System risk assessment method based on parallel operation

Abstract

The invention discloses a system risk assessment method based on parallel computation, which relates to the technical field of power system risk assessment and comprises the following steps: acquiring key operation parameters corresponding to a current moment of a power system; inputting the key operation parameters into a pre-trained load reduction prediction model at the same time to obtain a predicted value of the load reduction quantity of the power system at the current moment; according to the predicted value of the load reduction amount of the power system at the current moment, performing operation risk assessment on the power system to obtain an operation risk assessment result of the power system at the current moment; and when the operation risk assessment result reaches a set threshold value, sending an early warning signal or generating an operation adjustment suggestion. The method not only can fully consider the composite influence of the branch circuit fault and the multi-type equipment combination fault, but also can improve the response and judgment capability to the short-term risk of the power system under the complex working condition.

Description

Technical Field

[0001] The present invention relates to the technical field of power system risk assessment, and in particular to a system risk assessment method based on parallel computing. Background Art

[0002] With increasing global demands for carbon emissions control, new power systems based primarily on renewable energy are becoming a key development direction for the power industry. However, while the large-scale integration of renewable energy brings clean energy benefits, it also significantly increases the number of system states, leading to frequent and uncertain changes in system operating conditions. Consequently, operational risk assessment is gaining increasing attention due to the variability of system states.

[0003] Among the related technologies, the power system reliability assessment method faces the following technical bottlenecks: (1) The traditional Markov process-based steady-state probability assessment method realizes reliability evaluation through state enumeration and long-term probability analysis, but its modeling process relies on preset state transition rules, which makes it difficult to capture short-term dynamic characteristics such as the randomness of renewable energy output and the sudden change of load demand. Such methods can only reflect the long-term average reliability level of the system and cannot provide real-time risk warning and decision support for dispatchers. (2) Monte Carlo simulation (MCS) is a mainstream simulation method, and its assessment accuracy is positively correlated with the amount of calculation. In systems with a high proportion of renewable energy, the system state space grows exponentially with the number of renewable energy stations. Traditional MCS requires massive sampling to converge, resulting in the assessment taking hours or even days, which cannot meet the needs of online dispatching. (3) Most studies only consider a single fault type (such as generator failure) and fail to fully model the combined impact of branch circuit failures and multi-type equipment failures; (4) Existing methods mostly focus on static snapshot analysis, ignoring the temporal coupling relationship between renewable energy volatility and load demand, resulting in a lag in short-term risk prediction; (5) Traditional serial computing frameworks have difficulty processing high-dimensional state data, which limits the real-time reasoning capabilities of deep learning models in complex scenarios. Summary of the Invention

[0004] In view of this, the present invention provides a system operation risk assessment method based on parallel computing, which can not only fully consider the combined impact of branch circuit failures and combined failures of multiple types of equipment, but also improve the response and judgment capabilities to short-term risks of the power system under complex working conditions.

[0005] The present invention provides a system operation risk assessment method based on parallel computing, comprising:

[0006] S1. Obtain key operating parameters corresponding to the current moment of the power system; the operating parameters include: a node admittance matrix, a generator output matrix, and an energy demand matrix of each node;

[0007] S2. Constructing equivalent matrices corresponding to the real and imaginary parts of the node admittance matrix, respectively, and simultaneously inputting the equivalent matrix corresponding to the real part, the equivalent matrix corresponding to the imaginary part, the generator output matrix, and the energy demand matrix of each node into a pre-trained load reduction prediction model to obtain a predicted value of the load reduction amount of the power system at the current moment; the load reduction prediction model is determined based on a two-dimensional convolutional neural network including multiple feature input channels;

[0008] S3. Based on the predicted value of the load reduction amount of the power system at the current moment, an operation risk assessment of the power system is performed to obtain an operation risk assessment result of the power system at the current moment;

[0009] S4. When the operation risk assessment result reaches a set threshold, an early warning signal is issued or an operation adjustment suggestion is generated.

[0010] In an optional implementation, the number of characteristic input channels of the load reduction prediction model is ≥4.

