Written by | Yunxi
Editor| Brother Meng
Image source| Respondents and networks
On March 6, the last confirmed case of new crown pneumonia in Shijiazhuang was cured and discharged from the hospital, and the case was "cleared".
Two months ago, the outbreak of the second round of the epidemic made Shijiazhuang Gaocheng District a household name overnight. Fortunately, the whole people of Shijiazhuang have one heart, and in one month, Shijiazhuang, which is in the eye of the storm before the Spring Festival in 2021, will return to tranquility.
Although it is not as high-profile as Wuhan in 2020, the battle against the epidemic in Shijiazhuang is not easy. All sides rushed to the aid, including a group of engineers. Now life is back to calm, listening to the dusty people talk about those days.
"Here comes, it is to fight"
Mao Yajun
Alibaba Cloud Intelligent Digital Government Division
1
No one expected that the epidemic in Shijiazhuang would come so fiercely.
On January 5, my colleague Feng Qi and I drove from Zhangjiakou. The situation ahead was unclear, and Feng Qi wanted me to stay in Baoding to go by myself. I did not promise, at the moment of the epidemic, no one can stay out of the matter.
At noon in Shijiazhuang, there was no one on the street. At the moment, the only lively place is the anti-epidemic command center. The city will be locked down on the 6th, and we must set up an anti-epidemic team within one day. Hebei colleague Shu Feng immediately pulled a "Aliyun anti-epidemic front support group" and pulled the boss of various business units. Soon, the group approved free public cloud computing power resources all the way to the green light. Comrades-in-arms also came from various places, Yang Wenkui came from Shanghai; Ke Qiang and his wife and children were on vacation in Sanya, and they came to support without saying a word.
Mao Runner-up at work
Last year, Hebei was not the main battlefield, and the epidemic prevention system built at that time was no problem. But in the face of today's exponential explosion of high concurrency, the system has a single function and cannot withstand a huge concurrency, like a van that has to transport a train of goods overnight.
The surest approach, of course, would be to develop a new platform. But the epidemic can't wait, and it changes every moment. If the anti-epidemic command center cannot grasp the situation in time, it may miss the fighter.
We decided to build on the original platform and build on the new features. Nucleic acid detection, circulation traceability, isolation site control, and every link of epidemic disposal has complex logic.
It also meant that we had to build many new bodies while the vans were running non-stop. The original architecture could not meet the needs, and we had to redevelop the body from 0 to 1, and make it not only hanging on bread, but also adapting to the locomotive of the future. This also puts a huge pressure on the engineers in the rear.
In the last round of the epidemic, I provided remote support, and at that time I did not understand the urgency of the students in front of me. Now that I'm on the front line, I really understand the pressure.
But I was luckier than my classmates before. Last year's epidemic made the whole society see the energy of science and technology to fight the epidemic. From the idea to the idea, everyone has changed a lot.
In one and a half days, we built a comprehensive epidemic prevention system platform.
2
The virus spread rate is extremely fast, a confirmed patient's close contact personnel more than 100 people, of which there may be thirty people at risk of infection, they will contact hundreds of people in their own trajectory, suspected population is growing geometrically. The government has set up a large-scale outbound call team, but because the dialect has reduced the efficiency of communication and the number of problems that need to be understood, the average person can only make 10 phone calls per day, and everyone is anxious.
Feng Qi and I both thought of the intelligent voice technology of Dharma Academy, and the intelligent epidemic robot can replace the heavy manual outbound call.
Dharma Academy responded quickly and immediately organized a team to tune the algorithm. The key to intelligent outbound calls lies in whether speech recognition is accurate. Many places in Shijiazhuang speak dialects. In order to improve the accuracy rate, all of us have done the work of marking the semantics of the speech and helping the machine learn the local dialect.
But some of the problems are beyond imagination. One of the questions we set up was "whether to do nucleic acid testing." But the answers "do", "did not do", "negative", "masculine" are pronounced similarly in the local dialect, and even native Shijiazhuang people have to work hard to distinguish, not to mention machines.
This is a very critical part, so we set up a question "I don't understand, you can say yes or no". But someone replied and repeated the same answer as before.
In fact, if the inquiry call takes more than 2 minutes, the answerer will lose patience and hang up. So we ended up reducing the number of questions to less than 6.
No. 5 began to do testing, No. 6 tuned the algorithm, coordinated the resources of all parties, and the comrades in arms who were connected and moved also worked until 4 a.m., and the No. 7 intelligent epidemic robot was officially put into use.
Today, intelligent robots can make 400,000-500,000 calls a day, with a connection rate of more than 80%, and bear 80% of the outbound call workload.
But it's not the happiest. I remember Feng Qi saying that it was the happiest thing to learn that everyone was safe and sound.
We came to war.
"The city has come back to life, and I've never been in vain."
Kai Zhang, Intelligent Algorithm Engineer of Alibaba Cloud
"The resurgence of the epidemic in Shijiazhuang is in urgent need of support from the algorithm team." At 12:00 p.m. on January 4, I was still on a business trip in Qingdao, and I immediately packed up after receiving the call. Say it is clean up, in fact, there is nothing, just a computer, a pair of jeans, straight to the high-speed rail station in the morning.
