Detecting Anomalous Responses in Demand Response Surveys with Local Outlier Factor

In order to better carry out power demand response to utilize demand-side resources, researchers are conducting survey-based research to develop a quantitative and interpretable demand response behavior model. However, due to the characteristics of the demand response survey, using simple pre-set patterns can only identify a small number of abnormal respondents. This paper investigates the application of local outlier factor algorithm in detecting anomaly responses in demand response survey datasets to improve data quality. We test local outlier factor algorithm through the survey of customers’ electricity consumption habits from the Human and Energy Systems Laboratory, and investigate the results from both practical and theoretical perspectives. Our results reflect that the local outlier factor algorithm is effective. It can not only find abnormal points according to certain attribute pre-set patterns, but also automatically find abnormal points from the complex relationship between attributes, which can eliminate more abnormal responses.