Abstract: Due to the fluctuation of energy output and load of microgrid system, it is difficult to control its operating cost effectively. Therefore, a collaborative control method of source and network load and storage based on ensemble learning is proposed for microgrid system. After calculating the source(renewable energy generation), network(grid structure), load(electricity consumption load) and storage(energy storage equipment) of the microgrid system, an objective function based on minimizing the cost input of the microgrid system is constructed on the premise of meeting the load demand. In the control process, the integrated learning algorithm is introduced, and the integrated learning strategy set is set according to the executing scheduling parameters of the source network load and storage of the microgrid system. According to the TOU price information and the load curve prediction results under the power failure compensation mechanism, the parameters when the evolutionary integrated learning reaches the stable state are determined as the final control parameters. In the test results, the total cost of power purchase, energy storage and abandoned scenery is only 31,000 yuan, which is significantly lower than the control group.