In order to improve the prediction accuracy of photovoltaic power, this paper establishes an ultra-short-term photovoltaic power prediction model (KIBE model) based on Kmeans and IWOA-BiGRU-EC. Firstly, Kmeans clustering was used to divide historical data into similar daily samples of three types of weather types; secondly, in order to improve the prediction effect of BiGRU (Bidirectional Gated Recurrent Unit), an Improved Whale Optimization Algorithm (IWOA) was proposed by introducing improved Circle chaos map initialization, nonlinear convergence factor and adding inertia weight, and the IWOA-BiGRU model was constructed for preliminary prediction; finally, the Error Correction (EC) was carried out in combination with the preliminary prediction results, and the error sequence was decomposed into several components by using the variational mode decomposition. The IWOA-BiGRU prediction model is used to predict each component and superimpose to obtain the error prediction results. Calculate the arithmetic sum of the preliminary prediction result and the error prediction result to obtain the final power prediction result. The measured data of a photovoltaic power station in Jiangsu Province were selected for experiments. Results show that the KIBE model proposed in this paper can effectively improve the accuracy of prediction results under three types of weather conditions compared with other models.