Power distribution network line icing fault prediction method, device and equipment and storage medium

Abstract

Embodiments of the present application provide a power distribution network line icing fault prediction method, device and equipment and a storage medium, which comprises: performing feature extraction on original meteorological monitoring data, constructing a target time series set, the target time series set containing target time series samples and target time series labels; inputting the target time series samples into a pre-trained fault prediction model to obtain a fault classification result of a power distribution network line; if the fault classification result is fault data, reconstructing the fault data to determine a fault prediction result of the power distribution network line. By using a convolutional neural network to extract features from high-dimensional redundant meteorological monitoring data, then inputting the extracted features into a fault prediction model for fault prediction, the convergence speed of the model and the prediction accuracy are improved; finally, the prediction result is reconstructed, so that the reconstructed fault classification is closer to the actual value, and the prediction result accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of power grid disaster prevention, mitigation and prediction technology, and in particular to a method, device, equipment and storage medium for predicting icing faults on distribution network lines. Background Technology

[0002] In recent years, large-scale and continuous power outages caused by natural disasters have brought unprecedented challenges to the power system. Extreme weather events such as icing have dramatically increased the probability of line damage and the resulting losses. As the distribution network is located at the end of the power grid, compared to the transmission network, the distribution network has a lower level of automation, poorer remote measurement capabilities for ice thickness, lower redundancy, and a lack of control and protection measures, resulting in a weaker ability to cope with icing disasters. Therefore, predicting the icing fault status of distribution network lines is crucial to avoiding these problems.

[0003] In monitoring icing conditions on power distribution lines, the direct method directly measures changes in characteristics caused by ice deposition on the line surface, such as mass and dielectric constant, to reflect the icing condition. This requires additional sensors, increasing costs and is prone to equipment aging and data drift. The indirect method, on the other hand, obtains environmental factors such as temperature, wind speed, and air pressure, and uses data-driven methods combined with heuristic algorithms and machine learning to construct a line icing model to reflect the icing condition. The indirect method does not require the construction of complex physical models and does not require additional sensors, thus avoiding additional costs.

[0004] Data-driven approaches include shallow machine learning and deep machine learning. Shallow machine learning methods, such as random forests, support vector machines, and backpropagation neural networks, have relatively simple models, but they have the drawback of being unable to learn the correlations of long-term sequence data. Deep machine learning methods, such as recurrent neural networks and long short-term memory networks, have poor parallel computing capabilities and lower accuracy in processing multi-dimensional sequence data. Summary of the Invention

[0005] This invention provides a method, apparatus, equipment, and storage medium for predicting icing faults in power distribution lines, thereby improving the accuracy of icing fault prediction and shortening the prediction time.

[0006] In a first aspect, embodiments of the present invention provide a method for predicting icing faults on power distribution lines, comprising:

[0007] Feature extraction is performed on the raw meteorological monitoring data to construct a target time series set, which includes target time series samples and target time series labels;

[0008] The target time series samples are input into a pre-trained fault prediction model to obtain fault classification results for distribution network lines;

[0009] If the fault classification result indicates the existence of fault data, the fault data is reconstructed to determine the fault prediction result of the distribution network line.

[0010] Secondly, embodiments of the present invention provide a distribution network line icing fault prediction device, comprising:

[0011] The sequence construction module is used to extract features from the raw meteorological monitoring data and construct a target time series set, which includes target time series samples and target time series labels.

[0012] The prediction module is used to input the target time series samples into a pre-trained fault prediction model to obtain the fault classification results of the distribution network lines;

[0013] The reconstruction module is used to reconstruct the fault data if the fault classification result indicates the existence of fault data, and to determine the fault prediction result of the distribution network line.

[0014] Thirdly, embodiments of the present invention also provide an electronic device, comprising:

[0015] At least one processor; and

[0016] A memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the distribution network line icing fault prediction method provided in the first aspect embodiment.

[0018] Fourthly, embodiments of the present invention also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the distribution network line icing fault prediction method as described in the first aspect embodiment.

[0019] This invention provides a method, apparatus, device, and storage medium for predicting icing faults in power distribution lines. The method includes: extracting features from raw meteorological monitoring data to construct a target time series set, which includes target time series samples and target time series labels; inputting the target time series samples into a pre-trained fault prediction model to obtain fault classification results for the power distribution lines; if the fault classification result indicates the presence of fault data, reconstructing the fault data to determine the fault prediction result for the power distribution lines. This technical solution uses a convolutional neural network to extract features from high-dimensional redundant meteorological monitoring data to construct a target time series set; then inputs it into the fault prediction model to predict icing faults in power distribution lines, improving the convergence speed of the fault prediction model and the prediction accuracy for high-dimensional time series; finally, reconstructing the fault data from the prediction results makes the reconstructed fault classification closer to the actual value, improving the accuracy of the prediction results.

[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used 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 those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating a method for predicting icing faults in power distribution lines according to Embodiment 1 of the present invention.

[0023] Figure 2 This is a flowchart illustrating another method for predicting icing faults in power distribution lines provided in Embodiment 2 of the present invention.

