Intermittent and time-varying wind limits the power generation capacity of a wind farm. Accurate wind speed prediction will help reduce the impact of wind power connecting into a network on the operation modes arrangement of the power system. In view of the complex changes and chaotic characteristics of wind speed time series, in order to improve prediction accuracy and simplify the prediction mathematical model structure, this paper proposes a combined universal Gravitation Search Algorithm (GSA) full parameter continuous fraction prediction model with local optimization ability and the local fast convergence advantage of Particle Swarm Optimization (PSO). The n-term truncated continued fraction is transformed into a PSOGSA optimization parameter problem, and the function optimization in multidimensional space is carried out. Taking two groups of wind speed data collected from a wind farm as the prediction object, the complex wind speed time series is modeled and simulated, and the multi-scale prediction of the series is carried out using the full parameter continued fraction based on PSOGSA optimization. The simulation results are compared with the traditional BP neural network, RBF neural network, Long-Term Memory Network Algorithm (LSTM) that current time series prediction often uses, and the Particle Swarm Optimization Support Vector Machine (PSO-SVM). It is concluded that the full parameter continuous fraction prediction model based on PSOGSA has characteristics of high accuracy, simple structure and fast modeling speed, and stronger nonlinear prediction ability. This work is supported by National Natural Science Foundation of China (No. 51767022, No. 51967019 and No. 51575469) and Natural Science Foundation of the Xinjiang Uygur Autonomous Region (No. 2019D01C082).