Identification of disturbance sources based on random forest model

With more and more disturbance sources such as high-speed railway and renewable energy generation, the power quality problem has become increasingly complex, which seriously affects the reliable operation of the power grid. Identifying the types of disturbance sources that cause power quality events based on power quality monitoring data will support for targeted control disturbance sources, also provide evidences for determining contribution between customers and operators. This paper proposes a method for identification of disturbance sources based on random forests. Firstly, it chooses analysis indices and extracts both temporal and statistical characteristics of selected indicators from historical data. After balancing datasets, the data are used as eigenvectors of training random forests. Secondly, combining with OOB(out-of-bag), one of evaluation indicators, adjustment of random forest parameters in a closed-loop is used to construct cost-optimized random forest classifier. Thirdly, the type of disturbances is on-line identified using the classifer. Based on data from a power quality monitoring system in a regional grid of China, it is verified that the method has a high accuracy for identifying disturbance sources such as electric railways, converter stations, wind power, photovoltaics, and smelters.