[0011] In an optional embodiment, the training process of the load shedding prediction model includes:

[0012] Obtaining a sample data set; the sample data set includes: key operating parameters at multiple moments, and a label value corresponding to each moment; the label value represents the actual load reduction corresponding to each moment;

[0013] Divide the sample data set into training set and validation set according to the set ratio;

[0014] Inputting the key operating parameters at each moment in the training set into the load reduction prediction model in sequence for semi-offline training to obtain the predicted value of the load reduction amount at each moment;

[0015] Calculate the loss function corresponding to the predicted value of the load reduction at each moment. When the loss function meets the expected value, the pre-trained load reduction prediction model is obtained.

[0016] The key operating parameters at each moment in the validation set are sequentially input into the pre-trained load reduction prediction model to optimize the pre-trained load reduction prediction model.

[0017] In an optional embodiment, the key operating parameters at the multiple moments are obtained by performing multiple rounds of random sampling simulations on the annual load curve using a Monte Carlo simulation method.

[0018] In an optional embodiment, before dividing the sample data set into a training set and a validation set according to a set ratio, the training process of the load shedding prediction model further includes:

[0019] The sample data set is preprocessed; the preprocessing includes data classification, feature extraction, outlier removal and normalization.

[0020] In an optional implementation, the preprocessing of the sample data set includes:

[0021] According to whether there is load reduction, the sample data set is divided into reduction data and non-reduction data;

[0022] Extracting core variables that affect the system operating status from the reduction data and non-reduction data respectively; the core variables include: the real and imaginary parts of the node admittance matrix, the generator active power, and the instantaneous load demand of each node;

[0023] Invalid data and sampling data from abnormal sampling points are removed, and all core variables are normalized uniformly.

[0024] In an optional embodiment, the power system uses the expected power shortage amount and the power shortage probability of the power system as indicators for operation risk assessment, which are expressed as follows:

[0025]

[0026] Where EDNS is the expected power shortage; LOLP is the power shortage probability of the power system; N sum is the number of samples, y i Indicates the predicted value of the load reduction corresponding to each sample.

[0027] The present invention has the following beneficial effects:

[0028] The system operation risk assessment method based on parallel computing of the present invention utilizes a two-dimensional convolutional neural network to automatically extract and learn key features from system status data, significantly accelerating the computational efficiency of the risk assessment process, thereby achieving real-time reliability assessment of the system operation status. At the same time, considering key events that may occur during system operation, especially the impact of branch circuit fault information and generator fault information on system stability, a multi-input feature channel is constructed, and a deep learning model is used to model the relationship between system operation status and load curtailment (LC). By introducing a parallel computing mechanism (Parallel Training Method, PTM), system status data under large-scale fault scenarios is trained to determine the regression relationship between load curtailment and system status data. Furthermore, a fast load curtailment solver can be established for rapid risk assessment calculations to make high-precision predictions of load curtailment. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0030] Figure 1 1 is a flow chart of a system operation risk assessment method based on parallel computing according to an embodiment of the present invention;

[0031] Figure 2 2 is a schematic diagram of a two-dimensional convolutional neural network structure including multiple feature input channels according to an embodiment of the present invention;

[0032] Figure 3 is a diagram showing evaluation results of an expected power shortage indicator according to an embodiment of the present invention;

[0033] Figure 4 4 is a diagram showing evaluation results of the power shortage probability indicator according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0035] Operational risk assessment involves comprehensively assessing a system's ability to sustain energy supply when faced with various potential disturbance scenarios. During operation, power systems may be subject to a variety of external and internal uncertainties, such as drastic fluctuations in load levels, failures of critical equipment, the sudden shutdown of power generation units, and transmission line failures. These factors can cause the system to enter an unsafe state, resulting in inadequate load fulfillment and, consequently, load shedding. Therefore, appropriate measures must be taken to address the resulting load shedding. To rapidly determine the current load shedding amount, an embodiment of the present invention constructs a load shedding prediction model based on a two-dimensional convolutional neural network (2D-CNN). Using historical system operational data and a large number of simulation scenarios as training samples, the model automatically learns the deep nonlinear regression relationship between system status and load shedding during the training phase. This model eliminates the need to explicitly model system safety constraints. Instead, it leverages the learning capabilities of deep neural networks to integrate the constraint logic implicit in the operational data into the model in a data-driven manner. After training, the model can directly and rapidly infer real-time state inputs to predict the potential load shedding amount. Compared with traditional risk assessment methods that rely on state enumeration and power flow simulation, the present invention can perform rapid predictions in multi-state scenarios containing fault information, improve the coverage of risk identification, and also greatly improve the computational cost.

