ANALYSIS AND COMPARISON OF METHODS FOR FORECASTING THE PARAMETERSOF GRAIN DRYING PRODUCTION PROCESSES

Abstract

The paper presents a systematic analysis and comparison of methods for forecasting parameters of grain drying production processes. The methods considered include statistical forecasting, physical modelling, neural networks, fuzzy logic and adaptive neuro-fuzzy inference systems, and hybrid approaches. For each class of methods, advantages and limitations in application to grain drying control tasks are characterised. It is shown that classical methods (ARIMA, PID controllers) do not provide sufficient accuracy due to the nonlinear and non-stationary nature of the process. Fuzzy logic and ANFIS-based approaches demonstrate the possibility of interpretable control with a reduction in the number of model parameters. Deep learning methods (LSTM, GRU, GCN+Transformer) provide effective modelling of spatio-temporal dependencies of the temperature field. Hybrid approaches, in particular the combination of mechanistic and data-driven methods, allow achieving a maximum moisture deviation at the outlet within ±0.58–0.3%. A comparative table of methods and a classification block diagram are provided. The conclusion is drawn on the expediency of developing intelligent hybrid forecasting systems for grain dryer control tasks. The article systematises key results of published research and identifies directions for further development in intelligent drying control systems.