Limited diffusion model-based power grid catastrophe analog simulation method and system

Abstract

The invention discloses a power grid catastrophe analog simulation method and system based on a limited diffusion model, and the method comprises the steps: obtaining a state variable and an operation parameter of a power system according to a power grid line topological structure, and constructing a transient differential model of a power grid; solving a transient differential model of the power grid, calculating a target predicted value according to a solving result, and constructing a transient limited diffusion model of the power grid; constructing a neural network model to generate a predicted value, training the neural network model by taking the distance between the minimum predicted value and the target predicted value as a target, and performing an inverse diffusion process of a diffusion limiting model by using the trained neural network model to complete analogue simulation of a power grid catastrophe and recovery process; and adjusting parameters of the trained neural network model, and then applying the neural network model to a power grid system with a more complex topological structure. According to the method, the calculation process can be simplified, the calculation time can be shortened, the precision is remarkably improved, the method can be popularized to a more complex power grid system, the interpretability is high, and power grid catastrophe and black box in the recovery process are avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system fault simulation, and more specifically, relates to a power grid disaster simulation method and system based on a restricted diffusion model. Background Art

[0002] With the continuous development of new power systems, the complexity of power grids has increased, making power grid operation more complicated. Urban social and economic activities are highly dependent on electricity, and electricity has become the lifeline of cities. Once a large-scale failure occurs in the power grid, it may lead to serious political and social problems. The causes of power grid disasters usually include single or multiple failures, natural disasters, terrorist attacks and wars, all of which may undermine the safe and stable operation of the power grid. Despite technological advances, the risk of large-scale power outages and disasters in large urban power grids still exists, and these risks need to be studied in depth and actively addressed.

[0003] Prior art document 1 (CN113111512A) proposes a method for establishing a steady state using a transient simulation model, and prior art document 2 (CN104424510A) proposes a method for grid recovery after a disaster, both of which black-box the grid disaster and recovery process. The disadvantage is that, due to the black-boxing, the prior art cannot simulate the grid disaster and recovery process, it is difficult to accurately determine the nature and location of the fault, it is difficult to accurately predict whether the system can recover to a stable state during the transient process, the interpretability is poor, and there are safety hazards.

[0004] There are still some transient simulation algorithms in the prior art, which generally adopt numerical integration algorithms, etc., but their shortcomings are high complexity and limited accuracy, making them difficult to use in large-scale and complex power grid systems. In addition, traditional simulation methods based on steady-state power flow are difficult to efficiently simulate the dynamic process of power grid disasters and recovery. The existing technology cannot meet the needs of complex power grid disaster simulation and collaborative deduction, and there is still a lack of a simulation method that can efficiently simulate the dynamic process of complex power grid disasters and recovery. Therefore, it is necessary to further improve the simulation system to meet the needs of power grid disaster risk management. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the present invention provides a power grid disaster simulation method and system based on a restricted diffusion model to achieve more efficient and accurate simulation of power grid disaster and recovery process.

[0006] The present invention adopts the following technical solution.

[0007] A first aspect of the present invention provides a power grid disaster simulation method based on a restricted diffusion model, comprising the following steps:

[0008] Step 1, obtaining the state variable X and the operating parameter Y of the power system according to the topological structure of the power grid line, and constructing a transient differential model of the power grid;

[0009] Step 2: Solve the transient differential model of the power grid and calculate the target prediction value based on the solution results. Construct a transient restricted diffusion model for power grid;

[0010] Step 3: Build a neural network model to generate prediction values To minimize the predicted value and target prediction value The distance is used as the target to train the neural network model, and the trained neural network model is used to perform the inverse diffusion process of the restricted diffusion model in step 2 to complete the simulation of the power grid disaster and recovery process;

[0011] The method further includes: adjusting parameters of the trained neural network model and applying it to a power grid system with a more complex power grid line topology than that of step 1.