[0036] Figure 1 This is a system operation risk assessment method based on parallel computing according to an embodiment of the present invention. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0037] like Figure 1 As shown, the process includes the following steps:

[0038] S1. Obtain the key operating parameters corresponding to the current moment of the power system; the operating parameters include: node admittance matrix, generator output matrix and energy demand matrix of each node.

[0039] Specifically, during actual power system operation, the sensing module first collects node admittance information, generator output information, and load data for each node in real time. This data is then converted into a matrix to obtain the node admittance matrix, generator output matrix, and energy demand matrix for each node. The node admittance matrix is ​​in complex form, consisting of imaginary and real parts.

[0040] S2. Construct equivalent matrices corresponding to the real and imaginary parts of the node admittance matrix respectively, and simultaneously input the equivalent matrix corresponding to the real part, the equivalent matrix corresponding to the imaginary part, the generator output matrix, and the energy demand matrix of each node into a pre-trained load reduction prediction model to obtain a predicted value of the load reduction amount of the power system at the current moment; the load reduction prediction model is determined based on a two-dimensional convolutional neural network including multiple feature input channels.

[0041] Specifically, to enhance data representation and computational efficiency, the present invention extracts the real and imaginary parts of the node admittance matrix to construct corresponding equivalent matrices, ultimately forming four input matrices with consistent structures (the equivalent matrix corresponding to the real part, the equivalent matrix corresponding to the imaginary part, the generator output matrix, and the energy demand matrix of each node). These four input matrices are then input into multiple feature input channels of a two-dimensional convolutional neural network (2D-CNN).

[0042] In an optional implementation, the number of characteristic input channels of the load reduction prediction model is ≥4.

[0043] like Figure 2 As shown, the prediction process of the load reduction prediction model is described in detail below:

[0044] First, the four input matrices are subjected to multi-scale feature extraction through multiple convolutional layers to learn the potential relationship between the system operating status and the load reduction amount. Then, the multiple feature maps after multi-scale feature extraction are flattened and reconstructed into one-dimensional vectors. The vectors are input to the fully connected layer (FC) to output the predicted load reduction amount.

[0045] For example, after the four input matrices are acted upon by the convolution kernel, the convolutional neural network can mine the relevant feature system. The specific feature extraction formula is as follows:

[0046] L1=g(xh1+b1)

[0047] L i =g(L i-1 h i +b i )

[0048] Among them, L1 represents the output of the first convolutional layer, x represents the original input matrix. h1 is the convolution kernel of the first layer, b1 is the bias of the first layer, and g is the activation function of the first layer. Similarly, L i-1 The output of the previous layer is the result of the input of the next layer, L i Represents the output of the current layer.

[0049] The four input matrices first pass through two convolutional layers, and the corresponding results are reconstructed into one-dimensional data, which are then flattened and connected to the subsequent three fully connected layers. The output hidden in the connection layer can be expressed as follows.

[0050] FC=g(w f L last +b f )

[0051] Among them, L last is the output of the last convolutional layer, which is used as the input of the fully connected layer. f 、b f are the weights and biases in the fully connected layer respectively. FC represents the output of the fully connected layer.

[0052] In an optional embodiment, the training process of the load shedding prediction model includes:

[0053] Step A: Acquire a sample data set; the sample data set includes: key operating parameters at multiple moments and label values ​​corresponding to each moment; the label values ​​represent the actual load reduction corresponding to each moment.

[0054] In an optional implementation, the key operating parameters at multiple moments are obtained by performing multiple rounds of random sampling simulations on the annual load curve using a Monte Carlo simulation method.

[0055] Specifically, the system is in a stable operating state, no components have failed, and the system is operating under the N-0 operating condition. In order to obtain data that fully reflects the operating characteristics of the system, Monte Carlo Simulation (MCS) technology is used to generate a large amount of operating scenario data through multiple rounds of random sampling simulation under the annual load curve. The selection of this time scale can effectively eliminate the impact of short-term fluctuations, so that the collected data can better reflect the stability characteristics of the system under long-term operating conditions. The collection content covers key operating parameters such as the node admittance matrix (including its real and imaginary parts), the generator output matrix, and the energy demand matrix of each node. At the same time, if the system experiences load reduction during operation, the corresponding reduction amount and its corresponding system status will be recorded to ensure that each set of data has a clear risk label, which is convenient for the subsequent training model to identify and learn the reduction behavior.

