Energy Efficiency Prediction of Screw Chillers on BP Neural Network Optimized by Improved Genetic Algorithm

Abstract

To improve the accuracy of energy efficiency prediction of screw chillers, the optimized back propagation (BP) neural network through improved genetic algorithm(GA) was proposed. In this study, a modified genetic algorithm (MGA) based on a novel selection strategy was presented to optimize back propagation (BP) neural network (MGA-BP). Each individual in selection process was offered with a fitness value. The individuals whose fitness values are greater than the average fitness of people are retained to next generation. Additionally, an improved genetic algorithm (iGA) based on an optimal selection strategy is put forward to optimize back propagation (BP) neural network (iGA-BP). Each generation population is sorted by small to large according to the fitness degree, and individuals of the even bits are reserved to the next generation. Individuals are uniformly divided into two sections, namely, the front and tail sections. The larger part of the whole was retained. 300 sets of the acquired historical data from a screw chillers were used to train the network and 50 groups were selected to verify the model. The comparison of MGA-BP and iGA-BP with GA-BP whose selection operator is roulette- wheel selection operator, the results showed that the convergence rate of MGA-BP and iGA-BP increased by 27.03% and 43.24% respectively. Furthermore, the maximum relative errors predicted by MGA-BP and iGA-BP were 5.99%, 3.45% respectively, and the average relative error was within 2.0%. Therefore, the proposed prediction method can improve availably the accuracy of the energy efficiency prediction model of screw chillers.