Power distribution network fault positioning method and device, terminal equipment and storage medium

Abstract

The invention discloses a power distribution network fault positioning method and device, terminal equipment and a storage medium, and belongs to the technical field of power fault diagnosis, and the method comprises the steps: carrying out the data fusion of operation state data and operation environment data, and inputting the fused data into a fault detection model, and the fault detection model extracts deep features of the data through a deep belief network, extracts spatial features of the data through a convolutional neural network, further generates a prediction fault type and a prediction fault range, and further generates fault positioning information. The fault detection model extracts the deep features and the spatial features of the multi-source fusion data to perform fault prediction, generates the prediction fault type and the prediction fault range, and solves the problem that the deep features and the spatial features of the multi-source heterogeneous data are often ignored when the data is processed by a traditional method.

Description

Technical Field

[0001] The present invention relates to the technical field of power fault diagnosis, and in particular to a distribution network fault locating method, device, terminal equipment and storage medium. Background Art

[0002] In the field of distribution network fault location technology, the advancement of smart grid concepts has led to continuous advancements in related technologies. Early fault location techniques were simple, relying on relay protection device operation information and simple electrical quantity measurement principles. This was prone to misjudgment and inaccurate positioning. Although the subsequent installation of numerous sensors to monitor electrical and environmental data provided a foundation for precise positioning, and machine learning and deep learning technologies introduced new insights, traditional fault location techniques still had limitations. When processing heterogeneous data from multiple sources, traditional methods often overlooked the deep and spatial characteristics of this data, resulting in low fault location accuracy. Summary of the Invention

[0003] The present invention provides a distribution network fault location method, apparatus, terminal equipment and storage medium, which can solve the problem of ignoring the deep characteristics and spatial characteristics of data in the prior art.

[0004] An embodiment of the present invention provides a method for locating a distribution network fault, comprising:

[0005] Acquire first operating status data and first operating environment data of each node in the distribution network in real time; wherein the operating status data includes: voltage data, current data and temperature data; the operating environment data includes: ambient humidity data, ambient wind speed data and ambient gas concentration data;

[0006] Performing data fusion on the first operating state data and the first operating environment data to obtain multi-source fused data;

[0007] The multi-source fusion data is input into a fault detection model so that the fault detection model extracts deep features of the multi-source fusion data through a deep belief network, extracts spatial features of the multi-source fusion data through a convolutional neural network, and generates predicted fault types and predicted fault ranges based on the deep features and the spatial features.

[0008] Furthermore, the fusing the first operating status data and the first operating environment data to obtain multi-source fused data includes:

[0009] Completing the first operating state data and the first operating environment data respectively by a linear interpolation method to obtain second operating state data and second operating environment data;

[0010] Determine a lower limit based on the first quartile, determine an upper limit based on the second quartile, and remove data outside the lower limit and upper limit ranges of the second operating status data and the second operating environment data, respectively, to obtain third operating status data and third operating environment data;

[0011] The third operating state data and the third operating environment data are respectively standardized by a Z-Score method to obtain operating state standardized data and operating environment standardized data;

[0012] By using the principal component analysis method, the principal component features of the operating state data and the principal component features of the operating environment data are extracted from the standardized operating state data and the standardized operating environment data respectively;

[0013] The principal component features of the operating status data and the principal component features of the operating environment data are fused by the weighted average method to obtain fused data;

[0014] The fused data is restored to the same dimensional space as the first operating state data and the first operating environment data through inverse transformation processing to obtain the multi-source fused data.

[0015] Furthermore, generating the predicted fault type and predicted fault range based on the deep features and the spatial features includes:

[0016] Fusing the deep features and the spatial features to obtain fused features;

[0017] The fusion features are inferred by a classifier to generate the predicted fault type and predicted fault range.

[0018] Furthermore, it also includes:

[0019] Generate an optimized fault type based on the predicted fault type and the multi-source fusion data through a random forest algorithm;

[0020] Optimizing the fault occurrence probability distribution model based on the multi-source fusion data to obtain an optimized fault occurrence probability distribution model;

[0021] Inputting the predicted fault range into the optimized fault occurrence probability distribution model to generate an optimized fault range;

[0022] generating the fault location information according to the optimized fault type and the optimized fault range;

[0023] The method for constructing the fault occurrence probability distribution model includes:

[0024] Obtaining distribution network elements and distribution network association information of the distribution network; wherein the distribution network elements include device nodes and their operating status parameters; the distribution network association information is the association relationship between the distribution network elements;

[0025] Constructing a Bayesian network with the distribution network elements as nodes and the distribution network related information as edges;

[0026] According to historical fault data, basic probability values ​​of fault occurrence are assigned to the nodes of the Bayesian network to obtain a fault occurrence probability distribution model.

