Renewable Energy Power Generation Prediction Method Combining Spatio-Temporal Autocorrelation and Multi-Dimensional Decomposition

With the problems of energy security and environmental pollution becoming more and more prominent, how to increase the diversity of power sources, enhance the proportion of renewable energy as well as solve the problems of difficult prediction of renewable energy power generation are crucial for grid stability and management. Traditional time series forecasting models are difficult to cope with the volatility and spatial-temporal correlation of renewable energy generation. In this paper, we propose an innovative integrated forecasting model that incorporates spatial autocorrelation convolutional long-short time memory network and variational modal decomposition model (AHA-VMD) improved based on artificial hummingbird algorithm. Firstly, the generation power time-series signal is decomposed into several sub-signals using the decomposition model. Next, the spatio-temporal autocorrelation of the sub-signals is calculated separately, based on which they are input into the prediction model. The model enhances the capability of capturing the generation power under different time scales and weather conditions through multi-dimensional clustering of the sub-signals as well as spatial statistical properties. Experimental results show that the proposed model outperforms other single or simply integrated models.