China News Service, Hangzhou, September 25 (Qian Chenfei) In the workshops of industrial enterprises at night, machines are still running. Although the staff had left after get off work, the other pair of "eyes" did not stop.
Recently, at an industrial enterprise in Ruian, Wenzhou, Zhejiang, "Fire Eye Sentinel" continued to analyze factory surveillance footage. At 18:17:05, the system captured an abnormal fire in the corner of the screen and automatically triggered an alarm. The on-duty personnel confirmed the fire situation after calling real-time monitoring and review, and immediately contacted the person in charge of the company. Four minutes later, staff arrived at the scene and extinguished the initial fire.
From the discovery of anomalies to the arrival of personnel, the "Fire Eye Sentinel" eliminated a potentially expanding fire at its initial stage.
In Zhejiang, more and more such scenes are appearing. At the 2026 Fifth Global Digital Trade Expo held here, the Zhejiang Provincial Department of Emergency Management released four new "AI+ emergency management" scenarios: "AI+ electric bicycle charging and stopping supervision", "AI+ welding operation supervision", "AI+ urban and rural flood prevention" and "AI+ safety insurance services". It can be seen that Zhejiang is no longer exploring just letting AI "see risks", but also trying to let it participate in more aspects of security governance.
From "Human Staring" to "AI Staring"
Industrial enterprises have large production spaces and a lot of equipment, and it is difficult to achieve round-the-clock coverage by relying on traditional personnel inspections.
Zhejiang started from high-frequency risk scenarios and integrated AI visual recognition technology into the company's existing video surveillance system to create a "Fire Eye Sentinel" to identify abnormal situations such as smoke and flames in real time.
"I used to worry that no one would be on duty in the workshop after get off work, and I always had a stone hanging in my heart. This time, thanks to the 'Fire Eye Sentinel', which discovered the fire in time, otherwise I can't imagine how much damage it would cause." said the person in charge of a company that caught fire.
"AI on duty" as mentioned above has covered many enterprises. So far, Zhejiang's "Fire Eye Sentinel" has been connected to more than 125,000 enterprises and more than 613,000 video channels, and has successfully warned of 535 real fires, reducing enterprise losses by about 1.8 billion yuan.
If the risks of industrial enterprises are more concentrated in the production space, then the safety of electric bicycles has entered more dispersed life scenes such as residential areas and shops along the streets.
In May this year, the Zhejiang Provincial Department of Emergency Management and the Provincial Fire and Rescue Corps jointly established a special work team to launch the pilot of "AI+ electric bicycle charging and parking safety supervision". Hangzhou has undertaken the pilot mission of the Ministry of Emergency Management and is exploring the use of technologies such as AI video recognition, current sensing, and elevator control linkage to build a denser risk perception network for typical business formats such as old high-rise residential buildings, rural self-built houses, street shops, and commercial complexes.
Specifically, AI can identify electric bicycles and batteries entering elevators, and trigger voice warnings and linkage elevator controls; video recognition technology is used to detect illegal parking in corridors and fire escapes; current sensors try to identify battery charging behavior through the "power fingerprint" formed by residents' household electricity consumption. At present, the application's home charging recognition accuracy exceeds 80%, and the cumulative number of effective warnings exceeds 65,000.
In August this year, a fire broke out in the charging shed in Shiqiao Nanyuan, Gongshu District, Hangzhou. It only took 93 seconds from the fire to the warning. After the system issued an early warning, multiple forces quickly coordinated and the fire was extinguished in 10 minutes.
From industrial workshops to residential areas, AI is making up for the time and space that manual inspections cannot cover continuously, allowing risk perception to gradually move from "regular viewing" to "continuous viewing".
From "Risk Detection" to "Early Warning"
The discovery of risks is only the first step. How to further move the early warning threshold is another direction Zhejiang is exploring.
In Jiaxing, electric welding work is regarded as a key scene. Risks such as the scattered locations of welding machines, dynamic changes in the work process, illegal fires, and unsupervised operations are difficult to continuously control through manual inspections alone. The local government explores "AI + welding operation supervision" and uses Beidou and GPS dual-mode positioning and AI video analysis to dynamically supervise the location, operating status and operation process of the welding machine.
