Power grid dynamic load balancing and fault self-healing method based on digital twin fusion architecture

Abstract

The invention discloses a power grid dynamic load balancing and fault self-healing method based on a digital twinborn fusion architecture, and the method comprises the steps: constructing the digital twinborn fusion architecture which comprises a sensing layer, a data fusion layer, a digital twinborn model layer, a decision control layer and an execution layer, and achieving the deep fusion and virtual-real synchronous mapping of multi-source data of a power grid; through a particle swarm optimization algorithm, taking the minimum variance of the load rate of the whole network as a target, and combining constraint conditions such as line power and transformer capacity to realize dynamic load balancing of the power grid; and finally, based on fault feature extraction and digital twinborn simulation positioning, generating a fault isolation and load recovery strategy to complete fault self-healing. The method can respond to the power grid operation state change in real time, shortens the fault processing time, improves the power grid operation efficiency and the power supply reliability, and is suitable for a cooperative control scene of a modern smart power grid.

Description

Technical Field

[0001] This application relates to the field of power grid operation and management technology, specifically to a method for dynamic load balancing and fault self-healing of power grids based on a digital twin fusion architecture. Background Technology

[0002] With the rapid development of smart grids, the power grid scale continues to expand, and the proportion of renewable energy integration continues to increase. The widespread application of smart grids makes the power system operate more efficiently, flexibly, and reliably, meeting the ever-growing demand for electricity.

[0003] Traditional power grid load balancing methods often rely on static models and experience-based scheduling. Static models are built based on historical power grid data over a certain period, making it difficult to account for dynamic changes in the power grid's operating status. Experience-based scheduling relies on the personal experience and judgment of dispatchers, which is highly subjective. These methods struggle to respond in real-time to dynamic changes in the power grid's operating status. Furthermore, power grid fault diagnosis and self-healing primarily depend on manual inspections and offline analysis. Manual inspections require significant manpower and time and are difficult to perform in real-time. Offline analysis involves post-fault analysis of collected data, failing to detect and address faults promptly. While digital twin technology can construct virtual models synchronously mapped to the physical power grid, enabling real-time monitoring, simulation analysis, and predictive early warning of power grid operating status, current applications of digital twins in the power grid are mostly limited to single links or local areas, lacking deep integration of multi-source data and global collaborative optimization capabilities.

[0004] Considering that traditional power grid load balancing methods are unable to respond to dynamic changes in the power grid's operating status in real time, leading to uneven load distribution and potential local overloads or resource waste; and that power grid fault diagnosis and self-healing rely on manual inspections and offline analysis, resulting in time-consuming fault location and delayed self-healing decisions, they cannot meet the high requirements of modern power grids for power supply reliability and stability; and that the existing application limitations of digital twins in the power grid make it difficult to effectively support the integrated realization of dynamic load balancing and fault self-healing, there is an urgent need to provide a method for dynamic load balancing and fault self-healing of the power grid that solves the above problems. Summary of the Invention

[0005] In order to respond to the dynamic changes in the power grid operating status in real time, avoid local overload or resource waste, and perform timely and stable fault diagnosis and fault self-healing, this application provides a power grid dynamic load balancing and fault self-healing method based on a digital twin fusion architecture.

[0006] Firstly, this application provides a method for dynamic load balancing and fault self-healing of power grids based on a digital twin fusion architecture, including: Constructing a digital twin fusion architecture: The digital twin fusion architecture includes a perception layer, a data fusion layer, a digital twin model layer, a decision control layer, and an execution layer; A collaborative digital twin fusion architecture enables dynamic load balancing of the power grid. This includes: collecting power grid operating status parameters using multiple types of sensors deployed in the perception layer; performing noise reduction and standardization on the collected data using the data fusion layer; constructing a 3D twin model that synchronously maps the physical power grid using the digital twin model layer; monitoring the power grid operating status parameters output by the digital twin model layer using the decision control layer; and, when the load rate exceeds a preset threshold, initiating a dynamic load balancing algorithm based on particle swarm optimization to calculate the optimal load allocation ratio and send it to the execution layer; and finally, using the execution layer to perform dynamic load balancing allocation. A collaborative digital twin fusion architecture enables a fault self-healing decision-making algorithm, including: collecting fault signals using multiple types of sensors deployed in the perception layer; extracting and classifying fault features from the collected fault signals using wavelet transform in the data fusion layer; constructing a fault simulation model based on the acquired fault feature extraction and classification and traveling wave ranging algorithm in the digital twin model layer to achieve fault location; generating fault isolation commands and load restoration strategies using a fuzzy neural network in the decision control layer and sending them to the execution layer; and executing fault isolation and power restoration in the execution layer.

[0007] By adopting the above scheme, a digital twin fusion architecture is constructed to achieve deep fusion and virtual-real synchronous mapping of multi-source data of the power grid, which can reflect the power grid operation status in a timely and accurate manner and provide a reliable decision-making basis for dynamic load balancing and fault self-healing. A dynamic load balancing algorithm based on particle swarm optimization is adopted to adjust the load distribution in real time according to the changes in the power grid operation status, so as to avoid local overload and resource waste. Based on the fault simulation analysis of the digital twin model, the rapid and accurate fault location and the optimized generation of self-healing strategies are realized, achieving integrated and coordinated control of dynamic load balancing and fault self-healing of the power grid.

[0008] Preferably, the steps of the collaborative digital twin fusion architecture to achieve dynamic load balancing of the power grid further include: The system utilizes a perception layer to collect power grid topology data, electrical coupling data, and environmental and load characteristic data; a data fusion layer performs adaptive clustering based on power grid topology data and electrical coupling data to obtain the divided regions; and a digital twin model layer constructs three-dimensional twin models of the regions that are synchronously mapped to the physical power grid based on the divided regions. The system utilizes edge computing devices pre-designed in the decision control layer to perform local dynamic load balancing optimization based on the power grid operation status parameters within the sub-region where the edge computing devices are located, as output from the monitored digital twin model layer. This includes initiating a dynamic load balancing algorithm based on particle swarm optimization when the regional load rate exceeds a preset threshold, calculating the optimal load allocation ratio for the local region. The system then utilizes cloud-based systems designed in the decision control layer to perform interval-based collaborative dynamic load balancing optimization based on the power grid operation status parameters within all sub-regions, as output from the monitored digital twin model layer. This includes initiating a dynamic load balancing algorithm based on distributed collaborative optimization when the regional load rate, after local optimization, still exceeds a preset threshold for a preset duration, calculating the optimal load allocation ratio for each region. The calculated optimal load allocation ratios for each region are then distributed to the edge computing devices in each region. Each edge computing device re-optimizes the optimal load allocation ratio for its local region based on the new load allocation conditions, iteratively calculating until interval-based collaborative dynamic load balancing and regional dynamic load balancing are achieved.

[0009] By adopting the above scheme, the perception layer collects data such as power grid topology, the data fusion layer performs regional adaptive clustering, and the digital twin model layer constructs a three-dimensional twin model of the region, which more accurately reflects the power grid operation status of different regions. The edge computing devices of the decision control layer are used to perform local dynamic load balancing optimization, and the cloud can perform inter-regional collaborative dynamic load balancing optimization, realizing load balancing at different levels. Targeted adjustments are made according to the load conditions of different regions to make the load distribution more reasonable and effectively avoid local overload and resource waste.

