Transfer learning based multi-layer extreme learning machine for probabilistic wind power forecasting

With the increasing penetration of wind power, probabilistic forecasting becomes critical to quantifying wind power uncertainties and guiding power system operations. This paper proposes a transfer learning based probabilistic wind power forecasting method. Model-based transfer learning is utilized to construct the multilayer extreme learning machine (MLELM). The output mapping factors of MLELM are further optimized through the particle swarm optimization (PSO) with the objective of minimizing the quantile evaluation indexes. Joint distribution adaptation (JDA) is utilized to update the weights of MLELM to accommodate variable wind power output. Test results on the practical wind farms in China shows that the proposed method can provide more accurate quantile forecasting results with better nonlinear fitting ability compared with other quantile forecasting methods.

Automated summary

This paper proposes a transfer learning-based method for probabilistic wind power forecasting. It utilizes model-based transfer learning to construct a multilayer extreme learning machine, optimizes the output mapping factors using particle swarm optimization, and updates the weights through joint distribution adaptation. The method achieves more accurate quantile forecasting results and better nonlinear fitting ability compared to other methods.