Abstract
Distributed Energy Systems (DES) can play a vital role as the energy sector faces unprecedented changes to reduce carbon emissions by increasing renewable and low-carbon energy generation. However, current operational DES models do not adequately reflect the influence of uncertain inputs on operational outputs, resulting in poor planning and performance. This paper details a methodology to analyse the effects of uncertain model inputs on the primary output, the total daily cost, of an operational model of a DES. Global Sensitivity Analysis (GSA) is used to quantify these effects, both individually and through interactions, on the variability of the output. A Mixed-Integer Linear Programming model for the DES design is presented, followed by the operational model, which incorporates Rolling Horizon Model Predictive Control. A subset of model inputs, which include electricity and heating demand, and solar irradiance, is treated as uncertain using data from a case study. Results show reductions of minimum 25% in the total annualised cost compared to a traditional design that purchases electricity from the centralised grid and meets heating demand using boilers. In terms of carbon emissions, the savings are much smaller, although the dependency on the national grid is drastically reduced. Limitations and suggestions for improving the overall DES design and operation are also discussed in detail, highlighting the importance of incorporating GSA into the DES framework. Introduction The energy sector worldwide has taken on a transformative approach to meet current and future demand by increasing the proportion of renewable and low-carbon energy resources. In its wake, Distributed Energy Systems (DES) have attracted both large and small-scale customers , seeking to take advantage of government incentives whilst reducing greenhouse gas (GHG) emissions, distribution costs, and energy losses associated with transmission over long distances from power plants to consumers [1]. DES are the building block of holistic smart energy systems [2], as they can be connected and integrated into the national grid, with capabilities of introducing internal utility networks. Optimisation-based models are often employed for the design and operation of such systems, where the main aim is the minimization of its total annual cost considering the utilization of renewable energy generation technologies, such as solar panels, wind turbines, fuel cells, electrolysers, hydrogen tanks, etc. [3,4]. In this paper, a two-stage approach, which involves the optimisation of a DES design for a specified location or scale, and the subsequent optimisation of its operation based on the recommended structure [5], is implemented. The design includes the types, capacity and potential location for installation of the various generation and storage technologies to be used during the operation. DES, like most energy systems, have nonlinear characteristics, complex to model and time-consuming to solve [6]. This often demands a trade-off between complexity, accuracy, and solution time [7,8] for the models to be suitable for real-time operation. Therefore DES models are commonly linearized (or formulated as approximate linear models in the first place), to benefit from the efficiency of Mixed-Integer Linear Programming (MILP) solvers [9]. Although arguably more accurate, Mixed-Integer Nonlinear Programs (MINLP) are computationally very expensive and cannot ensure global optimality [7], guaranteed by MILPs. Linearization inevitably introduces differences between prediction and actual system performance for both the design and operational