[0010] Preferably, the steps of the collaborative digital twin fusion architecture to achieve dynamic load balancing of the power grid further include: The data fusion layer is used to perform multi-scale load forecasting based on historical load data, historical renewable energy output data, and meteorological data and user load type proportions in historical environmental and load characteristic data within the region. The data fusion layer uses multi-scale load forecasting error and renewable energy output fluctuations as random variables, and Monte Carlo simulation is used to generate several sets of node voltage and line power scenario data within the region. Each scenario is assigned a probability value, and the standardization processing and key indicator extraction of probabilistic power flow output are completed. Using the digital twin model layer, a load prediction submodule that synchronously maps load prediction within the region to multi-scale load prediction is constructed, and an uncertainty submodule that synchronously maps the standardized processing and key indicator extraction of probabilistic power flow output within the region is constructed. The edge computing device, pre-designed by the decision control layer, performs local dynamic load balancing optimization based on the standardized processing and key indicators of the power grid operation status parameters and probabilistic power flow output within the area where the edge computing device is located, according to the output of the monitored digital twin model layer. This includes initiating a dynamic load balancing algorithm fused with the dynamic PSO objective function when the predicted regional load rate exceeds a preset threshold, calculating the optimal load allocation ratio and sending it to the execution layer. This includes: designing a penalty term based on probabilistic indicators, incorporating it into the dynamic PSO objective function with the goal of minimizing the local load rate variance, optimizing the feasible domain of variables including power allocation ratio and adjusting resource output, constructing constraints based on probabilistic indicators, and solving to obtain the optimal load allocation ratio and resource output.

[0011] By adopting the above scheme, multi-scale load forecasting and Monte Carlo simulation based on historical data are carried out, relevant sub-modules are constructed, and local dynamic load balancing optimization is performed by combining power grid operating status parameters and probabilistic power flow indicators. This can more accurately respond to load changes and new energy fluctuations, achieve more precise load allocation, and improve the stability and reliability of power grid operation.

[0012] Preferably, the step of using the decision control layer to monitor the power grid operating status parameters output by the digital twin model layer, and when the load rate exceeds a preset threshold, to initiate a dynamic load balancing algorithm based on particle swarm optimization to calculate the optimal load allocation ratio further includes: The inertia weight in particle velocity updates is dynamically adjusted. Every N iterations, particle fitness is calculated. Compared to the previous fitness calculation, if the current particle fitness increases, the inertia weight is decreased, but not less than the minimum inertia weight threshold; if the particle fitness decreases, the inertia weight is increased, but not greater than the maximum inertia weight threshold. A velocity correction term is added based on the variance change rate: the variance change rate corresponds to a feedback gain, which is dynamically adjusted based on the absolute value of the difference between the current and previous variance change rates. If the absolute value of the difference between the current and previous variance change rates decreases, the feedback gain is increased; conversely, the feedback gain is decreased.

[0013] By adopting the above scheme, the inertia weight is dynamically adjusted in the dynamic load balancing algorithm based on particle swarm optimization, and a velocity correction term is added according to the variance change rate and the feedback gain is dynamically adjusted. This makes the algorithm more flexible in searching for the optimal solution during the iteration process, improves the accuracy and efficiency of load allocation ratio calculation, and thus more effectively realizes dynamic load balancing of the power grid.

[0014] Preferably, the collaborative digital twin fusion architecture for implementing the fault self-healing decision algorithm further includes: The data fusion layer determines the regional scene classification of the divided regions based on the topology parameters, environmental and load characteristics of the divided regions. The regional scene classification is pre-set based on the topology parameters, environmental and load characteristics of the divided regions. The data fusion layer uses a differentiated fault feature extraction and classification method based on regional scene classification instead of using wavelet transform to extract and classify fault features from the collected fault signals. A fault simulation model is constructed using a digital twin model layer based on the acquired fault feature extraction and classification and the traveling wave ranging algorithm adapted to the partition topology, so as to realize fault location; wherein, different regional scene classifications are set with corresponding traveling wave ranging algorithms. The decision control layer uses a fuzzy neural network with an embedded scenario rule base to generate fault isolation instructions and load recovery strategies within the region and sends them to the execution layer.

[0015] By adopting the above scheme, fault features are extracted and classified differently according to different scenarios in the divided areas. A fault simulation model is constructed by combining the traveling wave ranging algorithm with partition topology adaptation, so as to achieve more accurate fault location. Furthermore, a fuzzy neural network with embedded scenario rule base is used to generate fault isolation instructions and load recovery strategies that are more in line with the actual situation of the area, thereby improving the accuracy and efficiency of fault self-healing.

[0016] Preferably, the steps of the collaborative digital twin fusion architecture to achieve dynamic load balancing of the power grid further include: The decision control layer monitors the power grid operation status parameters output by the digital twin model layer. Combining the results of whether the load rate exceeds the preset threshold, the duration of the load rate exceeding the preset threshold, and whether there is overload or overvoltage, the severity of the load overload is comprehensively judged. When the load rate exceeds the preset threshold, a dynamic load balancing algorithm based on particle swarm optimization, adapted to the severity of the current load overload, or a dynamic load balancing algorithm based on improved particle swarm optimization is selected to replace the dynamic load balancing algorithm based on particle swarm optimization.

[0017] By adopting the above scheme, the decision control layer combines multiple factors to comprehensively judge the severity of load overload, and selects a dynamic load balancing algorithm according to the severity, so as to adjust the load distribution more accurately according to the actual load of the power grid, and further avoid local overload and resource waste.

[0018] Preferably, it also includes: a collaborative digital twin fusion architecture to achieve mutual auxiliary verification of dynamic load balancing and fault self-healing in the power grid; including: By statistically analyzing the historical power grid operation status parameters within the decision control layer, the range of normal load rate in the region is determined, and a baseline is generated. After fault location is performed using the digital twin model layer, the deviation between the real-time operation status parameters of the fault-located power grid node and the baseline is compared. If the deviation in the comparison result is greater than the preset deviation, the fault location is verified to be accurate; otherwise, it is determined that there is a false alarm in the fault location and the fault location is re-performed.

[0019] By adopting the above scheme, the normal load rate range is determined by statistically analyzing the power grid operation status parameters in the historical region at the decision control level to generate a baseline. The fault location data is then compared with the baseline to effectively verify the fault location results, avoid false alarms in fault location, and ensure the accuracy of fault location.

[0020] Preferably, it also includes: a collaborative digital twin fusion architecture to achieve mutual auxiliary verification of dynamic load balancing and fault self-healing in the power grid, and further includes: In the process of constructing a fault simulation model based on the acquired fault feature extraction and classification and traveling wave ranging algorithm in the digital twin model layer to realize fault location, the fault location of faults with a fixed fault type and a higher than preset fault probability is statistically analyzed, and typical fault cases containing fault features and fault probabilities are established. In the process of obtaining the optimal load allocation ratio in the decision control layer, the execution of the optimal load allocation ratio is verified based on the typical fault cases to see if there is a fault risk greater than the preset fault probability. If so, the current optimal load allocation ratio is removed and a new optimal load allocation ratio is obtained.

[0021] By adopting the above scheme, typical fault cases are established during fault location, and the presence of fault risks is verified based on these typical fault cases when obtaining the optimal load distribution ratio. Risky ratios are eliminated and the ratios are obtained again, which can effectively reduce the probability of faults during power grid load balancing and improve the safety and stability of power grid operation.

