Aiming at the problem of “ pseudo regression” in traditional machine learning, ignoring the long-term dependence between non-stationary sequences, the cointegration theory and vector error correction model were proposed to predict the performance degradation trend of rolling bearings, and then predicted the remaining useful life (RIL). Firstly, extracted the kurtosis value, peak-to-peak value and root mean square value from the vibration signal, and analyzed its stability. Then, tested the cointegration relationship of the time domain features and established vector error correction models based on the test results, and performed the residual analysis. The analysis results were close to the Gaussian white noise distribution, indicating that the models were good. Finally, used the models to predict the bearing performance degradation trend and applied RUL, and RMSE, MPE, MAPE and the unequal coefficients of Theil to quantitatively evaluate the prediction results. Experimental results show that the proposed vector error correction model has higher prediction accuracy than the differential average moving autoregressive-Kalman filter model, and simplifies the modeling process.