Traffic Flow Prediction Using DeepSeek Fine-Tuning and Dynamic Modeling

交通流预测是智能交通系统中的关键任务，对城市规划和交通管理具有重要意义。传统深度学习方法虽然提升了预测精度，但其可解释性较差，且常规大语言模型难以捕捉交通流的复杂时空动态特性。为此，本文提出了一种基于DeepSeek大语言模型微调与动态建模相结合的交通流预测方法(DynaSeek)。将动态建模技术和数据修正机制嵌入交通流预测任务中，量化时空因素(如地区、天气)对交通流的影响，采用Lora微调策略对DeepSeek模型进行优化，最后模型在预测时结合历史数据和实时时空信息进行动态修正。结果表明，本文方法在加利福尼亚州多模态数据集上表现优于基线模型，且在可解释性方面提供了更清晰的交通流变化规律。本文为交通流预测提供了一种新的多维度数据融合与动态建模框架，显著提升了模型的实用性和可靠性。Traffic flow prediction is a key task in intelligent transportation systems and is of great significance to urban planning and traffic management. Although traditional deep learning methods have improved prediction accuracy, their interpretability is poor, and conventional large language models are difficult to capture the complex spatiotemporal dynamic characteristics of traffic flow. To this end, this paper proposes a traffic flow prediction method (DynaSeek) based on the fine-tuning of the DeepSeek large language model and dynamic modeling. The dynamic modeling technology and data correction mechanism are embedded in the traffic flow prediction task, and the impact of spatiotemporal factors (such as region and weather) on traffic flow is quantified. The DeepSeek model is optimized using the Lora fine-tuning strategy. Finally, the model is dynamically corrected by combining historical data and real-time spatiotemporal information during prediction. The results show that the proposed method performs better than the baseline model on the California multimodal dataset and provides a clearer law of traffic flow changes in terms of interpretability. This paper provides a new multi-dimensional data fusion and dynamic modeling framework for traffic flow prediction, which significantly improves the practicality and reliability of the model.