Knowing which Americans have installed solar panels on their roofs, and why they do so, will be very useful for managing the changing American power system and for understanding the barriers to using renewable resources. But so far, all available data are basically estimates. To get accurate numbers, scientists at Stanford University used machine learning algorithms to analyze more than one billion high-resolution satellite images and identified almost all solar installations in these 48 states. The research results were published in the "Joule" magazine published on December 19. These data are publicly available on the project website. The analysis found that there were 1.47 million photovoltaic power generation devices, which is much higher than the two previously accepted estimates. The scientists will also combine the US census and other data with their solar facility catalog to determine the factors that led to the use of solar power. We can use the latest advances in machine learning to understand the location of all these assets, which is a huge problem, and we can have a deep understanding of the development direction of the power grid and how we can help bring it to a more favorable place, Ram Rajagopal said. He is a professor of civil and environmental engineering and oversees the project with Arun Majumdar, a professor of mechanical engineering. Who uses solar energy Data from these groups may be useful to power companies, regulatory agencies, solar panel marketers, and others. Knowing how many solar panels are nearby can help local power companies balance supply and demand, which is the key to grid reliability. The list highlights the catalysts and obstacles to solar deployment. For example, researchers found that family income is very important, but only to a certain extent. With an annual income of more than $ 150,000, income will soon no longer play a big role in people ’s decision-making. On the other hand, low- and middle-income families do not often install solar energy systems, even if the area where they live is profitable in the long run. For example, in areas with sufficient sunlight and high electricity bills, the electricity bill saved will exceed the monthly equipment cost. The author suspects that the obstacles for low- and middle-income families are upfront costs. This finding shows that solar installers can develop new financial models to meet unmet needs. To cover socio-economic factors, members of the research team used publicly available US census data. Each of these brochures covers approximately 1,700 households on average, about half of the zip code, or about 4% of a typical American county. They found other valuable things. For example, once solar energy penetrates to a certain level in a community, it will take off quickly, which is not surprising. However, if there are many families with unequal income in a community, this activator will usually not start. Using geographic data, the research team also found an important threshold, that is, how much sunlight a particular area needs to cross the important threshold for adoption. Majumdar said: "We found some insights, but we think this is just the tip of the iceberg that other researchers, power companies, solar developers, and decision makers can further discover. We will make this public so that others can discover Solar energy deployment model, and establish economic and behavioral models. " Discover the panel The team trained a machine learning program (AI) called DeepSolar to identify solar panels by providing approximately 370,000 images, each of which covers approximately 100 feet by 100 feet. Each photo is tagged with "is there a solar panel". Since then, DeepSolar has learned to identify features related to solar panels—for example, color, texture, and size. Yu Jiafen, a PhD student in electrical engineering who built the system together with Wang Zhecheng, a PhD student in civil and environmental engineering, said that we did not actually tell the machine which visual characteristics are important. All of this needs to be done through machine learning. In the end, DeepSolar was able to correctly identify images containing solar panels in 93% of cases, and about 10% of images with solar equipment installed were ignored. The author said in the report that DeepSolar is more accurate than the previous model in these two aspects. The research team then asked DeepSolar to analyze billions of satellite images to find solar installations—a work that would take years to complete using conventional technology. Through some new means to improve efficiency, DeepSolar completed the work within one month. The resulting database includes not only residential solar installations, but also installations on the roofs of enterprises, as well as solar farms owned by many large power companies. However, the scientists let DeepSolar skip the most sparsely populated areas, because the buildings in these rural areas are likely to either have no solar panels or not be connected to the grid. Scientists estimate from their data that approximately 5% of residential and commercial solar installations exist in uncovered areas. Wang said that the progress of machine learning technology is amazing. But off-the-shelf systems usually need to be adapted to specific projects, which requires expertise in project topics. Both Jiafen and I are focused on using this technology to serve renewable energy. Next, the researchers plan to expand the DeepSolar database to identify solar devices in rural areas and other areas with high-resolution satellite images. They also intend to add some features to calculate the angle and direction of the solar device, so as to accurately estimate the amount of electricity generated. DeepSolars' size measurements are currently only representative of potential outputs. The organization hopes to update the US database with new satellite images every year. This information will eventually be used to optimize power systems in various regions of the United States, including the Rajagopal and Yus projects, to help power companies visualize and further analyze distributed energy resources. Original paper: https: // S2542435118305701 (Originally from: Daily Solar China New Energy Network Comprehensive) Plastic Tap,Water Tap,Plastic Water Dispenser Faucet,Deck Mounted Plastic Faucet Cixi Ruisheng Electric Appliance Factory , https://www.rswatertap.com
The analysis found that there were 1.47 million photovoltaic power generation devices, which is much higher than the two previously accepted estimates. 