Grain storage temperature prediction based on chaos and enhanced RBF neural network

Area of Science:

• Agricultural Engineering
• Artificial Intelligence
• Time Series Analysis

Background:

• Grain storage necessitates precise temperature control due to strict requirements.
• Existing temperature prediction models suffer from nonlinear characteristics and low accuracy.
• Accurate temperature prediction is crucial for minimizing stored grain loss.

Purpose of the Study:

• To develop an advanced temperature prediction model for grain storage.
• To address the limitations of nonlinear dynamics and prediction inaccuracies in current methods.
• To enhance the efficiency and accuracy of grain storage temperature monitoring.

Main Methods:

• Application of chaos theory to determine embedding dimension and time delay of temperature sequences.
• Utilizing Lyapunov exponent to confirm chaotic properties and phase space reconstruction for data analysis.
• Development of a q-Normalized Least Mean Square Fourth (qXE-NLMF) algorithm to enhance radial basis function (RBF) neural networks for improved weight updating, prediction accuracy, and training speed.

Main Results:

• The enhanced RBF (ERBF) network demonstrated faster convergence and lower steady-state error than traditional RBF networks in Mackey-Glass chaotic time series prediction.
• The proposed C-ERBF model achieved higher prediction accuracy for grain storage temperature series compared to other time series prediction methods.
• The model successfully predicted grain pile temperatures in advance, enabling proactive control measures.

Conclusions:

• The C-ERBF model offers a significant improvement in predicting grain storage temperatures.
• Proactive temperature management based on C-ERBF predictions can substantially reduce stored grain consumption.
• This predictive capability helps prevent spoilage and ensures better grain quality preservation.