METHODS AND SYSTEMS FOR ELEVATOR CROWD PREDICTION

Abstract

A method for crowd prediction in an elevator system includes logging the number of calls across a period of time and generating a first time series for an average number of calls; collecting external influence data and generating a second time series for the external influence data; performing a cross-correlation test on the first and second time series; when a cross-correlation between the first and second time series is determined, performing a causality test on the first and second time series; and when a causal relationship between the first and second time series is determined, using the causal relationship to predict an expected number of calls in the elevator system.

Claims

1. A method for crowd prediction in an elevator system, the method comprising: logging the number of calls across a period of time and generating a first time series for an average number of calls; collecting external influence data and generating a second time series for the external influence data; performing a cross-correlation test on the first and second time series; when a cross-correlation between the first and second time series is determined, performing a causality test on the first and second time series; and when a causal relationship between the first and second time series is determined, using the causal relationship to predict an expected number of calls in the elevator system.

2. The method of claim 1, wherein the external influence data relates to one or more of: weather, epidemics, holidays, special events, urban transportation systems, and road traffic.

3. The method of claim 1, further comprising: performing a stationary test on the external influence data; and when the stationary test shows that the external influence data has a trend, de-trending the external influence data before generating the second time series.

4. The method of claim 1, further comprising: using the expected number of calls to predict an expected car occupancy level.

5. The method of claim 4, wherein predicting an expected car occupancy level further comprises: monitoring individual passenger journeys across a period of time; and implementing a machine learning process to identify passenger habits and further predict the expected car occupancy level.

6. The method of claim 5, wherein monitoring individual passenger journeys comprises: collecting passenger journey data using at least one of a sensor in a hallway and a sensor in car; and applying a time stamp to the passenger journey data.

7. The method of claim 6, wherein the passenger journey data includes one or more of: a boarding floor, an intended destination floor, a deboarding floor, a passenger identification, and occupancy volume.

8. The method of claim 5, wherein monitoring individual passenger journeys comprises: recognising an individual passenger's identity.

9. The method of claim 5, wherein monitoring individual passenger journeys comprises: using a hallway sensor to identify an individual passenger and using a car sensor to re-identify the same individual passenger.

10. The method of claim 1, further comprising: comparing the expected car occupancy level to an available car occupancy level; and issuing a crowd notification when the expected car occupancy level exceeds the available car occupancy level.

11. The method of claim 10, further comprising: controlling dispatch and/or stopping of at least one car in the elevator system in response to the crowd notification.

12. An elevator system comprising: a monitoring system arranged to log the number of calls across a period of time and generate a first time series for an average number of calls; a processor arranged to: receive external influence data and generate a second time series for the external influence data; perform a cross-correlation test on the first and second time series; when a cross-correlation between the first and second time series is determined, perform a causality test on the first and second time series; and when a causal relationship between the first and second time series is determined, use the causal relationship to predict an expected number of calls in the elevator system.

13. The elevator system of claim 12, wherein the monitoring system is arranged to: monitor individual passenger journeys across a period of time; and implement a machine learning process to identify passenger habits and predict an expected car occupancy level.

14. The elevator system of claim 12, further comprising a crowd detection system arranged to: compare the expected car occupancy level to an available car occupancy level; and issue a crowd notification when the expected car occupancy level exceeds the available car occupancy level.

15. The elevator system of claim 12, further comprising an elevator dispatch controller arranged to: redirect an incoming call to another elevator car when the expected car occupancy level exceeds the available car occupancy level; and/or avoid stopping an elevator car when the expected car occupancy level exceeds the available car occupancy level.

Description

DRAWING DESCRIPTION

[0030] Some examples of this disclosure will now be described, by way of illustration only, and with reference to the accompanying drawings, in which:

[0031] FIG. 1 is a schematic overview of a process for logging the number of calls and generating an associated time series in an elevator system according to an example of the present disclosure;

[0032] FIG. 2 is a schematic overview of a method for crowd prediction in an elevator system according to an example of the present disclosure;

[0033] FIG. 3 is a schematic overview of an elevator system according to an example of the present disclosure;

[0034] FIG. 4 is a schematic overview of a method for crowd prediction in an elevator system according to an example of the present disclosure; and

[0035] FIG. 5 is a schematic overview of a machine learning process in a monitoring system according to an example of the present disclosure.

DETAILED DESCRIPTION

[0036] As is generally known in the art, and seen in FIG. 1, a data source 100 relating to calls logged in an elevator system may be used as follows. At step 102, the cumulative number of calls per time interval (e.g. 10 minutes) is calculated. At step 104, the data are visualized e.g. using appropriate statistical tools as is well known in the art. At step 106, a time series is generated for the average number of calls. The calls log 100 represents a recording of past events, e.g. floor departure/destination and associated timestamps. The time series generated at step 106 may be used to predict an expected number of calls, at step 108, based on the average number of calls in the past.

