Fuzzy Theory Based Quality Assessment of Multivariate Electrical Measurements of Smart Grids

Electrical measurements of smart grids form the backbone of various data driven applications. Proper analysis of these measurements can result in improved system planning, operation, monitoring and protection. However, the efficacy of smart grid data mining is highly influenced by the quality of the data. Hence, quality assessment of smart grid data is essential prior to the usage of the data in various applications. A fuzzy assessment method is proposed in this paper for assessing quality of multivariate electrical measurements of smart grids. At first, relevant quality dimensions of smart grid data are identified. Then, based on certain desirable characteristics, novel membership functions are proposed for assessing the data quality with respect to each of the considered dimensions. The proposed membership functions are evaluated on the current and real power measurements obtained from the power flow analysis of the IEEE 14-bus system. In addition, the proposed method is also implemented on the voltage, current and the real power measurements obtained from the power flow analysis of an actual 34 node feeder located in Arizona. The impact of measurement noise is also investigated by polluting the original measurements with Gaussian noise. It is found that the quality of the noisy measurements worsens with the increase in variance of the added noise. The proposed method has also been validated on a database containing practical SCADA and PMU measurements of the Southern Regional Grid of India. It is found that the PMU datasets are relatively incomplete compared to the SCADA datasets. In addition, the obtained results indicate that PMU data are suitable for use in more number of applications compared to the SCADA datasets. Unlike the existing methods, the proposed method can be used for quantifying the quality of any smart grid dataset that contains electrical measurements of multiple power system variables including boolean variables such as circuit breaker status. Moreover, unlike the existing methods, the proposed method can measure the consistency among the measurements. In addition, the proposed method is found to be sensitive to the distribution of the bad measurements in a given database.