Development of an improved hybrid back propagation ANN for low wind speed prediction and wind energy evaluation

Abstract

Wind energy is clean, reliable, and affordable renewable energy which can be harnessed during the day and night. Before a wind turbine is installed, wind resource assessment (WRA) must be conducted in order to evaluate the wind power potential. The most important parameter in WRA is the wind speed values. The traditional ways of measuring wind speed could not be relied on, due to time constraint and cost. Because of these problems, a prediction model using deep learning is proposed in paper to solve the lingering problem. The objective of this paper is to develop a machine learning, prediction model using available data. Wind data were obtained from Malaysia Meteorological Department (MMD) for a period of ten years starting from 2008-2018. The wind energy evaluation was conducted at 10m-40m meters, respectively. In the areas with limited data or without data a prediction model was developed using different Artificial Neural Networks (ANNs) structures. The model was trained, tested, and validated using measured wind speed in the nearby location. The optimized model in terms of less structure with high prediction accuracy was selected for the final prediction has a correlation value of 0.952. A detailed wind resource assessment was conducted in the areas based on most fitted wind speed distribution model. It was found that Weibull and Rayleigh fitted the wind speed in the areas examined. At the end of the analysis, low wind speed turbine was selected for the wind farm sitting; the results show that wind energy can be harnessed for small Pico scale application such as rural electrification, and grain grinding. Because, in all the cases the wind power density falls within class 1 (PD≤100 W/m2). The outcomes of this study would be useful for policy makers to implement 3-Es Model (Earth-Energy and Empowerment) in rural and remote areas of Sarawak.