Long Short-term Memory modeling method with monotonicity analysis as constraints base on Spearman coefficient

Abstract

This paper proposes a new method of monotonicity, which is used to solve the overfitting problem of the Long-Short-Term Memory (LSTM) model. The main contribution of this paper is applying the monotonicity as priori knowledge to the modeling process. This study uses scatter plots to describe bivariate variables and the Spearman coefficient to extract the monotonicity of data. To exclude most noise point, the scatter diagram is filtered by a binary 0–1 liner program. Base on the monotonicity of data have known, an optimization problem with constraint is proposed to obtain the LSTM neural network model. An experiment of ethylene cracking show that the proposed method can achieve a good predicting performance and less overfitting effects.