[0022] Preferably, it also includes: a collaborative digital twin fusion architecture to achieve mutual auxiliary verification of dynamic load balancing and fault self-healing in the power grid, and further includes: The decision control layer uses a fuzzy neural network to generate fault isolation commands and load restoration strategies, which are then sent to the execution layer. During the execution layer's fault isolation and power restoration process, the self-healing time limit for the fault isolation and power restoration process is determined, and a self-healing time limit classification is established based on the self-healing time limit duration. Different self-healing time limit classifications have preset self-healing time limit duration ranges that match them. Based on the self-healing time limit classification, a load balancing coordination strategy is matched, including an emergency balancing mode that suspends the balancing adjustment of all preset non-critical grid nodes, a balancing coordination mode that retains balancing resources in half of the regions, and a normal balancing mode that performs load balancing normally. In the next moment, the decision control layer monitors the power grid operation status parameters output by the digital twin model layer. When the load rate exceeds the preset threshold, a decision is made on whether to start the dynamic load balancing algorithm based on particle swarm optimization based on the matching load balancing coordination strategy.

[0023] By adopting the above scheme, the load balancing coordination strategy can be matched according to the self-healing time limit of the fault self-healing. The load balancing strategy can be flexibly adjusted under different self-healing conditions, avoiding interference of load balancing operation on fault self-healing during fault handling. This ensures that the power grid can operate more stably and efficiently during the fault self-healing process and improves the overall ability of the power grid to cope with faults.

[0024] Preferably, the digital twin fusion architecture includes a support layer; the support layer provides network security protection and redundant communication network assurance.

[0025] By adopting the above scheme, security and redundant communication guarantees are provided for power grid operation data transmission, ensuring the security and reliability of data transmission.

[0026] In summary, this application has the following beneficial effects: 1. By constructing and coordinating a digital twin fusion architecture, deep fusion of multi-source data from the power grid and synchronous mapping between the virtual and real worlds are achieved, enabling real-time response to changes in the power grid's operating status, resolving uneven load distribution, and improving power grid operating efficiency. Dynamic load balancing algorithms based on particle swarm optimization are used to adjust load distribution in real time according to power grid load rates, avoiding local overload and resource waste. Furthermore, by leveraging fault simulation models in the digital twin model layer and fuzzy neural networks in the decision control layer, faults can be quickly and accurately located and self-healing strategies generated, shortening fault handling time and enhancing the power grid's fault self-healing capabilities. 2. Implement adaptive clustering and partitioning of regions, construct three-dimensional twin models of regions, and perform dynamic load balancing optimization through edge computing devices and the cloud for local and inter-regional collaboration, thereby improving the accuracy and efficiency of load balancing; implement multi-scale load prediction within regions, generate standardized processing and key indicators for probabilistic power flow output, construct load prediction submodule and uncertainty submodule, and perform local dynamic load balancing optimization based on probabilistic indicators to make load distribution more in line with actual operating conditions. 3. A collaborative digital twin fusion architecture is used to verify the accuracy of fault location based on the dynamic load status parameters of the power grid, and to adjust the dynamic load adjustment strategy of the power grid based on typical fault locations, so as to achieve mutual auxiliary verification of dynamic load balancing and fault self-healing of the power grid. Attached Figure Description

[0027] Figure 1 This is a flowchart of the power grid dynamic load balancing and fault self-healing method based on a digital twin fusion architecture as described in a specific embodiment; Figure 2This is a structural block diagram of the digital twin fusion architecture constructed in the power grid dynamic load balancing and fault self-healing method based on digital twin fusion architecture described in a specific embodiment. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0029] like Figure 1 As shown in the figure, this application discloses a method for dynamic load balancing and fault self-healing of power grids based on a digital twin fusion architecture, including steps such as constructing a digital twin fusion architecture, coordinating the operation of the architecture to achieve dynamic load balancing of the power grid, and coordinating the operation of the architecture to implement a fault self-healing decision algorithm. The details of each step will be described below.

[0030] S1. Construct a digital twin fusion architecture.

[0031] Specifically, such as Figure 2 As shown, the digital twin fusion architecture includes a perception layer, a data fusion layer, a digital twin model layer, a decision control layer, and an execution layer, with each layer working collaboratively.

[0032] In this embodiment, various types of sensors are deployed for the perception layer, such as the ACS712-5A current sensor with a sampling accuracy of 0.2 and the JDZ10-10 voltage sensor. The deployment follows a three-tiered "line-equipment-node" structure, with two current sensors and one voltage sensor deployed per 1km of transmission line. Each transformer, circuit breaker, and other critical equipment is equipped with one infrared temperature sensor (STH-10, measurement range -40℃-125℃) and one status monitoring terminal (DTU distribution terminal, supporting IEC 61850 protocol). All devices use RS485 or Ethernet interfaces, with sampling frequencies set according to data type differences. For example, electrical quantities (voltage, current) are sampled at 10Hz, while temperature and status quantities are sampled at 5Hz. Alternatively, external data from other power grid-related systems, such as meteorological data and user electricity consumption, can be received via Ethernet. The collected data is transmitted to the data fusion layer via 4G / 5G industrial modules or fiber optic Ethernet, with a data transmission delay not exceeding 200ms.

[0033] Specifically, for the data fusion layer, two edge computing nodes are deployed, such as NVIDIA Jetson AGX Xavier, with an 8-core CPU and a 512-core GPU, enabling dual-machine hot standby. A Kalman filter algorithm (Q=0.01, R=0.1) is used to remove sensor noise data, and data such as SCADA real-time data, equipment ledger static data, and meteorological data are converted into standardized datasets based on the JSON-LD protocol. Simultaneously, an industrial-grade database server (CPU Intel Xeon Gold 6330, memory 64GB) is configured, using PostgreSQL+PostGIS to store spatiotemporal data, with a data retention period of 90 days. Scheduled backups are also enabled, such as a full backup performed every morning, with a data write latency of no more than 100ms.

[0034] Specifically, for the construction of the twin digital model layer, based on the 1:500 precision GIS topology map of the physical power grid and equipment nameplate parameters (such as transformer model S11-M-1000 / 10, rated capacity 1000kVA), a geometric model is built using Unity3D, and a physical simulation model is built using PSCAD / EMTDC (version 4.6). The two models are linked through Python scripts to form a three-dimensional twin model of "geometry-physics-behavior". The OPC UA V1.04 protocol is used to realize the synchronization of virtual and real data, and the synchronization cycle is dynamically adjusted according to the operating status: 500ms during normal operation and 200ms during load fluctuations or faults. The model is calibrated monthly using measured data (such as selecting 100 sets of typical operating condition data), and the model error rate after calibration is ≤2%.

[0035] Specifically, the decision-making and control layer is built on two redundant industrial control computers (CPU Inteli7-12700K, memory 32GB, hard disk 1TB SSD), running on Ubuntu 20.04 LTS. It uses Docker containerization to deploy dynamic load balancing algorithms (such as particle swarm optimization) and fault self-healing decision algorithms (such as fuzzy neural networks). It supports parallel processing, can complete more than 1000 grid status assessments per second, and has a command generation latency of ≤300ms. It has a built-in algorithm parameter self-optimization module that can dynamically adjust parameters such as inertia weight and learning factor according to grid operating conditions (such as fluctuations in new energy output and load changes).

[0036] Among them, the execution layer construction specifically includes three types of intelligent actuators, all of which support the Modbus-RTU protocol (baud rate 9600bps, data bits 8 bits): (1) Switching equipment (vacuum circuit breaker ZW32-12 / 630A, opening time ≤60ms, closing time ≤80ms); (2) Voltage regulation / reactive power equipment (SVG static var generator SVG-1000kvar, response time ≤20ms, compensation accuracy ±1kvar); (3) Energy storage / power supply equipment (lithium battery energy storage system LFP-ESS-2000kWh, charging and discharging efficiency ≥92%, response time ≤200ms); The actuator adopts the "main and backup linkage" design, and the key equipment (such as the tie circuit breaker) is equipped with backup equipment, which automatically switches in case of failure, and the switching time is ≤1s.

