A Maximum Power Point Tracking in Wind Energy Conversion Systems using Machine Learning
S.Venkatesh Kumar1, P.Sebastian Vindro Jude2, K.Balamurugan3
1S.Venkatesh Kumar*, Assistant Professor(Sr.Gr), Department of EEE, Department of EEE Coimbatore, Tamilnadu, India.
2P.Sebastian Vindro Jude, Assistant Professor(Sr.Gr), Department of EEE, Department of EEE Coimbatore, Tamilnadu, India.
3K.Balamurugan, Assistant Professor(Sr.Gr), Department of EEE, Department of EEE Coimbatore, Tamilnadu, India
Manuscript received on March 18, 2020. | Revised Manuscript received on April 02, 2020. | Manuscript published on April 30, 2020. | PP: 717-721 | Volume-9 Issue-4, April 2020. | Retrieval Number: D7025049420/2020©BEIESP | DOI: 10.35940/ijeat.D7025.049420
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: In this paper, an efficient and feasible algorithm to extract the maximum power point (MPP) in wind energy conversion systems (WECS) by implementing machine learning (ML) into perturb and observe (P&O) algorithm is presented. The proposed algorithm is simulated on a separately-excited DC generator. This model uses instantaneous measurements of wind speed, humidity, temperature, pressure and generator speed to estimate a MPP by using ML at the end of each iteration. From this estimated power point, the controller follows quick perturbation to calculate the accurate MPP and is used as training data for further predictions in the next iteration. The controller learns from this training set and estimates the MPP closer to the maximum achievable power (MAP) which is corrected again through perturbation and is recorded. With the progress of time, the approximation of the maximum power point becomes more accurate whilst the time in further perturbation required for modification decreases. This model adapts to the versatile climatic conditions and yields an efficiency of 99.95% in predicting the MAP at the end of 1000 iterations corresponding to 2 hours 30 minutes.
Keywords: Wind energy conversion systems, Maximum power point tracking, Perturb and Observe, Machine Learning, Artificial Intelligence