Knowledge-data driven method for assessing disaster-bearing capacity of urban distribution network under rainstorm
By employing a knowledge- and data-driven approach, combining neural networks and Monte Carlo simulations, the disaster resilience of urban power distribution networks under heavy rain is assessed. This addresses the shortcomings of existing assessment methods and enhances the dynamic simulation and prediction performance of the impact of heavy rain and urban flooding on power distribution networks.
Patent Information
- Application Number
- CN202211658648.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-22
- Publication Date
- 2026-02-10
- Estimated Expiration
- 2042-12-22
AI Technical Summary
Existing technologies lack effective methods to assess the impact and dynamic evolution of rainstorms and flooding on urban power distribution networks, resulting in insufficient prediction of power distribution network failures under extreme rainstorm events. Existing assessment methods mainly focus on socio-economic losses and fail to reflect the extent of damage to the power distribution network.
A knowledge- and data-driven approach was adopted, combining feedforward neural networks and gate recurrent unit neural networks to extract static and dynamic variable features, establish a line outage prediction model, evaluate the disaster-bearing capacity of the distribution network under heavy rain through Monte Carlo simulation, and dynamically simulate the flood process using heavy rain intensity, rainfall process, and runoff generation and confluence models.
It improves the predictive performance and interpretability of the distribution network line breakage model, can dynamically simulate flood processes, is suitable for assessing the disaster-bearing capacity of distribution networks, fills the gaps in existing technologies, and is feasible.
Smart Images
Figure CN115907480B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of urban power distribution network, and particularly relates to a knowledge-data driven urban power distribution network disaster-bearing capacity evaluation method under rainstorm. BACKGROUND
[0002] In recent years, frequent extreme natural disasters have caused serious impact on people's daily life and safe operation of power systems. The power outage accidents caused by these extreme events have exposed the weakness of the power system in coping with low-probability and high-risk extreme events. At present, the research on the disaster-bearing capacity evaluation of power distribution network under rainstorm is relatively less at home and abroad. The existing researches are mostly concentrated in the fields of disaster science and urban safety. In the aspect of rainstorm risk evaluation, the existing researches use entropy weight method and TOPSIS method to construct a flood disaster-bearing capacity evaluation model. Some researches propose a standardized flood disaster risk index system for continuously monitoring the hydrological conditions of flood-prone areas and monitoring the potential threat of possible flood events. Considering the uncertainty of flood events, some researches use uncertain methods such as fuzzy mathematics method and grey system method to evaluate the flood disaster risk. The urban rainstorm waterlogging disaster is the result of the comprehensive action of the hazard of disaster-causing factors and the vulnerability of disaster-bearing body. The hazard of disaster-causing factors refers to the disaster intensity and its occurrence probability. The vulnerability refers to the resistance of the disaster-bearing body to the disaster-causing factors. The problems to be solved are specifically the rainstorm waterlogging model that can effectively depict the rainstorm intensity and the evaluation method that reflects the disaster situation of the power distribution network. There is no uniform standard for the establishment of the rainstorm waterlogging model at home and abroad. The existing researches use hydrological model, hydrodynamic model and simplified model. Some researches use the combination of cellular automata model and SWMM model to simulate the urban waterlogging. Some researches propose a flooding model suitable for large urban waterlogging areas by combining GIS technology, which provides technical support for simulating the flooding range and flooding depth of the research area. In the aspect of disaster situation evaluation of power distribution network, the existing researches are concentrated in the fault prediction of power system under typhoon disaster. Data mining method is mostly used to predict the number of power outage users. The existing evaluation methods are mostly used to reflect the social and economic losses and do not reflect the influence of rainstorm waterlogging on the power distribution network and the dynamic evolution process of rainstorm waterlogging. There is no fault prediction method for rainstorm in the field of power distribution network. Therefore, it is very necessary to provide a knowledge-data driven urban power distribution network disaster-bearing capacity evaluation method under rainstorm, which is simple and easy to implement, has good implementability, dynamically simulates the flood process, improves the prediction performance and interpretability. SUMMARY
[0003] The application aims to overcome the deficiencies of the prior art and provide a knowledge-data driven urban power distribution network disaster-bearing capacity evaluation method under rainstorm, which is simple and easy to implement, has good implementability, dynamically simulates the flood process, improves the prediction performance and interpretability.
[0004] The objective of this invention is achieved as follows: a knowledge-data-driven method for assessing the disaster resilience of urban power distribution networks under heavy rain, comprising the following steps:
[0005] Step 1: Based on the historical rainfall data of the selected study area, establish a calculation model for the rainfall intensity, a calculation model for the rainfall process, and a runoff generation and confluence model for the selected study area;
[0006] Step 2: Based on the static and dynamic data of the selected study area, extract the features of static and dynamic variables using feedforward neural networks and gate recurrent unit neural networks respectively, and establish the mapping relationship between dynamic and static variables and the total number of line interruptions in the distribution network of the selected study area to form a line interruption prediction model.
