Accurate prediction of ionospheric clutter is of great significance in improving the target detection performance of high-frequency surface wave radar. This paper proposes a short-term prediction model of ionospheric clutter using the Opposite Artificial Rabbits Optimization optimized Gated Recurrent Unit (OARO-GRU) network. Firstly, based on the a priori knowledge that ionospheric clutter received by high-frequency surface wave radar has chaotic characteristics, the input and output sample sets of the GRU network are constructed using the phase space reconstruction technique. Then, two improvement strategies, namely, the opposition-based learning and the Cauchy-based mutation, are incorporated to enhance the optimization capability of the original ARO, which is used to optimizthe GRU network with the values of three hyperparameters including the number of hidden layer nodes, the initial learning rate, and the maximum number of iterations. Finally, the optimized GRU network is retrained and fed into the test sample set for testing. The model is evaluated based on the given evaluation metrics. The experimental results show that compared with the other seven comparison forecast models, the proposed OARO-GRU network model has obvious superiority in prediction accuracy and reliability, and provides a new idea and method for effectively improving the target detection performance of high-frequency surface wave radar.