Sugarcane juice extraction prediction with physical informed neural networks

Abstract

Sugarcane crushing is a complex process with multiple factors, multiple objectives, strong coupling, large nonlinearity and uncertainty. Due to the ambiguity of the pressing mechanism and the unexplicability of data-driven, the process index of sugarcane pressing process is difficult to predict. In order to solve this problem, this paper combines deep learning with pressing mechanism, and establishes a process index prediction model of sugarcane pressing process based on physics-informed neural network (PINN). Firstly, the constitutive model of sugarcane was established based on the pressing mechanism. Combined with the porous medium control equation and numerical simulation, the sugarcane pressing mechanism model was established and verified, which provided high-quality simulation data for subsequent research. Secondly, the porous medium control equation is embedded into the PINN model as a physical law to establish a unique loss function of the sugarcane pressing process. Combining the historical data and simulation data of the workshop, a large sample data is made to train the model, and the model is compared with the common data-driven model to further illustrate the accuracy and stability of the established model.