[0024] Figure 3 This is a schematic diagram of the structure of a power distribution line icing fault prediction device provided in Embodiment 3 of the present invention;

[0025] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "original," "target," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] Example 1

[0029] Figure 1 This is a flowchart illustrating a method for predicting icing faults in distribution network lines according to Embodiment 1 of the present invention. This method is applicable to the prediction of icing faults in distribution network lines. The method can be executed by a distribution network line icing fault prediction device, which can be implemented in hardware and / or software and can be configured in an electronic device. For example... Figure 1 As shown, the method for predicting icing faults in distribution network lines provided in this embodiment can specifically include the following steps:

[0030] S110. Extract features from the raw meteorological monitoring data and construct a target time series set, which includes target time series samples and target time series labels.

[0031] In this embodiment, to predict icing faults in power distribution lines, it is necessary to acquire real-time meteorological data on icing of power distribution lines. Specifically, the raw meteorological monitoring data may include meteorological data and ground location information. For example, meteorological data includes wind direction, humidity, temperature, air pressure, and precipitation data at different altitudes. Ground location information includes the geographical coordinates and altitude of the power distribution line.

[0032] In this embodiment, to address the issues of redundancy and high dimensionality in meteorological data related to icing on power distribution lines, feature extraction of the raw meteorological monitoring data is necessary. A Convolutional Neural Network (CNN) model is preferably used for feature extraction. The CNN model is constructed through forward and backward propagation, and includes convolutional layers, pooling layers, and fully connected layers. The convolutional layers process the features of the raw meteorological monitoring data and output feature maps through activation functions. The pooling layers perform downsampling, reducing the spatial size of network elements and parameters. Max pooling is used to perform local maximum operations on the input features, obtaining features whose positions remain unchanged. The fully connected layers classify the features extracted by the preceding layers. The training objective of the CNN is to minimize the network's loss function, i.e., to minimize the difference between the predicted and actual values ​​obtained during forward propagation. During error backpropagation, gradient descent is used to minimize the loss function.

[0033] In this embodiment, a convolutional neural network model is constructed using forward and backward propagation to extract meteorological features from high-dimensional time-series meteorological data such as wind speed, wind direction, air pressure, and temperature at different altitudes. Specifically, feature extraction is performed on the raw meteorological monitoring data to obtain meteorological time-series features. After obtaining the meteorological time-series features, a training set is generated from these features. A time series is constructed based on the training set; in this embodiment, the constructed time series is denoted as the target time series. The target time series includes time series samples and time series labels.

[0034] For example, a training set (x1, x2, ... x) is generated from time-series meteorological data obtained through CNN training. n ), where x1, x2, ... x n These represent each meteorological time-series feature. A time series is constructed based on the training set, which includes constructing time series samples and time series labels. The samples and labels for the constructed time series are (X1, X2, ... X...). m ) and (y1, y2, ... y m ), where X1 is (x1, x2, ..., x k-1 The vector composed of x, y1 is the value of x. k X2 is (x2, x3, ..., x k The vector composed of x, y2 is the vector of x. k+1 And so on: X m For (x) m x m+1 , ..., x m+k-1 ), y m For x m+k , where m and k are positive integers.

[0035] S120. Input the target time series samples into the pre-trained fault prediction model to obtain the fault classification results of the distribution network lines.

[0036] In this embodiment, a fault prediction model is obtained by using Long Short-Term Networks (LSTM) and Random Forest (RF) models, trained on long-term meteorological feature data as input. The training process of the fault prediction model can be described as follows: First, a large amount of historical raw meteorological monitoring data is processed by a convolutional neural network model to extract meteorological time-series features. Then, fault-free meteorological time-series features are extracted from the meteorological time-series features to generate a training set. Based on this training set, time-series samples and time-series labels are constructed, and the time-series samples and time-series labels constitute a time series. An LSTM network model is constructed, with the time-series samples as input and the time-series labels as output. When the model loss rate tends to stabilize, training is stopped and model M1 is saved.

[0037] Similarly, fault meteorological time-series features are extracted from meteorological time-series characteristics to generate a training set. Then, time-series samples and time-series labels are constructed based on this training set, forming a time series. The LSTM model M1 generated in the above steps is used to predict the fault sequences. The model input is the fault sequence sample set, and the output is the predicted sequence set. Residual sequence data set (r1, r2, ... r j () is the prediction dataset With the labeled data set (l1, l2, ... l) m The difference between the residual data (r1, r2, ... r) is used to construct and train a random forest classifier, with the input being the residual data set (r1, r2, ... r). j ) and the tag set (l1, l2, ... l m The output is a fault classification. Each decision tree randomly selects the square root of the total number of features to train model M2. Models M1 and M2 are then used as fault prediction models.