[0027] Furthermore, generating an optimized fault type by using a random forest algorithm based on the predicted fault type and the multi-source fusion data includes:

[0028] Generate a corresponding one-hot encoding according to the predicted fault type;

[0029] According to the one-hot encoding, a number of decision trees are constructed using a random forest algorithm; wherein each decision tree is split according to the multi-source fusion data;

[0030] According to the decision tree, the optimized fault type is determined through a majority voting mechanism.

[0031] Furthermore, optimizing the fault occurrence probability distribution model based on the multi-source fusion data to obtain the optimized fault occurrence probability distribution model includes:

[0032] Calculating the similarity between the multi-source fusion data and the historical fault data;

[0033] When the similarity exceeds a preset similarity threshold, the basic probability value of the fault occurrence in the fault occurrence probability distribution model is increased to obtain an optimized fault occurrence probability distribution model.

[0034] Furthermore, inputting the predicted fault range into the optimized fault occurrence probability distribution model to generate the optimized fault range includes:

[0035] Inputting the predicted fault range into the optimized fault occurrence probability distribution model, so that the fault occurrence probability distribution model calculates the conditional probability value of the fault occurrence of each node according to the predicted fault range and the basic probability value of the fault occurrence;

[0036] The optimized fault range is generated based on nodes whose fault occurrence condition probability values ​​exceed a preset probability threshold.

[0037] Another embodiment of the present invention further provides a distribution network fault location device, comprising: a data acquisition module, a data fusion module and a fault location module;

[0038] The data acquisition module is used to acquire first operating status data and first operating environment data of each node in the distribution network in real time; wherein the operating status data includes: voltage data, current data and temperature data; the operating environment data includes: ambient humidity data, ambient wind speed data and ambient gas concentration data;

[0039] The data fusion module performs data fusion on the first operating state data and the first operating environment data to obtain multi-source fusion data;

[0040] The fault location module is used to input the multi-source fusion data into a fault detection model, so that the fault detection model extracts the deep features of the multi-source fusion data through a deep belief network, extracts the spatial features of the multi-source fusion data through a convolutional neural network, and generates a predicted fault type and a predicted fault range based on the deep features and the spatial features.

[0041] Another embodiment of the present invention further provides a terminal device, comprising: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps of the distribution network fault location method of the present invention are implemented.

[0042] Another embodiment of the present invention further provides a computer-readable storage medium item, comprising: a stored computer program, which controls the device where the computer-readable storage medium is located to execute the steps of the distribution network fault location method of the present invention when the computer program is executed.

[0043] The following beneficial effects are achieved by implementing the present invention:

[0044] The present invention obtains first operating status data and first operating environment data of each node in a distribution network in real time; wherein the operating status data includes voltage data, current data, and temperature data; and the operating environment data includes ambient humidity data, ambient wind speed data, and ambient gas concentration data; performs data fusion on the first operating status data and the first operating environment data to obtain multi-source fused data; and inputs the multi-source fused data into a fault detection model, so that the fault detection model extracts deep features of the multi-source fused data through a deep belief network, extracts spatial features of the multi-source fused data through a convolutional neural network, and generates a predicted fault type and a predicted fault range based on the deep features and the spatial features. The fault detection model of the present invention extracts deep features of the multi-source fused data through a deep belief network and extracts spatial features of the multi-source fused data through a convolutional neural network, generates a predicted fault type and a predicted fault range based on the deep features and the spatial features, and further generates fault location information. The present invention solves the problem that traditional methods often ignore the deep and spatial characteristics of multi-source heterogeneous data when processing data, and can improve the accuracy of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0046] Figure 1 This is a flow chart of a method for locating a distribution network fault provided by one embodiment of the present invention;

[0047] Figure 2 The figure is a schematic structural diagram of a distribution network fault location device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the term "include" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.

[0050] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "plurality" is more than two, unless otherwise clearly and specifically defined.

[0051] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0052] See also Figure 1To solve the problem of ignoring the deep and spatial characteristics of data in the prior art, an embodiment of the present invention provides a distribution network fault location method, including:

[0053] S1. Acquire first operating status data and first operating environment data of each node in the distribution network in real time; wherein the operating status data includes: voltage data, current data and temperature data; the operating environment data includes: ambient humidity data, ambient wind speed data and ambient gas concentration data.

[0054] In step S1, the first operating status data includes first voltage data, first current data, and first temperature data; the first operating environment data includes first ambient humidity data, first ambient wind speed data, and first ambient gas concentration data. The first voltage data refers to the voltage data of the bus or line, including the rated voltage and fluctuation range, and is collected through a voltage transformer. The first current data refers to the current data in the line, including normal operating current and abnormal conditions such as short-circuit current, and is collected through a current transformer. The first equipment temperature refers to the temperature data of key equipment, including the fluctuation range of the normal equipment temperature and the ambient temperature, and is collected through a temperature sensor. The first ambient humidity data is collected through a relative humidity sensor and expressed as a percentage, with a numerical range of 0% to 100%. The first ambient wind speed data is collected through a wind speed sensor, using ultrasonic speed measurement or three-cup mechanical speed measurement technology. The wind speed value is in meters per second, and the monitoring range is from static to high wind speed conditions. The first ambient gas concentration is collected through multiple types of gas sensors, which can detect methane, carbon dioxide, oxygen, or harmful gases, and the concentration value is in ppm or mg / m 3 The unit is gas, and the monitoring range is set according to the gas type.

