Modern road networks are critical in developing transportation infrastructures from the aspect of sustainability, thanks to the rapid increase in road users. The demand for mobility makes the existing infrastructure more crowded, boosting greenhouse gas emissions and delays in everyday commuting. Expanding the road network is only possible in some cases and is also not feasible, but Intelligent Transportation Systems (ITS) can enhance the efficiency of the existing transportation network. From a management point of view, a proven algorithm is Variable Speed Limit Control, which regulates the state of certain sectors by spatially distributing traffic in the form of dynamically varying speed limits. Combining this with a state-of-the-art predictive solution can make a big difference to performance. For the design of speed limits, this paper proposes an approach where deep reinforcement learning with the smallest industrial share not only resolves moving jams that arise during congested traffic situations, but also prevents them, thereby avoiding cumulative error from transients, all by abandoning physical equations and identified models.