[0038] The classification result is the fault status of the distribution network lines. Specifically, after obtaining the fault prediction model, the target time series samples in the target time series are input into the LSTM model M1 of the fault detection model, and the output is a set of predicted sequences. The difference between the predicted sequence set and the target time series labels in the target time series is calculated to obtain a set of residual sequence data. Using the random forest model M2, classification is performed based on the magnitude and order of the residual sequences in the residual sequence data set to obtain the classification result. The classification result may indicate that the classified data contains fault data or that the classified data does not contain fault data.

[0039] For example, an LSTM model M1 is used to predict a fault sequence, with the model input being a time series sample set (X1, X2, ... X...). m The model output is a set of predicted sequences (y1, y2, ... y). m The residual sequence data set (r1, r2, ... r) j () is the prediction dataset With the labeled data set (l1, l2, ... l) m The difference between the residual sequence magnitudes and their order is used to classify the data. The Random Forest Model M2 is then used to classify the data based on the residual sequence magnitudes and their order, yielding the classification results.

[0040] S130. If the classification result indicates the existence of fault data, then the fault data is reconstructed to determine the fault prediction result of the distribution network line.

[0041] Specifically, if the classification result indicates that faulty data exists within the categorized data, the faulty data is reconstructed using the predicted data set obtained in the above steps. This data reconstruction makes the output data more consistent with the actual data, improving the accuracy of the prediction results from the above steps. It's important to note that if the distribution network lines are in a non-faulty state, then faulty data reconstruction is unnecessary.

[0042] For example, if the classification data contains faulty data, then the predicted data set obtained using the LSTM model M1 is used. Reconstruct the faulty data.

[0043] This invention provides a method for predicting icing faults in power distribution lines. The method first extracts features from raw meteorological monitoring data to construct a target time series set, which includes target time series samples and labels. Then, the target time series samples are input into a pre-trained fault prediction model to obtain fault classification results for the power distribution lines. Finally, if the classification result indicates the presence of fault data, the fault data is reconstructed to determine the fault prediction result for the power distribution lines. This technical solution uses a convolutional neural network to extract features from high-dimensional redundant meteorological monitoring data to construct a target time series set. This set is then input into the fault prediction model to predict icing faults in power distribution lines, improving the convergence speed of the fault prediction model and the prediction accuracy for high-dimensional time series. Finally, the predicted results are reconstructed to make the reconstructed fault classification closer to the actual value, thus improving the accuracy of the prediction results.

[0044] As an optional embodiment of the present invention, based on the above embodiments, the training steps of the fault prediction model can be specifically described as including:

[0045] a1. Obtain a training sample set containing meteorological time series features.

[0046] Specifically, the training sample set is obtained by processing a large amount of historical meteorological monitoring data. Feature extraction is performed on this large amount of historical meteorological monitoring data using a convolutional neural network model to obtain meteorological time-series features. It can be understood that the meteorological time-series features should include both fault-free and fault-free meteorological time-series features. The large amount of historical meteorological monitoring data is then used to train the convolutional neural network model, and the fault-free time-series meteorological data obtained from the training of the convolutional neural network model are used to generate the training sample set (x1, x2, ... x...). n Take the fault time-series meteorological data from the training samples to generate a training sample set (e1, e2, ... e...). n ).

[0047] b1. Based on the fault-free meteorological time series characteristics in the training sample set, train the long-term and short-term neural network models to obtain the first sub-prediction model.

[0048] In this step, a time series is constructed based on the fault-free meteorological time series features in the training sample set. The time series includes time series samples and time series labels. An initial LSTM network model is constructed, with the time series samples as input and the time series labels as output. When the model loss rate tends to stabilize, training is stopped and model M1 is saved, which is denoted as the first sub-prediction model in this embodiment.

[0049] Furthermore, based on the fault-free meteorological time-series characteristics in the training sample set, a long-term and short-term neural network model is trained to obtain the first sub-prediction model, including:

[0050] b11. Obtain the fault-free meteorological time series features in the training sample set to form the first training set.

[0051] Specifically, fault-free meteorological time-series features obtained through CNN training are acquired to generate a training set, denoted as the first training set in this embodiment. For example, the first training set is represented as (x1, x2, ... x...). n ), where x1, x2, ... x n Let n represent each fault-free meteorological time series feature, where n is a positive integer.

[0052] b12. Based on the first training set, construct the first time series samples and the first time series labels, and combine the first time series samples and the first time series labels to form the first time series set.

[0053] Specifically, a time series is constructed based on the first training set, which is denoted as the first time series set in this embodiment. The first time series set includes first time series samples and first time series labels. For example, the first training set is represented as (x1, x2, ... x... nThe first time series samples and labels are constructed as (X1, X2, ... X). m ) and (y1, y2, ... y m ), where X1 is (x1, x2, ..., x k-1 The vector composed of x, y1 is the value of x. k X2 is (x2, x3, ..., x k The vector composed of x, y2 is the vector of x. k+1 And so on: X m For (x) m x m+1 , ..., x m+k-1 ), y m For x m+k .

[0054] To simplify calculations, the data is standardized; specifically, time series data is normalized using the following method: X′=(xx min ) / (x max -x min (), where x represents the value before normalization, and X′ represents the value after normalization, x min Let x represent the minimum value. max This represents the maximum value.

