Weather and circuit fault correlation analysis method based on back propagation neural network

Abstract

The invention discloses an extreme weather and fault correlation analysis and extraction method based on back propagation neural network analysis, and belongs to the field of meteorological research, and the method comprises the steps: obtaining regional meteorological station data and fault data; carrying out numerical statistical calculation on the meteorological station data according to a specified extreme weather definition index, and carrying out spatial correlation matching; dividing a sample set; constructing a neural network correlation analysis model, and training a network model by using the sample set to obtain a trained model; performing global testing by using the trained model to obtain a correlation analysis result of global extreme weather and faults, and forming security level classification; and performing spatial visualization and thematic map generation on a global association result by utilizing ArcMap geographic mapping software. According to the invention, through inputting the extreme weather index value and the fault quantity data, training based on the reverse neural network correlation analysis model is carried out, and rapid correlation analysis and visual result display of regional extreme weather and faults are realized.

Description

Technical Field

[0001] The invention belongs to the field of meteorological research and relates to a method for analyzing and extracting the correlation between extreme weather and circuit faults based on genetic algorithm back propagation neural network analysis. Background Art

[0002] my country has a vast territory and complex geographical conditions. It is often threatened by natural disasters such as typhoons, floods, high temperatures, and freezing. According to statistics, power grid failures caused by meteorological disasters have accounted for more than 60% of the total number of failures. In July 2021, Zhengzhou, Henan Province, suffered a rare torrential rainstorm in history, with extremely high rainfall in a short period of time. This rainstorm caused serious impact on the power system in Zhengzhou and surrounding areas, and the power grid almost collapsed, resulting in large-scale power outages. In July 2021, Typhoon Fireworks landed in Guangdong Province, causing strong winds and heavy rains, resulting in damage to power facilities, large-scale power outages, and millions of people were affected. In June 2020, Jiangxi Province suffered continuous heavy rainfall, which caused damage to the power system, submerged transmission lines, and large-scale power outages, posing a threat to the safe operation of the power grid. In 2019, a large-scale ice and snow disaster occurred in Yunnan Province, resulting in ice covering the transmission lines, the collapse of line towers, and a major impact on the power system, with large-scale power outages. Therefore, in response to such "low probability, high risk" extreme weather disasters, there is an urgent need to systematically and scientifically study the challenges and problems faced by power systems under extreme weather conditions such as extreme heat, extreme cold, and extreme drought, establish a boundary prediction model for extreme weather, and study the risk identification technology of power grid failures under extreme weather conditions.

[0003] At present, the ability to assess the risk situation of power grid operation under extreme weather conditions is insufficient. The research and analysis of circuit fault areas is still mainly based on manual field inspections. This method that relies on manual field inspections can directly and accurately observe the fault situation, but the inspection process requires a lot of manpower, material resources and time to conduct a comprehensive detection of power faults, which is costly. In addition, there is a lack of multi-time scale risk prediction under extreme weather conditions. The existing means fail to explore the correlation between various influencing factors and the operation status of the power grid, and it is difficult to achieve deep integration of multiple data sources and risk situation assessment through the current analysis mechanism. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides an extreme weather and fault correlation analysis and extraction method based on genetic algorithm back propagation neural network analysis. The present invention realizes sample set construction by inputting extreme weather index values ​​and fault quantity data, and conducts training based on the reverse neural network correlation analysis model. It can realize rapid correlation analysis between extreme weather and faults in the target area, help to analyze the spatial distribution of circuit faults in the target area, establish a boundary model of extreme weather in the target area, determine the weight of the impact of different extreme weather on faults, and carry out operation warning and auxiliary decision-making applications based on the results.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: The first aspect of the present invention provides an extreme weather and fault correlation analysis calculation based on genetic algorithm back propagation neural network analysis.

[0006] According to the current mechanism research theory of extreme weather events, obtain observation data of conventional meteorological observation stations for extreme weather and calculate the extreme weather research method: obtain the historical meteorological data observed and counted by the national meteorological observation station and the meteorological bureau in the target study area, including hourly rainfall, hourly average temperature, hourly maximum temperature, hourly minimum temperature, hourly maximum wind speed, instantaneous wind speed, average wind speed, etc. Obtain the shp data of the geographical boundary range of the study area. Obtain the observation numerical results provided by the meteorological observation station in the study area and calculate and process them according to the following extreme weather research indicators, including the following 9 input features: Maximum consecutive dry days CDD: The maximum number of consecutive days without precipitation in a year.

