科研之友
微信
新浪微博
Facebook
分享链接
摘要
Electric load forecasting, as a crucial way to tackle the energy tasks, helps to provide correct guidance and reduce energy consumption for an electric system. This paper introduces a novel forecasting method that combines wavelet transform (WT) and particle swarm optimization (PSO) with an extreme learning machine, phase space reconstruction (PSR), and a least squares support vector machine. Because the electric data that are easily interfered by many factors are unstable, directly adopting the data containing noise will affect the forecasting performance. Prior to the electric forecasting conduct, WT is adopted to reduce the noise. Because only a single method is not able to suit for all kinds of data, three individual models with a strong nonlinear learning capability are applied to deal with the whole time series that have complex nonlinear characteristics. They are used to obtain intermediate forecasting results. Besides, the C-C method is utilized to acquire the optimal delay time and embedding dimension of the PSR. PSO is employed to attain the best proportion of each single method of the hybrid model. By multiplying the corresponding optimal weights by all the three forecasting results and then adding them up, the final result of the hybrid method can be attained. In order to assess the accuracy and applicability of the proposed hybrid method, half-hourly electric load data of New South Wales and Victoria of Australia are used as case studies. The experiment results show that the hybrid method outperforms the individual methods and is appropriate for different kinds of data. Published by AIP Publishing.
- 出版日期2017-3
- 单位兰州大学
