The work aims to design a temperature setting model based on the improved BP multi input multi output (IBPMIMO) neural network to achieve the accurate control of beer sterilization. The temperature setting values of the spray nozzle in the main temperature zone of the beer sterilization machine were predicted by IBMIMO, and the predicted values were analyzed accurately by the PU control software in the beer sterilization machine. The maximum comprehensive error of the spray nozzle temperature predicted by (IBPMIMO) neural network was -1.09 ℃, while the result predicted by the general BP neural network method fluctuated greatly, and the local optimum was always got. Moreover, the preset data by IBPMIMO method were run in the PU control software of the sterilization machine, and the simulation results showed that the error between the PU value measured in the experiment and the expected PU value was only 0.4. Using IBPMIMO to set the parameters of the spray nozzle for the sterilization machine can achieve better results and solve the problem of setting the sterilization temperature of the new beer bottle.