Abstract: Settlement mechanism of real-time deviation power is an important part of electricity market system. The deviation power of electricity retailers in real-time market will bring certain risk of loss of income under this mechanism. In view of upcoming deviation power settlement mechanism in spot market of Guangdong Province， this paper proposes a control strategy to reduce real-time market deviation of electricity retailers. The constraints of power control strategy of electricity retailers under different risk acceptance levels are established. Case study is carried out on the basis of the prediction of extreme learning machine and the reduction of scene. Through sensitivity analysis， effects of prediction accuracy， user-side resource characteristics and cost on the real-time deviation of power control of power sales companies are studied.

Key words: spot market, real-time deviation power, electricity retailer, optimized scheduling