The throughput prediction of highway service areas is an essential method for enhancing operational efficiency and service satisfaction. Traditional throughput prediction models are based on time series forecasting. However, these models fail to consider the interdependencies between service areas. This paper proposes a Gravity Model Quadratic Assignment (GMAAN) method. Building upon traditional time series forecasting techniques, GMAAN utilizes the gravity model to reassign service area vehicle flow on highways and swiftly determines parameters through a meta-learning approach. Comparisons with three years of throughput data from highway service areas in Zhejiang Province demonstrate that the redistribution effect achieved by GMAAN outperforms traditional time series forecasting methods, demonstrating a MAPE reduction exceeding 5% compared to baseline models while maintaining real-time inference efficiency.