Latent thermal energy storage (TES) systems which store energy in a phase change material (PCM) can be used to alleviate disparities in energy supply and demand for a variety of applications from concentrated solar plants to building heating and cooling systems. Furthermore, multiple TES modules containing PCMs with different melt temperatures can be combined in series and parallel configurations to construct a multi-temperature TES assembly to allow for wider flexibility in system optimization. This study demonstrates the feasibility of using machine learning based control methods to operate multi-temperature TES assemblies to ensure both operational reliability and optimization of energy usage. Two different TES assemblies are considered: one with on/off valves controlling flow to individual modules and one with fully modulating valves. Theoretical models are developed to determine the best flow path for the working fluid through each TES assembly for a variety of inlet temperatures and flow rates based on a dual optimization of matching a target outlet temperature and minimizing exergy destruction. Artificial Neural Network (ANN) controllers are then developed for each of the systems to predict the best operating mode based on the inlet temperature and flow rate to the TES assembly. The results demonstrate that an ANN controller can be used to successfully operate a multi-temperature TES assembly and maintain high operational reliability by matching a desired outlet temperature while also improving efficiency by decreasing the fraction of exergy destroyed to total energy transferred.