[0007] Maximum Consecutive Wet Days (CWD): The maximum number of consecutive days with precipitation in a year.

[0008] Total annual precipitation: the total precipitation in a year.

[0009] Average annual precipitation intensity SDII: the ratio of total annual precipitation to wet days.

[0010] Heavy precipitation days R25mm and R50mm: days with daily precipitation greater than or equal to 25mm and 50mm.

[0011] Heavy precipitation rate R95p: the ratio of the annual cumulative precipitation with daily precipitation greater than the 95% percentile to the annual total precipitation.

[0012] High Temperature Days HTD: The number of days in a year with a maximum daily temperature greater than or equal to 35 degrees Celsius.

[0013] Frost days FD: The number of days in a year with a minimum daily temperature of less than or equal to 0 degrees Celsius.

[0014] Fault data collection: Obtain the location information, number of faults and corresponding fault voltage of each fault area in the study area, and record the time, location and type of fault.

[0015] The converted extreme weather index data is spatially correlated with the power failure data, usually processed using ArcMap geographic information processing software, to ensure that the fault data in each study area matches the corresponding meteorological data. To achieve sample set construction, the extreme weather index data and the corresponding power failure data are pre-classified to construct a sample set. The sample set is divided into training set, validation set and test set to ensure the generalization and stability of the model. Based on the definition of extreme weather, the pre-classification is mainly based on the index data and the number of faults in the fault area. All data sets are U. The classification is based on whether there is a spatial intersection between the national meteorological observation station and the meteorological bureau and the fault area, that is, whether a fault occurs at the location where the extreme weather numerical results are processed. Based on this, it is divided into two categories, one is U T Indicates that there is a spatial intersection, one type is U F Indicates that there is no spatial intersection. Select U T Serves as a sample range set for model testing.

[0016] The storage module is configured to: display the extreme weather index values ​​and circuit fault conditions in the form of geographic data, and store the numerical results in a database for subsequent analysis and query.

[0017] The extreme weather and fault correlation analysis based on back propagation neural network analysis builds a GA-BP neural network model based on genetic algorithm. The network structure is defined as: In the range U T Select the model test sample range set, construct input-output pairs based on weather indicators and fault data, take weather indicators A=[a1,a2,a3…,a9] as input, and the number of power faults Z=[z1,z2,z3,…,zi] under different voltages as output to construct a sample set. T Divide into training set, validation set and test set.

[0018] Input layer: contains 9 nodes, corresponding to the 9 extreme weather indicators you defined: CDD, CWD, total annual precipitation, SDII, R25mm, R50mm, R95p, HTD and FD.

[0019] Output layer: 1 node, the number of power failures divided by different voltages.

[0020] Hidden layer: Use 1 hidden layer. Use empirical methods to determine the appropriate neuron node Q and quantity. The empirical method formula is , select 7 neurons.

[0021] The method for optimizing the initial weights of the extreme weather and fault correlation analysis based on the genetic algorithm back propagation neural network analysis GA-BP neural network genetic algorithm is: Chromosome encoding. Encode the initial weights and biases of the BP neural network into chromosomes. The encoding can be binary, real valued, or mixed.

[0022] Population initialization. Randomly generate chromosomes of the initial population, each chromosome represents a possible combination of weights and biases.

[0023] Fitness function. The fitness function is used to evaluate the performance of each chromosome and calculate their effectiveness in minimizing the circuit fault prediction error, defined as .

[0024] Genetic operation. Selection: Select the best individual as the parent according to the fitness function.

[0025] Crossover: Combining genes from two parent chromosomes to produce new offspring.

[0026] Mutation: The introduction of random variation into some offspring to maintain genetic diversity.

[0027] Iteration: Through multiple generations of genetic operations, until the population converges to the optimal or near-optimal combination of weights and biases.

[0028] The parameter setting and training of the GA-BP neural network genetic algorithm for extreme weather and fault correlation analysis based on genetic algorithm back propagation neural network analysis are as follows: Activation function setting: The hidden layer uses the nonlinear activation function ReLU. The output layer uses a linear activation function because the output is a continuous value of the number of failures.

[0029] Loss function setting: Use mean square error MSE as the loss function, which is suitable for regression problems.

[0030] Learning rate setting: The learning rate is set to 0.001, Training process settings: Use the initial weights and biases optimized by GA as the initial state of the BP neural network; implement the backpropagation algorithm to further minimize the error by adjusting the weights and biases; use gradient descent and its advanced variants such as Adam to update the weights; to prevent overfitting, monitor the loss of the validation set and stop training early when the loss no longer decreases.

