Reliability evaluation method of power transmission network based on typhoon track and intensity simulation

Abstract

The application discloses a kind of based on typhoon path and intensity simulation transmission network reliability evaluation method, it is related to transmission network reliability evaluation technical field.The method is with continuous historical typhoon state vector as input, utilizes LSTM to extract path and intensity time evolution characteristics, and is in forward propagation link series connection differentiable physical layer module;Through physical-data feature fusion and joint loss training, realize the integration of typhoon path and intensity rolling simulation;Combining negative binomial distribution with Monte Carlo sampling generates millennium scale virtual typhoon sample, and wind speed, wind pressure extreme load input transmission network reliability evaluation model, obtains line failure probability, outage risk and wind resistance reliability index.The application will typhoon evolution simulation, tower line failure statistics and transmission network reliability calculation be through, can directly obtain line failure probability, system outage risk and wind resistance reliability index, provide basis for coastal transmission network disaster prevention reinforcement and planning evaluation.

Description

Technical Field

[0001] This invention relates to power transmission network reliability assessment technology, and more particularly to a power transmission network reliability assessment method based on typhoon path and intensity simulation. Background Technology

[0002] Coastal power transmission networks are frequently subjected to typhoon damage. Strong wind loads can easily trigger major power outages such as tower collapses and line breaks. Accurately assessing the wind resistance reliability of the transmission network is a core basis for power grid planning, operation, maintenance, and reinforcement. Current reliability assessments heavily rely on extreme typhoon load samples, but historical typhoon records only cover a few decades, resulting in insufficient sample size and large estimation biases for extreme loads, making it difficult to cover low-probability super typhoon scenarios.

[0003] Currently, typhoon simulation schemes are divided into two categories: Pure data-driven LSTM models: trained solely on historical observation data without meteorological dynamic constraints, the generated virtual typhoons are prone to abnormal samples that violate physical laws, such as path jumps, continuous enhancement upon landfall, and mismatch between wind speed and air pressure. Extreme loads are distorted, and the reliability of the evaluation results is low. Pure physical meteorological numerical models: These models simulate typhoons based on a complete set of atmospheric dynamic equations, but the computational cost is extremely high. They cannot generate a large number of typhoon samples spanning thousands of years in batches, and they are difficult to adapt to Monte Carlo reliability calculations for large-scale power transmission networks. In summary, the existing technology has the following shortcomings: the model cannot balance the efficiency of generating massive amounts of samples with the rationality of meteorological dynamics.

[0004] Therefore, there is an urgent need for a typhoon full-process simulation scheme that combines physics and data training to obtain extreme typhoon load samples for power transmission network reliability assessment. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a power grid reliability assessment method based on typhoon path and intensity simulation, which addresses the deficiencies in the existing technology.

[0006] The technical solution adopted by this invention to solve its technical problem is: a power grid reliability assessment method based on typhoon path and intensity simulation, comprising the following steps: 1) Construct a time-series prediction network model for predicting the path and intensity evolution of typhoons; The time-series prediction network model uses a Long Short-Term Memory (LSTM) network as the basic model for extracting continuous temporal evolution features. Its structure includes: Input layer, LSTM backbone network, embedded differentiable physical layer, feature fusion layer, output prediction layer; The model takes the state vector sequence of the typhoon center over k consecutive historical time steps as input; Wherein, at any time t, the state vector s of the typhoon center t Defined as:

