研究目的
Investigating the influence of an additional input parameter on the accuracy of an already tested and validated offline model for short-term photovoltaic power forecasting without the need of Numerical Weather predictions data.
研究成果
The addition of the photovoltaic power at time t as an input parameter improves the accuracy of the LS-SVR model for short-term photovoltaic power forecasting, with an MSE of 0.0083 and RMSE of 0.0913. This improvement suggests that incorporating additional relevant parameters can enhance model performance.
研究不足
The study focuses on short-term forecasts without the need for Numerical Weather Predictions data, which may limit its applicability for longer-term forecasts or in scenarios where weather data is available and could improve accuracy.
1：Experimental Design and Method Selection:
The study uses the Least Square Support Vector Regression (LS-SVR) algorithm for forecasting photovoltaic power. Two models are compared: one with two inputs (solar irradiation and temperature) and another with three inputs (adding photovoltaic power at time t).
2：Sample Selection and Data Sources:
Data is collected from a hybrid platform located at Moroccan School of Engineering Sciences (EMSI – Casablanca, Morocco), with a 15-minute sample rate, including solar irradiation, plate temperature, and photovoltaic power.
3：List of Experimental Equipment and Materials:
The platform includes a PV plant and 2 wind turbines. The PV plant characteristics include Voltec Solar cells and an SMA inverter.
4：Experimental Procedures and Operational Workflow:
The LS-SVR algorithm is trained with a Radial Basis Function kernel. Internal parameters (γ and σ2) are found using a 10-folds cross-validation based on MSE. The model is validated with data not used in training.
5：Data Analysis Methods:
Performance metrics include MAE, MBE, MSE, RMSE, and RRMSE to evaluate model accuracy.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
