为解决民航运输航空器火灾风险管理体系不健全、风险评价方法较主观、片面的问题，针对风险因素指标的分析与评估提出一种基于随机森林集成算法的综合风险评价方法。首先，根据“人-机-环-管”系统（man machine environment management, MMEM）理论以近20年的航空器火灾事故原因调查情况建立风险指标体系；其次，采集相关领域研究学者专家对指标之间关联关系的评分，采用网络分析（analytic network process, ANP）进行评价分析获取主观性的风险指标权重结果；统计数据库内20年中重大航空器火灾事故的致因情况并分类计算其先验概率，采用贝叶斯网络（Bayesian network, BN）动态分析方法进行逆向推理，获得各风险因素的概率分布情况；最后，引入随机森林算法综合分析指标主客观赋权情况，建立随机森林回归模型得到特征指标预测值及重要度，为运行单位进行火灾风险控制提出科学有效的建议。研究结果表明，5项指标安检遗漏危险品、隐患未及时排除、零部件故障、鸟击、地表温度过高是航空器火灾事故中最关键的风险因素。

In order to solve the problems of imperfect fire risk management system and subjective and one-sided risk assessment methods of civil aviation transport aircraft, a comprehensive risk assessment method based on random forest ensemble algorithm was proposed for the analysis and evaluation of risk factor indicators. Firstly, according to the man machine environment management(MMEM) theory, the risk index system was established based on the investigation of the causes of aircraft fire accidents in the past 20 years, and then the scores of scholars and experts in related fields on the correlation between the indicators were collected, and the subjective risk index weight results were obtained by using the analytic network process(ANP) method. The causes of major aircraft fire accidents in the database in the past 20 years were counted and their prior probabilities were calculated by classification, and the Bayesian network(BN) dynamic analysis method was used for reverse reasoning to obtain the probability distribution of each risk factor. The random forest regression model was established to obtain the predicted value and importance of the characteristic index, and put forward scientific and effective suggestions for the fire risk control of the operating unit. The results show that the five indicators of missing dangerous goods in security inspection, failure to eliminate hidden dangers in time, component failure, bird strike, and high surface temperature are the most critical risk factors in aircraft fire accidents.