Abstract: With the outbreak of the Covid-19 at the end of 2019 and the subsequent heavy blow on the civil aviation industry, in order to minimize the impact of public emergencies on the civil aviation industry and minimize economic losses in the future development, flight prediction of public emergencies has once again attracted the attention of civil aviation practitioners.
The biggest feature of public emergencies is that the time of their occurrence has great randomness, which leads to the uncertainty of the occurrence, prevention and control of the events and the degree of damage to the social economy. According to numerous historical events, major public emergencies can disrupt the normal flight operation order, or even make the entire civil aviation industry into a sustained downturn. Flight prediction under public emergencies refers to the quantitative assessment of flight change trend when disasters or disasters occur outside the scope of human expectation, which will cause or may cause serious social harm and last for a long time.
Through the analysis of the current situation, it is found that the impact of public emergencies on flights is mostly focused on qualitative management strategies, and there are few studies using data association analysis, deep quantitative mining and intelligent forecasting. Moreover, the trend prediction of traditional time series is easy to be affected by external interference, which makes it difficult to further improve the prediction accuracy. By taking public emergencies as individual guidance characteristics, ABI flight forecasting mechanism under the continuation of event life cycle is constructed to improve the anti-external interference ability of flight forecasting, change the disadvantages of traditional time series data trend forecasting, facilitate the formulation of operation strategy in advance and timely adjustment of transport capacity deployment, and reasonably ensure the safety and efficiency of flight operation.
The Agent Based Index (ABI) mechanism is introduced, public emergencies are taken as individual guiding characteristics, and ABI indexes of different operation scopes and types of flights are finally output to determine the correlation between public emergencies and flight operation trends. Then SPSS data analysis technology is used to make a statistical analysis of the impact of historical public emergencies on flights, machine learning of the Spearman correlation between flights and emergencies, and quantification of the direct positive and negative correlation between the two through correlation test, excluding irrelevant factors. Then, the flight prediction improvement model of BP neural network is established. The trend reflected by the historical data can be used as the input sample, emergencies and flight trend changes are used as the training function, and the improved Matlab.net is used to train the sample data, and the weight and threshold of each layer of neurons are constantly corrected to make the error function decline along the negative gradient direction. Flight projections close in on expectations. Finally, the historical flight data of 46 months before and after the Covid-19 outbreak are obtained through the information network of the Civil Aviation Administration of China as an example to verify the feasibility and prediction effect of the model and algorithm. In order to avoid the phenomenon of overfitting, 70% of the data is set as the training set to train the algorithm, 15% of the data is set as the verification set to verify the above training results, and the remaining 15% of the data is set as the test set to test the final model.
The results show that the computational search technique of BP neural network solves the complex nonlinear mapping relationship between event and flight to a certain extent, and the optimal verification set mean square error gets the prediction result closest to the expectation. The maximum training times are set as 1000 times and the autoregressive end as 3.The BP training mechanism for flight prediction based on the life cycle of public emergencies proposed in this paper is used. The running code in Matlab platform is iterated for 23 times and the prediction results of domestic, international and passenger and cargo flights in the next 8 months are obtained. With the small outbreak of the Covid-19 after the National Day holiday in 2020, flight data decreased in November and December 2020, and the total number of flights per month is expected to be about 300,000. In the next four months after January 2022, the total number of flights per month will remain at around 400,000.
Due to the limitation of data acquisition, only domestic epidemic data are selected as independent variables to participate in the correlation analysis, and the impact of foreign epidemic changes on Chinese international flights is not counted. In order to avoid the prediction deviation caused by the monotonously decreasing trend of the selected sample interval, the change characteristics of historical data will be further analyzed in the future, the scope of reference data will be expanded, and the anti-interference mechanism of prediction will be improved.

Key words: public emergency, agent based index mechanism, spearman correlation, BP neural network, flight forecast with Matlab

摘要： 为提升航空运输预测的抗外界干扰能力,改变传统时间序列的数据趋势预测弊端,以突发公共事件为个体引导特征,构建事件生命周期延续下的ABI飞行量预测机制。运用SPSS数据分析技术,统计分析历史公共突发事件对航班飞行量的影响,机器学习飞行量与突发事件的斯皮尔曼关联性,建立的BP神经网络的飞行量预测改善模型,利用突发事件与飞行量趋势变化作为训练函数,改进的Matlab.net对样本数据进行训练,修正各层神经元权值和阈值使误差函数沿负梯度方向下降,逼近期望预测值。以新冠疫情下的我国民用运输飞行量预测为例,验证模型和算法的可行性和预测效果;结果表明:BP神经网络的计算搜索技术一定程度内解决了突发公共事件与飞行量变化的复杂非线性映射关系,最优验证集均方误差得到最接近期望的预测结果。

关键词: 突发公共事件, ABI机制, 斯皮尔曼相关性, BP神经网络, Matlab飞行量预测