Method and system for constructing digital twinborn model of power distribution network, and method and system for generating toughness improvement measure

Abstract

The invention belongs to the technical field of smart power grids, and particularly relates to a power distribution network digital twinborn model construction and toughness improvement measure generation method and system. The toughness improvement measure generation method comprises the following steps: constructing and training a power distribution network toughness decision model based on a standardized multi-source data set and a power distribution network digital twin model; determining an estimated value of the node evaluation moment operation data under the to-be-predicted meteorological and hydrological data by using the digital twin model of the power distribution network; inputting the estimated value and the hydrological state into a trained power distribution network toughness decision-making model to obtain a corresponding reward value and an optimization decision-making action mapping strategy; and on the basis of to-be-predicted climate and hydrological data, optimizing a corresponding relation of decision action mapping strategies and a vulnerability index, and generating flood toughness improvement measures of'structure reinforcement + operation optimization + emergency deployment 'through an action-measure mapping rule. A digital twinning technology and a reinforcement learning technology are fused and applied to the power distribution network, and a flood scene is combined, so that power distribution network toughness improvement measures can be generated.

Description

Technical Field

[0001] This application belongs to the field of smart grid technology, and in particular relates to the construction of digital twin models of distribution networks, and the generation methods and systems for resilience enhancement measures. Background Technology

[0002] The power distribution network is a crucial link directly connecting the power system to users, and its resilience (i.e., its ability to withstand disruptions and recover) under flood disasters directly impacts social production and daily life. In recent years, extreme flood disasters have occurred frequently, and power outages caused by equipment immersion and line interruptions in the power distribution network have become commonplace. Therefore, accurate assessment of the flood resilience of the power distribution network has become an important requirement for ensuring the safety of the power system.

[0003] Traditional methods for assessing the flood resilience of distribution networks are mostly based on historical data or offline simulations, which cannot reflect the dynamic response of the distribution network in real time during flood evolution, resulting in assessment results that lag behind the actual disaster process. Existing models do not adequately characterize the interaction between the physical entities of the distribution network and the flood environment, making it difficult to simulate equipment failure mechanisms under complex hydrological conditions. Traditional assessment indicators rely heavily on empirical formulas and cannot be dynamically adjusted according to disaster intensity, distribution network topology, etc., resulting in weak universality. Existing methods can only output assessment results and cannot learn autonomously based on assessment data to generate resilience improvement strategies, making it difficult to support decision optimization. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a method and system for constructing a digital twin model of a distribution network and generating resilience enhancement measures. By integrating digital twin technology with reinforcement learning technology in the distribution network, it is possible to generate distribution network resilience enhancement measures and improve the resilience of the distribution network.

[0005] This application provides a method for constructing a digital twin model of a distribution network, including: A standardized multi-source dataset is defined, which includes: meteorological and hydrological data, equipment geographic information data, power distribution network archive data, power distribution network node operation status data, and node operation data; Determine time-series flood disaster scenario data based on meteorological and hydrological data and geographic information data; The distribution network state equation and observation equation are constructed. The input is the static parameters of the distribution network and the node operation data extracted from the standardized multi-source dataset. The solution method is the extended Kalman filter method. The output is the estimated value of the node operation data at the evaluation time. A coupled flood scenario model is constructed based on environmental factors and electrical characteristics, which includes a device failure probability model, an insulation degradation intensity model, and a cable damage risk model. Its input is time-series flood disaster scenario data, and its output is device failure probability index, insulation degradation intensity index, and cable damage risk index. If any index reaches the failure threshold, the switching status data of the device is set to 0. The topology of the distribution network is constructed based on the distribution network archive data and the node operation status data of the distribution network. The topology is updated based on the switching status data of the equipment. A Newton-Raphson agent power flow model is constructed, with the inputs being estimated values ​​of node operating data and updated topology, and the outputs being node time-series power and node time-series voltage. This completes the construction of the digital twin model of the distribution network.

[0006] Furthermore, meteorological and hydrological data include: surface water depth, water flow velocity, sediment density, and water flow temperature.

[0007] Furthermore, a standardized multi-source dataset is identified, including: Collect meteorological, hydrological, geographic information, equipment operating status and operation data associated with the power distribution network, and output a standardized multi-source dataset after spatial alignment, temporal alignment, anomaly detection, missing data imputation and feature normalization.

[0008] This application also provides a method for generating flood resilience enhancement measures for distribution networks. The digital twin model of the distribution network constructed based on the above method includes: A distribution network resilience decision-making model was constructed and trained based on a standardized multi-source dataset and a digital twin model of the distribution network. The distribution network resilience decision-making model adopts a shared coding layer architecture, equipped with a dual-path Critic network, and designs discrete action branches and continuous action branches respectively. Its state space includes: node operation data, hydrological status, and operational status; its action space includes: discrete actions including switching on / off operations and equipment isolation operations, and continuous actions including load peak shaving regulation and distributed power generation output regulation; the reward function is constructed based on load recovery rate, critical user guarantee level, operating cost, safety constraint penalty, and resilience improvement reward. Using a digital twin model of the power distribution network, the estimated values ​​of the operational data at the node assessment time under the meteorological and hydrological data to be predicted are determined; The estimated values ​​of the node's operating data at the assessment time and the hydrological status are input into the trained distribution network resilience decision model to obtain the corresponding reward value and the optimization decision action mapping strategy. Node vulnerability indicators are determined based on optimized decision-making actions; Based on the predicted climate and hydrological data, the correspondence between optimized decision-making action mapping strategies and the identified vulnerability indicators of nodes, flood resilience enhancement measures of "structural reinforcement + operational optimization + emergency deployment" are generated through action-measure mapping rules.

[0009] Furthermore, a graph convolutional neural network and an adjacency attention mechanism are used to encode the current state of nodes, and node-level and system-level features are extracted as inputs to the distribution network resilience decision model.

[0010] Furthermore, the reward function is:

[0011] in, This represents the reward value at time t. This represents the load recovery rate at time t. This indicates the critical user protection level at time t. This represents the operating cost at time t. This represents the safety constraint penalty at time t. This represents the resilience assessment index at time t. , , , and It is the weighting coefficient.

[0012] Furthermore, when training the distribution network resilience decision model, the following method is used to generate a time-series flood disaster scenario dataset: Based on meteorological and hydrological data, time-series flood scenario samples were obtained by using extreme value theory tail distribution and inverse transformation sampling. A temporal generative adversarial network was trained using temporal flood scene samples to obtain a conditional generative model; A time-series flood disaster scenario dataset is generated using a conditional generation model.

[0013] Furthermore, the resilience enhancement reward is determined based on the predicted resilience index of the current decision-making action and the flood scenario resilience index.