[0056] Step B: Divide the sample data set into a training set and a validation set according to the set ratio.

[0057] Step C: inputting the key operating parameters at each moment in the training set into the load reduction prediction model in sequence to perform semi-offline training to obtain the predicted value of the load reduction amount at each moment;

[0058] Step D: Calculate the loss function corresponding to the predicted value of the load reduction amount at each moment. When the loss function meets the expected value, obtain the pre-trained load reduction prediction model;

[0059] Step E: Input the key operating parameters at each moment in the validation set into the pre-trained load reduction prediction model in sequence to optimize the pre-trained load reduction prediction model.

[0060] Specifically, the load shedding prediction model, through multiple convolutional layers and pooling operations, effectively captures the complex local patterns and structural features of the system's operating state, thereby establishing a mapping between the amount of load shedding and the system's input state. During model training, the load shedding amount is used as the regression target, and multiple rounds of iterative learning are performed on the training set. The model's performance is dynamically evaluated using the validation set, continuously optimizing network parameters to improve the model's generalization and prediction accuracy under different operating conditions. After training, the load shedding prediction model's parameters are solidified and saved to support rapid loading and deployment in subsequent real-time systems.

[0061] In addition, when a major disturbance occurs during system operation (such as a topology change caused by a device failure), the system state needs to be adjusted in a timely manner. The specific process is as follows:

[0062] Fault state identification: Identify whether the system operating state has changed, for example, the system changes from the N-0 order fault state to the N-1 order fault state.

[0063] Resample collection: Under the new topology conditions, return to the sample collection phase, re-run the MCS simulation, and collect the updated system status data.

[0064] Model update: Repeat the process of data preprocessing, semi-offline training, and model application to generate a new regression model that adapts to the current system operation status.

[0065] Dynamic optimization closed loop: Through multiple rounds of adjustment and optimization, the system ensures that it can continuously conduct efficient risk assessment in a complex and changing operating environment.

[0066] The present invention can ensure the dynamic adaptability of system operation risk assessment and can quickly respond to changes in system operation conditions.

[0067] In an optional embodiment, before dividing the sample data set into a training set and a validation set according to a set ratio, the training process of the load reduction prediction model further includes:

[0068] Preprocess the sample data set; preprocessing includes data classification, feature extraction, outlier removal and normalization.

[0069] Specifically, the sample dataset contains both reduced and non-reduced data. In order to improve data availability and modeling efficiency, the collected data must be systematically preprocessed during the preprocessing phase. This phase primarily includes data classification, feature extraction, outlier removal, and normalization. Specifically,

[0070] According to whether there is load reduction, the sample data set is divided into load reduction data and non-load reduction data to enhance the model's ability to identify boundary states.

[0071] The core variables that affect the system operation status in the reduction data and non-reduction data are extracted as model input features, focusing on the real and imaginary parts of the node admittance matrix, the active power of the generator, and the instantaneous load demand of each node;

[0072] Removing invalid data and sampling data from abnormal sampling points, and performing unified normalization on various core variables, can significantly improve the model convergence speed and training stability, laying a solid foundation for efficient modeling of deep neural networks.

[0073] Finally, the preprocessed reduction data and non-reduction data are respectively input into the two-dimensional convolutional neural network (2D-CNN) model for offline training of the regression model.

[0074] S3. Based on the predicted value of the load reduction amount of the power system at the current moment, an operation risk assessment of the power system is performed to obtain an operation risk assessment result of the power system at the current moment.

[0075] In an optional embodiment, the power system uses the expected power shortage and the power shortage probability of the power system as indicators for operation risk assessment, which are expressed as follows:

[0076]

[0077] Where EDNS is the expected power shortage; LOLP is the power shortage probability of the power system; N sum is the number of samples, y i Indicates the predicted value of the load reduction corresponding to each sample.

[0078] S4. When the operational risk assessment result reaches the set threshold, an early warning signal is issued or operational adjustment suggestions are generated.

