Power distribution network fault prediction method and system

Abstract

The invention discloses a power distribution network fault prediction method and system, and the method comprises the steps: obtaining multi-dimensional feature data, carrying out the first preprocessing of the multi-dimensional feature data, and obtaining first multi-dimensional feature data; utilizing a first neural network and a second neural network to respectively extract spatial features and time features in the multi-dimensional feature data, and constructing a prediction model based on the spatial features and the time features; and defining a loss function of the prediction model, training the prediction model by using an optimizer to obtain a first prediction model, and evaluating the first prediction model through the evaluation index to realize dynamic evaluation and prediction of the power distribution network fault.

Description

Technical Field

[0001] The present invention relates to the field of power system disaster prevention and mitigation and artificial intelligence technology, and in particular to a distribution network fault prediction method and system. Background Art

[0002] The frequent occurrence of external natural disasters such as ice cover, wildfires, cold waves, rainstorms, floods, extreme temperatures and lightning has brought unprecedented challenges to the safe and stable operation of the power system. At the same time, due to the differences in the temporal and spatial distribution of the power system and the particularity of the life cycle of power equipment, the equipment failures in different geographical regions are currently diversified and specific. Extreme external disasters and potential defects of the equipment themselves can easily cause mass equipment failures, and may even further trigger various chain reactions, endangering the power supply and energy supply, thus affecting social stability. At present, Guizhou Power Grid Company has generated a large amount of data in the operation and maintenance of distribution network lines, but it has not been effectively utilized; at the same time, there is a lack of effective intelligent means to control distribution network failures, and it is difficult to improve the level of distribution network failure response strategies. In the daily operation and maintenance of the distribution network, predicting distribution network failures is the key path to achieve the goal of efficient disaster prevention and disaster resistance. The research on accurate distribution network fault prediction model can not only effectively guide the design and operation and maintenance of distribution network lines, but also significantly improve the ability of the power grid to resist natural disasters, which is of great significance to ensure the safe and stable operation of the power grid.

[0003] However, the current distribution network fault model faces many challenges. Traditional distribution network fault prediction methods based on experience and limited observation data often fail to fully and accurately reflect the fault characteristics of different regions under complex meteorological conditions. Existing methods are difficult to effectively integrate multiple influencing factors such as meteorological elements and environmental conditions, resulting in a high reliance on human experience in the distribution network fault prediction process and a lack of automated and scientific prediction methods. These problems directly affect the accuracy and applicability of distribution network auxiliary decision-making, making it difficult to adapt to complex terrain and changeable climate conditions, thereby limiting its ability to provide accurate decision support for power grid planning and operation and maintenance. In order to overcome these challenges, combining multi-source data, applying advanced numerical simulation and data assimilation techniques, and introducing machine learning methods have become the main trends in current research. In particular, digital distribution network fault prediction methods based on big data and machine learning have shown great potential. However, the current research on intelligent systems for automatic and accurate distribution network fault prediction under complex modal data in the industry is still a technical gap, which seriously restricts the improvement of the power grid's disaster prevention and mitigation capabilities. Existing methods cannot fully utilize the increasingly abundant multi-source data resources, nor can they adapt to rapidly changing climate conditions, resulting in the difficulty of distribution network fault prediction results in providing timely and accurate auxiliary decision support for distribution network planning and operation and maintenance. At the same time, although machine learning methods have made significant progress in the field of data analysis in recent years, there is still a lack of effective solutions for complex modal data. In particular, the potential of the 2DCNN-BLSTM algorithm in processing time series, high-dimensional, and nonlinear relationship data has not been fully explored and utilized. At present, there is an urgent need for a distribution network fault prediction method that can effectively integrate multi-source data and automatically process complex modal data. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a distribution network fault prediction method and system to solve the problems of low ultra-short-term prediction accuracy of distribution network faults under complex modes, inability to effectively integrate multi-source data, and lack of automation and intelligent processing capabilities of traditional methods.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a distribution network fault prediction method, comprising:

[0008] Acquire multidimensional feature data, and perform first preprocessing on the multidimensional feature data to obtain first multidimensional feature data;

[0009] Using a first neural network and a second neural network to respectively extract spatial features and temporal features from the multidimensional feature data, and constructing a prediction model based on the spatial features and temporal features;

[0010] A loss function of a prediction model is defined, and the prediction model is trained using an optimizer to obtain a first prediction model, and the first prediction model is evaluated using an evaluation indicator.

