CN-122020564-A - High-voltage cable equipment state prediction method and device based on deep learning, electronic equipment and storage medium

CN122020564ACN 122020564 ACN122020564 ACN 122020564ACN-122020564-A

Abstract

The invention relates to the technical field of power equipment state monitoring, and provides a high-voltage cable equipment state prediction method and device based on deep learning, electronic equipment and a storage medium. The method comprises the steps of obtaining historical monitoring data of high-voltage cable equipment, reconstructing the historical monitoring data to obtain a training data set, carrying out fusion processing on all features in the training data set to obtain first fusion features, carrying out enhancement processing on the first fusion features to obtain second fusion features based on local correlation and distribution discrete degree among the features, training an equipment state prediction model based on the second fusion features to obtain a target prediction model, and taking real-time monitoring data of the high-voltage cable equipment as input of the target prediction model to obtain a state prediction result output by the target prediction model. The embodiment of the invention can improve the accuracy and reliability of the state prediction result of the high-voltage cable equipment.

Inventors

GUO ZHIGANG
ZHU YIMENG
YAN FENG
LIU PENGYUE
QU TONG
YANG LINQING
ZHANG KAI
Liu Niming
JIA ZHANHAO
Chu Ziping

Assignees

国网陕西省电力有限公司西安供电公司

Dates

Publication Date: 20260512
Application Date: 20260226

Claims (10)

1. The high-voltage cable equipment state prediction method based on deep learning is characterized by comprising the following steps of: Acquiring historical monitoring data of high-voltage cable equipment, and reconstructing the historical monitoring data to obtain a training data set; performing fusion processing on each feature in the training data set to obtain a first fusion feature; Based on the local correlation and the distribution discrete degree among the features, enhancing the first fusion feature to obtain a second fusion feature; training the equipment state prediction model based on the second fusion characteristic to obtain a target prediction model; And taking the real-time monitoring data of the high-voltage cable equipment as the input of the target prediction model to obtain a state prediction result output by the target prediction model.
2. The method of claim 1, wherein the acquiring historical monitoring data of the high voltage cable device and reconstructing the historical monitoring data to obtain the training data set comprises: classifying the historical monitoring data to obtain each historical fault sample and each historical normal sample; extracting a plurality of key feature dimensions corresponding to each historical fault sample based on the operation condition records corresponding to each historical fault sample; Taking a plurality of key feature dimensions corresponding to each historical fault sample as a sample point, and screening each sample point based on Euclidean distance between each sample point to obtain a plurality of neighborhood sample points; randomly interpolating connecting lines between each sample point and each neighborhood sample point to generate a synthetic fault sample; Taking the synthetic fault samples and each historical fault sample as input of a preset condition generation type countermeasure network, and reconstructing the synthetic fault samples by taking a multi-factor coupling condition as constraint to obtain an enhanced fault sample; Mixing the enhanced fault samples with each historical normal sample according to a preset proportion, and carrying out equalization adjustment on the mixed sample distribution to obtain the training data set.
3. The method of claim 1, wherein the fusing each feature in the training dataset to obtain a first fused feature comprises: Taking a sample in the training data set as input of a preset neural network algorithm, and extracting multi-scale deep features of the sample layer by layer through forward propagation calculation to obtain a multi-dimensional first feature vector; Calculating importance scores of all feature channels channel by channel through a channel attention mechanism, wherein the channel attention mechanism comprises at least two fully-connected layers and a nonlinear activation function positioned between the two fully-connected layers, the original importance scores corresponding to all feature channels are obtained after the multi-dimensional first feature vectors are mapped through the two fully-connected layers, and normalization processing is carried out on all the original importance scores through the nonlinear activation function to obtain attention weight vectors; Multiplying the attention weight vector and the multidimensional first feature vector channel by channel to obtain a weighted feature vector; and carrying out channel dimension integration on the weighted feature vectors, and carrying out normalized fusion on the integrated features through a Softmax function to obtain the first fusion features.
4. The method of claim 1, wherein the enhancing the first fused feature based on the local correlation and the distribution dispersion degree between the features to obtain a second fused feature comprises: combining the feature vectors in the first fusion feature to obtain a plurality of feature vector groups; For each feature vector group, mapping three feature vectors in the feature vector group into coordinate points in a two-dimensional space respectively, and determining local correlation strength of the feature vector group based on triangle areas formed by the three coordinate points; Based on the local correlation strength of each feature vector group, recombining each feature vector to obtain local enhancement features; The second fusion feature is generated based on the local enhancement feature and the first fusion feature.
5. The method of claim 4, wherein the generating the second fused feature based on the local enhancement feature and the first fused feature comprises: Mapping each feature vector in the local enhancement features into feature points in a multidimensional space to obtain a feature point set; performing concave point elimination on the characteristic point set based on a Graham scanning method to generate a characteristic distribution concave packet; Generating global distribution description features based on the area values of the feature distribution recesses and the density distribution information of each feature point in the feature point set; and updating the first fusion feature based on the global distribution description feature and the local enhancement feature to obtain the second fusion feature.
6. The method of claim 5, wherein updating the first fused feature based on the global distribution description feature and the local enhancement feature results in the second fused feature, comprising: splicing the local enhancement features and the global distribution description features to obtain a combined feature vector; Carrying out standardization processing on the combined feature vector, and selecting a plurality of feature vector pairs from the standardized combined feature vector; Obtaining each cross product vector based on the cross product of the vectors between the feature pairs; Sequentially combining the cross product vectors to obtain an interaction matrix; Performing matrix multiplication operation on the normalized interaction matrix and the normalized combined feature vector to obtain a weighted and adjusted feature; Performing nonlinear transformation on the weighted and adjusted characteristics to obtain nonlinear transformed characteristics; and carrying out residual connection on the nonlinear transformed feature and the first fusion feature to obtain the second fusion feature.
7. The method of claim 1, wherein training the device state prediction model based on the second fusion feature results in a target prediction model, comprising: forward propagation calculation is carried out on the second fusion feature through the equipment state prediction model, so that a preliminary state prediction result is obtained; calculating a prediction error between the preliminary state prediction result and the real label through a cross entropy loss function; Calculating gradients of the prediction errors on all network parameters in the equipment state prediction model through a back propagation algorithm, and updating weight parameters of the equipment state prediction model by adopting an adaptive moment estimation algorithm; and repeatedly executing the forward propagation, loss calculation, backward propagation and weight parameter updating processes until the prediction error converges or reaches a preset training round to obtain the target prediction result.
8. High-voltage cable equipment state prediction device based on deep learning, characterized by comprising: The data reconstruction module is used for acquiring historical monitoring data of the high-voltage cable equipment and reconstructing the historical monitoring data to obtain a training data set; The fusion processing module is used for carrying out fusion processing on each feature in the training data set to obtain a first fusion feature; the enhancement processing module is used for enhancing the first fusion feature based on the local correlation and the distribution discrete degree among the features to obtain a second fusion feature; the model training module is used for training the equipment state prediction model based on the second fusion characteristic to obtain a target prediction model; And the state prediction module is used for taking the real-time monitoring data of the high-voltage cable equipment as the input of the target prediction model to obtain a state prediction result output by the target prediction model.
9. An electronic device, comprising: And a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.
10. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the method of any one of claims 1-7.

