An improved time delay neural network model for predicting dynamic heat and mass transfer characteristics of a packed liquid desiccant dehumidifier

Abstract

Researches on packed liquid desiccant dehumidifiers are abundant. However, only a few works concentrate on modeling the dynamic dehumidification process as it requires detailed heat and mass transfer coefficients along the dehumidifier, which are impacted by many factors such as flow pattern of liquid desiccant and hard to be obtained, hindering the development of dynamic models. Data driven modeling method could eliminate the requirement for detailed heat and mass transfer coefficient along the dehumidifier. A time delay neural network (TDNN) model is a general way to describe the dynamic process without the requirement of much knowledge on physical mechanism, yet the number of time steps for each inlet parameter is hard to be determined. In this paper, an improved time delay neural network (iTDNN) model was proposed. Through analyzing characteristics of dehumidification process, model inputs were simplified, which is the core of improvement of TDNN model. Besides, a new combined algorithm including back propagation (BP) algorithm and improved genetic algorithm (GA) was proposed to acquire the optimal weight values globally. The predicted results agree well with the experimental data for validation part. The method presented in the paper provides a reference for modeling multi-variables and large delay nonlinear dynamic systems.