A Study of Computer Vision, Deep Learning, and Machine Learning Techniques for Forecasting Solar Power and Renewable Energy

Introduction

Socio-economic development has led to progressive increase in energy demand that cannot be met with traditional technology with local resources only. Renewable energy is seen as a transformative solution to this increasing energy demand and economic challenges like rising coal prices. (Huang et. al., 2017; Jursa, 2007; Rohrig et. al., 2008) Further, renewable and natural sources are available easily, are in abundance and are also environment friendly. The nature of power generation from renewable energy plants is unpredictable because it is dependent on numerous variables, including location, wind speed, weather, and other climatic conditions. (M. Abdel-Nasser et al., 2019; Bouzerdoum et al, 2013) It is therefore pertinent to mention that integration of variable generation from solar and wind sources have become a real challenge for system operators in ensuring reliability and security of the power system network particularly with an exponentially increasing renewable energy share in our worlds energy mix.

Figure 1.

Renewable energy source

For variable energy sources to be integrated into the grid, renewable energy forecasting is crucial because it allows power systems to manage the intermittent nature of the energy supply at different spatiotemporal scales. To predict the future impact of cloud displacements on the energy generated by solar facilities, traditional modelling approaches rely on numerical weather prediction or physical models. (Bacher et al, 2009; Bapai et al., 2019). These models find it difficult to pick up systematic biases and integrate cloud data. By combining surface measurements from multiple sources with real-time cloud cover observations via machine learning, a few of these restrictions are overcome for improving computer vision. With a focus on deep learning, (AlKandari et al., 2019; Alsharif et al., 2019, Aslam et al., 2019) which offers the theoretical groundwork required to create architectures that can extract relevant data from data produced by weather stations, satellites, ground-level sky cameras, and sensor networks, this review gives an overview of recent developments in solar forecasting from multisensory observations of Earth. Taking everything into account, machine learning has enormous potential to improve the accuracy and robustness of solar energy meteorology. Nevertheless, further investigation is required to fully grasp this potential and tackle its constraints (Sheng, et al., 2018; Torres- Barran et al. 2019).

This chapter has been divided into various sections. The introduction and the purpose of the study were covered in section 1. In order to lessen the effects of solar intermittency, section 2 discusses solar power forecasting, which is the procedure for gathering and assessing data to predict the generation of solar power on different time horizons. The study's motivation and the list of current review articles are presented in section 3. Section 4 describes the computer vision of solar forecasting method. The various deep learning techniques utilised in vision-based solar energy modelling are covered in Section 5; the study's conclusion is presented in Section 6.