第四屆中國-中東歐國家（17+1）創新合作大會于2019年10月7日至9日在塞爾維亞共和國首都貝爾格萊德市舉辦，中國科技部部長王志剛、塞爾維亞創新與技術發展部部長奈納德·波波維奇等與會國部長級領導，以及國內相關產業協會、創新型企業和科研院校共聚貝爾格萊德，商議中國與中東國家科技創新合作。

銀江股份智慧交通研究院副院長徐甲，受中國智能交通協會之邀，代表股份公司和銀江研究院出席會議，推介公司最新創新戰略與技術產品，并接受塞爾維亞當代科學與數據科學家研究所所長——亞歷山大·林克·喬爾杰維奇（以下簡稱主持人）主持的專題訪談。

訪談中針對AI技術在城市交通治理方面的應用展開深入探討，對杭州交通信號配時中心的成功案例進行分析。杭州作為銀江股份在大型城市交通路網信號配時優化服務探索方面的先行城市，擁有最好的團隊和資源，結合AI技術全方位改善城市道路擁堵問題，科技與服務并行，在融合AI技術與交通治理的探索路程中具有深度研究價值。

主持人提出了4個企業在人工智能轉型以及應用方面的問題，以下為訪談內容（英文原文與中文翻譯）：

Djordjevic：What does an AI-powered organization stand for you and what experience do you and your company have with the topic of AI?

徐甲: Nice to meet you dear host of respect. As you may know, our company, Enjoyor is not an AI startup company, it has over 2 decades’ experience in the ITS construction, software engineering and system integration. During the past 2 decades, many projects of high quality have been settled down and lots of customers have been cultivated to pursue finer services. That means the service has to be more scientific to be able to tackle down their sophisticated realistic problems in a global view and also can be tailored to suit for their unique business processes. Under such background, our company has proposed and started up the AI strategy to meet the costumers’ needs. So our company is exactly the type of companies who is experiencing the AI transformation.As for the experience with AI topic, the most crucial thing we’ve realized is that we should always put the industrial practicality of AI techniques in the first place, which is to know the capability as well as the capability boundary of AI techniques in order to use them properly. Secondly, we should provide affordable AI techniques for the customers, in view of the necessary data and computational power.

Djordjevic:Dear Jia, You come from a country which has 42 cities with more than 2 million people living there. I can’t even imagine how something that we see as an easy fix and normal thing can become so exponentially complicated - just because of the numbers itself. That is why I cannot think better interlocutor for the topic of traffic, transportation and travel services than you.May I ask you to give us an overview of what are the major challenges you are facing and how does AI may solve problems in the transportation sector? Can you also tell us more about some concrete project your company is working on?

徐甲：Indeed, the city I live, Hangzhou has a population of 10 million, and more than 3 million of registered vehicles. The volume of vehicles in transit during daily peak is about 300,000 in average.The complexity of the problem comes from many ways. The scale of the problem is our first challenge. Hangzhou is a provincial capital city, which has a large amount of domestic or international activities and events. Meanwhile Hangzhou is a world-famous tourist city, the West Lake attracts huge amount visitors, the complex scenario caused by these factors is our second challenge. The traffic network we are facing to is consists of about 2000 signalized intersections, a considerable part of them are suffering from large traffic flow, imbalanced demand, and high dynamic in the same time, which is our third challenge. Last but not least, we are facing some engineering challenge, such as multiple source date collection and processing, modelling and managing the traffic channelization.As for the cases that we are using AI to tackled down real problem, I’d like to share with you an example. We use AI technologies in the traffic signal control to eliminate traffic congestion in Hangzhou and Nanchang. This sophisticated process generally includes 5 sections, that is perception, early warning, optimization, implementation, and security check.In the perception section, we use AI techniques to estimate the reliability of the raw data. Then to repair and provide the most reliable and direct date to the following sections. In the early warning section, the crucial thing is to estimate the trend but not the status of the traffic, therefore the predictive AI method. In the optimization section, we use to methods. One is to mimic the traditional manual operation in data analyzing and regulation induction and preparation of schedules, in order to address the general and recurrent problems. The second method is to mimic the traditional manual operation in investigating the traffic status, and adjust the plan to implement temporary control strategies.In the implementation section, we use an unified model in adapting our AI optimization methods to all existing controlling systems in the two cities. And finally in the security check section, we check the generated paraments again if they meet all experience constraints and span of control.The final effect is that we dramatically improved the productivity of operators, and improved traffic signal control performance of the two cities.

Djordjevic:What is your standpoint how does AI can benefit the public and the communities? Now and in lets say 2025, as we mentioned previously?

徐甲:1.The Big Data and AI technologies will assist the planning and construction of the infrastructures in a better way. Intelligentized infrastructure will emerge in a larger number, which would play the role as hub among the vehicles and information terminals of passengers.2.AI-powered platforms will connect all the devices, infrastructures, processing all the data, and to optimize the utilization of the infrastructures. Passengers will have better experience during travelling, feel more convenience.3.Cooperative vehicle and infrastructures will spread in some area. Autonomous vehicle driving technology will be wider applied. Autonomous vehicle driving and shared mobility will combine, in the scene of auto charging and auto scheduling.4. The AI-powered mobility service will be more personal, providing more precise travel guidance, travel reservation, with connecting the shared mobility with intelligent infrastructures.