The carriage was empty, and I slowed down and breathed into my home, "I may not be able to go back for the New Year." The wife was silent for a moment: "You go, there is me in the house." Do a good job, usually I will not call you, I will call us when I am homesick. ”
At 5 p.m. on January 5, the high-speed rail arrived at the station. The radio behind it sounded at the right time: "There is an epidemic in Shijiazhuang, please do not get off the bus." "I took a step down, and everyone around me looked at me with strange eyes.
It's false to say I'm not afraid, last year I was at home for remote support, this year is the first time to go to the front, all the way to the very nervous. But as soon as I arrived at the scene, I forgot all about it, the sound of keyboards and discussions came and went, and everyone looked tense.
On January 6, Shijiazhuang was locked down, and the entire area of Gaocheng District was adjusted to high risk. The city's primary and secondary schools and kindergartens are closed. The first round of nucleic acid testing was carried out by all employees, and 354 positive cases appeared.
Everyone understands that the great war is coming.
I originally planned to make a set of epidemic close connection algorithm models for Shijiazhuang, using the algorithm model to find close contacts, trace the source of the virus, and predict medium and high-risk areas. Before departure, Alibaba Cloud turned on the green light all the way and set up the cloud environment in advance within 2 hours. As long as the data is in place, the model will be able to make a big difference.
But I thought it was all too simple.
Data spans many departments and is difficult to reconcile in the short term.
The epidemic close connection algorithm model is a weapon, data is like ammunition, and the fight against the epidemic is like a battlefield. If the data is not used as quickly as possible, the model is meaningless.
I was confused at once, so what did I do?
The epidemic is changing every day, no one around me has flinched, and I can't stay out of the matter and decide to rebuild a system on the spot.
I dragged my local colleagues all over the CDC, the Health Commission, and the information offices of various departments to understand the situation. In the center of the storm, everyone is spinning in a row. Even so, I heard that we could help, and they had a hard time eating, and after a few bites, they put down and talked to us in detail.
I remember going to the Shijiazhuang Cd-operative Center one afternoon, the front-line personnel were still doing the flow adjustment in the front, specifically for us to rush back, and after less than 15 minutes of talking, we had to rush to the front line. After the end, I rushed back again and called other colleagues to explain together. Their eyes were bloodshot, the phone calls were one after the other, and they took the time to ask me, "How's it going, did it help you?" ”
We couldn't bear to delay for too long, and before we left, the other party kept saying to us, "Please, please, don't let Shijiazhuang wear out any longer." ”
"Please," that's the word we hear the most. The last epidemic made the whole society realize the importance of science and technology to fight the epidemic, and I know that everyone has high expectations for us, and we can't live up to this trust.
Back at the hotel, my colleagues and I continued to take stock, and our thinking gradually became clear - the richest thing now is the flow data, which we can use to make flow relationship analysis and global analysis models.
Zhang Kai (first from left)
If the epidemic close connection algorithm model is based on global data, finding close contacts and traceability, then the flow adjustment relationship analysis model is based on the existing flow adjustment reports of thousands of confirmed personnel, deducing the close contacts and conducting traceability analysis.
In the past, each circulation survey was presented separately in a file, containing detailed information and paths of confirmed personnel, with complicated information, and it was difficult to grasp the center by the naked eye, and it was difficult to dig out the internal links and assist decision-making.
We will extract all the important elements of the circulation report, analyze it with an algorithm, and associate all the reports, and through people and places, time, and organization, dig out the scattered associations to form a network, trace and display.
Once a confirmed person is found, through the flow relationship analysis model, it can lock in his close contacts.
3
On January 10, the program was officially launched. Although I am the only one on the front line, the algorithm team behind me is also shouldered together. Everyone is helping part-time, no matter how late they are still online. At Ali, it is incumbent upon everyone to do anything for the pandemic.
Two days later, 90 new cases appeared in Hebei, and the previously latent confirmed cases began to appear. This is the most new day since the lockdown. Fortunately, they all appear in the quarantine area.
On January 14, Shijiazhuang conducted the second nucleic acid test, and the positive case dropped to 247, which was different from the first time in the face of the epidemic's confusion and panic, this time everyone seemed to be in order.
On January 20, after 10 days, the flow adjustment relationship analysis model landed. With a click of the mouse, information is scattered in the network structure.
We continue to build a global analysis system. The flow-tone relationship analysis model can help frontline personnel track individual cases, while the global analysis algorithm model shows the global situation through three-dimensional images, locks key areas and super communicators, and helps medical experts and governments make decisions.
The third round of nucleic acid testing in Shijiazhuang arrived two days later, with 30 positive samples. Everything is good.
On January 25, the epidemic prevention command center decided to promote this system as a whole. Two days later, the algorithm model of the global analysis system, which took a week, was completed. The big stone in my heart finally landed.
On 3 February, there were no new cases in Shijiazhuang.
On February 8, Gaocheng changed from a high-risk area to a medium-risk area. The whole city is gradually coming to life. The station and high-speed rail station have resumed their hustle and bustle.
In the 30 days in Shijiazhuang, every day from 10 pm to 2 am is the meeting time, and after the end of the internal review work and planning. For us, there is no difference between day and night, coffee is picked up as water to drink, tired enough to take a nap in the conference room.
Hard work is true, but it is precious to have such an experience in life. After the circulation relationship analysis model was launched, the leader of the Health Commission was very satisfied, and I could feel that he really regarded us as comrades in arms in the same trench, not just partners.
Now, it's finally time to go home for the New Year and change into jeans.
The city came back to life, and I didn't come in vain.