A method for enhancing power system resilience combined with pre-disaster warning mechanism

Abstract

A method for enhancing the resilience of a power system combined with a pre-disaster warning mechanism collects various historical parameter data of power equipment before abnormal weather occurs, various historical meteorological data of abnormal weather, and corresponding data on the degree of damage to power equipment after abnormal weather, and trains a prediction model to predict the post-disaster damage of power equipment. This allows targeted emergency maintenance or reinforcement of power equipment before extreme weather arrives, thereby reducing the impact of extreme weather on the power system and enhancing the resilience of the power system. The present invention achieves pre-disaster warning for power equipment, effectively reduces the damage to power equipment caused by extreme weather, provides an important reference for the implementation of post-disaster recovery strategies, and can quickly perform simultaneous prediction and analysis on a large number of power equipment, fully and efficiently utilizing precious pre-disaster prevention time, and can enhance the overall resilience of the power system in a targeted manner.

Description

Technical Field

[0001] The present invention belongs to the field of power systems and is a method for enhancing the resilience of power systems combined with a pre-disaster warning mechanism. Background Art

[0002] With the rapid development of my country's economy, society's demand for electricity is increasing. At the same time, individuals and businesses are placing new demands on the stability of power systems. However, with the intensification of climate change in recent years, extreme weather events have become more frequent, potentially causing significant damage to power systems and leading to widespread and prolonged power outages. Large-scale power outages often result in adverse social impacts, significant economic losses, multiple secondary disasters, and even significant casualties.

[0003] Power system resilience refers to the ability of a power grid to minimize losses during a major disaster and quickly restore normal power supply to normal conditions. Improving power system resilience is clearly of great significance to modern society. The main measures for improving power system resilience include preventive measures and recovery measures. Preventive measures involve adjusting the distribution network's operating mode before a disaster strikes, based on disaster forecast information and the power system's operating status, to minimize the scope of power outages and maintain continuous power supply to critical loads. Recovery measures involve gradually restoring power to critical loads during or after a disaster, allowing the distribution network to approach or reach its pre-disaster normal state as quickly as possible. Currently, strategies for improving power system resilience primarily focus on system reconstruction and black start research, which fall under the recovery measures of resilience improvement. However, research on preventive measures before disasters is relatively limited.

[0004] Currently, power companies have accumulated a large amount of critical data on power equipment, but few are using this vast amount of data to prepare for extreme weather events and enhance power system resilience. Therefore, to address these deficiencies, providing a method for enhancing power system resilience by integrating pre-disaster warning mechanisms has become an urgent challenge in this field. Summary of the Invention

[0005] The purpose of this invention is to use a pre-disaster warning mechanism to provide a method for enhancing the resilience of the power system, predict the extent of post-disaster damage to each power equipment before the arrival of extreme weather, and make emergency adjustments to the equipment based on the prediction results, thereby reducing the damage to power equipment in extreme weather conditions and enhancing the resilience of the power system.

[0006] The technical solution of the present invention is: a method for enhancing the resilience of power systems in combination with a pre-disaster warning mechanism. The method collects historical parameter data of power equipment before abnormal weather occurs, historical meteorological data of abnormal weather, and data on the extent of damage to power equipment after the abnormal weather occurs. A prediction model is trained to predict the post-disaster damage to power equipment. This allows targeted emergency maintenance or reinforcement of power equipment before extreme weather occurs, thereby reducing the impact of extreme weather on the power system and enhancing the resilience of the power system.

[0007] The prediction model includes two layers of global information fusion modules. First, a meteorological-equipment two-dimensional tensor is constructed based on the parameters of the power equipment before the abnormal weather and the data of the abnormal weather. The meteorological-equipment two-dimensional tensor is input into the prediction model. The two-dimensional tensor is divided into M sequences in chronological order. The sequence is position-encoded once to give the sequence time series information, and the input sequence t containing the position information is obtained. i , i = 1, 2, ..., M, the input sequence containing position information is sent to the first global information fusion module, the correlation information between the input sequences is extracted and reconstructed, and the global sequence a containing global information is obtained i , add a flag sequence a0 before the global sequence, perform position encoding on the flag sequence and the global sequence again, and then input the second global information fusion module. The corresponding sequence output constitutes the global feature matrix. The first sequence of the global feature matrix is ​​taken as the feature sequence used to determine the predicted damage classification. Then, it passes through a multi-layer perceptron MLP to output the predicted post-disaster damage.