[0014] Furthermore, the predictive resilience index of current decision-making actions and the flood scenario resilience index are determined in the following ways: Based on the climate and hydrological data to be predicted, the time-series flood scenario is obtained by using extreme value theory tail distribution and inverse transformation sampling. The topology of the distribution network is updated based on the time-series flood scenario and the current decision action; The updated distribution network topology and its corresponding operational data at the evaluation time are input into the proxy power flow model to obtain the node time-series power of the current decision action. By inputting the distribution network topology of the flood scenario and its corresponding operational data at the assessment time into the proxy power flow model, the node time-series power of the flood scenario can be obtained. The weighted integral of the node temporal power based on the current decision action with the evaluation time is used as the predictive resilience index of the current decision action, and the weighted integral of the node temporal power in the flood scenario with the evaluation time is used as the predictive resilience index of the current decision action.

[0015] Furthermore, based on optimized decision-making actions, vulnerability indicators for nodes are determined, including: Extract the electrical rating parameters of the equipment from a multi-source dataset; Based on the node timing power and node voltage of the current decision action, determine the voltage vulnerability index, power carrying capacity vulnerability index, and branch power flow coupling vulnerability index. The weighted value of voltage vulnerability index, power carrying capacity vulnerability index, and branch power flow coupling vulnerability index is used as the vulnerability index of the node.

[0016] This application also provides a device for constructing a digital twin model of a distribution network, comprising: The dataset determination unit is used to determine the standardized multi-source dataset, which includes: meteorological and hydrological data, equipment geographic information data, power distribution network archive data, and the node operating status and node operating data of the power distribution network. The scenario data construction unit is used to determine time-series flood disaster scenario data based on meteorological and hydrological data and geographic information data; The distribution network equation construction unit is used to construct the distribution network state equation and observation equation. Its input is the static parameters of the distribution network and the node operation data extracted from the standardized multi-source dataset. The solution method is the extended Kalman filter method, and the output is the estimated value of the node operation data at the evaluation time. The flood scenario coupling model construction unit is used to construct a flood scenario coupling model based on environmental factors and electrical characteristics, which includes a device failure probability model, an insulation degradation intensity model, and a cable damage risk model. Its input is time-series flood disaster scenario data, and its output is device failure probability index, insulation degradation intensity index, and cable damage risk index. If any index reaches the failure threshold, the device's switch status data is set to 0. The topology building unit is used to construct the topology of the distribution network based on the distribution network archive data and the node operation status data of the distribution network. The topology is updated based on the switching status data of the devices. The proxy power flow model construction unit is used to build a Newton-Raphson proxy power flow model. Its inputs are the estimated values ​​of node operating data and the updated topology, and its outputs are the node time-series power and node time-series voltage. This completes the construction of the digital twin model of the distribution network.

[0017] This application also provides a distribution network flood resilience enhancement measure generation system, based on the above-mentioned distribution network digital twin model construction device, including: The decision model construction and training unit is used to construct and train a distribution network resilience decision model based on a standardized multi-source dataset and a distribution network digital twin model. The distribution network resilience decision model adopts a shared coding layer architecture, equipped with a dual-path Critic network, and designs discrete action branches and continuous action branches respectively. Its state space includes: node operating data, hydrological status, and operating status; its action space includes: discrete actions including shut-off / shutdown operations and equipment isolation operations, and continuous actions including load peak shaving and distributed power generation output regulation; the reward function is constructed based on load recovery rate, critical user assurance level, operating cost, safety constraint penalties, and resilience improvement rewards. The operation status estimation unit is used to determine the estimated value of the operation data at the node assessment time under the meteorological and hydrological data to be predicted by using the distribution network digital twin model construction device to construct the distribution network digital twin model. The correspondence determination unit is used to input the estimated values ​​of the node's operating data at the evaluation time and the hydrological status into the trained distribution network resilience decision model to obtain the corresponding reward value and the optimization decision action mapping strategy. The vulnerability index determination unit is used to determine the vulnerability index of nodes based on optimization decision actions. The measure generation unit is used to generate flood resilience enhancement measures, including "structural reinforcement + operational optimization + emergency deployment," based on the correspondence between the climate and hydrological data to be predicted, the optimized decision-making action mapping strategy, and the identified vulnerability indicators of nodes, through action-measure mapping rules.

[0018] Furthermore, the reward function is:

[0019] in, This represents the reward value at time t. This represents the load recovery rate at time t. This indicates the critical user protection level at time t. This represents the operating cost at time t. This represents the safety constraint penalty at time t. This represents the resilience assessment index at time t. , , , and It is the weighting coefficient.

[0020] Furthermore, the decision model construction and training unit is specifically used to obtain time-series flood scenario samples based on meteorological and hydrological data, using extreme value theory tail distribution and inverse transformation sampling when training the distribution network resilience decision model; to train a time-series generative adversarial network using the time-series flood scenario samples to obtain a conditional generative model; and to generate a time-series flood disaster scenario dataset using the conditional generative model.

[0021] This application also provides an electronic device, including: Memory, which stores execution instructions; and The processor executes the instructions stored in memory, causing the processor to perform the methods described above.

[0022] This application also provides a readable storage medium storing execution instructions, which are used to implement the above-described method when executed by a processor.

[0023] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0024] Compared with the prior art, this application has the following advantages: 1) A digital twin model building device for distribution networks, which includes a coupled model of flood scenarios, was constructed. The interaction between the physical entities of the distribution network and the flood environment was fully depicted. The digital twin model of the distribution network, which can be dynamically updated in real time, can simulate the failure mechanism of equipment under complex hydrological conditions, reflect the equipment degradation, topology changes and power flow changes under the influence of floods, and thus more closely reflect the real state of the physical entities of the distribution network under flood scenarios.

[0025] 2) Digital twin technology can construct a virtual mapping model of physical entities, enabling real-time state perception and dynamic simulation; reinforcement learning technology, through interactive iteration with the environment, completes the autonomous optimization of decision-making objectives. The solution in this application integrates the two and embeds them into the distribution network flood scenario, which can combine distribution network vulnerability indicators to autonomously generate optimized decision-making objectives and resilience enhancement measures, effectively improving the flood resistance resilience of the distribution network.

[0026] 3) A method for generating datasets for multiple flood scenarios is presented, which supports effective training of reinforcement learning.

[0027] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1A schematic diagram of a method for assessing the flood resilience of a power distribution network based on digital twins and reinforcement learning, according to an embodiment of this application, is shown. Figure 2 A schematic diagram of a distribution network digital twin model construction device according to an embodiment of this application is shown; Figure 3 A schematic diagram of a power distribution network flood resilience enhancement measure generation system according to an embodiment of this application is shown; Figure 4 A block diagram of an electronic device structure according to an embodiment of this application is shown. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] Figure 1 This paper illustrates a flowchart of a distribution network flood resilience assessment method based on digital twins and reinforcement learning according to an embodiment of this application. Figure 1 As shown in the figure, the flood resilience assessment method for distribution networks based on digital twins and reinforcement learning in this application includes the following steps: S1: Preprocessing of multi-source heterogeneous data.

