Rosa Basagoiti,^{ }Montragon University

Maite Beamurgia, Mondragon University

Richard Peters, Peters Research Ltd

Stefan Kaczmarczyk, Northampton University

This paper was presented at Symposium of Lift and Escalator Technologies, 2012. This web version © Peters Research Ltd 2012.
1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | |

1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |

2 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 |

3 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 3 | 0 | 0 |

4 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |

5 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 2 |

6 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |

7 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |

8 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |

9 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |

10 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |

11 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |

- Divide the log data, for the time period needed, in our case it was 5 minutes (also 2.5 minutes for forecasting purposes).
- For a given period, separate the lift movements into journeys.
- From those journeys, the following information is collected every time the lift stops:
- Number of passengers leaving and entering the lift at each floor. To obtain this number it is necessary to have three weight measures, just before the lift arrives at the floor where it is going to stop, after the passengers leave the lift and after the lift has left the floor. The information about the number of passengers boarding and alighting can be extracted from weight sensor data or maybe the processing of camera information.
- Time stamp associated with the landing calls coming from this floor.
- Car calls that have been registered after leaving that floor.

- The principle of passenger flow conservation equation, says that the number of passengers entering and leaving a lift during a journey is equal. Using that principle, it is possible to estimate the missing information about the real passenger movements in the journey. This calculation has been done using the symbolic calculation module of MATLAB. In some cases, some rules, extra rules, are needed to be applied in order to solve the equations. These calculations are completed for all the journeys across all the lifts.
- Information about the timestamp corresponding to the first request of the journey has to be collected to later aggregate the passenger movements across all the lifts for the five minutes time intervals. The final value is obtained by aggregating from the same time interval the different journey data in one OD matrix and then, aggregating the OD matrices of the different lifts.

Methods | Advantages | Disadvantages |

Moving Average | Simple | Not Good Fit |

Arima | Good Fit | Tedious Programming |

Kalman Filters | Good Fit | Tedious Programming |

Bayesian Networks | Good Fit | Tedious Programming |

Artificial Intelligence & Neural Networks | Well Known | Fall in Local Minima |

Support Vector Machines | Forecasting of Small Samples | Sensitive to the Selected Kernel Function |

Wavelets & ARIMA | Wavelets have Good Decomposition Power and ARIMA Good Linear Fitting | Tedious Programming |

5 minutes time interval OD matrix based prediction | 2.5 minutes time interval OD matrix based prediction | ||||||||

Periods | Real Calls | Estimated Calls | MSE | MRE | MAE | Estimated Calls | MSE | MRE | MAE |

11:40- 11:45 | 35 | 58 | 3,0682 | 67,42% | 0,0523 | 34 | 0,5019 | 34,47% | 0,0374 |

11:45- 11:50 | 31 | 49 | 1,6970 | 54,55% | 0,0302 | 32 | 0,3655 | 28,41% | 0,0163 |

11:50- 11:55 | 32 | 71 | 2,5076 | 73,48% | 0,0500 | 43 | 0,4754 | 30,68% | 0,0208 |

11:55- 12:00 | 33 | 59 | 2,2424 | 65,15% | 0,0560 | 44 | 0,4318 | 28,79% | 0,0278 |

12:00- 12:05 | 28 | 52 | 2,2727 | 62,12% | 0,0501 | 50 | 0,6155 | 26,14% | 0,0316 |

12:05- 12:10 | 47 | 55 | 3,0606 | 68,18% | 0,0710 | 47 | 0,6023 | 35,61% | 0,0420 |

12:10- 12:15 | 32 | 38 | 0,9242 | 34,85% | 0,0217 | 40 | 0,3902 | 23,48% | 0,0222 |

12:15- 12:20 | 52 | 79 | 1,9167 | 53,79% | 0,0490 | 52 | 0,5095 | 26,14% | 0,0272 |

12:20- 12:25 | 48 | 62 | 1,4848 | 53,03% | 0,0469 | 40 | 0,9811 | 31,82% | 0,0383 |

12:25- 12:30 | 50 | 35 | 2,1894 | 67,42% | 0,0489 | 48 | 0,6117 | 32,2% | 0,0309 |

12:30- 12:35 | 72 | 84 | 2,7576 | 72,73% | 0,0510 | 64 | 1,1420 | 39,02% | 0,0252 |

12:35- 12:40 | 50 | 48 | 1,5303 | 57,58% | 0,0624 | 54 | 0,4242 | 28,03% | 0,0300 |

12:40- 12:45 | 58 | 50 | 2,8485 | 69,7% | 0,0883 | 49 | 0,4337 | 26,89% | 0,0335 |

12:45- 12:50 | 79 | 42 | 3,1288 | 79,55% | 0,0586 | 57 | 0,9905 | 40,53% | 0,0243 |

Total | 647 | 782 | 31,6288 | 8,7955 | 0,7364 | 654 | 8,4753 | 4,3221 | 0,4075 |

- Juha-Matti Kuusinen, Janne Sorsa, Tuomas Susi, Marja-Liisa Siikonen, Harri Ehtamo
*A new model for vertical building traffic*. Transportation Research. 2010. - Li, B Markov
*models for bayesian analysis about transit route origin-destination matrices, 2009*. Transportation Reasearch Part B 43 (3), 301-310. - Hiri-O-Tappa, Kittipong; Pan-Ngum, Setha; Pattara-Atikom, Wasan; Narupiti, Sorawit
*A Novel Approach of Dynamic Time Warping for Traffic Congestion Detection and Short-Term Prediction*. 17th ITS World congress, 13p. 2010. - Hongbo Li, Yong Peng
*Traffic Flow Prediction Based on Wavelet Analysis and Artificial Neural Network*. ICLEM: Logistic for Sustained Economic Development: Infrastructure, Information, Integration. Section: Volume IV – System Optimization and Simulation Models, pp. 3528-3534. 2010. - Ni Lihua, Chen Xiaorong, Huang Qian
*ARIMA model for traffic flow prediction based on wavelet analysis*. 2nd International Conference on Information Science and Engineering (ICISE), pp. 1028 – 1031.2010. - Xiaoke Yan, Yu Liu, Zongyuan Mao, Zhong-Hua Li, Hong-Zhou Tan
*.SVM-based Elevator Traffic Flow Prediction*. Intelligent Control and Automation, 2006. WCICA 2006. - Mehdi Khashei , Mehdi Bijari.
*An artificial neural network (p, d,q) model for timeseries forecasting*. Expert systems with applications. 37 (2010) 479–489.