Forecast electricity demand in commercial building with machine learning models to enable demand response programs

Electricity load forecasting is an important part of power system dispatching. Accurately electricity have great impact on a number departments in systems. Compared to simulation (white-box model), (black-box model) does not require expertise building construction. The development cycle the model much shorter than design simulation. Recent developments machine learning lead creation models with strong fitting and accuracy deal nonlinear characteristics. Based real dataset, this paper evaluates compares two mainstream short-term techniques. Before experiment, first enumerates common methods explains principles Long Short-term Memory Networks (LSTMs) Support Vector Machines (SVM) used paper. Secondly, based characteristics data pre-processing feature selection takes place. This describes results controlled experiment study importance selection. LSTMs SVM are applied one-hour ahead one-day peak valley forecasting. predictive these calculated error between actual predicted loads, runtime recorded. show that higher prediction when sufficient. However, overall performance better train insufficient time cost prioritized.