基于模型和神经网络的绝热毛细管快速仿真方法

为简化毛细管模型，采用分相集中参数方法，将平均比容的权重因子作为两相区简化的特征参数，用人工神经网络方法建立特征参数与其影响参数之间的非线性映射。神经网络的学习样本采用工质Ｒ１２，检验样本包括Ｒ１２、Ｒ２２、Ｒ１３４ａ和Ｒ６００ａ等多种工质。在常用制冷空调工况范围内，该简化模型与分布参数模型相比，平均偏差０３％，计算速度提高１个数量级。

FAST SIMULATION METHOD FOR ADIABATIC CAPILLARY TUBES BASED ON MODEL AND ARTIFICIAL NEURAL NETWORK

A simple model of adiabatic capillary tubes is required for fast simulation of refrigeration and air-conditioning appliances, A lumped-parameter model simplified from a general distributed-parameter model and combined with neural network is established in this paper. The weighting factor of average specific volume in two-phase region is presented as a characteristic variable, and its correlative factors are transformed into dimensionless form. The nonlinear mapping between the characteristic variable and the correlative factors is set up with a forward neural network. The learning samples are generated with the working fluid R12, and the checking samples are done with R12, R22, R134a and R600a. In the common range of refrigeration and air-conditioning working conditions, the average deviation between the new model and the distributed-parameter model falls into 0.3%, while the maximum deviations are not greater than 5% except R600a as the working fluid. The computation speed of the new model is m6re than one order of magnitude higher than the distributed-parameter one. Therefore, the new simple model is of good precision and generalization.