An Enhanced LSTM Framework for Solar Radiation Forecasting

Using Hybrid Grey Wolf–Waterwheel Plant Optimization

Alaa Mohamed Abdel-Moati1,* Abdullah Muhammad Ibrahim2

1 Department of Architecture Engineering, Delta Higher Institute for Engineering & Technology (DHIET), Mansoura, Egypt

2 Department of Civil Engineering, Delta Higher Institute for Engineering & Technology (DHIET), Mansoura, Egypt

Received: August 22, 2025 Revised: October 18, 2025 Accepted: December 20, 2025 ⋆ Corresponding author

ABSTRACT

Accurate solar radiation prediction is critical for optimizing renewable energy utilization, yet traditional machinelearning

models struggle with suboptimal forecasting accuracy. To address this challenge, we propose a novel

hybrid optimization algorithm, the Grey Wolf and Water Whale Plant Optimizer (GWWWPA), to enhance Long

Short-Term Memory (LSTM) networks for improved solar radiation forecasting. The proposed method leverages the

exploration-exploitation synergy of the Grey Wolf Optimizer (GWO) and Water Whale Plant Algorithm (WWPA)

to optimize LSTM hyperparameters effectively. Experimental results on the NASA Space Apps Moscow dataset

demonstrate that GWWWPA-LSTM outperforms existing optimization techniques, achieving the lowest Mean

Squared Error (MSE) of 0.00019392 and the highest R2 score of 0.917262, surpassing standalone GWO, WWPA,

PSO, and WOA. These findings highlight the potential of hybrid metaheuristic approaches in enhancing predictive

accuracy, facilitating more reliable solar energy forecasting, and supporting sustainable energy management systems.

Keywords: Hybrid Optimization Grey Wolf Optimizer Water Whale Plant Algorithm Solar Radiation Prediction

Long Short-Term Memory (LSTM)

1. INTRODUCTION

The global transition toward renewable energy sources has

become a critical priority due to the depletion of fossil fuels

and the urgent need to address climate change. Solar energy,

in particular, has garnered significant attention due to its

abundance, sustainability, and potential to reduce dependence

on non-renewable energy sources. Solar photovoltaic (PV)

systems have been widely adopted for electricity generation;

however, their efficiency is highly dependent on the availability

of solar radiation, which varies due to atmospheric

and meteorological conditions [1, 2]. These fluctuations pose

significant challenges for power grid operators, energy planners,

and businesses that rely on solar power. Accurate solar

radiation forecasting is essential for optimizing energy production,

enhancing grid integration, reducing energy costs,

and ensuring the reliability of solar power plants [3, 4].

Despite the increasing availability of meteorological datasets

and advancements in predictive modeling techniques, accurately

forecasting solar radiation remains a complex challenge.

Various environmental factors, such as temperature, humidity,

wind speed, barometric pressure, and cloud cover, influence

the intensity and distribution of solar radiation [5]. The nonlinearity

and randomness of these variables make traditional

statistical models insufficient for capturing intricate dependencies

[6]. As a result, machine learning and deep learning

approaches have emerged as powerful alternatives due to their

ability to learn complex patterns from historical data [7].

Among machine learning techniques, deep learning models,

particularly Long Short-Term Memory (LSTM) networks,