[0011] As a preferred solution of the distribution network fault prediction method of the present invention, the first preprocessing of the multi-dimensional feature data includes:

[0012] Calculate the local anomaly factor value of each multidimensional feature data, and calculate the mean and standard deviation of the local anomaly factor values ​​of all multidimensional feature data;

[0013] If the local anomaly factor value of the multidimensional feature data is greater than the sum of the mean value of the local anomaly factor value of the multidimensional feature data and three times the standard deviation of the local anomaly factor value of the multidimensional feature data, then the multidimensional feature data is an outlier.

[0014] As a preferred solution of the distribution network fault prediction method described in the present invention, it also includes:

[0015] Calculate the mean and standard deviation of each multidimensional feature data, and calculate the standardized value of the mean of each multidimensional feature data;

[0016] Multiply the normalized value by the negative exponent of the nonlinear factor, add 1 to the exponential term, and get the denominator;

[0017] Divide the denominator by 2 and subtract 1 to get the normalized value.

[0018] As a preferred solution of the distribution network fault prediction method described in the present invention, it also includes:

[0019] The missing values ​​are filled by spatiotemporal data interpolation based on tensor decomposition. The optimal tensor rank is obtained by dividing the multidimensional feature data into K parts, taking one part as the test set and the remaining K-1 parts as the training set. This is repeated K times, and a different part is selected as the test set each time. The average performance index of the K evaluations is calculated to obtain the optimal tensor rank.

[0020] As a preferred solution of the distribution network fault prediction method of the present invention, wherein: constructing a prediction model based on the spatial characteristics and time characteristics includes:

[0021] For the input layer, the shape of the input data is expanded, and the spatial features of the expanded input data are extracted using the first neural network. The number of input channels is increased, the dimension is kept unchanged, and the weights are learned at the same time.

[0022] The pooling operation is used to reduce the spatial dimension and the computational complexity of subsequent layers. During the training process, a portion of the input units are randomly set to zero to reshape the data and obtain sequence data.

[0023] The second neural network is used to process the sequence data and output the final prediction result.

[0024] As a preferred solution of the distribution network fault prediction method described in the present invention, it also includes:

[0025] For the lth layer in the distribution network, the convolution kernel matrix is ​​used to perform convolution operation on the selected input map, and nonlinearity is introduced using a nonlinear activation function, and the feature map is operated through a subsampling function to output the map;

[0026] The temporal correlation in time series data is captured by introducing a gating mechanism.

[0027] As a preferred solution of the distribution network fault prediction method described in the present invention, the training prediction model includes:

[0028] The learning rate for the current time step is half the difference between the maximum and minimum learning rates. The adjustment factor is obtained by adding 1 to the result of the cosine function multiplied by the adjustment factor and adding the result to the minimum learning rate.

[0029] In a second aspect, the present invention provides a distribution network fault prediction system, comprising:

[0030] A preprocessing module, used for acquiring multidimensional feature data, performing first preprocessing on the multidimensional feature data, and obtaining first multidimensional feature data;

[0031] A model building module is used to use the first neural network and the second neural network to respectively extract the spatial features and the temporal features in the multidimensional feature data, and build a prediction model based on the spatial features and the temporal features;

[0032] The evaluation module is used to define the loss function of the prediction model, and use the optimizer to train the prediction model to obtain a first prediction model, and evaluate the first prediction model through the evaluation index.

