The future energy utilization will move towards the user-centered integrated energy system, which is of great significance to analyze the behaviour of user-side energy consumption. Firstly, this paper constructs the Spark-BIRCH algorithm based on balanced iterative reducing and clustering using hierarchies analysis(BIRCH) and Spark platform. Secondly, according to the clustering results of user's energy consumption behaviour, a muti-threaded prediction model (Spark-DBN) for user energy demand is constructed by combining the Spark platform and the Deep Belief Neural Network(DBN). Finally, the proposed parallel clustering algorithm and muti-threaded demand forecasting model are validated and simulated by numerical example, and the results show that the constructed Spark-BIRCH user behavior clustering method has high accuracy and validity, the method has excellent performance, which effectively saves time cost. In addition, the Spark-DBN prediction model has smaller prediction errors and higher accuracy than the single prediction model. The simulation results has verified the validity ad feasibility of the two models.