http://orcid.org/0000-0001-9584-879X
Figures
Abstract
Disruptive behavior by passengers delayed at airport terminals not only affects personal safety but also reduces civil aviation efficiency and passenger satisfaction. This study investigated the causal mechanisms of disruptive behavior by delayed passengers in three aspects: environmental, managerial, and personal. Data on flight delays at Shenzhen Airport in 2018 were collected and analyzed. The main factors leading to disruptive behavior by delayed passengers were identified, and an early warning model for disturbances was developed using multiple logistic regression and a back-propagation(BP) neural network. The results indicated that the proposed model and method were feasible. Compared to the logistic regression model, the BP neural network model had advantages in predicting disturbances by delayed passengers, showing higher prediction accuracy. The BP network weight analysis method was used to obtain the influence weight of each factor on behavior change of delayed passengers. The influence weight of different factors was obtained, providing an assistant decision-making method to address disruption from flight-delayed passengers.
Citation: Gu Y, Yang J, Wang C, Xie G (2020) Early warning model for passenger disturbance due to flight delays. PLoS ONE 15(9): e0239141. https://doi.org/10.1371/journal.pone.0239141
Editor: Qiang Zeng, South China University of Technology, CHINA
Received: June 15, 2020; Accepted: August 31, 2020; Published: September 21, 2020
Copyright: © 2020 Gu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The original data of the manuscript contain potentially sensitive information, and data are owned by Shenzhen Airport Co., Ltd. The data can not be publicly shared. Any researcher who wants to obtain data should contact Shenzhen Airport Co., Ltd. (E-Mail:TOC@szairport.com,Tel: +86-755-23458948, http://www.szairport.com/).
Funding: The author(s) received no specific funding for this work. Shenzhen Airport CO., Ltd., provided support in the form of salaries for the forth author [GX], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.
Competing interests: Shenzhen Airport CO., Ltd., provided support in the form of salaries for the forth author [GX], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Shenzhen Airport Co., Ltd. or any other institution does not develop related information products and apply for patents, does not provide test funds, and does not mind appearing as the author's unit. And it doesn't affect your decision to publish and share the results in journals.This does not alter our adherence to PLOS ONE policies on sharing data and materials.
1 Introduction
China has become a prominent force in civil aviation, and since 2005 its air-transport volume has ranked second highest in the world [1]. However, along with such rapid development, China also increasingly faces the problem of flight delays. Flight delays, especially long ones, can lead to disruptive behavior by passengers. This behavior not only affects the normal operation of the airport but also threatens the safety and service quality of civil aviation. According to statistics from Shenzhen Airport Terminal Operation Center, factors such as weather and air traffic control problems cause thousands of delayed flights annually, leading to many incidents of disruptive behavior. Due to a lack of effective early warning methods, the civil aviation industry is often reactive in dealing with disruptive behavior. This not only wastes manpower and material resources but also has little effect on reducing incidences of disruptive behavior. Such behavior can lead to a loss of control at the airport and seriously affect civil aviation safety.
Disruptive passenger behavior is a complex issue. Liu suggested that an event involving groups of people involves complex social phenomena, and it is often the case that a key person initiates the event; thus, that person should be identified and premanaged to control the occurrence of group events [2]. Using qualitative methods, Dell’Olio et al. classified key variables found to be related to human group behavior in certain situations [3]. Studies of group events caused by flight delays have mainly used qualitative analysis. Such research has investigated the main factors in incidents through passenger satisfaction surveys and has proposed measures to reduce incidents from economic and legal perspectives [4, 5]. Thengvall et al. suggested that when flights are delayed, airlines and airports do not adequately respond to passengers’ demands, which increases anxiety and may lead to group events [6]. At present, research on the main factors affecting disruptive passenger behavior has rarely considered the characteristics of terminal management, and quantitative studies are few. As such, attempts to apply an early warning system to terminal management have been limited.