[0012] Preferably, the transient differential model of the power grid satisfies the following relationship:

[0013]

[0014] Where:

[0015] X is the state variable of the dynamic element;

[0016] Y is the operating parameter of the power system;

[0017] Φ(X,Y) represents a function related to the properties of the dynamic element itself and the power system model;

[0018] F(X,Y) represents the dynamic element input and output current vectors.

[0019] Preferably, step 2 specifically includes:

[0020] Step 2.1, construct a diffusion model;

[0021] Step 2.2: Use the solution of the transient differential model of the power grid in step 1 within the set time period as the substitute mean μ of the restricted diffusion model. t ′, replacing the mean μ of the diffusion model in step 2.1 t , and set the variance β of the restricted diffusion model t ;

[0022] Step 2.3, according to the μ obtained in step 2.2 t ′ and the set β t , and then generate samples ∈ through standard normal distribution t , calculate the target prediction value of step t The restricted diffusion model of power grid transient is obtained.

[0023] Preferably, in step 2.3, the target prediction value of step t is obtained by the following formula:

[0024]

[0025] Where:

[0026] μ t ′ is the substitute mean;

[0027]

[0028] Preferably, step 3 specifically includes:

[0029] Step 3.1, initialize the neural network parameters θ and set the loss function L of the neural network model;

[0030] Step 3.2, with the goal of optimizing the loss function, train the model and update the model parameters θ;

[0031] Step 3.3, determine whether the convergence condition is met, and terminate the training if the condition is met;

[0032] Step 3.4: Generate predictions using the trained neural network model As the mean of the inverse diffusion process in the restricted diffusion model of the power grid transient in step 2, the initial conditions are input for reasoning to predict the changes of the state variables X of the dynamic elements in the power grid and the operating parameters Y of the power system over time.

[0033] Preferably, in step 3.1, the loss function is set to

[0034] Preferably, in step 3.3, the convergence condition includes that the changes of the state variable X and the operating parameter Y of the power system are both less than the set value, or The amount of change is less than the specified threshold ε, or the iteration exceeds the specified number of rounds R.

[0035] Preferably, the step of adjusting the parameters of the trained neural network model and applying it to a power grid system with a more complex power grid line topology than that of step 1 specifically includes:

[0036] Step A.1, according to the topological structure of the more complex power grid system, obtaining the state variables X0 of the dynamic elements of the more complex power grid and the operating parameters Y0 of the power system, and constructing a differential model of the more complex power grid system;

[0037] Step A.2, solving the state variable Xl and the operating parameter Yl within the set time period as the fine-tuning data set;

[0038] Step A.3, retraining the neural network model in step 3 using the fine-tuning data set in step A.2, and adjusting the parameters of the neural network model;

[0039] Step A.4, using the neural network model adjusted in step A.3 to simulate the more complex power grid system disaster and recovery process.

[0040] Preferably, the trained neural network model is trained on a subset of the more complex power grid system.

[0041] The second aspect of the present invention provides a power grid disaster simulation system based on a restricted diffusion model, which runs the power grid disaster simulation method based on a restricted diffusion model, including: a topology construction module, a probability diffusion module, a network training module and a parameter adjustment module;

[0042] The topology building module is used to build a transient differential model of the power grid according to the topological structure of the power grid line;

[0043] The probability diffusion module is used to construct a transient restricted diffusion model of the power grid based on the solution results;

[0044] The network training module is used to train the neural network model to generate the mean of the inverse diffusion process in the restricted diffusion model;

[0045] The parameter adjustment module is used to apply the neural network model generated by the network training module to other power grid systems.

[0046] Compared with the prior art, the beneficial effects of the present invention include at least:

[0047] 1. The power grid disaster simulation method and system based on the restricted diffusion model of the present invention can not only simplify the calculation process and shorten the calculation time, but also obtain the probability of results under various trends, avoiding the systematic deviations existing in traditional simulations based on numerical integration and steady-state power flow, and significantly improving the accuracy.

