Application of Machine Learning in Predicting Frozen Ground Temperatures Under a Building with Pile Foundation

Global warming processes have a particularly strong impact on northern regions with permafrost. In these areas, the rate of warming exceeds the global average, leading to permafrost degradation. One consequence of this process is the reduced bearing capacity of soils surrounding the pile foundations of critical infrastructure, including residential buildings with pile foundations and ventilated underground. To monitor changes in soil bearing capacity, automatic thermometric stations are installed in the ventilated undergrounds of residential buildings. Additionally, specialized software applications are developed for long-term forecasting of pile bearing capacity trends using temperature monitoring data. Typically, such software requires significant computational resources, often taking tens of hours to run on high-performance supercomputers. To address this challenge, we developed a neural network that efficiently predicts soil temperatures based on ventilated underground temperatures. The model is designed to run on conventional computers, making it accessible to researchers without advanced computational mathematics expertise. This paper describes the network’s architecture and presents prediction results that align well with monitoring data.