Department of Energy Science and Engineering, Indian Institute of Technology Delhi, India
*Corresponding author: Ravi Kumar K, Department of Energy Science and Engineering, Indian Institute of Technology Delhi, India
Submission: December 06, 2022;Published: January 26, 2023
The demand of energy is continuously increasing as a consequence of industrial growth and advancements
in both developed and developing countries. Among other renewable energy resources, solar energy is
one of the cleanest and abundantly available energy resources. However, it is sporadic and diurnal in
nature. The accurate forecasting of solar radiation becomes essential for the effective utilization of solar
energy. In this study, efforts are made to predict solar radiation on daily basis 24-hours ahead of time
for the location of New Delhi (28.54 °N, 77.19 °E) which can be beneficial for day-ahead energy trading
and grid operation. Statistical models such as random forest regression tree, recurrent neural network
based Long Short-Term Memory Model (LSTM) and a hybrid model are used for the prediction of Global
Horizontal Irradiance (GHI). The hourly dataset used for the prediction of GHI is obtained from NREL
(National Renewable Energy Laboratory) from 2015 to 2020. It includes input features such as year,
month, day, hour, temperature, pressure, relative humidity, wind speed and sky clearness index.
The input features are used to construct models for forecasting the global horizontal irradiance of New
Delhi, India (28.54 °N, 77.19 °E). The performance of the models is evaluated by metrics such as Mean
Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE) and coefficient of
determination (R2). The results of the hybrid model are obtained as MAE of 26.22W/m2, MSE of 1304.45W/
m2, RMSE of 36.11W/m2 and R2 of 0.98 W/m2. The results indicate that the hybrid model outperforms
the standalone models as it utilizes the features of both the random forest model and LSTM model to give
improved R2 by 0.40% and 1.02% with respect to the random forest and LSTM model respectively. MAE is
decreased by 2.42W/m2 and 5.97W/m2, MSE is decreased by 256.10W/m2 and 917.85W/m2 and RMSE is
decreased by 3.38W/m2 and 11.02W/m2 with respect to the random forest and LSTM model respectively.
Keywords:Solar radiation forecasting; Global horizontal irradiance; Random Forest; Long Short-Term Memory (LSTM); Hybrid model; Coefficient of determination
Abbreviations:AR: Auto Regressive; ANN: Artificial Neural Network; ADAM: Adaptive Moment Estimation; ARMA: Autoregressive Moving Average; ARIMA: Auto Regressive Integrated Moving Average; DHI: Direct Horizontal Irradiance; DNI: Direct Normal Irradiance; GHI: Global Horizontal Irradiance; LSTM: Long Short Term Memory; ML: Machine Learning; MAE: Mean Absolute Error; MSE: Mean Square Error; RMSE: Root Mean Square Error; RELU: Rectified Linear Unit