Djordjevic:Before we go to the questions from the public, I would like to ask you to give one present to our public - in the form of advice. What would be the first step if the company would like to start their AI-powered transformation today? And the second one is why to go into this process today and not tomorrow or in the near future?

徐甲:The first thing is to in-depth understand your business model, data basis and the capability of AI technology, also the capability boundaries as mentioned before.The second one is simple, that is because the productivity. If properly used AI techniques can bring significant productivity in comparison to the traditional methods, as a company, why not immediately?

主持人：您所在的公司——銀江股份是如何實現AI驅動的？您和貴公司在實踐AI議題時的體驗是怎樣的？

徐甲：主持人您好，銀江股份在智慧交通的工程建設、軟件開發和集成等領域有20多年的經驗。在過去的20多年里，我們沉淀了大量的優質項目，也培育了相當多的客戶，在獲得客戶信任的同時也對我們的服務提出了更高的要求。具體來說，客戶需要我們科學、專業的服務以從全局角度解決城市交通路網復雜的實際問題，并同時能夠量身定制方案從而滿足各個城市不斷變化的道路交通需求。正是在這樣的背景下，公司提出并發起了AI驅動戰略，以滿足客戶需求?？傮w而言，我們的公司正是在經歷AI轉型與變革的那一類公司。關于推行AI的體驗，我想我們最重要的經驗就是永遠把技術的工業實用性放在第一位，首先要知道AI的能力和該能力的邊界是什么，才可能將AI的技術用在合適的地方；其次是要提供客戶能夠負擔的AI技術，包括所需要的數據和算力成本。

主持人：親愛的徐博士，您所在的國家有42個人口超過兩百萬的城市。我甚至無法想象這些數字會給現實中的問題帶來多少指數級的復雜度，因此我非常想和您來聊一聊關于交通、運輸和出行服務的議題。我想請您全面介紹一下AI如何解決交通中的問題，以及面臨的主要挑戰是什么？最好能向我們多介紹一些夯實的案例？

徐甲：確實，在我所居住的城市——杭州，擁有超過一千萬的人口，機動車保有量超過300萬輛，日間最大機動車在途量達到約30萬輛。除了人口規模這個因素之外還有很多原因導致城市交通治理成為一個難題。杭州是省會城市，有很多國內、國際的大型活動，同時杭州是國際知名的旅游城市，西湖景區聞名遐邇，這些因素帶來的復雜場景是第二個挑戰。我們所面對的城市路網包含大約2000個信號控制交叉路口，這其中的相當一部分同時具有大流量、不均衡、高動態的特點，這是我們面臨的第三個挑戰。最后，還有一些工程上的挑戰，比如多源數據的采集處理，路口渠化的模型化管理等。關于我們使用AI技術在解決的交通問題，我想舉個例子，就是通過交通信號優化管理緩解城市交通擁堵的問題，這個技術在杭州和南昌都得到了應用。這個復雜過程大體上可以分為感知、預警、優化、執行、安全多個環節。感知環節主要在于數據的處理，我們通過AI方法判斷數據的可靠性，將直接、可靠的數據傳遞到后續環節；預警環節重點在于對趨勢的判斷，而不是狀態的判斷，我們使用了各種預測方法；優化的環節我們通過兩套方法，第一套方法模擬傳統人工方式中的分析歷史數據、總結規律和制定時刻表計劃的過程，處理普遍性的問題；第二套方法模擬傳統人工方式中分析實時局勢，采取臨時策略的過程，處理臨時性問題；執行的環節重點在于使用統一模型適配這兩個城市所有的信號系統底層設計；安全環節在于再次確認算法生成的方案中所有參數的有效性和控制幅度。最終的效果是我們顯著地提高了人工的處理時效和生產力，也改善了上述兩個城市的交通信號控制狀況。

主持人：你的觀點是什么？人工智能如何使公眾和社區受益？現在就讓我們設想到2025年，就會像我們之前提到的那樣了嗎？

徐甲：1、 大數據和AI輔助技術會讓基礎設施的規劃設計更加合理，智能化的基礎設施將大量涌現，對接那時的交通工具和信息終端；

2、 AI驅動的精細化管控與優化會讓基礎設施的利用率進一步提高；

3、 車路協同技術將在一些地區推廣，自動駕駛技術將得到更多應用；

4、 AI驅動的出行服務將提供更加個性化、更精準的出行指引、出行預約等服務，串聯起共享出行工具和智能基礎設施；

主持人：如果公司想轉型為AI驅動的模式，首先應該做什么事？為什么今天要立即做AI轉型，而不是等到明天再做？

徐甲：首先是深度了解自己的業務模型與數據基礎，以及如前所述，了解AI方法的能力范圍和能力邊界。第二個問題很簡單，那就是考慮到AI帶來的生產力提升。如果正確使用AI技術，與傳統方法相比，可以顯著提高生產力，降本提效，所以作為一個公司來說，為何要等到明天？