[0055] It should be noted that the collected analog signals will be converted into digital signals through a high-precision analog-to-digital conversion module. Subsequently, the collected first operating status data and first operating environment data will be transmitted to the processing center through wired or wireless communication methods, including optical fiber communication, wireless local area network or narrowband Internet of Things, to ensure real-time and reliability.

[0056] S2. Perform data fusion on the first operating status data and the first operating environment data to obtain multi-source fusion data.

[0057] In a preferred embodiment, the fusing the first operating status data and the first operating environment data to obtain multi-source fused data includes:

[0058] Completing the first operating state data and the first operating environment data respectively by a linear interpolation method to obtain second operating state data and second operating environment data;

[0059] Determine a lower limit based on the first quartile, determine an upper limit based on the second quartile, and remove data outside the lower limit and upper limit ranges of the second operating status data and the second operating environment data, respectively, to obtain third operating status data and third operating environment data;

[0060] The third operating state data and the third operating environment data are respectively standardized by a Z-Score method to obtain operating state standardized data and operating environment standardized data;

[0061] By using the principal component analysis method, the principal component features of the operating state data and the principal component features of the operating environment data are extracted from the standardized operating state data and the standardized operating environment data respectively;

[0062] The principal component features of the operating status data and the principal component features of the operating environment data are fused by the weighted average method to obtain fused data;

[0063] The fused data is restored to the same dimensional space as the first operating state data and the first operating environment data through inverse transformation processing to obtain the multi-source fused data.

[0064] In this embodiment, first, the first operating state data and the first operating environment data are complemented by a linear interpolation method to obtain the second operating state data and the second operating environment data.

[0065] Secondly, the second operating status data and the second operating environment data are sorted, and after sorting, the upper and lower interquartile ranges of the second operating status data and the upper and lower interquartile ranges of the second operating environment data are calculated. The upper and lower interquartile ranges are the difference between the third quartile and the first quartile. The third quartile refers to the critical value that divides the data into below 75% and above 25%, and the first quartile refers to the critical value that divides the data into below 25% and above 75%. Usually, the first quartile minus 1.5 times the upper and lower interquartile ranges is used as the lower limit, and the third quartile plus 1.5 times the upper and lower interquartile ranges is used as the upper limit. Values ​​outside this range are judged as outliers and are eliminated to obtain the third operating status data and the third operating environment data.

[0066] The formula for calculating the upper and lower interquartile range is:

[0067] IQR = Q3 - Q1;

[0068] Where IQR represents the interquartile range; Q3 represents the third quartile; and Q1 represents the first quartile.

[0069] Next, calculate the mean and standard deviation of the third operating status data and the third operating environment data. Taking the operating status data as an example, the mean is calculated by dividing the sum of the effective operating statuses of all nodes by the number of nodes, and the standard deviation is the average of the square differences of the operating status means of all nodes and then taking the square root. The same method is also applicable to the operating environment data. The third operating status data and the third operating environment data are standardized using the Z-Score formula, that is, each third operating status data and the third operating environment data is subtracted from the mean and divided by the standard deviation. Through this process, the third operating status data and the third operating environment data will be converted into operating status standardized data and operating environment standardized data with a mean of zero and a unit standard deviation as the distribution, eliminating the influence of different dimensions on subsequent analysis.

[0070] The expression for normalization is:

[0071]

[0072] In the formula, Z represents the data after standardization; O represents the data before standardization; μ represents the mean of the data; σ represents the standard deviation of the data.

[0073] Then, based on the standardized operating status data and operating environment data, the data covariance matrix was constructed, and the eigenvalues ​​and eigenvectors were calculated. The principal components with large eigenvalues ​​were selected as the main features, and the principal component features of the operating status data and the operating environment data were extracted. These principal components can reflect the main information of the original data and reduce the dimensionality of the distribution network operating status data and the distribution network operating environment data.

[0074] Next, the principal component features of the extracted operating status data and operating environment data are weighted averaged to generate fused data. The weight distribution can be adjusted based on the importance or contribution of the distribution network operating status data and the distribution network operating environment data to ensure that the fused features can comprehensively reflect the characteristics of multiple source data.

[0075] The expression of weighted fusion is:

[0076]

[0077] In the formula, G represents the fused data; p represents the number of principal component features; w i represents the weight of the i-th principal component feature; X i represents the i-th principal component feature.