[0055] b13. Construct an initial long short-term neural network model, and use the first time series samples as input and the first time series labels as output to train the initial long short-term neural network model.

[0056] Specifically, an LSTM network model is constructed, with the first time series samples as input and the first time series labels as output. The constructed LSTM model is trained using the back-propagation through time (BPTT) algorithm. The loss function is set to root mean square error, and the activation function is adam.

[0057] b14. After the training iteration ends, the trained long short-term neural network model is used as the first sub-prediction model.

[0058] Specifically, the termination condition for training iterations can be understood as the model loss rate stabilizing. Specifically, when the model loss rate stabilizes, training stops, and the trained long short-term neural network model is saved as model M1.

[0059] c1. Based on the fault meteorological time series characteristics in the training sample set, train the random forest model to obtain the second sub-prediction model.

[0060] In this step, a time series is constructed based on the fault meteorological time-series features in the training sample set. The time series includes time series samples and time series labels. The fault sequences are predicted based on the first sub-prediction model, i.e., model M1 obtained in the previous steps. The model input is the fault sequence sample set, and the output is the predicted sequence set. The residual sequence data set is determined based on the predicted sequence set and the time series labels. A random forest classifier is constructed and trained, with the input being the residual data set and the label set, and the output being the fault classification. Model M2 is obtained through training, which is denoted as the second sub-prediction model in this embodiment.

[0061] Furthermore, based on the faulty meteorological time-series characteristics in the training sample set, a random forest model is trained to obtain a second sub-prediction model, including:

[0062] c11. Obtain the fault meteorological time series features in the training sample set to form the second training set.

[0063] Specifically, the fault meteorological time-series features obtained through CNN training are acquired to generate a training set, which is denoted as the second training set in this embodiment. For example, the second training set is represented as (e1, e2, ... e... n ), where e1, e2, ... e n Let n represent the meteorological time series characteristics of each fault, where n is a positive integer.

[0064] c12. Based on the second training set, construct the second time series samples and the second time series labels, and combine the second time series samples and the second time series labels to form the second time series set.

[0065] Specifically, a time series is constructed based on a second training set, which in this embodiment is denoted as the second time series set. The second time series set includes second time series samples and second time series labels. For example, the second training set is represented as (e1, e2, ... e... n Construct a second time series sample and labels (E1, E2, ..., E...). m (l1, l2, ... l) and (l1, l2, ... l) m ), where E m For (e) m e m+1 , ..., e m+k-1 ), l m For e m+k , where m and k are positive integers.

[0066] c13. Input the second time series sample into the first sub-prediction model to obtain the second prediction series.

[0067] Specifically, the LSTM model M1 generated in the above steps is used to predict the fault sequence. The model input is the second time series sample set, and the output is the predicted sequence set, denoted as the second predicted sequence.

[0068] c14. Obtain residual sequence data by subtracting the second predicted sequence and the second time series label.

[0069] Specifically, residual sequence data is generated by subtracting the second predicted sequence from the second time series label. For example, the residual sequence data set (r1, r2, ... r... j () is the prediction dataset With the labeled data set (l1, l2, ... l) m The difference between ).

[0070] c15. Construct an initial random forest classifier, and use the residual sequence data and the second time series label as input, and the fault classification as output to train the initial random forest model.

[0071] Specifically, a random forest classifier is constructed and trained, with the input being the residual data set (r1, r2, ... r... j ) and the tag set (l1, l2, ... l m The output is the fault classification.

[0072] c16. After the training iteration ends, the trained random forest model is used as the second sub-prediction model.

[0073] Specifically, each decision tree randomly selects a number of features equal to the square root of the total number of features, and trains the random forest model as the second sub-prediction model, denoted as model M2.

[0074] d1. Use the first and second sub-prediction models as fault prediction models.

[0075] This optional embodiment specifies the training steps of the fault prediction model. Compared with traditional algorithms, it adopts long short-term neural networks and random forest algorithms, which have shorter iteration time, higher prediction accuracy, and prediction results that are closer to the actual values. It solves the defects of single intelligent algorithms and achieves faster convergence speed and prediction accuracy than ordinary traditional algorithms.

[0076] Example 2

[0077] Figure 2This is a flowchart illustrating another method for predicting icing faults in power distribution lines provided in Embodiment 2 of the present invention. This embodiment is a further optimization of the above embodiment. In this embodiment, the step of "extracting features from the original meteorological monitoring data and constructing a target time series set" is further optimized to use a set neural network model to extract features from the original meteorological monitoring data and determine the meteorological time series features; based on the meteorological time series features, a training set is determined; based on the training set, initial time series samples and initial time series labels are constructed, and the initial time series samples and initial time series labels constitute an initial time series set; the initial time series set is normalized to obtain the target time series set.

[0078] Furthermore, the phrase "inputting the target time series sample into a pre-trained fault prediction model to obtain the classification result of the distribution network line" is further optimized to: inputting the target time series sample into a first sub-prediction model to obtain a target prediction sequence; subtracting the target prediction sequence from the target time series label to determine the target residual sequence; and classifying the amplitude and order in the target residual sequence according to the first sub-prediction model to determine the fault classification result of the distribution network line.

