Energy-saving control method for air conditioning in mushroom room based on a simplified PINN model

This study addresses the limitations of current Model Predictive Control (MPC) methods in mushroom room energy management, particularly the neglect of spatial temperature heterogeneity and poor physical interpretability in temperature prediction models, by proposing an energy-saving control approach based on a simplified Physics-Informed Neural Network (PINN). A multi-dimensional indicator system integrating variance contribution rate, data entropy, temperature field reconstruction RMSE, spatial coverage, and sensor quantity was developed to efficiently select representative measurement points. A simplified PINN model with a fully connected network architecture was designed, incorporating Navier-Stokes equations and energy conservation equations as physical constraints to enhance prediction consistency and extrapolation capability. An MPC framework was established with a 3-minute minimum equipment on/off duration as a hard constraint, optimizing both temperature deviation and energy consumption. Experimental results demonstrate that with only six sensors, the method achieves a temperature field inversion accuracy of RMSE 0.267°C, outperforming gated recurrent unit (GRU) Neural Network and fully connected networks. Compared to conventional threshold control, energy savings of 9.8 %-14.6 % were achieved across different seasons and growth stages of Pleurotus citrinopileatus , confirming adaptability under variable thermal loads. Weight adjustment in the MPC objective function further explored operational ranges. The study provides an interpretable and practical solution for efficient climate control in industrial mushroom cultivation. • A multi-index system was built for efficient selection of key temperature points in the mushroom room. • A simplified Physics-Informed Neural Network was developed using Navier-Stokes and energy equations. • An MPC framework was set with a 3-min equipment cycle constraint to optimize temperature and energy use. • The method's energy-saving performance and stability were tested across seasons and temperature setpoints.