[0007] Step 3: Establish a probability model for rainstorms with different return periods, and based on the probability model, use Monte Carlo simulation to determine whether rainstorms with different return periods will occur in the current simulation using random numbers;
[0008] Step 4: Substitute the return period of the rainstorm in the current simulation into the rainstorm intensity calculation model, the rainfall process calculation model, and the runoff generation and confluence model to obtain the corresponding calculation results. Then, input the calculation results into the line interruption prediction model to obtain the total number of line interruptions corresponding to the current simulation.
[0009] Step 5: Based on the total number of line interruptions obtained from multiple simulations, assess the rainstorm disaster tolerance of the distribution network in the selected study area.
[0010] The specific model for calculating rainfall intensity is as follows: In the formula, A1 is the rainfall force parameter; C is the rainfall force variation parameter; P is the rainfall return period in years; b is the rainfall duration correction parameter; n is the rainfall attenuation parameter; q is the rainfall intensity in mm / min; and t is the rainfall duration in minutes.
[0011] The specific calculation model for the rainfall process is as follows: In the formula, t1 is the pre-peak duration in min; t2 is the post-peak duration in min; q(t1) represents the rainfall intensity at time t1 before the peak in mm / min; q(t2) represents the rainfall intensity at time t2 after the peak in mm / min; the values of parameters A, b, and n are empirical parameters; r∈(0,1) represents the peak rainfall coefficient, indicating the time of occurrence of the rainfall peak. The entire rainfall time series is divided into pre-peak and post-peak time series for separate calculation.
[0012] The specific runoff generation and confluence model is as follows: In the formula, W represents the total water volume of the selected study area; N represents the total number of raster cells in the selected study area after rasterization; E w E represents the elevation of the water surface in the selected waterlogged area of the study region. g (i) represents the ground elevation; i represents the i-th grid after the water accumulation area of the selected study area is rasterized; Δσ is the area element of the water accumulation area of the selected study area.
[0013] E, the elevation of the water surface in the selected study area of the runoff confluence model. w Calculate using the following steps:
[0014] Define E w The range of values for is E min E max E min The value of E is less than or equal to the minimum ground elevation of the waterlogged area under study. max Greater than or equal to the sum of the maximum ground elevation and the height of rainfall during that period;
[0015] Setting E w The initial value is E w =E a =(E min +E max ) / 2;
[0016] For the i-th cell, if E w >E g (i), W increases [E] w -E g (i)]Δσ; otherwise, the value of W remains unchanged;
[0017] After all grid cells have been calculated, update the total water accumulation W.
[0018] Compare the total water volume W with the surface runoff Q. If the allowable error is met, then the E used in this calculation is... w The initial value is used as the final water level elevation E. w If the condition is not met, determine the relationship between the total water volume W and the surface runoff Q. If Q > W, update E. min =E a Otherwise update E max =E a and return to setting E w The initial value calculation continues.
[0019] Based on the static and dynamic data of the selected study area, features of static and dynamic variables are extracted using feedforward neural networks and gate recurrent unit neural networks, respectively. A mapping relationship is established between the dynamic and static variables and the total number of line outages in the distribution network of the selected study area, forming a line outage prediction model, including the following steps:
[0020] By using a feedforward neural network and a gate recurrent unit neural network to extract features of static and dynamic variables from the static and dynamic data of the selected study area, and by using a multi-head attention network to fuse all features, a mapping relationship between static and dynamic variables and the total number of line interruptions is established, forming a line interruption prediction model with dynamic and static variables as inputs and response variables as outputs.
[0021] The static data mentioned include terrain type, forest coverage, number of users, and rainfall intensity.
[0022] The dynamic data refers to the rainfall per unit time, specifically the rainfall every five minutes, ten minutes, or twenty minutes.
[0023] Static and dynamic variables constitute the disaster-causing feature vector as input, while the response variable forms the predicted value as output. Static variables consist of stable quantities that remain unchanged during the duration of the rainstorm, while dynamic variables consist of quantities that change over time, represented by the total number of power grid line outages caused by each rainstorm. As a response variable, n i N represents a predictor variable unit, and N represents the total number of predictor units.
[0024] Before the steps of extracting static and dynamic features and fusing all features using a multi-head attention network to establish a mapping relationship between static and dynamic variables and the total number of line outages, the following steps are also included:
[0025] Obtain static and dynamic data from public information platforms or power grid companies, organize the static and dynamic data into historical datasets, remove outliers from the historical datasets, fill missing data through linear interpolation, and apply min-max normalization to all historical datasets.
[0026] The following steps are employed: A feedforward neural network and a gated recurrent unit neural network are used to extract features of static and dynamic variables from static and dynamic data of the selected study area, respectively. A multi-head attention network is then used to fuse all features.
