Accurate monitoring of lithium-ion battery temperature is crucial to ensuring efficient and safe operation. Traditional single temperature prediction models are somewhat limited when faced with large volumes of battery data. To address the challenge of feature extraction and deep temporal scale mining in lithium-ion battery temperature prediction with massive data, this paper proposes a surface temperature prediction model for lithium-ion battery packs in energy storage stations based on an improved Convolutional Long Short-Term Memory (ConvLSTM) network. In the Long Short-Term Memory network, a Convolutional Neural Network is introduced to extract temporal-scale features and the interrelationships among various temperature-influencing factors, with Inception_V3 selected as the optimal network structure through comparative experiments. To adapt to changes in lithium battery temperature data, a parametric deformable exponent is proposed to improve the activation function. Experimental results demonstrate that the proposed improved ConvLSTM model exhibits higher prediction accuracy compared to both the single LSTM model and the conventional ConvLSTM model.