Pin Wang, En Fan, Peng Wang, and Cheng Luo. RPE Algorithm of Recurrent Neural Network and Its Application In Modeling of Nonlinear Dynamic Systems. Dynamic Systems and Applications 30 (2021) No.5, 770-788

https://doi.org/10.46719/dsa20213057

ABSTRACT.
Recurrent neural network (RNN) has very sensitive feedback characteristics, so it is very suitable for the modeling of nonlinear dynamic systems, but the traditional RNN BP algorithm has problems, such as slow convergence speed and complicated training algorithm, which leads to there are great limitations in its real application. To this end, this paper proposes a recursive neural network for RPE algorithm, explores the principle and calculation process of its algorithm, and analyzes its real-time application in nonlinear dynamic system modeling, hoping to explore a fast recursive neural network learning algorithm. In this paper, through a large number of experiments, using mathematical statistical calculation methods, detailed analysis and discussion of the effects of different types of RNNs and different calculation methods on the modeling of nonlinear dynamic systems. At last, the results of research present that the dynamic recursive neural network takes less time to adjust the parameters, and the error value is only between 0.02-0.036, so that it is more suitable for nonlinear dynamic system modeling; compared with BP algorithm, traditional parameter estimation method-RPE algorithm, combined with RNN for nonlinear to establish the modeling of the dynamic system has optimized its effect, not only the convergence speed is fast, but also the maximum error from the expected output is only 0.01, and the accuracy of the model after construction is extremely high.

KEYWORDS: Recurrent Neural Network, RPE Learning Algorithm, Nonlinear Dynamic System, System Modeling