[0027] Features of static variables are extracted from static data of the selected study area using a feedforward neural network. The feedforward neural network consists of stacked linear layers with non-linear activation functions, and a batch normalization layer is added after each linear layer. The non-linear activation function is the LeakyReLU activation function.
[0028] Features of dynamic variables are extracted from dynamic data of the selected study area using a gated recurrent unit neural network. The gated recurrent unit neural network includes multiple hidden layers, with a lost layer added after each hidden layer.
[0029] The extracted features of static variables and dynamic variables are concatenated to form the input data of the multi-head attention network. The extracted features of static variables and dynamic variables are fused. The multi-head attention network consists of multiple representation subspaces, with a corresponding number of h. In each representation subspace, Q, K, and V are matrices formed by static and dynamic features.
[0030] The outputs of different representation subspaces are concatenated to obtain the output of a multi-head attention network, and the output of the multi-head attention network is used as the fusion result of all features.
[0031] In the process of constructing the line outage prediction model, the mean squared error is selected as the loss function to measure the difference between the predicted number of line outages and the actual number of outages. The formula is as follows: In the formula, L is the loss function value; N is the number of data points; y p This is the predicted value for the number of broken lines; y r The true number of line outages is given; the performance of the line outage prediction model is evaluated using two indicators: Mean Square Error (MSE) and Mean Absolute Error (MAE). The formula is as follows: In the formula, n is the number of data points in the dataset; y r y p These represent the actual and predicted values of the power distribution network disconnection data, respectively.
[0032] Based on the total number of line outages obtained from multiple simulations, the rainstorm disaster tolerance capacity of the distribution network in the selected study area is assessed, including the following steps:
[0033] Based on the total number of line interruptions obtained from multiple simulations, the mean and variance of the number of power transmission line interruptions in the distribution network are calculated. Based on the calculated mean and variance of the number of power transmission line interruptions in the distribution network, the rainstorm disaster resistance capacity of the distribution network in the selected study area is assessed.
[0034] A virtual device includes: a first model building module, used to build a rainstorm intensity calculation model, a rainfall process calculation model, and a runoff generation and confluence model for the selected study area based on historical rainfall data of the selected study area;
[0035] The second model building module is used to extract the features of static and dynamic variables based on the static and dynamic data of the selected study area, respectively using feedforward neural networks and gate recursive unit neural networks, and to establish the mapping relationship between dynamic and static variables and the total number of line interruptions in the distribution network of the selected study area, thus forming a line interruption prediction model.
[0036] The third model building module is used to build probability models of rainstorms with different return periods. Based on the probability models, Monte Carlo simulation is used to determine whether rainstorms with different return periods occur in the current simulation using random numbers.
[0037] The calculation module is used to substitute the return period of the rainstorm in the current simulation into the rainstorm intensity calculation model, the rainfall process calculation model and the runoff generation and confluence model to obtain the corresponding calculation results. The calculation results are then input into the line interruption prediction model to obtain the total number of line interruptions corresponding to the current simulation.
[0038] The evaluation module is used to assess the rainstorm disaster tolerance of the distribution network in the selected study area based on the total number of line interruptions obtained from multiple simulations.
[0039] An electronic device includes: at least one processor and a memory, the memory storing computer-executable instructions, the at least one processor executing the computer-executable instructions stored in the memory, causing the at least one processor to perform any of the methods described above.
[0040] A storage medium, which is a computer-readable storage medium, stores computer-executable instructions, which, when executed by a processor, implement any of the methods described above.
[0041] The beneficial effects of this invention are as follows: This invention is a knowledge- and data-driven method for assessing the disaster resilience of urban power distribution networks under heavy rain. In use, this invention takes the disaster resilience of urban power distribution networks under heavy rain as the target verification object, uses data-driven technology as technical support, and establishes a heavy rain flooding model using rainstorm intensity, single-peak rainfall process, and the equal volume method to characterize the dynamic process and static characteristics of heavy rain flooding, providing input data for the data-driven model. By combining expert domain knowledge and data-driven methods, a dynamic-static dual data-driven model is proposed, improving the predictive performance and interpretability of the power distribution network line breakage model. Through Monte Carlo simulation, the actual ability of the power distribution network to withstand a heavy rainstorm is predicted. This invention has the following advantages: First, the proposed heavy rain flooding model is a hydrodynamic model, which can dynamically simulate the flood process, is suitable for the disaster resilience assessment process of power distribution networks, and is simple. First, this invention is easy to implement and feasible. Second, based on the time-domain variation attributes of feature variables, it divides the input features of the distribution network line disconnection prediction model into static and dynamic variables. Static variables consist of stable features that remain unchanged during the duration of the rainstorm, while dynamic variables consist of changing features that change over time. Considering the different time-series characteristics of the two sets of variables, deep features are extracted through a multilayer perceptron model and a long short-term memory network, respectively. Finally, the number of line disconnections is obtained through multi-head attention network mapping. This invention improves the predictive performance and interpretability of the distribution network line disconnection model by combining expert domain knowledge and data-driven methods. Third, this invention realizes the assessment of the disaster-bearing capacity of urban distribution networks under rainstorms, filling a gap in this field. This invention has the advantages of being simple and easy to implement, having good feasibility, dynamically simulating the flood process, and improving predictive performance and interpretability. Attached Figure Description
[0042] Figure 1 This is a schematic diagram of a typical rainfall process in the Chicago rain pattern according to the present invention.
