Heat pump emulation using neural networks

A strategy for the control of variable-capacity heat pumps using neural networks is developed. The strategy is based on the emulation of the heat pump using a backpropagation (BP) network. The emulator, which has the same structure as the controller, may then be used for the training of the controller. The controller may then be applied to real-time control. Several emulators have been developed and tested. The emulators are either global or correspond to the various modules of the heat pump. For both cases a one-step predictor is developed and compared with full emulation. The emulators are trained using the BP algorithm with the data obtained on the heat pump. The outputs of the emulating neural network are compared with those of the heat pump. The behavior of the algorithm is presented.<>