[0007] In the formula: Longitude of the typhoon center; The latitude of the typhoon center; This represents the lowest atmospheric pressure at the center of the typhoon (indicating its intensity). This is a land-sea indicator (a static geographic attribute used to help the model identify the intensity attenuation mechanism during the land-sea transition). The output of the LSTM backbone network is: the extracted data features; After gating and updating the state vector sequence, the LSTM unit outputs a data hidden state ht that characterizes the temporal evolution of the typhoon's recent path and intensity. The hidden state is then used to map intermediate physical quantities, including the minimum central pressure, translational velocity, and translational direction. These intermediate physical quantities, along with the current typhoon latitude, radial distance grid, ambient pressure, and air density, are input into an embedded differentiable physical layer. After the LSTM unit calculates the data features, it embeds a differentiable physical layer in the network forward propagation link; the differentiable physical layer includes 4 sub-modules, which are spliced ​​together in series. Submodule 1: Holland radial pressure field solver module; Input: The minimum central pressure of the typhoon predicted by the LSTM backbone network. The current latitude of the typhoon; Output: Radial pressure gradient; Submodule 2, YanMeng gradient wind field solution module; The input is: radial pressure gradient; Output: Theoretical maximum wind speed, in R max At the location, the typhoon's maximum wind speed V max It must be strictly constrained by the physical mapping relationship of the gradient wind, and is no longer treated as an object that the neural network can predict completely freely, thus significantly reducing the model's degrees of freedom: Submodule 3, Cartesian relative vorticity calculation module; Output: Relative vorticity; Submodule 4, Absolute vorticity conservation residual solution module; Output: Global physics constraint scalar residuals (final output of the embedded differentiable physics layer) S2. A differentiable physical driving layer is serially embedded within the forward propagation link of the LSTM network. This differentiable physical driving layer comprises four fully differentiable, sequentially connected sub-modules. The data flow of each sub-module is unidirectionally progressive and follows a gradient chain-like return flow. Specifically, it includes: S21, Radial Pressure Field Reconstruction Submodule: Receives the LSTM-predicted minimum typhoon center pressure Pc and the current typhoon latitude. Using 1010 hPa as a fixed ambient pressure Pn, it calculates the typhoon center pressure difference. It combines the Earth's rotation angular velocity to solve for the Coriolis parameters and the Holland model pressure profile parameters B. Using the Holland analytical expression, it reconstructs the typhoon's radial pressure distribution and obtains the radial pressure gradient at any spatial location by taking the partial derivative with respect to the radial distance. Simultaneously, it solves for the typhoon's maximum wind speed radius Rmax using the Vickery empirical formula. Outputs the radial pressure gradient, Coriolis parameters, and maximum wind speed radius Rmax. S22, YanMeng Wind Field Equilibrium Velocity Inversion Submodule: Receives the radial pressure gradient, Coriolis parameters, and maximum wind speed radius Rmax output from S21. Combined with air density, overall typhoon translational velocity, and angular parameters, it decomposes air velocity into free atmospheric gradient wind speed and surface friction-corrected wind speed based on the boundary layer momentum equation. It derives the tangential and radial gradient wind speeds separately, synthesizing them to obtain the east-west wind speed u and north-south wind speed v in Cartesian coordinates. Based on the gradient wind physical mapping relationship, it solves for the theoretical maximum wind speed Vmax at Rmax, constructs the mapping error between the theoretical wind speed and the network predicted wind speed, realizes the physical hard constraint of the maximum wind speed, and reduces the network prediction degree of freedom. It outputs the two-dimensional wind field components u and v, and the wind speed physical mapping deviation. S23, Relative Vorticity Conversion Submodule: Receives the two-dimensional wind field components u and v output from S22, solves for the relative vorticity ζ corresponding to the typhoon's rotational motion based on the partial differential relation of the local Cartesian coordinate system, and converts the wind field vector characteristics into dynamic vorticity characteristics that govern the typhoon's path deviation; outputs the relative vorticity ζ. S24, Absolute Vortex Conservation Physical Residual Solving Submodule: Receives the relative vorticity ζ output from S23, combines it with the current typhoon latitude and the latitudinal change rate of the geostrophic effect, expands the absolute vortex conservation mass conservation equation under the quasi-geostrophic approximation into an Euler spatiotemporal evolution form, explicitly characterizing the Beta drift path shift effect caused by the typhoon's endogenous circulation; calculates the degree of deviation between the model simulation evolution results and the absolute vortex conservation theory, obtaining the globally differentiable physical residual Rphys; outputs the physical residual Rphys. The embedded differentiable physical layer calculates step by step according to the data flow direction of radial pressure field reconstruction, gradient wind field inversion, relative vorticity conversion, and absolute vorticity conservation physical residual solution, and outputs the theoretical maximum wind speed, two-dimensional wind field components, relative vorticity and global physical residual. The physical-data feature fusion layer integrates data features extracted by LSTM (pure data-driven trend prediction) and global physical constraint scalar residuals output by the differentiable physical layer (meteorological dynamic constraints). The model output layer takes the fused hidden state as input to the fully connected layer and outputs the complete typhoon state vector for the next time step. The model's output is the state increment target vector s for the next time step. t+1 : ; In the formula, and These are the latitude and longitude increments for the next moment; The lowest pressure increment at the center; This represents the maximum wind speed near the center at the next moment. By overlaying historical data with predicted increments, a rolling step simulation of the typhoon's path and intensity is achieved.

