Machine Learning-Based Demand and PV Power Forecasts

In recent years, predicting the performance of the power system is of high-priority to alleviate potential uncertainties. Achieving accurate load prediction at the customer and at the renewable energy resources levels will lead to more efficient balance of supply and demand. In this paper, we investigate different models including Support Vector Machines (SVM), Multi-Layer Perceptrons (MLP) and Long Short-Term Memory (LSTM). The models were used for two tasks: predicting the Load Demand and the PV generation. For the Load Demand prediction models, the dataset considered was the aggregated load demand for 40 randomly chosen homes in Austin, Texas and the weather in Austin, Texas. For the PV generation Model, the dataset considered was the PV generation of Yulara Plant in Australia and the weather conditions for the site. The MLP model gave us the best results for the Load Prediction closely followed by the LSTM and the LSTM model gave us the best result for the PV Prediction.