Digital twin operation monitoring system of power equipment

Abstract

The invention relates to the technical field of power equipment, and discloses a digital twin operation monitoring system for power equipment, which comprises a data sensing and acquisition system for acquiring key operation parameters of temperature, current, voltage, partial discharge, vibration and humidity of the power equipment in real time, and performing multi-dimensional data acquisition through a sensor and a data transmission module; the state evaluation and prediction system is used for performing equipment health evaluation and residual life prediction by using a prediction model LSTM based on the collected data, and updating a prediction result in real time; provided is a digital twin modeling system. Through the combination of edge calculation, an LSTM model and a digital twinning technology, the precision and real-time performance of health management of power equipment are improved, data quality is optimized through edge calculation, the LSTM model captures an equipment degradation trend, virtual-real fusion is realized through digital twinning, and accurate monitoring and early warning of the health state of the equipment are ensured, so that intelligent operation and maintenance decisions are optimized, the failure rate is reduced, and the safety of power equipment health management is improved. The equipment life is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment, and in particular to a digital twin operation monitoring system for power equipment. Background Art

[0002] With the increasing complexity and intelligence of power systems, the reliability and lifespan prediction of key equipment (such as transformers, circuit breakers, cables, and switchgear) have become core issues in power operations and maintenance management. Traditional operations and maintenance models rely primarily on regular inspections and empirical judgment. These issues include long detection cycles, delayed status perception, and untimely maintenance responses. This makes it difficult to promptly identify potential equipment failures, which can easily lead to sudden power outages, equipment damage, and even safety accidents. To address this, data-driven condition monitoring and health assessment methods have been gradually introduced in recent years. These methods use sensors to collect parameters such as current, voltage, temperature, humidity, vibration, and partial discharge, and combine them with specific models to perform lifespan assessments.

[0003] However, most existing life prediction methods rely on simple trend analysis or shallow neural networks, lacking the ability to deeply model the long-term time-series dependency characteristics of equipment operating status. In addition, the operating environment of power equipment is complex, and the data collected by sensors is often accompanied by noise, outliers or missing data. If it is directly input into the prediction model without effective processing, it will seriously affect the model training quality and actual prediction effect, resulting in large fluctuations and low accuracy in life prediction results. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a digital twin operation monitoring system for power equipment, which solves the problem that the existing prediction methods are difficult to deeply model long-term time series and have weak ability to process abnormal data, resulting in low prediction accuracy.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A digital twin operation monitoring system for power equipment, comprising: Data sensing and acquisition system: used to collect multi-dimensional operating data through current, voltage, temperature, humidity, vibration and partial discharge sensors deployed at key parts of the equipment; Condition assessment and prediction system: Based on the collected data, the LSTM model is used to assess the health status of the equipment and predict the remaining lifespan; Digital twin modeling system: Based on the collected data and equipment characteristics, CAD / CAE tools are used to build equipment geometry and physical models and synchronize them with real-time data; Linked response and control system: When equipment anomalies are detected, protective measures are automatically triggered to start backup power or adjust the load; Intelligent decision-making and optimization system: Generates optimization plans based on real-time data and forecast results, and automatically adjusts equipment operation; Visual display and platform system: Presents equipment operation, health assessment and warning information through WebGL / ECharts.

[0006] By adopting the above technical solution: the present invention solves the problems of delayed state perception and untimely fault warning in the operation and maintenance of traditional power equipment, uses multiple sensors to collect equipment operation data in real time, and combines it with LSTM-based deep time series modeling to effectively capture the dependency characteristics of the long-term operation status of the equipment, thereby improving the accuracy of health status assessment and remaining life prediction. At the same time, the system realizes virtual-real synchronization through digital twin technology, enhances the intuitive visualization and dynamic monitoring capabilities of equipment status, effectively filters abnormal noise data, improves model stability, and ensures stable and reliable prediction results, thereby significantly improving the intelligence level and response speed of equipment operation and maintenance, reducing the risk of sudden failures, and improving the safety and stability of the power system.

[0007] Preferably, the data perception and acquisition system includes a multi-source sensor subsystem, an edge pre-processing subsystem, a data transmission subsystem, and a time synchronization and labeling subsystem; The multi-source sensor subsystem collects information on current, voltage, temperature, humidity, vibration, and partial discharge of the device through sensors; The edge preprocessing subsystem uses 3σ anomaly elimination, KNN interpolation and completion, and Z-Score normalization; the data transmission subsystem uses 4G / 5G transmission and redundant storage; The time synchronization and tagging subsystem uses GPS / NTP to timestamp the data and add device IDs.

[0008] Preferably, the state assessment and prediction system includes a data processing subsystem for further statistical preprocessing of the collected operating data, including using the 3σ / Z-Score method to eliminate outliers, using the linear / KNN method to interpolate and fill missing values and Z-Score standardization, and generating time series samples according to a fixed time window; a prediction model subsystem for building an equipment health state prediction model based on a multi-layer LSTM unit and a linear output layer, outputting a health index HI(t); a degradation analysis subsystem for real-time calculation of the degradation rate ΔHI / Δt=[HI(t)–HI(t–1)] / Δt, where Δt is the interval between two predictions, ΔHI / Δt reflects the rate of change of the health index, and the distribution of the past N ΔHI / Δt. The kernel density estimation (KDE) method is used to fit the probability density. The fault risk level is determined based on the percentile of the current ΔHI / Δt in the historical distribution. The model update and optimization subsystem, after obtaining the real-time health status or actual life label, uses the corresponding input samples and labels as incremental data. The LSTM model weights are fine-tuned using the online gradient descent method and combined with old samples to stabilize training. The equipment comprehensive health index (CHI) assessment model combines electrical parameters (current I, voltage V), environmental parameters (temperature T, humidity H), and historical degradation rate (the average of the past M ΔHI / Δt values) into a feature vector. The CHI is output through normalization and weighted calculation, and is classified into normal, sub-healthy, and fault levels according to the set threshold, triggering early warning or maintenance strategies.

[0009] Preferably, the digital twin modeling system includes a virtual modeling subsystem, a data synchronization and fusion subsystem, a simulation and simulation analysis subsystem, and a real-time visualization subsystem. The virtual modeling subsystem uses SolidWorks / Ansys TwinBuilder to import equipment CAD and physical properties to generate modeling. The data synchronization and fusion subsystem fuses sensing and prediction data through Kalman filtering. The simulation and simulation analysis subsystem predicts fault evolution through the OpenModelica interface. The real-time visualization subsystem renders a three-dimensional twin based on Three.js / WebGL and superimposes a health color code.

[0010] Preferably, the linkage response and control system includes an anomaly detection subsystem, a linkage control subsystem, an execution feedback subsystem, a security policy subsystem and an edge computing module. The anomaly detection subsystem uses the IsolationForest algorithm to perform real-time classification and identification on the multi-dimensional time series features pre-processed by the data processing subsystem and output an anomaly label. The linkage control subsystem triggers circuit breaker tripping, backup power switching, and load limiting based on the Drools rule engine. The execution feedback subsystem collects PLC / SCADA receipt verification execution results; the security policy subsystem verifies instruction compliance using a security boundary model; The edge computing module deploys the TinyML model and rule engine to achieve local real-time diagnosis and control.

