In order to address the challenges of improving energy
efficiency and integration of renewable energy, multi-energy systems, composed
of electric, natural gas, heat and other energy networks, have received more
and more attention in recent years and have been rapidly developed. Through integration as a multi-energy system,
different energy infrastructures can be scheduled and managed as one unit. One
of the main stages in the optimal scheduling of a multi-energy system is the
predictions of various demands and sustainable energy in the scheduling
horizon. <a>This paper proposes a prediction model based on
adaptive random forest for demands and solar power of a real MES, Stone Edge
Farm, in California. </a><a>The adaptive random forest
model can provide a probability distribution of the prediction results. This
allows users to consider a variety of scenarios that may occur in the future
for further system operation optimization and help users evaluate the reliability
of the results.</a> Besides, an online self-adaptability feature is implemented
with the model so it can adapt to the new forecasting environment when new
observations are detected. The simulations show that the adaptive random forest
model is better than the benchmark models in terms of prediction accuracy.