[0032] This step aims to establish a standardized multi-source dataset consistent with the digital twin, achieving a unified representation of meteorological, hydrological, equipment operating status, geographic information, and operational data. Raw data is collected from heterogeneous sources such as data acquisition and monitoring control systems, synchronous phasor measurement devices, sensors, meteorological, hydrological, geographic information, and archives. This data undergoes spatial alignment, temporal alignment, missing data imputation, anomaly detection, feature engineering, and normalization. Uncertainty is quantified for each data item, resulting in a standardized multi-source dataset for use in steps S2, S3, S4, and S5.

[0033] S11: Data Access Data is acquired from source systems such as data acquisition and monitoring control systems, synchronous phasor measurement devices, sensors, meteorological monitoring platforms, hydrological databases, geographic information systems, and equipment file management systems. Data on meteorological and hydrological conditions, geographic information, file information, equipment operating status, and operational data are obtained, and a unique identifier is registered for each data source.

[0034] The specific data types are as follows: Meteorological and hydrological data: surface water depth Water flow velocity Sediment density Water flow temperature ; Geographic information data: Topographic elevation Latitude and longitude coordinates of power distribution network equipment and meteorological and hydrological monitoring stations Reference coordinate system type ; Archive information data: Topology (One device corresponds to two nodes in the topology), rated parameters Installation height Equipment parameters (Equipment type) , For the impedance between device nodes i and j, IP protection level, T p Protective layer type, For initial dielectric strength, (Electric field strength of the equipment). Equipment status data: Operating status of equipment and nodes such as transformers, lines, and switches. ; Equipment operating data: Equipment node voltage Branch current Active power at nodes Node reactive power ; S12: Spatial Alignment Geographic coordinates from different reference coordinate systems are converted into a unified reference coordinate system, and site measurements are mapped to distribution network nodes to ensure spatial consistency in subsequent simulations and evaluations.

[0035] Equipment geographical location alignment. Based on a unified coordinate system, the original spatial coordinates of power distribution network equipment and meteorological and hydrological monitoring stations are transformed, and the latitude and longitude coordinates of the e-th equipment in reference coordinate system g are obtained. Transform the target reference spatial coordinates in the G coordinate system .

[0036] Meteorological and hydrological data alignment. Water flow velocity with latitude and longitude coordinates. Sediment density Water flow temperature (Temp), climate and hydrological data The inverse distance weighting method is used to map the data to distribution network nodes. If the observation station is located within the range of the distribution network equipment, it is directly mapped to the corresponding distribution network equipment; otherwise, the weights are assigned according to the following formula:

[0037]

[0038] For surface water depth with latitude and longitude ,

[0039] In the formula, The water flow velocity of distribution network equipment e Sediment density Water flow temperature (Temp value) Let be the contribution weight of the o-th observation point to device e. For equipment e With the 0 The spatial distance between observation points (unit: m), n is the total number of neighboring observation points of device e, and p is the distance power parameter (default is 2, which can be adjusted to 3 in areas with complex terrain). For the first o Measured values ​​at each observation point This is the estimated value for the e-th device. The elevation of device e is in meters.

[0040] The remaining data is spatially aligned according to the device node coordinates, and the source data with a unified reference coordinate system is output to S13.

[0041] S13: Time Alignment For data with different sampling frequencies, low-frequency data is aligned to high-frequency data using linear interpolation, and high-frequency data is aligned to low-frequency data using timestamps. The time frequencies include minutes, hours, and days. Spatiotemporally aligned multi-source fusion data with a uniform time step is output to S14.

[0042] S14: Anomaly Detection Establish a three-tiered anomaly detection mechanism: statistical anomaly detection, rule-based threshold detection, and consistency detection, to identify and remove anomalous data. 1. Statistical anomaly detection The Z-score statistical test is performed using the following formula:

[0043] in, For the e-th observation, The mean of the observed data, The standard deviation of the observed data is used. When |ze| > τ is the anomaly detection threshold (τ is 3 by default, but can be adjusted according to the data distribution), it is judged as a statistical anomaly.

[0044] 2. Rule-based threshold detection A threshold range is set based on the device's rated parameters; if the range is exceeded, the rule is deemed abnormal. 3. Multivariate Consistency Detection: For the same distribution network equipment node e, there are at least two nodes i and j, with active power... reactive power Node voltage and branch current Verify the power conservation constraint ( ) and impedance constraints ( ), If the branch impedance is not met, it is judged as a conflict anomaly; check whether the geographic information data and the archive information data are completely consistent.

[0045] For abnormal data, continuous data is replaced with the mean, and discrete data is replaced with the nearest device or the same type of device. The multi-source fusion data after abnormal data processing is output to S15.

[0046] S15: Missing interpolation For data with a single missing point or consecutive missing data with a length ≤ 5 time steps, Gaussian process regression (GPR) is used for imputation, with the following formula:

[0047] in, For the target time point interpolated value, For the target time point With the set of observation times, The kernel function covariance vector, The covariance matrix of the observed data, These are the observed data values; If the length of consecutive missing data exceeds 5 time steps, the average of data from adjacent time periods is used to supplement the missing data, and the multi-source fused data is output to S16.

[0048] S16: Feature Normalization After spatial alignment, temporal alignment, anomaly detection, and missing data imputation are performed on continuous characteristic data such as equipment operation and weather / hydrological data, the Min-Max normalization method is used for non-normally distributed data, and the Z-score normalization method is used for approximately normally distributed data. The data units are unified to [0,1] or [-1,1]. Min-Max normalization: ; Z-score normalization: ; The predicted data are processed using the same calculation standard. The processed data is then output to a standardized multi-source dataset (SMSD), which includes feature data such as meteorological and hydrological data, geographic information, equipment operating status, and operational data, and serves as the input for steps S2, S3, S4, S5, and S6.

[0049] S2: Flood Scene Generation Based on extreme value theory and conditional generation model, flood scenarios covering extreme cases and including secondary disasters are generated. The rationality of the scenarios is ensured through physical consistency verification. The output of the flood scenarios provides diversified and highly reliable simulation inputs for digital twin simulation and reinforcement learning resilient decision-making.

[0050] S21: Tail Distribution Fitting of Extreme Value Theory The meteorological and hydrological data in the Standardized Multi-Source Data Set (SMSD) output in step S1, including data such as surface water depth, water flow velocity, sediment density, and water flow temperature, are sampled using extreme value theory tail distribution, inverse transformation sampling, and time-series flood scenario samples.

[0051] Using the output of step S1 The maximum likelihood estimation method is used to obtain the input. threshold The basic model is fitted using tail distribution fitting based on extreme value theory. We obtain the cumulative distribution function (CDF) of Y.

[0052] in, For surface depth, For water flow velocity, For sediment density data, The water temperature.

[0053] Using uniformly distributed random numbers, the excess value is restored to the original variable X using the inverse CDF formula to generate samples of the target distribution.

[0054] Considering the temporal autocorrelation and cross-correlation of meteorological and hydrological data, a uniform random sequence conforming to Gaussian Copula correlation is generated. For each time t, the time series is generated by edge inverse CDF. The ARMA time autocorrelation model is superimposed to obtain time-series flood scenario sample data, which is then output to S22.

