A Time Series Classification Method Based on 1DCNN-FNN
Zihao Zhao, Jie Geng, Wen Jiang · 2021
With the rise of deep learning technology, the use of one-dimensional convolutional neural network (1DCNN) to process time series has the advantages of higher classification accuracy and stronger generalization ability. However, the 1DCNN constructs a classification model by identifying the feature vector of the data distribution, which lacks the reasoning ability on digital features. Because Fuzzy Neural Network (FNN) combines fuzzy inference with neural network and has stronger ability of fuzzy information inference, this paper proposes a hybrid classification model combining 1DCNN and FNN. The hybrid model uses 1DCNN and FNN models to process two kinds of feature information separately and effectively merge them on the fully connected layer. In this paper, WISDM data set is used to train and test the proposed 1DCNN-FNN hybrid classification model, and the results are compared with the results of the 1DCNN model. Experimental results show that the proposed method has better classification effect.