Development of Evolutionary Design of Experiments to Find Significant Effects for Natural Disaster Prevention

For example, in Tsunami evacuation simulation for evaluating the routing factors, the number of simulation trials will increase exponentially, if we will take all possible combination of factors. It is strongly required that reducing the number of simulation run since such simulation based researches will be done in a local community. While it is well known that evolutionary computation will help us in that case if we are able to select a suitable fitness function. Thus, we consider an application of genetic algorithm based search methodology to find significant factors in this paper. The proprieties of tsunami evacuation simulation results are shown and a benchmark problem which has a similar proprieties of the actual simulation results is also introduced. And then, a search algorithm called "Evolutionary Design of Experiments" is introduced and the effectiveness is evaluated by using the benchmark problems. From the numerical simulation results, the proposed method can find significant factors with the smaller number of simulation executions although the proposed algorithm can not find enough significant factors when it was applied to actual tsunami evacuation simulation results. In the session of discussion, the difference of the properties between actual simulation results and benchmark problem is shown to improve the proposed method.

I. Introduction

A practical goal of the social problem like an evacuation simulation from earth quake trigged Tsunami is to find more sensitive factors which strongly affect the evacuation time. To the best of our knowledge, a study in which all potential factors are considered and more significant factors are to be found via social simulation executions is novel. Actually, many studies have explored significant factors via real data analyses without numerical simulations. However, because real disaster data can be obtained only when the disaster occurs in the real world, it is impossible to find significant factors from the real data in order to reduce the disaster damages. On the other hand, many studies have estimated the effectiveness of countermeasures against the damages by using numerical simulation with desirable conditions. For instance, Mas et al. integrated a model to treat Tsunami inundation as physical phenomena and evacuation as social phenomena in order to observe the behavior of individuals in the complex process of tsunami evacuation [1]. Moreover, they discussed the relationship between the start time decision of evacuation from a tsunami and the probability of evacuation success. However, since such a countermeasure is intentionally chosen among many factors, it is not enough to find the better and worse conditions.