The invention discloses a weight adjustment method and a power prediction method based on multiple numerical weather prediction sources. The weight adjustment method based on the multiple numerical weather prediction sources comprises the steps of 1, determining the type n of the numerical weather prediction sources and a weight coefficient [epsilon]i of each numerical weather prediction source; 2, setting the number of days [days], performing initialization, and enabling a to be 1; successively inputting data wi of the n types of the numerical weather prediction sources in the ath day; 4, calculating power prediction data fai; 5, obtaining a weight coefficient [epsilon]ai of the ith type of the numerical weather prediction source in the ath day; 6, performing adjustment and modification on the weight coefficient; 7, storing the modified weight coefficient [epsilon]ai, if a is greater than or equal to the [days], entering a step 8, otherwise, adding 1 to the value of a, and entering the step 3; 8, obtaining an averaged one-dimensional array array_avg_pre[]; and 9, performing normalization processing, obtaining a processed array array_avg[k1, k2,...kn], wherein k1, k2,...kn are elements in the array, and k1, k2,...kn correspond to the modified weight coefficients of the numerical weather prediction sources. The power prediction accuracy is improved according to the invention.