The optimization of short-term load forecasting simulation for the Grid power system can improve prediction accuracy and robustness of the results. Although the existing prediction models can meet the requirements of prediction speed, the accuracy and stability of the predicted results are always difficult to guarantee. In order to get more accurate and stable forecast results, this paper puts forward the bacterial foraging algorithm to optimize the new predicting model of the extreme learning machine. First, the training sample and forecast sample set are formed in the power load sampling data set. The bacteria foraging optimization algorithm is used to optimize the uncertain parameters in the prediction model of extreme learning machine algorithm. Then, the improved model for power load forecasting is used. Through the optimization of the new model simulation, the results show that the use of bacterial foraging algorithm optimization model to predict extreme learning machine precision and stability are superior to the traditional forecasting model prediction results, and the algorithm has good practicability.