LMP Prediction with Incomplete Information

Data analytics is changing the landscape of power system control as well as electricity market design. However, with the increasing public concern over privacy leakage, it becomes a delicate task to make the data public, affecting the performance of data analytic methods. Many independent system operators choose to anonymize or mask the system information to ensure privacy preservation. We seek to understand how and to what extent these privacy-preserving techniques affect the performance of the data analytic methods. This paper considers a classical task, locational marginal price prediction, with historical data. We compare the performance given diverse incomplete or anonymized historical data, which allows us to identify the effects of different privacy-preserving methods. We find that bidding information is more crucial than the demand information to enable accurate prediction through ablation experiments. In addition, given the incomplete data, we propose heuristic algorithms to improve prediction accuracy by embedding additional side information.