[0078] Finally, the fusion is restored to the original dimensional space of the first operating state data and the first operating environment data through inverse transformation to obtain multi-source fusion data, so that the multi-source fusion data retains the main feature information and conforms to the expression form of the original data.

[0079] S3. Input the multi-source fusion data into a fault detection model so that the fault detection model extracts deep features of the multi-source fusion data through a deep belief network, extracts spatial features of the multi-source fusion data through a convolutional neural network, and generates a predicted fault type and a predicted fault range based on the deep features and the spatial features.

[0080] It should be noted that the fault detection model combines the feature representation capability of DBN and the spatial information processing capability of CNN.

[0081] The method for constructing a fault detection model includes:

[0082] The feature representation capability of DBN and the spatial information processing capability of CNN are integrated through the splicing and fusion method to construct a hybrid deep learning model, which is optimized using the backpropagation algorithm.

[0083] Specifically, a concatenation and fusion method is used to sequentially arrange the deep semantic feature vectors of the DBN and the spatial information feature vectors of the CNN. For example, if the concatenation dimension is the feature axis, the two feature vectors are concatenated end-to-end in that direction to form a new fused feature vector. Based on the fused feature vector, the upper layer of the hybrid deep learning model is constructed, such as adding a fully connected layer to process and analyze the fused features for fault prediction. A loss function is defined based on the fault location and prediction objective, such as the cross-entropy loss function for classification tasks and the mean squared error loss function for regression tasks. All parameters in the hybrid deep learning model, except for the trained parameters of the DBN and CNN, are randomly initialized. The fused feature vector is then fed into the hybrid deep learning model for forward propagation to obtain a prediction result. The prediction result is then compared with the true label and the loss value is calculated using the loss function. Then, a backpropagation algorithm is used to calculate the gradient layer by layer from the output layer according to the chain rule. Based on the gradient, an optimization algorithm is applied to update all parameters of the hybrid deep learning model to minimize the loss value. The process of forward propagation, loss calculation, and optimization of the weight bias of each layer of the hybrid deep learning model is repeated continuously, and optimization is performed iteratively multiple times until the cross entropy loss function value converges or reaches the preset number of training rounds, completing the construction and optimization of the hybrid deep learning model.

[0084] The training process of the fault detection model includes:

[0085] Acquire a number of historical multi-source heterogeneous data and corresponding fault types and fault ranges; wherein each historical multi-source heterogeneous data includes the historical operating status data and historical operating environment data of each node in the distribution network at a certain historical moment;

[0086] Perform the same preprocessing and fusion operations as the first operating state data and the first operating environment data on each historical operating state data and historical operating environment data to obtain a multivariate fusion data sample; use the corresponding fault type and fault range as a label for the multivariate fusion data sample;

[0087] Based on multivariate fusion data samples, the hybrid deep learning model is trained using supervised learning methods.

[0088] These multivariate fusion data samples are divided into a training set and a validation set. The training set is input into the fault detection model, and the output is the predicted fault type or fault range. Supervised learning methods are used to calculate the error between the fault detection model's prediction and the true label. The cross-entropy loss function or mean squared error loss function is typically used to measure the magnitude of the error. The backpropagation algorithm is used to iteratively optimize the parameters of the fault detection model, adjusting the weight parameters of the deep belief network and convolutional neural network so that the hybrid deep learning model gradually approaches the true label. During the training process, the performance of the fault detection model is evaluated using the validation set to prevent overfitting. After multiple iterative training sessions, the fault detection model can learn the patterns and characteristics of historical data, thereby accurately predicting the type and range of distribution network faults.

[0089] In a preferred embodiment, generating the predicted fault type and predicted fault range based on the deep features and the spatial features includes:

[0090] Fusing the deep features and the spatial features to obtain fused features;

[0091] The fusion features are inferred by a classifier to generate the predicted fault type and predicted fault range.

[0092] In this embodiment, the multi-source fusion data is input into the trained fault detection model, and the deep feature representation is first extracted layer by layer through the deep belief network, and the nonlinear transformation ability of the deep belief network is used to capture hidden complex patterns and semantic information, such as the potential correlation between voltage and current or the regularity of temperature changes. The convolutional neural network performs a convolution operation on the input multi-source fusion data to extract spatial feature information and capture the local correlation and spatial distribution characteristics between different features, such as the spatial variation pattern of ambient humidity and wind speed. The deep features extracted by the deep belief network are combined with the spatial features extracted by the convolutional neural network through the feature fusion method. The deep feature vector extracted by the deep belief network and the spatial feature vector extracted by the convolutional neural network can be directly connected according to the corresponding dimensions into a longer feature vector to complete the splicing and fusion to obtain the fused feature.