[0008] The beneficial effects of the present invention are:

[0009] (1) A method for predicting the status of power equipment after a disaster is proposed.

[0010] At present, most of the methods for enhancing the resilience of power systems focus on post-disaster recovery, while a small number of pre-disaster prevention technologies for power equipment are mostly based on the historical data of individual devices to predict the future values ​​of the equipment, and obtain judgment results through predicted value analysis. However, the complexity of the power system requires accuracy of the prediction method on the one hand. The operation between devices is correlated, and the value of a single device is difficult to quickly reflect the damage of the power system, and further comprehensive analysis is required. On the other hand, the predicted device value cannot intuitively reflect the resilience state. Moreover, there are many devices in the power system. If the predicted device data is checked one by one, the workload is also very huge, and it is difficult to complete in a short time before the disaster. In response to the above shortcomings, the network model of the present invention combines historical power equipment data and extreme weather meteorological data for global prediction, and by introducing the design of the flag sequence, the numerical prediction is transformed into a category prediction, which intuitively shows the damaged state of the power equipment after the disaster, which is convenient for professionals to respond quickly, process quickly, and enhance the resilience of the power system in a targeted manner in advance. At the same time, the global information fusion module in the network is not only conducive to the extraction and fusion of the global deep information of the input data, improving the prediction accuracy of the model, but also the dimensionality reduction transformation operation in the module can effectively reduce the model calculation amount and accelerate the convergence speed of the model.

[0011] (2) The post-disaster status of equipment can be predicted in a short period of time.

[0012] The time between the release of an extreme weather forecast and its occurrence is often very short. Traditional pre-disaster prevention measures for power equipment involve comprehensive pre-disaster adjustments or reinforcement of equipment. This approach is time-consuming and lacks specificity, often preventing preventive measures from being completed before extreme weather has already occurred. The proposed pre-disaster warning scheme for power equipment consists of two parts: a network model training process and real-time network model prediction. The time-consuming network model training process is completed before extreme weather occurs, and the trained network model can quickly and simultaneously predict the impact of a large number of power equipment, thus fully and efficiently utilizing valuable pre-disaster prevention time. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of the present invention's pre-disaster warning and resilience enhancement process for power equipment.

[0014] Figure 2 Schematic diagram of a two-dimensional tensor of model input data in an embodiment of the present invention.

[0015] Figure 3 This is a structural diagram of the prediction model in an embodiment of the present invention.

[0016] Figure 4 Schematic diagram of the global information fusion module of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, the present invention provides a method for enhancing the resilience of power systems in combination with a pre-disaster warning mechanism. Specifically, various parameter data of power equipment before abnormal weather occurred in history, various meteorological data of abnormal weather in history, and data on the degree of damage to power equipment after abnormal weather are collected, and a prediction model is trained to predict the post-disaster damage to power equipment. In this way, targeted emergency maintenance or reinforcement of power equipment can be carried out before the extreme weather arrives, thereby reducing the impact of extreme weather on the power system and enhancing the resilience of the power system.

[0019] In this embodiment, typhoon weather is selected as a representative scenario of extreme weather, and transformers are selected as representative equipment in the power system. However, it should be noted that the extreme weather scenarios and power equipment to which the present invention can be applied include but are not limited to typhoons and transformers.

[0020] The embodiment of the present invention includes the following steps:

[0021] Step 1: Collect raw data for model training.

[0022] Meteorological data from historical typhoon periods is collected from the meteorological department. This data includes: maximum wind speed at the typhoon center, typhoon level, instantaneous maximum wind speed, maximum average wind speed, daily average rainfall, and daily average temperature. Transformer status parameter data from the two days before the historical typhoon is collected from the power company's Supervisory Control and Data Acquisition (SCADA) system. This status parameter data includes: internal temperature, internal humidity, internal SF6 gas concentration, operating power, operating current, and operating voltage. Data on the extent of transformer damage after the historical typhoon is collected. Because the specific manifestations of damage to power equipment vary, labels are required for each damage condition. Assuming that power equipment has Q levels of damage, Q types of labels are assigned. In this embodiment, the specific manifestations of transformer damage vary, which is not conducive to subsequent model training. Therefore, labels are assigned for each damage condition. Three types of damage labels are provided: undamaged normal operation, damaged abnormal operation, and damaged and stopped operation.