This study’s research team spent three months in a terminal management department to collect data on passenger disruption events related to flight delays. The factors causing passenger disruption were found to be varied, and the amount of each factor’s contribution was unclear. The relationships between factors could not be directly determined. In this regard, BP neural networks and multiple logistic regression have their own unique advantages for dealing with complex problems, and they have been widely used in research on industry, transportation, and medicine, among other fields [7–13]. Using a backpropagation (BP) neural network, Ahmed detected and classified faults in automobiles’ internal combustion engines and used experiments to verify the method’s stability and accuracy [14]. Wu used a BP neural network to analyze the measurement errors of an airborne laser and found that the method was beneficial for improving the accuracy of airborne ranging [15]. Zeng et al. developed a neural network (NN) model to explore the nonlinear relationship between crash frequency and risk factors[16]. Using four data-analysis methods, Reeve aimed to improve traditional logistic regression for analyzing treatment differences in incidence rates, thereby expanding upon logistic regression and generalizing the link function. Among the four methods, resampling based on the exact distribution function yielded the closest to nominal coverage rate [17].
Based on survey data, and drawing on prior experience providing on-site support in civil aviation, the present study analyzed the causes of disruptive behavior by delayed passengers to determine the key factors. Multiple logistic regression and a BP neural network were used to create a predictive model for disruptive passenger behavior; comparative experiments were then used to verify the model’s effectiveness.
2 Analysis of the main factors leading to disruptive behavior by delayed passengers
2.1 Causal mechanisms
Passenger behavior is an open, sudden, and complex system. When a complex system characterized by dynamic change is disturbed, it may change from a stable state to an unstable one. This process passes through a critical interface, which is the interface between two states. Between the stable state and the interface, the system is stable and controllable. Beyond the critical interface, the system is in an unstable state; it is then uncontrollable and cannot easily return to the original state [18].
When flights are delayed, passengers cannot travel normally, which results in psychological dissatisfaction. As passenger dissatisfaction becomes more serious, it reaches a critical state; once the critical state is exceeded, there is a risk of passenger disturbance. In this study, the critical factor refers to verbal disputes between passengers and staff in cases of flight delay. Disruptive passenger behavior is defined as extreme behavior that affects flight security or disrupts public order for the purpose of making demands or expressing dissatisfaction.
Many factors affect individual behavior in unconventional events, and the relationships among them are complex [19]. The causes of disruptive behavior are varied, and the circumstances and means of expression differ. This study collected flight delay data on-site from Shenzhen Airport in 2018, and passenger disturbance events were tracked and investigated in real time. Based on the obtained data, three main factors were identified: personal, environmental, and managerial.
(1) Personal.
With improvements in living standards, people tend to prefer quicker and more convenient means of travel; thus, civil aviation has become a popular choice for long-distance travel. Due to intrinsic personal differences (e.g., occupation, age, physical health, mental health), passengers can react differently to flight delays. Thus, the likelihood of disruptive behavior can vary among different passengers.
(2) Environmental.
1) Time. Passengers choose air travel mainly for convenience. Flight delays, however, offset the advantages and produce dissatisfaction. Generally, as the duration of flight delay increases, passenger dissatisfaction intensifies. Moreover, for physiological reasons, long-term delays at night can further increase dissatisfaction. Therefore, when analyzing the effect of time on passenger mood, we should consider both the duration and timing of the delay.
2) Space. Crowding in confined spaces can adversely affect people’s emotions and make them irrational. People are easily affected by the emotions of others around them, making it difficult to control their behaviors [20]. When flights are delayed, terminals may become crowded, and passenger disturbance incidents often occur in these crowded areas. Groups may form among passengers with similar demands. When a passenger begins to adopt language and actions reflecting the characteristics of a group leader, other passengers may unconsciously imitate the behavior. Under such a group dynamic, an individual who is usually calm and restrained may become disruptive.
(3) Managerial.
When flights are delayed, service staff must deal with passengers directly. Different airlines provide different kinds of training and management for their employees, and the level of service can therefore vary. The occurrence of disruptive behavior is usually sudden, contingent, time sensitive, and wide ranging; thus, efficient management systems and sound plans are needed to deal with it in a timely and efficient manner. If staff have insufficient training and professional knowledge for dealing with emergencies, they will not effectively respond to passengers’ psychological changes, which could easily worsen passenger behaviors. Meanwhile, when a flight is delayed, the airline may provide some supplies to passengers. However, since the law does not specify compensation for delayed passengers, compensation can vary from airline to airline. Some airlines, for example, provide free food while others do not. These different levels of service can have different psychological effects on passengers.
