关键词: 异位发酵床, 垫料, 可见—近红外光谱, 水分检测, 神经网络, 算法优化

Abstract: To address the need for rapid moisture detection in ectopic fermentation bedding materials, this study explored the feasibility of developing a moisture prediction model using visible-near infrared spectroscopy (Vis-NIR) technology. Over a period of 4-5 months, samples of bedding materials were collected, and spectral data ranging from 400 nm to 990 nm were obtained using a spectrometer. Key characteristic wavelengths were selected using the Competitive Adaptive Reweighted Sampling (CARS) algorithm. Subsequently, a Backpropagation (BP) neural network model was then constructed and optimized using three optimization algorithms: Grey Wolf Algorithm (GWO), Harris Hawk Algorithm (HHO), and the Guanhao Pig Algorithm (CPO). Among these, the CPO algorithm demonstrated the best optimization performance. To further enhance model accuracy, the Particle Swarm Optimization (PSO) algorithm was improved by integrating Chebyshev chaotic mapping, resulting in the CARS—ICPO model. The final model achieved R2 values of 0.993 5 and 0.995 6 on the validation and prediction sets,respectively, with corresponding RMSE values of 0.011 and 0.009, indicating excellent predictive accuracy and generalization capability. These findings confirmed the feasibility of Vis-NIR spectroscopy combined with advanced machine learning techniques for moisture control prediction in ectopic fermentation bedding materials, offering a novel approach and technical support for rapid detection and intelligent management of fermentation systems.

Key words: ectopic fermentation bed, padding material, visible near infrared spectroscopy, moisture content detection, neural network, algorithm optimization