Enhancing Solar Power Forecasting Accuracy: The Impact of Data Normalization on Machine Learning Models
نویسندگان
1 Department of Electrical and Electronic Engineering, International University of Business Agriculture and Technology (IUBAT), Dhaka, Bangladesh.
2 Faculty of Engineering, Technology and Built Environment, UCSI University, Kuala Lumpur, Malaysia.
3 Department of Electrical and Electronic Engineering, International University of Business Agriculture and Technology (IUBAT), Dhaka, Bangladesh.
4 Department of Electrical and Electronic Engineering, Dhaka University of Engineering and Technology (DUET), Gazipur, Bangladesh.
5 Department of Mechanical Engineering, International University of Business Agriculture and Technology (IUBAT), Dhaka, Bangladesh.
6 Department of Electrical and Electronic Engineering, Dhaka University of Engineering and Technology (DUET), Gazipur, Bangladesh.
doi
10.30501/jree.2025.517696.2357چکیده
Accurate forecasting of solar power generation is essential for efficient energy planning and grid stability in renewable energy systems. Machine Learning (ML) techniques have become prominent tools for this purpose; however, their effectiveness often depends heavily on data preprocessing steps, particularly normalization. This study investigates the impact of data normalization on the predictive accuracy of various ML models applied to solar power generation forecasting. Using solar and meteorological data from Cocoa, Florida, six widely used ML algorithms, including Linear Regression, SMOreg, Multilayer Perceptron, M5Rules, Random Forest, and k-Nearest Neighbors (Ibk), were trained and evaluated with both raw and normalized datasets across different training data percentages (60% to 85%). The models were assessed using Root Mean Squared Error (RMSE) as the performance metric. Results show that data normalization significantly improves model performance, particularly for algorithms sensitive to feature scaling, such as Multilayer Perceptron and SMOreg. Among all the models, Random Forest consistently achieved the lowest RMSE across both raw and normalized datasets, demonstrating robustness. Normalization not only reduced error but also enhanced consistency in model predictions. The findings emphasize that normalization is a crucial preprocessing step for developing reliable ML models in solar energy forecasting. Future research should explore hybrid normalization strategies and larger, multi-regional datasets to enhance predictive accuracy and model generalization further.