[0055] S22 Conditional Generation Model Construction 1. Constructing a Temporal Generative Adversarial Network (TimeGAN) architecture: Generator: 2-layer LSTM (64 hidden units) + 1-layer fully connected layer (3-dimensional output) + GPD extremum correction (fine-tuning with inverse GPD transform when exceeding the threshold); Discriminator: 2-layer bidirectional LSTM (64 hidden units) + 1-layer fully connected layer (outputs true / false probabilities); Reconstructor: 1-layer LSTM (32 hidden units) + 1-layer fully connected (reconstructing features).

[0056] 2. Model Training Pre-training. The generator is fixed, and random noise is input. Train the reconstructor until the reconstructor outputs the same as... Until the loss is less than 0.01; Alternate training. With the generator fixed, the input from step S21 is the output. The discriminator is trained to minimize its loss; the discriminator is then fixed, and random noise is input. The generator is trained by combining adversarial loss, temporal consistency loss, and extreme value loss until the multivariate correlation coefficient error is less than 5%, and the trained generator is saved.

[0057] Scene generation. Input random noise. Input the trained generator and generate A collection of time-series flood disaster scenarios.

[0058] Where M is the sample size. Let D be the time step, and let D be the dimension of the variable, which is set to 4.

[0059] Will The data is transmitted to S3 for building a digital twin of the distribution network.

[0060] S3: Construction of Digital Twin for Distribution Networks Using the standardized multi-source data from step S1 and the flood scenario data from S2, the model state is initialized, and a digital twin model of the coupled power grid and flood is constructed. This model can visualize the interaction between distribution network equipment, power flow changes, equipment operating status and the flood environment in real time, providing a high-fidelity simulation environment for scenario simulation and decision optimization.

[0061] Specifically, step S2 includes the following steps: S31: Model state initialization.

[0062] Extract the static parameters and operational data of the distribution network at the start of the evaluation from the standardized multi-source dataset SMSD in S1.

[0063] Static parameters are obtained by acquiring parameters that do not change over time, such as the distribution network topology, equipment parameters, installation height, and terrain elevation. Run data to obtain time-series dynamic data such as active power and reactive power of the distribution network nodes at the start time; S32: Equipment Operation Data Estimation The equipment operation data from step S31 is obtained, and the continuous state of the distribution network is dynamically estimated using an extended Kalman filter (EKF). The influence of measurement noise on the state estimation is corrected in real time, and high-precision, confidence-based equipment operation data is output to support subsequent topology updates and power flow calculations.

[0064] Distribution network state equations:

[0065] in, Let k be the system state vector (including node voltages) at time k. The control quantity at time k-1 includes power and switching state. This is the state transition function. This is process noise; Observation equation:

[0066] in, Here is the measurement vector at time k, from SMSD. For the observation function, To observe noise; EKF prediction steps:

[0067]

[0068] in, To predict the state, For the prediction error covariance matrix, State transition function exist Jacobian matrix at the location; EKF calibration steps:

[0069]

[0070]

[0071] in, For Kalman gain, For observation function exist Jacobian matrix at the location, This is the corrected state. To correct the error covariance matrix.

[0072] After extended Kalman filter state estimation, high-confidence distribution network equipment operation data are obtained, which are output to S35 for power flow calculation and to S41 for resilience assessment.

[0073] S33: Coupled Model of Flood Risk Obtain the time-series flood scenario data from step S22, construct a multi-factor dynamic coupling model of "surface water depth, water flow velocity, sediment content, and water flow temperature", quantify the destructive effect of the flood environment on equipment performance, and realize the accurate characterization of the "environment-equipment-power grid" coupling effect by the digital twin model.

[0074] By introducing environmental factors and electrical characteristic correction terms, and adopting the Weibull distribution, a device failure probability model is established:

[0075]

[0076] in, Let be the failure probability of device e at time t. Let t be the water depth at the location of device e at time t (in meters). Let t be the water flow velocity at the location of device e at time t (unit: m / s). Let be the density of sediment at the location of device e at time t (unit: kg / m³). For equipment lifespan, For failure type parameters, The critical water depth corresponding to the IP protection level. The critical water velocity corresponding to the IP protection level, is the mud and sand tolerance threshold corresponding to the IP protection level, and C is the structural integrity coefficient of the protection structure. The installation height of device e.

[0077] By introducing an environmental correction term and adopting a power-law degradation model, an insulation degradation intensity model is established:

[0078]

[0079]

[0080]

[0081] in, The initial dielectric strength, Let be the water depth correction factor for device e at time t. Let be the water velocity correction factor for device e at time t. Let be the temperature correction factor for device e at time t. For runtime, Let be the electric field stress coefficient, taken as 1.52. For time index, The water pressure aging coefficient. The abrasion coefficient is... The activation energy of insulating materials, The installation height of the e-th device.

[0082] when When the insulation breakdown probability is >50%, the degradation can be converted into a breakdown probability by combining the log-normal distribution:

[0083] It is the standard normal distribution function. The standard deviation of dielectric strength is taken as 0.1-0.2. This represents the operating electric field strength of the equipment.

[0084] Cable damage risk model: Cable damage risk

[0085] outer sheath breakage probability

[0086] Insulation breakdown probability

[0087] conductor breakage probability

[0088] Where D is the cable burial depth; T p For protective layer type, armor: T p =1, Non-armored: T p =0.5); For time t device e The vortex-induced vibration coefficient, .

[0089]

[0090] If any of the indicators reaches the failure threshold, the equipment is determined to be damaged, the equipment operation status data is set to 0, and the equipment operation status data is output to S34.

[0091] S34: Topology Event Detection and Update The system acquires equipment operating status data via S33, evaluates discrete topology events related to the operation of switches in the distribution network, and simulates and updates the timing topology.

[0092] 1. Event detection formula:

[0093] in, The operating status of device e at time t (0 = disconnected, 1 = normal). EKF estimates the equipment operating status. This is the threshold for state changes (the default value is 0.5; in binary mode, exceeding this value is considered a state change). 2. Topology update: Based on the status of the equipment, determine the connectivity between distribution network nodes (the same equipment has at least two nodes, and when the equipment's operating status is 0, the operating status of its own nodes and connected nodes is also 0), and update the distribution network topology.

[0094]

[0095] in, The data represents the distribution network topology at time t (rows and columns correspond to nodes, with 1 indicating a direct connection between nodes and 0 indicating a disconnection). The topological data is at time t-1. Let t be the set of topological change events. After the topology is updated, the time-series topology data is output to S35 for power flow calculation.

[0096] S35: Agent Current and High-Precision Current Calibration To balance the timeliness and accuracy of power flow calculation, a hybrid power flow calculation method is adopted, which combines real-time calculation using a proxy power flow model with periodic calibration using a high-precision Newton-Raphson (NR) calibrator.

[0097] The agent model is trained based on a graph neural network, with the input data being the active and reactive power data of nodes S1.