[0048] 2. The present invention obtains the state variables of dynamic components and the operating parameters of the power system by using differential equations, and converts them into the mean value in the diffusion model, thereby realizing the organic combination of the transient change process of the power system and the diffusion process, so that the diffusion model can be applied to the simulation of power grid disasters and recovery processes, with strong interpretability and avoiding black box.

[0049] 3. The present invention utilizes the generalization of the diffusion model. By training on small power grids and fine-tuning on complex power grids, the model can be extended to more complex power grid systems without significantly increasing the calculation time. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1is a schematic diagram of a diffusion model provided according to an embodiment of the present invention;

[0051] Figure 2 It is a flow chart of a power grid disaster simulation system based on a restricted diffusion model provided in accordance with an embodiment of the present invention;

[0052] Figure 3 The present invention is a schematic diagram of the main connection of a 110kV substation provided in accordance with an application example of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0054] The present invention modifies the diffusion model, restricts the diffusion process through prior conditions, and constructs a diffusion model suitable for the disaster and recovery process of the power system, thereby realizing high-precision simulation of the disaster and recovery process of the power grid. The present invention greatly simplifies the computational complexity by utilizing the efficient computing power of the diffusion model; significantly improves the simulation accuracy by utilizing the strong generation ability of the diffusion model, and provides a more efficient and accurate simulation of the disaster and recovery process of the power grid; and utilizes the generalization of the diffusion model to fine-tune the training results of the small power grid and generalize them to more complex power grid systems.

[0055] like Figure 2 As shown, embodiment 1 of the present invention provides a power grid disaster simulation method based on a restricted diffusion model, comprising the following steps:

[0056] Step 1: According to the topological structure of the power grid, the state variable X and the operating parameter Y of the power system are obtained, and a transient differential model of the power grid is constructed to simulate the disaster and recovery process of the power grid.

[0057] In a preferred but non-limiting embodiment of the present invention, in order to simulate the disaster and recovery process of the power grid, a transient simulation of the power grid needs to be performed, which can be reduced to solving a set of nonlinear differential algebraic equations, which can be expressed as the following formula:

[0058]

[0059] Wherein, equation (1) describes the differential equation of the dynamic characteristics of the relevant components of the power system, X is the state variable of the dynamic component, that is, the variable with mechanical inertia or electromagnetic characteristics (such as electromotive force, power angle, etc.), Y is the operating parameter of the power system (such as voltage U, current I, active power P, reactive power Q, etc.), Φ is the function of the state variable X and the operating parameter Y, which is related to the properties of the component itself and the model of the power system;

[0060] Formula (2) describes the network model of the system. F(X, Y) is the input and output current vector of the dynamic element related to X and Y. This formula expresses that for any element, the input and output currents should be equal and the sum is 0.

[0061] For a single component, Φ(X,Y) generally consists of several equations; for n components, Φ(X,Y) will consist of n times the number of equations. Therefore, when the scale of the power grid is large, the set of equations will also become very large and difficult to solve quickly and efficiently.

[0062] By simulating the disaster and recovery process of the power grid, we can accurately control the range and change direction of physical quantities such as the power angle, and then determine what to do next with the components in the system to ensure the stability and safety of the power system.

[0063] At present, there are two main algorithms for solving the system of equations (1) and (2): one is the simultaneous solution of differential algebraic equations, which uses the implicit trapezoidal formula to differentiate the differential equations and solve the differential equations and network equations simultaneously. The implicit integral simultaneous solution has good numerical stability, but it is relatively complex in program design and implementation. It is necessary to establish a simultaneous correction equation. The program has insufficient scalability and flexibility, so it is difficult to be widely used. The other algorithm is the alternating iteration method of differential equations and network equations based on the implicit integral method. The alternating iteration method can maintain the numerical stability of the implicit integral, but there is an interface error between the differential equation and the algebraic equation, which will cause error amplification in complex power grid systems. In addition, when there are too many related components in the power grid and the differential equation matrix is ​​too large, there are still problems such as high computational complexity. In addition, both methods have the problem of cumulative errors causing deviation from the true value after multiple time step iterations.

