International Journal of
Advanced Multidisciplinary Application

Volume 3, Issue 6

• Author(s)Arjun Verma, Pallabi Ghosh
• AffiliationDepartment of Electrical Engineering, Rajasthan Technical University, Kota, Rajasthan, India Department of Electronics and Electrical Engineering, Jalpaiguri Government Engineering College, Jalpaiguri, West Bengal, India
• Page No.117-124
• Volume, Issue & YearVolume 3, Issue 6, June 2026
• Published On2026/06/14
• JournalInternational Journal of Advanced Multidisciplinary Application (IJAMA)
• ISSN No.3048-9350

Abstract

Accurate short-term solar photovoltaic (PV) power forecasting is a prerequisite for safe and economically efficient integration of large-scale solar generation into national electricity grids. India's solar installed capacity reached 73.3 GW in March 2024, representing 18.4% of total installed generation capacity, with the National Solar Mission targeting 500 GW of renewable capacity by 2030 — a scale at which forecasting errors impose measurable ancillary service costs and grid stability risks. Existing forecasting models based on statistical methods (ARIMA), machine learning (SVR, MLP), or single-architecture deep learning (LSTM, CNN) fail to simultaneously capture the multi-scale temporal patterns and spatial feature correlations embedded in multi-variate meteorological and operational input data. This paper proposes a hybrid Convolutional Neural Network — Bidirectional Long Short-Term Memory (CNN-BiLSTM) model with an additive attention mechanism for 24-hour-ahead solar PV power forecasting using a 12-variable feature set including solar irradiance, ambient temperature, panel temperature, cloud cover fraction, aerosol optical depth, wind speed, humidity, precipitation, atmospheric pressure, dew point, hour-of-day, and day-of-year. The model is trained and evaluated on two years of hourly data (2021–2023) from a 5.5 MW ground-mounted PV plant in Jodhpur, Rajasthan — a high-irradiance semi-arid location representative of Rajasthan's dominant role in India's solar generation portfolio. The proposed model achieves RMSE of 0.143 MW, MAE of 0.109 MW, and MAPE of 3.7% on the held-out test set, outperforming ARIMA (MAPE 12.4%), SVR (8.9%), MLP (7.6%), standalone LSTM (6.3%), and CNN-LSTM without attention (5.1%). Permutation feature importance analysis confirms solar irradiance (34.2%), ambient temperature (18.7%), and hour-of-day (14.3%) as the three dominant predictors. Seasonal analysis reveals the largest absolute error in winter months (December–February, RMSE 0.168 MW) attributable to fog-induced irradiance attenuation in the Thar Desert region — a meteorological phenomenon not fully captured by standard numerical weather prediction inputs.

Keywords: solar PV forecasting, deep learning, CNN, BiLSTM, attention mechanism, renewable energy, short-term forecasting, Rajasthan, grid integration, MAPE

References

1. [1] Abdel-Nasser, M., & Mahmoud, K. (2019). Accurate photovoltaic power forecasting models using deep LSTM-RNN. Neural Computing and Applications, 31(7), 2727-2740.
2. [2] Bahdanau, D., Cho, K., & Bengio, Y. (2015). Neural machine translation by jointly learning to align and translate. Proceedings of ICLR 2015.
3. [3] CERC. (2022). Deviation Settlement Mechanism and Related Matters (Sixth Amendment) Regulations. Central Electricity Regulatory Commission, New Delhi.
4. [4] Fan, G. F., Peng, L. L., & Hong, W. C. (2021). Short term load forecasting based on phase space reconstruction algorithm and bi-square kernel regression model. Applied Energy, 224, 380-392.
5. [5] Kushwaha, V., & Pindoriya, N. M. (2019). A SARIMA-RVFL hybrid model assisted by wavelet decomposition for very short-term solar PV power generation forecast. Renewable Energy, 140, 124-139.
6. [6] Li, G., Wang, H., Zhang, S., Xin, J., & Liu, H. (2020). Recurrent neural networks based photovoltaic power forecasting approach. Energies, 12(6), 2538.
7. [7] MNRE. (2024). Annual Report 2023–24. Ministry of New and Renewable Energy, Government of India.
8. [8] Qin, Y., Song, D., Chen, H., Cheng, W., Jiang, G., & Cottrell, G. (2017). A dual-stage attention-based recurrent neural network for time series prediction. Proceedings of IJCAI-17, 2627-2633.
9. [9] Reikard, G. (2009). Predicting solar radiation at high resolutions: A comparison of time series forecasts. Solar Energy, 83(3), 342-349.
10. [10] Saini, L. M., Aggarwal, S. K., & Kumar, A. (2020). Parameter optimisation using AGO with Fuzzy Logic for short-term load forecasting in Indian power systems. Expert Systems with Applications, 37(2), 1689-1695.
11. [11] Shi, J., Lee, W. J., Liu, Y., Yang, Y., & Wang, P. (2012). Forecasting power output of photovoltaic systems based on weather classification and support vector machines. IEEE Transactions on Industry Applications, 48(4), 1064-1069.
12. [12] Voyant, C., Notton, G., Kalogirou, S., Nivet, M. L., Paoli, C., Motte, F., & Fouilloy, A. (2017). Machine learning methods for solar radiation forecasting: A review. Renewable Energy, 105, 569-582.

Explore Our Related Journals

Looking for the right journal for your next manuscript? Explore our international peer-reviewed journals covering engineering, management, computer science, artificial intelligence and multidisciplinary research.