遗传算法与修正的自适应矩估计优化循环神经网络的心音分类方法OA北大核心CSTPCD

Aiming at the problems of gradient explosion,gradient disappearance,and short-term memory in traditional recurrent neural networks(RNN)for identifying and classifying heart sound signals,a heart sound classification model is proposed combining genetic algorithm(GA)and rectified Adam(RAdam)optimized RNN without heart sound segmentation.The advantage of this model is that it integrates GA and RAdam optimizer in series into a RNN to improve its performance.Firstly,the selection,mutation and genetic operation of the GA are used to optimize the number of nodes in the input layer of the RNN,and the initial solution of the optimal individual of the heart sound feature vector is obtained.Secondly,according to the weight and bias matrix in the optimal individual,the initial weight and threshold are assigned to the model,and the optimal solution of the initial weight is obtained,and the entire model shares parameters.Finally,combined with the improved learning rate adaptive optimization algorithm,the RNN model is optimized.The results show that combining the classical Mel-frequency cepstral coefficient method to extract the eigenvectors of the heart sound signal,the classification accuracy of the heart sound signals reaches 90.29％,which is 17.79 percentage points higher than that of the unoptimized RNN model.