For sake of guaranteeing the stable operating of power system, synchronous condenser (SC) is configured in ultra-high voltage direct current (UHVDC) converter station for affording dynamic reactive power support to power system. Focusing on reactive power regulation system of SC with strong coupling, multivariable, and nonlinear, it is difficult for the universal analytic method to exactly build reactive power output model of SC. Reactive power output model of SC in UHVDC converter station based on interlaced superposition convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) is presented. The excitation voltage, excitation current of SC are used as inputs for training and test sampled data, and the reactive power of SC is used as outputs for training and test sampled data. CNN model based on interlaced superposition is superimposed of two convolution units with different structures to extract the feature of sample data, which are transformed into feature vectors and input into BiLSTM model. Bayesian optimization method is used to optimize its hyper parameters. The predicted results indicate that the proposed reactive power output model of SC in UHVDC converter station based on interlaced superposition CNN-BiLSTM improves modeling accuracy and generalization ability, and provides a reference for SC reactive power regulation control.