Current models for the prediction of the output power of photovoltaic (PV) clusters suffer from low prediction accuracy and are prone to overfitting. To address these problems, we propose an improved random forest (RF)-based method for ultra-short-term prediction of PV cluster output power. The total output power data for the PV clusters are used as the training dataset and fed into the RF model to obtain preliminary predictions. The error and accuracy of the preliminary predictions for individual sampling points concerning the actual values of the PV cluster output power are assessed. Each of the daily time series of preliminary predictions is divided into two phases according to whether the output power is increasing (morning) or decreasing (afternoon). The final ultra-short-term predictions of the PV cluster output power are obtained by correcting the two phases of preliminary predictions through trend correction and peak correction, respectively. The results show that, compared with the unimproved model, the accuracy of the stochastic forest model is 1.48% higher than that of the modified random forest model., which proves the effectiveness and practicability of the proposed method.
Short-term Prediction of Output Power and Constrained Optimal Control for Point-Absorber Type Wave-Energy Converters with Linear Generators