[0033] In a third aspect, the present invention provides a computing device, comprising:

[0034] Memory and processor;

[0035] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the distribution network fault prediction method are implemented.

[0036] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the distribution network fault prediction method.

[0037] Compared with the prior art, the present invention has the following beneficial effects: the present invention significantly improves the accuracy of distribution network fault prediction under complex modes by integrating two-dimensional convolutional neural network (2DCNN) and bidirectional long short-term memory (BLSTM) method; adopts multi-source distribution network data, meteorological elements and holiday data, combined with the feature extraction capability of deep learning model, so that the time resolution of distribution network fault prediction is improved from the original ultra-short term to the short term level, or even the medium term; introduces adaptive feature weight adjustment mechanism, so that the model can automatically adjust the weights of various environmental factors according to the data characteristics of different regions. This adaptability enables the model to maintain high accuracy in different meteorological environments, greatly expanding the scope of application of the method; by integrating time-series distribution network data and meteorological forecast data, dynamic evaluation and prediction of distribution network faults can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0039] Figure 1 A schematic diagram of the overall process logic of a distribution network fault prediction method according to an embodiment of the present invention;

[0040] Figure 2 A CNN-BLSTM structure diagram of a distribution network fault prediction method according to an embodiment of the present invention;

[0041] Figure 3 This is the LSTM layer of the distribution network fault prediction method described in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0043] Example 1

[0044] Reference Figure 1-Figure 3 , as an embodiment of the present invention, provides a distribution network fault prediction method, comprising:

[0045] S100: Acquire multidimensional feature data, and perform first preprocessing on the multidimensional feature data to obtain first multidimensional feature data;

[0046] S200: using the first neural network and the second neural network to respectively extract spatial features and temporal features from the multidimensional feature data, and constructing a prediction model based on the spatial features and the temporal features;

[0047] S300: define a loss function of a prediction model, and use an optimizer to train the prediction model to obtain a first prediction model, and evaluate the first prediction model using an evaluation indicator.

[0048] It should be noted that the fusion of two-dimensional convolutional neural network (2D CNN) and bidirectional long short-term memory (BLSTM) method significantly improves the accuracy of distribution network fault prediction under complex modes. Compared with the traditional single algorithm of a single factor, the error rate is significantly reduced, and the results are closer to the actual situation; the 2DCNN-BLSTM integrated learning algorithm architecture significantly improves the computational efficiency. Compared with the traditional manual experience division method, this method significantly reduces the calculation time and investigation time, while ensuring the accuracy of the results. This effect makes large-scale and high-frequency distribution network fault monitoring possible.

[0049] In the embodiment of the present application, the above step S100 includes the following sub-steps A1-A5;

[0050] In A1: calculate the local anomaly factor value of each multidimensional feature data, and calculate the mean and standard deviation of the local anomaly factor values ​​of all multidimensional feature data;

[0051] In A2: if the local outlier factor value of the multidimensional feature data is greater than the sum of the mean value of the local outlier factor value of the multidimensional feature data and three times the standard deviation of the local outlier factor value of the multidimensional feature data, then the multidimensional feature data is an outlier.

[0052] In A3: calculating the mean and standard deviation of each multidimensional feature data, and calculating the standardized value of the mean of each multidimensional feature data;

[0053] In A4: multiply the normalized value by the negative exponent of the nonlinear factor, add 1 to the exponential term, and obtain the denominator;

[0054] In A5: Divide the denominator by 2 and subtract 1 to obtain the normalized value.