[0008] 2) Design the joint loss function; The loss function consists of four weighted sub-losses, as follows: Path regression loss: used to constrain the fit between the typhoon's latitude and longitude trajectory and the observed samples; Pressure intensity regression loss: used to constrain the accuracy of central pressure intensity prediction; Gradient wind speed mapping loss: used to constrain the network output wind speed to satisfy the physical relationship of gradient wind balance; Absolute vorticity conservation physical residual loss: used to force the simulation process to satisfy the atmospheric vorticity conservation law; 3) Training the model; 4) Large-sample random initial scenario generation: The negative binomial distribution is used to probabilistically model the annual frequency of typhoons. The annual frequency sequence of typhoons on a millennial scale is generated by Monte Carlo sampling. Based on historical observation data of the active typhoon sea area in the Northwest Pacific, the initial state vector of typhoons is randomly sampled and used as the initial input for typhoon evolution simulation. 5) After the model training is completed, use the trained model to generate virtual typhoon samples in batches. Using Monte Carlo random sampling and a pre-trained model, a virtual typhoon spanning 1000 years was simulated at once, generating tens of thousands of complete typhoon lifecycle trajectories, which served as input for the power grid wind load.

[0009] S51. Probabilistic Modeling of Typhoon Annual Occurrence Frequency: Considering the distribution characteristic of the annual typhoon frequency variance in the Northwest Pacific being greater than the mean, a negative binomial distribution is used to probabilistically model the typhoon occurrence frequency Ni for each simulated year, and the fitting yields distribution parameters r=1170.93 and p=0.98; 1000 Monte Carlo samplings are performed on the probability distribution to generate a random occurrence frequency matrix of typhoon years spanning 1000 years; Since the variance of the annual frequency of typhoons in the Northwest Pacific is significantly greater than the mean, this scheme uses a negative binomial distribution to represent the total frequency N of typhoons generated in any simulated year i. i Modeling:

[0010] Based on the historical best path dataset, the discrete model parameters are set to [values ​​to be inserted]. r =1170.93, p =0.98. By performing 1000 Monte Carlo samplings on this probability function, a matrix of random annual occurrences spanning 1000 years can be generated.

[0011] S52. Random sampling of initial typhoon state: Based on the annual simulated typhoon frequency, a random repeated sampling method is used to randomly sample the initial state vector s0, which includes initial longitude, latitude, minimum central pressure, and maximum wind speed, from the historical typhoon initial sample library in the active typhoon sea area of ​​the Northwest Pacific Ocean within the range of 5°N~25°N and 110°E~170°E. This vector serves as the initial input for the physics-data dual-driven typhoon evolution model.

[0012] S53. Full lifecycle simulation of typhoon landfall degradation and multi-mechanism random termination determination: Typhoon landfall status is identified by land and sea indicators. The Vickery index model with regional parameterization is used to simulate the typhoon land surface intensity decay. The upper limit of the typhoon life cycle is set based on the Gamma distribution, and the typhoon dissipation pressure threshold is set based on the log-normal distribution. The typhoon rolling simulation is terminated by dual determination, generating a virtual sample of the entire typhoon life cycle that conforms to the real climate characteristics, and providing extreme load input for the reliability assessment of the power grid's wind resistance and disaster prevention.

[0013] 6) Extract the maximum near-surface wind pressure and wind speed extreme loads from each virtual typhoon sequence, and input the extreme loads in batches into the mechanical failure model of regional power transmission towers and overhead lines; Based on Monte Carlo sampling statistics of the failure frequency of transmission system lines and towers under all virtual typhoons, the failure probability of lines, the risk of system power outage, and the wind resistance reliability index of the power grid are calculated, and the typhoon failure probability and wind resistance reliability index of the regional power transmission network are calculated.

[0014] According to the above scheme, the main network of the model adopts a two-layer LSTM structure, with the number of hidden units set to 128 and 64 respectively. The LSTM uses an internal forget gate. Input gate Candidate memory states and output gate Selectively retain and update historical meteorological time-series information; According to the above scheme, the physics-data fusion layer receives the original temporal hidden state ht from the LSTM, the global physical residual Rphys output by the differentiable physics layer, and the gradient wind speed mapping bias. Rphys and the wind speed mapping bias are mapped into physical embedding vectors via a linear layer, and then concatenated or weighted with ht to obtain the corrected hidden state hc. The corrected hidden state hc is input into the fully connected output layer to obtain the incremental predictions of longitude, latitude, central minimum air pressure, and near-center maximum wind speed for the next time step. During the training phase, path regression loss, air pressure intensity loss, gradient wind speed mapping loss, and absolute vorticity conservation residual loss are backpropagated together, ensuring that the LSTM parameters and physics fusion parameters are updated synchronously.

[0015] According to the above scheme, the pathspace regression loss is:

[0016] Pressure intensity regression loss:

[0017] Gradient wind speed theoretical mapping loss:

[0018] Absolute vorticity conservation deviation (key physical loss term):

[0019] Formula with superscript obs All variables are real observations obtained from historical meteorological measurements. By deriving the gradient information of the network parameters from both sample statistical errors and physical dynamic residuals, it is ensured that even in small sample areas with historical extreme winds, the simulated trajectories still possess a high degree of physical consistency.