Fig 1 shows the main factors that lead to disruption by passengers in airport terminals.
2.2 Data sources and statistical analysis
The behavior states of delayed passengers in a terminal can be divided into three categories: emotional stability, quarrel, and disturbance. In the first state, passengers are very calm, waiting quietly without any disturbing behavior or verbal disputes. In the second state, there are verbal disputes between passengers and staff but no disturbances. In the third state, there are passenger disturbances.
The Civil Aviation Administration of China stipulates that a flight delay occurs when the actual departure time is at least 20 minutes later than the planned departure time. Accordingly, we collected data for 856 delayed flights from the terminal management department of Shenzhen Airport from June 1 to August 30, 2018. These included 355 flights without disruption, 391 flights with verbal disputes only, and 110 flights with disruptions. The specific effects of each category are analyzed below.
(1) Length of the flight delay.
The 856 delayed flights, including 110 with passenger disruptions, were statistically analyzed based on the length of the delays (Table 1).
(2) Time of flight delay.
The likelihood of disturbance varies according to the time of day in which a delay occurs. Passengers become more anxious at night because they worry about whether they will arrive the same day and whether the airport can solve transportation and accommodation problems. The delayed flights, including 110 with passenger disruptions, were statistically analyzed based on the time of the flight delay (Table 2).
Since flight delay is a cumulative process, delayed flights in different time periods were partially repeated. Following this method, the number of delayed flights was more than 856.
(3) Passenger density at boarding gates.
When there is a delay, flights can be allocated to different boarding gates. Each boarding gate has a different number of flights and a different passenger density, leading to differing likelihoods of disturbance. Table 3 shows the number of delayed flights allocated to boarding gates and the number of disturbances at Shenzhen Airport in 2018. The statistical analysis indicated that the more flights allocated to the same gate, the higher the passenger density, and the greater the likelihood of disturbance incidents.
(4) Service level of aviation ground service companies.
Airline passenger satisfaction scores for 2018 were collected from the website of the Civil Aviation Passenger Service Satisfaction Survey (https://www.capse.net/). The data were classified and counted, and Table 4 shows the statistical results.
(5) Average age of delayed flight passengers.
The average age of delayed flight passengers was 30–45. Table 5 shows the data for the average age of delayed passengers and the number of delayed flights with disturbances.
3 Prediction model based on multiple logistic regression
3.1 Multiple logistic regression
Logistic regression is a kind of linear regression. The basic principle is to use the function f(z) as the function to predict by linearly summing the prediction factors. The value of f(z) is [0, 1]. When the event occurs, probability is p; when the event does not occur, probability is 1-p. Maximum likelihood estimation is used for parameter calculation. Multiple logistic regression is a multivariate statistical analysis method used to analyze the relationship between dependent (reaction) and independent (observation) variables in multiclassification situations [21]. The number of independent variables is n, and the number of dependent variables is m, where m≥3. The principle of multiple logistic regression is to divide different factor classifications into multiple binary logistic regressions. Its mathematical model is described as follows:
(1)
(2)
(3)
(4)
(5)
(6)
where, P is the probability of the occurrence of reaction variable type j, xk is observation variable k, βjk is the regression coefficient, αj is the regression intercept of reaction variable j, and πj is the conditional probability of the occurrence of event j.
3.2 Establishment of delayed flight dataset
Through the analysis set forth in Section 2.2, we obtained five factors that affect the incidence of passenger disturbance. Based on these factors, we can establish the following delayed flight data set for analysis:
(7)
where Xi is the vector of data for delayed flight i, xi(1) is length of the delay for flight i, xi(2) is time of flight delay i, xi(3) is passenger density at the boarding gate for delayed flight i, xi(4) is service level of aviation ground service companies for delayed flight i, and xi(5) is average age of delayed flight passengers for flight i.
3.3 Experimental analysis
Passenger behavior state is the dependent variable, and the influencing factor is the independent variable. The confidence interval of the model is 95%. Multiple logistic regression analysis was carried out using SPSS analysis tool.
The significance level of each factor and model was less than 0.05. The model had statistical significance, and the selected factors were effective (Tables 6 and 7). The overall prediction accuracy of the model was 71.03%, as shown in Table 8.