[0079] Specifically, the pre-trained load reduction prediction model rapidly calculates the corresponding load reduction forecast value based on the current system operating status, further deriving the system's current operational risk indicators. Based on the assessment results, the system can provide dynamic early warning signals or operational adjustment recommendations, assisting operations and maintenance personnel in promptly responding to potential risks, preventing incidents from escalating, and improving the safety and economic efficiency of system operations. This phase effectively couples the historically trained model with the real-time operating environment, significantly enhancing the practicality and timeliness of power system risk assessment.

[0080] In addition, the hardware configuration of the present invention includes a computer equipped with a 6-core i7-9750H CPU and NVIDIA GeForce1650GPU, the compilation environment is Python 3.6.10, the operating system is Windows 10, the simulation tool uses MATLAB 2021b, and the MATPOWER toolkit is used to complete the traditional optimal power flow (OPF) calculation to determine the optimal load reduction.

[0081] The sample dataset used in the experiment was set up as follows: a total of 6,000 samples were generated, of which the training set contained 4,000 samples and the test set contained 2,000 samples. The training set contained 2,000 successful and unsuccessful instances, and the test set contained 1,000 successful and failed samples. After model training was completed, no additional state filtering was required for the subsequent calculation of risk assessment indicators. Finally, 1,000 data points were selected for the calculation of risk assessment indicators. When training the 2D-CNN model, the convolutional network was set up with two convolutional layers, a convolution kernel size of 3, a padding and stride of 1 and 3, respectively, 150 neurons per layer, and ReLU as the activation function. The training parameters were set to Epoch 200, Batch Size 64, Adam as the optimizer, the learning rate remained at the default value of 0.001, and the loss function was MAE.

[0082] To obtain the training set X that represents the system operation status and corresponding risk information train and verify X test , and a label dataset Y used to represent the optimal load reduction amount train and validation set Y test First, a set of possible system operating states is generated through Monte Carlo simulation, and the load reduction amount under each fault state is solved using the traditional OPF model as the target output. Subsequently, the generated data set is classified according to whether load reduction occurs and a data warehouse is established. Before inputting the data into the model of the present invention, the system state data needs to be preprocessed and the preprocessed data is passed to the model in a multi-channel form. At the same time, the corresponding load reduction amount is used as a label for model training. Finally, based on the trained model, the load reduction amount is predicted through real-time system information to achieve effective assessment of system risks.

[0083] To verify the technical effectiveness of the present invention, the system operation risk assessment method based on parallel computing was tested on the IEEE-RTS79 system. The IEEE-RTS79 system includes 24 nodes, 32 generator sets, and 38 branches, with a peak load of 2850MW. Specifically, the 38 branches include 5 transformer branches, 1 cable branch, and 32 transmission branches. Some of the generator sets were replaced with renewable energy, accounting for 20%, of which photovoltaics accounted for 47%.

[0084] The expected shortage of power supply (EDNS) and the power shortage probability (LOLP) of the power system are used as indicators for evaluating the risk of the power system (i.e., reliability indicators). In the reliability index calculation results, different line types represent different prediction models obtained when the current system is in different fault states, such as Figure 3 and Figure 4 As shown. Among them, "----MCS" represents the reliability index calculated by Monte Carlo simulation (MCS for short) and optimal power flow (OPF); "·····N-0" represents the model evaluation result when the current system is in the N-0 fault state; "-·-·-N-0(br36)" represents the model evaluation result when the No. 36 transmission line of the current system is in the N-1 fault state; "-··-··N-2" represents the model evaluation result when the current system is in the N-2 fault state; "-------N-1(G23)" represents the model evaluation result when the No. 23 generator is in the N-1 fault state; "···········PTM" represents the model evaluation result under the system operation risk assessment method based on parallel computing of the present invention, in which the optimizer is Adam, the maximum number of training cycles is 85, the learning rate is 0.01, and MSE is used as the loss function in the training process.

[0085] From time 0 to time 3, the system is in normal operation. At time 3, transmission line 36 fails and requires repair. The system continues operating until time 10. At this point, generator 23 fails, placing the system in the N-2 state. At time 15, transmission line 36 is repaired, leaving only generator 23 under maintenance. Finally, at time 22, generator 23 is repaired.

