Prediction Model of Thermal Comfort Based on Tabu Genetic Neural Network

Abstract

The calculation process of thermal comfort index has the characteristics of nonlinearity and high computational complexity, so that the real-time controller of air conditioning can’t be used directly. To solve this problem, based on the thermal comfort equation provided by professor Fanger, an improved tabu genetic neural network (TGA-BPNN) is proposed to generate a prediction model for PMV index. The improvements include: training the initial population using train( ) function and using tabu tables to optimize selection, crossover, and mutation. The simulation experiment shows that, compared with the traditional BP network and the genetic neural network, TGA-BPNN can quickly find the global optimal solution and make the prediction model more accurate under the condition of maintaining the diversity of the population. Of course, there are still deficiencies in the simulation experiment. The deficiencies include that initial population needs to be trained repeatedly, which increases algorithm’s running time compared to random population. The algorithm spatial complexity is increased, so that the efficiency of TGA-BPNN algorithm is not as high as that of GA-BPNN algorithm.