After the welding machine enters the key area, the system can automatically verify; prior reporting for hot work, dedicated supervision, equipment verification and other aspects are included in online closed-loop control; AI conducts a round of video capture analysis every 10 seconds to identify risky behaviors such as illegal operations and unsupervised operations.
At present, "Welding Guardian" has gathered data from more than 90,000 companies, more than 310,000 welding machines, and more than 160,000 welders to dynamically monitor hot work conditions and automatically warn more than 180,000 times.
The same is true in the field of flood control. Jinhua relies on the "Jiu'an" large model to create an "AI + urban and rural waterlogging prevention" application, which brings together data from multiple sources such as meteorology, water conservancy, construction, and public security, and uses more than 15,000 real water samples to train the AI model to identify and evaluate the risk of waterlogging.
Through AI video recognition and IoT monitoring two-way verification, combined with weather radar and historical water accumulation data, the system can predict high-risk waterlogging points one hour in advance, and the early warning information is pushed to relevant responsible personnel within 30 seconds. After actual testing during this year's plum flood season and multiple rounds of typhoons, the application has issued 438 early warnings in advance, shortened the arrival time of disposal forces by 80%, and increased disposal efficiency by more than three times.
The scope of AI intervention also extends from risk monitoring to production safety liability insurance services.
About 99,000 companies in Zhejiang have purchased safety production liability insurance, with premiums of 480 million yuan and service fees of 130 million yuan. Focusing on enterprise risk assessment, hidden danger identification, report review and other aspects, many places in Zhejiang have piloted "AI+safety and liability insurance services" to correlate multi-source business data, regulatory basis, etc., and develop two assistants, service and assessment.
Data shows that the recall rate of key inspection items for identifying hidden dangers has now reached over 90%; the average response time of intelligent question and answer has been shortened to less than 10 seconds; the review of a single report has been shortened from 30 minutes to less than 3 minutes; and the service evaluation has been reduced from 15 person-days to 0.5 person-days.
From the operating status of a welding machine, to the changing trend of water accumulation on a road, to the rapid response of a safety and liability insurance service, the time node for AI-empowered emergency management is constantly moving forward.
From "a scene" to "a set of tactics"
Behind each specific scenario is Zhejiang’s continuous exploration of the digital transformation path of emergency management.
Since the beginning of this year, the Zhejiang Provincial Department of Emergency Management has relied on the Ministry of Emergency Management's "Jiu'an" large model to promote the integration of artificial intelligence and front-line needs of emergency management. It has launched five rounds of scenario releases, and 12 provincial benchmark applications have been implemented at the grassroots level. Hangzhou, Jiaxing, Jinhua, Wenzhou and other places combine different risk characteristics to explore and form unique application scenarios.
Observing these explorations, their common feature is that they start from real needs.
What industrial enterprises need to solve is risk monitoring during unmanned periods, electric bicycle supervision is concerned with the identification of hidden dangers in residents’ living scenes, welding operations focus on risks during electrical welding operations, and urban and rural waterlogging prevention requires the gathering of scattered meteorological, water conservancy, road and other information...
The technical paths in different scenarios have different focuses, but they are all exploring and solving the same problem: allowing risks to be perceived earlier, allowing information to flow faster, and making disposal more accurate.
At the same time, the data generated from real scenarios continue to feed back algorithm training and model optimization, pushing applications from "usable" to "easy to use" and from single-point exploration to more scenarios.
The relevant person in charge of the Zhejiang Provincial Department of Emergency Management said that the next step will continue to focus on the actual needs of emergency management, iteratively optimize the online scenarios, promote the further integration of artificial intelligence into the entire chain of risk monitoring and early warning, safety supervision and law enforcement, and emergency command and rescue, and improve the practicality, applicability and promotion value of the application.
From "small incisions" to the front lines of safety production and disaster prevention and reduction, AI is changing not only the way of risk discovery, but also promoting the transformation of Zhejiang's emergency management from manual inspection and experience judgment to human-machine collaboration and data support, providing new technical paths for building a strong safety defense line.
