A data-centric stochastic model for simulation of occupant-related energy demand in buildings

Abstract

If greenhouse emission reduction targets are to be met worldwide, not only will there need to be major investment in decarbonisation of the electricity supply network, but there will also need to be a significant reduction in energy demand. The built environment offers opportunities for demand reduction that can help to achieve the necessary targets. For buildings yet to be built there is a real opportunity to design low-energy environments, but the existing building stock will largely still be in existence in 50 years and so retrofit options must also be explored. No matter how efficient a building it is the occupants that drive the energy consumption - whether by requiring comfortable conditions or by using electrical equipment. While in the residential sector owner-occupiers have particular responsibility for consumption, in the non-domestic sector the financial responsibility may not lie with the building occupants and hence it is harder to target demand reduction interventions. In addition, while in a residential building the behaviour of the individual has a direct and significant impact on energy consumption, in a non-domestic building it is the collective behaviour that is important to understand. The impact of the building occupants on internal loads is critical to assessing the energy efficiency of a design or retrofit. Building energy simulation offers a means to assess the potential benefits of different options without requiring costly in-situ tests. In order for the approach to be viable, however, a simulation needs to demonstrably replicate the building performance. This has proved to be difficult not only pre-construction but even for operational demand, in part because individual and collective occupant behaviour is difficult to quantify. Typically, building energy simulation packages require occupant-related internal loads to be input into the simulation via a deterministic schedule consisting of a peak daily demand and a diversity schedule that describes how the demand varies over a 24hr period. The stochastic nature of occupant-related energy demand is well known but as yet there is no accepted methodology for generating stochastic loads for building energy simulation. A new approach is required. The aim of this thesis is to develop a new model for the definition of occupant-related building internal loads for input into building energy simulation. Early studies showed that a model must not only be able to generate good estimates of the key parameters of interest with a measure of the uncertainty, but must also be able to assimilate data, be able to simulate operational change and be straightforward to use. All buildings generate monitored data of some form, even if it is just monitored consumption for purposes of billing. Since the start of the century there has been a rapidly increasing pool of monitored data at increasing time and spatial resolutions for both residential and non-domestic buildings. Increasing monitoring of electricity consumption generates an opportunity to gain an in-depth understanding of the nature of occupant-related internal loads. The requirement for a model to be able to assimilate these data make a data-centric model a natural choice. This study focuses on non-domestic buildings and the collective stochastic behaviour of the occupants as evidenced by monitored plug loads and lighting demand. Using monitored data from four sub-metered buildings across the Cambridge University building stock a functional data analysis approach has been used to extract the underlying structure of the data in a way which facilitates generation of new data samples that encompass the observed behaviour without replication. A key assumption in simulation of non-domestic buildings is that the internal loads are in some way related to the activity that takes place in a building zone. This is problematic both because the definition of activity is indeterminate and because building sub-metering strategies rarely align with the specified activities. Deconstruction of the data allows exploration of this fundamental assumption and leads to the conclusion that activity per se is not a good indicator of internal loads. Instead, for plug loads it is the expected variability of the data that is important, whereas for lighting the control strategy of each individual building zone defines the stochasticity of the demand. The model has been developed into a practical online tool for generation of plug loads and lighting demand in the form of annual hourly time histories of internal load that can be input directly into a building energy simulation. As a design tool the modeller can select an expected level of variability in demand and use estimated base load and load range to generate synthetic demand profiles. The beauty of the approach is that if monitored data are available - for example when optimising retrofit designs - the data can be used to generate synthetic time histories that encompass observed demand but can also be modified to account for operational change - a reduction in minimum daily demand for example. Finally this thesis suggests a potential alternative to the activity-based deterministic approach for the specification of occupant-related building internal loads. Rather than generating loads for each new simulation on a case by case basis, the suggestion is to use an approach similar to that used for the specification of weather data - another stochastic input. The proposal is to create annual hourly stochastic samples of typical demand according to the expected variability. These would be used with user-defined energy use intensity values with scenarios for extreme demand in much the same way that typical and future weather scenarios are modelled. The methodology presented here is one such way to generate annual hourly stochastic sample data and provides an initial step towards the specification of such typical load profiles.