Share-Cycle Station Placement Prediction Project

Overview

This project, part of the MDA program at the University of Tsukuba in collaboration with the Tsukuba City Government, aims to predict user demand and determine optimal locations for setting up share-cycle stations across the city of Tsukuba. By leveraging AI technologies and machine learning models, the project seeks to solve real-world problems related to urban mobility and share-cycle infrastructure planning.

The goal is to use advanced prediction models to place share-cycle stations in locations that maximize accessibility and meet user needs. This project also addresses the challenge of vehicle redistribution by analyzing data on bicycle availability, demand, and usage patterns.

Key Contributions

1. Research I: User Demand Prediction Model

Model Improvement: Enhancing Explanatory Variables

1. Port Density: We incorporated variables related to port density within specific buffers (e.g., the number of ports within certain distances). This variable, added by ZHENG, plays a crucial role in understanding the concentration of share-cycle stations and user access points.

2. Weather Data: Weather data was integrated into the model, an improvement led by Inaba-san, to account for external factors affecting bicycle usage patterns.

3. Excess-Zero Problem: Watanabe-san addressed the issue of excess zeros in the data, optimizing the model to handle cases where the demand for bicycles was zero at certain locations or times.

4. Human Flow Data: Additional variables, such as human flow within a 500-meter mesh, were included by ZHENG to enrich the explanatory power of the model.

5. Reanalysis with XGBoost: Li-san reanalyzed the dataset using XGBoost, improving the prediction accuracy by better capturing nonlinear relationships between the explanatory variables.

6. Transformer Model Construction: Li-san also worked on building a Transformer-based model as one of the candidate models to further improve the predictive power.

Data Processing and Application

• Port Data Organization: We organized variables related to port locations, such as land use type, distance to the nearest train station or bus stop, population data, and human flow data. This process ensures a more structured approach to predicting where share-cycle stations should be placed.

2. Research II: Vehicle Redistribution Problem

Data Collection and Analysis

1. Bicycle Data: Data on the number of available bicycles (collected from 2023.08 to 2024.10) was obtained from Tanaka-san, forming the basis for analyzing current usage patterns and vehicle distribution.

2. Bicycle Usage Data: Inaba-san worked on organizing the data related to bicycle usage frequency, a key factor in understanding station demand and usage trends.

3. Vehicle Data Processing: ZHENG contributed by analyzing the availability of bicycles at different times, focusing on empty-port scenarios and understanding the factors behind them.

4. Potential Demand Analysis: Li-san prepared to analyze latent demand, with daily potential demand data being processed. ZHENG visualized the current empty port situations, making it easier to interpret usage trends.

Output and Scenarios

1. Demand vs. Supply Analysis: We identified the gaps between daily potential demand and the number of available bicycles, aiming to highlight mismatches between the number of vehicles and the actual demand.

2. Scenario Proposals:

   • Scenario 1: Adjust the number of redistributed bicycles based on the demand.
   • Scenario 2: Propose the relocation of certain share-cycle stations.
   • Scenario 3: Adjust the frequency of vehicle redistribution operations.
   • Scenario 4: Add more share-cycle ports, bicycles, and racks to meet increasing demand.

AI Technology Application and Data Scraping

The integration of AI technologies like XGBoost and Transformer-based models plays a central role in addressing complex urban mobility challenges. Through advanced data analysis, these models can predict user demand patterns, optimize vehicle redistribution strategies, and ensure that share-cycle stations are strategically positioned throughout Tsukuba. This data-driven approach not only enhances the efficiency of the share-cycle system but also significantly improves user accessibility and satisfaction.

To support this initiative, automated data scraping is used to collect real-time information from relevant sources such as share-cycle usage data, traffic conditions, and user behaviors. By employing a custom test.py file, we can automate the data collection process from official websites and APIs, ensuring that the most up-to-date information is always available for AI model training and evaluation. This method allows for continuous monitoring and adjustment of AI predictions based on real-time trends, leading to more responsive and adaptive urban infrastructure planning.

This project exemplifies the application of AI in real-world settings, where data-driven solutions directly contribute to enhancing public transportation systems and creating smarter, more efficient urban environments. The success of these technologies in Tsukuba's share-cycle program highlights the potential for similar approaches to revolutionize mobility in other cities.