[0023] Step 2: Preprocess the raw data.

[0024] The transformer status parameter data for the two days before a historical typhoon was collected every two hours, meaning each status parameter had 24 sets of values. These values ​​may contain outliers and missing values, so data cleaning is required.

[0025] For the outliers, we first use the 3sigma method to determine whether it is an outlier point in a day. The judgment formula is as follows:

[0026]

[0027] Among them, q n,t represents the value at time t on day n, q n represents the mean of all sampling moments on day n, and σ represents the standard deviation of all sampling moments on day n. For points that meet the above formula, they are judged as outliers, and the outlier points are corrected based on the data of the two moments before and after them. The formula for correcting outliers is as follows:

[0028]

[0029] Among them, q n,t ′ is the corrected value at time t on day n, q n,t+1 and q n,t-1 are the values ​​at time t+1 and time t-1 on the nth day, respectively, q n,t+2 and q n,t-2 are the values ​​at time t+2 and time t-2 on day n, respectively. α1 and α2 are weight coefficients, and their sum should be equal to 1.

[0030] For missing values ​​in the original data, the sampling mean of the date is used to fill the missing values. The formula for filling missing values ​​is as follows:

[0031]

[0032] Among them, q n,t ′ is the value at time t on the nth day after filling, q n Represents the mean of all sampling times on day n.

[0033] After completing the correction and filling of abnormal values ​​and missing values ​​in the transformer state parameter data, it is necessary to normalize the six data sets of daily box temperature, box humidity, box SF6 gas concentration, box oxygen concentration, operating current, operating voltage, and operating power to eliminate the impact of the dimensions between the features and improve the accuracy of the model. The normalization method adopts the min-max normalization method. For any set of data, the normalization formula is as follows:

[0034]

[0035] in, represents the normalized value at time t on day n, G n,t Indicates the original value at time t on day n, maxG n and minG nRepresent the maximum and minimum values ​​of all sampling moments on day n, respectively.

[0036] Step 3: Build and train the prediction model.

[0037] The network structure of the prediction model is as follows Figure 3 As shown in FIG, it mainly includes: deploying two layers of global information fusion units, performing two position encodings, adding a flag sequence, and using a multi-layer perceptron with a Softmax function in the output layer to output categories.

[0038] The input data of the model is a two-dimensional tensor of weather-equipment. Each day is divided into T time points with equal time intervals. Any parameter of the power equipment N days before the abnormal weather is represented as a one-dimensional tensor with a length of P = N * T. If the number of parameters of the power equipment is K, then the parameter data of each device is represented as a two-dimensional tensor with P rows and K columns. The W meteorological data of the abnormal weather are spliced ​​to the end of each row of the two-dimensional tensor to form a two-dimensional tensor of weather-equipment with P rows, W + K columns. The schematic diagram can be referred to Figure 2 This embodiment samples the transformer status parameters 2 days before the occurrence of historical abnormal weather every 2 hours, and the time information is Figure 2 The left half of the line indicates the transformer status parameters include: box temperature, box humidity, box SF6 gas concentration, working power, working current, working voltage, these 6 groups of status parameters are represented by Figure 2 The left half of the column represents the historical typhoon meteorological data, including: the maximum wind speed at the typhoon center, typhoon level, instantaneous maximum wind speed, maximum average wind speed, daily average rainfall, and daily average temperature. These six sets of meteorological data are Figure 2 The right half of the column represents. Figure 2 The left and right parts are concatenated row by row, and finally a two-dimensional tensor (24 rows and 12 columns) of model input data is obtained.

[0039] The two-dimensional tensor is divided into M sequences in time order, and the sequence is position-encoded once to give the sequence time sequence information, and the input sequence t containing position information is obtained. i , i = 1, 2, ..., M, the input sequence containing position information is sent to the first global information fusion module, the correlation information between the input sequences is extracted and reconstructed, and the global sequence a containing global information is obtained i , add a flag sequence a0 before the global sequence, perform position encoding on the flag sequence and the global sequence again, and then input the second global information fusion module. The corresponding sequence output constitutes the global feature matrix. The first sequence of the global feature matrix is ​​taken as the feature sequence used to determine the predicted damage classification. Then, it passes through a multi-layer perceptron MLP to output the predicted post-disaster damage.

