The invention discloses a time series data completion system based on social relation perception. A time series data completion system adopted by the method comprises a social attention network and atime series attention network. In a social attention network, firstly, a memory representation matrix with a fixed size of time series data corresponding to each neighbor node is calculated, and the memory matrixes of all neighbor nodes are spliced to be combined with the state of a current hidden layer to obtain a social relation context vector; in the time sequence attention network, memory representation of the user is directly calculated, and a time sequence context vector is obtained in combination with the current hidden layer state; and finally, the current hidden layer state vector, the social relation context vector and the time sequence context vector are spliced and completion value is acquired through nonlinear change. The method is self-adaptive to various time series data scenes with social relation contexts, so that a more universal scheme is provided, and the problem of time series data completion in generalized social relation contexts is solved.