Long Short-term Power Load Forecasting Algorithm using Long Short-Term Memory Neural Network with Density-Based Spatial Clustering

Abstract

In pursuance of refined accuracy of short-term power load forecasting, this paper proposes Long Short-Term Memory Neural Network with Density-Based Spatial Clustering (LSTMNNDBSC) to forecast the short-term power load. The proposed algorithm facilitates the extraction of original power data features using 8-dimensional load features; while the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) was employed to classify the extracted data and further determine the power load cluster. Furthermore, the Long Short-Term Memory Neural Network (LSTMNN), was used to forecast the cluster of power load on the day of forecasting as well as to forecast the short-term power load. It is evident from the experimental results that the algorithm based on 8-dimensional load curve features exhibited a better clustering effect and required lesser amount of processing time compared to the 96-dimensional load features of the original data. Similarly, the LSTMNNDBSC algorithm employed lesser training time and achieved higher forecasting accuracy compared to the comparative forecasting algorithm.