[0055] In the embodiment of the present application, an advanced multi-source data acquisition system is used to collect multi-dimensional feature data including historical distribution network faults, meteorological elements, and holidays;

[0056] Specifically, historical distribution network fault information includes the time, location, impact range and fault handling information of the fault. The data types include line information, time information and string information. Meteorological elements include the consideration that temperature rise will lead to a decrease in the voltage withstand level and insulation characteristics of the equipment. High humidity can easily reduce the insulation performance of the equipment, leading to current leakage, flashover and other situations. Lightning may directly cause the equipment to withstand lightning overvoltage, etc. It includes time classification, regional classification, average temperature, average high temperature, extremely high temperature, average low temperature, extremely low temperature, rainy days, thunderstorm days, snowy days, haze days, foggy days, 3-5 wind days, 5 wind days or above, relative humidity, average wind speed, monthly average maximum wind speed, and maximum continuous wind speed. Holidays include holidays that indirectly affect the operation and faults of the distribution network.

[0057] In the embodiment of the present application, the first preprocessing includes outlier detection and removal, data normalization and missing value processing;

[0058] Outlier detection and removal includes using an improved local outlier factor (LOF) algorithm to identify and remove outlier data points;

[0059] Specifically, for feature j, the criterion for determining outliers is expressed as:

[0060]

[0061] in, is the mean of the LOF values ​​of feature j, is the standard deviation of the LOF value of feature j;

[0062] Use the kd-tree algorithm to speed up the LOF calculation process to handle large-scale data sets;

[0063] It should be noted that the improved local outlier factor (LOF) algorithm is suitable for anomaly detection in high-dimensional data, which can effectively identify local outliers and improve the reliability of data.

[0064] In an optional embodiment, data normalization may also include a linear transformation of the original data using a traditional Min-Max normalization method, so that the result value is mapped to between [0,–1]. The conversion function is expressed as:

[0065]

[0066] Among them, x max is the maximum value of the sample data, x min is the minimum value of the sample data;

[0067] In an optional embodiment, data normalization may also include Z-score normalization, which is to normalize the data by giving the mean and standard deviation of the data. The processed data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1. The conversion function is expressed as:

[0068]

[0069] Among them, x mean is the mean of all sample data, σ is the standard deviation of all sample data;

[0070] In the embodiment of the present application, data normalization includes:

[0071] Using the improved Min-Max normalization method, nonlinear transformation is introduced to enhance the distinguishability of features as follows:

[0072]

[0073] Among them, α is an adjustable nonlinear factor, usually set between 0.1 and 10, μ j and σ j are the mean and standard deviation of feature j respectively;

[0074] It should be noted that the present method preferably uses a nonlinear normalization method to better handle features of different scales and distributions, and is particularly effective for features with long-tail distributions.

[0075] In the embodiment of the present application, after completing steps A1-A5 in the above step S100, the following step A6 is also included;

[0076] In A6: Using tensor decomposition-based spatiotemporal data interpolation to fill missing values, the optimal tensor rank is obtained by dividing the multidimensional feature data into K parts, using one part as the test set and the remaining K-1 parts as the training set, repeating K times, selecting a different part as the test set each time, and calculating the average performance index of K evaluations to obtain the optimal tensor rank.

[0077] In an optional embodiment, the missing value processing may include a sliding average window method. If the i-th position data in a list a is missing data, the average value of the window data before and after is taken as the interpolation data.

[0078] For example, the temperature data is: a=[13,10,11,5,None,4,9,7,8], window=3, then the data at the None position is: (10+11+5+4+9+7) / 6=7.67;

[0079] In an optional embodiment, missing value processing may also include regression interpolation, which is based on a complete data set, and a regression equation is established. For objects containing null values, known attribute values ​​are substituted into the equation to estimate unknown attribute values, and the estimated values ​​are used to fill in the missing values. When the variables are not linearly related, biased estimates will result.

