A New Energy Power Forecasting Method using Neural Networks with Autocorrelated Errors

With the transformation and upgrading of the energy structure, the installed capacity and power generation of new energy have continued to grow in recent years. Due to the randomness of new energy, the impact on the grid will increase with its large-scale grid connection. Therefore, in order to ensure the stability of the power grid, it is necessary to make a reasonable forecast of the new energy power. So far, there have been many authors using deep methods to perform new energy power forecasting. However, these methods usually assume the errors of the model across different time steps are independent, which is inaccurate because of the temporality of the series. In this paper, a method that combines neural networks with autocorrelation errors is proposed for new energy power forecasting. Through the ablation experiments in difference datasets, we verify the effectiveness of the methods.