The wavelet transform and rescaled range analysis were employed to reveal the multi-scale interactions between particles and fluids in a gas-solid cyclone reactor. A wavelet-threshold criterion was subsequently specified to identify clusters. By the established threshold, cluster dynamics can be determined from the transient signal. Experimental results show that the micro-, meso-, and macro-scale characteristics can be described by different wavelet decomposed sub-signals. The normalized energy of original solid holdup signals is mainly distributed on mesoscale signals. Cluster dynamics is found to be rather complex near the outer wall. The relationship between time fraction of clusters and average cluster concentration near the outer wall was further analyzed, dividing the cyclone reactor into dilute cluster regions and dense cluster regions. Finally, the prediction model between average cluster concentration and local average solids holdup in the vicinity of the outer wall was established. (C) 2020 Published by Elsevier B.V.