[0037] Furthermore, to ensure the secure operation of the digital twin converged architecture, a support layer is also set up to provide network security protection and redundant communication network assurance. Specifically, this includes: a security protection system: deploying industrial firewalls (e.g., Hillstone Networks SG-6000), an intrusion detection system (IDS), using AES-256 encryption algorithm to encrypt transmitted data, and assigning operation permissions based on role-based access control (RBAC); and a communication network system: using the MQTT V3.1.1 protocol to achieve cross-layer data communication, with a gigabit fiber optic backbone network and 4G / 5G redundant backup for branch networks, achieving a network packet loss rate ≤0.1% and latency ≤50ms.

[0038] S2, a collaborative digital twin fusion architecture, enables dynamic load balancing of the power grid.

[0039] Specifically, the system utilizes various types of sensors deployed in the perception layer to collect power grid operating status parameters, such as electrical quantities (voltage, current, and electrical coupling indicators), temperature, and status parameters; and simultaneously accesses external data such as power grid topology data (node ​​connection relationships, line impedance parameters), environmental data (meteorological data), and user electricity consumption records (industrial / residential load characteristics).

[0040] The data fusion layer performs noise reduction and standardization on the collected data; for example, Kalman filtering is used to remove sensor noise, and spatiotemporal alignment technology is used to unify the timestamps of multi-source data to build a standardized data set, such as the data set of the operating status of each node of the power grid, such as the load rate and power flow distribution, to support the subsequent call of the digital twin model layer.

[0041] The digital twin model layer is used to construct a three-dimensional twin model that is synchronously mapped to the physical power grid. This includes: constructing a three-dimensional twin model (such as building a visualized power grid topology based on BIM+GIS, dynamically updating switch status and line connection relationships), acquiring data processed by the data fusion layer, and synchronously mapping it to the digital twin model layer; and implementing electrical simulation based on the three-dimensional twin model (such as embedding a Newton-Raphson power flow calculation module with an error rate ≤3%) to support the digital twin model layer in performing electrical simulation operations at various nodes in the power grid.

[0042] The decision control layer monitors the power grid operation status parameters output by the digital twin model layer. When a load factor exceeding a preset threshold is detected, a dynamic load balancing algorithm based on particle swarm optimization is initiated. This algorithm aims to minimize the variance of the overall network load factor, comprehensively considering constraints such as line power, transformer capacity, and node voltage, to calculate the optimal load allocation ratio and send it to the execution layer. The specific formula is as follows: In the formula, n is the total number of power grid nodes (n 50), (Actual power of node i / rated power) This represents the average load rate across the entire network.

[0043] Constraints: A 10% margin is reserved for line power; The transformer capacity has a 15% margin; 0.95 The node voltage fluctuation range is ±5% of the rated voltage.

[0044] PSO Algorithm Iterative Process: Particle Velocity Update Particle position update: ;in, Inertial weight (0.4≤ ≤0.9); c1 and c2 are learning factors (both take the value of 2); r1 and r2 are random numbers in the range [0,1]. The optimal position for the individual; This is the globally optimal position.

[0045] The execution layer is used to perform dynamic load balancing, such as controlling the charging and discharging power of energy storage devices, adjusting the output of distributed power sources, and switching interconnection lines. Among these, the charging and discharging power of energy storage devices (response time ≤ 200ms), the adjustment range of distributed power source output (step size ≤ 5% of rated power), and the switching of interconnection lines (action time ≤ 500ms) are used to achieve load balancing.

[0046] S3. A collaborative digital twin fusion architecture enables fault self-healing decision-making algorithms. Based on the existing load balancing twin model, fault detection and self-healing solutions are implemented. Specific steps include: utilizing multiple types of sensors deployed in the perception layer to collect fault signals, such as voltage transient components and current surge data during a fault.

[0047] The data fusion layer utilizes wavelet transform to extract and classify fault features from the acquired fault signals. Specifically, wavelet transform is used to extract transient fault features, and the wavelet coefficient calculation formula is as follows: ;in, scale factor ( b is the translation factor. Let f(t) be the db4 wavelet basis function, and f(t) be the voltage / current time-domain signal at the fault moment. A BP neural network classifier outputs the fault type probability, and the classifier output formula is: ;in, It is the Sigmoid activation function. For neuron connection weights, The input feature vector consists of 8-dimensional features, including the mean and peak values ​​of wavelet coefficients. For bias terms, Outputs the probability of fault type (such as short circuit / grounding / overload).

[0048] A fault simulation model is constructed using a digital twin model layer based on acquired fault feature extraction and classification, and traveling wave ranging algorithms, to achieve fault location. Specifically, a fault simulation sub-model is built based on PSCAD / EMTDC, embedding topology parameters and mapping fault types (such as single-phase grounding, two-phase short circuit, three-phase short circuit, etc.) and transient feature libraries and traveling wave propagation law data for different fault types, providing model support for fault identification and location. The digital twin model layer constructs the fault simulation model based on the traveling wave ranging principle; the traveling wave propagation speed formula is: Where L is the inductance per unit length of the line (H / km), and C is the capacitance per unit length of the line (F / km); the formula for calculating fault distance is: d = v × t / 2, t represents the time difference (s) between the arrival of the fault traveling wave at the sensors at both ends of the line. The reflection and refraction process of the traveling wave at the fault point is simulated, and the theoretical traveling wave waveform and arrival time are output. The fault location is calculated using a traveling wave ranging algorithm, and the calculation results are input into the digital twin topology model to verify whether the fault point is located on the line segment of the locked area.

[0049] The decision control layer utilizes a fuzzy neural network to generate fault isolation commands and load recovery strategies, which are then sent to the execution layer. Specifically, self-healing decision input parameters are constructed, including fault information (type, location, severity, e.g., permanent faults are assigned a value of 0.8, transient faults 0.3) and regional load status (fault area load rate, load rate variance, critical load percentage, regulating resource status (regional energy storage SOC, distributed power output, tie-line transmission capacity), and safety constraints (node ​​voltage upper and lower limits, line power upper limit). These input parameters are transformed into fuzzy linguistic variables using a membership function (triangular membership), followed by rule-based reasoning and defuzzification to output self-healing decision indicators (e.g., fault isolation switch action sequence, load shedding ratio, energy storage charging and discharging power, tie-line power adjustment, etc.). A fuzzy neural network is used to generate the load recovery strategy, with the fuzzy reasoning rule formula as follows: ;in, For membership function, As cluster center, This is the width parameter. The load recovery priority weights are output through a neural network, and the load transfer is ultimately performed according to the weights, achieving fault self-healing.

[0050] Fault isolation and power restoration are performed using the execution layer.

[0051] Using the above method, taking a 110kV distribution network in a certain city as an example, the digital twin fusion architecture is constructed as follows: The perception layer deploys voltage sensors, current sensors, temperature sensors, and smart meters on 10 substations, 50 lines, and 200 distribution transformers in the distribution network to collect real-time operating data at a sampling frequency of 1 second; the data fusion layer uses edge computing nodes to process the collected data, remove abnormal data caused by sensor failures, and convert data of different formats into standardized datasets in JSON format; the digital twin model layer uses the Unity3D engine to build a virtual twin model based on the GIS topology map and equipment parameters of the distribution network, and achieves data synchronization with the physical power grid through the OPC UA protocol, with a synchronization cycle of 2 seconds; the decision control layer uses an industrial control computer equipped with a dynamic load balancing algorithm and a fault self-healing decision algorithm based on Python; the execution layer includes actuators such as circuit breakers, SVG static var generators, and energy storage batteries, and receives control commands through the Modbus protocol.

