![]() ![]() We will cover these in more detail in the Visualize Data section. It's obvious that there are other anomalies in the sample dataset in addition to the synthetic anomaly. Between 03-18-1997 and 03-19-1997, an anomaly is created by replacing a small fraction of original data (spaning 19 hours) with a constant region. The blue represents the original data, and the green represents the data after an anomaly was introduced. The figure above shows how we created a visually obvious anomaly. This is a very interesting dataset, because it contains both a synthetic anomaly, as well as a natural change from winter regime to summer regime at the end. The sample dataset for this tutorial is taken and modified from the Italian power demand dataset in, representing the hourly electrical power demand in an Italian city for a total of 42 days (1,008 hours) from 02-23-1997 to 04-05-1997. In the next section, we will also illustrate the concept of the k-th discord with more specific examples.įor more information on Matrix Profile, please refer to this blog post. This blog post provides you with a good intuitive example of top-k discord. You can imagine that to find the k-th discord, you first need to calculate distances between all subsequences and their nearest neighbors, then sort these distances from large to small, and finally select the subsequences with the k-th largest distance. If we utilize a distance measure to quantify the dissimilarity between a time series subsequence and its nearest neighbor, the k-th time series discord is a time series subsequence with the k-th furthest distance from its nearest neighbor. The definition of time series discord also leads us to the discussion of the k-th time series discord. We can easily spot the most significant discord from a Matrix Profile: the higher the Matrix Profile value, the greater the dissimilarity between the corresponding subsequence and its nearest neighbor, so the maximum value within a Matrix Profile indicates the time series discord (or the most significant discord). In a narrow sense, a time series discord is a subsequence that is most dissimilar to its nearest neighbor, also known as the most significant discord (defined in ). Since the time series discord and the k-th time series discord have different definitions, it is necessary to keep these two concepts distinct. Time Series Discord and K-th Time series Discord ¶ The primary use of time series discords is to detect anomalies in a long time series. , time series discords refer to the most unusual time series subsequences: those which are maximally different from all other subsequences in the same time series. Time series discords have become one of the most effective and competitive methods in anomaly detection.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |