Please use this identifier to cite or link to this item:
|Title:||Holistic optimization framework for the operation of urban energy systems||Authors:||Chiam, Zhonglin||Keywords:||Engineering::Industrial engineering::Operations research
Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
|Issue Date:||2020||Publisher:||Nanyang Technological University||Source:||Chiam, Z. (2020). Holistic optimization framework for the operation of urban energy systems. Doctoral thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/147327||Abstract:||The push for greater sustainability in urban energy use has led to increasingly complex systems - favoring greater extents of centralization and integration. Centralization of energy systems offers unrivaled levels of energy efficiencies and cost-effectiveness, a consequence of economies of scale. Shared infrastructures from centralized systems, further act as an enabler for facilitating the integration of alternative energy sources. These sources include renewable energy and the harvest of waste energy. The incongruity between the design and operating conditions of these complex energy systems will always persist; it is inevitable to make assumptions given data limitations. When these conditions differ too vastly, the expected gains could easily be negated. Measures to improve the operation of urban energy systems could be explored through mathematical optimization on the appropriate models. Components of urban energy systems are often operated at predefined setpoints or independently optimized - a repercussion of complexity. Localized control strategies inhibit the ability to adapt well under less-than-ideal scenarios, through the disregard of cascading effects on the system. Formulation of optimization problems that concurrently express these systems in its tuneable variables and captures the tight-coupling between the numerous components, i.e., holistically, usually results in a mixed integer non-linear program which is large and difficult to solve. To address this issue, abstraction techniques of judiciously selected models enabled the accompanying optimization problem to be formulated as a mixed-integer linear program. However, this technique has limited applications as it is highly reliant on case-study specific information. For generic purposes, a hierarchical optimization framework for the operation of urban energy systems is introduced in the current work. In this framework, model abstraction techniques are first applied so that the resultant problem could be solved using a combination of a genetic algorithm and a mixed-integer linear program solver. The metaheuristic is introduced to handle important decision variables that cannot be linearized. Using the mixed-integer linear program in tandem with the genetic algorithm effectively alleviates the computational effort, by reducing the search space of the latter. Since the reliance on the metaheuristic is reduced, the likelihood of achieving global optimality is increased. The existence of energy storage systems, sanctions for energy to be stored in more favorable periods to be available later, thus enhancing cost-effectiveness and/or energy efficiency. Multi-period considerations demand optimization across several periods, burgeoning the size of the problem. Thus, the sliding-window technique is used in tandem with the hierarchical framework. This technique permits the trade-off between solution accuracy and problem size, hence resolution time and solvability. When the case-specific, size of the window (number of periods) is suitably chosen, the sacrifice in accuracy becomes justifiable, for the above-mentioned reasons. Genetic algorithms are population-based metaheuristics and can require a considerable amount iterations to converge. This issue is further exacerbated in the framework where a mixed-integer linear program has to be solved in each iteration. For online applications, the genetic algorithm was substituted with a reinforcement learner. Implementation of the reinforcement learner allows part of the optimization load (learning) to be shifted offline, hugely speeding up the resolution time, when deployed online. Once trained, the framework only requires a single iteration to generate close-to-optimal solutions. Finally, the three versions of the optimization framework were applied to case studies for illustration purposes. These case studies were based on an existing district cooling system and ground-coupled heat pump system serving a building. Results generated suggest substantial energy and cost savings of up to 31.9% and 12.7% respectively, when the operations were optimized for given energy demand.||URI:||https://hdl.handle.net/10356/147327||DOI:||10.32657/10356/147327||Rights:||This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).||Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Theses|
Updated on May 28, 2022
Updated on May 28, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.