The invention discloses a history segmented sequence search and time sequence rarefaction-based wind power forecast method. The method comprises the following steps of: normalizing wind power historydata and determining window width optimum values of searched segmented time sequences according to power time sequence fluctuation features and basic statistical characteristics; calculating matchingdegrees between all the history segmented time sequence and a segmented time sequence at the current moment by taking the newest segmented time sequence at the current moment as a standard and synthesizing a correlation index and a similarity distance index; sorting the matching degrees according to a big-to-small sequence, and determining a number of optimum history segmented sequences accordingto an average matching degree aggregation principle; aiming the current segmented time sequence at each moment, determining a number of corresponding optimum history segmented sequences and a number of corresponding optimum average history segmented sequences; aiming at all the moments for training the time sequences, establishing a time sequence rarefaction-based power forecast model; and solvingthe model by adoption of a multiplier alternating direction method to obtain parameters of the model, wherein the parameters are used for future power forecast.