A Deep Reinforcement Learning Approach to Solving the Digital Billboard Optimal Placement Problem Based on Multi-Source Spatiotemporal Data Fusion

Abstract

The study presents a study on optimizing the spatial location of digital billboards in urban environments using multi-source big data and deep reinforcement learning methods. Focusing on the Fifth Ring Road in Beijing, China, the study aims to solve the Maximum Coverage-Digital Billboard Localization Problem (MC-DBLP) model. This paper describes the data collection and preprocessing process, the use of geo-detectors and attention models, and an evaluation of the effectiveness of deep reinforcement learning methods compared to traditional optimization solvers. The authors propose a new solution model for the digital billboard placement problem. The method is universal and scalable, which can provide a reference for similar problems in other fields. Multi-sourced spatiotemporal data are incorporated to unify the data format of factors and, an attention model multi-decoder (AMMD), Gurobi solver, and a heuristic algorithm are used to solve the problem. The results show that the Gurobi solver produces the best solution results, and the optimal digital billboard locations are obtained. In the future, the authors plan to select more various and complex influencing factors and explore more efficient and accurate algorithm models to better solve location optimization problems.