Advanced monitoring and communication infrastructure allows distribution system operators to design optimal operation strategies through the centralised control of Distributed Energy Resources (DERs). However, most distribution networks today lack such communication infrastructure and in many countries the cost of building the required capacity is prohibitive. On the contrary, conventional, local DER control schemes offer a robust, cheap, communication-free solution, but with sub-optimal performance. In this presentation, we investigate the use of data-driven control design algorithms to derive local DER controllers that can emulate the optimal behaviour without the need for communication. This is achieved by using historical data to capture expected operating conditions, in combination with off-line optimization techniques and machine learning methods.