Traffic data is characterized by complexity in structure, high collection frequency, large volume, and uneven spatiotemporal distribution. Efficient range querying on traffic data poses a significant challenge in intelligent transportation system research. Utilizing spatiotemporal indices to accelerate queries is an effective approach. However, traditional spatiotemporal indices suffer from issues where index efficiency is negatively correlated with data volume, and index storage size is positively correlated with data volume. To address these challenges, this paper proposes an efficient range query method for traffic data based on a learning-based spatiotemporal index. The Flood index is applied and enhanced with the following improvements: Firstly, a fully connected neural network constructs a time-cost model to predict query times and determine the optimal grid layout. Secondly, a recursive model index learns data distribution within each grid for query range prediction and proposes an index updating strategy. Experimental results using real data demonstrate that our method maintains comparable query efficiency to Flood index while reducing index storage space by a factor of 10. Furthermore, compared to baseline methods, our approach improves query speed for traffic spatiotemporal data by 3 to 8 times.