A Decimal Scaling Normalization Encoder-Decoder Deep Network for Modeling in the Nonlinear Systems

Deep learning predictive networks are extensively employed for modeling and predicting nonlinear systems due to their superior nonlinear modeling capabilities. They are able to capture the hidden characteristics given the presence of noisy and complex nonlinearities in data collected by sensors, and provide satisfactory performance such as modeling accuracy. In this paper, we have proposed the so-called Decimal Scaling Normalization Encoder-Decoder Deep Network (DSED) to deal with the modeling problem for certain complex nonlinear systems. Our approach relies on a codec structure, and the data undergoes decimal scaling normalization, significantly enhancing modeling accuracy. Experiments involving the modeling of tank vibration reaction forces have been conducted to verify the effectiveness of the proposed method, and the results confirm the model's strong modeling capability.