[0093] The fused features, formed by fusing the deep features extracted by the deep belief network and the spatial features extracted by the convolutional neural network, are input into the classifier. The classifier can be a random forest, support vector machine, or other classification model. Its core task is to predict the fault type and scope based on the input feature data. The multi-source fused features, extracted based on distribution network operating status data and distribution network operating environment data, contain deep semantic information and spatial distribution characteristics of distribution network operations. The classifier analyzes the fused features and, using its internal learning rules and decision-making mechanism, maps the input features to corresponding fault type and scope categories. During inference, the classifier calculates the similarity or probability distribution between the input features and each fault category based on the pattern matching rules learned during the training phase. It ultimately outputs the predicted fault type, such as short circuit, overload, or ground fault, and the predicted fault scope, such as the specific line segment or equipment area, providing the foundation for subsequent precise analysis and location.

[0094] In a preferred embodiment, the distribution network fault location method further includes:

[0095] Generate an optimized fault type based on the predicted fault type and the multi-source fusion data through a random forest algorithm;

[0096] Optimizing the fault occurrence probability distribution model based on the multi-source fusion data to obtain an optimized fault occurrence probability distribution model;

[0097] Inputting the predicted fault range into the optimized fault occurrence probability distribution model to generate an optimized fault range;

[0098] generating the fault location information according to the optimized fault type and the optimized fault range;

[0099] The method for constructing the fault occurrence probability distribution model includes:

[0100] Obtaining distribution network elements and distribution network association information of the distribution network; wherein the distribution network elements include device nodes and their operating status parameters; the distribution network association information is the association relationship between the distribution network elements;

[0101] Constructing a Bayesian network with the distribution network elements as nodes and the distribution network related information as edges;

[0102] According to historical fault data, basic probability values ​​of fault occurrence are assigned to the nodes of the Bayesian network to obtain a fault occurrence probability distribution model.

[0103] It should be noted that the specific process for constructing a fault probability distribution model involves extracting distribution network elements and their relationships based on real-time, multi-source, heterogeneous data, constructing a semantic knowledge graph, defining the nodes and edges of a Bayesian network, and building a fault probability distribution model. First, distribution network elements are extracted from real-time operating status data and operating environment data. These distribution network elements include device nodes such as transformers, switches, and lines, as well as operating status parameters. Furthermore, the relationships between distribution network elements are analyzed, such as the connectivity between devices, the dependency between voltage and current, or the potential link between ambient humidity and fault occurrence. By semanticizing these distribution network elements and their relationships, a knowledge graph centered around devices, states, and environments is constructed. First, the related elements of devices, states, and environments are organized to identify various entities (e.g., different types of devices, various state parameters, and various environmental factors). Next, the relationships between distribution network elements are analyzed, such as the causal relationship between device operating status and environmental factors, and the mutual influence between devices. Finally, these organized entities (e.g., different devices, state parameters, environmental factors, etc.) and their relationships are assigned clear semantics in accordance with the knowledge graph construction specifications. Specifically, for each entity, its category is clearly defined, such as the specific type of a device. The entity's attributes, such as its power and capacity, are also determined. Relationships between entities must also be clearly defined, such as causal, dependency, or parallel relationships, and their directions must be specified. Semanticized entities and relationships are then stored using a triplet model. Through data collection, cleaning, and conversion, the data in the knowledge graph is continuously enriched, ultimately constructing a knowledge graph centered around devices, states, and environments. Nodes are defined in a Bayesian network, each representing a distribution network element, such as a device state or environmental parameter. Edges represent causal relationships or conditional dependencies between nodes, such as the association between device failure and temperature rise. The knowledge graph is then used to deeply mine various semantic information related to distribution network elements, including the functional characteristics of the equipment, the interactions between its operating state and the environment, and other aspects. Historical fault data is meticulously cleaned, classified, and statistically analyzed to extract the characteristic manifestations and correlation patterns of each element under different fault scenarios. Then, based on this information, for each node in the Bayesian network, considering its own attributes and conditional dependencies with other nodes, probabilistic statistical methods and related algorithms are used to assign reasonable probability values ​​to the nodes, thereby constructing a fault probability distribution model.

[0104] In a preferred embodiment, generating an optimized fault type by a random forest algorithm based on the predicted fault type and the multi-source fusion data includes:

[0105] Generate a corresponding one-hot encoding according to the predicted fault type;

[0106] According to the one-hot encoding, a number of decision trees are constructed using a random forest algorithm; wherein each decision tree is split according to the multi-source fusion data;

[0107] According to the decision tree, the optimized fault type is determined through a majority voting mechanism.

[0108] In this embodiment, the predicted fault type is analyzed by the random forest algorithm to output the optimized fault type. Specifically:

[0109] The predicted fault type is converted into a one-hot encoding format and input into the random forest classification algorithm to build multiple decision trees. Each tree is split based on multi-source fusion data, and the optimized fault type is output through a majority voting mechanism.

