In order to solve the conflict between the interests of the participants and the reliable operation of the equipment in the integrated energy system, an integrated energy system optimization strategy based on non-cooperative game is proposed. For the uncertainty of wind, light and load, the Latin hypercube sampling method and K-means clustering method were used to generate typical models of predicted output. Considering the benefits and reliability of source, network, load and storage, the model leads into flexible load of electric vehicle on the user side to enhance the utilization rate of energy and analyzes the balanced interaction strategy of all parties in the pursuit of benefits and operational reliability. For the problems such as uneven distribution of wolves in the traditional Gray Wolf Algorithm and weak ability to search for prey, a more uniform initial Wolf pack was generated based on the Hammersley sequence, and the processing of off-limit individuals was adjusted to produce high-quality wolves, enrich the sample species, and reduce the time and times of optimization . The effectiveness of the model and the algorithm is verified.