[0079] Furthermore, the method of "if the classification result indicates the existence of fault data, then reconstructing the fault data to determine the fault prediction result of icing on the distribution network line" includes: if the classification result indicates the existence of fault data, then reconstructing the fault data according to the target prediction sequence to obtain estimated data; and determining the fault prediction result of icing on the distribution network line based on the estimated data and the fault data.

[0080] like Figure 2 As shown in the figure, this embodiment 2 provides a method for predicting icing faults in power distribution lines, which specifically includes the following steps:

[0081] S210. Use the established neural network model to extract features from the original meteorological detection data and determine the meteorological time series characteristics.

[0082] Preferably, the neural network model is a convolutional neural network model, which is built based on forward propagation and backward propagation. The convolutional neural network model includes convolutional layers, pooling layers, and fully connected layers. The convolutional layers are used to process the raw meteorological detection data and output feature maps through activation functions; the pooling layers are used for downsampling operations to reduce the spatial size of network elements and parameters; and the fully connected layers are used to classify the features extracted by the previous layers.

[0083] In this embodiment, the construction process of the convolutional neural network model includes:

[0084] Forward propagation

[0085] Convolutional layer: The convolutional layer processes the input features of meteorological data and geographical location information, and outputs a feature map through an activation function.

[0086] For example, the input data includes meteorological data such as wind direction, wind speed, temperature, humidity, air pressure, and precipitation data at different altitudes from 10 meters to 150 meters along the route, and geographic location information such as the geographical coordinates and altitude of the route obtained from a geographic information system. The specific convolution process of the kernel function is as follows:

[0087] In the formula, This is the output of the j-th channel in layer l-1; Output for the j-th channel of layer l; This is the weight matrix corresponding to the convolution kernel; f is the bias term; f is the activation function, and this paper chooses the ReLU function, which is expressed as f(x) = max(0, x).

[0088] Pooling Layers: Pooling layers perform downsampling, reducing the spatial size of network elements and parameters. This paper uses max pooling, which performs local maximum operations on the input features to obtain features whose positions remain unchanged. The max pooling transformation is as follows:

[0089] Fully connected layer: The fully connected layer is used to classify the features extracted by the preceding multi-layer network. The input is a weighted sum of one-dimensional vectors expanded from all feature maps, obtained by inputting a one-dimensional vector into an activation function. The mathematical expression is as follows: y l =f(w l x l-1 +b l In the formula, y l The output of the fully connected layer; x l-1 This is a one-dimensional vector expanded from the feature map. This paper uses softmax as the classification function for the fully connected layer, as shown below: In the formula, It is an exponential function, where k is the number of categories, and q(x) is the number of categories. j ) represents the function's output value.

[0090] Backpropagation

[0091] The training objective of a convolutional neural network is to minimize the network's loss function, that is, to minimize the difference between the predicted value obtained during forward propagation and the actual value. In this embodiment, the cross-entropy function is selected as the loss function, and its expression is as follows: In the formula, n is the number of samples of this type of fault; d i and y iLet w and b be the predicted and true values ​​of the i-th sample, respectively. During error backpropagation, gradient descent is used to minimize the loss function, and the adaptive parameters w and b of the network are updated layer by layer by taking the derivative of the loss function. The specific calculation method is as follows: In the formula, w′ and b′ are the updated weights and biases; w and b are the existing weights and biases; η is the learning rate, used to control the step size of weight updates.

[0092] In this embodiment, a convolutional neural network model is constructed through forward and backward propagation to extract meteorological features from high-dimensional time-series meteorological data such as wind speed, wind direction, air pressure, and temperature at different altitudes. This enables the high-dimensional time-series feature quantities to be reduced in dimensionality and input into the RF-LSTM, thus solving the inherent problems of low prediction accuracy and slow convergence speed of time-series prediction networks when faced with high-dimensional data input.

[0093] Specifically, a set neural network model is used to extract features from the raw meteorological data to determine the meteorological time series characteristics.

[0094] S220. Determine the training set based on meteorological time series characteristics.

[0095] Specifically, the meteorological time-series features obtained from the above steps of CNN training are used to generate a training set. For example, the training set can be represented as (x1, x2, ... x...). n ).

[0096] S230. Based on the training set, construct the initial time series samples and initial time series labels, and combine the initial time series samples and initial time series labels to form the initial time series set.

[0097] Specifically, time series are constructed based on the training set. The initial time series samples and initial time series labels are (X1, x2, ... X...). m ) and (y1, y2, ... y m ), where X1 is (x1, x2, ..., x k-1 The vector composed of x, y1 is the value of x. k X2 is (x2, x3, ..., x k The vector composed of x, y2 is the vector of x. k+1 And so on: X m For (x) m x m+1 , ..., x m+k-1 ), y m For x m+k .

[0098] S240. Normalize the initial time series set to obtain the target time series set.

