Novel and Efficient Hybrid Deep Learning Approach for Solar Photovoltaic Power Forecasting Considering Meteorological Parameters

Abstract

The power generation from photovoltaic plants depends on varying meteorological conditions. These meteorological conditions such as solar irradiance, temperature, and wind speed, are non-linear and stochastic thus affect estimation of photovoltaic power. Accurate estimation of photovoltaic power is essential for enhancing the functioning of solar power installations. The paper aims to develop a novel deep learning based photovoltaic power forecasting model on different weather conditions. The proposed model utilizes a two-stage deep learning framework for accurate solar power forecasting, which combines the long short-term memory (LSTM) and convolutional neural network (CNN) deep learning architectures. The key role of CNN layer is to identify the weather conditions, i.e., sunny, cloudy and rainy while the LSTM layer learns the patterns of solar power generation that depend on weather variations to estimate photovoltaic power. The proposed hybrid models consider meteorological factors, such as wind speed, sun irradiations, temperature, and humidity, including cloud cover and UV index to provide precise solar power forecasting. The presented hybrid model, a Root Mean Square Error of 0.0254, 0.03465 and 0.0824, Mean Square Error of 0.000645, 0.00120 and 0.00679, R2 of 0.9898, 0.9872 and 0.9358, Mean Average Error of 0.0163 and 0.0236 and 0.2521 for sunny, cloudy and rainy weather conditions respectively. The results demonstrate that presented deep learning based novel solar photovoltaic (SPV) power forecasting model can accurately forecast solar power based on instantaneous changes in generated power patterns, and aid in the optimization of PV power plant operations. The paper presents an effective methodology for forecasting solar power that can contribute to the improvement of solar power generation and management.

Estimating hourly air temperature in an Amazon-Cerrado transitional forest in Brazil using machine learning regression models

“…In particular, the feature of hourly temperature is strongly connected with the thermodynamic processes that influence the energy and matter exchanges between biosphere and atmosphere in natural ecosystems [23,24]. However, accurately measuring and predicting temperature can be a challenge due to its spatial and temporal variability [25][26][27].…”