Traffic accidents occurred frequently on roads, which bring huge losses to society. The purpose of this article is to extract the important influence factors of traffic accidents, reduce the probability of road traffic accidents, and support the basis for the decision-making of unmanned vehicles. For all this, an improved artificial neural network method is proposed to predict and analyze the severity of vehicle accidents. The impact of different factors, such as road, weather, road surface, time, etc., on the traffic accident is huge. Hence, the factors of traffic accidents are quantified and the redundancy between factors is reduced. To obtain the relationship between influence factors and severity, a multiple-input and multiple-output neural network prediction model is constructed. The hyperparameters of the network model are optimized, and the model prediction efficiency is improved. Besides, the correlation between factors and severity is also calculated, through which the impact of factors on the accuracy of the accident model is explored. To verify the validity of the method, the accuracy of the proposed method is counted and analyzed based on the data of more than 10 000 traffic accidents. The results show that the accuracy of the proposed method is higher than other traditional methods and model has high stability. The findings provide effective support to improve safety for unmanned system.

中文翻译：

---

无人系统安全决策支持：道路交通事故分析与评估

道路交通事故频发，给社会带来巨大损失。本文的目的是提取交通事故的重要影响因素，降低道路交通事故的可能性，为无人驾驶车辆的决策提供依据。为此，提出了一种改进的人工神经网络方法来预测和分析车辆事故的严重性。道路，天气，路面，时间等不同因素对交通事故的影响是巨大的。因此，交通事故的因素得以量化，并且减少了因素之间的冗余。为了获得影响因素与严重性之间的关系，构建了多输入多输出神经网络预测模型。优化了网络模型的超参数，从而提高了模型预测效率。此外，还计算了因素与严重性之间的相关性，从而探讨了因素对事故模型准确性的影响。为了验证该方法的有效性，根据超过1万起交通事故的数据对所提方法的准确性进行了计数和分析。结果表明，该方法的准确性高于其他传统方法，模型具有较高的稳定性。这些发现为改善无人驾驶系统的安全性提供了有效的支持。基于10000多起交通事故的数据对所提方法的准确性进行了统计和分析。结果表明，该方法的准确性高于其他传统方法，模型具有较高的稳定性。这些发现为改善无人驾驶系统的安全性提供了有效的支持。基于10000多起交通事故的数据对所提方法的准确性进行了统计和分析。结果表明，该方法的准确性高于其他传统方法，模型具有较高的稳定性。这些发现为改善无人驾驶系统的安全性提供了有效的支持。