| dc.description.abstract | In response to the challenge of improving energy production and consumption efficiencies due to environmental problems and energy crisis, multi-energy systems composed of electrical power, natural gas, heating power, cooling power networks and energy storage are attracting more attention and are being developed rapidly in recent years. Traditionally, different energy infrastructures are scheduled and operated independently, which results in less efficient energy usage and resource wasting. Through integrating as a multi-energy system, different energy carriers can be coupled and optimized as one unit to improve overall energy utilization efficiency, reduce system operating cost, and improving solar power integration.
In this thesis, optimal scheduling methods based on machine learning and optimization techniques of a real multi-energy system, Stone Edge Farm, CA, are proposed from an economic point of view. Specifically, Random Forest forecasting model is applied and further improved with online adaptability feature to provide input for the subsequent optimization. Besides, a new two-stage optimization formulation is proposed, which help greatly reduce computation time comparing with traditional integrated methods in the literature. Thus, the scheduling of MES operation can be conducted in much shorter time interval while considering more possible future scenarios.
Simulation results suggest that the proposed scheduling methods can help quantify the daily operating cost, balance real-time power demands and PV output solar power, and achieve considerable operating cost savings by appropriately arranging and utilizing all the devices in the multi-energy system. | |
