In the last week of November, the National Supercomputer Forum was held in Pereslavl-Zalessky. For three days people told and showed how things are going with the development of supercomputers in Russia and how technologies tested on supercomputers are turned into goods.
Corresponding Member of the Russian Academy of Sciences Sergei Abramov spoke about the “Sensitive House” project (November 27). Developing the concept of a “smart home,” he suggests observing home equipment, building and remembering patterns of its behavior, learning from its mistakes, and predicting its condition and problems in advance.
The Institute of Software Systems of the Russian Academy of Sciences, under the leadership of Sergei Abramov, began creating “sensitive houses” in 2014, when the reform of the Academy of Sciences required bringing academic projects to the commercial market. By this time, the IPS RAS had good developments in sensor networks and equipment control, and was developing cloud technologies and machine learning.
According to Sergei Abramov, residential and industrial buildings are filled with equipment on which the well-being of the home and the quiet work of people depend. Although this “smart” equipment develops into a “smart home”, it does not have automatic control. Owners do not know the status of the devices and cannot conveniently monitor them. All that remains is to manually care for the entire infrastructure, like a huge Tamagotchi, regularly checking and adjusting the machines.
Is the smart home working correctly? Or is it time to intervene? Will there be an accident soon? By itself, no “smart home” solves this problem; to answer such questions, automatic supervision and analysis is needed. Therefore, the computer system created at the Institute collects statistics from sensors, builds behavior patterns of household machines and learns to recognize these patterns. By distinguishing normal behavior from problematic behavior and detecting abnormal operation, artificial intelligence will alert the homeowner in time to a potential threat.
A “sensitive home” is a “smart home”, to which sensitivity, the ability to self-learn, the ability to accumulate a pattern of correct behavior, the ability to predict and react have been added.
(Sergey Abramov, Corresponding Member of the Russian Academy of Sciences)
We are accustomed to the way a “smart home” maintains its parameters: set temperature and illumination, constant air humidity, stable mains voltage. A “smart home” can operate according to a script depending on the time of day or event (for example, it will close a gas tap on a command from a gas analyzer). “Sensitive Home” takes the next step - analyzes sensory data and builds new scenarios for classification: everything is going as before or there are surprises. It reacts to changes in the external environment and predicts possible failures, guessing anomalies in the simultaneous actions of different devices. “Sensitive House” monitors the results of its work, warns of problems and changes the scenario, giving hints to the owner and allowing the owner to turn off faulty appliances.
We solve the problem of atypical behavior of equipment.
(Sergey Abramov, Corresponding Member of the Russian Academy of Sciences)
The proposed system relies on a sensor network that provides time-based measurements. For example, a diesel boiler occasionally turns on and heats up the water, a circulation pump drives hot water through heating pipes, and primary sensors report how these devices consume electricity. Based on a series of readings, the secondary sensor (program) compares them with the normal profile and diagnoses failures. The tertiary sensor (program) receives the outside air temperature and predicts the future operation of the system, assesses its load and efficiency - how the heating of the boiler and the weather relate. Maybe the windows are open and the boiler is heating the street, or maybe the efficiency has dropped and it’s time for preventative repairs. Based on the drift of derived parameters, one can predict at what time they will go beyond the norm.
By assessing the simultaneous readings of the sensors, the “sensitive house” is able to notice that the water pump does not turn off because it is pouring water back into the well (through a faulty valve) or directly onto the floor (through a burst pipe). The diagnosis will be even more reliable if the motion sensors are silent and the pump pumps water into an empty house.
Sensor networks are also found in smart homes. Cloud infrastructure is also available in smart homes. But what “smart homes” do not have is artificial intelligence, machine learning, the accumulation of patterns of correct behavior, classification and forecasting.
(Sergey Abramov, Corresponding Member of the Russian Academy of Sciences)
The cloud part of the “sensitive house” is based on the NoSQL database Riak or the Akumuli database, where time series of readings are stored. Receiving and issuing data is done on the Erlang/OTP platform, it allows you to deploy the database on many nodes. A program for mobile applications and a web interface is deployed above it to inform the customer via the Internet and telephone, and next to it is a program for data analysis and behavioral control. You can connect any time series analysis here, including those based on neural networks. Thus, all control over the “sensitive home” systems is placed in a separate management layer. Access to it is provided through your personal account in the cloud service.
Erlang provides all the benefits of a functional approach. It has mechanisms for distributed operation, and the easiest way to make a parallel distributed program is to use Erlang. Our architecture contains software “secondary sensors”; there can be several of them per physical sensor, and if we count on tens of thousands of clients with dozens of devices, we will have to process a huge flow of data. They need lightweight processes that can be launched in huge quantities. Erlang allows you to run tens of thousands of processes on a single core; this system scales well.
(Sergey Abramov, Corresponding Member of the Russian Academy of Sciences)
According to the developer, Erlang is easy to organize a diverse team of programmers, in which students and luminaries create one system. Individual fragments of the software system crash with an error, but the entire system continues to work, which allows you to correct erroneous areas on the fly.
The “sensitive home” system uses all the technologies that IPS RAS used to control supercomputers. This includes electronic sensors, monitoring and remote control systems. Currently, the sensitive program runs on its own sensors and can connect to fire department loops, but there is a plan to collect data from the sensors of any “smart homes.”
“Sensitive Home” is interesting because complex intelligent solutions for the city, neighborhood and home are coming to the forefront. What’s interesting here is not to build a supercomputer, but to construct a social-computer complex, introducing a supercomputer into everyday life, so that the machine changes people’s lives.
(Olga Kolesnichenko, Ph.D., senior lecturer at Sechenov University)
By the spring of 2020, developers will prepare a basic set of programs and equipment to assemble systems of various sizes in buildings and apartments. They promise that the result will be easy to set up, no more complicated than a robot vacuum cleaner. The basic kit will support any supervised equipment: heating boilers, water heaters, refrigerators, water pumps and septic tanks. Then it will be the turn of small-scale sales, then fabless production, addition of new sensors and modules. And in the future, all kinds of diversification and adaptation are possible - a sensitive farm, a sensitive hospital, a sensitive ship, and even a very sensitive tank.
Text:
Portrait:
Source: habr.com