On-line solution of constrained optimization problems is an essential requirement for a secure, reliable and economic smart grids operation. In this context, the stream of data acquired by the grid sensors should be promptly analyzed in order to identify proper control actions aimed at mitigating the effect of system perturbations, or adapting the grid state to new load and/or generation patterns. To solve this computationally intensive task, case-based reasoning could play a key role in extracting actionable knowledge from historical datasets. Due to time constraints, the search of similar patterns, although simple in principle, requires to face the large size of the historical smart grid data, which increase dynamically once new measured information is available. However, recent advances in Big Data offer efficient schemes for processing massive data. As an example, in this paper we present a solution aimed at searching k most similar cases based on the MapReduce paradigm. This paradigm allows to distribute data and jobs over a compute cluster of commodity hardware. Experimental results are intended to describe the solution behavior under different conditions, such as the dataset size, the number of available nodes/jobs, the block size.