Research On Device Anomaly Detection Based On GA-CNN-LSTM Model

In industrial production,the stable and efficient operation of equipment is critical for improving manufacturing efficiency and save manufacturing costs.However,losses and even threats to employee safety due to equipment failures occur from time to time.Therefore,it is necessary to detect abnormalities in the operating status of equipment.The development of hardware technology makes sensor data collection more and more often used in the field of equipment anomaly detection,but the data collected by sensors are usually high-dimensional and closely related to each other,and the data is noisy.Although deep learning algorithms have been widely used in the field of equipment anomaly detection in recent years and have strong feature extraction capability.However,it still faces problems such as difficulty in defining network parameters,and the development of equipment anomaly detection still faces a series of difficulties.In this regard,this paper applies principal component analysis,deep learning,and genetic algorithm to device anomaly detection to improve fault detection accuracy and speed,and the main work of this paper is as follows.（1）A PCA-LSTM based device anomaly detection model is proposed to solve the problems of high dimensionality of device sensor data and difficulty in feature extraction.Firstly,the PCA algorithm is used to reduce the dimensionality of the device sensor data and replace the highdimensional variables with low-dimensional variables by linear combination to reduce the complexity of computation during training,and then the LSTM network is used to obtain the abnormal features of the reduced-dimensional sensor data and detect the device abnormalities.（2）A GA-CNN-LSTM based device anomaly detection model is proposed to improve the training accuracy.The model first uses 1DCNN to extract the overall features of the device sensor data,where the convolutional and pooling layer are used to extract data features and dimensionality reduction,and the output of the last pooling layer is processed by the Flatten layer as the input of the next LSTM layer,and the device sensor data is extracted through the LSTM The temporal features of sensor data are extracted through LSTM,and the nonlinear relationship between the data state and the anomaly is learned,and finally the anomaly of the device is output after the fully connected layer and softmax regression processing.To further optimize the algorithm and enhance the detection performance,the number of LSTM layers and the number of neurons,and the number of fully connected layers and the number of neurons in the model are therefore determined using a genetic algorithm.and finally the optimal parameters of the GACNN-SLTM model are searched.The comparative experimental verification shows that the model achieves 98% fault detection rate and 97% accuracy rate with the best results.（3）To address the problem that the traditional genetic algorithm has low search efficiency and easily falls into local optimum,this paper improves the genetic algorithm in the GA-CNNLSTM-based equipment anomaly detection model,adjusts the crossover rate and variation rate in the genetic algorithm from fixed values to change with the degree of adaptation,and improves the encoding method and the range of values of the genetic algorithm in the model according to the optimization goal,reduces the probability of local.（4）Finally,based on the equipment anomaly detection method proposed in this paper,a WEB equipment anomaly detection system is designed and realization to achieve the purpose of equipment abnormality detection.