Pattern detection in time-series data
Systems and methods for detecting patterns in data from a time-series are provided. In one implementation, a method for pattern detection includes obtaining data in a time-series and creating one-dimensional or multi-dimensional windows from the time-series data. The one-dimensional or multi-dimensional windows are created either independently or jointly with the time-series. The method also includes training a deep neural network with the one-dimensional or multi-dimensional windows utilizing historical and/or simulated data to provide a neural network model. Also, the method includes processing ongoing data with the neural network model to detect one or more patterns of a particular category in the ongoing data, and localizing the one or more patterns in time.
1. A method for pattern detection, comprising:
obtaining data in a time-series;
creating multi-dimensional windows from the time-series data, wherein the multi-dimensional windows are created from the time-series based on sliding windows such that the multi-dimensional windows include chunks of the time-series for pattern detection thereon;
training a deep neural network with the multi-dimensional windows utilizing historical and/or simulated data to provide a neural network model;
processing ongoing data with the neural network model to detect one or more patterns of a particular category in the ongoing data; and
localizing the one or more patterns in time based on the multi-dimensional windows.
2. The method of claim 1 , wherein the time-series data is obtained by monitoring a telecommunications network.
3. The method of claim 2 , wherein the time-series data and the ongoing data are any of logs, Performance Monitoring (PM) data, packet buffer measurements, and physical layer impairment measurements.
4. The method of claim 1 , wherein the neural network model is a convolutional neural network (CNN) including one or more CNN blocks, each CNN block including one or more CNN layers.
5. The method of claim 4 , wherein each CNN block further includes one or more non-CNN layers.
6. The method of claim 5 , wherein each CNN layer is sandwiched between the non-CNN layers.
7. The method of claim 5 , wherein the one or more non-CNN layers include one or more pooling layers following the one or more CNN layers.
8. The method of claim 4 , wherein the neural network model is configured to perform a convolution process on flexible, non-rectangular boundaries to surround anomalies in a feature map.
9. The method of claim 4 , wherein the neural network model is a regional CNN (R-CNN).
10. The method of claim 1 , wherein the neural network model is a deconvolutional neural network.
11. The method of claim 1 , further comprising optimizing hyper-parameters and selecting a best transformation for the optimized hyper-parameters.
12. The method of claim 1 , wherein the neural network model utilizes a heat map or Fourier transform to define bounding boxes around detected anomalies.
13. The method of claim 12 , wherein a masking module is configured to mask the bounding boxes.
14. The method of claim 13 , further comprising utilizing a meta-learning process on the masked bounding boxes.
15. The method of claim 13 , wherein the masking module uses tensors.
16. The method of claim 1 , further comprising the step of classifying the data as normal or anomalous based on thresholds.
17. A non-transitory computer-readable medium configured to store a program executable by a processing system, the program including instructions to cause the processing system to:
obtain time-series data;
create multi-dimensional windows from the time-series data based on sliding windows such that the multi-dimensional windows include chunks of the time-series for pattern detection thereon;
train and optimize hyper-parameters of one or more machine learning algorithms with the multi-dimensional windows obtained from historical data to create one or more machine learning models;
determine an algorithm among the one or more machine learning algorithms with the best performance;
utilize the machine learning model created from the algorithm determined to have the best performance to classify future windows as containing a pattern of a particular category; and
localize the pattern in time in ongoing data based on the multi-dimensional windows.
18. The non-transitory computer-readable medium of claim 17 , wherein the time-series data is obtained by monitoring a communication network.
19. The non-transitory computer-readable medium of claim 18 , wherein the time-series data and the ongoing data are any of logs, Performance Monitoring (PM) data, packet buffer measurements, and physical layer impairment measurements.
20. The non-transitory computer-readable medium of claim 17 , wherein the time-series data is obtained from a vehicular traffic monitoring system, a vehicle maintenance system, a medical monitoring system, a financial monitoring system, or a natural disaster sensing system.