Climate Change Time Series: Predicting Travel Demands Based on Origin and Destination

In the heart of California, San Francisco's bustling streets are a testament to the city's vibrant life. The taxi fleet in this city, provided by Uber, a leading ride-hailing service, forms the basis of an intriguing case study in forecasting passenger demand.

The city map is meticulously divided into 10,000 grid cells for spatial decomposition, allowing for a detailed analysis of the taxi trips. Each cell represents a sub-region, and the origin and destination of each occupied taxi trip are recorded. The resulting data set is then reconstructed to contain origin, destination, and the origin timestamp of each passenger trip.

Mining floating car data, a crucial task in intelligent transportation systems, presents a challenging problem due to the spatial and temporal dependencies inherent in GPS data. However, the advent of graph neural networks has provided a promising solution. These networks have been increasingly used to forecast traffic conditions, as each pair is correlated with the neighboring OD pairs or surrounding roads.

The goal of this problem is to model where people are moving to, given their origin. This complex task can be broken down into four sub-tasks: spatial grid decomposition, selection of origin-destination pairs, temporal discretization, and modeling and forecasting. For simplicity, a subset of the top 50 OD grid cell pairs with the most trips is taken into consideration.

Temporal discretization is achieved by counting how many trips occur in each hour for each top pair. The time series resulting from this process shows a daily seasonality, primarily driven by rush hours. This pattern, when understood, can help reduce congestion, overall transportation activity, and greenhouse gas emissions.

A Keras example is available to learn how to implement this method using graph neural networks. For those interested, this tutorial uses GPS data from 536 taxis over a period of 21 days in San Francisco, California, USA. The data allows for counting how many trips go from cell A to cell B, providing valuable insights into the city's mobility patterns.

Moreover, an ARIMA method can be used for simplicity in building the forecasting model. This approach, when combined with graph neural networks, offers a promising solution for forecasting how many passengers want to make the trip relative to a given OD pair.

In conclusion, the use of graph neural networks and spatial decomposition techniques in forecasting taxi passenger demand in San Francisco presents a significant step forward in understanding and optimizing urban mobility patterns. This research not only has practical applications for reducing congestion and emissions but also paves the way for future advancements in intelligent transportation systems.

Climate Change Time Series: Predicting Travel Demands Based on Origin and Destination