[0098]

[0099] in, Output for the proxy model (including node voltage and device power). For proxy power flow functions, Inject vectors into the active and reactive power of nodes. Topology data; Training and Validation: Training data: 1000 sets of different topologies and operational data; Evaluation metric: Voltage prediction error Convergence criterion: The MAE no longer decreases and remains stable for 100 consecutive rounds after training iterations on the validation set; NR high-precision calibration:

[0100] in, For the current imbalance vector Jacobian matrix, For state vectors, The state correction quantity is iteratively solved until... ( The convergence threshold is set to 10 by default. -6 Output high-precision power flow results ; Error feedback calibration:

[0101] in, The learning rate for the surrogate model parameters (default value is 0.001) is used to minimize the error between the surrogate model's predictions and the NR result through gradient descent. The optimal proxy flow model is obtained.

[0102] Input the distribution network equipment operation data output in step 32 and the topology data output in step S34 into the optimal proxy power flow model, and quickly output the time-series power flow results. , , The time-series power flow results are output to S41 for resilience assessment and S42 for node vulnerability assessment.

[0103] S4: Resilience and Vulnerability Assessment The distribution network simulation operation data based on step S35 also includes the time sequence power flow results, which quantifies the distribution network resilience assessment and node vulnerability, identifies key weak links, and provides a basis for the precise formulation of resilience improvement measures.

[0104] S41 Weighted Resilience Index Calculation Input the time-series power flow data from S35, which is the time-weighted integral of the load in the simulated scenario. A higher weight is given to the initial recovery phase to reflect the contribution of rapid recovery capability to overall resilience.

[0105]

[0106] Wherein, RI is the weighted resilience index, with a value range of [0,1]; For time step, For time weighting function ( (In the early recovery phase, the smaller t is, the greater the weight). Let be the load value of node i at time t after the topology update; The integral is discretized using the trapezoidal rule. RI ≥ 0.8 is considered high toughness, 0.6 < RI < 0.8 is medium toughness, and RI < 0.6 is low toughness.

[0107] Output the toughness assessment results.

[0108] S42 node vulnerability assessment A vulnerability assessment model for distribution network nodes is constructed by inputting the time-series power flow calculation results of S35 and the equipment electrical rated parameters of the standardized multi-source dataset of S16, quantifying the flood resistance vulnerability of each node, and improving the accuracy of vulnerability identification.

[0109]

[0110] Voltage vulnerability index

[0111] Power load vulnerability index, load load vulnerability Power supply vulnerability

[0112] Branch power flow coupling vulnerability index

[0113]

[0114] in, is the vulnerability index of node i (the value ranges from [0,1], and the closer it is to 1, the more vulnerable it is). Let i be the voltage at node i. The rated voltage of node i, These represent the active power and its maximum value for the load at node ij, respectively. These represent the active power and its maximum value of the power source at node ij. Let represent the active and reactive power of node ij. The apparent power of node ij Let be the maximum apparent power of node ij. The number of connected branches of node i; Let i be the set of neighboring nodes of node i; Output node vulnerability indicators and ranking.

[0115] S5: Reinforcement Learning Resilient Decision Model Using time-series standardized climate and hydrological data preprocessed by S1 as the core input features, a structured input that the model can recognize is formed. With discrete and continuous mixed actions as output, a resilient decision-making model that is trained to maximize the reward value and adapt to the flood scenario of the power distribution network is calculated.

[0116] S51 Graph Neural Network State Encoding A graph convolutional neural network (GCN) and an adjacency attention mechanism are used to encode the structured state of the distribution network, extract node-level and system-level features, retain topological correlation information, adapt to the input requirements of reinforcement learning models, and provide comprehensive state support for policy optimization.

[0117] Node running feature vector definition:

[0118] in, For distribution network nodes The running feature vector, Normalized node voltage values Normalized value of node active power This is the normalized value of the node reactive power. This indicates the node's running status. Among them, For node runtime data: The hydrological status of the node.

[0119] Graph convolution operation:

[0120] in, The feature vector encoded for node i Let W be the ReLU activation function, W be the convolution weight matrix, and b be the bias term. Let i be the set of adjacent nodes of node i. This represents the number of adjacent nodes; The adjacency attention mechanism assigns attention weights to the features of adjacent nodes, and the weights are determined by both the importance of the node and the connection strength.

[0121] in, Let be the attention weight of node j to node i. It is a multilayer perceptron. This is the normalized value of the branch impedance between nodes i and j; After encoding, the features of all nodes are averaged and pooled to output a system-level feature vector, which serves as the state input for the reinforcement learning agent.

[0122] S52 Hybrid Motion Space The historical case data from step S1 is used to form a hybrid action space containing discrete and continuous actions, clarifying the action constraints and laying the foundation for subsequent action-measure mapping.

[0123]

[0124] a d Discrete actions, including the opening and closing operations of switches such as tie switches and sectionalizing switches, and equipment isolation operations; a c Continuous regulation, including load peak shaving regulation and distributed power generation output regulation.

[0125] S53 reward function Construct a multi-factor weighted reward function to balance load recovery, critical user protection, operating costs, safety constraints, and historical case matching rewards. Input the equipment load, status, and power flow of S32, S34, and S35 to carry out comprehensive benefit and risk calculations for decision-making actions.

[0126] Reward function:

[0127] Where w1=0.35, w2=0.25, w3=0.2, w4=0.1, and w5=0.1 are weighting coefficients. The definitions of each item are as follows: (1) Load recovery rate

[0128] in, Let represent the normal load (kW) of node i at time t. Let N represent the operating state of node i at time t, and N be the total number of nodes in the distribution network. (2) Key User Protection

[0129] Where K is the set of key user nodes.

[0130] (3) Operating costs:

[0131] in, Cost of switching operation (yuan / time). Cost of compensation for peak load shaving (RMB / kW). The cost of deploying emergency resources (in yuan) has been normalized to [0,1]. (4) Safety constraint penalties:

[0132] in, (The default percentage is 1%, so we take 10). Rated voltage (kV). For voltage tolerance, Let J be the branch current (kA). Let J be the rated current of branch j (kA), and L be the total number of branches.

[0133] (5) Resilience Enhancement Rewards

[0134]

[0135] in, This is a predictive resilience index for the model's current decision action. This represents the actual resilience index under flood conditions.

[0136] S54 Improved SAC Algorithm Input the feature codes of S51 and S52 and the hybrid action space, and use the improved soft Actor-Critic (SAC) algorithm to train the decision model to solve the collaborative optimization of the hybrid action space and achieve a balance between stable optimization and exploration utilization of the strategy.

[0137] 1. Algorithm hyperparameter settings Entropy temperature (Initial value), adjusted through an adaptive formula optimization( ); Discount factor ; Policy Network Learning Rate Critic network learning rate ; Experience replay pool capacity 10 6 ; Batch sampling size: 64.

