2020/04/02 by Mehdi Neshat, Neshat, Mehdi, Meysam Majidi Nezhad +15 · 1 citation
Engineering · #Energy Load and Power Forecasting #Wind Energy Research and Development #Wind Turbine Control Systems
paper · pdf · doi:10.48550/arxiv.2004.12794
Reliable wind turbine power prediction is imperative to the planning,\nscheduling and control of wind energy farms for stable power production. In\nrecent years Machine Learning (ML) methods have been successfully applied in a\nwide range of domains, including renewable energy. However, due to the\nchallenging nature of power prediction in wind farms, current models are far\nshort of the accuracy required by industry. In this paper, we deploy a\ncomposite ML approach--namely a hybrid neuro-evolutionary algorithm--for\naccurate forecasting of the power output in wind-turbine farms. We use\nhistorical data in the supervisory control and data acquisition (SCADA) systems\nas input to estimate the power output from an onshore wind farm in Sweden. At\nthe beginning stage, the k-means clustering method and an Autoencoder are\nemployed, respectively, to detect and filter noise in the SCADA measurements.\nNext, with the prior knowledge that the underlying wind patterns are highly\nnon-linear and diverse, we combine a self-adaptive differential evolution\n(SaDE) algorithm as a hyper-parameter optimizer, and a recurrent neural network\n(RNN) called Long Short-term memory (LSTM) to model the power curve of a wind\nturbine in a farm. Two short time forecasting horizons, including ten-minutes\nahead and one-hour ahead, are considered in our experiments. We show that our\napproach outperforms its counterparts.\n