[0031] The GA-BP neural network genetic algorithm test scheme for extreme weather and fault correlation analysis based on genetic algorithm back propagation neural network analysis is set as follows: The dataset is divided into training set, validation set and test set, with 70% for training, 15% for validation and 15% for testing.

[0032] Model evaluation: Evaluate model performance on the test set and use R², MAE, and MSE indicators to evaluate the effect of the model.

[0033] In the above formula is the predicted value, is the true value is the mean, and m is the number of samples.

[0034] The importance assessment of the global weight of the extreme weather and fault correlation analysis GA-BP neural network genetic algorithm based on genetic algorithm back propagation neural network analysis: After training is completed, the contribution of each input feature to the output failure count is calculated, and the importance of each weather indicator is evaluated by analyzing the trained weights. .

[0035] The contributions of each layer are accumulated to quantify the relative importance of each input feature.

[0036] Assume that the 9 nodes in the input layer are A=[a1,a2,a3…,a9], and the 7 nodes in the hidden layer are H=[h1,h2,h3…,h7]. The weight matrix from the input layer to the hidden layer is ,in Indicates that from the input node To the hidden layer nodes The weight of .

[0037] in, is the activation function, is a hidden layer node The bias.

[0038] Assume that the output layer nodes are , the weight matrix from the hidden layer to the output layer is ,in Represents the hidden layer nodes The weight to the output layer node z.

[0039] in, Represents the bias of the output layer node.

[0040] Specifically, the contribution of each input feature to the output can be calculated by the following steps.

[0041] Calculate the contribution of each hidden layer node to the output: Among them, C j is a hidden node Contribution to output.

[0042] Calculate the contribution of each input feature to the hidden layer nodes: Among them, C ij is the input feature For hidden nodes contribution.

[0043] Compute the total contribution of each input feature to the output: Among them, C i is the input feature The total contribution to the output z.

[0044] Normalization is performed. In order to compare the contribution of each input feature, each C i Normalize so that all C i The sum is 1: Contribution like If it is 0, it means that the meteorological index has no effect on the fault. If it is 1, it means that the meteorological indicator has 100% correlation influence. The closer it is to 0, the smaller the risk of the event. The closer it is to 1, the greater the risk of the event. Using this function as a confidence calculation method can effectively express the credibility of this type of factor. Through the above steps, the relative importance of each weather indicator on the number of power failures can be quantified, and the global weights of each input feature can be obtained. These weights reflect the influence in the entire network and can provide a basis for further decision analysis. The fault correlation pre-storage module is configured to store and calculate the trained extreme weather index values, circuit fault prediction conditions and node weight values. The trained GA-BP neural network model is used to batch process the data of all observation points in the study area to generate global fault prediction results.

[0045] The second aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the extreme weather and fault correlation analysis and calculation steps based on back-propagation neural network analysis as described in the first aspect of the present invention are implemented.

[0046] The third aspect of the present invention provides a method for dividing the model security level based on the correlation analysis results of the entire domain, dividing the security level based on the influence of each feature on the model output, and visualizing the thematic map. The inverse distance weighted interpolation method is used to generate thematic maps or heat maps in the study area. The z-score and p-value obtained can be used to know the locations where high-value or low-value elements are clustered in space. The steps and methods are used to intuitively show the degree of influence of different weather indicators on power failures.

[0047] Data input. Import the global analysis results into ArcMap geographic mapping software, including the geographic boundary range data of the study area in shp format, the calculated values ​​of extreme weather research indicators, fault prediction data, and the weight analysis results of each indicator.

[0048] According to the total contribution weight C of each input feature i , statistically analyze the weights of all features, divide the weights into several levels, and use the standard deviation method to divide them. The following is the specific division method: High Impact: Features with weights greater than the mean plus one standard deviation are considered high impact.

[0049] Medium Impact: Features with weights between plus or minus one standard deviation from the mean are considered to have medium impact.

[0050] Low Impact: Features with weights less than the mean minus one standard deviation are considered low impact.

[0051] in, is the standard deviation, is the weighted average.

[0052] Generate thematic maps. Use the spatial analysis tool of the software and adopt the inverse distance weighted method to analyze the number of faults and their weight impact values. Interpolation analysis is performed to obtain global numerical results, and a heat map or spatial distribution map of the impact of different weather indicators on power failures is generated. The heat map can show the correlation strength between the fault density in different areas and specific weather indicators.