[0138] 2. Network Structure The policy network has a shared coding layer, discrete action branches, continuous action branches, and adopts a dual Critic network, including two fully connected layers and a one-dimensional output. 3. Policy network output:

[0139]

[0140] in, For discrete actions, For continuous actions, For discrete action branching networks, For the mean and variance network of continuous action branches, For the corresponding branch parameters; Value function update:

[0141]

[0142] in, It is a dual Critic network, and d is the termination status flag (1 = terminated, 0 = not terminated). For the next state, For the next action; 3. Introduce a distribution network security constraint attention mechanism to carry out strategy updates.

[0143] in, =5 represents the attention weight. This mechanism reduces the probability of selecting a default action, representing the normalized value of the voltage or current over-limit amplitude. 4. Target network soft update:

[0144] in, This is the soft update coefficient.

[0145] Input flood scenario climate and hydrological data, output a mapping strategy of "climate and hydrological scenario - optimized decision-making action". Proceed to step S61.

[0146] S6: Generation and Verification of Resilience Enhancement Measures Based on the optimized mapping strategy output by S54 and the node vulnerability ranking by S42, comprehensive resilience enhancement measures of "structural reinforcement + operation optimization + emergency deployment" are generated through action-measure mapping rules. These measures are adapted to different flood levels and secondary disaster scenarios and are then verified by digital twins to output feasible solutions.

[0147] S61 resilience enhancement measures generated 1. Action and Measure Mapping Rules Based on the hybrid action space (S54) output by reinforcement learning, and combining the execution frequency and value contribution of actions, the following mapping rules are formulated to realize the transformation from abstract action sequences to specific measures: 1) Discrete action a d : For switch on / off actions / path switching actions (execution frequency ≥ 5 times / scenario), implement strategies and measures to switch to backup paths in advance; Equipment isolation actions (for nodes with water depth > 0.5m), operational strategies and measures, and proactive isolation of risk areas; Pole and tower operations, structural reinforcement measures, and specialized pole and tower reinforcement in secondary disaster scenarios; 2) Continuous action a c : Load peak shaving regulation (regulation range ≥10%), operation strategies and measures, and pre-peak load shaving; Distributed power generation output regulation, operation strategies and measures, and distributed power generation scheduling; High-value actions (single action reward ≥ 0.8) should be prioritized as core measures, key user recovery actions, and emergency resource deployment.

[0148] 2. Logic for determining the priority of measures The priority of measures is determined by a triple assessment of "flood level + action value + vulnerability level" to ensure that resources are allocated to higher priority measures. First-level priority: High-value actions and corresponding measures under severe flooding, vulnerability index Structural reinforcement measures for nodes ≥0.8; Secondary priority: Operational strategies and measures under moderate flooding, vulnerability index 0.6 ≤ Local modifications to nodes <0.8; Level 3 Priority: Routine inspections under mild flooding conditions, and preventative measures for low-vulnerability nodes.

[0149] 3. Flood Level - Response Adaptation Matrix

[0150] 4. Specific measures 1) Structural reinforcement measures: Equipment lifting: V i The installation height of node equipment with a depth of ≥0.8 meters has been increased to the maximum historical water accumulation depth + 0.5m; Waterproofing upgrade: IP67-rated waterproofing encapsulation for outdoor switches, transformers, and other equipment; Improved drainage capacity: Drainage pumps are added to areas with concentrated water accumulation, increasing drainage capacity by more than 30%. Special reinforcement for secondary disasters: Towers in landslide risk areas (slope > 30°) adopt reinforced foundations, increasing the overturning moment by 50%; anti-galloping devices are installed on lines in high-risk areas of tree fall (vegetation coverage > 40%), and vegetation within 10m of the line corridor is cleared; 2) Operational strategies and measures: Early path switching: The power supply path of critical and vulnerable nodes is switched to the backup path, which is determined by the action sequence output by the strategy model; Pre-peak load shaving: Implement 10%-20% pre-peak shaving for non-critical users to reduce line load rate; Preemptive isolation of high-risk areas: Areas with a depth greater than 0.8m or a high risk of secondary disasters should be isolated in advance to avoid cascading failures. Distributed power dispatch: Adjust the output of distributed power sources (photovoltaic, wind power) in advance and reserve backup capacity; Secondary disaster early warning linkage: Upon receiving a landslide / tree fall warning, the load in the corresponding area is transferred to the backup path in advance to shorten the recovery time after the failure; 3) Emergency resource deployment: Mobile energy storage deployment: Deploy mobile energy storage near key user nodes, with a capacity ≥ 1.2 times that of the key load; Emergency power generation vehicle deployment: Deploy emergency power generation vehicles at the top 5 most vulnerable nodes to ensure rapid power supply after an outage; Pre-positioning of repair teams: Pre-positioning repair teams and materials near high-risk areas for secondary disasters to shorten repair time.

[0151] S62 Digital Twin System Verification Input the comprehensive resilience enhancement measures generated by S61; the full-scenario flood dataset generated by S2; and the vulnerability ranking results of nodes in S42. Values; S54 optimized operation strategy; S3 distribution network digital twin model, including equipment parameters, topology, power flow calculation module, and flood environment coupling module, to carry out scheme verification.

[0152] 1. Validation metrics: Load recovery speed: After implementing the measures, the time to reach a 90% load recovery rate is shortened by ≥30% compared to when no measures were taken; Node risk reduction level: Vulnerability index of critical vulnerable nodes Average reduction ≥40%, equipment failure probability P f,i Reduced by ≥50%; Power flow safety: During the simulation, the time percentage of voltage and current exceeding the limit is ≤0.5%, and no equipment cascading failure occurs; Cost-effectiveness ratio: The input cost for a unit resilience improvement (RI improvement of 0.1) is ≤ 500,000 yuan; Recovery rate formula:

[0153] in, The load recovery increase (must be ≥30%). The average load recovery rate after the implementation of the measures. The average load recovery rate before any measures were implemented; 2. Adjustment Mechanism: If any verification indicator fails to meet the standard, the measures parameters shall be iteratively adjusted according to the following rules, and the data shall be re-entered into the digital twin environment for verification until all indicators meet the judgment criteria.

[0154] Output the verified resilience enhancement measures up to S63.

[0155] S63 can implement toughness enhancement solutions. Integrating the resilience enhancement measures validated by S62, we output a standardized and implementable comprehensive resilience enhancement solution, comprising four core modules: Structural reinforcement plan: Clearly define the equipment lifting list, the scope of waterproofing modifications, the location of drainage facility additions, and construction process requirements; Operational optimization plan: Develop flood warning thresholds (e.g., trigger warning when rainfall intensity ≥ 50 mm / h), advance path switching operation procedures, and pre-peak load shaving allocation plan; Emergency resource deployment plan: clearly define the deployment location, quantity, access method, and start-up procedure of mobile energy storage / emergency power generation vehicles; Dispatch Strategy Manual: This manual details the action sequences (switching operations, load regulation), execution timing, and division of responsibilities under different flood scenarios.

