Research On Anomaly Data Recognition Algorithm Based On Clustering And Neural Network

With the rapid expansion of data,more and more researchers are paying attention to the research of abnormal data detection.Abnormal data detection has been used in various abnormal data scenarios.According to the characteristics of power big data,studying an anomaly detection algorithm that can perform unsupervised learning,combined with clustering algorithm.The neural network method has its own advantages and is mainly applied to the power big data anomaly detection scenario.Abnormal data greatly reduces the quality of the data.Corresponding to different application fields,the anomaly data shows different characteristics.Grid data is time sensitive.It is a very effective detection method to use the combination of density clustering algorithm and Long-term and Short-term Memory Neural Network(LSTM)to find the solution to the problem of abnormal data conforming to the grid.This paper selects Density-Based Spatial Clustering of Applications with Noise(DBSCAN)and Local Outlier Factor(LOF)for initial identification of anomalous data.It mainly studies the LSTM and its variants in neural networks,and realizes accurate prediction of time-series data and abnormal data detection.The process and specific practices of LSTM training,timing prediction and threshold setting method for power quality anomaly data identification and verification are completed and analyzed.Finally,combined with density clustering and LSTM,a fusion algorithm is proposed,which considers the change law of data and realizes the hierarchical recognition of outlier.The density-based clustering algorithm is used to automatically divide the normal data anomaly data labels;then the LSTM is used to give the timing input determined as abnormal data,match the optimal output neuron number,and train through a series of data.Learn to correct the weight coefficient until the end of the training.After the neural network reaches the optimal memory state,it can effectively determine the specific outliers in the abnormal data sequence through the learned data change law.Finally,the anomaly detection results are analyzed.The comparison of clustering algorithms verifies the advantages of the improved fusion method in detection performance,and proves that the improved fusion algorithm is suitable for the detection of power quality anomaly data.