[0040] In this example, the two-dimensional tensor of input data is segmented chronologically and can be viewed as 24 sequences, each of length 12, including six transformer parameter data and six typhoon meteorological data. The sequences are first positionally encoded to impart temporal information, enabling the subsequent global information fusion module to better extract deeper insights from the data. Specifically, absolute position encoding is employed, whereby all sequences are assigned a learnable pairing sequence of equal length and added to the original sequence to produce a sequence containing positional information.

[0041] The sequence containing position information is sent to the global information fusion module to extract the correlation information between the sequences and reconstruct it to obtain a fusion sequence containing global information. Figure 4 This module includes feature normalization, association mechanism unit, fully connected layer and residual connection. First, the input sequence of the module is feature normalized, that is, the standardized sequence value is calculated based on the mean E and variance V of each sample data. The calculation formula is as follows:

[0042]

[0043] Where X represents the two-dimensional tensor consisting of the input sequence of a certain sample, E(X) represents the mean of the sample two-dimensional tensor, V(X) represents the variance of the sample two-dimensional tensor, ε is a protection constant, which is a very small constant used to prevent the denominator from being zero during the calculation process, and Y represents the two-dimensional tensor consisting of the output sequence.

[0044] After the normalization, the sequence is calculated through the association mechanism unit to carry out the correlation information calculation and data fusion between the sequences. The main process is that the input sequence consists of a 24*12 two-dimensional tensor Y and is multiplied by three different parameter matrices W of 12*12 in size. q 、W k 、W v , we get the query matrix Q, key matrix K and value matrix V, all of size 24*12, which can be expressed as follows using matrix multiplication:

[0045] Q=YW q

[0046] K=YW k

[0047] V=YW v

[0048] Next, the key matrix K and the value matrix V are subjected to dimensionality reduction operations to reduce the amount of network computation and accelerate the convergence of the model. The specific formula is as follows:

[0049] Z′=R(Z,a)W X '

[0050] Among them, Z represents the key matrix or value matrix before transformation, a represents the dimension reduction coefficient, R represents the shaping operation, and W X ' represents the transformation matrix, and Z' represents the transformed key matrix or value matrix. In this embodiment, the original 24*12 key matrix or value matrix is ​​reshaped into Then let it be combined with 12a*12 W X ′, and finally we get The reduced dimension key matrix or value matrix.

[0051] Multiply the query matrix Q by the transpose of the dimension reduction key matrix K′ to obtain The correlation matrix A is obtained; after performing a Softmax operation on A, the normalized correlation matrix A′ is obtained.

[0052] Finally, the association matrix A′ is multiplied by the dimension reduction value matrix V′ to obtain the output matrix O. The row height of the output matrix O is 24 and the column width is 12. The output sequence can be obtained by splitting it in row order.

[0053] In order to supplement the shallow information lost during the calculation process and enable the output sequence of the global information fusion module to take into account both global and local features, the output sequence of the association mechanism unit is residually added to the input sequence of the original global information fusion module, and then input into the fully connected layer and the output is standardized. Finally, the residual is added again to obtain the final output sequence of the first-layer global information fusion module.

[0054] The calculation process of the position encoding and global information fusion module of the second layer is the same as that of the first layer. The difference is that the input sequence of the second layer global information fusion module, in addition to the output sequence of the first layer global information fusion module, also adds a flag sequence, which is located at the first position of the second layer position encoding, that is, the 0th position. The initial value of the flag sequence can be random, for example, all 0s at the beginning. The flag sequence is a learnable sequence. The learnability refers to the W to be multiplied after the flag sequence enters the global information fusion unit. q 、W k 、W v Parameter matrices, the parameters in these matrices are changing, that is, they will change in the direction of making the loss function value smaller. The role of introducing the flag sequence is that when performing global information fusion network operations, the input sequence will interact with all other sequences, and the flag sequence is a group of sequences outside the original data, which can avoid the problem of being biased towards certain specific original sequences during the operation process, objectively aggregate the information of all input sequences, and thus better fuse the global information. The output of the second global information fusion module in the prediction model of the present invention is also a sequence with the same number of input sequences. All output sequences constitute a global feature matrix, and its first sequence, that is, the sequence output at the corresponding position of the flag sequence, is taken as the feature sequence.

