District energy systems are an alternative to conventional national-scale networks to meet demand in urban areas. Decentralising electricity production reduces distribution losses, while centralising thermal energy production capitalises on economies of scale. Furthermore, geographic proximity facilitates the use of waste heat from electricity production to meet thermal demand. However, it is difficult to assess which configuration of technologies will best suit a cluster of users. Mathematical optimisation techniques have been extensively researched as a method to resolve this, as they can simplify the design of investment portfolios and operation schedules for a given set of geolocated demands. However, they are not yet practically applicable.

This thesis uses the open-source, mixed integer linear programming framework Calliope to present three methodological enhancements which address model simplification, parameter uncertainty, and conflicting decision-maker objectives. These enhancements enable the practical design of district energy systems through data-centric workflows which can readily represent real system complexities in a tractable optimisation model.

This thesis first examines the impact of parameter simplification in a linear model. Piecewise linearisation is applied to nonlinear part-load energy consumption curves and a pre-processing step is developed to optimise breakpoint positioning along a piecewise curve. Second, a three-step method is proposed to handle demand uncertainty in linear models, using historical demand data. These steps are scenario generation, scenario reduction, and scenario optimisation. Using out-of-sample tests, robustness of optimal investments to unmet demand is quantified. Furthermore, system resilience to unexpected events is explored by introducing interruptions to the national electricity grid availability for a district in Bangalore, India. Scenario optimisation is extended to account for these interruptions as well as to mitigate unfavourably high levels of CO$2$ emissions in system design. Finally, this thesis identifies eight possible decision makers, who each hold a different objective in the design of a district energy system. Optimal technology configuration and out-of-sample test results are compared across all decision-maker objectives.

These methodological enhancements demonstrate the capability of optimisation models to be reflective of reality whilst being transparent concerning the impact of simplifications, uncertainty, and conflicting objectives. Calliope is extended in this thesis to be practically applicable for district energy systems. Moreover, its extensibility facilitates the continuation of development, including possible future work into data validation and spatial dimension simplification.