4 Prediction model based on BP neural network
4.1 BP neural network
Section 3 analyzes the factors related to disruptive delayed-passenger behavior, finding complex interaction between the factors and disturbance-incident occurrence. This section establishes a disturbance early warning model.
A BP neural network consists of a multilayer feed-forward neural network based on an error backpropagation algorithm. It was pioneered by Rumelhart and McClelland in 1986 [22] and has become the most widely used neural network learning algorithm [23–26]. The BP neural network consists of one input layer, one or more hidden layers, and one output layer. Each layer has one or more neurons. Information is transmitted from the input layer to the output layer through the hidden layer or layers. The strength of the connection between neurons in different layers is represented by a connection weight [27]. Taking a BP neural network with a single hidden layer as an example, its logical structure is shown in Fig 2.
The principle of the BP neural network is as follows:
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
where Xk is the input vector, Yk is the output vector, m is the number of learning mode pairs, n is the number of units of the input layer, q is the number of units of the output layer, p is the number of units of the hidden layer, Sk is the net input vector of the hidden layer, Bk is the net output vector of the hidden layer, Lk is the net input vector of the output layer, Ck is the actual output vector of the output layer, W is the connection weights of the input and hidden layers, and V is the connection weights of the hidden and output layers.
4.2 Designing the BP neural network
Based on the analysis in Section 3.2, a quantization process was performed to obtain a matrix of impact factors. As a result, there are five nodes on the input layer and one node on the output layer. Since a single hidden-layer neural network can theoretically approximate any continuous function (as long as there are a sufficient number of neurons in the hidden layer) [28], this study used a single hidden layer to construct the early warning model.
(1) Design of the hidden layer.
Determining the number of neurons in the hidden layer is very important in the network design process. Too many neurons will increase the amount of processing and lead to over fitting. If the number of neurons is too small, it will affect network performance and may not achieve the expected results. The number of hidden-layer neurons is related to the complexity of the problem, the number of neurons in the input and output layers, and the setting of the expected error. In this study, an empirical formula was used as a guideline for the initial selection of the number l of neurons in the hidden layer [29]:
(16)
(17)
where n is the number of neurons in the input layer, m is the number of neurons in the output layer, and a is a constant adjusted based on the accuracy and convergence speed of the verification stage.
(2) Selection of activation function.
A BP neural network often uses sigmoid-type differentiable functions or linear functions as the activation functions for the network. TANSIG and LOGSIG are types of sigmoid functions. This study chose the TANSIG function as the transfer function for neurons in the hidden layer. Since the output of the network was normalized to the range [−1, 1], the prediction model also applied an S-shaped logarithm function TANSIG as the transfer function for neurons in the output layer.
4.3 Weight analysis of influencing factors
The weight analysis of influencing factors refers to calculating the proportion of the connection weights of input nodes related to input factors in the total weights of all input nodes to the output contribution of the network. According to the weight contribution rate, the degree of influence of the input factors on the output is judged, so as to determine its importance. A single hidden layer BP neural network with 5 input nodes and 1 output node is constructed. The weight calculation formula [30] is as follows:
(18)
(19)
where bi is the weight contribution rate of input node i, Wij is the connection weights of input layer node i and hidden layer node j, Vj is the connection weight between hidden layer node j and output node q (q = 1), and ci is the weight contribution rate of input node i.
4.4 Data standardization
Input data standardization was conducted on the basis of Sections 2.2 and 3.2 (Table 9). Whether disruptive behavior occurred was the target for the output prediction of the BP neural network. This is represented by the variable Y with the range [0.25, 0.5, 0.75]. An output of 0.25 indicates emotional stability, 0.5 quarrel, and 0.75 disturbance.
4.5 Results
In accordance with Section 2.2, data were collected for 856 delayed flights. Data for 227 flights were randomly selected as test data. Target error was set to 0.01, learning rate to 0.02, and training accuracy to 0.001. Error, defined as the absolute value of the difference between test output Y’ and expected value Y, should not be greater than 0.125. If this value is greater than 0.125, the prediction is counted as an error.
In fact, if there are noise data in the input neural network data, too many neurons will make the noise data influence amplification, leading to reduced prediction accuracy of the model and the over-fitting phenomenon. The results showed that for the single hidden-layer BP network, the accuracy of the trained model increased with the number of neurons and the number of training iterations. However, when the increase reached a certain level, the rate of change leveled off. Table 10 shows the results.
