负荷预测, 时间序列分解, 在线学习, 组合算法, 隐马尔可夫模型

Since it is troublesome for conventional machine learning methods to extract the main features relevant to the uncertainties and variations of electrical load, in this paper, a recently proposed hidden Markov model based online learning algorithm is used to solve the load forecasting problems, extracting the uncertainties and variations from the load data. By combining with the decomposition algorithm, the variation features can be estimated more precisely and forecasting accuracy can be improved. Based on the hidden Markov model, the proposed algorithm is updated once new samples are received, thus adapting to real-time data; the STL algorithm is implemented to decompose the load data, leading to the separation of components with different trends. The online learning algorithm is then applied to each component of data, composing the hybrid load forecasting algorithm. Validated by three public datasets, it is shown that the proposed algorithm can improve the forecasting accuracy and reduce the relative error up to $27\%$ when compared with the existing technique.

Load forecasting, Time series decomposition, Online learning, Combined algorithm, Hidden Markov model