[0156] Based on the above method, this application also provides a distribution network digital twin model construction device corresponding to the above method. Figure 2 A block diagram of a distribution network digital twin model construction device according to an embodiment of this application is shown, such as... Figure 2 As shown, a device for constructing a digital twin model of a power distribution network includes: The dataset determination unit is used to determine the standardized multi-source dataset, which includes: meteorological and hydrological data, equipment geographic information data, power distribution network archive data, power distribution network node operation status data, and node operation data; The scenario data construction unit is used to determine time-series flood disaster scenario data based on meteorological and hydrological data and geographic information data; The distribution network equation construction unit is used to construct the distribution network state equation and observation equation. Its input is the static parameters of the distribution network and the node operation data extracted from the standardized multi-source dataset. The solution method is the extended Kalman filter method, and the output is the estimated value of the node operation data at the evaluation time. The flood scenario coupling model construction unit is used to construct a flood scenario coupling model based on environmental factors and electrical characteristics, which includes a device failure probability model, an insulation degradation intensity model, and a cable damage risk model. Its input is time-series flood disaster scenario data, and its output is device failure probability index, insulation degradation intensity index, and cable damage risk index. If any index reaches the failure threshold, the device's switch status data is set to 0. The topology building unit is used to construct the topology of the distribution network based on the distribution network archive data and the node operation status data of the distribution network. The topology is updated based on the switching status data of the devices. The proxy power flow model construction unit is used to build a Newton-Raphson proxy power flow model. Its inputs are the estimated values ​​of node operating data and the updated topology, and its outputs are the node time-series power and node time-series voltage. This completes the construction of the digital twin model of the distribution network.

[0157] Based on the above method, this application also provides a system for generating flood resilience enhancement measures for power distribution networks, corresponding to the above method. Figure 3 This diagram illustrates a system block diagram for generating flood resilience enhancement measures for a power distribution network according to an embodiment of this application. Figure 3 As shown, a system for generating flood resilience enhancement measures for power distribution networks includes: The decision model construction and training unit is used to construct and train a distribution network resilience decision model based on a standardized multi-source dataset and a distribution network digital twin model. The distribution network resilience decision model adopts a shared coding layer architecture, equipped with a dual-path Critic network, and designs discrete action branches and continuous action branches respectively. Its state space includes: node operating data, hydrological status, and operating status; its action space includes: discrete actions including shut-off / shutdown operations and equipment isolation operations, and continuous actions including load peak shaving and distributed power generation output regulation; the reward function is constructed based on load recovery rate, critical user assurance level, operating cost, safety constraint penalties, and resilience improvement rewards. The operation status estimation unit is used to determine the estimated value of the operation data at the node assessment time under the meteorological and hydrological data to be predicted by using the distribution network digital twin model construction device to construct the distribution network digital twin model. The correspondence determination unit is used to input the estimated values ​​of the node's operating data at the evaluation time and the hydrological status into the trained distribution network resilience decision model to obtain the corresponding reward value and the optimization decision action mapping strategy. The vulnerability index determination unit is used to determine the vulnerability index of nodes based on optimization decision actions. The measure generation unit is used to generate flood resilience enhancement measures, including "structural reinforcement + operational optimization + emergency deployment," based on the correspondence between the climate and hydrological data to be predicted, the optimized decision-making action mapping strategy, and the identified vulnerability indicators of nodes, through action-measure mapping rules.

[0158] Furthermore, the reward function is:

[0159] in, This represents the reward value at time t. This represents the load recovery rate at time t. This indicates the critical user protection level at time t. This represents the operating cost at time t. This represents the safety constraint penalty at time t. This represents the resilience assessment index at time t. , , , and It is the weighting coefficient.

[0160] Furthermore, the decision model construction and training unit is specifically used to obtain time-series flood scenario samples based on meteorological and hydrological data, using extreme value theory tail distribution and inverse transformation sampling when training the distribution network resilience decision model; to train a time-series generative adversarial network using the time-series flood scenario samples to obtain a conditional generative model; and to generate a time-series flood disaster scenario dataset using the conditional generative model.

[0161] Based on the same inventive concept disclosed above, this application also provides an electronic device, the structural block diagram of which is shown below. Figure 4 As shown. An electronic device according to an embodiment of this application includes at least one processor and at least one memory electrically connected to the processor. The memory stores instructions executable by the at least one processor, which, when executed, enable the at least one processor to perform the method described above.

[0162] It should be noted that the electrical connection between the above-mentioned units does not necessarily mean the connection between lines. The indirect connection method can be applied to the embodiments of this application as long as it achieves the purpose of this application.

[0163] Based on the same inventive concept, this application also provides a computer storage medium storing a computer program, which, when executed by a processor, implements the above-described method.

[0164] Based on the same inventive concept, this application also provides a computer program product, including a computer program that implements the above-described method when executed by a processor.

[0165] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for constructing a digital twin model of a power distribution network, characterized in that, include: A standardized multi-source dataset is defined, which includes: meteorological and hydrological data, equipment geographic information data, power distribution network archive data, power distribution network node operation status data, and node operation data; Determine time-series flood disaster scenario data based on meteorological and hydrological data and geographic information data; The distribution network state equation and observation equation are constructed. The input is the static parameters of the distribution network and the node operation data extracted from the standardized multi-source dataset. The solution method is the extended Kalman filter method. The output is the estimated value of the node operation data at the evaluation time. A coupled flood scenario model is constructed based on environmental factors and electrical characteristics, which includes a device failure probability model, an insulation degradation intensity model, and a cable damage risk model. Its input is time-series flood disaster scenario data, and its output is device failure probability index, insulation degradation intensity index, and cable damage risk index. If any index reaches the failure threshold, the switching status data of the device is set to 0. The topology of the distribution network is constructed based on the distribution network archive data and the node operation status data of the distribution network. The topology is updated based on the switching status data of the equipment. A Newton-Raphson agent power flow model is constructed, with the inputs being estimated values ​​of node operating data and updated topology, and the outputs being node time-series power and node time-series voltage. This completes the construction of the digital twin model of the distribution network.

2. The method according to claim 1, characterized in that, The meteorological and hydrological data include: surface water depth, water flow velocity, sediment density, and water temperature.

3. The method according to claim 1, characterized in that, Determine the standardized multi-source dataset, including: Collect meteorological and hydrological information, geographic information, equipment operating status and operation data associated with the power distribution network, and output a standardized multi-source dataset after spatial alignment, temporal alignment, anomaly detection, missing data imputation and feature normalization.

4. A method for generating flood resilience enhancement measures for power distribution networks, characterized in that, The digital twin model of the distribution network constructed based on the method described in claim 1 includes: A distribution network resilience decision-making model was constructed and trained based on a standardized multi-source dataset and a digital twin model of the distribution network. The distribution network resilience decision-making model adopts a shared coding layer architecture, equipped with a dual-path Critic network, and designs discrete action branches and continuous action branches respectively. Its state space includes: node operation data, hydrological status, and operational status; its action space includes: discrete actions including switching on / off operations and equipment isolation operations, and continuous actions including load peak shaving regulation and distributed power generation output regulation; the reward function is constructed based on load recovery rate, critical user guarantee level, operating cost, safety constraint penalty, and resilience improvement reward. Using a digital twin model of the power distribution network, the estimated values ​​of the operational data at the node assessment time under the meteorological and hydrological data to be predicted are determined; The estimated values ​​of the node's operating data at the assessment time and the hydrological status are input into the trained distribution network resilience decision model to obtain the corresponding reward value and the optimization decision action mapping strategy. Node vulnerability indicators are determined based on optimized decision-making actions; Based on the predicted climate and hydrological data, the correspondence between optimized decision-making action mapping strategies and the identified vulnerability indicators of nodes, flood resilience enhancement measures of "structural reinforcement + operational optimization + emergency deployment" are generated through action-measure mapping rules.