[0080] For example, the temperature data to be interpolated is: a=[13,10,11,5,None,4,9,7,8], construct the time x-temperature y regression model y=-0.774x+11.856, substitute x=5, and the missing value of temperature is 7.99;

[0081] In the embodiment of the present application, a spatiotemporal data interpolation method based on tensor decomposition is used to fill in missing values. ij , the fill value is expressed as:

[0082]

[0083] in, is the core tensor, a i 、b j , c are space, time and characteristic pattern factors respectively;

[0084] Use alternating least squares (ALS) to solve the tensor decomposition problem and select the optimal tensor rank through cross-validation;

[0085] It should be noted that the spatiotemporal data interpolation method based on tensor decomposition is suitable for data with spatiotemporal correlation, can simultaneously consider the spatial and temporal correlation of data, and provide more accurate missing value estimation. Compared with other interpolation methods, the tensor decomposition method can better handle nonlinear relationships, provide more accurate missing value filling, and can adapt to various complex data patterns, including nonlinear and high-dimensional data, which increases the flexibility and robustness of the method.

[0086] In the embodiment of the present application, the above step S200 includes the following sub-steps B1-B3;

[0087] In B1: For the input layer, the shape of the input data is expanded, the spatial features of the expanded input data are extracted using the first neural network, and the number of input channels is increased, the dimension is kept unchanged, and the weights are learned at the same time;

[0088] In B2: Pooling is used to reduce spatial dimensions and computational complexity of subsequent layers. During training, a portion of the input units are randomly set to zero to reshape the data and obtain sequence data.

[0089] In B3: the second neural network is used to process the sequence data and output the final prediction result.

[0090] Specifically, the first neural network is a two-dimensional convolutional neural network, and the second neural network is a bidirectional long short-term memory. The two-dimensional convolutional neural network is used to learn and extract the spatial features of the data set, and the bidirectional long short-term memory is used to learn and extract the temporal features of the data set;

[0091] like Figure 2 As shown, the repeated vector layer expands the input shape from (16,6) to (16,6,1) to accommodate the two-dimensional convolution layer. The two-dimensional convolution layer increases the number of channels from 1 to 64 while maintaining the dimension of (16,6) and learns 640 weights. The two-dimensional maximum pooling layer reduces the spatial dimension, retains important features and reduces computational complexity. The regularization layer alleviates overfitting by randomly setting a portion of the input units to zero during training. The repeated vector layer reshapes the data to (16,384) to be compatible with the subsequent bidirectional long short-term memory layer. The bidirectional long short-term memory layer processes sequence data in both forward and reverse directions, capturing time dependencies with 31,600 parameters. A regularization layer is applied to the output of the bidirectional long short-term memory layer to further prevent overfitting. The fully connected layer converts the output of the bidirectional long short-term memory layer into the final prediction result, reducing the output shape from 20 to 1 and outputting a predicted distribution network fault with 21 parameters.

[0092] It should be noted that the two-dimensional convolutional neural network effectively extracts spatial features from the input data, which is crucial for understanding the geometry and structural information of the data. Reducing the spatial dimension through pooling operations helps to reduce the computational complexity of the model while retaining important features, making the model more efficient. The use of regularization layers helps to reduce overfitting and enhance the generalization ability of the model, ensuring that the model performs well on unseen data. The bidirectional long short-term memory network processes sequence data and captures time dependencies. By combining the two-dimensional convolutional neural network and the bidirectional long short-term memory network, high-precision prediction of distribution network faults is achieved while ensuring computational efficiency and the generalization ability of the model.

[0093] In the embodiment of the present application, after completing steps B1-B3 in the above step S200, the following steps B4-B5 are also included;

[0094] In B4: for the lth layer in the distribution network, the convolution kernel matrix is ​​used to perform a convolution operation on the selected input map, and a nonlinear activation function is used to introduce nonlinearity, and the feature map is operated through a subsampling function to output the map;

[0095] In B5: Temporal correlation in time series data is captured by introducing a gate control mechanism.

