grid equipment, the equipment failure probability is predicted, and an optimized maintenance plan is generated using a heuristic algorithm combined with maintenance resource constraints.

13. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 10, characterized in that: The smart grid operation and maintenance system adopts a three-layer interaction mechanism, including data collection layer interaction, model training interaction and maintenance decision interaction; In the interaction with the data acquisition layer, the smart meter enhancement module establishes a real-time data channel with the edge computing layer. The edge computing layer obtains the power quality parameters and environmental data collected by the smart meter enhancement module, and performs data preprocessing and power quality anomaly detection. During model training interaction, the federated learning module of the edge computing layer maintains periodic model updates with the cloud. Local nodes encrypt and transmit model parameters to the cloud, which then aggregates the model and securely distributes the updated global model to local nodes. In the maintenance decision interaction, the edge computing layer is responsible for real-time power quality anomaly detection and early warning, and the cloud analysis platform is responsible for long-term grid status prediction.

14. The smart grid operation and maintenance system based on federated learning and edge computing according to claim 10, characterized in that: The smart grid operation and maintenance system is further configured with a visual display interface, which includes: The data display module is used to support data browsing from global to local through multi-level display methods; The trend analysis module provides trend display and prediction at multiple time scales, including: the changing trends of power quality parameters such as voltage, current, and harmonic content, the evolution trends of equipment health indicators, changes in load levels and distribution, potential fault risks, and the impact of environmental parameters on grid operation; The maintenance recommendation module is used to convert trend analysis results into specific maintenance recommendations and sort them by priority.

Images