Voltage Quality Pattern Recognition in Microgrids Using a Data-Driven Machine Learning-Based Methodology

The paper proposes a methodology for voltage quality evaluation in low-voltage microgrids using a data-driven machine-learning process. The data inputs of this process are the voltage quality indicators on each phase (average voltage deviation, the variance of voltage deviation, and mean squared value of voltage deviation) at the level of each node from the microgrid. The indicators are determined based on the voltage profiles from the power flow calculations, considering the data from the smart meters installed at each end-user. All nodes are classified into one of the following four voltage quality patterns (“Unsatisfactory”, “Satisfactory, “Good”, and “Very Good”) using the K-means clustering algorithm based on a global quality coefficient calculated as the weighted sum of the voltage indicators. A real low-voltage microgrid with 36 nodes and 53 end-users (consumers and prosumers) without voltage control devices integrated has been used to test the proposed methodology. Finally, two voltage quality patterns, “Satisfactory” and “Good”, associated with the first two phases, “Good” and “Very Good” for the third, have been recognized. However, the distribution network operator should take technical measures to improve the voltage quality such that the ranking of each node is inside the “Good” and “Very Good” patterns.