To improve the accuracy and stability of lithium-ion battery core temperature estimation and achieve better battery state monitoring, in this paper, the performance of different machine learning models in lithium-ion battery core temperature estimation is estimated. The experimentally obtained core temperature data of the batteries are preprocessed using three methods: empirical mode decomposition (EMD), complementary ensemble empirical mode de-composition with adaptive noise (CEEMDAN), and variational mode decomposition (VMD). Under different temperature and cycling test conditions, the root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and maximum absolute error (MAXE) metrics of the traditional convolutional neural network-long short-term memory (CNN-LSTM) model are compared with those of the EMD-CNN-LSTM, CEEMDAN-CNN-LSTM, and VMD-CNN-LSTM models. The results show that models processed by EMD and CEEMDAN perform better in RMSE, MAE and MAXE compared with the traditional CNN-LSTM model, while the VMD-CNN-LSTM model performs optimally in RMSE, MAE, MAPE and MAXE, showing high efficiency and stability in signal extraction and noise suppression.