Online updated CSI feedback method and device

The invention relates to the technical field of CSI (Channel State Information) feedback algorithms, and particularly provides an online updated CSI feedback method and an online updated CSI feedback device, which are based on incremental learning and specifically comprise the following steps: S1, generating downlink CSI data; s2, performing data cleaning; s3, dividing the data set into training data of five tasks; s4, outputting a batch of training data Bn belonging to the task c from the data stream, and inputting the Bn and each sample of the replay subset Msub into an encoder to obtain feature sets gamma n and gamma m; s5, calculating the average similarity between each characteristic value in gamma n and gamma m; s6, mixing the input batch Bn and the replay batch Bm, and inputting the mixture into the model for updating; s7, updating the replay set of the existing category; s8, scoring the samples according to the distance between each sample and the class center; and S9, repeating the steps S4 to S8 unt