[0064] Step 2: Solve the transient differential model of the power grid in step 1 and calculate the target prediction value based on the solution results. Construct a restricted diffusion model for power grid transients.

[0065] In a preferred but non-limiting embodiment of the present invention, step 2 specifically comprises:

[0066] Step 2.1, construct a diffusion model, such as Figure 1 shown.

[0067] The Diffusion Model is a deep learning algorithm that is mainly used in the field of generative models, especially in image generation. The core idea of ​​this model is to simulate a diffusion process from order to disorder and then to order, and to generate data by gradually increasing and then removing noise.

[0068] From the above description, it can be seen that the diffusion model is divided into two processes:

[0069] 1) Diffusion process (forward process), from target distribution to noise distribution, from the original data distribution x 0 Continue to add Gaussian noise until it finally becomes an opposing Gaussian distribution x T ;

[0070] 2) Reverse diffusion process (reverse process), from x T Starting from, through the process of denoising and continuous recovery, we gradually obtain x 0 .

[0071] The diffusion process can generally be realized by the following formula:

[0072]

[0073] This formula refers to the use of x t-1 Predict x t When , it is assumed to obey a conditional probability distribution, and the conditional probability distribution is a Gaussian distribution is the mean, β t is the variance, and I corresponds to the standard deviation of the standard normal distribution of 1.

[0074] At this time x t It can be easily calculated from the previous moment: where ∈ t is a sample generated from a standard normal distribution. Let α t =1-β t , but

[0075] The inverse diffusion process needs to be implemented through a neural network. The essence of denoising is to obtain the variance and mean of the inverse operation when adding noise. The variance is generally a constant in the diffusion model and is related to x. 0 The mean can be calculated using the following formula:

[0076]

[0077] And x t It is known in the reverse diffusion process, so the only unknown quantity is ∈ t, so the neural network only needs to fit a ∈ θ (x t ,t), so that∈ θ (x t ,t) as close as possible to∈ t That's it.

[0078] The loss function of the neural network in the tth step of the reverse diffusion process can be expressed as L t =||∈ t -∈ θ (x t ,t)|| 2 , the total loss function can be expressed as

[0079] Step 2.2, improve the diffusion model of step 2.1, and use the solution results of the transient differential model of the power grid in step 1 within the set time period, that is, the state variables X of the dynamic components and the operating parameters Y of the power system, as the substitute mean μ of the restricted diffusion model t ′, replacing the mean μ of the diffusion model in step 2.1 t , and set the variance β of the restricted diffusion model t .

[0080] It is worth noting that the diffusion process in the diffusion model uses Gaussian distribution to generate a series of noisy data to achieve a gradual noise addition process. However, the power system disaster and recovery process does not conform to the gradual noise addition model of existing technology, so the diffusion model cannot be directly applied.

[0081] Further preferably, the noise-added process formula (3) is improved, and the diffusion process in the power system is set to comply with the following conditions:

[0082]

[0083] Where:

[0084] μ t ′ is the substitute mean of the Gaussian distribution, corresponding to the solution obtained by solving the transient differential equation of the power system;

[0085] β t is the variance, corresponding to the potential error of the solution, which can be set artificially.

[0086] It can be understood that the diffusion model is expressed by the formula The mean value of the tth step is automatically generated, and the present invention obtains the actual X and Y in each time step by solving the transient differential equation, using μ t ′ instead of μ t Designing restricted diffusion models for power systems.

[0087] Step 2.3, according to the μ obtained in step 2.2 t ′ and the set β t , and then generate samples ∈ through standard normal distribution t , calculate the target prediction value instead of the mean at step t The restricted diffusion model is obtained.

[0088] Further preferably, the target prediction value replacing the mean value in step t can be obtained by the following formula:

[0089]

[0090] Where:

[0091]

[0092] Step 3: Build a neural network model to generate predicted values ​​instead of the mean To minimize the predicted value and target prediction value The distance is used as the target to train the neural network model, and the trained neural network model is used to perform the inverse diffusion process of the restricted diffusion model in step 2 to complete the simulation of the power grid disaster and recovery process.