[0043] Figure 2 This is a schematic diagram of dynamic and static data of the present invention.
[0044] Figure 3 This is a schematic diagram of the dual-channel prediction model structure of the present invention.
[0045] Figure 4 This is the overall flowchart of the present invention. Detailed Implementation
[0046] The present invention will now be further described with reference to the accompanying drawings.
[0047] Example 1
[0048] like Figures 1-4 As shown, a knowledge- and data-driven method for assessing the disaster resilience of urban power distribution networks under heavy rainstorms is described, and the method includes the following steps:
[0049] A knowledge- and data-driven method for assessing the disaster resilience of urban power distribution networks under heavy rainstorms includes the following steps:
[0050] Step 1: Based on the historical rainfall data of the selected study area, establish a calculation model for the rainfall intensity, a calculation model for the rainfall process, and a runoff generation and confluence model for the selected study area;
[0051] Step 2: Based on the static and dynamic data of the selected study area, extract the features of static and dynamic variables using feedforward neural networks and gate recurrent unit neural networks respectively, and establish the mapping relationship between dynamic and static variables and the total number of line interruptions in the distribution network of the selected study area to form a line interruption prediction model.
[0052] Step 3: Establish a probability model for rainstorms with different return periods, and based on the probability model, use Monte Carlo simulation to determine whether rainstorms with different return periods will occur in the current simulation using random numbers;
[0053] Step 4: Substitute the return period of the rainstorm in the current simulation into the rainstorm intensity calculation model, the rainfall process calculation model, and the runoff generation and confluence model to obtain the corresponding calculation results. Then, input the calculation results into the line interruption prediction model to obtain the total number of line interruptions corresponding to the current simulation.
[0054] Step 5: Based on the total number of line interruptions obtained from multiple simulations, assess the rainstorm disaster tolerance of the distribution network in the selected study area.
[0055] A virtual device includes: a first model building module, used to build a rainstorm intensity calculation model, a rainfall process calculation model, and a runoff generation and confluence model for the selected study area based on historical rainfall data of the selected study area;
[0056] The second model building module is used to extract the features of static and dynamic variables based on the static and dynamic data of the selected study area, respectively using feedforward neural networks and gate recursive unit neural networks, and to establish the mapping relationship between dynamic and static variables and the total number of line interruptions in the distribution network of the selected study area, thus forming a line interruption prediction model.
[0057] The third model building module is used to build probability models of rainstorms with different return periods. Based on the probability models, Monte Carlo simulation is used to determine whether rainstorms with different return periods occur in the current simulation using random numbers.
[0058] The calculation module is used to substitute the return period of the rainstorm in the current simulation into the rainstorm intensity calculation model, the rainfall process calculation model and the runoff generation and confluence model to obtain the corresponding calculation results. The calculation results are then input into the line interruption prediction model to obtain the total number of line interruptions corresponding to the current simulation.
[0059] The evaluation module is used to assess the rainstorm disaster tolerance of the distribution network in the selected study area based on the total number of line interruptions obtained from multiple simulations.
[0060] An electronic device includes: at least one processor and a memory, the memory storing computer-executable instructions, the at least one processor executing the computer-executable instructions stored in the memory, causing the at least one processor to perform any of the methods described above.
[0061] A storage medium, which is a computer-readable storage medium, stores computer-executable instructions, which, when executed by a processor, implement any of the methods described above.
[0062] This invention presents a knowledge- and data-driven method for assessing the disaster resilience of urban power distribution networks under heavy rain. In practice, this invention uses the disaster resilience of urban power distribution networks under heavy rain as the target verification object, employs data-driven technology as technical support, and establishes a heavy rain-induced flooding model using rainstorm intensity, single-peak rainfall processes, and the equal-volume method. This model characterizes the dynamic process and static characteristics of heavy rain-induced flooding, providing input data for the data-driven model. By combining expert domain knowledge and data-driven methods, a dynamic-static dual data-driven model is proposed, improving the predictive performance and interpretability of the power distribution network line breakage model. Monte Carlo simulation is used to predict the actual ability of the power distribution network to withstand a heavy rainstorm. This invention has the following advantages: First, the proposed heavy rain-induced flooding model is a hydrodynamic model, capable of dynamically simulating flood processes, suitable for assessing the disaster resilience of power distribution networks, and is simple and easy to implement. First, this invention ensures feasibility. Second, based on the time-domain variation attributes of feature variables, the input features of the distribution network line disconnection prediction model are divided into static and dynamic variables. Static variables consist of stable features that remain unchanged during the duration of the rainstorm, while dynamic variables consist of changing features that change over time. Considering the different time-series characteristics of the two sets of variables, deep features are extracted using a multilayer perceptron model and a long short-term memory network, respectively. Finally, the number of line disconnections is obtained through mapping using a multi-head attention network. This invention improves the predictive performance and interpretability of the distribution network line disconnection model by combining expert domain knowledge and data-driven methods. Third, this invention enables the assessment of the disaster-bearing capacity of urban distribution networks under rainstorms, filling a gap in this field. This invention has the advantages of being simple and easy to implement, having good feasibility, dynamically simulating the flood process, and improving predictive performance and interpretability.
