Abstract: In the context of the rapid development of new energy and energy storage systems, moment, while a 1D-CNNextracts local temporal features from a sliding window of multivariate sequences. The features from both branches are subsequently concatenated to achieve fused prediction, enabling end-to-end SOC estimation. Experimental results show that the proposed model achieves a root mean square error (RMSE) of 1.48 and a coefficient of determination (R 2 ) of 0.996 on the test set. Compared with single-branch CNN, MLP, and long short-term memory (LSTM) models, the CNN-MLP fusion model demonstrates significant improvements in mean absolute error (MAE), RMSE, and R 2 , validating the effectiveness of the dual-branch decoupling and feature fusion strategy in enhancing the robustness and accuracy of SOC estimation. high-precision estimation of the state of charge (SOC) of lithium-ion batteries is crucial for battery safety and energy management. To address the challenges posed by strong nonlinearity, dynamic hysteresis, and noise interference under complex operating conditions, this paper proposes a SOC estimation method based on a convolutional neural network-multilayer perceptron (CNN-MLP) dual-branch fusion network. The method employs a decoupled modeling approach: a multilayer perceptron captures the nonlinear mapping of current, voltage, and temperature at the present

Key words: lithium-ion battery; SOC estimation; 1D-CNN; MLP; feature fusion; deep learning

摘要： 在新能源与储能系统快速发展的背景下，锂离子电池荷电状态（SOC）的高精度估计 对于电池安全与能量调度具有重要意义。针对复杂工况下 SOC 存在的强非线性、动态滞后及 噪声干扰等问题，文章提出了一种基于卷积神经网络-多层感知机（CNN-MLP）双分支融合 网络的 SOC 精确估计方法。该方法通过解耦建模，一方面采用多层感知机对当前时刻的电流、 电压及温度进行非线性映射；另一方面利用一维卷积神经网络对滑动窗口内的多变量序列进 行局部时序特征提取。随后，通过特征拼接实现融合预测，实现 SOC 的端到端估计。实验结 果显示，该模型在测试集上的均方根误差（RMSE）为 1.48，决定系数（R 2）达到 0.996。与 单支 CNN、MLP 及长短期记忆网络（LSTM）模型对比分析结果表明，CNN-MLP 融合模型 在平均绝对误差（MAE）、RMSE 和 R 2 等指标上均有显著提升，验证了双分支解耦与特征融 合策略在提升 SOC 估计鲁棒性与精度方面的有效性。

关键词: 锂离子电池；SOC 估计；一维卷积神经网络；多层感知机；特征融合；深度学习