Deep Neural Network Pattern Detection in Time-Series Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional performance monitoring and pattern detection in time-series data are manual, costly, and time-consuming, especially in large and complex networks, and current machine learning approaches are limited in scalability and accuracy, primarily focusing on reactive anomaly detection and failing to handle non-stationary data.
Innovation Solution
The method involves creating one-dimensional or multi-dimensional windows from time-series data, training deep neural networks using historical and simulated data to develop neural network models, and processing ongoing data to detect patterns and localize anomalies in time, utilizing techniques like Convolutional Neural Networks (CNNs) and Regional Convolutional Neural Networks (R-CNNs) for enhanced pattern detection and anomaly localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual performance monitoring and pattern detection are used, then expertise and knowledge of the system are utilized, but the process becomes expensive, time-consuming, and not scalable to large and complex networks
Solution Approach 1:
The system performs preliminary actions by training machine learning models offline with historical data before deployment. The models are pre-trained to recognize patterns and anomalies, enabling rapid real-time detection without requiring manual expert analysis during critical events. This preliminary preparation resolves the contradiction by shifting computational work from real-time manual analysis to offline automated model training.
Solution Approach 2:
The invention creates a virtual copy of the network system through machine learning models that replicate expert detection capabilities. Instead of relying on human experts to analyze each event, the system uses trained models that copy and encode expert knowledge, enabling scalable automated pattern detection across large networks while maintaining detection accuracy. This copying approach eliminates the time loss associated with manual expert analysis.
2Ease of manufacture
If rule-based engines with hard-coded statements are used for anomaly detection, then known failures can be detected, but the approach cannot find failures that span multiple network elements and does not scale
Solution Approach 1:
The system transitions from fixed rule-based parameters to adaptive learned parameters through machine learning. The model learns optimal detection parameters from historical data, enabling it to detect complex failures spanning multiple network elements. This parameter change from static rules to dynamic learned parameters resolves the contradiction by maintaining ease of implementation while dramatically increasing adaptability to complex failure patterns.
Solution Approach 2:
The invention moves from one-dimensional simple threshold rules to multi-dimensional pattern recognition. The machine learning models analyze multiple network elements and parameters simultaneously across different dimensions, enabling detection of complex failures that span multiple components. This dimensional expansion allows the system to maintain simple implementation while achieving versatile detection capabilities.
3Ease of operation
If simple one-dimensional thresholding rules are used, then implementation is straightforward, but accuracy is limited
Solution Approach 1:
The system merges multiple simple detection rules into a unified machine learning model that combines information from multiple network elements and parameters. Instead of using isolated one-dimensional thresholds, the model integrates signals across multiple dimensions, maintaining operational simplicity while dramatically improving detection accuracy through combined pattern recognition.
Solution Approach 2:
The invention creates a composite detection system that combines multiple detection approaches and data sources into a unified model. The machine learning model acts as a composite structure integrating various input features and detection strategies, achieving high accuracy while maintaining ease of operation through automated model deployment. This composite approach resolves the contradiction by combining simplicity with precision.
Data Source
AI summary
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.