Linear prediction is a commonly used method in time series analysis. For the traditional one-dimensional linear prediction spectrum estimation algorithm, only the source angle or signal frequency can be estimated. A space-time two-dimensional linear prediction algorithm is proposed. The data are extracted and arranged by the data received by the space-time two-dimensional array, and the data covariance matrix is reconstructed. The space-time two-dimensional linear prediction weight is obtained and the peak search is performed. The paper focuses on the principle of space-time two-dimensional forward prediction, backward prediction and bidirectional prediction algorithm, focuses on the data structure of the constructed space-time two-dimensional linear prediction covariance matrix, and discusses the forward and backward directions. The mutual relationship between two-way prediction and the relationship between two dimension and one dimension is compared and analyzed with the space-time two-dimensional minimum variance algorithm and the space-time two-dimensional MUSIC algorithm. Theoretical analysis and simulation show that the forward, backward and bidirectional prediction of one-dimensional spatial and one-dimensional time domain algorithms are special cases of the space-time two-dimensional prediction algorithm, and the space-time two-dimensional prediction algorithm not only overcomes the shortcomings of the coherent signal source that the space-time two-dimensional minimum variance algorithm and the space-time two-dimensional MUSIC algorithm can not solve, but also has a good direction finding frequency measurement capability.