[0052] Dynamic load balancing is implemented on a 10kV outgoing line (model YJV22-8.7 / 15kV-3×400mm², rated current carrying capacity 630A, corresponding rated power) of a 110kV substation. =10.9kW) Load rate reaches 85% (actual power) =9.27kW), triggering the equilibrium algorithm; PSO algorithm parameter settings: number of particles 50, number of iterations 30. =0.6, c1=c2=2; the solution shows that 20% of the load (1.85kW) of this line needs to be transferred to the 10kV standby line of the adjacent substation (same model as the main line, current load rate 55%); execution layer actions: control the 2MWh / 500kW energy storage battery (model LFP-ESS-2000kWh) to discharge at 1.5kW power to supplement the power gap; adjust the output of two nearby 100kW photovoltaic inverters (model SG125HX) to increase by 0.35kW; close the tie circuit breaker (model VS1-12 / 630A) to transfer the load; after the adjustment, the load rate of the main line drops to 65% (P_i=7.1kW), the load rate of the standby line rises to 64%, the variance of the load rate of the whole network drops from 0.12 to 0.05, the voltage fluctuation is controlled within the range of 0.98U_N-1.02U_N, and the total response time is 1.2s.

[0053] Fault self-healing implementation: A three-phase short-circuit fault occurred on a 10kV line (model JKLYJ-10-240, length 8km). The current sensor (model ACS712-5A) in the sensing layer detected a peak fault current of 12kA (rated current 630A), and the voltage sensor (model JDZ10-10) detected a voltage drop of 0.3U_N. The data fusion layer used wavelet transform (db4 wavelet basis, decomposition layer 3) to extract fault features, and identified it as a three-phase short-circuit fault through a BP neural network classifier (8 neurons in the input layer, 12 in the hidden layer, and 3 in the output layer), with an initial location error ≤1km. The digital twin model layer is based on PS. The CAD / EMTDC simulation platform was used to construct a fault model. Voltage and current waveforms at the time of the fault were input, and the simulation step size was set to 50µs. After iterative calculation, the cable joint at the fault point, 5.2km from the substation, was accurately located (location error ≤50m). The decision control layer generated the following strategy: disconnect the vacuum circuit breakers (model ZW32-12 / 630A, opening time ≤60ms) on both sides of the fault point; restore power according to load importance level (hospital / school > residential > commercial), prioritizing the transfer of 1.2MW of important loads via the 10kV tie line (2MW spare capacity); start three 500kVA diesel generators (model GF-500) to supplement temporary power supply; fault isolation took 0.3s, important load power restoration took 2min, full load restoration took 4.5min, and the power restoration rate reached 99.2%.

[0054] In one specific embodiment, by adaptively clustering and dividing the power grid into regions and optimizing load balancing at multiple scales, dynamic load balancing is performed locally and across regions based on the load conditions of different regions. This improves the accuracy and efficiency of load balancing and achieves more precise dynamic load balancing of the power grid. The method further includes the following steps in achieving dynamic load balancing of the power grid using the collaborative operation digital twin fusion architecture: collecting power grid topology data, electrical coupling data, and environmental and load characteristic data using the perception layer.

[0055] A data fusion layer is used to perform adaptive clustering of regions based on power grid topology data and electrical coupling data to obtain the partitioned regions. Specifically, firstly, the power grid topology data dimension indicators include node degree, shortest path length, and connectivity; the electrical coupling degree indicators include voltage correlation coefficient, power flow transfer coefficient, and absolute value of mutual impedance between nodes; the analytic hierarchy process (AHP) is used, with topology indicator weights m (0.4) and electrical coupling degree indicator weights 1-m (0.6). Secondly, the indicators are normalized, converting the topology and electrical indicators into dimensionless values ​​in the [0,1] interval; the K-means clustering algorithm is improved to partition the regions, aiming to maximize the weighted sum of indicators within the region and minimize the weighted sum of indicators between regions. The initial cluster center is selected as the topology core node (with the highest node degree). During the iteration process, power grid operation constraints are introduced (e.g., maximum regional load ≤ regional power capacity + tie line transmission limit). Finally, for the initially partitioned regions, the "mean electrical coupling degree within the region" and the "mean electrical coupling degree between regions" are calculated. If the difference is ≥0.3, the partition is valid; otherwise, the cluster centers are readjusted.

[0056] Using the digital twin model layer, regional three-dimensional twin models that are synchronously mapped to the physical power grid are constructed based on the division of regions; specifically, based on the above regional division results, regional three-dimensional twin models are constructed, and virtual and real data are synchronized within the region.

[0057] The decision control layer is used to perform local dynamic load balancing optimization. To prevent computational resource strain, several edge computing devices and a cloud layer with abundant computing resources are deployed at the decision control layer. Then, using the edge computing devices pre-designed at the decision control layer, local dynamic load balancing optimization is performed based on the power grid operation status parameters of the sub-region where the edge computing devices are located, output from the monitored digital twin model layer. This includes initiating a particle swarm optimization-based dynamic load balancing algorithm to calculate the optimal load distribution ratio for the local area when the regional load rate exceeds a preset threshold.

[0058] Inter-regional collaborative dynamic load balancing optimization is performed using the decision control layer. The cloud-based computing devices within the decision control layer continue to utilize the monitored digital twin model layer to perform inter-regional collaborative dynamic load balancing optimization based on the power grid operation status parameters output from all sub-regions. This includes triggering a distributed collaborative optimization-based dynamic load balancing algorithm when the regional load rate, even after local optimization, still exceeds a preset threshold for a preset duration, to calculate the optimal load allocation ratio for each region. Specifically, optimization objectives are set to minimize the expected value of the overall network load rate variance and the expected value of tie-line network losses while satisfying the constraints of local loads in each region. This results in the output of tie-line power adjustment commands (e.g., transferring 10MW of power from region 1 to region 2). Each region corrects its local optimization strategy based on these commands to achieve inter-regional power complementarity and calculates the optimal load allocation ratio for each region. The calculated optimal load allocation ratios for each region are then distributed to the edge computing devices in each region. These edge computing devices re-optimize the optimal load allocation ratio for their local regions based on the new load allocation conditions, iteratively calculating until inter-regional collaborative dynamic load balancing and regional dynamic load balancing are achieved.

[0059] In a specific embodiment, to more effectively cope with the complex and ever-changing power grid operating environment and further improve the efficiency and reliability of power grid operation, multi-scale load forecasting and probabilistic power flow analysis are introduced to consider the uncertainties of new energy output fluctuations and load changes, making the load balancing algorithm more accurate. The method also includes the following steps in achieving dynamic load balancing of the power grid using the collaborative digital twin fusion architecture: A data fusion layer is used to perform multi-scale load forecasting based on historical load data, historical renewable energy output data, and meteorological data and user load type proportions from historical environmental and load characteristic data within the region. An improved Transformer model can be employed, with input features including historical load data (15-minute granularity, past 30 days), meteorological data (wind speed, sunshine, temperature), historical renewable energy output data, and user load type proportions. The output is multi-timescale load forecasts for the next 15, 30, and 60 minutes. Furthermore, an attention mechanism can be introduced to strengthen the correlation weights between meteorological data and renewable energy output, and transfer learning can be used to adapt to changes in load type, thereby keeping the forecast error below 4%.