[0011] Preferably, the intelligent decision-making and optimization system includes a decision analysis subsystem, an optimization scheduling subsystem, a strategy generation subsystem and an auxiliary operation and maintenance subsystem; The decision analysis subsystem uses PCA dimension reduction + Bayesian network to fuse multi-source data; The optimization scheduling subsystem balances efficiency, energy consumption and risk based on the NSGA-II algorithm; The strategy generation subsystem converts the scheduling results into a frequency converter speed regulation and maintenance plan; The auxiliary operation and maintenance subsystem generates a decision report through NLG and pushes it to the operation and maintenance terminal.

[0012] Preferably, the visualization display and platform system includes a graphical display subsystem, an interactive monitoring subsystem, an alarm push subsystem and a multi-terminal access subsystem. The graphical display subsystem draws real-time trends and RUL curves based on ECharts, the interactive monitoring subsystem supports timeline and multi-dimensional filtering, the alarm push subsystem sends warnings to the Web and mobile terminals through WebSocket / MQTT, and the multi-terminal access subsystem adopts ReactNative and Web responsive design to achieve unified access.

[0013] Preferably, the health status and life prediction model is constructed based on the long short-term memory neural network LSTM algorithm, and the key parameters of the power equipment during operation are sampled at high frequency through the physical perception module. The collected data include multi-dimensional operating parameters such as current, voltage, temperature, humidity, vibration, partial discharge signal, and insulation resistance, forming a high-dimensional data set with time series characteristics; The edge computing module preprocesses the raw data, including outlier removal, missing value filling, and Z-score normalization, to ensure that the input data is complete, valid, and has a unified dimension, reducing the impact of abnormal interference on model training and prediction accuracy. The preprocessed data is input into the LSTM model in the form of a tensor. The structure of the input tensor is X∈ℝ^(B×T×N), where B is the number of samples input into the model at one time, T is the length of the predicted historical time, and N is the feature dimension, including the following features: Current I, current change rate ΔI / Δt, voltage V, voltage change rate ΔV / Δt, temperature T, temperature change rate ΔT / Δt, humidity H, humidity change rate ΔH / Δt, vibration A, vibration change rate ΔA / Δt; Among them, Δ represents the difference of the variable between adjacent time steps, Δx / Δt=x(t)-x(t−1), which is used to characterize the instantaneous change trend of the feature; The LSTM model models the multi-dimensional time series data in the intelligent prediction and diagnosis module, and uses the mean square error as the loss function for gradient optimization to learn the time series evolution law of equipment degradation. The model output structure is Y∈ℝ^(B×1), which represents the remaining life prediction value of each sample. The prediction value can be given in time units. The subsequent collected data will also continue to be input into the trained LSTM model in the form of tensors to achieve real-time prediction of the current remaining life of the equipment, and the prediction results will be synchronously transmitted to the digital twin modeling module to drive the update and visualization of the virtual equipment status, thereby realizing dynamic monitoring of the operating status of the power equipment and displaying the degradation trend.

[0014] Preferably, the equipment comprehensive health index assessment model combines the equipment operating parameters current, voltage, temperature rise, partial discharge intensity, vibration and environmental parameters temperature and humidity collected by the physical perception module into a multidimensional feature vector X∈ℝⁿ, and constructs input samples through normalization preprocessing; The feature vector X is then weighted and fused based on the feature item weight vector W∈ℝⁿ set by the expert knowledge rule base to calculate the comprehensive equipment health index (CHI=W·Xᵀ). The output is a real-valued structure CHI∈ℝ, which represents the health score of the equipment in its current state. The expert knowledge rule base is a set of feature weight specifications developed based on power equipment operating experience, historical fault analysis, and domain expert experience. It is used to define the relative importance of different operating parameters in health assessment. The model classifies CHI values according to set multi-level threshold intervals and automatically triggers corresponding maintenance alarm information or equipment control linkage strategies when the CHI value exceeds the set threshold. The evaluation results are synchronously fed back to the digital twin modeling module to achieve three-dimensional visualization of health level status and dynamic warning of fault risks.

[0015] Preferably, the edge computing module deploys a TinyML neural network model and a Drools rule control engine, wherein the lightweight TinyML neural network model is used to run on resource-constrained edge node embedded gateways, PLCs, and industrial controllers, and the input is a feature vector: x=[I,V,T,H,A,ΔI,ΔV,ΔT,ΔH,ΔA]∈ , Where I is the effective value of current, V is the effective value of voltage, T is temperature, H is humidity, A is vibration intensity, ΔI is the current change rate, ΔV is the voltage change rate, ΔT is the temperature change rate, ΔH is the humidity change rate, and ΔA is the vibration change rate. The model consists of an input layer, two layers of 32 ReLU activated hidden units each, and a linear output layer, and outputs a Boolean vector: ; in =1 indicates overload warning =1 indicates contact failure =1 indicates that the insulation performance is reduced; The Drools rule control engine includes the following control rules: And the duration of this condition is Δ ≧Bs, the cooling fan is triggered to start. If H> And the partial discharge signal PD> Duration Δ ≧Bs, the main power supply is disconnected and the backup power supply is switched, where is the temperature change rate threshold, is the humidity threshold, is the PD intensity threshold, Δ and Δ are the duration of each condition respectively, and Bs is the set time threshold. The neural network and rule engine have been deployed at the edge node of the device to complete the real-time processing of input features, fault diagnosis and control decision-making locally; when it is detected =1 or any rule condition is met, a control signal is generated locally to drive the circuit breaker to trip, the cooling device to start, the load to limit, and the backup power supply to be put into use; and in the event of network interruption or high latency, the edge computing module continues to independently execute the above model reasoning and rule triggering.

[0016] The present invention provides a digital twin operation monitoring system for power equipment. It has the following beneficial effects: 1. The present invention uses an LSTM neural network to model multidimensional time series data, explores the long-term dependencies and evolution laws between equipment operating parameters, introduces the change rate of each parameter as a feature input, enhances the model's sensitivity to instantaneous changes in parameters, and uses edge computing to remove outliers, fill in missing values, and perform Z-score standardization on the original collected data to ensure data reliability, improve anti-interference ability, and improve prediction accuracy.

[0017] 2. The present invention integrates the multi-dimensional operating parameters of power equipment into a feature vector, and introduces an expert knowledge rule base based on power operation and maintenance experience to assign weights, thereby building a comprehensive and accurate state perception foundation, realizing quantitative assessment of the health status of equipment, and combining multi-level threshold judgment and digital twin three-dimensional visualization to achieve real-time health presentation, intelligent alarm and control linkage, thereby improving assessment accuracy, abnormal response capability and system intelligent operation and maintenance efficiency, and effectively reducing the risk of sudden failures.

