Damage to road network links caused by a large-scale disaster is considered to be a problem because it can cause serious life-threatening problems, such as isolation of the affected area and delay in relief efforts. Therefore, in order to create a robust network that ensures connectivity between locations even in the event of a disaster, it is essential to develop roads with disaster resistance that takes into account "selection and concentration". In this study, we develop a model for optimizing the maintenance order by combining deep reinforcement learning with the "disaster prevention function evaluation of roads" used by the Ministry of Land, Infrastructure, Transport and Tourism to evaluate road maintenance. As a result of applying the model to a simple network, it was confirmed that the model can search for an effective maintenance sequence for the total number of maintenance sequence patterns with less search than the total search. In the future, it is hoped that the introduction of a maintenance period and application to large-scale networks will enable us to make proposals that aim to achieve a disaster-resistant effect as soon as possible within a limited budget.
