Mostly, miRNA's are identified by short-RNA sequencing and then running pipelines like miRDeep. I guess it is possible in principle to blastN all miRbase precursors against your genome, then further check the hits using cmscan against all Rfam. However, this is not a good approach because it will both suffer from large numbers of false positives and low sensitivity. The question is what you want to do with the predicted miRNAs. If you run target prediction on hypothetical data the result will be totally unreliable, for differential expression analysis, you would need small-RNA seq anyway.

If you want to try how well this works, you could try this pipeline on a well characterized organism:

• download genome sequence
• download all miRBase pre-miRNA
• remove all pre-miRNA that are from this species and keep as test set
• BlastN all precursors against the genome
• extract hit sequences
• annotate hits using cmscan and keep seqs annotated mi-RNA
• compare with test set and evaluate the performance