[0018] 3. The present invention deploys the lightweight TinyML neural network model and the Drools rule control engine on edge devices in a coordinated manner, enabling the system to perform real-time processing, intelligent diagnosis, and control decisions on multi-dimensional operating parameters locally. Even in network interruptions or high-latency environments, edge nodes still have independent operation and control capabilities to ensure the continuous operation and safety protection of power equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the architecture of the present invention; Figure 2 Schematic diagram of the subsystem architecture of the present invention. DETAILED DESCRIPTION

[0020] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] Please see the attached Figure 1 , an embodiment of the present invention provides a digital twin operation monitoring system for electric power equipment, comprising: Data sensing and acquisition system: used to collect multi-dimensional operating data through current, voltage, temperature, humidity, vibration and partial discharge sensors deployed at key parts of the equipment; Condition assessment and prediction system: Based on the collected data, the LSTM model is used to assess the health status of the equipment and predict the remaining lifespan; Digital twin modeling system: Based on the collected data and equipment characteristics, CAD / CAE tools are used to build equipment geometry and physical models and synchronize them with real-time data; Linked response and control system: When equipment anomalies are detected, protective measures are automatically triggered to start backup power or adjust the load; Intelligent decision-making and optimization system: Generates optimization plans based on real-time data and forecast results, and automatically adjusts equipment operation; Visual display and platform system: Presents equipment operation, health assessment and warning information through WebGL / ECharts.

[0022] Specifically, the data perception and acquisition system collects multi-dimensional data from the equipment's operation in real time by deploying temperature sensors, current sensors, voltage sensors, partial discharge detectors, vibration sensors, and humidity sensors on the power equipment itself and key nodes. The collected raw data is initially processed and uploaded to the system platform via the data transmission module, providing basic data support for the analysis and subsequent prediction of the equipment's operating status. The system has multi-channel parallel acquisition capabilities, enabling high-frequency, high-precision, synchronous acquisition of key parameters. The condition assessment and prediction system uses real-time operating data provided by the acquisition system and an LSTM long-short-term memory network prediction model to perform health status analysis and remaining life prediction. This model has time series learning capabilities and can comprehensively analyze key characteristics such as the equipment's historical operating trajectory, current state evolution trends, and cyclical loads to dynamically predict the future operating status of power equipment. The prediction results are presented in the form of health index and remaining life value, and are updated in real time as the equipment operates, providing operation and maintenance personnel with a scientific reference for maintenance and replacement. Supported by the aforementioned data, the digital twin modeling system constructs a three-dimensional virtual model corresponding to the actual power equipment. This model is dynamically driven based on the equipment's structural parameters, operating conditions, environmental factors, and state prediction results, achieving virtual-real synchronization. Through simulation analysis, the system can simulate equipment performance under different operating conditions, identify potential abnormal trends, and enable early intervention. The digital twin model is not only used to display status but also serves as an auxiliary decision-making basis for the control system. The linked response and control system uses status assessment results and digital twin models as triggers to form an intelligent response mechanism. When abnormal equipment operation is detected or prediction results indicate an impending failure, the system automatically initiates protection strategies, controls backup power supply access or load switching, and ensures continuous and stable operation of the power system. The control system is embedded in an edge computing unit, which enables localized rapid response control, reduces response latency, and improves fault tolerance. The intelligent decision-making and optimization system integrates equipment operation data, prediction results, and historical fault models to build a multi-objective optimization algorithm to dynamically adjust the operation strategy of power equipment. The system comprehensively considers equipment health status, power quality, load balance, and energy consumption level, automatically generates the optimal operating parameter configuration plan, and links the execution system for closed-loop control, thereby improving overall operating efficiency, extending equipment service life, and reducing sudden failure rates. The visualization and platform system centrally displays the equipment operating status, health assessment indicators, prediction results and control response information through a graphical user interface. The platform supports multiple display formats such as trend charts, heat maps, health level charts, and integrates digital twin 3D visualization functions to facilitate operation and maintenance personnel to conduct real-time monitoring, remote diagnosis and auxiliary decision-making. At the same time, the platform has permission management and data traceability functions, and supports enterprise-level equipment management and analysis.

[0023] Please see the attached Figure 2 ,The data perception and acquisition system includes a multi-source sensor subsystem, an edge pre-processing subsystem, a data transmission subsystem, and a time synchronization and labeling subsystem; The multi-source sensor subsystem collects information on the device's current, voltage, temperature, humidity, vibration, and partial discharge through sensors; The edge preprocessing subsystem uses 3σ anomaly elimination, KNN interpolation and completion, and Z-Score normalization; the data transmission subsystem uses 4G / 5G transmission and redundant storage; The time synchronization and tagging subsystem uses GPS / NTP to timestamp data and provide device IDs.

[0024] Specifically, the multi-source sensor subsystem is deployed at the key parts of the power equipment and is equipped with various types of sensors for current, voltage, temperature, humidity, vibration and partial discharge signals to collect the electrical and physical environmental parameters of the power equipment in real time during operation. By deploying a heterogeneous sensor matrix, the system can achieve multi-dimensional perception of the operating status of the power equipment and provide basic data support for subsequent fault diagnosis and health assessment. Edge acquisition and computing subsystem: This subsystem is deployed at the edge of the site to perform local pre-processing on the raw data from multi-source sensors. First, noise filtering and outlier detection are performed on the raw data collected from the sensors. Using the Z-score algorithm, the system can effectively identify and remove unreasonable noise data to ensure that the input data is more reliable. For missing data that may occur during the acquisition process, the linear interpolation method is used in the calculation to fill in the values of the two valid data points before and after. At the same time, all data from different types of sensors need to be converted into a unified format through normalization, which includes unit conversion and timestamp standardization. The edge computing subsystem also converts parameters of different dimensions (such as temperature and current) into a unified scale. To reduce the amount of computation and effectively extract valid information, the edge computing subsystem performs data dimensionality reduction by extracting features from sensor data, such as the signal's mean, variance, and frequency characteristics. It then uses set thresholds to monitor key parameters in real time. When indicators such as temperature and current exceed their limits, an early warning mechanism is triggered. After data processing is completed, the data transmission subsystem is responsible for stably and quickly transmitting the edge-processed data to the central monitoring platform. This subsystem supports multiple communication protocols, including 4G / 5G, fiber optic, and Ethernet, and uses set thresholds to monitor key parameters in real time. When indicators such as temperature and current exceed their limits, an early warning mechanism is triggered. To ensure the time consistency of all collected data and the identifiability of source devices, the time synchronization and labeling subsystem accurately synchronizes the collection system using GPS timing and the Network Time Protocol, attaching a high-precision timestamp and unique device identifier to each data item. This allows for precise location and sequential recovery of the data source during data fusion, analysis, and fault tracing.

