Bearings as one of the essential components of the motor, the influence of operating conditions of bearings complex and variable, resulting in diversity in the characterization of monitoring data and difficulty in state prediction. To address the challenge of forecasting the remaining useful life of bearings under varying operating conditions, this paper introduces a prediction model grounded in a nonlinear stochastic process that utilizes fusion degradation indicator. Firstly, essential features of the signal data are extracted from multiple perspectives to form an initial feature set. Subsequently, candidate feature set is grouped and clustered based on cosine similarity. Then the typical features are selected among the same trend feature classes, and a fusion degradation indicator is constructed by using weighting algorithm. Finally, we estimate the bearings remaining useful life using a nonlinear stochastic process, with model parameters estimated based on parameter initialization and maximum likelihood estimation method. The experimental results obtained from the PHM2012 dataset show that the designed degradation indicator has clear nonlinear trend and similar failure thresholds under different operating conditions, with excellent generalization and improved prediction accuracy.