An evolution-dependent multi-objective ensemble model of vanishing moment with adversarial auto-encoder for short-term wind speed forecasting in Xinjiang wind farm, China

Abstract

Wind speed forecasting can enhance the safety and economy of wind energy integration and conversion. The characteristic of the wind speed evolves over time. In this paper, a novel evolution-dependent multi-objective ensemble model of vanishing moment is proposed to solve the above problem. The proposed model can assign time-varying ensemble weights to base predictors with the different vanishing moments, so as to achieve better forecasting performance. In this model, the deep Adversarial Auto-Encoder (AAE) is firstly utilized to convert the wind speed into a two-dimension Gaussian distribution feature space. The Bat Algorithm (BA) is applied to partition the feature space into several sectors. Each sector represents a cluster of wind speed. Outlier Robust Extreme Learning Machines (ORELMs) improved by Maximal Overlap Discrete Wavelet Packet Transform (MODWPT) with the different vanishing moments are applied as base predictors, and the Multi-Objective Feasible Enhanced Particle Swarm Optimization (MOFEPSO) is used to obtain the best ensemble weights for each cluster. At last, the suitable ensemble model is applied for forecasting at each time. Four actual wind speed series collected from Xinjiang wind farm are used to verify the effectiveness of the proposed model. The experimental results indicate the proposed model outperforms other benchmark models.