Control of a simulated dual-temperature hydronic system using a neural network approach

This paper discusses the application of a neural network approach toward operation of a dual-temperature hydronic system that the authors initially studied in a course project. The hydronic system uses water as the working medium to provide heating and cooling simultaneously to a process plant. The operation consists of temperature setup and control, and it is accomplished by adjusting the 15 valves on the line. A neural network-based expert system was developed to simulate such an operation. It consists mainly of two subsystems, one for temperature setup, the other for temperature control. Each subsystem is composed of a front end and a neural network base. The neural network was trained with thermal demands (heating and cooling temperature in Fahrenheit) as inputs and valve adjustment (percentage of each valve's opening) as outputs. The training facts were given by thermodynamic considerations. The function of the front end is to communicate with the neural network base, so that the inputs can be sent to it and the outputs can be taken from it. The control operator is also prompted with instructions on how to adjust the valves from the front end.