When hazardous materials are detected, the most important thing is to quickly and accurately determine their source and predict the direction in which they will spread. Scientists from the National Center for Nuclear Research’s Complex Systems Division are working to speed up this process. They propose models based on artificial neural networks (ANN), which will allow pollution monitoring in real time.
Materials from nanotubes will stop balls and space debris
The result was a light, solid and durable material made of nanotubes, which preserves projectiles more effectively than the already known polymer – Kevlar or steel plates …
For several years, the HACCP has been working on algorithms to determine the source of pollution, based on data on substance concentrations released from a network of detectors. The main task of emergency response groups located in all cities is to respond quickly to all threats to people and the environment. The main factor in the success or failure of any activity is reaction time – Report to NCBJ.
Today, a variety of chemicals are used in most industries, making the transportation and storage of toxic substances in constant danger of being released into the atmosphere and causing them to occur. radioactive contamination. Situations where sensors located throughout the city report a non-zero concentration of a hazardous substance, of unknown origin, present a major challenge. In such cases, it is important that the system be able to estimate the most likely location of the pollution source in real time, based only on concentration data from the sensor network.
Algorithms that deal with the task can be divided into two categories. The first are algorithms based on the backward approach, i.e. analysis of the problem starting from its last stage, but intended for open areas or a problem on a continental scale. The second class are algorithms that rely on sampling the relevant parameters of the dispersion model (parameters such as the location of the source) to determine which class gives the smallest difference between the output data and the actual concentration measurements made by the detector network. This approach is to use sampling algorithms to find the optimal parameters of the dispersal model, based on the comparison of the model results and the detection of contamination. Due to the effectiveness of the parameter scanning algorithm used, each rebuild requires multiple typical runs. Urban reconstruction, of primary interest to researchers, requires advanced dispersal models that take into account wind field disturbances around buildings. Computational fluid dynamics models are the most reliable and accurate. However, it poses a very difficult computational challenge. We must realize that in order to find the most likely source of pollution, the dispersal model must be run tens of thousands of times, the NCBJ points out.
This means that the model used must be fast so that it can be applied to a real-time backup system. Assuming, for example, that the average time to perform scattering model calculations in an urban area is 10 minutes, it would be difficult to perform a full reconstruction using it in an acceptable short time.
The quasar surprised researchers. It has a black hole that is 117 million times heavier than the sun.
Determining the distance to space objects allows you to estimate how much energy they produce and the processes responsible for their illumination. Really…
artificial nerve network
The solution to this problem, which Dr. Anna Warziuchak-Saban of the Menhas Center for Risk Analysis at NCBJ is working on, in collaboration with the Institute of Computer Science of the University of Life Sciences and Humanities in Siedels, is to use an artificial neural network in the reconstruction system, rather than the scattering model. . The point is for the artificial neural network to be effective in simulating the transport of pollutants in urban air. If this works, the ANN can act as a dispersion model in a system that identifies the source of the pollution in real time. – The main advantage of ANN is a very short response time – notes Wawrzyńczak-Szaban.
– Of course, the ANN must be trained on the static structure of the city, using real meteorological conditions, using an appropriate and documented (documented and consistent with assumptions) dispersal model. This process requires several simulations that act as training datasets for the ANN. Training an ANN is computationally expensive, but after training, the method will be a quick tool for estimating point concentrations for a given pollution source. – He explains.
The center presented the results of neural network training based on the data, teaching the spread of airborne toxins in the center. LondonUsing the test field for the DAPPLE field experiment 2). The ANN training data was generated using the Quick Urban and Industrial Complex (QUIC) dispersion model. – We tested different ANN structures, i.e. the number of its layers, neurons and activation functions. Tests conducted confirmed that a trained ANN can simulate the turbulent transport of airborne toxins in a highly urban area, Wawrzyńczak-Szaban explains.
Moreover, we have shown that the use of ANN can reduce the response time of the reconstruction system. The time required by the ANN presented in the paper to estimate thirty minute gas concentrations at 196,000 sensor points was 3 seconds. In the case of the QUIC model, the time was estimated to be at least 300 seconds, which gives us a 100-fold acceleration of the computations. With this in mind, the reconstruction time in an actual emergency situation may be short, resulting in rapid locating of the pollution source, explains the researcher.
“Infuriatingly humble musicaholic. Problem solver. Reader. Hardcore writer. Alcohol evangelist.”