The invention discloses a lncRNA-protein association prediction method based on projected neighborhood non-negative matrix decomposition. First, according to the lncRNA sequence, lncRNA expression profile data, protein sequence, GO function annotation data of protein and the interaction network of lncRNA and protein, calculate A variety of lncRNA features, protein features, lncRNA similarity matrix and protein similarity matrix are obtained. Secondly, the integrated lncRNA similarity network is obtained by fusing multiple lncRNA similarity networks, and the integrated protein similarity network is obtained by fusing multiple protein similarity networks. Finally, combining the integrated lncRNA(protein) similarity network and multiple lncRNA(protein) features, a feature-projected neighborhood nonnegative matrix factorization algorithm is proposed to predict potential lncRNA-protein interactions. The present invention can not only accurately predict new lncRNA-protein interactions, but also predict new proteins (lncRNA) that are not associated with any lncRNA (protein), effectively avoiding high human and material resource consumption caused by biochemical experiments.