Current methods for predicting the remaining useful life of relay protection device suffer from issues such as the lack of accurate assessment and prediction for individual device states, and the inability to timely correct prediction data according to actual operating conditions, resulting in unreliable prediction results. To address this, a method for predicting the remaining useful life of relay protection devices based on the IRIME-BP-LSTM model is proposed. First, operational experience and procedural requirements are summarized to establish a set of state assessment indicators for protection devices, forming the initial input vector. Then, the Cauchy mutation strategy is introduced to improve the rime optimization algorithm, which is used to optimize the initial parameters of the backpropagation (BP) neural network. Next, the initial input vector is assigned to the optimized neural network to assess the condition of the protection device, forming a representation vector of the device’s operating state and constructing a time series. Finally, the constructed time series is fed to the long short-term memory (LSTM) network for predicting the remaining useful life of the protection device. Case study results show that the proposed method has higher accuracy in predicting the remaining useful life of protection devices and can provide theoretical guidance for relay maintenance and operation decision-making.