基于两段式物理信息神经网络的液位测量技术OA

Liquid level measurement plays a crucial role in industrial production and safety monitoring.Howev-er,the capacitive liquid level gauges are prone to systematic measurement errors under complex operating con-ditions involving material adhesion,temperature drift,and variations in medium properties.To address this is-sue,a physics-informed fusion network(PIF-Net)is proposed.The features related to adhesion-induced capac-itance are firstly extracted through a physics-encoding layer,then prior physical knowledge is introduced for a-daptive fusion,and finally the corrected liquid level value is output through an error-compensation layer.To e-valuate the effectiveness of the approach,an experimental dataset covering multiple liquid media and tempera-ture conditions is constructed,and PIF-Net is compared with 4 mainstream machine learning and deep learning baselines.Experimental results demonstrate that PIF-Net consistently achieves the lowest mean absolute per-centage error(MAPE)across various media,exhibiting superior robustness and generalization compared to purely data-driven models.Furthermore,ablation studies reveal that the physics-fusion layer significantly im-proves both convergence speed and final accuracy,confirming the effectiveness of the design,and providing a promising new avenue for high-precision liquid level measurement.