[0053] Where Z(x,y) represents the interpolation result of the number of power failures at the point in the area with the longitude and latitude (x,y). is the distance between the known sample point and the point to be interpolated, is the number of known sample power failures, k is 2, Indicates contribution.

[0054] Interpretation of results. Combined with the generated thematic maps, analyze which areas of power failure are most vulnerable to certain extreme weather events. In-depth interpretation of the results in key areas, and provide targeted disaster prevention measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0056] Figure 1 This is a calculation flow chart of extreme weather and fault correlation analysis based on back propagation neural network analysis in an embodiment of the present invention.

[0057] Figure 2 It is a schematic diagram of a BP neural network model construction in extreme weather and fault correlation analysis based on back propagation neural network analysis in an embodiment of the present invention.

[0058] Figure 3 It is a network model framework for extreme weather and fault correlation analysis based on back propagation neural network analysis in an embodiment of the present invention.

[0059] Figure 4 This is a schematic diagram of a data table for spatial correlation matching between the converted extreme weather research indicators and power failure data in an embodiment of the present invention.

[0060] Figure 5 Schematic diagram of the classification mode of weight levels of various features in an embodiment of the present invention.

[0061] Figure 6 A schematic diagram of the spatial distribution of fault prediction results for the entire research domain generated in an embodiment of the present invention.

[0062] Figure 7 To generate a thermal map of fault prediction results for the entire research domain in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0064] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0065] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0066] Embodiment 1 like Figure 1 As shown, this embodiment discloses a flowchart of the entire process of extreme weather and fault correlation analysis based on genetic algorithm back propagation neural network analysis, including: Step 1: Collect statistical data of meteorological observation stations in the study area, the scope of the study area, and fault data in the study area.

[0067] Step 2: Perform numerical statistics and conversion on the statistical data of meteorological observation stations according to the specified extreme weather definition indicator format, and spatially associate and match the fault data with the geographical scope of the study area to ensure that each fault record can be aligned with the corresponding meteorological data.

[0068] Step 3: Spatial correlation matching is performed between the fault data and the geographic scope of the study area to ensure that each fault record can be aligned with the corresponding meteorological data.

[0069] Step 4: Divide and mark the extreme weather index data and the corresponding fault quantity data, and divide the sample set into a training set, a validation set, and a test set, which are used for model training, model tuning, and model performance evaluation, respectively.

[0070] Step 5: To analyze the correlation between extreme weather and power failures, a GA-BP neural network correlation analysis model is constructed. The input layer contains 9 nodes, corresponding to 9 extreme weather indicators; the hidden layer contains 7 neuron nodes; the output layer is 1 node, representing the number of failures. Then the model is trained, and the genetic algorithm is used to optimize the initial weights and biases of the BP neural network, and the model is trained with data from the sample set. During the training process, the model continuously adjusts the parameter learning rate and the number of neurons in the hidden layer to minimize the error, and finally obtains the importance evaluation of the global weight, and the model is verified and evaluated with the test set.

[0071] Step 6: Use the trained GA-BP neural network model to conduct global training and construction for the entire study area to obtain the correlation analysis results between each extreme weather index and faults in the study area. The analysis results cover the fault prediction of all relevant locations in the study area.

[0072] Step 7: Analyze and visualize the results, analyze the weight impact analysis, and build a model and safety level classification. Analyze the global weight of the model to evaluate the relative importance of each extreme weather indicator to power failures; based on the impact analysis of the global weight, build a safety level classification of power failure risks to provide a reference for power grid disaster prevention and mitigation; use ArcMap geographic information system software to perform spatial analysis on the correlation analysis results in the study area and generate thematic maps and heat maps. These graphical results intuitively show the impact area of ​​extreme weather on power failures. As an embodiment, a method for analyzing the correlation between extreme weather and faults based on back propagation neural network analysis is specifically as follows: Step 1: Collect data.

[0073] In this example, we selected a coastal area as the research area and collected the following data for this area: Meteorological observation station statistics: Statistical data of meteorological observation stations in the region were collected, including meteorological parameters of hourly rainfall, hourly average temperature, hourly maximum temperature, hourly minimum temperature, hourly maximum wind speed, instantaneous wind speed, and average wind speed over the past decade.

[0074] Study area scope data: The shp file of the geographical boundary scope data of the area was obtained.

[0075] Fault data in the study area: The power fault records in the area were collected, including the time, location, type of fault and the corresponding fault voltage level.

[0076] Step 2: Data processing.

