The factors of shades and fault nodes will cause that location effect is not the optimal

when all AP nodes are as feature inputs. The traditional methods select AP nodes by comparing signal feature of the whole area

without considering correlation between different sub area and AP nodes. Aiming at this problem

put forward a distributed AP selection algorithm

which can effectively remove at AP nodes with big noise and weak ability of position resolution. This method divides indoor area into several sub areas and calculates the correlation between the sub area and AP nodes for selecting AP nodes with optimal correlation as training nodes of this sub area. Deep belief networks（ DBN） model is used for training location model. According to the result of the experiments

average location error of the proposed model is 0. 415 1 m at the range of 12 m × 12 m. At the same time

comparing with BP algorithm

RADAR algorithm

the proposed model performs better in wireless indoor location

which reduces positioning error by 35. 95 % and 46. 78 % and decreases running time by 22. 8 % and 32. 9 %respectively.