MLSEV Virtual. Optimization of Passengers Waiting Time in Elevators
0 likes270 views
This document summarizes a project to optimize passenger waiting times in elevators using machine learning. Data was collected from elevators and analyzed, revealing subgroups that could be improved by establishing parking floors. Machine learning models were developed and one was implemented, achieving an 8% reduction in waiting times for hotel elevators. Further improvements to the models are planned, along with expanding the use of machine learning to other applications like predictive maintenance and demand forecasting.
MLSEV Virtual. Optimization of Passengers Waiting Time in Elevators
- 1. 2nd edition
- 2. #MLSEV 2 Optimization of passengers waiting time in elevators using ML Delio Tolivia R&D&i, Talento Corporativo
- 4. #MLSEV 4 Business lines TRAINING We design and implement your e-Learning plans. CUSTOM TECHNOLOGY e-Commerce, APPs, CRMs, Corporate portals, Public Administration proyects and Citizen Participation. TECHNOLOGICAL INNOVATION IoT, Machine Learning, Blockchain, Augmented and Mixed RalityRealidad DIGITAL COMMUNICATION Online Presence, Digital Strategy, Digital Marketing, Content and Multimedia
- 5. #MLSEV 5 Technological innovation Virtual Reality Blockchain Internet of things Machine learning
- 9. #MLSEV 9 Project IALift Optimization of passengers waiting time in elevators using ML
- 11. #MLSEV 11 Project • Data: We got data from the elevators • Service: Thyssenkrupp wants to improve their services • Prediction: We can use ML to predict • Waiting time: The time that an user of the elevator waits untill it arrives. Can we improve the waiting time and make an “intelligent” elevator?
- 12. #MLSEV 12 Project Data Exploration We have taken data from 5 street elevators and a group of 2 elevators from an hotel. On the Street elevators we can see two groups (call count/seconds wait): • Four of them are very similar (1,2,3,4) • One of them has 4 floors and it´s behaviour is quite different from the others (5)
- 13. #MLSEV 13 Project Have aslo divides the Street elevator group into two more subgroups (1,4 and 2,3) Data Exploration
- 14. #MLSEV 14 Project In one of the subgroups (2,3) we have detected that all calls came from the same floor Data Exploration
- 15. #MLSEV 15 Project For the another subgroup (1,4) we realized that the elevator was receiving the 90% of the calls from the same floor Data Exploration
- 16. #MLSEV 16 Project The elevator whose behaviour is different (5) is more difficult to model Data Exploration
- 17. #MLSEV 17 Project We have also studied the time distribution by hour, day, week. Domingo 3 Domingo 10 Domingo 24 Domingo 17 Data Exploration
- 18. #MLSEV 18 Project Taken all previous into account… what have we done to solve the problem and improve the tk elevators system? • In the subgroup (2,3): we decided to establish a “parking” at floor 0 • In the subgroup (1,4) we also decided to establish a “parking” at floor 0 • In the more complicated one (5) we decided to make an ML model
- 19. #MLSEV 19 Project Results: • Thanks to the data analysys carried out we detected a malfunction in the subgroup of elevators (2,3) -> Success • We have achive an improvement of 12% in terms of reducing waitting time for the other subgroup (1,4). The energy measures confirm that with the new parking system the elevators consume the same-> Success Data exploration is very important!
- 20. #MLSEV 20 Project ML Model (I): • We have taken the data, and have created some models using OptiML. We keep on working and have created some new variables to improve the models: ü SecondsWait_1: The seconds that the user have wait in the previous call. ü last_hour_calls_2: Number of calls from the second floor in the last hour (It is the floor with the largest number of calls) ü last_three_minutes_calls_0: Number of calls from the floor 0 in the last 3 minutes We have also added data from work calendar and meteorology o improve models OPTIML
- 21. #MLSEV 21 Project ML Model (II): The best model was an ensemble (We focus on accuracy)
- 22. #MLSEV 22 Project Implementation: Whe have put into production a docker containing an API. The elevator control application make calls trough this API to get the probability of each floor for the next call so the system can choose the best option to move the elevator.
- 23. #MLSEV 23 Project Results: • At the time we were going to test the model, the elevator was out of service, so we couldn´t use it. That´s when we decided to use the same type of model and variables for the hotel elevators. We have taken the new data, have trained the model and have tried it in the real installation. • We achive an 8% of waiting times improvement!. • However this model isn't perfectly designed for this group of elevators and we would love to improve it by calculating, adding and testing new variables.
- 25. #MLSEV 25 Forthcoming use cases We keep on working with the elevators but we also have another use cases: • Predictive Maintenance (PdM) • Production planning • Quality control • Demand forecasting • Etc. • . What’s next?: