Mobike Scientist: Why do you need artificial intelligence to launch shared bicycles?

AsiaIndustrial NetNews: On April 13th, we can see a large number of shared bicycles every day. In some places, shared bicycles are placed in rows, but in some places, these small cars are not seen. For platforms, where and how many shared bicycles are placed is a question worthy of study. If no one needs the car where it is placed, the delivery effect will be greatly reduced; if people cannot find the car.Mobike Scientist: Why do you need artificial intelligence to launch shared bicycles?

But how to determine where is the right place to put it? After the bicycles are ridden to all corners of the city, how to schedule them to ensure that they can be ridden by more people? Thousands of pictures of faulty vehicles are reported to the platform every day. How to distinguish between true and false? Recently, Yin Dafei, a data scientist at Mobike, shared a discussion on the artificial intelligence methods used in solving the above problems.

The following is a transcript of Yin Dafei’s speech:

good afternoon everyone! I am Dafei Yin, the data scientist of Mobike. I am very honored to share with you some data analysis and data mining, and even some artificial intelligence research that we have done on the Mobike platform in the past few months.

Today’s lecture mainly talks about how we can use artificial intelligence or big data to solve operational problems. Because we have millions of vehicles, how to dispatch scientifically and efficiently, almost all friends who care about us will ask such questions. Anyone who has any questions during my speech can raise their hands at any time.

Our data department not only supports operations, we also support the company’s financial department, customer department, including operation and maintenance. Therefore, it should be said that the artificial intelligence mentioned today is only the application of one department of our company. Today I will share with you how to solve the problem of managing millions of bicycles in the city through scientific scheduling methods and forecasting methods.

First of all, after a period of operation, we found that the supply and demand of bicycles in various cities and regions are extremely unbalanced.For example, taking me as an example, my home is near Jiangtai Road, Jiuxianqiao, and I will find that the subway exit A is very dense to the north.industryThere are a series of large industrial parks such as Hengtong Commercial Institute, 360 and 58 in the same city. Port C is a traditional residential area. Every morning, port A of the bicycle is quickly taken away by people who go to work, and the car at port C is often stagnated. Therefore, our data platform will find through simple analysis that port A is a hot spot, while port C is a cold spot. After doing some simple analysis, in fact, our operation colleagues have optimized the direction and strategy of car investment under our guidance. When putting in a car, put in more vehicles at port A, and intentionally put in less vehicles at port C.

In addition to using historical data to do some regression and strategic calculations, it also does some prediction-related things. We all know that the biggest uncertain factor affecting the supply and demand of bicycles is the weather. In the weather, we do not distinguish whether it is rain and snow, PM2.5, temperature, or wind speed. The amount of influence has the greatest impact.

Mobike Scientist: Why do you need artificial intelligence to launch shared bicycles?

Impact of climate change on order volume

We use GDPP here, which is a machine learning algorithm to budget for weather and order volume. The blue dots are actual orders, and the red dots are forecasts. As you can see, the basic forecast is still in line with this trend.

Mobike Scientist: Why do you need artificial intelligence to launch shared bicycles?

Mobike’s budget for order volume

Our vehicles have GPS, and not only GPS sensors, but also vehicle voltage sensors and other sensors that monitor the state of our vehicles, we can know in real time whether each vehicle is a faulty vehicle or even how long it has not been ridden Row. In this case, we will assist the operators to do some vehicle recovery and optimize operations. One of the other questions is, if we have a lot of vehicles that are faulty, how do we recycle them? There are vehicles in charge of operation. One question is, we have many such dispatch vehicles, how to optimize their paths. Because every car walking aimlessly in the middle of the city is a waste of resources. What our colleagues use is to first determine the location of the faulty car, and then use an optimization technique to minimize the difference between these places and the source of the vehicle’s convenience, so that the operating vehicle can maximize its economic benefits.

In addition, there are still many enthusiastic users to help us report, that is, there is a function of uploading pictures under the report of faults in the APP. Every day, thousands of pictures are uploaded to our server in this way. Our customer service staff has very limited energy. How can he identify tens of thousands of pictures? We used a method, we used TensorFLOW again, a deep learning method for image recognition. For example, the picture on the right is illegally parked in the community. Through learning and accumulating experience, our learning performance has exceeded 97%. That is to say, if such a picture is uploaded to our system, we have more than 97% of the pictures. Probability to judge that it is indeed a picture of illegal parking, we will reward those who take pictures of illegal parking, and we will punish those who ride the car into the illegal parking area.

The main algorithm aspects are introduced to you so much. Our research is still going on. We sincerely welcome everyone to give us valuable opinions on our operation and our algorithm.

Published on 10/09/2022