Anomaly Detection via Zonal Parameter Characteristics and Non-Linear Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection systems rely on reactive solutions with fixed thresholds, requiring manual expert intervention and being data domain-specific, which limits their ability to pre-emptively identify and remediate anomalies in real-time across various industries.
Innovation Solution
A computer-implemented method using zonal parameter characteristics and non-linear scoring with a plurality of artificial intelligence models, such as hierarchical temporal memory, to detect and classify anomalies into categories without fixed thresholds, automatically determining root causes and recommending remediation steps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If fixed threshold methods are used for anomaly detection, then the detection process is simple and fast, but the system requires manual expert intervention and cannot adapt to different data domains
Solution Approach 1:
The patent implements a universal anomaly detection system that works across multiple data domains (IoT, IT, healthcare, finance) by using domain-agnostic deep learning models. The system processes time series data from any domain through the same neural network architecture (LSTM, GRU, or vanilla RNN), eliminating the need for domain-specific customization while maintaining high detection accuracy.
Solution Approach 2:
The system automatically adapts to different domains by changing internal parameters through deep learning model training. The neural networks learn domain-specific patterns and thresholds dynamically during training, replacing fixed manual thresholds with adaptive learned parameters that automatically adjust to each data domain's characteristics.
2Ease of manufacture
If reactive anomaly detection with fixed thresholds is used, then implementation is straightforward, but the system cannot pre-emptively identify anomalies
Solution Approach 1:
The patent implements pre-emptive anomaly detection by training deep learning models on historical time series data to learn normal patterns and predict future states. The system detects anomalies before they occur by identifying deviations from predicted normal behavior, enabling preventive rather than reactive responses. The models continuously learn from past data to improve future predictions.
Solution Approach 2:
The system replaces manual threshold-setting mechanisms with automated deep learning-based prediction mechanisms. Instead of experts manually defining fixed thresholds, the patent uses neural networks to automatically learn optimal detection thresholds and patterns from data, substituting mechanical rule-based systems with intelligent adaptive systems.
3Measurement precision
If multiple artificial intelligence models are used for anomaly detection, then detection accuracy improves, but system complexity increases
Solution Approach 1:
The patent segments the anomaly detection task into distinct model types (LSTM for sequential dependencies, GRU for compressed sequential processing, vanilla RNN for baseline patterns). Each model type addresses specific aspects of time series analysis, and the system evaluates multiple segments to determine the optimal model for each scenario, improving accuracy while maintaining manageable complexity through modular design.
4Measurement precision
If manual expert intervention is required for anomaly detection, then detection can be domain-specific, but productivity and response time decrease
Solution Approach 1:
The patent implements self-service anomaly detection where the deep learning models automatically perform detection without manual expert intervention. The system autonomously trains on historical data, identifies patterns, detects anomalies in real-time, and generates alerts. This automation maintains domain-specific accuracy through learned patterns while dramatically improving productivity by eliminating manual review requirements.
Solution Approach 2:
The system incorporates feedback loops where detection results and ground truth data continuously improve the models through retraining. The models learn from correct and incorrect predictions, automatically adjusting their parameters to improve accuracy over time. This feedback mechanism enables the system to maintain or improve domain-specific detection accuracy while operating autonomously at high speed.
Data Source
AI summary
Anomaly detection using zonal parameter characteristics and non-linear scoring is provided. Anomalous behavior characteristics of a parameter and group of parameters are identified within time series data using an optimal artificial intelligence model of a plurality of artificial intelligence models. Anomalies are detected based on the anomalous behavior characteristics of the parameter and the group of parameters within the time series data. The anomalies are classified into a corresponding anomaly category. A root cause of the anomalies is determined based on the corresponding anomaly category. One or more action steps are performed to remediate the root cause of the anomalies.