TensorTraffic - traffic prediction using machine learning - Pawel Gora
Description
Together with an international group of scientists, I am developing TensorFlow-based tool, TensorTraffic, for predicting traffic simulation outcomes, such as delays / total waiting times on a red signal, as a function of traffic control settings. We achieved great results, which may find applications in transport planning, real-time traffic management and many other fields.
Abstract
Together with an international group of scientists, I am developing TensorFlow-based tool, TensorTraffic, for predicting traffic simulation outcomes, such as delays / total waiting times on a red signal, as a function of traffic control settings. We achieved great results, which may find applications in transport planning, real-time traffic management and many other fields.
First, I will talk about a goal of the research and problem we would like to solve, which is: approximating traffic simulation outcomes fast and with an acceptable accuracy. The motivation behind this goal is to have a meta-model of traffic, which can allow to effectively evaluate quality of a large number of settings (e.g., traffic control strategies or road infrastructure settings).
Then, I will explain our approach and experiemnts. We built a dataset consisting of almost 120 000 traffic simulation scenarios, using Traffic Simulation Framework software, each case corresponded to a different traffic signal control strategy, simulations computed the total waiting time of all cars on a red signal in a given region of Warsaw (Stara Ochota district, 15 crossroads with traffic signals). Then, we built meta-models based on machine learning, in order to approximate outcomes of traffic simulations faster and with an acceptable accuracy. First, we obtained very good accuracy of predictions using neural networks (average error 1.18%), later we even improved these results using XGBoost (average error lower than 1%). What is important, results of approximations can be obtained a few orders of magnitude faster than by running traffic simulations.
Finally, I will tell about conclusions and possible directions of a future research. I will show how achieved results may be applied in real-time traffic management and transport planning, which may lead to significant reduction of the required time and cost of traffic engineering works, as well as better results in terms of traffic efficiency. Thus, this may lead to developing AI-based tools able to support traffic engineers in their work. An interesting conclusion is that similar approach may be used in many other cases, in which it is required to run a large number of simulations, e.g., weather forecast, medicine, bioinformatics. There may be also important theoretical conclusions, related to automata theory and computational complexity.
www.pydata.org
PyData is an educational program of NumFOCUS, a 501(c)3 non-profit organization in the United States. PyData provides a forum for the international community of users and developers of data analysis tools to share ideas and learn from each other. The global PyData network promotes discussion of best practices, new approaches, and emerging technologies for data management, processing, analytics, and visualization. PyData communities approach data science using many languages, including (but not limited to) Python, Julia, and R.
PyData conferences aim to be accessible and community-driven, with novice to advanced level presentations. PyData tutorials and talks bring attendees the latest project features along with cutting-edge use cases.
Видео TensorTraffic - traffic prediction using machine learning - Pawel Gora канала PyData
Together with an international group of scientists, I am developing TensorFlow-based tool, TensorTraffic, for predicting traffic simulation outcomes, such as delays / total waiting times on a red signal, as a function of traffic control settings. We achieved great results, which may find applications in transport planning, real-time traffic management and many other fields.
Abstract
Together with an international group of scientists, I am developing TensorFlow-based tool, TensorTraffic, for predicting traffic simulation outcomes, such as delays / total waiting times on a red signal, as a function of traffic control settings. We achieved great results, which may find applications in transport planning, real-time traffic management and many other fields.
First, I will talk about a goal of the research and problem we would like to solve, which is: approximating traffic simulation outcomes fast and with an acceptable accuracy. The motivation behind this goal is to have a meta-model of traffic, which can allow to effectively evaluate quality of a large number of settings (e.g., traffic control strategies or road infrastructure settings).
Then, I will explain our approach and experiemnts. We built a dataset consisting of almost 120 000 traffic simulation scenarios, using Traffic Simulation Framework software, each case corresponded to a different traffic signal control strategy, simulations computed the total waiting time of all cars on a red signal in a given region of Warsaw (Stara Ochota district, 15 crossroads with traffic signals). Then, we built meta-models based on machine learning, in order to approximate outcomes of traffic simulations faster and with an acceptable accuracy. First, we obtained very good accuracy of predictions using neural networks (average error 1.18%), later we even improved these results using XGBoost (average error lower than 1%). What is important, results of approximations can be obtained a few orders of magnitude faster than by running traffic simulations.
