ANALYSIS OF THE IMPACT OF INPUT PARAMETERS ON THE ACCURACY OF GRAIN TEMPERATURE FORECASTING USING NEURO-FUZZY SYSTEMS

Abstract

The article analyzes the impact of input parameters on the accuracy of grain temperature forecasting in storage facilities using an Adaptive Neuro-Fuzzy Inference System (ANFIS). Particular attention is given to identifying the most significant parameters that ensure the high efficiency of the model, including incorporating the current grain temperature as an integral indicator reflecting the energy state of the grain mass. Adding this parameter allowed the model to account for initial conditions and significantly improve the prediction of thermal processes.

In addition to numerical parameters, such as air temperature, relative humidity, and wind speed, the model also considers time series formed based on previous values of grain temperature and environmental conditions. For this purpose, a sliding window with a width of 8 intervals (2 hours) was used, enabling the model to analyze short-term temperature changes and the dynamics of external conditions. This approach enhances forecasting accuracy by accounting for dependencies between the current and previous states of the grain mass.

Special attention is paid to integrating qualitative parameters, such as the type and variety of grain, represented as linguistic variables using fuzzy logic. The use of expert data facilitated the creation of a rule-based system that adapts to the specifics of different grain crops and ensures the model's high flexibility under various storage conditions.

The modeling results demonstrated that the developed ANFIS model with an optimal set of parameters achieves significantly lower RMSE values compared to baseline models such as ARIMA and LSTM. In particular, the model confirmed its superiority in forecasting accuracy for various storage zones, ensuring temperature stability and timely detection of thermal self-heating risks. The obtained results highlight the importance of considering both numerical and qualitative parameters to improve the efficiency of automated grain storage temperature monitoring systems.