In this article we present a model for the interaction of distributed energy resources (DER) with the electricity system, using reinforcement learning.Our method relaxes the requirements for information necessary to train and engage in Pareto improving trading, and can directly incorporate the inherent intermittency of variable renewable energy sources.The distributed resources include consumers of electricity, energy storage systems, and variable renewable energy.We modify the algorithms to improve the scheduling of the resources.In our empirical application, we use data from Colombia subject to large variability due to El Niño Southern Oscillation and illustrate the use of the model under large variations in the data used to train the model.