[0099] Specifically, to standardize the data and facilitate calculation, the initial time series set is normalized using the method X′=(xx min ) / (x max -x min ).

[0100] S250. Input the target time series sample into the first sub-prediction model to obtain the target prediction sequence.

[0101] Specifically, the target time series samples are input into the first sub-prediction model, that is, the LSTM model M1 is used to predict the fault sequence. The model input is the time series sample set (X1, X2, ... X... m The model output is a set of predicted sequences (y1, y2, ... y). m In this embodiment, denoted as the target prediction sequence.

[0102] S260. Subtract the target predicted sequence from the target time series to determine the target residual sequence.

[0103] Specifically, the target residual sequence is the difference between the target predicted sequence and the target time series label.

[0104] S270. Based on the second sub-prediction model, classify the magnitude and order in the target residual sequence to determine the fault classification results of the distribution network lines.

[0105] Specifically, the Random Forest Model M2 is used to classify faults based on the magnitude and order of the target residual sequence. The fault classification result is the fault status of the distribution network lines. The fault classification result may indicate that the classified data contains faults or that the classified data does not contain faults.

[0106] S280. If the fault classification result indicates the existence of fault data, then the fault data is reconstructed based on the target prediction sequence to obtain the estimated data.

[0107] Specifically, nominally normal measurements are estimated based on the measured values ​​of process variables already affected by the fault, and the severity of the fault can be estimated through fault reconstruction. If the fault classification result indicates that faulty data exists within the categorized data, the faulty data is reconstructed based on the target prediction sequence to obtain estimated data that more closely approximates the true values. The estimated data can be understood as the normal data values ​​corresponding to the faulty data.

[0108] S290. Based on the estimated data and fault data, determine the fault prediction results for icing on distribution network lines.

[0109] Specifically, by comparing the predicted data with the fault data, the difference between the fault data and the predicted data can be further determined, and the fault prediction result of icing on the distribution network lines can be determined based on the difference.

[0110] Specifically, if the classification data contains faulty data, then the target prediction sequence obtained in the above steps, i.e., the prediction data set obtained by the LSTM model M1, is used. Reconstruct the fault data. Data reconstruction makes the output data more consistent with the actual data, improving the accuracy of LSTM prediction results. If the operating state of the distribution network lines is not a fault state, reconstruction is not required.

[0111] Example 3

[0112] Figure 3 This is a schematic diagram of a distribution network line icing fault prediction device provided in Embodiment 3 of the present invention. This device is applicable to situations involving chip layout and can be configured in electronic devices, such as... Figure 3 As shown, the device includes: a sequence construction module 31, a prediction module 32, and a reconstruction module 33; wherein,

[0113] The sequence construction module 31 is used to extract features from the raw meteorological monitoring data and construct a target time series set, which includes target time series samples and target time series labels.

[0114] The prediction module 32 is used to input the target time series samples into the pre-trained fault prediction model to obtain the fault classification results of the distribution network lines;

[0115] The reconstruction module 33 is used to reconstruct the fault data if the fault classification result indicates that fault data exists, and to determine the fault prediction result of the distribution network line.

[0116] This invention provides a device for predicting icing faults in power distribution lines. First, features are extracted from raw meteorological monitoring data to construct a target time series set, which includes target time series samples and labels. Then, the target time series samples are input into a pre-trained fault prediction model to obtain fault classification results for the power distribution lines. Finally, if the fault classification result indicates the presence of fault data, the fault data is reconstructed to determine the fault prediction result for the power distribution lines. This technical solution uses a convolutional neural network to extract features from high-dimensional redundant meteorological monitoring data to construct a target time series set. This set is then input into the fault prediction model to predict icing faults in power distribution lines, improving the convergence speed of the fault prediction model and the prediction accuracy for high-dimensional time series. Finally, the predicted results are reconstructed to make the reconstructed fault classification closer to the actual value, thus improving the accuracy of the prediction results.

[0117] Optionally, the device also includes a model training module, comprising:

[0118] The acquisition unit is used to acquire a training sample set containing meteorological time-series features;

[0119] The first model determination unit is used to train a long-term and short-term neural network model based on the fault-free meteorological time-series characteristics in the training sample set to obtain the first sub-prediction model.

[0120] The second model determination unit is used to train a random forest model based on the faulty meteorological time series features in the training sample set to obtain the second sub-prediction model.

[0121] The model determination unit is used to use the first sub-prediction model and the second sub-prediction model as the fault prediction model.

[0122] Optionally, the first model determines the unit, specifically used for:

[0123] Obtain fault-free meteorological time-series features from the training sample set to form the first training set;

[0124] Based on the first training set, construct the first time series samples and the first time series labels, and combine the first time series samples and the first time series labels to form the first time series set;

[0125] An initial long short-term neural network model is constructed, and the first time series sample is used as input and the first time series label is used as output to train the initial long short-term neural network model;

[0126] After the training iteration ends, the trained long short-term neural network model is used as the first sub-prediction model.

