[Purposes]Given the shortcomings of existing construction alignment prediction methods for bridges, a alignment prediction method for continuous rigid frame bridge based on back propagation (BP) neural network optimized by adaptive moment estimation (Adam) was proposed. [Methods]With the project of Xiaowujiang Bridge taken as the research object, the sensitive parameters of bridge construction alignment were determined through the orthogonal test, namely concrete bulk density, concrete elastic modulus, tension control stress, and temperature. With the root mean square error, average absolute error, coefficient of determination, and operation time taken as evaluation indexes, under the same incipient learning rate, the four optimization algorithms of gradient descent, minimum gradient descent, root mean square propagation, and Adam were compared. [Findings]The results indicate that the operation time for the BP neural network optimized by Adam when it finishes convergence is 0.518 s and that the Adam algorithm has a faster convergence speed and a higher fitting accuracy than the other three optimization algorithms. [Conclusions] The proposed method can predict the alignment of rigid frame bridge construction in a relatively accurate way.