Time Series Prediction Based on T-S Fuzzy Deep Network

Time series are widely existed in various fields such as industry, agriculture and finance, and time series prediction has been a classical problem for research. In order to improve prediction accuracy and model interpretability, this paper proposes a time series prediction model based on T-S fuzzy deep network. The fuzzy deep network model is based on the principle of stack generalization and is composed of several simple fuzzy model in series by referring to the layer connection structure of neural network. Each fuzzy subsystem is, each model contains multiple fuzzy if-then rules, and each fuzzy rule describes a subsystem. The proposed T-S fuzzy model compares five prediction models such as support vector machine regression, random forest regression, ridge regression, linear regression and T-S regression using open data sets. The results show that the prediction performance based on fuzzy deep network model is better in multiple data sets and has stronger prediction stability.