[0025] Please see the attached Figure 2The status assessment and prediction system includes a data processing subsystem, which is used to perform further statistical preprocessing on the collected operating data, including using the 3σ / Z-Score method to eliminate outliers, using the linear / KNN method to interpolate and fill missing values and Z-Score standardization, and generating time series samples according to a fixed time window. The prediction model subsystem builds an equipment health status prediction model based on multi-layer LSTM units and linear output layers, and outputs the health index HI(t). The degradation analysis subsystem calculates the degradation rate ΔHI / Δt=[HI(t)–HI(t–1)] / Δt in real time, where Δt is the interval between two predictions, ΔHI / Δt reflects the rate of change of the health index, and uses the distribution of the past N ΔHI / Δt to calculate the degradation rate. The kernel density estimation (KDE) method fits the probability density and determines the fault risk level based on the percentile of the current ΔHI / Δt in the historical distribution. The model update and optimization subsystem, after obtaining the real-time health status or actual life label, uses the corresponding input samples and labels as incremental data. The LSTM model weights are fine-tuned using the online gradient descent method and combined with old samples to stabilize training. The equipment comprehensive health index (CHI) assessment model combines electrical parameters (current I, voltage V), environmental parameters (temperature T, humidity H), and historical degradation rate (the average of the past M ΔHI / Δt values) into a feature vector. The CHI is output through normalization and weighted calculation, and is classified into normal, sub-healthy, and fault levels according to set thresholds to trigger early warning or maintenance strategies.

[0026] Specifically, the data processing subsystem is used to perform further statistical preprocessing operations on the original operating data collected from the perception system to improve data quality and ensure that the input of the prediction model is accurate and effective. First, the system uses the 3σ method to identify and eliminate outliers to avoid the impact of extreme data on model performance; secondly, for missing data, the system uses the KNN interpolation method to fill in the missing data according to the data characteristics to maintain the continuity of the time series samples; finally, the system performs Z-Score standardization on the processed data to eliminate the influence of different parameter dimensions and unify the data scale for model training and prediction. Subsequently, the system stores the preprocessed data according to the set time window. The data is constructed into time series samples as the basis for subsequent predictive modeling. The prediction model subsystem builds an equipment health status prediction model based on a multi-layer LSTM long short-term memory network structure and a linear output layer. The model can capture the medium- and long-term dependencies of time series data and realize the prediction of the health index HI(t) of power equipment. The cross entropy loss function is used in the model training process, and the model parameters are continuously optimized through back propagation and gradient descent methods to improve the prediction accuracy and generalization ability. Through the prediction output of continuous time points, the changing trend of the equipment health status can be intuitively understood. The degradation analysis subsystem further analyzes the prediction results and calculates the health index in real time. The degradation rate ΔHI / Δt=[HI(t)-HI(t–1)] / Δt, where Δt is the interval between two predictions. This rate reflects the speed of change of the equipment health status and is an important indicator for evaluating the equipment degradation trend. The system continuously records the past N times of ΔHI / Δt and uses the kernel density estimation KDE method to fit its probability density distribution. On this basis, the risk level of the equipment is judged according to the quantile position of the current ΔHI / Δt in the distribution. For example, different status levels are divided to respond to potential failure risks in a timely manner. The model update and optimization subsystem is used to improve the adaptability and accuracy of the model. The system obtains real-time health status labels or real-time health status labels. After the actual operating life label is obtained, the current input sample and the corresponding label are automatically input into the LSTM model as incremental data. The online gradient descent algorithm is used to fine-tune the model weights, and the training process is stabilized by combining some historical samples to prevent model drift or overfitting, thereby ensuring that the model continues to have predictive capabilities in a dynamic environment. The equipment comprehensive health index (CHI) evaluation model is used to quantitatively evaluate the overall health status of the equipment. The model integrates multi-source data to construct a comprehensive feature vector, including the current current I, voltage V, ambient temperature T, humidity H, and the average of the past M degradation rates ΔHI / Δt. The CHI value is calculated through weighted normalization. , where x_i is the standardized value of each feature item, is the corresponding weight, Reflecting the overall operational health level of the equipment, the system divides CHI into normal, sub-healthy, and fault levels based on set thresholds, and triggers corresponding early warning prompts or maintenance strategies, such as scheduling maintenance, replacing parts, etc., ultimately achieving effective management of power equipment operation risks and optimal configuration of operation and maintenance resources.

[0027] Please see the attached Figure 2 The digital twin modeling system includes a virtual modeling subsystem, a data synchronization and fusion subsystem, a simulation and simulation analysis subsystem, and a real-time visualization subsystem. The virtual modeling subsystem uses SolidWorks / Ansys TwinBuilder to import equipment CAD and physical properties to generate models. The data synchronization and fusion subsystem uses Kalman filtering to fuse sensor and prediction data. The simulation and simulation analysis subsystem predicts fault evolution through the OpenModelica interface. The real-time visualization subsystem renders three-dimensional twins based on Three.js / WebGL and overlays healthy color labels.

[0028] Specifically, the virtual modeling subsystem uses SolidWorks to generate a solid model based on equipment structural information such as CAD models, structural drawings, BIM data, as well as dynamic information such as historical operating data and real-time sensor data. It also depicts the connection relationships between components in the form of a network graph. At the same time, it uses a graph neural network (GNN) to learn the characteristics of component topology and capture the coupling effects under complex structures. Then, through the state-space model, it defines the dynamic equations of each component to express the transmission and interaction laws of physical quantities such as current, voltage, and temperature, realizing the "structure + behavior" dual modeling of the digital twin. The data synchronization and fusion subsystem can acquire heterogeneous data such as electrical parameters such as current and voltage, environmental parameters such as temperature and humidity, vibration signals, and partial discharge signals from multi-source sensor subsystems in real time, as well as the predicted output of the state assessment subsystem. It uses the Kalman filter algorithm to perform weighted fusion of multi-channel data by timestamp, remove noise, and fill missing values to improve data quality. At the same time, through the MQTT protocol and the edge gateway as the intermediary, the fused data is pushed to the central twin model at high speed, realizing high-frequency, low-latency "physical to virtual" state mapping; The simulation and analysis subsystem, based on virtual modeling results, connects to the OpenModelica multi-physics simulation platform to perform dynamic simulations of the equipment's operating behavior under different operating conditions, such as high temperature, high load, and humidity. Key performance indicators derived from the simulations, such as temperature rise curves, insulation resistance decay, and mechanical vibration modes, are then input into the LSTM prediction model. Combined with the Markov Chain Monte Carlo (MCMC) method, the model generates fault degradation paths and corresponding probability distributions, assessing the equipment's possible failure modes at different future time points to assist in formulating operation and maintenance strategies. The real-time visualization and monitoring subsystem uses Three.js / WebGL technology to render a 3D twin. By binding the real-time data stream after Kalman filtering, it can realize the dynamic status update of each component of the equipment, such as the intuitive display of temperature color scale, pressure isosurface, displacement vector, etc. By embedding the hierarchical analysis method (AHP) risk scoring model, the simulation prediction results are integrated with the real-time health index HI / CHI, and the risk level color bands are calculated and superimposed. In the human-computer interaction interface, it is presented in multi-dimensional views such as curve charts, radar charts, and timelines, supporting operation and maintenance personnel to conduct multi-angle, interactive status queries and early warning responses.

