Joint entropy and multi-objective evolutionary optimization of water distribution networks

Tanyimboh, Tiku T. and Czajkowska, Anna M. (2018) Joint entropy and multi-objective evolutionary optimization of water distribution networks. Water Resources Management. ISSN 0920-4741 (https://doi.org/10.1007/s11269-017-1888-y)

[thumbnail of Tanyimboh-Czajkowska-WRM-2018-Joint-entropy-and-multi-objective-evolutionary-optimization-of-water-distribution-networks]
Preview
Text. Filename: Tanyimboh_Czajkowska_WRM_2018_Joint_entropy_and_multi_objective_evolutionary_optimization_of_water_distribution_networks.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo

Download (865kB)| Preview

Abstract

It is essential to consider resilience when designing any water distribution network and surrogate measures of resilience are used frequently as accurate measures often impose prohibitive computational demands in optimization algorithms. Previous design optimization algorithms based on flow entropy have essentially employed a single loading condition because the flow entropy concept formally has not been extended to multiple loading conditions in water distribution networks. However, in practice, water distribution networks must satisfy multiple loading conditions. The aim of the research was to close the gap between the prevailing entropy-based design optimization approaches based on one loading condition essentially and water distribution practice that must address multiple loading conditions. A methodology was developed and applied to a real-world water distribution network in the literature, based on the concept of the joint entropy of independent probability schemes. The results demonstrated that the critical loading conditions were design specific. In other words, the critical loading and operating conditions cannot readily be determined beforehand. Consequently, maximizing the joint entropy provided the most consistently competitive solutions in terms of the balance between cost and resilience. The results were derived using a penalty-free genetic algorithm with three objectives. Compared to previous research using flow entropy based on a single loading condition and two objectives, there was a substantial increase of 274% in the number of non-dominated solutions achieved.