Bus Load Forecasting Based on K-means and Long Short-Term Memory Networks

Bus load forecasting is essential for power system generation planning and dispatching operations. To improve the accuracy of bus load prediction, this paper proposes a bus load prediction method based on principal component analysis, K-means clustering, and long short-term memory networks. Firstly, the bus load is pre-processed by denoising and normalization. Then, the load characteristics and the correlation of numerous influencing factors with the daily load are analyzed. Principal component analysis is performed on the multidimensional features to obtain linearly uncorrelated new features. Finally, K-means clustering of historical loads by feature observations is performed, and different LSTM models are constructed for each subset for prediction. The experimental results show that the proposed hybrid prediction method has better performance.