基于深度学习与电化学模型的锂电池SOC估算方法OA

To address the challenge of balancing model accuracy and robustness in lithium-ion state of charge(SOC)estimation,a multimodal synergistic estimation framework integrating electrochemical mecha-nisms and deep learning was proposed.Firstly,a high-fidelity multiphysics coupled model was constructed to solve nonlinear partial differential equations,such as solid-phase diffusion and charge conservation,generating training data that captured microscopic kinetic features like electrode phase transitions and electrolyte gradients.Subsequently,a hybrid neural network architecture integrating a Convolutional Neural Network(CNN)and a multi-head attention mechanism was innovatively designed.Using terminal voltage,current,and neural net-work-inferred variables such as negative electrode potential and open-circuit potential as inputs,this architecture captured dynamic features through multi-scale dilated causal convolutions and outputted the SOC estimates via fully connected layers.Furthermore,an Iterated Unscented Kalman Filter was introduced for online correction,effectively suppressing noise and model errors.Experimental results demonstrated that the proposed framework exhibited excellent performance under complex operating conditions,such as wide temperature ranges and high C-rates.Specifically,under dynamic conditions,the Mean Squared Error was reduced to 0.17%,and the Mean Absolute Error was maintained within±1.3%.Consequently,this framework significantly enhances both the ac-curacy and robustness of SOC estimation.