With the development of wireless communication technology and the Industrial Internet, Software Defined Network (SDN) technology has been introduced to wireless sensor networks due to its agility and flexibility. This meets the potential scalability and flexibility requirements of the Internet of Things. Thus, a new Industrial Internet architecture, called SDN-WSN, was formed. As the scale of SDN-WSN increases, efficient routing protocols with low latency and high security are required, while the standard routing protocol of SDN is still vulnerable to dynamic changes in traffic control rules, especially when the network is under attack. To address the above issues, a network node credibility evaluation model based on D-S evidence theory was constructed to evaluate the trust value of wireless sensor network nodes. A trustworthy secure routing algorithm based on Q-learning (QTSRA) was proposed. This method extracts knowledge from historical traffic demands by interacting with the underlying network environment to evaluate the trustworthiness of network nodes. Simultaneously, it implements dynamic optimising routing strategies based on deep reinforcement learning algorithms. We conducted simulation experiments for several network performance metrics, and the results showed that the proposed QTSRA routing algorithm exhibited good performance. In most of the cases, the QTSRA had an improved relative performance gain as compared to the traditional AODV and OLSR routing algorithms.