A pattern classification neural network suitable for machine control

We propose a pattern classification neural network (CCNN), which is suitable for incremental learning, has simple learning algorithm, and uses less memory. Then we show an application of CCNN to adaptive learning control in an air conditioner. We confirm that the predictive control with this model allows the machine to learn and reproduce user's preference. We also propose an enhanced model CCNN2, which adjusts the position of the reference vectors. Finally, we give an experimental comparison between two adjacent categories which were normally distributed in the two-dimensional space, and demonstrate that CCNN and CCNN2 are more effective than conventional models in pattern classification.