[0096] Specifically, the two-dimensional CNN layer has excellent spatial feature extraction capabilities, which greatly simplifies the prediction process and can use convolution kernels to extract spatial features from data; a typical convolutional neural network (CNN) consists of a convolutional layer (CL), a pooling layer (PL), and a fully connected layer (FL), which extracts the features of input samples through the convolutional layer and the pooling layer. The equations of the convolutional layer and the pooling layer are expressed as:

[0097]

[0098] in, is the convolution layer, l represents the lth layer in the network, k represents the convolution kernel matrix of size S×S, M j is the selection of input mapping, f is the nonlinear activation function, and the rectified linear unit (ReLU) is used as the nonlinear activation function; represents a pooling layer, down() represents a subsampling function that sums all different n×n blocks in the feature map, so that the output feature map is reduced by a factor of n in both spatial dimensions, and each output map has a corresponding additive bias b and multiplicative bias β;

[0099] Specifically, Figure 3 As shown, the LSTM layer is used to capture time correlation and is expressed as:

[0100] f t =σ(W f [h t-1 ,x t ]+b f )

[0101] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0102]

[0103] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0104] h t =o t *tanh(C t )

[0105] Among them, f t 、i t , o t They are forget gate, input gate and output gate, C tis the cell state, h t is the hidden state, W f , W i , W o , W C is the weight matrix of the forget gate, input gate, output gate, and unit state candidate, b f 、b i 、b o 、b C is the bias term, is the unit state candidate, σ is the activation function, h t-1 is the hidden state of the previous time step, x t is the input of the current time step, C t-1 is the cell state at the previous time step, f t and i t They respectively control the update of the previous unit state and the unit state candidate.

[0106] It should be noted that through the two-dimensional CNN layer, rich features can be extracted from the spatial data, and the introduction of nonlinear activation functions enhances the model's ability to handle nonlinear problems. The pooling layer reduces the spatial dimension of the feature map through the subsampling function, effectively reducing the computational complexity of the model while retaining important features. The introduction of the LSTM layer enables the model to capture the time correlation in time series data, and the comprehensive analysis of spatial and temporal features makes the final prediction results more accurate, thereby improving the reliability of distribution network fault prediction.

[0107] In the embodiment of the present application, the above step S300 includes the following sub-step C1;

[0108] In C1: The learning rate for the current time step is half the difference between the maximum and minimum learning rates, get the adjustment factor, add 1 to the result of the cosine function multiplied by the adjustment factor, and add the result to the minimum learning rate.

[0109] Specifically, the mean square error is used as the loss function and is expressed as:

[0110]

[0111] Among them, y i is the true value of the i-th sample, is the predicted value of the model for the i-th sample, and n is the total number of samples;

[0112] The Adam optimizer is used for model training, and the learning rate adopts the cosine annealing strategy as follows:

[0113]

[0114] Among them, η tis the learning rate at time t, η max , η min are the maximum and minimum learning rates respectively, and T is the total training step length;

[0115] The model effect is verified using a pre-divided prediction test set, and the ROC curve is used to evaluate the effect. The area under the ROC curve (AUC) measures the quality of the algorithm. The closer the curve is to the upper left corner, the larger the AUC and the better the classification effect.

[0116] It should be noted that the cosine annealing strategy makes the learning rate change dynamically during the training process, which helps the model converge quickly in the early stage and stabilize and refine in the later stage to avoid falling into the local minimum. By adjusting the combination of factors and cosine functions, it can gradually approach the optimal solution while maintaining a certain degree of randomness, balancing exploration and utilization. The cosine annealing strategy helps avoid training instability caused by too high a learning rate or training stagnation caused by too low a learning rate.

[0117] The above is a schematic scheme of a distribution network fault prediction method of this embodiment. It should be noted that the technical scheme of the distribution network fault prediction system and the technical scheme of the above distribution network fault prediction method belong to the same concept, and the details not described in detail in the technical scheme of the distribution network fault prediction system in this embodiment can be referred to the description of the technical scheme of the above distribution network fault prediction method.