[0020]

[0021] In the formula, w 1, w 2, w 3, w 4. Weighting coefficients to balance the contributions of each task; L track L press L v and L phys These represent the path space regression loss, air pressure intensity regression loss, gradient wind speed theoretical mapping loss, and absolute vorticity conservation deviation, respectively; N is the number of training samples; λ, φ, and φ are the longitude and latitude of the typhoon center, respectively; P is the minimum central pressure, and V... max These represent the maximum wind speed near the center; the superscript obs indicates historical observations; u and v represent the east-west and north-south wind speed components in Cartesian coordinates, respectively; ζ is the relative vorticity; and β is the local planetary vorticity gradient.

[0022] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described scheme.

[0023] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method in the above-described scheme.

[0024] The beneficial effects of this invention are: This invention serially embeds differentiable physical modules such as Holland pressure field, YanMeng wind field, relative vorticity and absolute vorticity conservation residuals into the LSTM time series prediction network, so that the typhoon path and intensity simulation are simultaneously constrained by historical statistical laws and meteorological dynamics, reducing unreasonable results such as path abrupt changes and wind speed and pressure mismatch that may occur in pure data-driven models. This invention integrates typhoon evolution simulation, near-ground wind field load extraction, tower and line failure statistics, and power transmission network reliability calculation, which can directly obtain line failure probability, system power outage risk, and wind resistance reliability indicators, providing a basis for disaster prevention reinforcement and planning assessment of coastal power transmission networks. Attached Figure Description

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the model in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0027] like Figure 1 As shown, a power transmission network reliability assessment method based on typhoon path and intensity simulation includes the following steps: 1) Construct a time-series prediction network model for predicting the path and intensity evolution of typhoons; The time-series prediction network model uses a Long Short-Term Memory (LSTM) network as the basic model for extracting continuous temporal evolution features. Its structure includes: Input layer, LSTM backbone network, embedded differentiable physical layer, feature fusion layer, output prediction layer; The model takes the state vector sequence of the typhoon center over k consecutive historical time steps as input; Wherein, at any time t, the state vector s of the typhoon center t Defined as:

[0028] In the formula: Longitude of the typhoon center; The latitude of the typhoon center; This represents the lowest atmospheric pressure at the center of the typhoon (indicating its intensity). This is a land-sea indicator (a static geographic attribute used to help the model identify the intensity attenuation mechanism during the land-sea transition). The output of the LSTM backbone network is: the extracted data features; like Figure 2 After the LSTM unit performs gated updates on the state vector sequence, it outputs a data hidden state ht that characterizes the temporal evolution of the typhoon's recent path and intensity. The hidden state is then used to map intermediate physical quantities, including the minimum central pressure, translational velocity, and translational direction. These intermediate physical quantities, along with the current typhoon latitude, radial distance grid, ambient air pressure, and air density, are input into an embedded differentiable physical layer. The LSTM unit calculates the data characteristics and embeds a differentiable physical layer in the network forward propagation link; the differentiable physical layer consists of 4 sub-modules, which are spliced ​​together in series. Submodule 1: Holland radial pressure field solver module; Input: The minimum central pressure of the typhoon predicted by the LSTM backbone network. The current latitude of the typhoon; Output: Radial pressure gradient; Submodule 2, YanMeng gradient wind field solution module; The input is: radial pressure gradient; Output: Theoretical maximum wind speed, in R max At the location, the typhoon's maximum wind speed V max (i.e., in Equation 9) V t+1 The model must be strictly constrained by the physical mapping relationship of the gradient wind and is no longer treated as an object that the neural network can predict completely freely, thus significantly reducing the model's degrees of freedom. Submodule 3, Cartesian relative vorticity calculation module; Output: Relative vorticity; Submodule 4, Solving for the absolute vorticity conservation residual; Output: Global physics constraint scalar residuals (final output of the embedded differentiable physics layer) S2. A differentiable physical driving layer is serially embedded within the forward propagation link of the LSTM network. This differentiable physical driving layer comprises four fully differentiable, sequentially connected sub-modules. The data flow of each sub-module is unidirectionally progressive and follows a gradient chain-like return flow. Specifically, it includes: S21, Radial Pressure Field Reconstruction Submodule: Receives the LSTM-predicted minimum typhoon center pressure Pc and the current typhoon latitude. Using 1010 hPa as a fixed ambient pressure Pn, it calculates the typhoon center pressure difference. It combines the Earth's rotation angular velocity to solve for the Coriolis parameters and the Holland model pressure profile parameters B. Using the Holland analytical expression, it reconstructs the typhoon's radial pressure distribution and obtains the radial pressure gradient at any spatial location by taking the partial derivative with respect to the radial distance. Simultaneously, it solves for the typhoon's maximum wind speed radius Rmax using the Vickery empirical formula. Outputs the radial pressure gradient, Coriolis parameters, and maximum wind speed radius Rmax. S22, YanMeng Wind Field Equilibrium Velocity Inversion Submodule: Receives the radial pressure gradient, Coriolis parameters, and maximum wind speed radius Rmax output from S21. Combined with air density, overall typhoon translational velocity, and angular parameters, it decomposes air velocity into free atmospheric gradient wind speed and surface friction-corrected wind speed based on the boundary layer momentum equation. It derives the tangential and radial gradient wind speeds separately, synthesizing them to obtain the east-west wind speed u and north-south wind speed v in Cartesian coordinates. Based on the gradient wind physical mapping relationship, it solves for the theoretical maximum wind speed Vmax at Rmax, constructs the mapping error between the theoretical wind speed and the network predicted wind speed, realizes the physical hard constraint of the maximum wind speed, and reduces the network prediction degree of freedom. It outputs the two-dimensional wind field components u and v, and the wind speed physical mapping deviation. S23, Relative Vorticity Conversion Submodule: Receives the two-dimensional wind field components u and v output from S22, solves for the relative vorticity ζ corresponding to the typhoon's rotational motion based on the partial differential relation of the local Cartesian coordinate system, and converts the wind field vector characteristics into dynamic vorticity characteristics that govern the typhoon's path deviation; outputs the relative vorticity ζ. S24, Absolute Vortex Conservation Physical Residual Solving Submodule: Receives the relative vorticity ζ output from S23, combines it with the current typhoon latitude and the latitudinal change rate of the geostrophic effect, expands the absolute vortex conservation mass conservation equation under the quasi-geostrophic approximation into an Euler spatiotemporal evolution form, explicitly characterizing the Beta drift path shift effect caused by the typhoon's endogenous circulation; calculates the degree of deviation between the model simulation evolution results and the absolute vortex conservation theory, obtaining the globally differentiable physical residual Rphys; outputs the physical residual Rphys. The physical-data feature fusion layer integrates data features extracted by LSTM (pure data-driven trend prediction) and global physical constraint scalar residuals output by the differentiable physical layer (meteorological dynamic constraints). The model output layer takes the fused hidden state as input to the fully connected layer and outputs the complete typhoon state vector for the next time step. The model's output is the state increment target vector s for the next time step. t+1 : ; In the formula, and These are the latitude and longitude increments for the next moment; The lowest pressure increment at the center; This represents the maximum wind speed near the center at the next moment. By overlaying historical data with predicted increments, a rolling step simulation of the typhoon's path and intensity is achieved.

[0029] 2) Design the joint loss function; The loss function consists of four weighted sub-losses (weights w1=0.3, w2=0.3, w3=0.2, w4=0.2): Path regression loss: constrains the typhoon's latitude and longitude trajectory to fit the observed samples; Pressure intensity regression loss: accuracy of pressure intensity prediction at the constraint center; Gradient wind speed mapping loss: constrains the network output wind speed to satisfy the physical relationship of gradient wind balance; Absolute vorticity conservation physical residual loss (core term): Forces the simulation process to satisfy the law of atmospheric vorticity conservation; 3) Training the model; 4) Large-sample random initial scenario generation: The negative binomial distribution is used to probabilistically model the annual frequency of typhoons. The annual frequency sequence of typhoons on a millennial scale is generated by Monte Carlo sampling. Based on historical observation data of the active typhoon sea area in the Northwest Pacific, the initial state vector of typhoons is randomly sampled and used as the initial input for typhoon evolution simulation. 5) After the model training is completed, use the trained model to generate virtual typhoon samples in batches. Using Monte Carlo random sampling and a pre-trained model, a virtual typhoon spanning 1000 years was simulated at once, generating tens of thousands of complete typhoon lifecycle trajectories, which served as input for the power grid wind load.

[0030] S51. Probabilistic Modeling of Typhoon Annual Occurrence Frequency: Considering the distribution characteristic of the annual typhoon frequency variance in the Northwest Pacific being greater than the mean, a negative binomial distribution is used to probabilistically model the typhoon occurrence frequency Ni for each simulated year, and the fitting yields distribution parameters r=1170.93 and p=0.98; 1000 Monte Carlo samplings are performed on the probability distribution to generate a random occurrence frequency matrix of typhoon years spanning 1000 years; Since the variance of the annual frequency of typhoons in the Northwest Pacific is significantly greater than the mean, this scheme uses a negative binomial distribution to represent the total frequency N of typhoons generated in any simulated year i. i Modeling:

[0031] Based on the historical best path dataset, the discrete model parameters are set to [values ​​to be inserted]. r =1170.93, p =0.98. By performing 1000 Monte Carlo samplings on this probability function, a matrix of random annual occurrences spanning 1000 years can be generated.