[0060] Using a data fusion layer, multi-scale load prediction errors (normally distributed, mean 0, standard deviation = predicted value × 5%) and renewable energy output fluctuations (parameters fitted based on historical fluctuation data) are treated as random variables. Monte Carlo simulation is used to generate 1000 sets of regional node voltage and line power scenario data (organized according to the dimensions of "node / line - index type - scenario value - scenario probability"). Each scenario is assigned a probability value P, such as voltage value U1 for node 1, voltage value U2 for node 2, scenario type 1, and scenario probability 0.1. Standardization processing and key index extraction of probabilistic power flow output are completed, including: mean, standard deviation, 95% confidence upper limit, and probability of exceeding the limit.

[0061] Using the digital twin model layer, a load prediction submodule that synchronously maps load prediction within the region to multi-scale load prediction is constructed, and an uncertainty submodule that synchronously maps the standardized processing and key indicator extraction of probabilistic power flow output within the region is constructed.

[0062] The edge computing devices, pre-designed using the decision control layer, are used to perform local dynamic load balancing optimization based on standardized processing and key indicators of power grid operation status parameters and probabilistic power flow output within the area where the edge computing devices are located, according to the monitored digital twin model layer output. This includes initiating a dynamic load balancing algorithm fused with the dynamic PSO objective function when the predicted regional load rate exceeds a preset threshold, calculating the optimal load allocation ratio, and sending it to the execution layer. The dynamic load balancing algorithm fused with the dynamic PSO objective function, based on the original dynamic PSO objective function, considers uncertainty, correlates probabilistic power flow, and incorporates a penalty term based on probabilistic indicators into the dynamic PSO objective function, which aims to minimize the local load rate variance. The formula is as follows: In the formula, f is the original objective function. The line power penalty term for normalization is calculated by weighted summation based on the "probability of exceeding the limit + confidence upper limit deviation" in the key indicators; For the node voltage penalty term in the normalized processing, a distinction is made between overvoltage and undervoltage overvoltage design for nodes with voltage over-limit probability. Under the constraints of PSO deterministic upper and lower limits, the feasible region of variables is further optimized, and the power allocation ratio and resource output of the variables are defined, such as: Power adjustment amounts of n regulating resources (energy storage, flexible loads, distributed power sources); constraints Correspondingly, constraints based on probability indices are constructed, including: high-confidence constraints on line power (using a 95% confidence upper limit U). 95 As a constraint upper limit for line power (the mapping relationship with the optimization variables can be obtained through machine learning fitting to ensure that overload does not occur in 95% of scenarios), and a high-confidence constraint for node voltage (using a 5% confidence lower limit L5 and a 95% confidence upper limit U). 95Define the high-probability fluctuation range of voltage (the mapping relationship with x is fitted by a BP neural network to ensure that the range falls completely within the safe range) and load rate fluctuation constraints (the standard deviation of the node load rate reflects the fluctuation risk, the node load rate constraint does not exceed 10% of the mean, and the mapping relationship between the standard deviation and mean of the node i load rate and x is established by multiple linear regression) to solve for the optimal load allocation ratio and resource output. In the iteration stage, particles that violate the constraints are given a high penalty in the objective function to force the particles to converge to the feasible region.

[0063] In a specific embodiment, a dynamic parameter mechanism and feedback loop are introduced to optimize the performance of the PSO algorithm, reduce computational overhead, and improve convergence efficiency. The method further includes: utilizing the decision control layer to monitor the power grid operating status parameters output by the digital twin model layer; when a load rate exceeds a preset threshold, initiating a dynamic load balancing algorithm based on particle swarm optimization; and calculating the optimal load allocation ratio. The step further includes: dynamically adjusting the inertia weight in the particle velocity update; setting that after every N iterations, the particle fitness (e.g., load variance) is calculated; compared to the previous particle fitness calculation, if the current particle fitness increases, the inertia weight is reduced but not less than the minimum inertia weight threshold; if the particle fitness decreases, the inertia weight is increased but not greater than the maximum inertia weight threshold. For example, every 5 iterations, a higher particle fitness indicates closer proximity to the swarm optimum, so the inertia weight is reduced to accelerate convergence; conversely, a lower particle fitness increases the inertia weight to expand the global search. A velocity correction term is added based on the variance change rate, with the formula: Among them, the rate of change of variance Corresponding weight The feedback gain is dynamically adjusted based on the absolute value of the difference between the current rate of change of variance and the rate of change of variance at the previous time step. If the absolute value of the difference between the current rate of change of variance and the rate of change of variance at the previous time step decreases, the feedback gain is increased; otherwise, the feedback gain is decreased.

[0064] In a specific embodiment, to improve the efficiency and reliability of fault self-healing and better adapt to the characteristics of power grids in different regions, this method addresses the shortcomings of traditional fault self-healing methods in handling faults in different regional power grids. It employs a regional scenario classification method to more accurately extract fault features, locate faults, and formulate self-healing strategies. The collaborative digital twin fusion architecture described in this method, which enables the fault self-healing decision-making algorithm, further includes: The data fusion layer determines the regional scenario classification of the divided regions based on the topology parameters, environmental and load characteristic data of the divided regions. The regional scenario classification is preset based on the topology parameters, environmental and load characteristic data of the divided regions. Examples include: high new energy penetration regional scenarios (PV / wind power fluctuation ±20%), industrial heavy load regional scenarios (high power motor load ratio ≥40%), residential light load regional scenarios (multiple dispersed loads, random load fluctuations), thunderstorm weather regional scenarios, power grid maintenance regional scenarios, and critical load regional scenarios (preset critical type load ratio ≥60%) and their combinations.

[0065] A data fusion layer employs a differentiated fault feature extraction and classification method based on regional scenarios to replace wavelet transform for fault feature extraction and classification of collected fault signals. Fault feature extraction includes: for high-energy penetration areas, new differentiated features such as harmonic distortion rate and sequence component amplitude ratio of transient current are added; for heavy-load industrial areas, new differentiated features such as peak surge current and current change rate are added; for thunderstorm areas, new differentiated features such as transient energy entropy and waveform similarity are added. Fault classification is achieved by using a CNN-LSTM model for high-energy and normal weather combinations, and a gradient boosting tree model for heavy-load industrial and any weather scenarios.

[0066] A fault simulation model is constructed using a traveling wave ranging algorithm based on the acquired fault feature extraction and classification and regional topology adaptation in a digital twin model layer. The ranging parameters are dynamically corrected according to the regional scene classification to achieve fault location. Specifically, different regional scene classifications are set with corresponding traveling wave ranging algorithms and dynamic correction of ranging parameters. For example, single-end traveling wave ranging is adapted to the residential light load scene; multi-end traveling wave ranging and weighted averaging are adapted to the industrial heavy load scene; traveling wave propagation velocity correction is based on the new energy penetration rate calculation for the high new energy penetration scene; and temperature-wave velocity curve correction is adapted to the thunderstorm weather scene.

[0067] The decision control layer uses a fuzzy neural network with an embedded scenario rule base to generate fault isolation instructions and load recovery strategies within the region and send them to the execution layer. Specifically, the scenario rule base is embedded in the fuzzy neural network, such as adding a constraint that the self-healing renewable energy consumption rate is ≥95% in the high renewable energy region scenario.

[0068] In a specific embodiment, to optimize the operation of the power grid and enhance its ability to cope with varying degrees of load changes, the method comprehensively assesses the severity of load overload and selects an appropriate load balancing algorithm to more effectively address different levels of load overload. The method further includes the following steps in achieving dynamic load balancing of the power grid using the collaborative digital twin fusion architecture: The decision control layer monitors the power grid operation status parameters output by the digital twin model layer. Combining the results of whether the load rate exceeds the preset threshold, the duration of the load rate exceeding the preset threshold, and whether there is overload or voltage anomaly, the severity of the load overload is comprehensively judged. When the load rate exceeds the preset threshold, a dynamic load balancing algorithm based on particle swarm optimization, adapted to the severity of the current load overload, or a dynamic load balancing algorithm based on improved particle swarm optimization is selected to replace the dynamic load balancing algorithm based on particle swarm optimization.