Finally, I will tell about conclusions and possible directions of a future research. I will show how achieved results may be applied in real-time traffic management and transport planning, which may lead to significant reduction of the required time and cost of traffic engineering works, as well as better results in terms of traffic efficiency. Thus, this may lead to developing AI-based tools able to support traffic engineers in their work. An interesting conclusion is that similar approach may be used in many other cases, in which it is required to run a large number of simulations, e.g., weather forecast, medicine, bioinformatics. There may be also important theoretical conclusions, related to automata theory and computational complexity.
www.pydata.org
PyData is an educational program of NumFOCUS, a 501(c)3 non-profit organization in the United States. PyData provides a forum for the international community of users and developers of data analysis tools to share ideas and learn from each other. The global PyData network promotes discussion of best practices, new approaches, and emerging technologies for data management, processing, analytics, and visualization. PyData communities approach data science using many languages, including (but not limited to) Python, Julia, and R.
PyData conferences aim to be accessible and community-driven, with novice to advanced level presentations. PyData tutorials and talks bring attendees the latest project features along with cutting-edge use cases.
Видео TensorTraffic - traffic prediction using machine learning - Pawel Gora канала PyData
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
![Predictive Maintenance & Monitoring using Machine Learning: Demo & Case study (Cloud Next '18)](https://i.ytimg.com/vi/4wneZDEB3VA/default.jpg)
![Day 1: Applications and Techniques of Artificial Intelligence GIS](https://i.ytimg.com/vi/nYmg9_uBrXw/default.jpg)
![Modeling Road Safety with Machine Learning](https://i.ytimg.com/vi/xfx7gqD3ILk/default.jpg)
![Introduction to Forecasting in Machine Learning and Deep Learning](https://i.ytimg.com/vi/bn8rVBuIcFg/default.jpg)
![](https://i.ytimg.com/vi/e37GOzb1krk/default.jpg)
![Intelligent Transportation Systems: At A Glance](https://i.ytimg.com/vi/OnjX0O9dPMc/default.jpg)
![CUDA Explained - Why Deep Learning uses GPUs](https://i.ytimg.com/vi/6stDhEA0wFQ/default.jpg)
![The basics of spatio-temporal graph neural networks](https://i.ytimg.com/vi/RRMU8kJH60Q/default.jpg)
![How A.I. Traders Will Dominate Hedge Fund Industry | Marshall Chang | TEDxBeaconStreetSalon](https://i.ytimg.com/vi/lzaBbQKUtAA/default.jpg)
![Applications of Machine Learning in Supply Chain and Logistics](https://i.ytimg.com/vi/bIOHUltu_r8/default.jpg)
![Intelligent Traffic system](https://i.ytimg.com/vi/6L05EbzlPxM/default.jpg)
![Using Survival Analysis to understand customer retention - Lorna Brightmore](https://i.ytimg.com/vi/aKZQUaNHYb0/default.jpg)
![VIRTUAL MEETING | Machine Learning Applications in the Atmospheric Sciences](https://i.ytimg.com/vi/zZ1WYfEppRA/default.jpg)
![Deep Learning In 5 Minutes | What Is Deep Learning? | Deep Learning Explained Simply | Simplilearn](https://i.ytimg.com/vi/6M5VXKLf4D4/default.jpg)
![Hanna Meyer: "Machine-learning based modelling of spatial and spatio-temporal data"](https://i.ytimg.com/vi/QGjdS1igq78/default.jpg)
![How AI/ML Is Transforming Cybersecurity | Introduction To Cybersecurity | Edureka](https://i.ytimg.com/vi/d31eg547Uuo/default.jpg)
![Prediction Machines: The Simple Economics of Artificial Intelligence](https://i.ytimg.com/vi/Q4o56nufXTw/default.jpg)
![Time Series Forecasting With RNN(LSTM)| Complete Python Tutorial|](https://i.ytimg.com/vi/S8tpSG6Q2H0/default.jpg)
![Machine Learning with Encrypted Data | Homomorphic Encryption](https://i.ytimg.com/vi/nlsd2LO-S50/default.jpg)
![How Computers See the Earth: A ML Approach to Understanding Satellite Imagery (Cloud Next '18)](https://i.ytimg.com/vi/5PNnPagENxQ/default.jpg)