Recurrent Penalized Splines for Real-Time Forecasting of the Parameters of Technical Systems
Елена Алексеевна Кочегурова, Anastasiia Kaida, Maria Galkina · 2023
Forecasting time series (TS) from their historical data is used in many preventive monitoring and decision support systems in industry. For systems operating in real-time, an urgent problem is to reduce forecasting time without compromising other quality indexes (accuracy and trend continuation). The purpose of this paper is to describe a forecasting model based on a recurrent penalized P-spline. The spline model has a compact computational scheme where the efficiency is achieved by using a short input of historical data a simple mathematical description. The model was compared to 11 most well-known forecasting algorithms. The performance of the proposed model compared quite well to the competitive forecasting algorithms. The model was used to forecast a real-world TS of a technological parameter of a boiler plant.