[0118] The distribution network fault prediction system in this embodiment includes:

[0119] A preprocessing module, used for acquiring multidimensional feature data, performing first preprocessing on the multidimensional feature data, and obtaining first multidimensional feature data;

[0120] A model building module is used to use the first neural network and the second neural network to respectively extract the spatial features and the temporal features in the multidimensional feature data, and build a prediction model based on the spatial features and the temporal features;

[0121] The evaluation module is used to define the loss function of the prediction model, and use the optimizer to train the prediction model to obtain a first prediction model, and evaluate the first prediction model through the evaluation index.

[0122] This embodiment also provides a computing device, which is applicable to distribution network fault prediction, including:

[0123] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the distribution network fault prediction method proposed in the above embodiment.

[0124] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for predicting distribution network faults as proposed in the above embodiment is implemented.

[0125] The storage medium proposed in this embodiment and the method for implementing distribution network fault prediction proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0126] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (Random Access Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0127] Example 2

[0128] Different from the first embodiment, a verification test of a distribution network fault prediction method is provided to verify and illustrate the technical effects adopted in the method.

[0129] If there are 7 types of independent variable data, including cableization rate, commissioning date, line type, line area characteristics, weather, temperature, and wind speed in the time period from 2021.01.01 to 2024.09.01, and 1 target variable data of whether there is a fault.

[0130] It was subjected to outlier detection, data normalization and missing value removal processing to obtain nine fault factors including cabling rate, commissioning date, line type, line area characteristics, weather, maximum temperature, minimum temperature, average temperature and wind speed.

[0131] The correlation analysis between the nine fault factors and whether the distribution network is faulty was performed. After PCA processing, the three fault factors with correlation less than 0.3 and the five fault factors with correlation greater than 0.3 formed the six final distribution network fault factors.

[0132] The six final distribution network failure factors and one target variable of whether there is a failure in the time period from 2021.01.01 to 2024.09.01 are used to form the distribution network failure model construction and prediction data set.

[0133] The dataset is divided into 16 time steps as samples, and the samples in the time period of 2021.01.01-2024.07.31 are used as samples for the construction and evaluation of the distribution network fault prediction model (the ratio of training set to validation set is 8:2), and the samples in the time period of 2024.08.01-2024.09.01 are used as the prediction sample set.

[0134] The distribution network fault prediction model construction and evaluation training set is used as input (input layer (16, 6)), and a RepeatVector layer is used to expand the input shape from (16, 6) to (16, 6, 1) to adapt to the Conv2D layer; a Conv2D layer is used to increase the number of channels from 1 to 64 while maintaining the dimension of (16, 6) and learn 640 weights; a MaxPooling2D layer is used to reduce the spatial dimension, retain important features and reduce computational complexity; a Dropout layer is used to randomly set part of the input units to zero during training. To mitigate overfitting, another RepeatVector layer is used to reshape the data to (16,384) to be compatible with the subsequent BLSTM layer. A Bidirectional Long Short-Term Memory (BLSTM) layer processes the sequence data in both the forward and reverse directions to capture the temporal dependencies with 31,600 parameters. A Dropout layer is applied to the BLSTM output to further prevent overfitting. Finally, a fully connected Dense layer is used to transform the BLSTM output into the final prediction result, reducing the output shape from 20 to 1 and outputting the predicted distribution network fault with 21 parameters. The loss function is set to mean square error (MSE).

[0135] Use the trained model to predict the validation set, and finally use AUC to evaluate the accuracy of the validation set to complete the model construction. Use the trained model to predict the prediction set, and use AUC to evaluate the accuracy of the validation set to complete the distribution network fault prediction method and system.