[0032] S52. Random sampling of initial typhoon state: Based on the annual simulated typhoon frequency, a random repeated sampling method is used to randomly sample the initial state vector s0, which includes initial longitude, latitude, minimum central pressure, and maximum wind speed, from the historical typhoon initial sample library in the active typhoon sea area of ​​the Northwest Pacific Ocean within the range of 5°N~25°N and 110°E~170°E. This vector serves as the initial input for the physics-data dual-driven typhoon evolution model.

[0033] S53. Full lifecycle simulation of typhoon landfall degradation and multi-mechanism random termination determination: Typhoon landfall status is identified by land and sea indicators. The Vickery index model with regional parameterization is used to simulate the typhoon land surface intensity decay. The upper limit of the typhoon life cycle is set based on the Gamma distribution, and the typhoon dissipation pressure threshold is set based on the log-normal distribution. The typhoon rolling simulation is terminated by dual determination, generating a virtual sample of the entire typhoon life cycle that conforms to the real climate characteristics, and providing extreme load input for the reliability assessment of the power grid's wind resistance and disaster prevention.

[0034] To ensure that the generated virtual random typhoon samples closely match real climate characteristics in terms of lifespan and landfall dissipation process, the scheme designs a clear random termination mechanism: 1. Land Surface Intensity Topographic Attenuation Model (Vickery Index Model): When the longitude and latitude of the typhoon in the rolling simulation trigger the land area determination (identified by the land-sea indicator Mt), the model automatically triggers the intensity index attenuation equation caused by the friction of the underlying land surface.

[0035] In the formula: The central pressure difference at the moment of initial landing; t is the cumulative time since landing; This is the residual correction term. The attenuation constants a0 and a1 are strictly extracted based on the geographical latitude range of the typhoon (for example, a0=0.0323 and a1=0.00061 for the central region of 22°N~25.3°N).

[0036] 2. Random Determination of Lifespan Upper Limit (Gamma Distribution): The termination step in the continuous rolling simulation of the entire typhoon process depends on the upper limit threshold T of the lifespan extracted from the Gamma distribution obtained by fitting historical lifespans. Its lifespan probability density function is:

[0037] Based on the maximum likelihood estimation of historical long-series data, the shape parameter of the Gamma distribution is taken as follows: k =4.1452, scale parameter taken =42.4689.

[0038] 3. Dissipation Threshold Determination (Log-Normal Distribution): When the typhoon dissipates and the near-center pressure difference falls below the set termination threshold, the rolling simulation is also forcibly stopped. The central pressure difference at the time of typhoon dissipation approximately follows a log-normal distribution on a logarithmic scale, and its dissipation determination probability density function is:

[0039] Statistical fitting control parameters =2.5987, =0.5455.

[0040] 7) Extract the maximum near-surface wind pressure and wind speed extreme loads from each virtual typhoon sequence, and input the extreme loads in batches into the mechanical failure model of regional power transmission towers and overhead lines; Based on Monte Carlo sampling statistics of the failure frequency of transmission system lines and towers under all virtual typhoons, the failure probability of lines, the risk of system power outage, and the wind resistance reliability index of the power grid are calculated, thus completing the calculation of the typhoon failure probability and wind resistance reliability index of the regional transmission network. It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for assessing the reliability of power transmission networks based on typhoon path and intensity simulation, characterized in that, Includes the following steps: 1) Construct a time-series prediction network model for predicting the path and intensity evolution of typhoons; The time-series prediction network model uses a Long Short-Term Memory (LSTM) network as the basic model for extracting continuous temporal evolution features. Its structure includes: Input layer, LSTM backbone network, embedded differentiable physical layer, feature fusion layer, output prediction layer; The model input layer takes the state vector sequence of the typhoon center over k consecutive historical time steps as input; The output of the LSTM backbone network is: the extracted data features; After the LSTM unit calculates the data features, a differentiable physical layer is embedded in the network forward propagation link; The physical-data feature fusion layer fuses data features extracted by LSTM and global physical constraint scalar residuals output by the differentiable physical layer. The model output layer takes the fused hidden state as input to the fully connected layer and outputs the complete typhoon state vector for the next time step. The model's output is the state increment target vector s for the next time step. t+1 : ； In the formula, and These are the latitude and longitude increments for the next moment; The lowest pressure increment at the center; The maximum wind speed near the center at the next moment; by overlaying historical data with predicted increments, a rolling step simulation of the typhoon's path and intensity is achieved. 2) Design the joint loss function; The loss function consists of four weighted sub-losses, as follows: Path regression loss: used to constrain the fit between the typhoon's latitude and longitude trajectory and the observed samples; Pressure intensity regression loss: used to constrain the accuracy of central pressure intensity prediction; Gradient wind speed mapping loss: used to constrain the network output wind speed to satisfy the physical relationship of gradient wind balance; Absolute vorticity conservation physical residual loss: used to force the simulation process to satisfy the atmospheric vorticity conservation law; 3) Training the model; 4) Large-sample random initial scenario generation: Probabilistic modeling of the frequency of typhoon occurrences is performed, and a millennial-scale typhoon annual frequency sequence is generated by sampling. Based on historical observation data, the initial state vector of typhoons is randomly sampled and used as the initial input for typhoon evolution simulation. 5) After the model training is completed, use the trained model to generate virtual typhoon samples in batches. 6) Conduct quantitative assessment of the wind resistance reliability of the power transmission network based on the near-surface maximum wind pressure and extreme wind speed loads of the virtual typhoon sequence.

