Japanese mobile operator NTT Docomo is using artificial intelligence (AI) for allocating shared micromobility vehicles and replacing depleted batteries used by these vehicles.

The Sharing Operation Optimisation System was adopted last month by Docomo Bike Share, the provider of a bicycle-sharing service that allows users to reserve electric-assist bikes at the most convenient bike station, ride around the city, and then choose their preferred return station. The system will be gradually deployed throughout Tokyo to manage the company’s shared-bike fleet.

Micromobility-sharing services let users rent bicycles and other small, lightweight vehicles and then return them to any station operated by the service. As the number of sharing-service users grows globally, the corresponding increases in vehicles and renting and returning stations is making it difficult to ensure vehicles in each fleet are optimally allocated and equipped with charged batteries at all times.

Docomo’s system uses AI to generate optimised plans for collecting and reallocating vehicles and replacing spent batteries. The system uses machine learning to simulate vehicle movements to predict the availability of vehicles and charged batteries at each station. Maintenance personnel can then use tablets or other mobile devices to view precisely which vehicles need to be trucked to other stations and which batteries need to be replaced to improve operational efficiency.

The system forecasts rental and return trends based on data such as in-use and returned vehicles, weather forecasts, date and time, travelling distances between stations, and each truck’s storage capacity as well as quantities of vehicles and batteries on board any truck at any time. Using this information, it generates a reallocation plan, including the best transport routes. The system enables personnel with less experience to function as efficiently as experienced staffers. It is also expected to help operators develop efficient operating routes in new territories.

Going forward, Docomo will continue to assess the system’s performance, including its forecasting accuracy and the effectiveness of its recommended routes for bike reallocation, based on which the company expects to upgrade and adapt the technology for supply-and-demand optimisation in other fields.

There are three parts – demand forecasting, simulations and reallocation planning.

Demand forecasting technology makes hourly predictions of how many vehicles will be in use and how many will be available at each station over the next 24 hours. In addition to real-time data, the system takes into account other data such as weather forecasts and dates and times. It has been shown to have the capacity to forecast changing variations in vehicle demand accurately.

Simulation technology uses multi-agent simulations to project the precise movements among stations by the shared vehicles and their logistical-support trucks. The movement of each vehicle is forecasted using real-time data and statistics, including the probability of vehicles moving between specific stations and predicted demand at each station. Using these input values, the system forecasts the number of vehicles at each station and the remaining charge of each vehicle’s battery. Simulations are run every ten minutes to enable the reallocation plan to be continuously updated.

Reallocation planning technology uses the simulation results to generate joint optimisation plans for vehicle collection and reallocation and battery replacement at each station, including the order in which trucks should visit stations and which transport routes to take. Forecasts of conditions in coming hours support the prioritisation of battery replacements at the busiest stations, collections at stations where vehicle returns are expected to spike, and so on, helping inexperienced personnel work more efficiently and operators develop efficient operating routes in new territories.

NTT Docomo is Japan’s leading mobile operator with over 86 million subscriptions.