| Power quality(PQ)issues not only affects safety and stability of power system,but also causes great economic losses,therefore,it is necessary for utilities to strengthen technical supervision for PQ issues.Assessment of steady PQ phenomena is one of important components in the technical supervision for PQ issues.Assessment of steady PQ issues is a typical complex multi-index decision problem.With the increasement of PQ monitoring data,there is a novel approach for assessment of steady PQ issues by mining the information from data.Therefore,based on a large amount of monitoring data of PQ monitoring systems,this thesis focuses on the data-driven assessment methods of steady PQ issues from grid side.The main work is as follows:i)Steady PQ monitoring data is analyzed statistically to find the characteristics of data.There are three noteworthy characteristics for steady PQ monitoring data which are the existence of bad data,an imbalanced structure,and the limitations of single steady PQ index.ii)A method based on isolation forest is proposed to detect the bad data in the monitoring data.According to distribution characteristic of PQ index,three attributes are selected to generate isolation forest.Hyperparameters and anomaly score threshold of isolation forest are determined by the detection effectiveness of test samples.The proposed method can detect the bad data in the monitoring data effectively.iii)To analyze the data with imbalanced structure,quantum clustering is applied to the assessment of steady PQ issues.Various abnormal patterns hidden in data can be mined by quantum clustering effectively.Limit-exceeding index for each abnormal pattern is defined by the difference between the monitoring data and limits.Severity index of each region was defined by the proportion of regional monitoring data in different clusters,which could be used to compare the steady PQ level in different regions.iv)To analyze the massive and high-dimensional monitoring data,a method of dimension reduction based on aggregation is proposed.At the same time,an assessment method based on ranking principal curve is proposed with the aggregation results.To reduce the dimension,the rank attributes describing PQ level of each region are obtained by aggregation.Ranking principal curve is applied to calculate the rankings of each region according to the rank attributes.The method is an unsupervised ranking algorithm for multi-attribute objects.The two assessment methods proposed in this thesis make full use of the value of rich monitoring data from different perspectives.The methods play a significant role in supporting decision-making for the technical supervision and management of steady PQ. |