[0110] It should be noted that the predicted fault type, such as a short circuit, overload, or ground fault, must first be converted into a numerical value using the one-hot encoding rule. Assuming there are three fault types, each fault type is represented as a binary vector of length three, where the index corresponding to the fault type is 1 and the remaining positions are 0. Subsequently, this one-hot encoded fault type data is input into the random forest classification algorithm along with the multi-source fusion data. The random forest algorithm constructs multiple decision trees, each of which splits based on the multi-source fusion data. The optimal split point is selected based on the information gain or Gini index of the features, gradually dividing the predicted fault types into smaller subsets. Each decision tree independently classifies the input predicted fault type and outputs the corresponding fault type prediction. Finally, a majority voting mechanism is used to combine the predictions of all decision trees and select the fault type with the most votes as the optimized fault type output.

[0111] The initial predicted fault type is directly output by the fault detection model constructed by fusing the feature representation capability of DBN with the spatial information processing capability of CNN, which has certain errors. The final optimized fault type is a more reliable result after secondary optimization of the preliminary prediction results by random forest and correction of feature correlation. Both the preliminary predicted fault type and the final optimized fault type are based on real-time multi-source fusion data, and the final optimized fault type takes the preliminary prediction as input, and improves the classification robustness through integrated learning.

[0112] In a preferred embodiment, optimizing the fault occurrence probability distribution model based on the multi-source fusion data to obtain the optimized fault occurrence probability distribution model includes:

[0113] Calculating the similarity between the multi-source fusion data and the historical fault data;

[0114] When the similarity exceeds a preset similarity threshold, the basic probability value of the fault occurrence in the fault occurrence probability distribution model is increased to obtain an optimized fault occurrence probability distribution model.

[0115] It should be noted that the optimization process of the fault probability distribution model specifically involves converting distribution network elements and semantic relationships extracted from the knowledge graph, such as connectivity between devices and correlations between operating status parameters and environmental parameters, into node and edge definitions in a Bayesian network. The semantic information in the knowledge graph provides the Bayesian network with a structured foundation for causal and dependency relationships, such as the causal relationship between device failure and temperature anomalies. Historical fault data is analyzed to calculate the frequency of faults occurring at each node under the combination of distribution network operating status data and distribution network operating environment data, which serves as a reference for base probabilities. Furthermore, by integrating real-time multi-source heterogeneous data and focusing on changes in distribution network operating status data and distribution network operating environment data, the base probabilities are modified and refined based on the degree of match with similar situations in historical fault data. For example, when real-time monitored parameters such as device temperature and load approach historical fault thresholds, the conditional probability of a node fault is correspondingly increased, thereby more accurately characterizing the node fault probability distribution and representing the impact of changes in device status or environmental parameters on fault occurrence. By incorporating semantic information from the knowledge graph into the Bayesian network, the fault probability distribution model's ability to understand complex relationships is enhanced. During the reasoning process, the probability calculation capability of the Bayesian network is combined with the conditional probability distribution and Bayes' theorem to dynamically update the probability distribution of fault occurrence, thereby optimizing the reasoning capability of the fault probability distribution model.

[0116] In a preferred embodiment, inputting the predicted fault range into the optimized fault occurrence probability distribution model to generate the optimized fault range includes:

[0117] Inputting the predicted fault range into the optimized fault occurrence probability distribution model, so that the fault occurrence probability distribution model calculates the conditional probability value of the fault occurrence of each node according to the predicted fault range and the basic probability value of the fault occurrence;

[0118] The optimized fault range is generated based on nodes whose fault occurrence condition probability values ​​exceed a preset probability threshold.

[0119] In this embodiment, it is first necessary to pass the initial predicted fault range, such as the line segment or equipment area where the fault may occur, as input to the optimized fault occurrence probability distribution model. The fault occurrence probability distribution model is based on the probabilistic reasoning ability of the Bayesian network and the semantic information of the knowledge graph, comprehensively considering the operating status data and the operating environment data, and dynamically evaluating the possibility of the fault occurrence. In the reasoning process of the fault occurrence probability distribution model, the probability distribution of the fault range is further adjusted in combination with the semantic rules defined in the knowledge graph, such as the connection relationship between devices, the fault propagation law or the impact of environmental parameters on the device status. By calculating the conditional probability of each area through the Bayesian theorem, the optimized fault occurrence probability distribution model can more accurately identify the high probability area of ​​fault occurrence. Finally, combining the reasoning results of the probability distribution and semantic rules, the specific range of the distribution network fault is output, that is, the optimized fault range.