[0063] Example 2
[0064] like Figures 1-4 As shown, a knowledge- and data-driven method for assessing the disaster resilience of urban power distribution networks under heavy rainstorms is described, and the method includes the following steps:
[0065] Step 1: Based on the historical rainfall data of the selected study area, establish a calculation model for the rainfall intensity, a calculation model for the rainfall process, and a runoff generation and confluence model for the selected study area;
[0066] Step 2: Based on the static and dynamic data of the selected study area, extract the features of static and dynamic variables using feedforward neural networks and gate recurrent unit neural networks respectively, and establish the mapping relationship between dynamic and static variables and the total number of line interruptions in the distribution network of the selected study area to form a line interruption prediction model.
[0067] Step 3: Establish a probability model for rainstorms with different return periods, and based on the probability model, use Monte Carlo simulation to determine whether rainstorms with different return periods will occur in the current simulation using random numbers;
[0068] Step 4: Substitute the return period of the rainstorm in the current simulation into the rainstorm intensity calculation model, the rainfall process calculation model, and the runoff generation and confluence model to obtain the corresponding calculation results. Then, input the calculation results into the line interruption prediction model to obtain the total number of line interruptions corresponding to the current simulation.
[0069] Step 5: Based on the total number of line interruptions obtained from multiple simulations, assess the rainstorm disaster tolerance of the distribution network in the selected study area.
[0070] The specific model for calculating rainfall intensity is as follows: In the formula, A1 is the rainfall force parameter; C is the rainfall force variation parameter; P is the rainfall return period in years; b is the rainfall duration correction parameter; n is the rainfall attenuation parameter; q is the rainfall intensity in mm / min; and t is the rainfall duration in minutes.
[0071] In this embodiment, the parameters to be determined in the calculation model of rainstorm intensity vary depending on the historical rainfall data of different regions. Taking the historical rainfall data of a city in southern China as an example, the calculation model is as follows:
[0072] The specific calculation model for the rainfall process is as follows: In the formula, t1 is the pre-peak duration in min; t2 is the post-peak duration in min; q(t1) represents the rainfall intensity at time t1 before the peak in mm / min; q(t2) represents the rainfall intensity at time t2 after the peak in mm / min; the values of parameters A, b, and n are empirical parameters; r∈(0,1) represents the peak rainfall coefficient, which indicates the time of occurrence of the peak rainfall. The entire rainfall time series is divided into pre-peak and post-peak time series for calculation. The value of the peak rainfall coefficient r is obtained from local rainfall data statistics. Figure 1 Schematic diagrams of typical rainfall events in Chicago with seven different return periods are provided.
[0073] The specific runoff generation and confluence model is as follows: In the formula, W represents the total water volume of the selected study area; N represents the total number of raster cells in the selected study area after rasterization; E w E represents the elevation of the water surface in the selected waterlogged area of the study region. g (i) represents the ground elevation; i represents the i-th grid after the water accumulation area of the selected study area is rasterized; Δσ is the area element of the water accumulation area of the selected study area.
[0074] E, the elevation of the water surface in the water accumulation area of the selected study region in the runoff generation and confluence model. w Calculate using the following steps:
[0075] Define E w The range of values for is E min E max E min The value of E is less than or equal to the minimum ground elevation of the waterlogged area under study. max Greater than or equal to the sum of the maximum ground elevation and the height of rainfall during that period;
[0076] Setting E w The initial value is E w =E a =(E min +E max ) / 2;
[0077] For the i-th cell, if E w >E g (i), W increases [E] w -E g (i)]Δσ; otherwise, the value of W remains unchanged;
[0078] After all grid cells have been calculated, update the total water accumulation W.
[0079] Compare the total water volume W with the surface runoff Q. If the allowable error is met, then the E used in this calculation is... w The initial value is used as the final water level elevation E. w If the condition is not met, determine the relationship between the total water volume W and the surface runoff Q. If Q > W, update E. min =E a Otherwise update E max =E a and return to setting E w The initial value calculation continues.