[0136] By integrating the two-dimensional convolutional neural network (2D CNN) and the bidirectional long short-term memory (BLSTM) method, the accuracy of distribution network fault prediction under complex modes is significantly improved. The 2DCNN-BLSTM integrated learning algorithm architecture significantly improves the computational efficiency and introduces an adaptive feature weight adjustment mechanism, which enables the model to automatically adjust the weights of various environmental factors according to the data characteristics of different regions. By integrating time-series distribution network data and meteorological forecast data, dynamic evaluation and prediction of distribution network faults can be achieved.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A distribution network fault prediction method, characterized in that: include: Acquire multidimensional feature data, and perform first preprocessing on the multidimensional feature data to obtain first multidimensional feature data; Using a first neural network and a second neural network to respectively extract spatial features and temporal features from the multidimensional feature data, and constructing a prediction model based on the spatial features and temporal features; A loss function of a prediction model is defined, and the prediction model is trained using an optimizer to obtain a first prediction model, and the first prediction model is evaluated using an evaluation indicator.

2. The distribution network fault prediction method according to claim 1, characterized in that: The first preprocessing of the multidimensional feature data includes: Calculate the local anomaly factor value of each multidimensional feature data, and calculate the mean and standard deviation of the local anomaly factor values ​​of all multidimensional feature data; If the local anomaly factor value of the multidimensional feature data is greater than the sum of the mean value of the local anomaly factor value of the multidimensional feature data and three times the standard deviation of the local anomaly factor value of the multidimensional feature data, then the multidimensional feature data is an outlier.

3. The distribution network fault prediction method according to claim 2, characterized in that: Also includes: Calculate the mean and standard deviation of each multidimensional feature data, and calculate the standardized value of the mean of each multidimensional feature data; Multiply the normalized value by the negative exponent of the nonlinear factor, add 1 to the exponential term, and get the denominator; Divide the denominator by 2 and subtract 1 to get the normalized value.

4. The distribution network fault prediction method according to claim 1 or 3, characterized in that: Also includes: The missing values ​​are filled by spatiotemporal data interpolation based on tensor decomposition. The optimal tensor rank is obtained by dividing the multidimensional feature data into K parts, taking one part as the test set and the remaining K-1 parts as the training set. This is repeated K times, and a different part is selected as the test set each time. The average performance index of the K evaluations is calculated to obtain the optimal tensor rank.

5. The distribution network fault prediction method according to claim 4, characterized in that: Constructing a prediction model based on the spatial features and temporal features includes: For the input layer, the shape of the input data is expanded, and the spatial features of the expanded input data are extracted using the first neural network. The number of input channels is increased, the dimension is kept unchanged, and the weights are learned at the same time. The pooling operation is used to reduce the spatial dimension and the computational complexity of subsequent layers. During the training process, a portion of the input units are randomly set to zero to reshape the data and obtain sequence data. The second neural network is used to process the sequence data and output the final prediction result.

6. The distribution network fault prediction method according to claim 5, characterized in that: Also includes: For the lth layer in the distribution network, the convolution kernel matrix is ​​used to perform convolution operation on the selected input map, and nonlinearity is introduced using a nonlinear activation function, and the feature map is operated through a subsampling function to output the map; The temporal correlation in time series data is captured by introducing a gating mechanism.

7. The method for predicting distribution network faults according to claim 6, characterized in that: Training a prediction model involves: The learning rate for the current time step is half the difference between the maximum and minimum learning rates. The adjustment factor is obtained by adding 1 to the result of the cosine function multiplied by the adjustment factor and adding the result to the minimum learning rate.

8. A system using the distribution network fault prediction method according to any one of claims 1 to 7, characterized in that: include: A preprocessing module, used for acquiring multidimensional feature data, performing first preprocessing on the multidimensional feature data, and obtaining first multidimensional feature data; A model building module is used to use the first neural network and the second neural network to respectively extract the spatial features and the temporal features in the multidimensional feature data, and build a prediction model based on the spatial features and the temporal features; The evaluation module is used to define the loss function of the prediction model, and use the optimizer to train the prediction model to obtain a first prediction model, and evaluate the first prediction model through the evaluation index.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the distribution network fault prediction method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the distribution network fault prediction method according to any one of claims 1 to 7.

Images