To comprehensively consider the actual spatial-temporal transfer process of electric vehicles (EVs) and enhance the computation efficiency of scheduling, this article proposes a spatial-temporal transfer model of EVs and an improved Lagrange dual relaxation method (ILDRM) for the decentralized scheduling of a charging-storage station (CSS). Specifically, with the application of trip chain technology, Monte Carlo, and Markov decision process (MDP), the spatial-temporal transfer model of EVs is constructed, taking into account multiple factors including temperatures, traffic conditions, and transfer randomness. Subsequently, by introducing ILDRM, a decentralized optimization model is proposed which converts the traditional centralized optimization model into a set of sub-problems. Moreover, the optimization model aims to maximize the profit of CSS under the constraints of vehicle-to-grid behavior and the operation of both CSS and distribution network. To validate the proposed spatial-temporal transfer model and the decentralized optimization method for CSS, a series of simulations in various scenarios are performed regarding the load curve and computation efficiency. The comprehensive and systematical study indicates that the proposed spatial-temporal transfer model enables to reflect EVs transfer randomness and it is more factually practical than the classical Dijkstra algorithm. Besides, ILDRM can provide a high computationally efficient solution to the operation of CSS.