Novosibirsk State Technical University (NSTU) has developed an algorithm for constructing meteorological forecasting models using neural networks to improve the effectiveness of weather forecasting.
Industry, agriculture, urban planning, tourism, and many other industries require accurate and up-to-date information about weather conditions, climate change, and the state of the environment. It is based mainly on observations and data obtained from meteorological stations, satellites, sensors and other devices. However, existing approaches have their limitations due to the large amount of data, the complexity of its processing, and space and time constraints.
According to Boris Malozemov, Associate Professor of the Department of Electrical Engineering at NSTU-NETI, Candidate of Technical Sciences, new approaches to weather analysis and forecasting have become possible due to the development of neural networks and machine learning. Neural networks can process large amounts of data, and are able to account for complex interactions between various factors such as temperature, atmospheric pressure, humidity, and wind, allowing for more accurate weather forecasts for short and long periods of time.
Scientists at NSTU-NETI have proposed using a method of grouping neural networks, in which several independent neural networks are trained to perform the same task in order to improve the quality of forecasting. It is important to select carefully their structure and parameters and train them on a sufficient amount of diverse data to achieve good results. These datasets should include information about past weather events, as well as data collected in real time.
"One of the ways to use grouping neural networks for weather forecasting is to use the batch method. In this case, neural networks are trained on different subsets of the source data with different characteristics, such as time intervals and geographical areas. Each forecasts of theirs are then combined to produce the final weather forecast. This approach allows us to take into account the nuances of weather conditions in different areas, which is necessary for planning daily life, providing agricultural work, construction, effective management of water, energy and other resources," Boris Malozemov said.
An important aspect in weather forecasting is the use of various inputs such as weather station data, satellite observations, geographical and historical data. Grouping neural networks allows you to combine information from these sources.
The scientist explained that in order to achieve higher forecast accuracy in all weather seasons, averaging or weighing the contributions of each neural network can be applied. This allows smoothing out possible errors in individual models and taking into account different forecast scenarios for different weather seasons.
"We built a predictive model using the method of grouping neural networks, which allowed us to take into account a larger number of dependent variables than individual neural models, and ultimately improve the accuracy of forecasting. To date, the accuracy of most existing weather forecasting models is limited to 120 hours, while we have managed to increase this period to 180 hours," Boris Malozemov emphasized.
He also added that this method may have the potential to be applied in other areas requiring accurate forecasting, such as financial markets, transportation systems, and energy.
In the future, scientists plan to continue studying meteorological forecasting models in order not only to improve the order and accuracy of input data, but also to change the mathematical basis for building the model itself.