An Ensemble-Optimization-Based Artificial Neural Network Approach for Predicting Global Horizontal Irradiance

Harnessing the Power of Metaheuristics for Solar Energy Forecasting

In the realm of renewable energy, harnessing the sun’s boundless energy has emerged as a beacon of hope. Among the various solar energy parameters, Global Horizontal Irradiance (GHI) stands tall as a crucial factor, dictating the amount of solar radiation reaching Earth’s surface. Accurate GHI prediction is the linchpin for efficient solar energy system design, operation, and integration into the grid. This article unveils a novel approach that leverages the synergy of metaheuristic algorithms and artificial neural networks (ANNs) to predict GHI with remarkable precision.

The Fusion of Metaheuristics and Artificial Neural Networks

Our proposed approach ingeniously employs a quartet of metaheuristic algorithms – Equilibrium Optimizer (EO), Whale Optimization Algorithm (WDO), Salp Swarm Algorithm (SOSA), and Orca Inspired Optimization (OIO) – to optimize the hyper-parameters of a Feed-Forward Artificial Neural Network (FFANN) model. This FFANN architecture comprises an input layer, a hidden layer, and an output layer. The input layer ingests eight meteorological and temporal factors, including temperature, humidity, pressure, wind direction, wind speed, day of the year, hour of the day, and month of the year. The hidden layer consists of six neurons, while the output layer yields the predicted GHI value.

The metaheuristic algorithms meticulously fine-tune the weights and biases of the FFANN, along with the population size of each algorithm. This optimization odyssey aims to minimize the Root Mean Square Error (RMSE) between the predicted and observed GHI values, paving the way for highly accurate GHI forecasting.

Data Preparation and Feature Selection

To train and validate our models, we meticulously curated a comprehensive dataset comprising 8803 records of meteorological and GHI data, meticulously collected from a solar radiation monitoring station in Bandar Abbas, Iran. This dataset was then judiciously partitioned into two subsets: a training set (7042 records) and a testing set (1761 records).

To unravel the intricate relationships between the input factors and GHI, we employed a battery of statistical tools, including the Pearson correlation coefficient and principal component analysis (PCA). This analysis unveiled the most influential factors, enabling us to optimize the dataset and enhance the prediction accuracy of our models.

Performance Evaluation: A Tale of Superiority

The performance of our EO-FFANN, WDO-FFANN, SOSA-FFANN, and OIO-FFANN models was meticulously evaluated using a triumvirate of accuracy metrics: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R). The results echoed the exceptional prowess of the EO-FFANN model, which outshone its counterparts in both the training and testing stages, achieving the lowest RMSE, MAE, and the highest R-value.

Taylor diagrams were meticulously crafted to visualize the models’ performance in terms of correlation and error. Once again, the EO-FFANN model stood out, exhibiting the lowest error and highest correlation, followed closely by WDO-FFANN and OIO-FFANN. The SOSA-FFANN model, however, displayed a noticeable weakness compared to the other three models.

PCA Importance Analysis: Unveiling the Key Players

PCA, a powerful statistical technique, was employed to identify the input factors that hold the greatest sway over GHI prediction. This analysis revealed that temperature, humidity, pressure, wind direction, day of the year, and month of the year are the most influential factors. On the other hand, wind speed and hour of the day can be judiciously discarded, thereby optimizing the dataset and expediting the GHI prediction process.

A Monolithic Formula: Simplicity at Your Fingertips

To facilitate the use of our proposed model without the need for computer-based programs, we derived a mathematical expression from the EO-FFANN model. This monolithic formula consists of two steps and requires the input of the six most influential factors identified by PCA. With this formula, GHI prediction becomes accessible and effortless, empowering users to harness the sun’s energy with ease.

Strengths, Limitations, and Future Horizons

Our study meticulously highlights the strengths and limitations of our proposed approach, charting a course for future research endeavors. The EO-FFANN model stands as a testament to its superior performance compared to previous studies, offering a simplified formula for practical applications. Future efforts could focus on reducing the dataset inputs to further lighten the computational burden and refine the GHI prediction solution.

Call to Action: Embrace the Solar Revolution

The sun’s boundless energy awaits your embrace. With our groundbreaking approach, harnessing solar power has never been simpler. Step into the future of renewable energy, where GHI prediction is a breeze, and the sun’s power is at your fingertips.