[0037] FIG. 2 provides an overview of an exemplary method for crowd prediction in an elevator system. As already seen in FIG. 1, a calls log 100 may be used to generate a first time series for the average number of calls 106. According to examples of the present disclosure, a data source 120 relating to external influences is also connected to the elevator system. At step 122, a second time series is generated for the external influence data. As is shown in dotted outline, the external influence data 120 optionally undergoes a stationary test 132 and de-trending 134.

[0038] As is further seen from FIG. 2, a cross-correlation test 124 is performed on the first and second time series. When a cross-correlation between the first and second time series is determined, a causality test 126 is then performed on the first and second time series. If the outcome of both the cross-correlation test 124 and the causality test 126 is positive, i.e. when a causal relationship between the first and second time series is determined, the causal relationship is used to predict an expected number of calls in the elevator system at step 128. This prediction of the expected number of calls is based on the effect of external influences. On the other hand, if the outcome of either the cross-correlation test 124 or the causality test 126 is negative, then the expected number of calls is predicted based on the average number of calls in the past, as seen at step 130, and this may be the same outcome as seen at step 108 in FIG. 1.

[0039] As will be described further below, the prediction of expected calls from steps 128,130 may be converted to an expected car occupancy level and compared with an available car occupancy level. The result of this comparison can be used to inform car dispatching in an elevator system.

[0040] FIG. 3 shows in general how a monitoring system 1 collects data 20 from a hallway sensor 8 and data 30 from a car sensor 10. In this example, the monitoring system 1 is in an elevator system that comprises a plurality of elevator cars, such as the car 2, controlled by a dispatch controller 4. The monitoring system 1 is able to track an individual passenger 6, and output real time passenger data, using multiple sensors. The hallway sensor 8 is arranged to monitor the individual passenger's approach to an elevator car. The car sensor 10 is arranged to monitor the individual passenger 6 inside the elevator car 2. In this example, the hallway sensor 8 also acts as a passenger volume sensor arranged to determine the individual passenger's occupancy volume. The hallway sensor 8, such as a video camera and/or depth sensor, may utilise object detection algorithms to recognise a suitcase 12 associated with the passenger 6. The individual passenger's occupancy volume includes the size of the passenger 6 and the size of the suitcase 12.

[0041] In this example, the monitoring system 1 is arranged to recognise the identity of the passenger 6. At least one hallway sensor 8 is arranged to monitor the passenger's interaction with a call input device 7, depicted here as a kiosk. This interaction may be used to identify the passenger 6, for example if the passenger presents an identity card to input an elevator call. From information received from the hallway sensor 8, the monitoring system 1 starts to build up real time passenger data including one or more of: an elevator call request, a boarding floor, an intended destination (e.g. as entered at the call input device 7), a passenger identification, and occupancy volume.

[0042] The monitoring system 1 includes a processor that is configured to calculate an available car occupancy level based on the real time data 20, 30. Turning to FIG. 4, there is seen an example wherein the monitoring system 1 is used to provide a refinement to the prediction of expected calls from steps 128,130 (in FIG. 2). The monitoring sensors 14, as already described in relation to FIG. 3, provide real time passenger journey data to a time stamping system 16. In the monitoring system 1 there is a processor 18 arranged to implement a machine learning process. The processor 18 is able to build up a picture of the whole elevator system's traffic pattern over a period of time of days, weeks or even months. The processor 18 may be at least partly combined with a processor relating to the monitoring sensors 14 of the monitoring system 1, or implemented as a physically separate processor, or the processor 18 may be implemented in the cloud. The processor 18 is arranged to determine an expected car occupancy level.

[0043] As is further seen in FIG. 4, the expected car occupancy level is compared with the available car occupancy level determined by the monitoring system 1. This comparison may be carried out by a crowd detection processor 20 that is separate to, or combined with, the processor 18 in the monitoring system 1. A crowd notification is then sent to the dispatch controller 4 when the expected car occupancy level exceeds the available car occupancy level.

[0044] With reference to FIG. 5, it is further disclosed how the monitoring system 1 may be used to learn passenger habits and build up a traffic pattern for the elevator system. In this example, an identity test 124 is performed on the real time passenger journey data. If an individual passenger is identified by the monitoring system 1, the machine learning process 218 can immediately start to learn passenger habits for each identified passenger. If an individual passenger is not identified by the monitoring system 1, the real time passenger journey data is still logged at 200 and fed into the machine learning process 218 and, over time, this enables the machine learning process 218 to identify repeats or trends in individual passenger journey data so that passenger habits are learnt. The passenger habits are stored at 202. Furthermore, the machine learning process 218 integrates individual passenger journey data received over the course of many hours, days, weeks or months to build up traffic patterns 204 that is specific to the elevator system. The machine learning process 218 can use the passenger habits 202 and traffic patterns 204 to more accurately determine the expected car occupancy level at any given moment in time. The expected car occupancy level may be refined in this way before undergoing the crowd detection comparison 20 seen in FIG. 4.

[0045] It will be appreciated by those skilled in the art that the disclosure has been illustrated by describing one or more specific examples thereof, but is not limited to these aspects; many variations and modifications are possible, within the scope of the accompanying claims.