To solve the problem of large volume and storage difficulty of bridge health monitoring data, the deep convolutional autoencoder is proposed to compress data. By designing a damage indicator, structural damage location is identified from the compressed data, thereby ensuring the effectiveness of the data compression. Firstly, an appropriate deep convolutional autoencoder model is designed, and the autocorrelation functions of acceleration responses under the healthy state of the bridge are used as training data to obtain appropriate model parameters. Secondly, the real-time monitored acceleration autocorrelation functions are input into the trained deep convolutional autoencoder model to obtain compressed data. Then, the Euclidean distance between the compressed data under both healthy and real-time states is calculated as a damage indicator, with each damage indicator corresponding to a specific structural location. Next, the damage condition at the corresponding location is determined according to whether the indicator changes. Finally, numerical models of a simply supported beam and a continuous beam, as well as a simply supported beam test in laboratory, are adopted to verify the effectiveness and noise robustnessof the proposed method.