Energy system operation and planning with data analytics : toward a reliable, economic, and environmental-friendly future

Abstract

The worldwide push for decarbonization of energy systems is largely expected to be accomplished by the high penetration of renewable energy resources (RESs) and end-use electrification. Meanwhile, with the developments of data storage devices and computing ability, massive data are available and data analytics, the science of analyzing data to draw meaningful insights, provides with us new opportunities for situation awareness, prediction, and decision-making. Therefore, it is critical to investigate how data analytics can benefit us in energy system operation and planning for a reliable, economic, and environmental-friendly future.

On the operation side, the RESs introduce huge uncertainties and raise the need for operational flexibility. In this context, more and more utilities take actions to increase time granularity in daily operations but encounter the “fast computation or a good balance” dilemma. This is because a good economic-environmental balance, which is a point on a Pareto front (PF) of a multi-objective problem, is inherently time-consuming to calculate. We solve the dilemma from the following two aspects: On the solution method side, we propose a new multi-objective algorithm, basis changing boundary intersection, to improve computational efficiency. Specifically, a data-based PF shape awareness method is proposed to identify effective searching directions and thus reduce computational burden; On the modelling side, we establish an intraday offline-online coordination (IOOC) dispatch framework, where the time-consuming PF calculation is moved to offline. We also provide rigorous applicable conditions under which the offline data can safely describe the online situations despite the forecast errors.

On the planning side, the growing penetration of RES and end-use electrification cause two issues in transmission network expansion planning (TNEP): First, a crisis of data credibility, because the climate change will largely affect the RES output and load demands, making historical data not qualified for future planning. To solve this issue, we establish a computationally efficient index to quantify the effects of climate change using practical data from international authorities. Then, a climate-adaptive TNEP is developed to improve operational security under the climate change while reducing investment costs. Second, a huge computational burden, because the diversity of operational scenarios grows significantly as the pace of decarbonization keeps accelerating, driving TNEP to consider far more scenarios than ever. we propose a physics-informed deep k-medoid clustering method to get a subset of the scenarios and thus develop a computationally efficient reduced TNEP problem. Specially, we give analytical sufficient conditions (SCs) under which the reduced TNEP problem can have the same line addition decision as the full TNEP problem. We show the incapacity of the pure physical-based clustering method to satisfy the SCs and justify the use of a data-based method, termed deep k-medoid clustering network (DKCN). To train the DKCN, a physics-informed loss function is developed where the physical insights are derived from the SCs. In conclusion, this thesis investigates the potential of data analytics in energy system operation and planning from different viewpoints and explores the applications.