Optimized Design and Energy Management of Heating, Ventilating and Air Conditioning Systems by Evolutionary Algorithm

In the study of heating, ventilating and air conditioning (HVAC) systems, it is becoming increasingly common to adopt a combined simulation-optimization approach to handle the design and energy management problems. Due to the large variety of mathematical models and the complexity in solving the differential-algebraic equation set, it is normal to use a commercial plant simulation package for modeling purpose. The plant simulation models developed can be a useful tool to study “what if” variations to the system parameters. However if a large number of problem variables are involved, this would be demanding and reliant on the intuition of the operator to consider different scenarios. In this case, suitable optimization techniques can be introduced in order to ease the workload and determine reliable and optimal solution. However due to the typically multimodal, multidimensional, nonlinear, mixed continuous-discrete, and highly constrained nature of HVAC problems, this renders to the traditional analytical and gradient-based optimization methods not effective in these cases. As a result, other numerical and heuristic optimization methods have been considered. Among such methods, evolutionary algorithm (EA) has been investigated for HVAC problems by researchers in the last decade. Although there are primarily three paradigms in EA – genetic algorithm, evolutionary programming and evolution strategy – only genetic algorithm has been widely applied and developed for engineering applications, but few studies were concerned with the other two paradigms for HVAC problems.

In this research, a plant simulation model of HVAC system was developed using the TRNSYS simulation environment. An optimization platform, referred to EA Suite, was established using MATLAB based on the paradigms of evolutionary programming and evolution strategy. The EA Suite included the major EA operators of mutation, selection, recombination and constraint handling, as well as a coupling linkage with the TRNSYS for simulation-optimization functionality. In-depth and systematic qualitative studies were carried out to understand the performances of different choices and combinations of EA operators, in order to develop a robust EA (REA). A major problem in analysis of HVAC simulation problems was the length of time taken to evaluate the objective function by calling the related external simulation program. This meant that the number of evaluation function calls was a very significant factor in determining the search efficiency, which was not normally the case in optimization research. Finally, by analyzing a number of local HVAC design and energy management optimization problems, the REA was demonstrated to be effective in tackling real-life engineering applications where there would be limited evaluation function calls.