In order to improve the performance of elastic optical networks, a combined time-frequency domain and blocking threshold triggered periodic defragmentation(PDT-TFDBT) strategy is proposed. In this strategy, the blocking delay triggering mechanism with improved fragmentation index is adopted to reduce the load of ineffective defragmentation in non-blocking conditions. In terms of sorting order, the time domain and frequency domain characteristics of services are comprehensively considered, and the reconfiguration order of services is reasonably allocated. At the same time, the heuristic deep learning-assisted routing and spectrum allocation(DKA-RSA) algorithm is introduced. The simulation results show that PDT-TFDBT strategy can not only reduce the number of reconstructed traffic and spectrum fragmentation, but also further reduce the traffic blocking rate and improve the spectrum utilization rate.