It has been a scientific consensus that the climate-warming trends due largely to the release of large amount of ‘greenhouse gas’; such as carbon dioxide, which is produced by boiling fossil fuel to provide energy. One effective way to reduce greenhouse gas is to reduce energy on the demand side by improving the energy efficiency of facilities, such as cooling systems of buildings.

Extensive electric power is required to maintain indoor thermal environments using heating, ventilation, and air conditioning (HVAC) systems. The water-cooled chiller plants of HVAC systems consume more than 50% of the total electric power. Adjustments to the interactions between components could reduce the total power used, in the premise of helping systems to meet building thermal load targets. Studies have shown that significant energy efficiency can be improved by adopting different kinds of optimisation techniques to different variables, namely supervisory optimisation control strategies. The controlled variables are usually optimised according to the instant building cooling load and ambient wet bulb air temperature at regular time intervals. In this way, the energy efficiency of chiller plants can be improved. However, with an inherent assumption that the instantaneous building cooling load and ambient wet bulb temperature remain constant in a coming time interval, the energy efficiency potential has not been fully realised, especially when cooling loads vary suddenly and extremely during the coming time interval.

To solve this problem, a dynamic optimisation framework, which incorporates cooling load forecasting into supervisory optimisation, has been proposed. Rather than trying to minimise the instantaneous system power according to the instantaneous building cooling load and ambient wet bulb temperature, the controlled variables are derived to minimise the sum of the current system power and the possible future system power according to both current and forecasted future building cooling loads and weather conditions.

To make the proposed dynamic optimisation framework successful, a robust and accurate cooling load forecasting model is necessary. Numerous studies have been carried out to develop cooling load forecasting models, and great achievements have been made. However, limitations in their applicability persist because most previous models are calendar- and time-based data-driven models, which may fail when unexpected issues occur, or special schedules are adopted. What’s more, the inputs that were selected passively from the available source data pools rather than via active exploration may be insufficient and impair the accuracy of forecasting models. A novel cooling load forecasting model is developed to overcome these drawbacks; it combines an artificial neural network with an ensemble approach. Based on physical principles other than the available data source, the inputs are explored actively and are independent from either calendar or time indicators, which makes the forecasting model can dealing with irregular occasions and unexpected schedules with high accuracy.

Another aspect of ensuring the success of the proposed dynamic optimisation framework is the application of optimisation. Frequent simultaneous updates of different variables with large fluctuations impede operational stability of chiller plants and make it difficult to apply optimisation strategies into the industry. In practice, making fewer and more moderate variable updates is a more practical optimisation strategy. To address this problem, this thesis proposes using varying searching bounds in the optimisation processes. Also, the contribution of each variable to power reduction is analysed to determine the priority of optimised variables. The results show that applying varying searching bounds is an effective way to avoid frequent extreme fluctuations in optimised variables without sacrificing the optimisation performance. The optimisations of the condenser water mass flow rate and condenser water supplying temperature make greater contributions to power reduction.

Incorporated the novel cooling load forecasting model with the improved optimisation process, the proposed dynamic optimisation framework is validated by a case study. Through which, it is proved that the energy efficiency potential of a chiller plant can be further released without shortening the operation time interval by a dynamic optimisation, compared with the traditional optimisation strategy. 80% of the redundant energy consumption has been reduced; tremendous energy can be saved for chiller plants that work for years, especially whose cooling loads vary suddenly and extremely.