[0093] In a preferred but non-limiting embodiment of the present invention, step 3 specifically comprises:

[0094] Step 3.1, initialize the neural network parameters θ and set the loss function L of the neural network model.

[0095] Further preferably, the loss function is set to

[0096] Step 3.2, with the goal of optimizing the loss function, train the model and update the model parameters θ.

[0097] Step 3.3, determine whether the convergence condition is met, and terminate the training if the condition is met.

[0098] Further preferably, the convergence condition is that the changes in X and Y are both less than the set value, or The amount of change is less than the specified threshold ε, or the iteration exceeds the specified number of rounds R.

[0099] Step 3.4: Generate predictions using the trained neural network model As the mean of the inverse diffusion process in the restricted diffusion model of the power grid transient in step 2, the initial conditions are input for reasoning to predict the changes of the state variables X of the dynamic elements in the power grid and the operating parameters Y of the power system over time.

[0100] It can be understood that by improving the diffusion model, the present invention can significantly shorten the calculation time, not only can a series of future prediction values ​​(X and Y) be generated through the initial system state of the disaster, but also the probability of results under various trends can be obtained.

[0101] In addition, the generalization of the diffusion model can be extended to more complex power grid systems. As mentioned in step 1, the transient differential equations of complex power grids are difficult to solve, so a large number of μ t 'There are difficulties. The present invention proposes that training can be performed on a small power grid (such as a subset of a complex power grid, or other relatively small power grid systems) first. Then, only a small number of data points are taken from the complex power grid as a fine-tuning data set, and sometimes a fine-tuning data set can achieve good results with dozens of samples. Therefore, by utilizing the generalization of the diffusion model, the diffusion model of power grid simulation can be promoted to complex systems.

[0102] The steps of fine-tuning the trained neural network model and applying it to a grid system with a more complex topology include:

[0103] Step A.1, according to the more complex power grid line topology and corresponding data, obtain the state variable X0 of the dynamic element of the complex system and the operating parameter Y0 of the power system, and construct a differential equation suitable for this complex power grid system.

[0104] Step A.2, solve the state variable Xl and operating parameter Yl in a short period of time as fine-tuning data.

[0105] Step A.3, fine-tune the model parameter θ for the data set. The fine-tuning process adopts a restricted diffusion model construction method and a neural network training method similar to Step 2 and Step 3 in Example 1.

[0106] Step A.4 completes model fine-tuning and can be used for simulation of complex power grid disasters and recovery processes.

[0107] Embodiment 2 of the present invention provides a power grid disaster simulation system based on a restricted diffusion model, which runs the power grid disaster simulation method based on a restricted diffusion model described in embodiment 1, including: a topology construction module, a probability diffusion module, a network training module and a parameter adjustment module;

[0108] The topology building module is used to build a transient differential model of the power grid according to the topological structure of the power grid line;

[0109] The probability diffusion module is used to construct a transient restricted diffusion model of the power grid based on the solution results;

[0110] The network training module is used to train the neural network model to generate the mean of the inverse diffusion process in the restricted diffusion model;

[0111] The parameter adjustment module is used to apply the neural network model generated by the network training module to other power grid systems.

[0112] In order to further clearly introduce the technical solution of the present invention and the beneficial technical effects brought by it, the following introduces an application example, taking the handling process after a 110kV substation main transformer fault trip as an example to illustrate the specific implementation plan of this system. Figure 3 shown.

[0113] The power grid disaster simulation method based on the restricted diffusion model described in Example 1 of the present invention is used to simulate the disaster and recovery process after the internal fault of the main transformer of the 110kV substation, and migrate to a complex power grid system, specifically including the following steps:

[0114] Step B.1: According to the topological structure of the power grid and the corresponding data, the state variables X0 of the dynamic elements and the operating parameters Y0 of the power system are obtained, and the differential equations applicable to this power grid system are constructed;

[0115] Step B.2: Solve the differential equation to obtain all X and Y data sets within time T, and set the data in the data set to μ in each time step i . Set β i , β i It can be set based on the trade-off between accuracy and efficiency of the task, or it can be set manually based on experience.