[0120] It's important to note that the initial predicted fault range is a direct output of the fault detection model, constructed by fusing the feature representation capabilities of DBN with the spatial information processing capabilities of CNN. This model is subject to uncertainty. The final optimized fault range, however, combines the semantic rules of the knowledge graph with the probabilistic reasoning of the Bayesian network, optimizing the accuracy and reliability of the prediction. Both the initial predicted fault range and the final optimized fault range are based on real-time, multi-source heterogeneous data. The optimized fault range uses the predicted fault range as input and is further refined and refined by optimizing the probabilistic reasoning of the Bayesian network through the semantic rules and associations provided by the knowledge graph.

[0121] In one embodiment, generating the fault location information according to the optimized fault type and the optimized fault range includes:

[0122] Based on multi-source fusion data and historical fault data, the distribution network connection structure is sorted out, and through the rule mapping method, it is combined with the optimized fault type and the optimized fault range to form comprehensive distribution network related data;

[0123] Through genetic algorithms, the integrated distribution network correlation data and real-time multi-source heterogeneous data are iteratively optimized to output fault location information.

[0124] It should be noted that information related to the distribution network connection structure, such as the connection relationships between devices and the direction of the lines, is extracted from multi-source fusion data and historical fault data to construct an initial distribution network connection structure model. Based on a pre-defined rule mapping method, the elements in the distribution network connection structure are associated with information related to the distribution network fault type and the distribution network fault occurrence range. For example, the possible fault type and approximate range are determined based on factors such as device type and location. Through a data fusion algorithm, the associated distribution network fault type and distribution network fault occurrence range information is deeply integrated with the original distribution network connection structure model to form comprehensive distribution network association data.

[0125] The expression of fault location information output by genetic algorithm is:

[0126]

[0127] Where x * represents fault location information; & represents decision variables; M represents integrated distribution network related data; D represents multi-source fusion data; F represents fault information, namely, optimized fault type and optimized fault range; P represents fault probability distribution; and R represents the set of all fault locations.

[0128] It should be noted that the parameters for initializing the genetic algorithm include population size, number of iterations, crossover probability, and mutation probability. The integrated distribution network-related data and real-time multi-source heterogeneous data are encoded to construct an initial population, with each individual representing a potential fault location solution. A fitness function is defined to evaluate individual fitness based on metrics such as the degree of match between the integrated distribution network-related data, real-time multi-source heterogeneous data, and the fault location solution, as well as fault location accuracy. During iterations, crossover operations are performed on top-performing individuals based on their fitness values ​​to generate new individuals. Mutation operations are then performed on these new individuals with a certain probability to increase population diversity. The fitness of all individuals is recalculated in each generation, and individuals with high fitness are retained to form a new population. Iterations continue until the preset number of iterations is reached or convergence conditions are met. The individual with the highest fitness in the final population is selected as the optimal solution, which is decoded and output as fault location information.

[0129] The fault detection model of the present invention extracts deep features from the multi-source fused data using a deep belief network and spatial features from the multi-source fused data using a convolutional neural network. Based on these deep and spatial features, it generates predicted fault types and predicted fault ranges, and further generates fault location information. This solves the problem that traditional methods often overlook the deep and spatial characteristics of multi-source heterogeneous data when processing data, thereby improving the accuracy of fault detection.

[0130] like Figure 2 As shown, based on the above method embodiment, a corresponding device embodiment is provided;

[0131] An embodiment of the present invention provides a distribution network fault location device, comprising:

[0132] Data acquisition module, data fusion module and fault location module;

[0133] The data acquisition module is used to acquire first operating status data and first operating environment data of each node in the distribution network in real time; wherein the operating status data includes: voltage data, current data and temperature data; the operating environment data includes: ambient humidity data, ambient wind speed data and ambient gas concentration data;

[0134] The data fusion module performs data fusion on the first operating state data and the first operating environment data to obtain multi-source fusion data;

[0135] The fault location module is used to input the multi-source fusion data into a fault detection model, so that the fault detection model extracts the deep features of the multi-source fusion data through a deep belief network, extracts the spatial features of the multi-source fusion data through a convolutional neural network, and generates a predicted fault type and a predicted fault range based on the deep features and the spatial features.

[0136] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present invention, and can implement the distribution network fault locating method provided by any of the above-mentioned method embodiments of the present invention.

[0137] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. Furthermore, in the drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which may be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement the present invention without inventive effort.

[0138] Based on the above-mentioned method embodiment, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the distribution network fault location method of any embodiment of the present invention.

[0139] For example, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more module elements may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0140] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0141] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0142] Based on the above method embodiments, another embodiment of the present invention provides a computer-readable storage medium, including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the distribution network fault location method described in any one of the above method embodiments of the present invention.