[0077] After data collection is completed, the following processing operations are performed: Meteorological data processing: According to the definition indicators of extreme weather, the statistical data of meteorological observation stations are numerically counted and converted to form 9 extreme weather indicators: Maximum consecutive dry days CDD: The maximum number of consecutive days without precipitation in a year.

[0078] Maximum Consecutive Wet Days (CWD): The maximum number of consecutive days with precipitation in a year.

[0079] Total annual precipitation: the total precipitation in a year.

[0080] Average annual precipitation intensity SDII: the ratio of total annual precipitation to the number of wet days (i.e. days with daily precipitation greater than 1 mm).

[0081] Heavy precipitation days R25mm and R50mm: days with daily precipitation greater than or equal to 25mm and 50mm.

[0082] Heavy precipitation rate R95p: the ratio of the annual cumulative precipitation with daily precipitation greater than the 95% percentile to the annual total precipitation.

[0083] High Temperature Days HTD: The number of days in a year with a maximum daily temperature greater than or equal to 35 degrees Celsius.

[0084] Frost days FD: The number of days in a year with a minimum daily temperature of less than or equal to 0 degrees Celsius.

[0085] Spatial association of fault data: Spatial association and matching of fault data with the geographic scope of the study area ensures that each fault record can be aligned with the corresponding meteorological data.

[0086] Figure 5 The numerical results of the spatial correlation matching between the extreme weather index data and the power failure data after partial conversion in the experimental use case are shown.

[0087] Step 3: Sample set construction.

[0088] Based on the processed extreme weather index data and the corresponding fault data, a sample set is constructed. Each sample contains the following contents: Based on the definition of extreme weather, the pre-classification is mainly based on the indicator data and the number of faults in the fault area. The use case data set is U. The classification is based on whether there is a spatial intersection between the national meteorological observation station and the meteorological bureau and the fault area, that is, whether the location of the extreme weather numerical result processed by the classification has a fault. Based on this, it is divided into two categories, U T Indicates that there is a spatial intersection, U F Indicates that there is no spatial intersection. Select U T As the model test sample range set. In the range U T Select the model test sample range set and build the sample set. T Divide into training set, validation set and test set, with 70% training, 15% validation and 15% testing.

[0089] Step 4: Build a GA-BP neural network model and use the test set to evaluate the trained GA-BP neural network model.

[0090] The GA-BP neural network model is constructed based on the correlation analysis between extreme weather and faults through back propagation neural network analysis. The specific network structure of this use case is defined as follows: In the range UT Select the model test sample range set, construct input-output pairs based on weather indicators and fault data, take weather indicators A=[a1,a2,a3…,a9] as input, and the number of power faults Z=[z1,z2,z3,…,zi] under different voltages as output to construct a sample set. T Divide into training set, validation set and test set.

[0091] Input layer: contains 9 nodes, corresponding to the 9 extreme weather indicators you defined: CDD, CWD, total annual precipitation, SDII, R25mm, R50mm, R95p, HTD and FD.

[0092] Output layer: 1. The number of power failures at each node divided by different voltages.

[0093] Hidden layer: Use 1 hidden layer. Use empirical methods to determine the appropriate neuron nodes. The empirical method formula is , so 7 neurons are chosen.

[0094] The step 4 is based on constructing a GA-BP neural network model genetic algorithm to optimize the initial weight method: Chromosome encoding. Encode the initial weights and biases of the BP neural network into chromosomes. The encoding uses real values.

[0095] Population initialization. Randomly generate chromosomes of the initial population, each chromosome represents a possible combination of weights and biases.

[0096] Fitness function. Calculate their effectiveness in minimizing the circuit fault prediction error and define it as .

[0097] Perform genetic operations in the genetic algorithm to build the GA-BP neural network model. Selection: Select the best individual as the parent according to the fitness function.

[0098] Crossover: Combining genes from two parent chromosomes to produce new offspring.

[0099] Mutation: The introduction of random variation into some offspring to maintain genetic diversity.

[0100] Iteration: Through multiple generations of genetic operations, until the population converges to the optimal or near-optimal combination of weights and biases.

[0101] The step 4 is based on the construction of the GA-BP neural network model genetic algorithm parameter setting and training according to the following settings: Activation function setting: The hidden layer uses the nonlinear activation function ReLU.

[0102] Loss function setting: Use mean square error MSE as the loss function.

[0103] Learning rate setting: The initial learning rate is set to 0.001.

