Research On Anomaly Detection And Early Warning Of Generator Based On Deep Learning

In recent years,with the development of China’s industrialization process,the demand for energy is increasing and the problem of pollution is becoming more and more serious.As a clean energy,wind power and water power will become an important part of the energy structure in the future.Currently,the technology of anomaly detection in generator has not been a major breakthrough,by wind and water power equipment as a breakthrough point,this paper devotes to do research on the equipment health status of the generator,and finds the abnormal state of the equipment as soon as possible in order to facilitate staff maintenance and reduce the risk of equipment damage.The following research works are carried out in this paper:Firstly,research on anomaly detection based on the Isolation Forest and PCA process detection method.After analyzing the causes of abnormal data and the correlation of each dimension,the third chapter uses Isolation Forest based on ensemble learning method and PCA process detection method for abnormal data detection and early warning research,and discusses the deficiency of anomaly detection ability of the Isolation Forest method and poor modeling stability of the PCA process detection method.Then,an improved method of attention-iForest was proposed to solve the shortage of Isolation Forest.In the fourth chapter,the attention module is added to increase the attention of the model to the abnormal vector,and the sliding window method is added to solve the insensitivity of the model to local abnormal points.The improvement of Isolated Forest can effectively solve the problem of its insufficient ability of anomaly detection and improve its applicability in multi-dimensional data.Then,an improved network structure PAR-LSTM is proposed for anomaly detection.The fifth chapter of this paper uses LSTM network to conduct training model in normal data,constructs the normal state model and applies it to anomaly detection.Long Short-Term Memory(LSTM)is a neural network for processing sequence data.Compared with general neural networks,it can extract the context of data effectively.As a special Recurrent Neural Network(RNN),the structure of input gate,forget gate and output gate can solve the problem of gradient disappearance in optimization of RNN network.At the same time,in view of the poor performance of LSTM in some scenarios,this paper proposes an improved network structure PAR-LSTM as an anomaly detection model.Its parallel network structure solves the problem that LSTM can not learn the "normal state" of the equipment,which leads to the decline of detection ability in some scenarios.Finally,in order to verify the effectiveness of the above method,this paper uses the real data set of wind motor and water motor to test.By comparing the similar algorithms,the effectiveness and applicability of the improved method in this paper are proved.