The measurement and control device of a UHV converter station is a piece of secondary equipment with nonlinear analog and high requirements for transmission and conversion. The current evaluation and prediction methods are not fully suitable for the analysis of such a device. A health assessment and prediction method based on wavelet kernel principal component analysis (KPCA) and a bidirectional long-term and short-term memory network (Bi-LSTM) is proposed. The wavelet kernel function is introduced to improve KPCA's feature extraction of influencing factors on the state of health. A health index is established through the first nuclear principal component to evaluate the state change of the device. The purpose of health prediction is achieved by constructing a Bi-LSTM network model to input characteristics information. Taking the real data collected by a converter station in Zhejiang as the sample, the experimental data are compared and analyzed. The results show that this method can improve the accurate evaluation and prediction accuracy of multidimensional health monitoring data, and provide a scientific reference for maintenance personnel in formulating maintenance strategies. This work is supported by the National Natural Science Foundation of China (No. 51777119).