[0116] Step B.3: Based on μ i and β i Construct a transient restricted diffusion model for the power grid.

[0117] Step B.4: Initialize the neural network parameters θ and set the loss function of the neural network model. A feasible loss function can be set as

[0118] Step B.5: Train the model and update the model parameters θ with the goal of optimizing the loss function L.

[0119] Step B.6: Determine whether the convergence condition is met. The convergence condition can be that X and Y do not change significantly, or The amount of change is less than the specified threshold ε, or the iteration exceeds the specified number of rounds R. If the condition is met, the training is terminated.

[0120] Step B.7: Input a new target initial condition. For example, if you want to predict the disaster and recovery process after the internal fault of #1 main transformer, you can set the internal fault of #1 main transformer, take the state variable X0 and the operating parameter Y0 of the power system at this time as the initial value, and use the neural network model trained in step B.6 combined with the restricted diffusion model in step B.3 for reasoning to obtain a series of data of the state variable X in the subsequent time period.

[0121] Step B.8: If you want to migrate this model to a complex power grid system, you can first obtain the state variables X0' of the dynamic elements of the complex system and the operating parameters Y0' of the power system based on the power grid line topology and corresponding data, and construct differential equations suitable for this complex power grid system.

[0122] Step B.9: Solve the state variables Xl and operating parameters Yl within a short period of time, using a small amount of labels as fine-tuning data.

[0123] Step B.10: Fine-tune the model parameters θ for the dataset. The fine-tuning process uses the restricted diffusion model construction method and loss function setting method similar to the training stage.

[0124] Step B.11: Complete model fine-tuning and can be used for simulation of complex power grid disasters and recovery processes.

[0125] It can be seen from the above model training and actual application process that the present invention only needs to solve the numerical solutions of some simple differential equations as basic data before training. When performing power grid disaster simulation afterwards, it is only necessary to change the initial conditions and input the trained restricted diffusion model to directly obtain the calculation results, replacing the complex solution process of differential equations, thereby simplifying the calculation, and after fine-tuning the model parameters with a small number of data points, it can be migrated to complex power grid systems, with good generalization performance.

[0126] Compared with the prior art, the beneficial effects of the present invention include at least:

[0127] 1. The power grid disaster simulation method and system based on the restricted diffusion model of the present invention can not only simplify the calculation process and shorten the calculation time, but also obtain the probability of results under various trends, avoiding the systematic deviations existing in traditional simulations based on numerical integration and steady-state power flow, and significantly improving the accuracy.

[0128] 2. The present invention obtains the state variables of dynamic components and the operating parameters of the power system by using differential equations, and converts them into the mean value in the diffusion model, thereby realizing the organic combination of the transient change process of the power system and the diffusion process, so that the diffusion model can be applied to the simulation of power grid disasters and recovery processes, with strong interpretability and avoiding black box.

[0129] 3. The present invention utilizes the generalization of the diffusion model. By training on small power grids and fine-tuning on complex power grids, the model can be extended to more complex power grid systems without significantly increasing the calculation time.

[0130] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A power grid disaster simulation method based on a restricted diffusion model, characterized in that: The following steps are involved: Step 1, obtaining the state variable X and the operating parameter Y of the power system according to the topological structure of the power grid line, and constructing a transient differential model of the power grid; Step 2: Solve the transient differential model of the power grid and calculate the target prediction value based on the solution results. Construct a transient restricted diffusion model for power grid; Step 3: Build a neural network model to generate prediction values To minimize the predicted value and target prediction value The distance is used as the target to train the neural network model, and the trained neural network model is used to perform the inverse diffusion process of the restricted diffusion model in step 2 to complete the simulation of the power grid disaster and recovery process; The method further includes: adjusting parameters of the trained neural network model and applying it to a power grid system with a more complex power grid line topology than that of step 1.

