Fuzzy Learning Time-Series Extraction via FAM Bank
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional time-series analysis methods, such as ARIMA and neural networks, are inefficient for extracting behavior in scenarios with weak relationships between data points, particularly in RDBMS-based systems like CPU utilization, where human behavior influenced by time-related values is significant, and require extensive training data and are computationally intensive.
Innovation Solution
The method employs fuzzy learning to extract time-series behavior by loading data, dividing it into fuzzy regions, assigning fuzzy membership functions, generating non-conflicting fuzzy rules, and creating a Fuzzy Associated Memory (FAM) bank to build a model that can predict trends without requiring re-examination of initial data points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional methods like ARIMA are used for time-series analysis, then the analysis can be performed with a structured approach, but the method becomes computationally intensive and time-consuming
Solution Approach 1:
The patent replaces the mechanical iterative computation of ARIMA with a neural network-based system that learns temporal patterns through adaptive weight adjustments. The neural network computes predictions in parallel without requiring multiple iterative passes, thereby maintaining analysis reliability while significantly improving computational speed and productivity.
Solution Approach 2:
The patent transforms the fixed parameter structure of ARIMA into dynamic parameters through the neural network's learning process. The network automatically adjusts weights and thresholds based on training data, allowing the system to adapt to varying temporal patterns without manual parameter tuning, thus resolving the contradiction between reliable analysis and computational efficiency.
2Reliability
If ARIMA models are used for time-series prediction, then the method provides a comprehensive analysis framework, but it requires a relatively large amount of training data
Solution Approach 1:
The neural network system performs partial learning by focusing on the most significant temporal patterns and features in the data rather than requiring comprehensive training on all possible scenarios. This allows the system to achieve reliable predictions with smaller training datasets by capturing the essential dynamics without overfitting to noise or minor variations.
Solution Approach 2:
The neural network performs self-learning and automatic feature extraction from the training data, eliminating the need for manual data preprocessing and feature engineering that ARIMA requires. The system autonomously identifies relevant patterns and adjusts its internal parameters, thereby achieving reliable predictions with reduced training data requirements.
3Reliability
If conventional time-series analysis tools are used, then the methods can handle strong relationships between consecutive data points, but they perform poorly when the relationship between data points is weak
Solution Approach 1:
The neural network provides a dynamic modeling approach that adapts to the actual relationship strength in the data. Through its learning mechanism, the network automatically adjusts its sensitivity to temporal patterns, making it effective for both strong and weak relationships between consecutive data points. This dynamic adaptability resolves the contradiction by allowing the system to optimize its behavior based on the specific characteristics of the input data.
Solution Approach 2:
The neural network transforms the static assumptions of conventional methods into dynamic parameter adjustments. By learning from training data, the network adapts its internal parameters to match the relationship strength in the specific application, whether strong or weak, thereby achieving reliable behavior extraction across diverse data types without requiring different methods.
4Reliability
If ARIMA and neural networks are used for time-series analysis, then the methods can capture complex patterns, but they are still relatively slow and computationally intensive
Solution Approach 1:
The system performs preliminary learning during a training phase where the neural network pre-computes and stores optimal weight configurations for capturing temporal patterns. During actual prediction operations, the system applies these pre-learned weights directly without requiring intensive computation, thereby maintaining high pattern recognition capability while significantly reducing computational resource consumption during deployment.
Data Source
AI summary
Systems and methods for extracting or analyzing time-series behavior are described. Some embodiments of computer-implemented methods include generating fuzzy rules from time series data. Certain embodiments also include resolving conflicts between fuzzy rules according to how the data is clustered. Some embodiments further include extracting a model of the time-series behavior via defuzzification and making that model accessible. Advantageously, to resolve conflicts between fuzzy rules, some embodiments define Gaussian functions for each conflicting data point, sum the Gaussian functions according to how the conflicting data points are clustered, and resolve the conflict based on the results of summing the Gaussian functions. Some embodiments use both crisp and non-trivially fuzzy regions and/or both crisp and non-trivially fuzzy membership functions.