Integrated Regression Approach for Prediction of Solar Irradiance Based on Multiple Weather Factors

1. Introduction

The Solar irradiance is renewable source of energy provided by sun naturally. It is electromagnetic radiation emitted by the sun, in generic manner. It is converted into useful electric energy using feasible technology. However, the economic feasibility of this type of conversion setup depends on ambient weather condition and available solar resource at that location. After accurate solar irradiance prediction at any particular location, electric power grids or Solar PVs can be deployed so as to minimize the physical expenses. At present, actually when these systems are installed at majority areas of our country India, it is seen that though these modules are placed after the analysis of ambient weather conditions and sun direction, output power generated does not meet the desired level. Some significant reasons are, modules are manufactured at STC, and deployed by taking into consideration only solar PV panel orientation. Also Solar irradiance varies with meteorological data longitude, latitude, wind speed and change of weather at different areas. Therefore, deviation from standard test conditions affect generation of output power. So this paper proposes the methodology to predict the collection of energy for particular location, by taking into consideration ambient parameters, before deployment and similarly after deployment also, this methodology can be helpful in predicting solar energy collection in advance for its proper utilization and distribution.

The case study is done for Bhopal (MP), location in the central region of India with diverse climate but solar radiation in a finite range. Thus the climate of this city depicts the climate of the country and is suitable for solar radiation based prediction model. Existing Literature provided different benchmark forecasting techniques in the domain of statistical methods, machine learning and deep learning, for predicting solar PV energy output, efficiency, solar radiation with the help of historical time series data, either endogenous or exogenous dataset. Existing methods produce forecasts every 6 hours, mostly based on historical data at a location. Some limitations of existing research is the training set is reduced for longer time horizons as one cannot use unavailable weather forecasts and majority study depicts prediction of PV panel output, so in fact, PV panel deployment setup is required for experimentation. Majority of work reflects locations worldwide, but a few case studies are there from the country like India where there is more need of renewable energy sources. Some recent work is also focused on use of weather factors for prediction of solar radiation, that too by applying neural network, however, benchmark weather factors are difficult to find out.

The contribution of the paper is:

• •

  Presented, preprocessing of standard meteorological dataset for solar radiance prediction, which is actual weather for the given location and interdisciplinary research opportunity is explored

• •

  Identified the most significant environment variables, innovative combination of these variables is provided and train hybrid machine learning model with generic setting and hyper parameter tuning

• •

  Demonstrated hybrid model accuracy for irradiance prediction and significance of correlation of weather factors for solar irradiance prediction

• •

  Demonstrated App setup for predicting solar irradiance at any location based on GPS and hybrid machine learning model in the background to verify the site for deployment of PV panel/solar farm. This is innovative holistic approach for sustainable energy sources.

The paper is organized into four sections. Current section is an overview of the paper title. Subsequent section, existing work, gives literature reviews followed by proposed methodology in third section. Section four demonstrates experimental setup and work, results, discussions followed by conclusion.

2. Background

The authors of (Alzahrani, A, et al.(2014)) has described various methodologies for obtaining parameterized model to estimate generated power in PV generation systems, as these systems are weather independent and hard to predict, site parameters at the hourly level. Time series historical data are relatively rare, so in the recent years statistical approaches and NWP model based exogenous datasets have been developed to help the research community to work on solar radiance prediction.