2. The power grid disaster simulation method based on the restricted diffusion model according to claim 1, characterized in that: The transient differential model of the power grid satisfies the following relationship: Where: X is the state variable of the dynamic element; Y is the operating parameter of the power system; Φ(X,Y) represents a function related to the properties of the dynamic element itself and the power system model; F(X,Y) represents the dynamic element input and output current vectors.

3. The power grid disaster simulation method based on the restricted diffusion model according to claim 1, characterized in that: Step 2 specifically includes: Step 2.1, construct a diffusion model; Step 2.2: Use the solution of the transient differential model of the power grid in step 1 within the set time period as the substitute mean μ of the restricted diffusion model. t ′, replacing the mean μ of the diffusion model in step 2.1 t , and set the variance β of the restricted diffusion model t ; Step 2.3, according to the μ obtained in step 2.2 t ′ and the set β t , and then generate samples ∈ through standard normal distribution t , calculate the target prediction value of step t The restricted diffusion model of power grid transient is obtained.

4. The power grid disaster simulation method based on the restricted diffusion model as claimed in claim 3, characterized in that: In step 2.3, the target prediction value of step t is obtained by the following formula: Where: μ t ′ is the substitute mean; 5. The power grid disaster simulation method based on the restricted diffusion model according to claim 1, characterized in that: Step 3 specifically includes: Step 3.1, initialize the neural network parameters θ and set the loss function L of the neural network model; Step 3.2, with the goal of optimizing the loss function, train the model and update the model parameters θ; Step 3.3, determine whether the convergence condition is met, and terminate the training if the condition is met; Step 3.4: Generate predictions using the trained neural network model As the mean of the inverse diffusion process in the restricted diffusion model of the power grid transient in step 2, the initial conditions are input for reasoning to predict the changes of the state variables X of the dynamic elements in the power grid and the operating parameters Y of the power system over time.

6. The power grid disaster simulation method based on the restricted diffusion model according to claim 5, characterized in that: In step 3.1, the loss function is set to 7. The power grid disaster simulation method based on the restricted diffusion model according to claim 5, characterized in that: In step 3.3, the convergence conditions include that the changes in the state variable X and the operating parameter Y of the power system are both less than the set value, or The amount of change is less than the specified threshold ε, or the iteration exceeds the specified number of rounds R.

8. The power grid disaster simulation method based on the restricted diffusion model according to claim 1, characterized in that: The steps of adjusting the parameters of the trained neural network model and applying it to a power grid system with a more complex power grid line topology than step 1 specifically include: Step A.1, according to the topological structure of the more complex power grid system, obtaining the state variables X0 of the dynamic elements of the more complex power grid and the operating parameters Y0 of the power system, and constructing a differential model of the more complex power grid system; Step A.2, solving the state variable Xl and the operating parameter Yl within the set time period as the fine-tuning data set; Step A.3, retraining the neural network model in step 3 using the fine-tuning data set in step A.2, and adjusting the parameters of the neural network model; Step A.4, using the neural network model adjusted in step A.3 to simulate the more complex power grid system disaster and recovery process.

9. The power grid disaster simulation method based on the restricted diffusion model according to claim 1, characterized in that: The trained neural network model is trained on a subset of the more complex power grid system.

10. A power grid disaster simulation system based on a restricted diffusion model, running a power grid disaster simulation method based on a restricted diffusion model as described in any one of claims 1 to 9, characterized in that: include: Topology building module, probability diffusion module, network training module and parameter adjustment module; The topology building module is used to build a transient differential model of the power grid according to the topological structure of the power grid line; The probability diffusion module is used to construct a transient restricted diffusion model of the power grid based on the solution results; The network training module is used to train the neural network model to generate the mean of the inverse diffusion process in the restricted diffusion model; The parameter adjustment module is used to apply the neural network model generated by the network training module to other power grid systems.

Images