Through the analysis of the principle of refrigeration and air conditioning system, the paper establishes the simulation model of each module, and then combines into the general model of the refrigeration system, analysis of the fault categories and the causes of these failures, then finds out the characteristic parameters, and builds the BP neural network model with particle swarm optimization according to a large number of historical data in the database. At last the results show that the combined use of swarm optimization algorithm and BP neutral network can achieve better fault detection efficiency than simply use the BP neutral network.