Abstract: In recent years, the number of motor vehicles has been increasing and traffic accidents have occurred frequently, making accident analysis an important part of ensuring traffic safety. Among them, the data analysis of the event data recording (EDR) system is particularly crucial. This study aims to alleviate the problems of heavy workload and low efficiency in traditional manual analysis. Using partial EDR table data of the involved vehicle as samples, based on the DeepSeek model, a "main task–sub task–micro task" prompt method is adopted to analyze the accident process in combination with information such as airbag status, vehicle speed, acceleration, etc.Then, by combining the changes in speed and acceleration points, driver operation, and various system data of the vehicle, the factors affecting the changes in driving status are identified. The results show that the vehicle accelerated from an initial speed of about 60 km/h to 180 km/h before colliding, with collision speeds of 171 km/h, 122 km/h, and 71 km/h respectively. Adaptive cruise control (ACC) is activated before the accident, and the driver accidentally stepped on the accelerator pedal throughout the entire process, resulting in almost no braking during the accident. This result is consistent with reality, indicating the feasibility of DeepSeek technology in EDR data analysis and providing direction for more specific analysis in the future.

Key words: DeepSeek; traffic accident; data analysis; artificial intelligence

摘要： 近年来机动车保有量不断增加且交通事故频发，使得事故分析成为保障交通安全的重 要环节，其中事件数据记录（EDR）系统的数据解析更是至关重要。为缓解传统人工分析时 工作量大、效率低的问题，以肇事车的部分 EDR 表格数据为样本，基于 DeepSeek 模型，采 用“主任务－子任务－微任务”的提示方式，结合气囊状态、车辆速度、加速度等信息对事 故过程进行了分析；然后结合对通过速度、加速度变化点、驾驶员操作和车辆各系统数据， 明确行驶状态变化因素。结果显示，车辆从初始约 60 km/h 加速至 180 km/h 后发生碰撞，依 次碰撞时速分别为 171、122、71 km/h。事故前自适应巡航（ACC）激活，驾驶员全程误踩加 速踏板，事故中几乎未制动。该结果与实际相符，说明了 DeepSeek 技术在 EDR 数据解析中 的可行性，可为后续更具体分析提供方向。

关键词: DeepSeek；交通事故；数据解析；人工智能