Application of artificial neural network and genetic algorithm to the optimization of load distribution for a multiple-type-chiller plant

Abstract

Numerous methods are available in various studies that investigate the optimization of central air conditioning systems. However, the majority of these studies focus on a single chiller or multiple chillers of identical capacity and type. Although the data indicated substantial energy conservation, all relevant findings were optimized under continuous concave functions, which cannot resolve the optimal load distributions of multiple-chiller plants commonly seen today due to their discontinuous properties. This study targets a large public exhibition venue, which uses a central air conditioning system consisting of heterogeneous chillers: fixed /variable frequency screw chillers, and centrifugal chillers. This system adopts a manual method (MM) for controlling the on/off operation of chillers; after an extended period of monitoring and collecting data, the operational characteristics of each chiller are better understood. This study uses the neural network (NN) to establish a chiller power consumption model, which considers the operating constrains of various types of chillers to ensure stable operation of the chiller plant. Genetic algorithm (GA) is also integrated while satisfying cooling load conditions to optimize chiller loading and determine the minimum power consumption of chiller plant. This is achieved by overcoming the deficiency of the traditional method, which fails to converge at low loads, and liberating from the restrictions limited by discontinuous functions. The results show that integrating NN and GA (NNGA) between 95% and 55% cooling loads improved power savings compared to MM load distributions, saving total power consumption by approximately 14.55%. A simulation data analysis for a one-year operation period showed that the proposed method can save up to 16.68% in power consumption over the MM method on a monthly basis and is conclusively superior to the MM controlled chiller model. This proved that the NNGA method proposed in this study provides an effective solution for optimizing load distributions of common multiple-chiller plants in parallel operation.