Van Phu Tran - Post-graduate Student, Volgograd State Technical University E-mail: vanphu.vstu.russia@gmail.com M.V. Shcherbakov - Dr.Sc. (Eng.), Volgograd State Technical University E-mail: maxim.shcherbakov@gmail.com Nguyen Tuan Anh - Post-graduate Student, Volgograd State Technical University D.A. Skorobogatchenko - Dr.Sc. (Eng.), Volgograd State Technical University

Proactive decision support systems are the next step of computer decision-making support systems (DSS) which implements the principles of proactive computing. The developing and implementing of preventive decisions can lead to cost savings in processes management, in particular, urban processes management based on the big heterogeneous data analysis. Often, data sources provide data with different types, so collecting and pre-processing of variation data types is an unstructured problem it requires significant time costs. There are two types of data sources are considered: log files from vehicles and video recording system. There are several typical approaches for data collecting, transferring and storing. The method which is the most commonly used (marked as P1) is based on the storing of the raw data in a distributed file system with metadata. The proposed in the paper method includes the following steps: (i) defining the required data scheme for data storing, (ii) describing the schemes of data sources and setting options for data collecting (iii) setting the relationships between schemes of data (sources and required), (iv) implementing data transforming algorithms as distributed software (v) inserting data into database. The proposed method (PM) was implemented for traffic situation evaluating based on vehicles data movement (as logs files) and video streams. The user makes requests for data about the movements of vehicles in a given time interval for a certain geolocation from all available data sources, i.e. log files and from video streams. Experiments were made for a various intensity of the incoming data: log files up to 1 million messages per second and a single video stream. The average times of data processing by PM method are 0.14 seconds for logs and 0.75 seconds for video, i.e. in total 0.89 sec. The frequency of users requests is varied: 10, 30, 300 and 3,000 requests per second. So for 10 RPS, the P1 shows performance 0.703 ± 0.16 secs and PM 1.087 ± 0.32 sec, which differs from each other not too much. However, for 30 RPS, for P1 it takes 1.925 ± 0.09 sec, and for the PM it takes 1.368 ± 0.25 sec, which is 1.4 times less than P1. For 300 PPS runtime of PM is 0.33 ± 2,904 sec, which is 1.7 times less than P1 (4.995 ± 0.28 sec). For 3000 RPS run time was 14,153 ± 1.04 sec, which is 3.24 times lower than P1 (45.899 ± 2.56 sec). To confirm the superiority of PM the Siegel\'s nonparametric test for two independent samples was used (with a significance level a = 0.01).

 

1. Tennenhouse D. Proactive computing // Communications of the ACM. 2000. V. 43. P. 43-50.
2. SHHerbakov M.V., Sadovnikova N.P., Parygin D.S., Golubev A.V., CHechetkin I.A. Avtomatizacija podderzhki prinjatija reshenijj po razrabotke marshrutov obshhestvennogo transporta na osnove analiza dannykh o korrespondencijakh zhitelejj // Vestnik kompjuternykh i informacionnykh tekhnologijj. 2016. № 8. C. 29-33.
3. Opher Etzion, Peter Niblett Event Processing in Action // Manning Publications Company 2010. P. I-XXIV, 1-360.
4. De Gooijer J.G., Hyndman R.J. 25 years of time series forecasting // International Journal of Forecasting. 2006. V. 22. № 3. R. 443-473.
5. Bengio Y. Practical recommendations for gradient-based training of deep architectures. 2012. https://arxiv.org/pdf/1206.5533v2.pdf
6. Stewart R.J., Trinder P.W., Loidl H-W. Comparing high level mapreduce query languages, advanced parallel processing technologies: 9^th International Symposium, APPT 2011, Shanghai, China, September 26-27, 2011. Proceedings.R. 58-72.
7. GolubevA.V., CHechetkinI.A., Solnushkin K.S., SadovnikovaN.P., ParyginD.S., SHHerbakovM.V. Strategway: web solutions for building public transportation routes using big geodata analysis // The 17^th International Conference on Information Integration and Web-based Applications & Services (iiWAS2015) (Brussels, Belgium, 11-13 December 2015): Proceedings / ed. by Maria Indrawan-Santiago [etc.]. [New York : The Association for Computing Machinery]. 2015. P. 665-668.
8. Van Phu Tran, Shcherbakov M.V. Anh Tuan Nguyen EVGEN: a Framework for Event Generator in Proactive System Design // The 7^th International Conference on Information, Intelligence, Systems and Applications (IISA). 2016.