Short-range particle simulations have a wide variety of applications in physics, chemistry, and materials science. Therefore, various techniques have evolved to efficiently simulate short-range forces, but no single algorithm is best suited to simulate all scenarios. Selecting the best algorithm for a simulation is a non-trivial task, often requiring extensive benchmarking. This thesis implements an automated algorithm selection strategy in the AutoPas short-range particle simulation library. The strategy uses a deep reinforcement learning approach that requires little data for the initial parameter tuning. During a simulation, it uses two stages to select the best algorithm. In the first stage, the exploration phase, it evaluates the runtime of a small number of algorithms and updates the internal neural network with the data collected. In the second stage, the exploitation phase, the neural network predicts the best algorithm based on the collected run-time information. This thesis shows that the new approach regularly outperforms previously existing strategies in terms of runtime. Furthermore, the deep reinforcement learning tuning strategy also demonstrates great generalizability to different types of problems, despite the limited amount of data available for tuning the initial state.