Genetic Hybrid Optimization of a Real Bike Sharing System

In recent years there has been a growing interest in resource sharing systems as one of the possible ways to support sustainability. The use of resource pools, where people can drop a resource to be used by others in a local context, is highly dependent on the distribution of those resources on a map or graph. The optimization of these systems is an NP-Hard problem given its combinatorial nature and the inherent computational load required to simulate the use of a system. Furthermore, it is difficult to determine system overhead or unused resources without building the real system and test it in real conditions. Nevertheless, algorithms based on a candidate solution allow measuring hypothetical situations without the inconvenience of a physical implementation. In particular, this work focuses on obtaining the past usage of bike loan network infrastructures to optimize the station’s capacity distribution. Bike sharing systems are a good model for resource sharing systems since they contain common characteristics, such as capacity, distance, and temporary restrictions, which are present in most geographically distributed resources systems. To achieve this target, we propose a new approach based on evolutionary algorithms whose evaluation function will consider the cost of non-used bike places as well as the additional kilometers users would have to travel in the new distribution. To estimate its value, we will consider the geographical proximity and the trend in the areas to infer the behavior of users. This approach, which improves user satisfaction considering the past usage of the former infrastructure, as far as we know, has not been applied to this type of problem and can be generalized to other resource sharing problems with usage data.