RBM-Based Back Propagation Neural Network with BSASA Optimization for Time Series Forecasting
Huiyuan Li, Lina Pan, Mei Chen, Xiaoyun Chen, Yufan Zhang · 2017
Time series forecasting plays a vital role in stable developments of economy and industry, the precaution of natural hazard, even adjustment of energy policy. Back Propagation Neural Network (BPNN) is always viewed as a predictor which powers in dealing with nonlinear problems associated with time series forecasting. However, BPNN easily falls into local minimum. To overcome this disadvantage, a novel hybrid approach called RBM-BSASA-BP is presented. Restricted Boltzmann Machine (RBM) firstly provides rough parameters for BPNN via pre-training process. The improved Backtracking Search Algorithm with Simulated Annealing (BSASA) is then employed to fine-tune the parameters for the purpose of seeking out the optimal weights and biases. Finally, optimized BPNN is utilized to accomplish forecasting tasks. Verified by artificial time series and real-world time series, the proposed hybrid approach defeats the state-of-the-art methods and it is proved to be a promising forecasting algorithm.