Fitness evaluation for nurse scheduling problems

When applying evolutionary algorithms to difficult real-world problems, the fitness function routinely needs evaluating for a very high number of intermediary cases. The paper is concerned with real-world nurse rostering problems with highly constrained resources. We consider a particular approach, which allows for a quick evaluation and is general enough to deal with other kinds of resource planning problems with time-related constraints. The model developed for this approach handles the constraints in a modular way and the addition of new constraints is relatively straightforward. Simple constraints (such as those affecting the personal wishes of employees) and global constraints (such as balancing the workload among people) can be formulated easily using this approach. Our approach can also handle very complex time-related constraints as well as conditions that are related to previously planned work. Moreover, it provides clear feedback about violation of constraints. The approach has been implemented successfully in a nurse rostering program entitled "Plane" which is used in hospitals all over Belgium. It can tackle a high number of specific and modifiable constraints of a very different nature. The benefits from this approach (in terms of software requirements) are small memory use and a computationally simple, single evaluation function allowing for the simultaneous rostering of several hospital wards at the same time.