2. The power grid reliability assessment method based on typhoon path and intensity simulation according to claim 1, characterized in that, A differentiable physical driving layer is serially embedded within the forward propagation link of the LSTM network. This differentiable physical driving layer comprises four fully differentiable, sequentially connected sub-modules. The data flow of each sub-module is unidirectionally progressive and follows a gradient chain-like return flow. Specifically, it includes: S21, Radial Pressure Field Reconstruction Submodule: Receives the LSTM-predicted minimum typhoon center pressure Pc and the current typhoon latitude. Using 1010 hPa as a fixed ambient pressure Pn, it calculates the typhoon center pressure difference. Combining the Earth's rotation angular velocity, it solves for the Coriolis parameters and Holland model pressure profile parameters B. It reconstructs the typhoon's radial pressure distribution using the Holland analytical expression and obtains the radial pressure gradient at any spatial location by taking the partial derivative with respect to the radial distance. Simultaneously, it solves for the typhoon's maximum wind speed radius Rmax using the Vickery empirical formula. Outputs the radial pressure gradient, Coriolis parameters, and maximum wind speed radius Rmax. S22, YanMeng Wind Field Equilibrium Velocity Inversion Submodule: Receives the radial pressure gradient, Coriolis parameters, and maximum wind speed radius Rmax output from S21. Combined with air density, overall typhoon translational velocity, and angular parameters, it decomposes air velocity into free atmospheric gradient wind speed and surface friction-corrected wind speed based on the boundary layer momentum equation. It derives the tangential and radial gradient wind speeds separately, synthesizing them to obtain the east-west wind speed u and north-south wind speed v in Cartesian coordinates. Based on the gradient wind physical mapping relationship, it solves for the theoretical maximum wind speed Vmax at Rmax, constructs the mapping error between the theoretical wind speed and the network predicted wind speed, realizes the physical hard constraint of the maximum wind speed, and reduces the network prediction degree of freedom. It outputs the two-dimensional wind field components u and v, and the wind speed physical mapping deviation. S23, Relative Vorticity Conversion Submodule: Receives the two-dimensional wind field components u and v output from S22, solves for the relative vorticity ζ corresponding to the typhoon's rotational motion based on the partial differential relation of the local Cartesian coordinate system, and converts the wind field vector characteristics into dynamic vorticity characteristics that govern the typhoon's path deviation; outputs the relative vorticity ζ. S24, Absolute Vortex Conservation Physical Residual Solving Submodule: Receives the relative vorticity ζ output from S23, combines it with the current typhoon latitude and the latitude change rate of the geostrophic effect, expands the absolute vortex conservation mass conservation equation under the quasi-geostrophic approximation into an Euler spatiotemporal evolution form, explicitly characterizing the Beta drift path shift effect caused by the typhoon's endogenous circulation; calculates the degree of deviation between the simulation evolution results of the model and the absolute vortex conservation theory, obtaining the globally differentiable physical residual Rphys; outputs the physical residual Rphys.

3. The power grid reliability assessment method based on typhoon path and intensity simulation according to claim 1, characterized in that, The model input layer takes the state vector sequence of the typhoon center over k consecutive historical time steps as input; Wherein, at any time t, the state vector s of the typhoon center t Defined as: In the formula: Longitude of the typhoon center; Latitude of the typhoon center; This is the lowest atmospheric pressure at the center of the typhoon; For both land and sea indicators.

