Multicore Parallel Genetic Algorithm with Tabu Strategy for Rainfall-Runoff Model Calibration

Abstract

Conceptual rainfall-runoff models (CRRMs) are widely used for flood forecasting and hydrologic simulations. However, parameter calibration poses a major challenge for using CRRMs, especially given that climate changes and effects of human activities often necessitate recalibration of CRRMs. Genetic algorithms (GAs) are one of the most widely used optimization techniques for hydrological model calibration, and have been widely and successfully used in model calibration; however, the complexity and high dimensionality of parameter calibration make them time-consuming and prone to local optima. Moreover, repetitive computation of fitness values in the GAs greatly reduces the efficiency. Therefore, new methods must be explored to improve the computational efficiency. Multicore parallel technology, which enables resource sharing and has low parallel costs and computing burdens, offers considerable benefits for parameter calibration. Thus, a multicore parallel genetic algorithm (MCPGA) based on the fork–join parallel framework with a tabu strategy is proposed in this paper for CRRM calibration. The method uses a multicore to divide the original task into several small subtasks and complete them concurrently, and adopts tabu strategy to avoid redundant computation in the GA to enhance the computing efficiency. The current methodology is applied to parameter calibration for the Xinanjiang model (which is a typical CRRM and has extensive applications in humid and semi-humid regions in China) of the Shuangpai Reservoir. Result comparisons between the MCPGA and serial GA indicate that the proposed method ensures a high degree of accuracy and significantly improves the computational efficiency.