Abstract
A novel approach to automatic climate control employing neural networks is proposed. The Neural Network based ACC (NNACC) learns the desirable input/output of ACC from training samples and realizes flexible ACC control in various environmental conditions. NNs have a serious drawback, as being black boxes. We introduce a NN verification method using interval arithmetic to guarantee the I/O of the trained NN is correct, making the NNs industrially applicable. A typical effect of the NNACC is the increased blower speed in mild ambient conditions, which is difficult by the conventional linear control. The feeling test reveals NNACC significantly improves amenity score from 3.9 to 5.2, where 7 and 1 stand for very comfortable and very uncomfortable, respectively. Shorter development period is also achieved by the flexible control provided by the NNs.