It is difficult to describe the randomness and volatility of the KELM (Kernel Extreme Learning Machine) parameters dependent and deterministic prediction of PV (Photovoltaic) power. In this article, a short-term PV power interval prediction model based on IAO (Improved Aquila Optimizer) to optimize KELM is proposed. First of all, the correlation analysis was applied to analyze the meteorological factors affecting the PV power output, and the input of the prediction models was obtained. In view of the situation that aquila optimizer was prone to fail into the local optimal solution, the logistic-sine chaotic mapping and adaptive weight factor were introduced to perfect aquila optimizer, and the improved algorithm was applied to optimize the parameters of the KELM, so IAO-KELM PV power deterministic prediction model was built. Finally, based on the basis of deterministic prediction, combined with Bootstrap method, the prediction error was analyzed to establish the PV power interval prediction model. The example verification shows that the IAO-KELM-Bootstrap interval prediction model proposed in this article has higher prediction accuracy, and can obtain the power fluctuation interval with high reliability and narrow interval.