[0104] Training process settings: Use the initial weights and biases optimized by GA as the initial state of the BP neural network; implement the backpropagation algorithm to further minimize the error by adjusting the weights and biases; use the advanced variant of gradient descent Adam to update the weights.

[0105] The evaluation of the extreme weather and fault correlation analysis GA-BP neural network genetic algorithm model based on back propagation neural network analysis is set as follows: Evaluate the model performance on the test set, and use indicators such as R², MAE, and MSE to evaluate the effect of the model.

[0106] In the formula is the predicted value, is the true value is the mean, and m is the number of samples.

[0107] Step 5: Calculate the input feature contribution weight.

[0108] After training is completed, the connection weights from the input layer to the hidden layer and then to the output layer nodes are calculated to evaluate the importance of each weather indicator. . Accumulate the contribution of each layer and quantify the relative importance of each input feature. The specific steps include: Calculate the contribution of each hidden layer node to the output.

[0109] Calculate the contribution of each input feature to the hidden layer nodes.

[0110] The total contribution weight of each input feature to the output is calculated and normalized to compare the relative importance of each feature.

[0111] Assume that the 9 nodes in the input layer are A=[a1,a2,a3…,a9], and the 7 nodes in the hidden layer are H=[h1,h2,h3…,h7]. The weight matrix from the input layer to the hidden layer is ,in Indicates that from the input node To the hidden layer nodes The weight of .

[0112] in, is the activation function, is a hidden layer node The bias.

[0113] Assume that the output layer nodes are , the weight matrix from the hidden layer to the output layer is ,in Represents the hidden layer nodes The weight to the output layer node z.

[0114] in, Represents the bias of the output layer node.

[0115] Specifically, the contribution of each input feature to the output can be calculated by the following steps.

[0116] Calculate the contribution of each hidden layer node to the output: Among them, C j is a hidden node Contribution to output.

[0117] Calculate the contribution of each input feature to the hidden layer nodes: Among them, C ij is the input feature For hidden nodes contribution.

[0118] Compute the total contribution of each input feature to the output: Among them, C i is the input feature The total contribution to the output z.

[0119] Normalization is performed. In order to compare the contribution of each input feature, each C i Normalize so that all C i The sum is 1: Contribution Step 6: Classify the model into security levels.

[0120] According to the total contribution weight of each input feature, it is divided into three security levels: high, medium and low: it provides the model security level classification based on the correlation analysis results of the entire domain, and divides the security level according to the influence of each feature on the model output, as well as thematic map visualization.

[0121] According to the total contribution weight C of each input feature i, statistically analyze the weights of all features, divide the weights into several levels, and use the standard deviation method to divide them. The following is the specific division method: High Impact: Features with weights greater than the mean plus one standard deviation are considered high impact.

[0122] Medium Impact: Features with weights between plus or minus one standard deviation from the mean are considered to have medium impact.

[0123] Low Impact: Features with weights less than the mean minus one standard deviation are considered low impact.

[0124] in, is the standard deviation, is the weighted average.

[0125] Step 7: Based on the global weight impact analysis, ArcMap geographic information system software is used to perform spatial analysis on the correlation analysis results within the study area, generate thematic maps and heat maps, and visually display the impact areas of extreme weather on power failures. Generate thematic maps. Use the spatial analysis tool of the software and adopt the inverse distance weighted method to analyze the number of faults and their weight impact values. Interpolation analysis is performed to obtain global numerical results, and heat maps and spatial distribution maps of the impact of different weather indicators on power failures are generated, showing the correlation strength between fault density in different areas and specific weather indicators.

[0126] Where Z(x,y) represents the interpolation result of the number of power failures at the point in the area with the longitude and latitude (x,y). is the distance between the known sample point and the point to be interpolated, is the number of known sample power failures, k is 2, Indicates contribution.

[0127] The specific steps for correlation analysis and mapping in ArcGIS are as follows: Data preparation: First prepare the data, including vector data and raster data. Make sure the coordinate system and projection of the data are correct to perform accurate spatial analysis.

[0128] Loading data: Use ArcMap's Catalog function or the "Add Data" tool to load data into the map. This can be done by dragging the data onto the layer or using the "Add Data" button.

[0129] Data Clipping: Use the Bulk Clipping Tool to clip data to focus analysis on specific areas.

[0130] Symbolization: Symbolize different layers to clearly distinguish them on the map. Set the symbolization through the layer's properties panel.

[0131] Association analysis: For vector data, use the Spatial Join tool to join related data tables to the map layer. Use the Hot Spot Analysis tool to identify spatial clustering of high or low values ​​in the dataset.