[0029] Please see the attached Figure 2 The linkage response and control system includes an anomaly detection subsystem, a linkage control subsystem, an execution feedback subsystem, a security policy subsystem, and an edge computing module. The anomaly detection subsystem uses the IsolationForest algorithm to perform real-time classification and recognition on the multi-dimensional time series features pre-processed by the data processing subsystem and output anomaly labels. The linkage control subsystem triggers circuit breaker tripping, backup power switching, and load limiting based on the Drools rule engine. The execution feedback subsystem collects PLC / SCADA receipts to verify execution results; the security policy subsystem uses the security boundary model to verify instruction compliance; The edge computing module deploys TinyML models and rule engines to achieve local real-time diagnosis and control.

[0030] Specifically, the anomaly detection subsystem receives the multi-dimensional time series feature data pre-processed by the data processing subsystem, uses the IsolationForest algorithm to perform real-time anomaly detection and classification, outputs specific anomaly labels, and identifies abnormal events of unknown patterns based on the isolation of the data. It is suitable for early fault detection and sudden event identification. At the same time, the subsystem can be deployed on the edge computing node to respond to sudden changes in equipment status in real time, such as current overload, voltage drop, partial discharge surge and other abnormal conditions, and achieve second-level identification. After the linkage control subsystem receives the abnormal label, it triggers the linkage operation through the Drools rule engine based on the preset control logic. For example, when a voltage anomaly is detected, the corresponding rule can be automatically called to implement backup power switching, or when the equipment is identified as overloaded, load limiting or active power off can be implemented. After the control operation is issued, the execution feedback subsystem collects the receipt information of the PLC / SCADA system and the feedback data of the equipment sensor to verify the actual execution status in real time. If a control execution failure is detected, such as the circuit breaker not tripping, The system automatically records faults and notifies the linked control subsystem to attempt redundancy strategies or alarm intervention, forming a complete closed-loop feedback process. To avoid misoperation or triggering cascading failures, the safety strategy subsystem verifies the compliance of all control instructions. This subsystem uses the SafetyEnvelopeModel safety boundary modeling method to construct a safe operating domain based on historical equipment operating data. This subsystem verifies in real time whether current control operations are within the permitted range. For example, it limits the triggering of high-voltage closing commands in maintenance mode or prohibits the activation of high-energy-consuming loads under high-temperature conditions, thereby ensuring the rationality and safety of the control strategy. The edge computing module is deployed in edge devices at the power site and has raw data processing and intelligent decision-making capabilities. It uses the TinyML model to predict the status and conduct preliminary diagnosis of local data. Combined with the Drools rule engine, it implements local response after anomaly detection. For example, in the event of communication failure or failure of the main control platform, it can still independently complete key control tasks such as emergency power outages, alarm reminders, and equipment isolation, improving the system's response efficiency.

[0031] Please see the attached Figure 2 ,The intelligent decision-making and optimization system includes decision analysis subsystem, ,optimization scheduling subsystem, strategy generation subsystem and auxiliary operation and maintenance ,subsystem; The decision analysis subsystem uses PCA dimensionality reduction + Bayesian network to fuse multi-source data; The optimization scheduling subsystem balances efficiency, energy consumption and risk based on the NSGA-II algorithm; The strategy generation subsystem converts the scheduling results into inverter speed regulation and maintenance plans; The auxiliary operation and maintenance subsystem generates decision reports through NLG and pushes them to the operation and maintenance terminal.

[0032] Specifically, the decision analysis subsystem performs dimensionality reduction processing on multi-source data based on the PCA algorithm, extracts key features, improves data processing efficiency, and further uses the Bayesian network to fuse multi-source heterogeneous data from monitoring systems, health assessment systems, historical fault libraries, etc., to achieve comprehensive modeling and dynamic evaluation of equipment status, and provide highly reliable prior information for subsequent scheduling and strategy formulation. On this basis, the optimization scheduling subsystem constructs a multi-objective optimization problem covering three major goals: operating efficiency, energy consumption, and equipment risk based on the state model output by the decision analysis subsystem, and uses the NSGA-II algorithm to solve it. The algorithm can find multiple feasible solutions, balance the trade-offs between various goals, and dynamically adjust load distribution, power scheduling, and equipment risk. The strategy generation subsystem then automatically converts the optimized scheduling results into control instructions and operation and maintenance strategies that can be directly issued, including: automatically adjusting the inverter speed according to the equipment operating status to achieve energy saving and consumption reduction; generating customized maintenance plans based on the predicted equipment life to avoid potential risks in advance and ensure stable and reliable system operation. In addition, the auxiliary operation and maintenance subsystem automatically generates equipment operation and maintenance decision reports based on natural language generation (NLG) technology, and pushes them to the operation and maintenance personnel through the operation and maintenance terminal. The report content includes current equipment status assessment, scheduling optimization suggestions, operating parameter adjustment suggestions and future risk warnings, so that the operation and maintenance personnel can grasp the equipment operation status and system optimization path in real time.

[0033] Please see the attached Figure 2 The visualization display and platform system includes a graphical display subsystem, an interactive monitoring subsystem, an alarm push subsystem and a multi-terminal access subsystem. The graphical display subsystem draws real-time trends and RUL curves based on ECharts. The interactive monitoring subsystem supports timeline and multi-dimensional filtering. The alarm push subsystem sends warnings to the Web and mobile terminals through WebSocket / MQTT. The multi-terminal access subsystem uses ReactNative and Web responsive design to achieve unified access.

[0034] Specifically, the graphical display subsystem is built based on the ECharts visualization framework. It receives real-time equipment operation data from the digital twin data layer and renders key indicators such as equipment temperature, voltage, current, load rate, and remaining life (RUL) in the form of dynamic charts, trend lines, and heat maps. The system supports continuous drawing and trend comparison of RUL prediction results, enabling operation and maintenance personnel to clearly understand equipment degradation trends and life cycle changes. The data update process is pushed frequently through the Kafka message stream combined with the WebSocket mechanism. The interactive monitoring subsystem provides powerful interactive control functions, supporting timeline backtracking and multi-dimensional screening such as device number, device type, time interval, and risk level, allowing users to quickly locate target devices or abnormal conditions from a large group of devices. This subsystem combines the PCA algorithm to perform dimensionality reduction and feature extraction on high-dimensional monitoring data. The alarm push subsystem is responsible for monitoring abnormal events and sending timely notifications. When the status assessment system or anomaly detection model determines that a device is at risk, the subsystem uses the Drools rule engine to determine and classify the alarm level. It then pushes the alarm information to web-based and mobile operation and maintenance terminals based on the WebSocket or MQTT protocol, enabling multi-terminal synchronization and instant response. The multi-terminal access subsystem uses the cross-platform development framework ReactNative and Web responsive design technology to build a unified access platform, achieving interface consistency, operational continuity and data interoperability between Web and mobile terminals.