5. The method according to claim 4, characterized in that, A graph convolutional neural network and an adjacency attention mechanism are used to encode the current state of nodes, and node-level and system-level features are extracted as inputs to the distribution network resilience decision model.

6. The method according to claim 4, characterized in that, The reward function is: in, This represents the reward value at time t. This represents the load recovery rate at time t. This indicates the critical user protection level at time t. This represents the operating cost at time t. This represents the safety constraint penalty at time t. This represents the resilience assessment index at time t. , , , and It is the weighting coefficient.

7. The method according to claim 4, characterized in that, When training the distribution network resilience decision model, the following method is used to generate a time-series flood disaster scenario dataset: Based on meteorological and hydrological data, time-series flood scenario samples were obtained by using extreme value theory tail distribution and inverse transformation sampling. A temporal generative adversarial network was trained using temporal flood scene samples to obtain a conditional generative model; A time-series flood disaster scenario dataset is generated using a conditional generation model.

8. The method according to claim 4, characterized in that, The resilience enhancement reward is determined based on the predicted resilience index of the current decision-making action and the flood scenario resilience index.

9. The method according to claim 8, characterized in that, The predictive resilience index of current decision-making actions and the flood scenario resilience index are determined in the following ways: Based on the climate and hydrological data to be predicted, the time-series flood scenario is obtained by using extreme value theory tail distribution and inverse transformation sampling. The topology of the distribution network is updated based on the time-series flood scenario and the current decision action; The updated distribution network topology and its corresponding operational data at the evaluation time are input into the proxy power flow model to obtain the node time-series power of the current decision action. By inputting the distribution network topology of the flood scenario and its corresponding operational data at the assessment time into the proxy power flow model, the node time-series power of the flood scenario can be obtained. The weighted integral of the node temporal power based on the current decision action with the evaluation time is used as the predictive resilience index of the current decision action, and the weighted integral of the node temporal power in the flood scenario with the evaluation time is used as the predictive resilience index of the current decision action.

10. The method according to claim 9, characterized in that, Vulnerability indicators for nodes are determined based on optimized decision-making actions, including: Extract the electrical rating parameters of the equipment from a multi-source dataset; Based on the node timing power and node voltage of the current decision action, determine the voltage vulnerability index, power carrying capacity vulnerability index, and branch power flow coupling vulnerability index. The weighted value of voltage vulnerability index, power carrying capacity vulnerability index, and branch power flow coupling vulnerability index is used as the vulnerability index of the node.

11. A device for constructing a digital twin model of a power distribution network, characterized in that, include: The dataset determination unit is used to determine the standardized multi-source dataset, which includes: meteorological and hydrological data, equipment geographic information data, power distribution network archive data, power distribution network node operation status data, and node operation data; The scenario data construction unit is used to determine time-series flood disaster scenario data based on meteorological and hydrological data and geographic information data; The distribution network equation construction unit is used to construct the distribution network state equation and observation equation. Its input is the static parameters of the distribution network and the node operation data extracted from the standardized multi-source dataset. The solution method is the extended Kalman filter method, and the output is the estimated value of the node operation data at the evaluation time. The flood scenario coupling model construction unit is used to construct a flood scenario coupling model based on environmental factors and electrical characteristics, which includes a device failure probability model, an insulation degradation intensity model, and a cable damage risk model. Its input is time-series flood disaster scenario data, and its output is device failure probability index, insulation degradation intensity index, and cable damage risk index. If any index reaches the failure threshold, the device's switch status data is set to 0. The topology building unit is used to construct the topology of the distribution network based on the distribution network archive data and the node operation status data of the distribution network. The topology is updated based on the switching status data of the devices. The proxy power flow model construction unit is used to build a Newton-Raphson proxy power flow model. Its inputs are the estimated values ​​of node operating data and the updated topology, and its outputs are the node time-series power and node time-series voltage. This completes the construction of the digital twin model of the distribution network.

12. A system for generating flood resilience enhancement measures for power distribution networks, characterized in that, The distribution network digital twin model construction device according to claim 11 includes: The decision model construction and training unit is used to construct and train a distribution network resilience decision model based on a standardized multi-source dataset and a distribution network digital twin model. The distribution network resilience decision model adopts a shared coding layer architecture, equipped with a dual-path Critic network, and designs discrete action branches and continuous action branches respectively. Its state space includes: node operating data, hydrological status, and operating status; its action space includes: discrete actions including shut-off / shutdown operations and equipment isolation operations, and continuous actions including load peak shaving and distributed power generation output regulation; the reward function is constructed based on load recovery rate, critical user assurance level, operating cost, safety constraint penalties, and resilience improvement rewards. The operation status estimation unit is used to determine the estimated value of the operation data at the node assessment time under the meteorological and hydrological data to be predicted by using the distribution network digital twin model construction device to construct the distribution network digital twin model. The correspondence determination unit is used to input the estimated values ​​of the node's operating data at the evaluation time and the hydrological status into the trained distribution network resilience decision model to obtain the corresponding reward value and the optimization decision action mapping strategy. The vulnerability index determination unit is used to determine the vulnerability index of nodes based on optimization decision actions. The measure generation unit is used to generate flood resilience enhancement measures, including "structural reinforcement + operational optimization + emergency deployment", based on the correspondence between the climate and hydrological data to be predicted, the optimized decision-making action mapping strategy, and the identified vulnerability indicators of nodes, through action-measure mapping rules.

13. The system according to claim 12, characterized in that, The reward function is: in, This represents the reward value at time t. This represents the load recovery rate at time t. This indicates the critical user protection level at time t. This represents the operating cost at time t. This represents the safety constraint penalty at time t. This represents the resilience assessment index at time t. , , , and It is the weighting coefficient.

14. The system according to claim 12, characterized in that, The decision model construction and training unit is specifically used to obtain time-series flood scenario samples based on meteorological and hydrological data, using extreme value theory tail distribution and inverse transformation sampling when training the distribution network resilience decision model; to train a time-series generative adversarial network using the time-series flood scenario samples to obtain a conditional generative model; and to generate a time-series flood disaster scenario dataset using the conditional generative model.

15. An electronic device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes execution instructions stored in the memory, causing the processor to perform the method of any one of claims 1 to 10.

16. A readable storage medium, characterized in that, The readable storage medium stores execution instructions, which, when executed by a processor, are used to implement the method of any one of claims 1 to 10.

17. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 10.

Images