[0055] In this embodiment, after position encoding is completed for the flag sequence and the output sequence of the first-layer global information fusion module, the same calculation steps as described above for the first layer are used to obtain an output matrix with a row height of 25 and a column width of 12. The elements of the first row are taken as the feature sequence and fed into a multi-layer perceptron with a Softmax function in the output layer to obtain the probabilities of the three transformer states: namely, the probabilities of the transformer being in normal operation (undamaged), operating abnormally (damaged), and stopped (damaged).

[0056] After the historical typhoon weather ended, the labels of the actual damage levels of the transformers were one-hot encoded and binarized. That is, the true value for an intact transformer operating normally was "100," the true value for a damaged transformer operating abnormally was "010," and the true value for a damaged transformer not operating properly was "001." The prediction model was trained by calculating the loss function using the cross-entropy formula. The loss function formula is as follows:

[0057]

[0058] Where m is the number of samples, Indicates the value of the jth bit of the true value of the kth sample, The j-th bit value represents the predicted probability value of the k-th sample.

[0059] Through the gradient descent algorithm, the prediction model parameters are continuously updated to make the loss function value continue to decrease. When the loss function value tends to be stable or less than the set threshold, the parameters in the model stop updating and the model training process ends.

[0060] Step 4: Predict equipment status before a disaster.

[0061] When receiving extreme weather forecast information from the meteorological department, the power company can input the predicted extreme weather meteorological data and the status parameters of all transformers under its jurisdiction two days ago into the trained prediction model, and output the predicted damage labels of the transformers in its jurisdiction. Based on the different damage conditions described by the labels, the power company can arrange professionals to carry out emergency inspections or reinforcements on transformers with the prediction label of "damaged and stopped working", thereby enhancing the resilience of the power system in advance.

Claims

1. A method for enhancing the resilience of power systems combined with a pre-disaster warning mechanism, characterized by: Collect historical parameter data of power equipment before abnormal weather occurs, historical meteorological data of abnormal weather, and corresponding data on the extent of damage to power equipment after abnormal weather. Train a prediction model to predict the post-disaster damage of power equipment. This allows targeted emergency maintenance or reinforcement of power equipment before extreme weather arrives, thereby reducing the impact of extreme weather on the power system and enhancing the resilience of the power system. The prediction model includes two layers of global information fusion modules. First, a meteorological-equipment two-dimensional tensor is constructed based on the parameters of the power equipment before the abnormal weather and the data of the abnormal weather. The meteorological-equipment two-dimensional tensor is input into the prediction model. The two-dimensional tensor is divided into M sequences in chronological order. The sequence is position-encoded once to give the sequence time series information, and the input sequence t containing the position information is obtained. i , i = 1, 2, ..., M, the input sequence containing position information is sent to the first global information fusion module, the correlation information between the input sequences is extracted and reconstructed, and the global sequence a containing global information is obtained i , add a flag sequence a0 before the global sequence, perform position encoding on the flag sequence and the global sequence again, and then input the second global information fusion module. The corresponding sequence output constitutes the global feature matrix. The first sequence of the global feature matrix is ​​taken as the feature sequence used to determine the predicted damage classification. Then, it passes through a multi-layer perceptron MLP to output the predicted post-disaster damage.

2. The method for enhancing the resilience of a power system combined with a pre-disaster warning mechanism according to claim 1 is characterized by global The information fusion module performs feature standardization on the sequence of module input, that is, according to the mean E and variance V of each sample data, the standardized sequence value is calculated. The calculation formula is as follows: Among them, X represents the two-dimensional tensor composed of the sequence of a certain sample, E(X) represents the mean of the two-dimensional tensor, V(X) represents the variance of the two-dimensional tensor, ε is a protection constant used to prevent the denominator from being 0 during the calculation process, and Y represents the two-dimensional tensor composed of the output sequence. The standardized sequence is calculated through the association mechanism unit to perform inter-sequence correlation information and data fusion. The process is: the two-dimensional tensor Y composed of the input sequence is multiplied by three different parameter matrices W q 、W k 、W v , we get the query matrix Q, key matrix K and value matrix V: Q=YES q K=YES k V=YES v Next, the key matrix K and the value matrix V are subjected to dimensionality reduction operations to reduce the amount of network computation and accelerate the convergence of the model. The specific formula is as follows: Z′=R(Z,a)W X ′ Among them, Z represents the key matrix or value matrix before transformation, a represents the dimension reduction coefficient, R represents the shaping operation, and W X ′ represents the transformation matrix, and Z′ represents the key matrix or value matrix after transformation; The query matrix Q is multiplied by the transpose of the dimensionality reduction key matrix K′ to obtain the association matrix A; a Softmax operation is then performed on A to obtain the normalized association matrix A′; finally, the association matrix A′ is multiplied by the dimensionality reduction value matrix V′ to obtain the output matrix O, which is split in row order to obtain the sequence output by the global information fusion module.

