A prediction method based on simulation is proposed to reduce the difficulty and the large error in measuring the power of variable-speed compressor. The threshold and weight of a back propagation (BP) neural network were initialized by particle swarm optimization to measure the power of the variable-speed compressor. In this study

a total of three kinds of simulation models were established for comparison

i.e.

a BP neural network model

a genetic algorithm (GA)-BP neural network model

and a particle swarm optimization (PSO)-BP neural network model. Then

the power of variable-speed compressor was predicted through the interpolation of three models

as well as the extrapolation of evaporation temperature and condensation temperature. The predicted results and the average relative fitting degree error were compared and analyzed. The results showed that the BP neural network model based on the particle swarm algorithm optimization was superior to the other two models. For the extrapolation tests of condensation temperature

in particular

the relative error of BP neural network model was reduced by 1.11% and 2.64%

respectively

compared with the other two neural networks. For the three methods

the average relative error was within 1% and the fitting degree was above 0.9

indicating that the BP neural network model based on the particle swarm algorithm optimization can adequately obtain the power of variable-speed compressor and has a strong generalization ability.