[0080] Based on the static and dynamic data of the selected study area, features of static and dynamic variables are extracted using feedforward neural networks and gate recurrent unit neural networks, respectively. A mapping relationship is established between the dynamic and static variables and the total number of line outages in the distribution network of the selected study area, forming a line outage prediction model, including the following steps:
[0081] By using a feedforward neural network and a gate recurrent unit neural network to extract features of static and dynamic variables from the static and dynamic data of the selected study area, and by using a multi-head attention network to fuse all features, a mapping relationship between static and dynamic variables and the total number of line interruptions is established, forming a line interruption prediction model with dynamic and static variables as inputs and response variables as outputs.
[0082] The static data mentioned include terrain type, forest coverage, number of users, and rainfall intensity.
[0083] The dynamic data refers to the rainfall per unit time, specifically the rainfall every five minutes, ten minutes, or twenty minutes.
[0084] Static and dynamic variables constitute the disaster-causing feature vector as input, while the response variable forms the predicted value as output. Static variables consist of stable quantities that remain unchanged during the duration of the rainstorm, while dynamic variables consist of quantities that change over time, represented by the total number of power grid line outages caused by each rainstorm. As a response variable, n i N represents a predictor variable unit, and N represents the total number of predictor units.
[0085] Before the steps of extracting static and dynamic features and fusing all features using a multi-head attention network to establish a mapping relationship between static and dynamic variables and the total number of line outages, the following steps are also included:
[0086] Obtain static and dynamic data from public information platforms or power grid companies, organize the static and dynamic data into historical datasets, remove outliers from the historical datasets, fill missing data through linear interpolation, and apply min-max normalization to all historical datasets.
[0087] The following steps are employed: A feedforward neural network and a gated recurrent unit neural network are used to extract features of static and dynamic variables from static and dynamic data of the selected study area, respectively. A multi-head attention network is then used to fuse all features.
[0088] Features of static variables are extracted from static data of the selected study area using a feedforward neural network. The feedforward neural network consists of stacked linear layers with non-linear activation functions, and a batch normalization layer is added after each linear layer. The non-linear activation function is the LeakyReLU activation function.
[0089] Features of dynamic variables are extracted from dynamic data of the selected study area using a gated recurrent unit neural network. The gated recurrent unit neural network includes multiple hidden layers, with a lost layer added after each hidden layer.
[0090] The extracted features of static variables and dynamic variables are concatenated to form the input data of the multi-head attention network. The extracted features of static variables and dynamic variables are fused. The multi-head attention network consists of multiple representation subspaces, with a corresponding number of h. In each representation subspace, Q, K, and V are matrices formed by static and dynamic features.
[0091] The outputs of different representation subspaces are concatenated to obtain the output of a multi-head attention network, and the output of the multi-head attention network is used as the fusion result of all features.
[0092] In the process of constructing the line outage prediction model, the mean squared error is selected as the loss function to measure the difference between the predicted number of line outages and the actual number of outages. The formula is as follows: In the formula, L is the loss function value; N is the number of data points; y p This is the predicted value for the number of broken lines; y r The true number of line outages is given; the performance of the line outage prediction model is evaluated using two indicators: Mean Square Error (MSE) and Mean Absolute Error (MAE). The formula is as follows: In the formula, n is the number of data points in the dataset; y r y p These represent the actual and predicted values of the power distribution network disconnection data, respectively.
[0093] Based on the total number of line outages obtained from multiple simulations, the rainstorm disaster tolerance capacity of the distribution network in the selected study area is assessed, including the following steps:
[0094] Based on the total number of line interruptions obtained from multiple simulations, the mean and variance of the number of power transmission line interruptions in the distribution network are calculated. Based on the calculated mean and variance of the number of power transmission line interruptions in the distribution network, the rainstorm disaster resistance capacity of the distribution network in the selected study area is assessed.
[0095] A virtual device includes: a first model building module, used to build a rainstorm intensity calculation model, a rainfall process calculation model, and a runoff generation and confluence model for the selected study area based on historical rainfall data of the selected study area;
[0096] The second model building module is used to extract the features of static and dynamic variables based on the static and dynamic data of the selected study area, respectively using feedforward neural networks and gate recursive unit neural networks, and to establish the mapping relationship between dynamic and static variables and the total number of line interruptions in the distribution network of the selected study area, thus forming a line interruption prediction model.
[0097] The third model building module is used to build probability models of rainstorms with different return periods. Based on the probability models, Monte Carlo simulation is used to determine whether rainstorms with different return periods occur in the current simulation using random numbers.
[0098] The calculation module is used to substitute the return period of the rainstorm in the current simulation into the rainstorm intensity calculation model, the rainfall process calculation model and the runoff generation and confluence model to obtain the corresponding calculation results. The calculation results are then input into the line interruption prediction model to obtain the total number of line interruptions corresponding to the current simulation.