Description

High-voltage cable equipment state prediction method and device based on deep learning, electronic equipment and storage medium Technical Field The invention relates to the technical field of power equipment state monitoring, in particular to a high-voltage cable equipment state prediction method and device based on deep learning, electronic equipment and a storage medium. Background At present, the state monitoring and operation and maintenance of a high-voltage cable mainly depend on two modes, namely periodic inspection and preventive test, the mode needs to interrupt power supply and has fixed detection period, hidden defects of dynamic change in the operation process of the cable are difficult to sense in real time, and inspection results are unreliable. Secondly, an intelligent monitoring method developed by depending on the Internet of things (Internet of Things, ioT) and big data technology is used, however, when the existing intelligent monitoring method is applied to state prediction of the old urban high-voltage cable, the problem that the cable fault belongs to a small probability event, and the special acidic soil corrosion, high-temperature operation and short-term overload multifactor coupling faults of the old urban cable are caused, and because the occurrence condition is complex, the monitoring coverage difficulty is high, effective fault samples which can be used for model training are few, the serious class imbalance problem exists in a training dataset, and the model prediction result is inaccurate is caused. Disclosure of Invention The invention provides a state prediction method and device for high-voltage cable equipment based on deep learning, electronic equipment and a storage medium, and can solve at least one technical problem. In a first aspect, an embodiment of the present invention provides a method for predicting a state of a high-voltage cable device based on deep learning, including: Acquiring historical monitoring data of high-voltage cable equipment, and reconstructing the historical monitoring data to obtain a training data set; performing fusion processing on each feature in the training data set to obtain a first fusion feature; Based on the local correlation and the distribution discrete degree among the features, enhancing the first fusion feature to obtain a second fusion feature; training the equipment state prediction model based on the second fusion characteristic to obtain a target prediction model; And taking the real-time monitoring data of the high-voltage cable equipment as the input of the target prediction model to obtain a state prediction result output by the target prediction model. In a second aspect, an embodiment of the present invention provides a high-voltage cable device state prediction apparatus based on deep learning, including: The data reconstruction module is used for acquiring historical monitoring data of the high-voltage cable equipment and reconstructing the historical monitoring data to obtain a training data set; The fusion processing module is used for carrying out fusion processing on each feature in the training data set to obtain a first fusion feature; the enhancement processing module is used for enhancing the first fusion feature based on the local correlation and the distribution discrete degree among the features to obtain a second fusion feature; the model training module is used for training the equipment state prediction model based on the second fusion characteristic to obtain a target prediction model; And the state prediction module is used for taking the real-time monitoring data of the high-voltage cable equipment as the input of the target prediction model to obtain a state prediction result output by the target prediction model. In a third aspect, an embodiment of the present invention further provides an electronic device, including at least one processor, and a memory communicatively coupled to the at least one processor, where the memory stores instructions executable by the at least one processor, the instructions being executable by the at least one processor to enable the at least one processor to perform the method according to any one of the embodiments of the present invention. In a fourth aspect, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method of any one of the embodiments of the present invention. The method and the device realize high-quality construction of the training data set by systematically reconstructing and fusing historical monitoring data of the high-voltage cable device, wherein each feature forms a first fused feature through preliminary fusion, multidimensional operation information can be comprehensively expressed in a unified feature space, relevance information among different features is enhanced, then, the first fused feature is selectively enhanced and recombin