Machine Learning-Based Prediction of Photovoltaic Power Generation A Case Study Using Two-year Time Series Amsterdam Weather Data and SAM Simulations

This study investigates the performance of five machine learning (ML) algorithms for predicting photovoltaic (PV) power generation using weather data for Amsterdam in 2018 and 2019. These ML algorithms are Random Forest (RF), Gradient Boosting (GB), k-Nearest Neighbors (kNN), Neural Networks (NN) and Linear Regression (LR). The dataset of PV power that is used in this study is artificially generated with simulations by System Advisor Model (SAM). Further this data cover 2-year time series of relevant meteorological variables speed with the recording half-hour interval such as Direct Normal Irradiance (DNI), Diffuse Horizontal Irradiance (DHI), Global Horizontal Irradiance (GHI), ambient temperature, and wind. The SAM simulation generates 50 data points per hour, providing high-resolution sub-hourly calculations across the entire period. The dataset of PV power is divided into training (50%) and testing (50%) sets, with data from the year 2018 used for training and data from the year 2019 used for testing. This ensures a structured approach for model development and evaluation across two distinct years. Finally, the models are evaluated based on four performance indicators: Mean Squared Error (MSE), Mean Absolute Error (MAE), R-squared (R2), and Explained Variance Score (EVS). Among the five models tested— The RF algorithm demonstrated the best performance for this specific dataset and case study.