Thermal comfort sensor based on probabilistic energy neural network
Toshikazu Takemori, N. Miyasaka, S. Hirose · 1990
A description is given of a new type of neural network for pattern recognition, the probabilistic energy neural network (PENN), and of the thermal comfort sensor (TCS) (P.O. Fanger, 1970) using PENN. PENN is based on Bayes' rule, and the learning mechanism is motivated by such conventional neural networks as restricted coulomb energy (RICE). PENN is a supervised three-layered feedforward network. It can be regarded as a network that outputs a posteriori probability after learning a priori probability and state conditional probability density distribution. The special features of PENN are real-time learning capability, pattern classification ability on nonlinearly separable data, and probabilistic nature of the decision rule. The TCS developed is a computer simulation system