Statistical Validation and Machine Learning Prediction of Solar Power Generation in a Developed Automated Multi-Source Power Generation System
1Akinrinade, N. A., 2,3*Onawumi, A. S., 2,3Sangotayo, E. O., 4Oke, A. M., 1Omoniyi, E. B., 5Ajibowu, S. B., 6Oni, A. A. and 6Ajayi, M. O.
1Department of Mechanical Engineering, College of Engineering, Bells University of Technology, Ota, Ogun State, Nigeria.
2Department of Mechanical Engineering, Faculty of Engineering, Ladoke Akinola University of Technology, Ogbomoso. Nigeria.
3Department of Mechanical and Mechatronics Engineering, College of Engineering and Technology, Achievers University, Owo. Nigeria.
4Department of Agricultural and Bio-resources Engineering, Bells University of Technology, Ota, Ogun State, Nigeria.
5Department of Electrical, Electronics and Telecommunication Engineering, Bells University of Technology, Ota, Ogun State, Nigeria.
6Department of Computer Engineering, College of Engineering, Bells University of Technology, Ota, Ogun State, Nigeria.
DOI: 10.36108/laujoces/6202.71.0170
Abstract
The rising integration of renewables into modern energy grids, optimising operational management and reliability in Automated Multi-Source Power Generation Systems (AMPGS), is critical for grid stability. Models of AMPGS currently available are mostly not validated with statistical testing and do not provide accurate prediction tools for operation. This paper represents statistical validation and machine learning predictive modelling of an AMPGS. Two data set consist of 1000 experimental and simulated observations were analysed using six (6) variables, which include Solar Power Current (SPC), Battery Level (BL), Temperature (Temp), Irradiance (Ir), Response Time (RT) and Solar Power Generation (SPG). Kolmogorov-Smirnov (KS) test, t-test, histogram comparison and correlation test were considered for statistical analysis. Eight (8) machine learning algorithms: Artificial Neural Network (ANN), Random Forest (RF), Extra Trees, Gradient Boosting (GB), AdaBoost, Decision Tree, K-Nearest Neighbour (kNN) and XGBoost were trained using 80:20 training-to-testing split. These machine learning algorithms were evaluated by prediction accuracy values such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R2). The KS test results were given by statistics as 0.045 and a p-value of 0.1383. The p-value from the t-test was 0.3329, proving no significant differences between experimental and simulation data. The Extra Trees machine learning algorithm yielded the best results with R2 of 0.993 for the experimental and simulation datasets. The results show that the developed AMPGS simulation is able to replicate the measured SPG behaviour with good statistical agreement and that Extra Trees is an accurate data-driven model for SPG prediction.
Keywords: Hybrid Energy System, Machine Learning, Extra Trees Algorithm, Statistical Validation, Renewable Energy Management.