[0127] Optionally, the second model determining unit is specifically used for:

[0128] Obtain the fault meteorological time series features from the training sample set to form a second training set;

[0129] Based on the second training set, construct a second time series sample and a second time series label, and combine the second time series sample and the second time series label to form a second time series set;

[0130] The second time series sample is input into the first sub-prediction model to obtain the second prediction series;

[0131] The residual sequence data is obtained by subtracting the second predicted sequence and the second time series labels.

[0132] An initial random forest classifier is constructed, and the residual sequence data and the second time series labels are used as inputs, with fault classification as the output, to train the initial random forest model;

[0133] After the training iteration ends, the trained random forest model is used as the second sub-prediction model.

[0134] Optional, sequence building module 31, specifically used for:

[0135] A predefined neural network model was used to extract features from the raw meteorological data to determine the meteorological time series characteristics.

[0136] The training set was determined based on the meteorological time series characteristics;

[0137] Based on the training set, construct initial time series samples and initial time series labels, and combine the initial time series samples and initial time series labels to form an initial time series set;

[0138] The initial time series set is normalized to obtain the target time series set.

[0139] Optionally, the neural network model is set as a convolutional neural network model. The convolutional neural network model is built based on forward propagation and backward propagation. The convolutional neural network model includes convolutional layers, pooling layers, and fully connected layers. The convolutional layers are used to process the raw meteorological detection data and output feature maps through activation functions. The pooling layers are used for downsampling operations to reduce the spatial size of network elements and parameters. The fully connected layers are used to classify the features extracted by the previous layers of the network.

[0140] Optionally, the target time series samples are input into a pre-trained fault prediction model to obtain the classification results of the distribution network lines, including:

[0141] The target time series sample is input into the first sub-prediction model to obtain the target prediction series;

[0142] The target residual sequence is determined by subtracting the target predicted sequence and the target time series.

[0143] The second sub-prediction model is used to classify the magnitude and order of the target residual sequence to determine the fault classification results of the distribution network lines.

[0144] Optionally, refactor module 33, specifically used for:

[0145] If the classification result indicates the presence of faulty data, the faulty data is reconstructed based on the target prediction sequence to obtain the estimated data.

[0146] Based on the estimated data and fault data, the fault prediction results for icing on distribution network lines are determined.

[0147] The distribution network line icing fault prediction device provided in this embodiment of the invention can execute the distribution network line icing fault prediction method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0148] Example 4

[0149] Figure 4 This is a schematic diagram of an electronic device according to Embodiment 4 of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0150] like Figure 4 As shown, the electronic device 40 includes at least one processor 41 and a memory, such as a read-only memory (ROM) 42 or a random access memory (RAM) 43, communicatively connected to the at least one processor 41. The memory stores computer programs executable by the at least one processor. The processor 41 can perform various appropriate actions and processes based on the computer program stored in the ROM 42 or loaded into the RAM 43 from storage unit 48. The RAM 43 may also store various programs and data required for the operation of the electronic device 40. The processor 41, ROM 42, and RAM 43 are interconnected via a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0151] Multiple components in electronic device 40 are connected to I / O interface 45, including: input unit 46, such as keyboard, mouse, etc.; output unit 47, such as various types of monitors, speakers, etc.; storage unit 48, such as disk, optical disk, etc.; and communication unit 49, such as network card, modem, wireless transceiver, etc. Communication unit 49 allows electronic device 40 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0152] Processor 41 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 41 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 41 performs the various methods and processes described above, such as the method for predicting icing faults in power distribution lines.

[0153] In some embodiments, the distribution network line icing fault prediction method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 48. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 40 via ROM 42 and / or communication unit 49. When the computer program is loaded into RAM 43 and executed by processor 41, one or more steps of the distribution network line icing fault prediction method described above may be performed. Alternatively, in other embodiments, processor 41 may be configured to perform the distribution network line icing fault prediction method by any other suitable means (e.g., by means of firmware).

