In order to make the smart home system to have the ability of learning user behavior actively and provide services spontaneously, this paper introduced user behavior prediction model which combined back propagation neural network (BPNN) with Hadoop parallel computing to the traditional smart home system, numerous user-generated behavior and environmental parameters data are packaged in particular data frame format and uploaded to the cloud platform through 4G or WLAN by the home gateway. According to the received historical data, repeated parallel training of BPNN which run on cloud platform was utilized to achieve user behavior prediction. Case study on smart home validated that the proposed model is valid for user behavior prediction with accuracy elevated, it can help user to complete equipment operating independently in the corresponding cases. Another comparison, time efficiency experiment on the parallelized neural network algorithm also showed that the suggested method is excellent in convergence speed and accuracy.