4. The power grid reliability assessment method based on typhoon path and intensity simulation according to claim 1, characterized in that, The loss function is as follows: Pathspace regression loss: Pressure intensity regression loss: Gradient wind speed theoretical mapping loss: Absolute vorticity conservation deviation: In the formula, there are superscripts. obs All variables are real observation values ​​obtained from historical meteorological measurements; In the formula, w 1, w 2, w 3, w 4. Weighting coefficients to balance the contributions of each task; L track L press L v and L phys These represent the path space regression loss, air pressure intensity regression loss, gradient wind speed theoretical mapping loss, and absolute vorticity conservation deviation, respectively; N is the number of training samples; λ, φ, and φ are the longitude and latitude of the typhoon center, respectively; P is the minimum central pressure, and V... max These represent the maximum wind speed near the center; the superscript obs indicates historical observations; u and v represent the east-west and north-south wind speed components in Cartesian coordinates, respectively; ζ is the relative vorticity; and β is the local planetary vorticity gradient.

5. The power grid reliability assessment method based on typhoon path and intensity simulation according to claim 1, characterized in that, In step 3), the training process is as follows: S31. Organize the historical best path data into supervised training samples in chronological order. Each sample takes the typhoon state vector sequence of k consecutive historical time steps as input and the next moment's longitude increment, latitude increment, minimum central pressure increment, and maximum wind speed increment as labels. S32. After normalizing the input samples, input them into the LSTM backbone network. During forward propagation, perform data flow calculation of the differentiable physical layer simultaneously to obtain state increment prediction, gradient wind speed mapping deviation and absolute vorticity conservation physical residual. S33. Construct a joint loss function based on path space regression loss, air pressure intensity regression loss, gradient wind speed theoretical mapping loss, and absolute vorticity conservation physical residual loss. S34. The time backpropagation algorithm is used to jointly update the trainable parameters of the LSTM backbone network, the physical-data fusion layer and the output prediction layer, and the validation samples are used to select the model until the joint loss converges or the preset number of iterations is reached.

6. The power grid reliability assessment method based on typhoon path and intensity simulation according to claim 1, characterized in that, In step 5), the specific details are as follows: S51. Probabilistic modeling of annual typhoon frequency: The negative binomial distribution is used to probabilistically model the typhoon frequency Ni of each simulated year, and the distribution parameters r and p are obtained by fitting. The probability distribution was sampled 1000 times using Monte Carlo sampling to generate a random occurrence frequency matrix of typhoon years spanning 1000 years. The total frequency N of typhoons generated in any simulated year i is calculated using a negative binomial distribution. i Modeling: S52. Random sampling of the initial state of typhoons: Based on the annual simulated typhoon frequency, a random repeated sampling method is used to randomly sample the initial state vector s0, which includes the initial longitude, latitude, minimum central pressure, and maximum wind speed, from the historical typhoon initial sample database within the active typhoon sea area of ​​the Northwest Pacific Ocean, within the range of 5°N~25°N and 110°E~170°E. This vector is used as the initial input of the physical-data dual-driven typhoon evolution model. S53. Set the typhoon simulation termination conditions and conduct a virtual sample simulation of the entire typhoon life cycle: A virtual sample of the entire life cycle of a typhoon, conforming to real climate characteristics, is generated to provide extreme load input for the reliability assessment of the power grid's wind resistance and disaster prevention.

7. The power grid reliability assessment method based on typhoon path and intensity simulation according to claim 1, characterized in that, In step S53), Typhoon landfall status is identified by land and sea indicators, the land surface intensity decay of typhoons is simulated by a regional parameterized Vickery index model, the upper limit of typhoon life cycle is set by Gamma distribution, the typhoon dissipation pressure threshold is set by log-normal distribution, and multiple mechanisms work together to determine the typhoon simulation termination conditions. The Vickery exponential decay model uses the central pressure difference at the moment of first landfall, the cumulative time of landfall, and the residual correction term as core variables, and the decay constants a0 and a1 are taken according to the latitude interval of the typhoon. The Gamma distribution is used to constrain the upper limit of the typhoon's lifespan, and the maximum number of simulation steps for the typhoon is constrained by randomly sampling the upper limit threshold of the lifespan. The log-normal distribution is used to limit the critical pressure difference for typhoon dissipation and simulate the natural dissipation process of typhoons; finally, virtual samples of the entire life cycle of typhoons on a millennium scale are generated in batches, which conform to the laws of real climate statistics and atmospheric physics.

8. The power grid reliability assessment method based on typhoon path and intensity simulation according to claim 1, characterized in that, Step 6) includes: The maximum near-surface wind pressure and extreme wind speed loads of all virtual typhoon sequences are extracted, and the extreme loads are batch-input into the mechanical failure models of transmission towers and overhead lines. Based on Monte Carlo sampling statistics of the failure frequency of transmission lines and towers under multiple scenarios, the failure probability of lines, the risk of system power outage and the wind resistance reliability index of the power grid are calculated, and the full probability assessment of the reliability of the regional power grid under typhoon disasters is completed.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.

Images