[0086] like Figure 2 and Figure 3As shown, when a system failure occurs, the proposed system operation risk assessment method based on parallel computing demonstrates excellent accuracy in capturing the degradation of system reliability caused by N-1 failures, substantially overlapping with the MCS results. This demonstrates high accuracy in assessing single-point failures and effectively reflects the surge in LOLP caused by multiple-point failures, with the curve consistent with the MCS. PTM exhibits excellent stability and adaptability when handling dynamic system failures. Compared to MCS, which often relies on extensive random simulations and can introduce random fluctuations, this is particularly true when sampling is insufficient or computing resources are limited.

[0087] The parallel computing method (PTM) proposed in this paper has significant advantages over traditional methods in calculating power system operation risks. By comparing it with Monte Carlo simulation (MCS), PTM can provide high-precision results similar to MCS in complex scenarios such as single-point failure (N-1) and multiple-point failure (N-2), while significantly reducing computational complexity and time overhead. Its calculation results are smoother, more stable and adaptable, and it is particularly suitable for scenarios with high real-time requirements, such as online reliability assessment and dynamic risk warning. It can quickly respond to changes in system faults and improve the efficiency and accuracy of system operation reliability assessment.

[0088] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A system operation risk assessment method based on parallel computing, characterized in that: include: S1. Obtain key operating parameters corresponding to the current moment of the power system; The operating parameters include: node admittance matrix, generator output matrix and energy demand matrix of each node; S2. Constructing equivalent matrices corresponding to the real and imaginary parts of the node admittance matrix, respectively, and simultaneously inputting the equivalent matrix corresponding to the real part, the equivalent matrix corresponding to the imaginary part, the generator output matrix, and the energy demand matrix of each node into a pre-trained load reduction prediction model to obtain a predicted value of the load reduction amount of the power system at the current moment; the load reduction prediction model is determined based on a two-dimensional convolutional neural network including multiple feature input channels; S3. Based on the predicted value of the load reduction amount of the power system at the current moment, an operation risk assessment of the power system is performed to obtain an operation risk assessment result of the power system at the current moment; S4. When the operation risk assessment result reaches a set threshold, an early warning signal is issued or an operation adjustment suggestion is generated.

2. The method according to claim 1, characterized in that The number of characteristic input channels of the load reduction prediction model is ≥4.

3. The method according to claim 1, characterized in that The training process of the load shedding prediction model includes: Obtaining a sample data set; the sample data set includes: key operating parameters at multiple moments, and a label value corresponding to each moment; the label value represents the actual load reduction corresponding to each moment; Divide the sample data set into training set and validation set according to the set ratio; Inputting the key operating parameters at each moment in the training set into the load reduction prediction model in sequence for semi-offline training to obtain the predicted value of the load reduction amount at each moment; Calculate the loss function corresponding to the predicted value of the load reduction at each moment. When the loss function meets the expected value, the pre-trained load reduction prediction model is obtained. The key operating parameters at each moment in the validation set are sequentially input into the pre-trained load reduction prediction model to optimize the pre-trained load reduction prediction model.

4. The method according to claim 3, characterized in that The key operating parameters at the multiple moments are obtained by performing multiple rounds of random sampling simulations on the annual load curve using the Monte Carlo simulation method.

5. The method according to claim 3, characterized in that Before dividing the sample data set into a training set and a validation set according to a set ratio, the training process of the load reduction prediction model further includes: The sample data set is preprocessed; the preprocessing includes data classification, feature extraction, outlier removal and normalization.

6. The method according to claim 5, characterized in that The preprocessing of the sample data set includes: According to whether there is load reduction, the sample data set is divided into reduction data and non-reduction data; Extracting core variables that affect the system operating status from the reduction data and non-reduction data respectively; the core variables include: the real and imaginary parts of the node admittance matrix, the generator active power, and the instantaneous load demand of each node; Invalid data and sampling data from abnormal sampling points are removed, and all core variables are normalized uniformly.

7. The method according to claim 1, characterized in that The power system uses the expected power shortage and the power shortage probability of the power system as indicators for operation risk assessment, which are expressed as follows: Where EDNS is the expected power shortage; LOLP is the power shortage probability of the power system; N sum is the number of samples, y i Indicates the predicted value of the load reduction corresponding to each sample.

Images