[0099] The evaluation module is used to assess the rainstorm disaster tolerance of the distribution network in the selected study area based on the total number of line interruptions obtained from multiple simulations.
[0100] An electronic device includes: at least one processor and a memory, the memory storing computer-executable instructions, the at least one processor executing the computer-executable instructions stored in the memory, causing the at least one processor to perform any of the methods described above.
[0101] A storage medium, which is a computer-readable storage medium, stores computer-executable instructions, which, when executed by a processor, implement any of the methods described above.
[0102] This invention presents a knowledge- and data-driven method for assessing the disaster resilience of urban power distribution networks under heavy rain. In practice, this invention uses the disaster resilience of urban power distribution networks under heavy rain as the target verification object, employs data-driven technology as technical support, and establishes a heavy rain-induced flooding model using rainstorm intensity, single-peak rainfall processes, and the equal-volume method. This model characterizes the dynamic process and static characteristics of heavy rain-induced flooding, providing input data for the data-driven model. By combining expert domain knowledge and data-driven methods, a dynamic-static dual data-driven model is proposed, improving the predictive performance and interpretability of the power distribution network line breakage model. Monte Carlo simulation is used to predict the actual ability of the power distribution network to withstand a heavy rainstorm. This invention has the following advantages: First, the proposed heavy rain-induced flooding model is a hydrodynamic model, capable of dynamically simulating flood processes, suitable for assessing the disaster resilience of power distribution networks, and is simple and easy to implement. First, this invention ensures feasibility. Second, based on the time-domain variation attributes of feature variables, the input features of the distribution network line disconnection prediction model are divided into static and dynamic variables. Static variables consist of stable features that remain unchanged during the duration of the rainstorm, while dynamic variables consist of changing features that change over time. Considering the different time-series characteristics of the two sets of variables, deep features are extracted using a multilayer perceptron model and a long short-term memory network, respectively. Finally, the number of line disconnections is obtained through mapping using a multi-head attention network. This invention improves the predictive performance and interpretability of the distribution network line disconnection model by combining expert domain knowledge and data-driven methods. Third, this invention enables the assessment of the disaster-bearing capacity of urban distribution networks under rainstorms, filling a gap in this field. This invention has the advantages of being simple and easy to implement, having good feasibility, dynamically simulating the flood process, and improving predictive performance and interpretability.
Claims
1. A knowledge- and data-driven method for assessing the disaster resilience of urban power distribution networks under heavy rain, characterized by: The method includes the following steps: Step 1: Based on the historical rainfall data of the selected study area, establish a calculation model for the rainfall intensity, a calculation model for the rainfall process, and a runoff generation and confluence model for the selected study area; Step 2: Based on the static and dynamic data of the selected study area, extract the features of static and dynamic variables using feedforward neural networks and gate recurrent unit neural networks respectively, and establish the mapping relationship between dynamic and static variables and the total number of line interruptions in the distribution network of the selected study area to form a line interruption prediction model; including the following steps: By using a feedforward neural network and a gate recurrent unit neural network to extract features of static and dynamic variables from the static and dynamic data of the selected study area, and by using a multi-head attention network to fuse all features, a mapping relationship between static and dynamic variables and the total number of line interruptions is established, forming a line interruption prediction model with dynamic and static variables as inputs and response variables as outputs. The static data mentioned include terrain type, forest coverage, number of users, and rainfall intensity. The dynamic data mentioned refers to the rainfall every ten minutes; Static and dynamic variables constitute the disaster-causing feature vector as input, while the response variable forms the predicted value as output. Static variables consist of stable quantities that remain unchanged during the duration of the rainstorm, while dynamic variables consist of quantities that change over time, represented by the total number of power grid line outages caused by each rainstorm. As a response variable, n i N represents a predictor variable unit, and N represents the total number of predictor units. Features of static variables are extracted from static data of the selected study area using a feedforward neural network. The feedforward neural network consists of stacked linear layers with non-linear activation functions, and a batch normalization layer is added after each linear layer. The non-linear activation function is the LeakyReLU activation function. Features of dynamic variables are extracted from dynamic data of the selected study area using a gated recurrent unit neural network. The gated recurrent unit neural network includes multiple hidden layers, with a lost layer added after each hidden layer. The extracted features of static variables and dynamic variables are concatenated to form the input data of the multi-head attention network. The extracted features of static variables and dynamic variables are fused. The multi-head attention network consists of multiple representation subspaces, with a corresponding number of h. In each representation subspace, Q, K, and V are matrices formed by static and dynamic features. The outputs of different representation subspaces are concatenated to obtain the output of the multi-head attention network, and the output of the multi-head attention network is used as the fusion result of all features; Step 3: Establish a probability model for rainstorms with different return periods, and based on the probability model, use Monte Carlo simulation to determine whether rainstorms with different return periods will occur in the current simulation using random numbers; Step 4: Substitute the return period of the rainstorm in the current simulation into the rainstorm intensity calculation model, the rainfall process calculation model, and the runoff generation and confluence model to obtain the corresponding calculation results. Then, input the calculation results into the line interruption prediction model to obtain the total number of line interruptions corresponding to the current simulation. Step 5: Based on the total number of line interruptions obtained from multiple simulations, assess the rainstorm disaster tolerance of the distribution network in the selected study area.