[0143] Wherein, the module / unit integrated in the device / terminal equipment, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0144] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A distribution network fault location method, characterized in that: include: Acquire first operating status data and first operating environment data of each node in the distribution network in real time; wherein the operating status data includes: voltage data, current data and temperature data; the operating environment data includes: ambient humidity data, ambient wind speed data and ambient gas concentration data; Performing data fusion on the first operating state data and the first operating environment data to obtain multi-source fused data; The multi-source fusion data is input into a fault detection model so that the fault detection model extracts deep features of the multi-source fusion data through a deep belief network, extracts spatial features of the multi-source fusion data through a convolutional neural network, and generates predicted fault types and predicted fault ranges based on the deep features and the spatial features.

2. The distribution network fault location method according to claim 1, characterized in that: The step of fusing the first operating state data and the first operating environment data to obtain multi-source fused data includes: Completing the first operating state data and the first operating environment data respectively by a linear interpolation method to obtain second operating state data and second operating environment data; Determine a lower limit based on the first quartile, determine an upper limit based on the second quartile, and remove data outside the lower limit and upper limit ranges of the second operating status data and the second operating environment data, respectively, to obtain third operating status data and third operating environment data; The third operating state data and the third operating environment data are respectively standardized by a Z-Score method to obtain operating state standardized data and operating environment standardized data; By using the principal component analysis method, the principal component features of the operating state data and the principal component features of the operating environment data are extracted from the standardized operating state data and the standardized operating environment data respectively; The principal component features of the operating status data and the principal component features of the operating environment data are fused by the weighted average method to obtain fused data; The fused data is restored to the same dimensional space as the first operating state data and the first operating environment data through inverse transformation processing to obtain the multi-source fused data.

3. The distribution network fault location method according to claim 1, wherein: The generating of the predicted fault type and predicted fault range according to the deep features and the spatial features includes: Fusing the deep features and the spatial features to obtain fused features; The fusion features are inferred by a classifier to generate the predicted fault type and predicted fault range.

4. The distribution network fault location method according to claim 1, wherein: Also includes: Generate an optimized fault type based on the predicted fault type and the multi-source fusion data through a random forest algorithm; Optimizing the fault occurrence probability distribution model based on the multi-source fusion data to obtain an optimized fault occurrence probability distribution model; Inputting the predicted fault range into the optimized fault occurrence probability distribution model to generate an optimized fault range; generating the fault location information according to the optimized fault type and the optimized fault range; The method for constructing the fault occurrence probability distribution model includes: Obtaining distribution network elements and distribution network association information of the distribution network; wherein the distribution network elements include device nodes and their operating status parameters; the distribution network association information is the association relationship between the distribution network elements; Constructing a Bayesian network with the distribution network elements as nodes and the distribution network related information as edges; According to historical fault data, basic probability values ​​of fault occurrence are assigned to the nodes of the Bayesian network to obtain a fault occurrence probability distribution model.

5. The distribution network fault location method according to claim 4, characterized in that: Generating an optimized fault type by using a random forest algorithm based on the predicted fault type and the multi-source fusion data includes: Generate a corresponding one-hot encoding according to the predicted fault type; According to the one-hot encoding, a number of decision trees are constructed using a random forest algorithm; wherein each decision tree is split according to the multi-source fusion data; According to the decision tree, the optimized fault type is determined through a majority voting mechanism.

6. The distribution network fault location method according to claim 4, characterized in that: Optimizing the fault occurrence probability distribution model based on the multi-source fusion data to obtain the optimized fault occurrence probability distribution model includes: Calculating the similarity between the multi-source fusion data and the historical fault data; When the similarity exceeds a preset similarity threshold, the basic probability value of the fault occurrence in the fault occurrence probability distribution model is increased to obtain an optimized fault occurrence probability distribution model.

7. The distribution network fault location method according to claim 4, characterized in that: Inputting the predicted fault range into the optimized fault occurrence probability distribution model to generate the optimized fault range includes: Inputting the predicted fault range into the optimized fault occurrence probability distribution model, so that the fault occurrence probability distribution model calculates the conditional probability value of the fault occurrence of each node according to the predicted fault range and the basic probability value of the fault occurrence; The optimized fault range is generated based on nodes whose fault occurrence condition probability values ​​exceed a preset probability threshold.

8. A distribution network fault location device, characterized in that: include: Data acquisition module, data fusion module and fault location module; The data acquisition module is used to acquire first operating status data and first operating environment data of each node in the distribution network in real time; wherein the operating status data includes: voltage data, current data and temperature data; the operating environment data includes: ambient humidity data, ambient wind speed data and ambient gas concentration data; The data fusion module performs data fusion on the first operating state data and the first operating environment data to obtain multi-source fusion data; The fault location module is used to input the multi-source fusion data into a fault detection model, so that the fault detection model extracts the deep features of the multi-source fusion data through a deep belief network, extracts the spatial features of the multi-source fusion data through a convolutional neural network, and generates a predicted fault type and a predicted fault range based on the deep features and the spatial features.

9. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the method for locating a distribution network fault according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that include: A stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the distribution network fault location method according to any one of claims 1 to 7.

Images