The growing popularity of intelligent manufacturing is driven by deterministic transmission demands of applications in industrial cyber-physical systems (ICPS).However,the ossified shortages of in-dustrial wireless communication such as diverse quality of service (QoS) and complex signaling processes incur a severe long tail of transmission delay distribution.As a solution,the 5th generation (5G) wire-less communication technology provides ultra-reliable and low-latency communication (URLLC) for industry scenarios.Moreover,the newly proposed time-sensitive networking (TSN) standards guarantee the trans-mission determinacy by gate mechanism.In this paper,we propose a heterogeneous time-sensitive network(HTSN) co-designed by 5G and TSN.We first develop a predictive multi-priority wireless scheduling mecha-nism based on semi-persistent scheduling (SPS) to reduce signaling delay by reserving resources in advance.Then we propose an adaptive data injection mechanism for TSN based on per-stream filtering and policing(PSFP),which dynamically adjusts the priority of data for queue injection in TSN.To further reduce the long tail of delay,we employ a risk-sensitive learning method to improve the worst-case delay.Simulations on a hot rolling production scenario demonstrate that the proposed mechanisms under HTSN achieve great performance in terms of integrated delay and resource utilization.