When the number of neurons was 7, and the number of training cycles was 10,000, the number of errors was 77 (33.92%). When the number of neurons was 13, and the number of training cycles was 100,000, errors were reduced to 25.99%, and accuracy reached 74.01%. When the number of neurons was 20, and the number of training cycles was 120,000, errors were reduced to 18.50%, and accuracy reached 81.50%. Fig 3 shows the corresponding actual network error at this time.
In accordance with Section 4.3, the influence weight of each factor is shown in Table 11. When the number of nerves was 20 and the number of training cycles was 120,000, influence weight of length of flight delays was 0.26, the influence weight of passenger density at boarding gates was 0.25, and the influence weight of average age of delayed flight passengers was 0.12, the smallest influence weight.
5 Conclusion
Combining on-site data and a situational assessment of civil aviation, this study analyzed the causal mechanisms of delayed-passenger disruptions in terms of personal, environmental, and managerial aspects. Based on 2018 data for flight delays at Shenzhen Airport, the main factors leading to delayed-passenger disruption were identified as follows: length of flight delay, time of flight delay, passenger density at boarding gates, service level of aviation ground service companies, and intrinsic passenger factors. A BP neural network and multiple logistic regression were used to establish prediction models. The experimental results indicated that the prediction accuracy of the BP neural network reached 81.5%, showing a better prediction effect than multiple logistic models. According to the influence weight analysis of the BP neural network, the length of flight delays and passenger density at boarding gates have a great impact on the behavior of delayed passengers. In practice, airport staff should pay more attention to passengers experiencing long delays. When flight density at the gate is too high, measures such as increasing service personnel should be taken to avoid passenger disturbance behavior.
As shown in the statistics, disturbances caused by passenger’s own reasons were difficult to quantify, only the average age of passengers was quantified. In the early warning analysis of passenger disturbances, priority was given to factors that had a greater impact on actual operating processes. Meanwhile, factors that were more difficult to observe—such as the physical and mental states of passengers—were not considered. In the future, more individual passenger factors should be integrated into the research, and the training parameters should be further optimized. Thus, there is a need for further research on the selection of impact factors. In addition, more methods will be used to establish the prediction model, such as ordered logistic regression, which is a direction of the future research. The prediction accuracy will be compared with that of BP neural network.
References
- 1. Liu X. P. Research on the Diffusion Model of Passenger Mass Emergencies under Flight Delay Situation. 4th International Conference on Applied Materials and Manufacturing Technology, Materials Science and Engineering, 2018, 423:1–6.
- 2.
Liu X. P. Research on Formation Mechanism of Passenger Emergencies Based on Theory of Social Network Analysis. 2018 4th International Conference on Information Management (ICIM), Oxford, 2018, pp. 62–65. https://doi.org/10.1109/INFOMAN.2018.8392810
- 3. Luigidell'Olio AngelIbeas, RosaBarreda RobertoSañudo. Passenger behavior in trains during emergency situations. Journal of Safety Research,2013,46:157–166. pmid:23932697
- 4. McCollough Michael A., Berry Leonard L., Yadav Manjit S. An empirical investigation of customer satisfaction after service failure and recovery. Journal of Service Research, 2000,3(2):121–137.
- 5. Jasper James M. The emotions of protest: affective and reactive emotions in and around social movements. Sociological Forum, 1998,13(3):397–424.
- 6. Thengvall Benjamin G., Bard Jonathan F., Yu Gang. Balancing user preferences for aircraft schedule recovery during irregular operations. IIE Transactions,2000,32(3):181–193.
- 7. Schmidhuber J. Deep Learning in Neural Networks: An Overview. Neural Networks, 2015,61:85–117. pmid:25462637
- 8. Fischer Aaja, Igel Christian. Training Restricted Boltzmann Machines: An Introduction. Pattern Recognition, 2014,47(1):25–39.
- 9. Liu W. F., Liu H. L., Tao D. P., Wang Y. J., Lu K. Multiview Hessian regularized logistic regression for action recognition. Signal Processing,2015,110:101–107.