[0154] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0155] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0156] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0157] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0158] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0159] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0160] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0161] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A power distribution network line icing failure prediction method characterized by, The method comprises the following steps: feature extraction is performed on original meteorological monitoring data to construct a target time series set, which contains target time series samples and target time series labels; the target time series samples are input into a pre-trained fault prediction model to obtain a fault classification result of a power distribution network line; if the fault classification result is fault data, the fault data is reconstructed to determine a fault prediction result of the power distribution network line; wherein the training step of the fault prediction model comprises: obtaining a training sample set containing meteorological time sequence features; training a long short-term neural network model according to the fault-free meteorological time sequence features in the training sample set to obtain a first sub-prediction model; training a random forest model according to the fault meteorological time sequence features in the training sample set to obtain a second sub-prediction model; using the first sub-prediction model and the second sub-prediction model as the fault prediction model; wherein the training of the long short-term neural network model according to the fault-free meteorological time sequence features in the training sample set to obtain the first sub-prediction model comprises: obtaining fault-free meteorological time sequence features in the training sample set to form a first training set; constructing first time series samples and first time series labels according to the first training set, and forming a first time series set from the first time series samples and the first time series labels; constructing an initial long short-term neural network model, inputting the first time series samples, and outputting the first time series labels to train the initial long short-term neural network model; after meeting the training iteration end condition, using the trained long short-term neural network model as the first sub-prediction model; wherein the training of the random forest model according to the fault meteorological time sequence features in the training sample set to obtain the second sub-prediction model comprises: obtaining fault meteorological time sequence features in the training sample set to form a second training set; constructing second time series samples and second time series labels according to the second training set, and forming a second time series set from the second time series samples and the second time series labels; inputting the second time series samples into the first sub-prediction model to obtain a second prediction sequence; obtaining residual sequence data by differencing the second prediction sequence and the second time series labels; constructing an initial random forest model, inputting the residual sequence data and the second time series labels, and outputting fault classification to train the initial random forest model; after meeting the training iteration end condition, using the trained random forest model as the second sub-prediction model; wherein the inputting of the target time series samples into the pre-trained fault prediction model to obtain the fault classification result of the power distribution network line comprises: inputting the target time series samples into the first sub-prediction model to obtain a target prediction sequence; determining a target residual sequence by differencing the target prediction sequence and the target time series labels; According to the second sub-prediction model, the amplitudes and orders in the target residual sequence are classified to determine a fault classification result of the power distribution network line.

2. The method of claim 1, wherein, The feature extraction on the original meteorological monitoring data and the construction of the target time sequence include: The feature extraction on the original meteorological monitoring data is performed by using a set neural network model to determine meteorological time sequence features; According to the meteorological time sequence features, a training set is determined; According to the training set, an initial time sequence sample and an initial time sequence label are constructed to form an initial time sequence set; The initial time sequence set is normalized to obtain a target time sequence set.

3. The method of claim 2, wherein, The set neural network model is a convolutional neural network model, which is constructed based on forward propagation and back propagation. The convolutional neural network model includes a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer is used to process the original meteorological detection data and output a feature map through an activation function. The pooling layer is used for downsampling operation to reduce the spatial size of network elements and parameters. The fully connected layer is used to classify the features extracted by the previous layers.

4. The method of claim 1, wherein, If the classification result is fault data, the fault data is reconstructed to determine the fault prediction result of the power distribution network line icing, including: If the classification result is fault data, the fault data is reconstructed according to the target prediction sequence to obtain estimated data; According to the estimated data and the fault data, the fault prediction result of the power distribution network line icing is determined.

5. A power distribution network line icing failure prediction apparatus characterized by comprising: It includes: A sequence construction module is configured to extract features from original meteorological monitoring data and construct a target time sequence set, which includes a target time sequence sample and a target time sequence label; A prediction module is configured to input the target time sequence sample into a pre-trained fault prediction model to obtain a fault classification result of a power distribution network line; A reconstruction module is configured to reconstruct the fault data if the fault classification result is fault data to determine the fault prediction result of the power distribution network line. The device further includes a model training module, which includes: An acquisition unit is configured to acquire a training sample set containing meteorological time sequence features; A first model determination unit is configured to train a long short-term neural network model according to the fault-free meteorological time sequence features in the training sample set to obtain a first sub-prediction model; A second model determination unit is configured to train a random forest model according to the fault meteorological time sequence features in the training sample set to obtain a second sub-prediction model; A model determination unit is configured to use the first sub-prediction model and the second sub-prediction model as a fault prediction model; The first model determination unit is specifically configured to: Acquire the fault-free meteorological time sequence features in the training sample set to form a first training set; According to the first training set, a first time sequence sample and a first time sequence label are constructed to form a first time sequence set; constructing an initial long short-term neural network model, inputting the first time sequence sample as input and the first time sequence label as output, and training the initial long short-term neural network model; after meeting the training iteration end condition, taking the trained long short-term neural network model as the first sub-prediction model; The second model determination unit is specifically configured to: obtain the fault weather time sequence features in the training sample set to form a second training set; constructing a second time sequence sample and a second time sequence label according to the second training set, and forming a second time sequence set from the second time sequence sample and the second time sequence label; inputting the second time sequence sample into the first sub-prediction model to obtain a second prediction sequence; obtaining residual sequence data by differencing the second prediction sequence and the second time sequence label; constructing an initial random forest model, inputting the residual sequence data and the second time sequence label as input and taking fault classification as output, and training the initial random forest model; after meeting the training iteration end condition, taking the trained random forest model as the second sub-prediction model; The prediction module is specifically configured to: inputting the target time sequence sample into the first sub-prediction model to obtain a target prediction sequence; determining a target residual sequence by differencing the target prediction sequence and the target time sequence label; classifying the amplitude and order in the target residual sequence according to the second sub-prediction model to determine the fault classification result of the distribution network line.

6. An electronic device, comprising: comprises: at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the distribution network line icing fault prediction method of any one of claims 1-4.

7. A storage medium containing computer-executable instructions, wherein: The computer executable instructions, when executed by the computer processor, are used to execute the distribution network line icing fault prediction method of any one of claims 1-4.

Images