[0035] Please see the attached Figure 2 The health status and life prediction model is built based on the long short-term memory neural network (LSTM) algorithm. The physical perception module performs high-frequency sampling of key parameters during the operation of power equipment. The collected data includes multi-dimensional operating parameters such as current, voltage, temperature, humidity, vibration, partial discharge signals, and insulation resistance, forming a high-dimensional data set with time series characteristics. The edge computing module preprocesses the raw data, including outlier removal, missing value filling, and Z-score normalization, to ensure that the input data is complete, valid, and has a unified dimension, reducing the impact of abnormal interference on model training and prediction accuracy. The preprocessed data is input into the LSTM model in the form of a tensor. The structure of the input tensor is X∈ℝ^(B×T×N), where B is the number of samples input into the model at one time, T is the length of the predicted historical time, and N is the feature dimension, including the following features: Current I, current change rate ΔI / Δt, voltage V, voltage change rate ΔV / Δt, temperature T, temperature change rate ΔT / Δt, humidity H, humidity change rate ΔH / Δt, vibration A, vibration change rate ΔA / Δt; Among them, Δ represents the difference of the variable between adjacent time steps, Δx / Δt=x(t)-x(t−1), which is used to characterize the instantaneous change trend of the feature; The LSTM model models the multi-dimensional time series data in the intelligent prediction and diagnosis module, and uses the mean square error as the loss function for gradient optimization to learn the time series evolution law of equipment degradation. The model output structure is Y∈ℝ^(B×1), which represents the remaining life prediction value of each sample. The prediction value can be given in time units. The subsequent collected data will also continue to be input into the trained LSTM model in the form of tensors to achieve real-time prediction of the current remaining life of the equipment, and the prediction results will be synchronously transmitted to the digital twin modeling module to drive the update and visualization of the virtual equipment status, thereby realizing dynamic monitoring of the operating status of the power equipment and displaying the degradation trend.

[0036] Specifically, first, the data perception and acquisition system is deployed at the key position of the power equipment to perform high-frequency real-time acquisition of key parameters of current, voltage, temperature, humidity, vibration, and partial discharge signals during operation. The collected data has typical time series characteristics and constitutes a high-dimensional time series data set. The collected raw data is then transmitted to the edge computing module for the following three types of processing operations: outlier removal to remove extreme values caused by sensor errors or sudden interference; missing value filling, using linear interpolation or sliding average strategies to make up for lost data; Z-score standardization, unifying data of different dimensions to improve the convergence speed and prediction stability of model training, so that the preprocessed data structure standard is unified. After processing, the data is organized into tensor form and input into the trained LSTM prediction model. The input tensor is denoted as X∈ℝ^(B×T×N), where B is the number of samples input to the model at one time, and T is the predicted The historical time length, N is the feature dimension, and features include current, voltage, temperature, humidity, vibration, and their respective rates of change, specifically in the form of Δx / Δt=x(t)-x(t−1), to enhance the model's perception of instantaneous feature trends. The LSTM model runs within the intelligent prediction and diagnosis module and uses the mean square error (MSE) as the loss function for optimization. The model models multidimensional time series data through multi-layer stacked LSTM units, learning the potential temporal evolution of equipment state degradation. The final output tensor Y∈ℝ^(B×1) represents the predicted remaining life of each sample under the current operating conditions, which can be expressed in hours, days, or other time units. The prediction result is transmitted synchronously in real time to the digital twin modeling module to drive the dynamic update of the virtual model state of the power equipment, enabling the model to continuously receive new data for inference and support a rolling prediction mechanism, thereby achieving continuous tracking of health index change trends and online estimation of remaining life.

[0037] Please see the attached Figure 2,The equipment comprehensive health index assessment model combines the equipment operating parameters of current, voltage, temperature rise, partial discharge intensity, vibration and environmental parameters of temperature and humidity collected by the physical perception module into a multidimensional feature vector X∈ℝⁿ, and constructs the input sample through normalization preprocessing; The feature vector X is then weighted and fused based on the feature item weight vector W∈ℝⁿ set by the expert knowledge rule base to calculate the comprehensive equipment health index (CHI=W·Xᵀ). The output is a real-valued structure CHI∈ℝ, which represents the health score of the equipment in its current state. The expert knowledge rule base is a set of feature weight specifications developed based on power equipment operating experience, historical fault analysis, and domain expert experience. It is used to define the relative importance of different operating parameters in health assessment. The model classifies CHI values according to set multi-level threshold intervals and automatically triggers corresponding maintenance alarm information or equipment control linkage strategies when the CHI value exceeds the set threshold. The evaluation results are synchronously fed back to the digital twin modeling module to achieve three-dimensional visualization of health level status and dynamic warning of fault risks.

[0038] Specifically, the data sensing and acquisition system is used to collect parameters such as current, voltage, temperature rise, partial discharge intensity, vibration, temperature and humidity in real time. The collected data is then used to form a multi-dimensional feature vector X∈ℝⁿ of the equipment operation. Normalization pre-processing is then performed in the intelligent prediction and diagnosis module to ensure that all types of data have a unified dimension and are relatively comparable, thus avoiding evaluation bias due to inconsistent data scales. Subsequently, the system calls the preset feature item weight vector W∈ℝⁿ in the expert knowledge rule base, performs weighted fusion on the normalized feature vector X, and calculates the comprehensive equipment health index CHI at the current moment. =W·Xᵀ, CHI is a real-valued structure that represents the quantitative score of the equipment health status. The expert knowledge rule base is constructed based on the power equipment operation experience, historical fault analysis and industry expert knowledge. It can assign reasonable weights according to the importance of each feature to the health of the equipment, so that the health assessment has a certain professional basis and adaptability. The system divides the CHI value into levels according to the set multi-level threshold interval, such as normal (CHI>80), warning (50<CHI≤80), fault (CHI≤50) and other health status levels. When the CHI value exceeds the set safety threshold, the system will automatically trigger The system generates maintenance alerts and initiates related control linkage strategies, such as adjusting equipment operating parameters, enabling redundant equipment, and sending remote operation and maintenance instructions, to reduce potential failure risks. Assessment results are synchronized in real time to the digital twin modeling module, driving status updates for the virtual equipment model and presenting the equipment's current health level on a 3D visualization platform. Operation and maintenance personnel can use this interface to intuitively view the equipment's health score, trend changes, and warning information, enabling real-time monitoring of equipment operating status, risk prediction, and decision support, thereby improving the accuracy and response efficiency of operation and maintenance. By integrating data parameters from power equipment to form a multidimensional feature vector and introducing an expert knowledge rule base based on power operation and maintenance experience to assign weights to each feature item, a more comprehensive and accurate state perception foundation can be established, enabling quantitative assessment of equipment health status. This not only improves the professionalism and accuracy of health assessments, but also enhances the system's responsiveness to abnormal trends and the foresight of fault prediction, effectively reducing the risk of sudden failures. Combined with a multi-level threshold judgment mechanism and the digital twin 3D visualization platform, real-time presentation of health assessment results, intelligent alarm and control linkage are possible, further enhancing the system's intelligent operation and maintenance capabilities and fault warning efficiency.

