Increased availability of high resolution metered consumption data shows clear spatio-temporal variability in energy demand, both in terms of magnitude and time. This variability is rarely captured in district energy modelling and optimisation. In this paper, we demonstrate a modelling approach that integrates the stochastic variability of energy demand in energy system optimisation. In our set-up, energy demand is a stochastic function over time, separated into weekdays and weekends in a year. We consider cooling and electricity as end-uses. We implement the district energy optimisation using the mixed integer linear programming (MILP) Scenario optimisation (SO) framework. The stochastic variability of hourly demand is represented by 500 scenarios for 24 typical days in the year. For computational efficiency, we implement a scenario reduction step, resulting in 16 reduced scenarios as representative of the full scenario set. These 16 scenarios are used to formulate an SO model for a group of office buildings in Bangalore, India. The objective in this model is to minimise the Conditional Value at Risk (CVaR) associated with each scenario, weighted by the probability of that scenario being realised. A scenario can have some demand unmet, but this will incur a financial penalty. To better understand the necessary parametrisation of the model, the penalty for unmet demand is tested by sensitivity analysis.