[0132] Mapping: Set the map layout, including adding legends, compasses, scale bars, and grids; use "Layout View" to adjust the layout and design of the map; export the map to image format through the "Export Map" function.

[0133] Interpretation of analysis results: After the analysis is completed, interpret the results to determine whether the association between the data is statistically significant.

[0134] Documentation and Reporting: Finally, integrate the analysis results and maps into a report or documentation.

[0135] Interpretation of results. Combined with the generated thematic maps, analyze which areas of power failure are most vulnerable to certain extreme weather events. In-depth interpretation of the results in key areas, and provide targeted disaster prevention measures.

[0136] Embodiment 2 This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps in the method for analyzing the correlation between extreme weather and faults based on back propagation neural network analysis as described in the first embodiment above are implemented.

[0137] Embodiment 3 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for analyzing the correlation between extreme weather and faults based on back-propagation neural network analysis as described in the first embodiment above are implemented.

[0138] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for analyzing the correlation between extreme weather indicators and faults based on genetic algorithm back propagation, characterized by the following steps: Step 1: Collect statistical data of meteorological observation stations in the study area, the scope of the study area and fault data in the study area; Step 2: Perform numerical statistics and conversion on the statistical data of the meteorological observation station according to the specified extreme weather definition index format, and spatially associate and match the fault data with the study area; Step 3: Divide and mark the extreme weather index data and the corresponding fault quantity data to construct a sample set; Step 4: Build a GA-BP neural network association analysis model, then use the sample set to train the association analysis model, predict the number of faults, obtain the importance evaluation of the global weight, perform model verification and evaluation, and finally obtain a trained model; Step 5: Use the trained model to conduct global training and construction in the research area to obtain the correlation analysis results between global extreme weather and faults; Step 6: Analyze and visualize the results, analyze the weight impact analysis, build models and safety level classification; use ArcMap geographic mapping software to perform spatial analysis and generate thematic maps for the global correlation results; Based on this correlation analysis model method, a rapid correlation analysis of extreme weather and faults in the entire target area can be achieved.

2. The extreme weather index correlation fault analysis method based on genetic algorithm back propagation as claimed in claim 1 is characterized in that The step 1 mentioned includes: Based on the results of the mechanism research of extreme weather events, obtain conventional meteorological data of extreme weather and methods for calculating extreme meteorological data; obtain historical meteorological data observed and counted by the national meteorological observation station and meteorological bureau in the target study area, including hourly rainfall, hourly average temperature, hourly maximum temperature, hourly minimum temperature, and hourly maximum wind speed; obtain shp data of the geographical boundary range of the study area; obtain the location information of each fault area in the study area, the number of faults occurring, and the corresponding fault voltage size.

3. The extreme weather index correlation fault analysis method based on genetic algorithm back propagation as claimed in claim 1, characterized in that The step 2 mentioned includes: At present, the main indicators for defining extreme rainfall and temperature weather in research include: the maximum number of consecutive dry days CDD, the maximum number of consecutive wet days CWD, the total annual precipitation, the average annual precipitation intensity SDⅡ, that is, the ratio of the total annual precipitation to the number of wet days (i.e., the number of days with daily precipitation greater than 1 mm), the number of heavy precipitation days R25mm and R50mm (i.e., daily precipitation greater than or equal to 25mm and 50mm), the heavy precipitation rate R95p (%), that is, the ratio of the annual cumulative precipitation with daily precipitation greater than the 95% percentile to the total annual precipitation, the number of high temperature days HTD (i.e., the number of days with the highest daily temperature greater than or equal to 35 degrees Celsius in the year), and the number of frost days FD. That is, the number of days with a daily minimum temperature of less than or equal to 0 degrees Celsius in a year, the extreme high value of the daily maximum temperature TXx, that is, the maximum value of the daily maximum temperature in each month, and the extreme low value of the daily minimum temperature TNn, that is, the minimum value of the daily minimum temperature in each month; the above research indicators are calculated annually; the data collected by the National Meteorological Observation Station and the Meteorological Bureau are calculated according to the above extreme weather definition indicators; the geographical location information of the obtained research area is spatially matched with the fault data, generally at the county level, and the matching method can be artificial geographic statistics or ArcMap geographic information processing software for spatial association processing; ArcMap geographic information processing software was used to integrate and match data in the same area. The specific process was to import the geographic boundary range shp data of the study area into the ArcMap software, import the extreme weather definition mainly based on indicator data processed by the National Meteorological Observation Station and the Meteorological Bureau observation station in the form of point data, and import the vector surface data containing the number of faults in each fault area according to the county-level surface range. Then, using the spatial association tool, the extreme weather definition mainly based on indicator data and the vector surface data containing the number of faults in the fault area were spatially associated to obtain a data format containing both attributes.