- 10. Yann LeCun Yoshua Bengio, Hinton Geoffrey. Deep learning. Nature,2015,521:436–444. pmid:26017442
- 11. Deng L., Yu D. Deep Learning: Methods and Application. Foundations and Trends in Signal Processing, 2014,7(3–4):197–387.
- 12. Chen W., Xie X. S., Wang J. L., Biswajeet Pradhan, Hong H. Y. Dieu Tien Bui, et al. A comparative study of logistic model tree, random forest, and classification and regression tree models for spatial prediction of landslide susceptibility. CATENA, 2017,151:147–160.
- 13. Wang J. J., Shi P., Jiang P., Hu J. W., Qu S. M, Chen X. Y., et al. Application of BP Neural Network Algorithm in Traditional Hydrological Model for Flood Forecasting. Water, 2018,9(48):1–16.
- 14. Ahmed Ryan, Mohammed El Sayed, S. Andrew Gadsden, Jimi Tjong, Saeid Habibi. Automotive Internal-Combustion-Engine Fault Detection and Classification Using Artificial Neural Network Techniques. IEEE Transactions on Vehicular Technology, 2015,64(1): 21–33.
- 15. Wu B., Han S. J., Xiao J., Hu X. G., Fan J. X. Error compensation based on BP neural network for airborne laser ranging. Optik, 2016,127(8):4083–4088.
- 16. Zeng Q, Huang H. L, Xin P., Wong S. C., Gao M. Y. Rule extraction from an optimized neural network for traffic crash frequency modeling. Accident Analysis and Prevention, 2016,97:87–95. pmid:27591417
- 17. Russell R. Confidence Interval of Difference of Proportions in Logistic Regression in Presence of Covariates. Statistical Methods in Medical Research,2018,27(2):451–465. pmid:26988932
- 18.
Saunders P T. An Introduction to Catastrophe Theory. London: Cambridge University Press,.1980.
- 19. Virginia P S.Application of traffic simulation modeling for improved emergency preparedness planning. Management and Response in Urban planning and Development,2007,133(1) 51–60. pmid:26988932
- 20. Stokols D. On the Distinction between Density and Crowding: Some Implication for Future Research. Psychological Review (S0033-295X), 1972,79(3):275+. pmid:5056743
- 21. Zhang H, Yao D, Lu H. F., Xue L., Zhu N. N. Study on Water Source Discrimination Based on Multinomial Logistic Regression Model Using Principal Component Analysis. Geological Journal of China Universities, 2017,23(2):366–372.
- 22. Segal , Michael M. Explorations in Parallel Distributed Processing: A Handbook of Models, Programs, and Exercises.Science,1998,241(4869):1107+.
- 23. Abhijit Suresh K.V Harish, N. Radhika. Particle swarm optimization over back propagation neural network for length of stay prediction. Procedia Computer Science,2015,46:268–275.
- 24. Hu F. Y., Wang L., Wang S. S., Liu X. L., He G. X. A Human Body Posture Recognition Algorithm Based on BP Neural Network for Wireless Body Area Networks. China Communications, 2016,13(8):198–208.
- 25. Xu Z. H., Liu J. X., Chen X. X., Wang Y. L., Zhao Z. Continuous blood pressure estimation based on multiple parameters from eletrocardiogram and photoplethysmogram by Back-propagation neural network. Computers in Industry, 2018,89:50–59.
- 26. Wang L., Zeng Y., Chen T. Back propagation neural network with adaptive differential evolution algorithm for time series forecasting. Expert Systems with Applications, 2015,42(2):855–863.
- 27.
Li J. S., Tian Y., Liu Y. F., Shu T., Liang M. H. Applying a BP neural network model to predict the length of hospital stay. HIS 2013, London, UK, pp.18–29. https://doi.org/10.1007/978-3-642-37899-7_2
- 28. Hou M. Z., Han X. L., Gan Y. X. Constructive approximation to real function by wavelet neural networks. Neural Computing and Applications, 2009,18(883):883–889.
- 29. Sun F., Shi X. Q. Design of BP neural networks based on MATLAB. Computer and Digital Engineering, 2007,35(8):124–126.
- 30. Li Z. Y., Ding H. K., Ding J. A model of sources apportionment of atmospheric particulates based on weight analysis of networks. Journal of Sichuan University (Natural Science Edition),2004,41(5):1026–1029.