Research And Implementation Of Deep Learning-based Fault Prediction Method For Power Grid Equipment

With the increasing scale and complexity of power grids,the frequency of power grid equipment failures and accidents is also increasing,which brings great challenges to the safe and stable operation of power grids.Therefore,it is important to carry out research on grid equipment fault prediction to improve the reliability of grid equipment,ensure the safe and stable operation of the grid,and reduce maintenance costs.In recent years,with the continuous development of information technology and artificial intelligence,big data,deep learning and other methods for grid equipment fault prediction have become hot spots and frontiers of research.However,low quality raw data and lack of data correlation analysis lead to insufficient accuracy of grid equipment fault prediction,and in this paper,traditional fault prediction models judge faults based on thresholds and ignore spatial relationships in time series,which further affect the accuracy of prediction.To address the above problems,the research work in this paper is conducted as follows.Firstly,missing values and noisy data of long time series data generated by power grid equipment are analyzed,and a data pre-processing method for long time series data of power grid equipment is proposed,which includes a missing value interpolation algorithm based on generating adversarial imputation network and a noisy data processing algorithm based on fast Fourier transform.The data quality is improved and the foundation for the subsequent prediction algorithm is laid by this method.Secondly,a hybrid neural network model based on a self-attentive mechanism and a spatio-temporal graph convolutional network is proposed.The attention of the model to the key data of the long time series data is improved through the selfattention mechanism,while the temporal and spatial relationships of the long time series are comprehensively considered through the spatio-temporal graph convolutional network,and the correlation analysis of the data from different systems is realized through the full-connected fusion module.Furthermore,the fault prediction accuracy is further improved by predicting the remaining available life of equipment.Finally,a prototype system for grid equipment fault prediction is designed and implemented,deploying and validating the proposed algorithm or model.The results of data pre-processing and model prediction can be visualized through data visualization in the system,and the reliability and practicality of the hybrid neural network model proposed in this paper are fully verified.In summary,the research results of this paper provide a new idea and method for grid equipment fault prediction,which has high practical application value.