[0039] Please see the attached Figure 2 The edge computing module deploys a TinyML neural network model and a Drools rule control engine. The lightweight TinyML neural network model is used to run on resource-constrained edge nodes, embedded gateways, PLCs, and industrial controllers. The input is a feature vector: x=[I,V,T,H,A,ΔI,ΔV,ΔT,ΔH,ΔA]∈ , Where I is the effective value of current, V is the effective value of voltage, T is temperature, H is humidity, A is vibration intensity, ΔI is the current change rate, ΔV is the voltage change rate, ΔT is the temperature change rate, ΔH is the humidity change rate, and ΔA is the vibration change rate. The model consists of an input layer, two layers of 32 ReLU activated hidden units each, and a linear output layer, outputting a Boolean vector: ; in =1 indicates overload warning =1 indicates contact failure =1 indicates that the insulation performance is reduced; The Drools rule control engine includes the following control rules: And the duration of this condition is Δ ≧Bs, the cooling fan is triggered to start. If H> And the partial discharge signal PD> Duration Δ ≧Bs, the main power supply is disconnected and the backup power supply is switched, where is the temperature change rate threshold, is the humidity threshold, is the PD intensity threshold, Δ and Δ are the duration of each condition respectively, Bs is the set time threshold, the neural network and rule engine have been deployed at the edge node of the device, and the real-time processing of input features, fault diagnosis and control decision-making are completed locally; when a fault is detected =1 or any rule condition is met, a control signal is generated locally to drive the circuit breaker to trip, the cooling device to start, the load to limit, and the backup power supply to be put into use; and in the event of network interruption or high latency, the edge computing module continues to independently execute the above model reasoning and rule triggering.

[0040] Specifically, the edge computing module first collects the operating parameters of the equipment in real time from the data perception and acquisition system, including current (I), voltage (V), temperature (T), humidity (H), vibration intensity (A) and the change rate of each parameter ΔI, ΔV, ΔT, ΔH, ΔA, to form a multidimensional feature vector x=[I,V,T,H,A,ΔI,ΔV,ΔT,ΔH,ΔA]∈ , is input into the deployed TinyML neural network model, which consists of an input layer, two hidden layers with 32 ReLU activation units each, and a linear output layer that can output a Boolean vector , which correspond to three typical fault types of the equipment: overload warning (y1=1), contact fault (y2=1), and insulation performance degradation (y3=1). At the same time, the Drools rule control engine runs on the same edge node to determine whether the manually formulated operation rules are met. If the temperature change rate ΔT exceeds the set threshold , and the duration Δt≥Bs, the system triggers the cooling fan to start; if the humidity H exceeds the threshold , and at the same time the partial discharge signal PD exceeds the set threshold When the duration Δt≥Bs, the control strategy of disconnecting the main power supply and switching to the backup power supply is executed. The above thresholds and durations are preset by system parameters and can be adjusted and optimized according to actual working conditions. At the edge node, when the TinyML model outputs any alarm signal y1 / y2 / y3 as 1, or the Drools engine rule is triggered, the system immediately generates a corresponding control signal locally to execute safety protection actions such as circuit breaker tripping, cooling system startup, load limiting or backup power switching, thereby achieving rapid response and avoiding the delay risk brought by relying on cloud decision-making. In addition, when the communication network is interrupted or cloud processing is unavailable, When the system is not fully operational, the edge computing module can still operate independently to maintain the equipment's continuous monitoring and control decision-making capabilities. This mechanism ensures the safe operation and emergency response capabilities of the equipment in extreme or emergency environments, and enhances the stability and intelligence level of the entire power system. By collaboratively deploying the lightweight TinyML neural network model and the Drools rule control engine on edge devices, the system can perform real-time processing, intelligent diagnosis, and control decisions on multi-dimensional operating parameters locally. Even in network outages or high-latency environments, edge nodes still have independent operation and control capabilities to ensure the continuous operation and safety protection of power equipment.

[0041] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A digital twin operation monitoring system for power equipment, characterized in that: include: Data sensing and acquisition system: used to collect multi-dimensional operating data through current, voltage, temperature, humidity, vibration and partial discharge sensors deployed at key parts of the equipment; Condition assessment and prediction system: Based on the collected data, the LSTM model is used to assess the health status of the equipment and predict the remaining lifespan; Digital twin modeling system: Based on the collected data and equipment characteristics, CAD / CAE tools are used to build equipment geometry and physical models and synchronize them with real-time data; Linked response and control system: When equipment anomalies are detected, protective measures are automatically triggered to start backup power or adjust the load; Intelligent decision-making and optimization system: Generates optimization plans based on real-time data and forecast results, and automatically adjusts equipment operation; Visual display and platform system: Presents equipment operation, health assessment and warning information through WebGL / ECharts.

2. A digital twin operation monitoring system for electric power equipment according to claim 1, characterized in that: The data perception and acquisition system includes a multi-source sensor subsystem, an edge pre-processing subsystem, a data transmission subsystem, and a time synchronization and labeling subsystem; The multi-source sensor subsystem collects information on current, voltage, temperature, humidity, vibration, and partial discharge of the device through sensors; The edge preprocessing subsystem uses 3σ anomaly elimination, KNN interpolation and completion, and Z-Score normalization; the data transmission subsystem uses 4G / 5G transmission and redundant storage; The time synchronization and tagging subsystem uses GPS / NTP to timestamp the data and add device IDs.

3. The digital twin operation monitoring system for electric power equipment according to claim 1, characterized in that: The state assessment and prediction system includes a data processing subsystem for further statistical preprocessing of the collected operating data, including using the 3σ / Z-Score method to eliminate outliers, using the linear / KNN method to interpolate and fill missing values and Z-Score standardization, and generating time series samples according to a fixed time window; a prediction model subsystem for building an equipment health state prediction model based on a multi-layer LSTM unit and a linear output layer, and outputting a health index HI(t); a degradation analysis subsystem for real-time calculation of the degradation rate ΔHI / Δt=[HI(t)–HI(t–1)] / Δt, where Δt is the interval between two predictions, ΔHI / Δt reflects the rate of change of the health index, and uses the distribution of the past N ΔHI / Δt to calculate the degradation rate. The kernel density estimation (KDE) method fits the probability density and determines the fault risk level based on the percentile of the current ΔHI / Δt in the historical distribution. The model update and optimization subsystem, after obtaining the real-time health status or actual life label, uses the corresponding input samples and labels as incremental data. The LSTM model weights are fine-tuned using the online gradient descent method and combined with old samples to stabilize training. The equipment comprehensive health index (CHI) assessment model combines electrical parameters (current I, voltage V), environmental parameters (temperature T, humidity H), and historical degradation rate (the average of the past M ΔHI / Δt values) into a feature vector. The CHI is output through normalization and weighted calculation, and is classified into normal, sub-healthy, and fault levels according to set thresholds to trigger early warning or maintenance strategies.

4. A digital twin operation monitoring system for electric power equipment according to claim 1, characterized in that: The digital twin modeling system includes a virtual modeling subsystem, a data synchronization and fusion subsystem, a simulation and simulation analysis subsystem, and a real-time visualization subsystem. The virtual modeling subsystem uses SolidWorks / Ansys TwinBuilder to import equipment CAD and physical properties to generate a model. The data synchronization and fusion subsystem uses Kalman filtering to fuse sensor and prediction data. The simulation and simulation analysis subsystem predicts fault evolution through the OpenModelica interface. The real-time visualization subsystem renders a three-dimensional twin based on Three.js / WebGL and overlays a healthy color code.