3. The method for enhancing the resilience of a power system combined with a pre-disaster warning mechanism according to claim 1 is characterized by: The position encoding in the prediction model uses absolute position encoding.

4. The method for enhancing the resilience of a power system combined with a pre-disaster warning mechanism according to claim 1 is characterized by: The construction of the meteorological-equipment two-dimensional tensor is as follows: divide each day into T time points with equal time intervals. Any parameter of the power equipment N days before the abnormal weather is expressed as a one-dimensional tensor with a length of P = N*T. Let the parameter number of the power equipment be K, then the parameter data of each device is expressed as a two-dimensional tensor with P rows and K columns. The W meteorological data of the abnormal weather are spliced ​​to the end of each row of the two-dimensional tensor to form a meteorological-equipment two-dimensional tensor with P rows, W and K columns.

5. The method for enhancing power system resilience in combination with a pre-disaster warning mechanism according to claim 1, wherein collecting data comprises: Meteorological data of historical abnormal weather events are collected from the meteorological department, and status parameter data of power equipment N days before the historical abnormal weather events are collected from the power company's data acquisition and monitoring control system SCADA. At the same time, data on the degree of damage to power equipment after the abnormal weather events end are collected. Among them, damage level labels are set for the damage to power equipment, indicating Q types of damage levels of power equipment.

6. The method for enhancing the resilience of a power system combined with a pre-disaster warning mechanism according to claim 1 is characterized by: The collected data is pre-processed. First, the status parameter data of the power equipment N days before the historical abnormal weather is cleaned, including abnormal data correction and missing data filling; Secondly, each state parameter is normalized to eliminate the influence of each parameter dimension.

7. The method for enhancing the resilience of a power system combined with a pre-disaster warning mechanism according to claim 6 is characterized by: Data cleaning is specifically as follows: 1) Abnormal data correction: First, use the 3sigma method to determine whether it is an abnormal value point. The judgment formula is as follows: Among them, q n,t It represents the value at time t on day n. represents the mean of all sampling moments on day n, and σ represents the standard deviation of all sampling moments on day n. For points that meet the above formula, they are judged as outlier points, and the outlier points are assigned correction values ​​based on the data of the two moments before and after them. The formula for correcting outliers is as follows: Among them, q n,t ′ is the corrected value at time t on day n, q n,t+1 and q n,t-1 are the values ​​at time t+1 and time t-1 on the nth day, respectively, q n,t+2 and q n,t-2 are the values ​​at time t+2 and time t-2 on day n, respectively. α1 and α2 are weight coefficients, and their sum should be equal to 1. 2) Missing data filling: For missing values ​​in the original data, the sampling mean of the date is used to fill the missing values. The formula for filling missing values ​​is as follows: Among them, q n,t ′ is the value at time t on the nth day after filling, Represents the mean of all sampling times on day n.

8. The method for enhancing the resilience of a power system combined with a pre-disaster warning mechanism according to claim 6 is characterized by: The min-max normalization method is used to normalize the state parameters. The normalization formula for any set of data is as follows: in, represents the normalized value at time t on day n, G n,t Indicates the original value at time t on day n, maxG n and minG n Represent the maximum and minimum values ​​of all sampling moments on day n, respectively.

9. The method for enhancing the resilience of a power system combined with a pre-disaster warning mechanism according to claim 1 is characterized by: When receiving extreme weather forecast information from the meteorological department, the predicted extreme weather meteorological data and the status parameters of the power equipment within the jurisdiction N days ago are input into the trained prediction model, and the maximum probability of damage level of each power equipment within its jurisdiction is predicted and output. According to different damage situations, the power system conducts targeted emergency maintenance or reinforcement of the power equipment.

Images