Hydrological Time Series Prediction Model Based on Deep Belief Network

Abstract

Continuous hydrological time series have the characteristics of randomness and mutagenicity, which reduce the accuracy of its prediction model. In this paper, a hydrological time series-deep belief network (HTS-DBN) based on the continuous restricted Boltzmann machine is proposed. The continuous Boltzmann machine can process the continuous hydrological time series better than the traditional Boltzmann machine. In addition, the number of input layer nodes and the optimal structure of the network in HTS-DBN are determined by similarity correlation method. At the same time, the LM algorithm is used to optimize the HTS-DBN model. Experiments show that HTS-DBN has better accuracy in predicting hydrological time series.