Efficient Methods for Constraint-Handling in Evolutionary Algorithm-Based Management of Irrigation Canal Network Operations

Handling constraints in evolutionary algorithms (EAs) is a challenging issue, since an EA does not handle constraints directly. Thus, developing a heuristic that can guide the search toward feasible and good-performing solutions has become a topic of considerable interest in the field of evolutionary computation in recent years. If a heuristic is applied that is not appropriate or properly fine-tuned, the EA may converge on infeasible solutions, or on feasible solutions with poor objective function values. This paper presents a new constraint-handling technique based on stochastic tournament selection (STS). The performance of this technique is compared to various other constraint-handling techniques applied to a genetic algorithm- (GA-) based irrigation canal management problem. The problem involves finding an efficient strategy of irrigation canal network operations that can minimize the total water consumed while preventing problems such as water shortage and flooding. The most common way to handle constraints in EAs is to incorporate penalty functions into the fitness criterion to degrade the fitness of solutions that violate one or more constraints. Alternative techniques have also been reported in the literature, including multiobjective optimization techniques that treat the constraints in single objective problems as additional objectives and stochastic techniques. This paper investigates the performance of five constraint-handling techniques (four techniques from the literature, and the new proposed STS technique). Three scenarios of the case study, with different levels of difficulty and different numbers of constraints, are tested. A comparison of the performances is made to help provide guidelines for the most promising constraint-handling techniques for EA-based management of water supply and irrigation canal operations, based on the level of difficulty of the problem.