吸收式制冷机组逆神经网络设定优化
A Set Point Optimization Method for Absorption Chiller Based on Inverse Neural Network

针对吸收式制冷机组非线性、难以控制的特点,提出了一种基于逆神经网络模型的设定点优化方案。首先,以11.5 kW单效溴化锂吸收式制冷机组为对象,使用人工神经网络方法建立了机组模型,通过对溴冷机制冷原理的分析,建立了系统结构为5？？6？？2的网络模型,该神经网络模型的相关系数大于0.99且方均根误差小于0.2％,与实验数据取得了良好的拟合效果;然后,利用该模型对溴冷机的各个输入参数进行灵敏度分析,并据此选择热水供水温度与冷却水流量作为优化方法的控制输入参数;最后,以冷冻水输出温度作为系统控制输出,对其进行优化计算,并采用改进的粒子群优化算法与逆神经网络相结合的方法,计算制冷机组的最优控制输入参数。通过实验与仿真分析,可知该算法的计算时间在30 s以内,低于吸收式制冷机组的稳定时间;溴冷机的目标输出与仿真计算结果间的误差小于0？？02％,表明该方案可以应用于吸收式制冷机组的在线控制。
In view of the nonlinearity and difficulty in control of absorption chiller, a set point optimization method based on inverse neural network model is proposed. Firstly, taking an 11.5 kW single？？effect lithium bromide absorption chiller as the research object, an artificial neural network method is used to establish a model of the unit. Through the analysis of the chiller, a network model with a 5？？6？？2 structure is established. The correlation coefficient of the neural network model is more than 0.99 and the root mean square error is less than 0.2％, so the experimental data are well fitted. Subsequently, the sensitivity analysis of each input parameter of the chiller is conducted to select the hot water supply temperature and the cooling water flow rate as the control input parameters to be estimated. Finally, as the control output of the system, the chilled water output temperature is optimized. The optimal control input parameters of the chilling system are estimated by combination of an improved particle swarm optimization and the inverse neural network algorithm. Through the analysis of experiment and simulation, the calculation time of this method is within 30 s, which is shorter than the stable time of absorption chiller. Moreover, the error between target output and simulation calculation is less than 0？？02％. These results show that the proposed scheme is suitable for online control of absorption chiller