2. The knowledge-data-driven disaster resilience assessment method for urban power distribution networks under heavy rain as described in claim 1, characterized in that: The specific model for calculating rainfall intensity is as follows: In the formula, A1 is the rainfall force parameter; C is the rainfall force variation parameter; P is the rainfall return period in years; b is the rainfall duration correction parameter; n is the rainfall attenuation parameter; q is the rainfall intensity in mm / min; and t is the rainfall duration in minutes.
3. The knowledge-data-driven disaster resilience assessment method for urban power distribution networks under heavy rain as described in claim 2, characterized in that: The specific calculation model for the rainfall process is as follows: In the formula, t1 is the pre-peak duration in min; t2 is the post-peak duration in min; q(t1) represents the rainfall intensity at time t1 before the peak in mm / min; q(t2) represents the rainfall intensity at time t2 after the peak in mm / min; the values of parameters A, b, and n are empirical parameters; r∈(0,1) represents the peak rainfall coefficient, indicating the time of occurrence of the rainfall peak. The entire rainfall time series is divided into pre-peak and post-peak time series for separate calculation.
4. The knowledge-data-driven disaster resilience assessment method for urban power distribution networks under heavy rain as described in claim 3, characterized in that: The specific runoff generation and confluence model is as follows: In the formula, W represents the total water volume of the selected study area; N represents the total number of raster cells in the selected study area after rasterization; E w E represents the elevation of the water surface in the selected waterlogged area of the study region. g (i) represents the ground elevation; i represents the i-th grid after the water accumulation area of the selected study area is rasterized; Δσ is the area element of the water accumulation area of the selected study area.
5. The knowledge-data-driven disaster resilience assessment method for urban power distribution networks under heavy rain as described in claim 4, characterized in that: E, the elevation of the water surface in the water accumulation area of the selected study region in the runoff generation and confluence model. w Calculate using the following steps: Define E w The range of values for is E min E max E min The value of E is less than or equal to the minimum ground elevation of the waterlogged area under study. max Greater than or equal to the sum of the maximum ground elevation and the height of rainfall during that period; Setting E w The initial value is E w =E a =(E min +E max ) / 2; For the i-th cell, if E w >E g (i) W increases Conversely, the value of W remains unchanged. After all grid cells have been calculated, update the total water accumulation W. Compare the total water volume W with the surface runoff Q. If the allowable error is met, then the E used in this calculation is... w The initial value is used as the final water level elevation E. w If the condition is not met, determine the relationship between the total water volume W and the surface runoff Q. If Q > W, update E. min =E a Otherwise update E max =E a and return to setting E w The initial value calculation continues.
6. The knowledge-data-driven disaster resilience assessment method for urban power distribution networks under heavy rain as described in claim 1, characterized in that: Before the steps of extracting static and dynamic features and fusing all features using a multi-head attention network to establish a mapping relationship between static and dynamic variables and the total number of line outages, the following steps are also included: Obtain static and dynamic data from public information platforms or power grid companies, organize the static and dynamic data into historical datasets, remove outliers from the historical datasets, fill missing data through linear interpolation, and apply min-max normalization to all historical datasets.
7. The knowledge-data-driven disaster resilience assessment method for urban power distribution networks under heavy rain as described in claim 1, characterized in that: In the process of constructing the line outage prediction model, the mean squared error is selected as the loss function to measure the difference between the predicted number of line outages and the actual number of outages. The formula is as follows: In the formula, L is the loss function value; N is the number of data points; y p This is the predicted value for the number of broken lines; y r The true number of line outages is given; the performance of the line outage prediction model is evaluated using two indicators: Mean Square Error (MSE) and Mean Absolute Error (MAE). The formula is as follows: In the formula, n is the number of data points in the dataset; y r y p These represent the actual and predicted values of the power distribution network disconnection data, respectively.
8. The knowledge-data-driven disaster resilience assessment method for urban power distribution networks under heavy rain as described in claim 1, characterized in that: Based on the total number of line outages obtained from multiple simulations, the rainstorm disaster tolerance capacity of the distribution network in the selected study area is assessed, including the following steps: Based on the total number of line interruptions obtained from multiple simulations, the mean and variance of the number of power transmission line interruptions in the distribution network are calculated. Based on the calculated mean and variance of the number of power transmission line interruptions in the distribution network, the rainstorm disaster resistance capacity of the distribution network in the selected study area is assessed.