[0069] If the severity of the current overload is mild (the load exceeds the preset threshold, the duration of the load rate exceeding the preset threshold does not exceed the preset duration, and there is no overload or overvoltage), then the dynamic load balancing algorithm based on particle swarm optimization is suitable.

[0070] If the current overload severity is classified as moderate overload (load exceeds a preset threshold, load rate exceeds the preset threshold for a duration exceeding a preset threshold, or no overload exists), then an improved particle swarm optimization (PSO) dynamic load balancing algorithm integrating the genetic algorithm (GA) is adopted. Specifically, based on the original algorithm, a global GA exploration is performed beforehand: using the output real-time load rate and power flow data, 50 initial populations (power allocation ratios) are randomly generated. Through roulette wheel selection, single-point crossover (crossover probability 0.8, mutation (mutation probability 0.05) operations, 20 globally optimal solutions are selected as initial PSO particles. Subsequently, PSO local convergence is performed. If, after N iterations of PSO, the load rate variance decrease rate is ≤5% (determined as a local optimum), the GA mutation operation is reintroduced to update the particles, thus escaping the local optimum trap.

[0071] If the current load overload severity is classified as severe overload (load exceeds preset threshold, load rate exceeds preset threshold for duration exceeding preset threshold, and overload or overvoltage occurs), then an improved particle swarm optimization dynamic load balancing algorithm with a safety constraint priority mechanism will be applied. Specifically, based on the original algorithm, in the particle fitness calculation, safety constraints (line power ≤ 90% of rated power, voltage ≥ 0.95Ue) are first checked; if not met, the lowest fitness is directly assigned (excluding it from the optimal solution).

[0072] A specific embodiment differs from the above embodiments in that it adds a collaborative digital twin fusion architecture to achieve mutual auxiliary verification functions for dynamic load balancing and fault self-healing of the power grid. Specifically, it includes the following aspects, and the method further includes: First, by statistically analyzing historical power grid operating status parameters within the region using the decision control layer, the range of normal load rates for the region is determined, and a baseline is generated. After fault location is performed using the digital twin model layer, the deviation between the real-time operating status parameters of the fault-located power grid node and the baseline is compared. If the deviation in the comparison result is greater than the preset deviation (10%), the fault location is verified to be accurate; otherwise, it is determined that there is a false alarm in the fault location, and the fault location is re-performed. For example, power fluctuations (deviation ≤10%) caused by the execution of a load balancing strategy in a certain region can be eliminated by baseline comparison to rule out false fault judgments.

[0073] Second, in the process of constructing a fault simulation model based on the acquired fault feature extraction and classification and traveling wave ranging algorithm using the digital twin model layer to achieve fault location, statistical analysis is performed on fault locations with a probability greater than the preset fault and a fixed fault type. Typical fault cases containing fault features and fault probabilities are established (such as a line short circuit caused by a long-term high load rate in a certain area). In the process of obtaining the optimal load allocation ratio using the decision control layer, the risk of faults exceeding the preset fault probability is verified based on typical fault cases (such as the optimal load ratio causing a line to be in a high load state for a long time (load rate ≥ 80%)). If such a risk exists, the current optimal load allocation ratio is removed and a new optimal load allocation ratio is obtained.

[0074] Third, the decision control layer uses a fuzzy neural network to generate fault isolation commands and load restoration strategies, which are then sent to the execution layer. During the execution layer's fault isolation and power restoration process, a self-healing time limit is determined, and a self-healing time limit classification is established based on the duration. Different self-healing time limit classifications have pre-set self-healing time limit ranges, such as a relaxed level matching the first pre-set range, a normal level matching the second, and an emergency level matching the third. Load balancing coordination strategies are matched according to the self-healing time limit classification, including emergency balancing mode, coordinated balancing mode, and normal balancing mode. Emergency balancing mode refers to suspending all pre-set non-critical grid node balancing adjustments, releasing resources within all regions, and pre-calling the maximum power of neighboring area tie lines (without considering balancing network losses). Coordinated balancing mode refers to retaining balancing resources in half of the regions and scheduling neighboring area resources according to the principle of minimizing balancing network losses. Normal balancing mode refers to performing load balancing normally, with self-healing resource requirements only embedded as constraints. In the next moment, the decision control layer monitors the power grid operation status parameters output by the digital twin model layer. When the load rate exceeds the preset threshold, a decision is made on whether to start the dynamic load balancing algorithm based on particle swarm optimization based on the matching load balancing coordination strategy.

[0075] This application also discloses a computer-readable storage medium.

[0076] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed, such as the above-described method for dynamic load balancing and fault self-healing of power grids based on a digital twin fusion architecture. The computer-readable storage medium includes, for example, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0077] This application also discloses a computer device.

[0078] Specifically, the computer device includes a memory and a processor. The memory stores a computer program that can be loaded and executed by the processor to implement the aforementioned dynamic load balancing and fault self-healing method for power grids based on a digital twin fusion architecture.

[0079] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A power grid dynamic load balancing and fault self-healing method based on a digital twin fusion architecture, characterized in that, The method comprises the following steps: constructing a digital twin fusion architecture, wherein the digital twin fusion architecture comprises a perception layer, a data fusion layer, a digital twin model layer, a decision control layer, and an execution layer; cooperating with the digital twin fusion architecture to realize dynamic load balancing of the power grid, comprising: collecting power grid operation state parameters by using multiple types of sensors deployed in the perception layer; performing denoising processing and standardization conversion on the collected data by using the data fusion layer; constructing a three-dimensional twin model that is virtually and virtually synchronized with the physical power grid by using the digital twin model layer; monitoring the power grid operation state parameters output by the digital twin model layer by using the decision control layer, and when the load rate exceeds the preset threshold, starting a dynamic load balancing algorithm based on particle swarm optimization to calculate the optimal load distribution ratio and sending it to the execution layer; executing dynamic load balancing distribution by using the execution layer; cooperating with the digital twin fusion architecture to realize a fault self-healing decision algorithm, comprising: collecting fault signals by using multiple types of sensors deployed in the perception layer; extracting and classifying fault features from the collected fault signals by using wavelet transform in the data fusion layer; constructing a fault simulation model based on the obtained fault feature extraction and classification and traveling wave distance measurement algorithm to realize fault location in the digital twin model layer; generating fault isolation instructions and load recovery strategies by using fuzzy neural networks in the decision control layer and sending them to the execution layer; executing fault isolation and power supply recovery by using the execution layer.

2. The power grid dynamic load balancing and fault self-healing method based on the digital twin fusion architecture according to claim 1, characterized in that, The steps of cooperating with the digital twin fusion architecture to realize dynamic load balancing of the power grid further comprise: collecting power grid topology data, electrical coupling data, and environmental and load characteristic data by using the perception layer; performing regional adaptive clustering division based on the power grid topology data and the electrical coupling data by using the data fusion layer to obtain divided regions; constructing regional three-dimensional twin models that are virtually and virtually synchronized with the physical power grid based on the divided regions by using the digital twin model layer; previously using the edge computing device designed by the decision control layer, performing local dynamic load balancing optimization according to the monitored power grid operation state parameters in the sub-region where the edge computing device is located, including when the regional load rate exceeds the preset threshold, starting a dynamic load balancing algorithm based on particle swarm optimization to calculate the optimal load distribution ratio of the local region; continuing to use the cloud designed by the decision control layer to perform interval collaborative dynamic load balancing optimization according to the monitored power grid operation state parameters in all sub-regions output by the digital twin model layer, including when the regional load rate after local optimization still exceeds the preset threshold for a preset period of time, starting a dynamic load balancing algorithm based on distributed collaborative optimization to calculate the optimal load distribution ratio of each region; downloading the calculated optimal load distribution ratio of each region to each regional edge computing device, and each regional edge computing device re-optimizes the local optimal load distribution ratio according to the new load distribution condition, and iteratively calculates until the interval collaborative dynamic load balancing and the regional dynamic load balancing are achieved.