4. The extreme weather index correlation fault analysis method based on genetic algorithm back propagation as claimed in claim 1, characterized in that The step 3 mentioned includes: Based on the definition of extreme weather, the index data and the number of faults in the fault area are pre-classified. All data sets are U. The classification is based on whether there is a spatial intersection between the national meteorological observation station and the meteorological bureau and the fault area, that is, whether there is a fault at the location of the extreme weather numerical results processed. Based on this, it is divided into two categories: U T Indicates that there is a spatial intersection, one type is U F Indicates that there is no spatial intersection; select U T Serves as a sample range set for model testing.

5. The extreme weather index correlation fault analysis method based on genetic algorithm back propagation as claimed in claim 1, characterized in that The step 4 mentioned includes: (1) Define the network structure In the range U T Select the model test sample range set, construct input-output pairs according to weather indicators and fault data, take weather indicators A=[a1,a2,a3…,a9] as input, and the number of power faults under five different voltages B=[b1,b2,b3,b4,b5] as output to construct a sample set; T Divide into training set, validation set and test set; (2) Use genetic algorithm to optimize initial weights Chromosome encoding: Encode the weights and thresholds of the BP neural network into chromosomes of the genetic algorithm; Population initialization: randomly generate several chromosomes as the initial population; Fitness function: Define the fitness function to evaluate the fitness of each chromosome by training the BP neural network and calculating its error; Genetic operations: including selection, crossover, mutation, iterative optimization of weights and thresholds until the global optimal solution is found; (3) Perform BP neural network training Use the initial weights and thresholds optimized by genetic algorithm to train BP neural network; The back propagation algorithm is used to further adjust the network weights and thresholds until the network error converges; (4) Evaluate the importance of global weights Weight extraction: Extract the global weights after training, including all weights from the input layer to the hidden layer and from the hidden layer to the output layer; Global weight analysis: Calculate the total weight contribution of each input variable in the entire network; specifically, the impact of each input variable on the output can be accumulated layer by layer, and the comprehensive weight of each input variable on the final output can be calculated by analyzing the weight from the input layer to the hidden layer and combining the weight from the hidden layer to the output layer; These combined weights are normalized so that their sum is 1, which quantifies the relative importance of each input variable; The GA-BP neural network is used to extract effective eigenvalues ​​from the extreme weather definition based on indicator data, and the extracted extreme weather indicator numerical eigenvalues ​​are used to construct the corresponding input vector according to the structural characteristics of the BP neural network input layer; the input vector is sent to the GA-BP neural network model, and the BP neural network obtains the regional fault data according to the output value of the neuron.

6. The extreme weather index correlation fault analysis method based on genetic algorithm back propagation as claimed in claim 1, characterized in that The step 5 mentioned comprises: The steps of constructing the global training using the above-trained GA-BP neural network correlation analysis model include batch processing of the data of all observation points in the study area to generate global fault prediction results; then model validation and evaluation of the validation set are performed; Validation set evaluation: Use validation set data to evaluate the performance of the model and observe its performance on unseen data to ensure the generalization ability of the model; Test set verification: Use the test set to perform final verification of the model, evaluate the accuracy and stability of the model, and calculate the mean square error (MSE) and R-square value of various performance indicators.

7. The extreme weather index correlation fault analysis method based on genetic algorithm back propagation as claimed in claim 1, characterized in that The step 6 mentioned comprises: Weight impact analysis: Based on the importance assessment results of the global weight, analyze the impact of various weather indicators on power failures and determine which indicators have the greatest impact on the occurrence of failures; After calculating the total contribution weight of each input feature to the output, the model security level is divided according to the influence of each feature on the model output; Visual analysis: ArcMap is used to visualize the fault data and weight analysis results of the entire region. The inverse distance weighted method is used to interpolate the number of faults and weight influence values. The global numerical results are interpolated to generate a heat map or spatial distribution map of the influence weights, which intuitively displays the spatial distribution of faults in the entire region and the influence weights of various indicators.

8. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the extreme weather index and fault correlation analysis method based on genetic algorithm back propagation as described in any one of claims 1 to 7 are implemented.

Images