Abstract: In order to quickly and accurately determine the strength of cemented backfill，a strength prediction model of cemented backfill based on PSO-BP was constructed and verified with the strength test data of cemented backfill at 7 d and 28 d of curing. The results show that the initial weights of BP neural network are optimized by combining particle swarm optimization algorithm，which greatly improves the ac⁃ curacy and reliability of the prediction model. The relative error of neural network optimized by particle swarm optimization algorithm is 0.77 %，which is 3.42 % lower than the average relative error of BP neural network prediction，showing good prediction accuracy.

Key words: cemented backfill, PSO-BP, strength prediction model