5. The digital twin operation monitoring system for electric power equipment according to claim 1, characterized in that: The linkage response and control system includes an anomaly detection subsystem, a linkage control subsystem, an execution feedback subsystem, a security policy subsystem, and an edge computing module. The anomaly detection subsystem uses the IsolationForest algorithm to perform real-time classification and identification on the multi-dimensional time series features pre-processed by the data processing subsystem and output an anomaly label. The linkage control subsystem triggers circuit breaker tripping, backup power switching, and load limiting based on the Drools rule engine. The execution feedback subsystem collects PLC / SCADA receipt verification execution results; the security policy subsystem verifies instruction compliance using a security boundary model; The edge computing module deploys the TinyML model and rule engine to achieve local real-time diagnosis and control.

6. A digital twin operation monitoring system for electric power equipment according to claim 1, characterized in that: The intelligent decision-making and optimization system includes a decision analysis subsystem, an optimization scheduling subsystem, a strategy generation subsystem and an auxiliary operation and maintenance subsystem; The decision analysis subsystem uses PCA dimension reduction + Bayesian network to fuse multi-source data; The optimization scheduling subsystem balances efficiency, energy consumption and risk based on the NSGA-II algorithm; The strategy generation subsystem converts the scheduling results into a frequency converter speed regulation and maintenance plan; The auxiliary operation and maintenance subsystem generates a decision report through NLG and pushes it to the operation and maintenance terminal.

7. The digital twin operation monitoring system for electric power equipment according to claim 1, characterized in that: The visualization display and platform system includes a graphical display subsystem, an interactive monitoring subsystem, an alarm push subsystem and a multi-terminal access subsystem. The graphical display subsystem draws real-time trends and RUL curves based on ECharts. The interactive monitoring subsystem supports timeline and multi-dimensional filtering. The alarm push subsystem sends warnings to the Web and mobile terminals through WebSocket / MQTT. The multi-terminal access subsystem adopts ReactNative and Web responsive design to achieve unified access.

8. The digital twin operation monitoring system for electric power equipment according to claim 3, characterized in that: The health status and life prediction model is built based on the long short-term memory neural network (LSTM) algorithm. The physical perception module performs high-frequency sampling of key parameters during the operation of power equipment. The collected data includes multi-dimensional operating parameters such as current, voltage, temperature, humidity, vibration, partial discharge signal, and insulation resistance, forming a high-dimensional data set with time series characteristics. The edge computing module preprocesses the raw data, including outlier removal, missing value filling, and Z-score normalization, to ensure that the input data is complete, valid, and has a unified dimension, reducing the impact of abnormal interference on model training and prediction accuracy. The preprocessed data is input into the LSTM model in the form of a tensor. The structure of the input tensor is X∈ℝ^(B×T×N), where B is the number of samples input into the model at one time, T is the length of the predicted historical time, and N is the feature dimension, including the following features: Current I, current change rate ΔI / Δt, voltage V, voltage change rate ΔV / Δt, temperature T, temperature change rate ΔT / Δt, humidity H, humidity change rate ΔH / Δt, vibration A, vibration change rate ΔA / Δt; Among them, Δ represents the difference of the variable between adjacent time steps, Δx / Δt=x(t)-x(t−1), which is used to characterize the instantaneous change trend of the feature; The LSTM model models the multi-dimensional time series data in the intelligent prediction and diagnosis module, and uses the mean square error as the loss function for gradient optimization to learn the time series evolution law of equipment degradation. The model output structure is Y∈ℝ^(B×1), which represents the remaining life prediction value of each sample. The prediction value can be given in time units. The subsequent collected data will also continue to be input into the trained LSTM model in the form of tensors to achieve real-time prediction of the current remaining life of the equipment, and the prediction results will be synchronously transmitted to the digital twin modeling module to drive the update and visualization of the virtual equipment status, thereby realizing dynamic monitoring of the operating status of the power equipment and displaying the degradation trend.

9. The digital twin operation monitoring system for electric power equipment according to claim 3, characterized in that: The equipment comprehensive health index assessment model combines the equipment operating parameters collected by the physical sensing module, including current, voltage, temperature rise, partial discharge intensity, and vibration, with the environmental parameters of temperature and humidity, to form a multidimensional feature vector X∈ℝⁿ, and constructs input samples through normalization preprocessing. The feature vector X is then weighted and fused based on the feature item weight vector W∈ℝⁿ set by the expert knowledge rule base to calculate the comprehensive equipment health index (CHI=W·Xᵀ). The output is a real-valued structure CHI∈ℝ, which represents the health score of the equipment in its current state. The expert knowledge rule base is a set of feature weight specifications developed based on power equipment operating experience, historical fault analysis, and domain expert experience. It is used to define the relative importance of different operating parameters in health assessment. The model classifies CHI values according to set multi-level threshold intervals and automatically triggers corresponding maintenance alarm information or equipment control linkage strategies when the CHI value exceeds the set threshold. The evaluation results are synchronously fed back to the digital twin modeling module to achieve three-dimensional visualization of health level status and dynamic warning of fault risks.

10. The digital twin operation monitoring system for electric power equipment according to claim 5, characterized in that: The edge computing module deploys a TinyML neural network model and a Drools rule control engine, wherein the lightweight TinyML neural network model is used to run on resource-constrained edge node embedded gateways, PLCs, and industrial controllers, and the input is a feature vector: x=[I,V,T,H,A,ΔI,ΔV,ΔT,ΔH,ΔA]∈ , Where I is the effective value of current, V is the effective value of voltage, T is temperature, H is humidity, A is vibration intensity, ΔI is the current change rate, ΔV is the voltage change rate, ΔT is the temperature change rate, ΔH is the humidity change rate, and ΔA is the vibration change rate. The model consists of an input layer, two layers of 32 ReLU activated hidden units each, and a linear output layer, and outputs a Boolean vector: ; in =1 indicates overload warning =1 indicates contact failure =1 indicates that the insulation performance is reduced; The Drools rule control engine includes the following control rules: And the duration of this condition is Δ ≧Bs, the cooling fan is triggered to start. If H> And the partial discharge signal PD> Duration Δ ≧Bs, the main power supply is disconnected and the backup power supply is switched, where is the temperature change rate threshold, is the humidity threshold, is the PD intensity threshold, Δ and Δ are the duration of each condition respectively, and Bs is the set time threshold. The neural network and rule engine have been deployed at the edge node of the device to complete the real-time processing of input features, fault diagnosis and control decision-making locally; when it is detected =1 or any rule condition is met, a control signal is generated locally to drive the circuit breaker to trip, the cooling device to start, the load to limit, and the backup power supply to be put into use; and in the event of network interruption or high latency, the edge computing module continues to independently execute the above model reasoning and rule triggering.

Images