According to the website of New Scientist magazine on the 20th, British companies such as Deep Thinking have developed an artificial intelligence (AI) to optimize the cooling system of buildings, which can control the cooling system of buildings under different weather conditions while minimizing energy consumption. The test results show that it reduces the energy cost of cooling buildings by about 10%. Relevant research is published on the preprint website. Deep Thinking, Google and Trane Technologies, a manufacturer of building control systems, have developed this AI system to control the cooling system of university campus buildings and buildings with apartments, restaurants and stores. The team trained AI to constantly find ways to minimize overall energy consumption without sacrificing the comfort of indoor personnel. Scientists first "fed" the AI system with one year's historical data of each building's standard control system and trained it. AI system can further understand the situation by controlling the cooling device of the building, while monitoring its impact, weather patterns and changing cooling demand level. On this basis, AI uses the information learned to generate decisions and score them. The test results show that the system can save 9% to 13% of the cooling cost of the studied buildings in three months from summer to autumn in 2021. The researchers said that most buildings still rely on simple control systems, which require humans to manually adjust settings and cannot automatically adapt to climate conditions and other changing factors. One of the key benefits of using AI in air-conditioning systems is that it can constantly monitor and adapt to the changing external world. Cooling homes and buildings accounts for about 10% of the world's energy demand. Therefore, although the energy saving of a single building is relatively small, large-scale promotion may have a significant impact. Researchers said that it is not easy to do this. A lot of manpower is needed to help AI understand specific control systems in different buildings, and some knowledge needs to be pre programmed into AI to create learning shortcuts for it. (Liu Xinshe)