3. The power grid dynamic load balancing and fault self-healing method based on the digital twin fusion architecture according to claim 2, characterized in that, The steps of cooperating with the digital twin fusion architecture to realize dynamic load balancing of the power grid further comprise: The data fusion layer is used to perform multi-scale load prediction according to historical load data, historical new energy output data, meteorological data and user load type proportion in historical environmental and load characteristic data in the region; the data fusion layer is used to take the multi-scale load prediction error and new energy output fluctuation as random variables, and a plurality of sets of regional node voltage and line power scene data are generated by using Monte Carlo simulation, each scene is assigned a probability value, and the standardized processing and key index extraction of the probability power flow output are completed; The digital twin model layer is used to respectively construct a load prediction submodule that is synchronously mapped with the multi-scale load prediction in the region, and respectively construct an uncertainty submodule that is synchronously mapped with the standardized processing and key index extraction of the probability power flow output in the region; The edge computing device designed by the decision control layer is used to perform local dynamic load balancing optimization according to the monitored digital twin model layer output regional power grid operation state parameters and the standardized processing and key index of the probability power flow output; including starting a dynamic load balancing algorithm fused with a dynamic PSO objective function when the predicted regional load rate exceeds a preset threshold, calculating an optimal load distribution ratio and sending the optimal load distribution ratio to the execution layer; including: a penalty term based on a probability index is designed and fused into the dynamic PSO objective function with the minimum local load rate variance as the target, the feasible region of the optimization variable includes the power distribution ratio and the adjustment resource output, and a constraint based on the probability index is constructed to obtain the optimal load distribution ratio and the resource output.

4. The power grid dynamic load balancing and fault self-healing method based on digital twin fusion architecture according to claim 1, characterized in that, The step of monitoring the digital twin model layer output power grid operation state parameters by the decision control layer, and starting a dynamic load balancing algorithm based on particle swarm optimization when the load rate exceeds the preset threshold to calculate the optimal load distribution ratio further includes: The inertia weight in the dynamic adjustment of the particle velocity update is adjusted, the particle fitness is calculated every N iterations, compared with the particle fitness calculated last time, if the current particle fitness increases, the inertia weight is reduced and is not less than the minimum inertia weight threshold, and if the particle fitness decreases, the inertia weight is increased and is not greater than the maximum inertia weight threshold; and a speed correction term is added according to the variance change rate: wherein the corresponding weight of the variance change rate is the feedback gain, the corresponding feedback gain is dynamically adjusted according to the absolute value of the difference between the current variance change rate and the previous time variance change rate; if the absolute value of the difference between the current variance change rate and the previous time variance change rate decreases, the feedback gain is increased; otherwise, the feedback gain is decreased.

5. The method of claim 2, wherein, The collaborative operation digital twin fusion architecture, which realizes the fault self-healing decision algorithm, further includes: The data fusion layer is used to determine the regional scene classification of the divided region according to the topological parameters, environmental and load characteristic data of the divided region; the regional scene classification is pre-set according to the topological parameters, environmental and load characteristic data of the divided region; the data fusion layer uses a differentiated fault feature extraction and classification method for the divided regional scene classification to replace the wavelet transform for fault feature extraction and classification of the collected fault signals; A fault simulation model is constructed based on the acquired fault feature extraction and classification and the traveling wave distance measurement algorithm adapted to the partition topology by using the digital twin model layer to realize fault positioning; wherein, the traveling wave distance measurement algorithm adapted to the classification of different regional scenarios is set; The decision control layer adopts a fuzzy neural network embedded with a scene rule library to generate fault isolation instructions and load recovery strategies in the region and sends them to the execution layer.

6. The method of claim 2, wherein, The steps for realizing the dynamic load balancing of the power grid by the synergistically operating digital twin fusion architecture further include: The decision control layer monitors the output power grid operation state parameters of the digital twin model layer, combines whether there is a load rate exceeding a preset threshold, the duration of the load rate exceeding the preset threshold, and whether there is an overload or overvoltage, and comprehensively judges the severity of the load overload; when there is a load rate exceeding the preset threshold, the particle swarm optimization-based dynamic load balancing algorithm or the improved particle swarm optimization-based dynamic load balancing algorithm adapted to the severity of the current load overload is selected to replace the particle swarm optimization-based dynamic load balancing algorithm.

7. The power grid dynamic load balancing and fault self-healing method based on the digital twin fusion architecture according to claim 2, characterized in that, Further including: The synergistically operating digital twin fusion architecture realizes mutual auxiliary verification of the dynamic load balancing and fault self-healing of the power grid; including: The decision control layer statistically analyzes historical regional power grid operation state parameters to determine the range of normal load rate in the region and generate a baseline; after fault positioning by the digital twin model layer, the deviation degree of the real-time operation state parameters of the fault positioning power grid node from the baseline is compared, and if the deviation degree in the comparison result is greater than a preset deviation degree, it is verified that the fault positioning is accurate, otherwise, it is determined that there is a false alarm in the fault positioning to reposition the fault.

8. The power grid dynamic load balancing and fault self-healing method based on the digital twin fusion architecture according to claim 2, characterized in that, Further including: The synergistically operating digital twin fusion architecture realizes mutual auxiliary verification of the dynamic load balancing and fault self-healing of the power grid, further including: In the process of fault positioning by the digital twin model layer based on the acquired fault feature extraction and classification and the traveling wave distance measurement algorithm to construct a fault simulation model, fault positioning with a fault probability greater than a preset fault probability and a fixed fault type is statistically analyzed to establish a typical fault case containing fault features and fault probability; in the process of acquiring the optimal load distribution ratio by the decision control layer, whether there is a fault risk greater than the preset fault probability in executing the optimal load ratio is verified according to the typical fault case; if so, the current optimal load distribution ratio is excluded and the optimal load distribution ratio is reacquired.

9. The power grid dynamic load balancing and fault self-healing method based on the digital twin fusion architecture according to claim 2, characterized in that, Further including: The synergistically operating digital twin fusion architecture realizes mutual auxiliary verification of the dynamic load balancing and fault self-healing of the power grid, further including: In the process of generating fault isolation instructions and load recovery strategies by the decision control layer using a fuzzy neural network and sending them to the execution layer, the self-healing time limit for executing the fault isolation and power supply recovery process is determined, and the self-healing time limit classification is determined according to the self-healing time limit duration, different self-healing time limit classifications are preset with matching preset self-healing time limit duration ranges; the load balancing coordination strategy is matched according to the self-healing time limit classification, including the emergency balancing mode of suspending the balancing adjustment of all preset non-critical power grid nodes, the cooperative balancing mode of reserving half of the balancing resources in the region, and the normal balancing mode of normally executing load balancing; In the next moment, the decision control layer monitors the power grid operation state parameters output by the digital twin model layer, and when there is a load rate exceeding a preset threshold, based on a matching load balancing coordination strategy, it is decided whether to start a dynamic load balancing algorithm based on particle swarm optimization.

10. The method of claim 1, wherein, The digital twin fusion architecture includes a support layer; the support layer provides network security protection and redundant communication network guarantee.

Images