Seasonal Pattern Detection Using Outlier Space Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine-learning models for seasonal pattern detection and forecasting, such as the Holt-Winters model, are limited in handling variable seasonal patterns and irregularities, leading to inaccurate forecasts and inefficient resource utilization due to their assumption of fixed seasonal periods and inability to model complex seasonal patterns.
Innovation Solution
The approach involves converting time-series data to an outlier space to detect long-term seasonal patterns, using encoding spaces to map timestamps to multiple encodings, and training models based on detected seasonal patterns, allowing for flexible and scalable detection of variable seasonal patterns without the need for multiple models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If Holt-Winters model with fixed seasonal period is used, then model simplicity is maintained, but ability to represent complex seasonal patterns is limited
Solution Approach 1:
The patent applies dynamics by making the seasonal period variable rather than fixed. The model learns the seasonal period from data and allows it to change over time, enabling adaptation to varying seasonal patterns while maintaining a unified model structure. This resolves the contradiction by allowing the model to be simple in structure but flexible in behavior through learned parameters.
Solution Approach 2:
The patent changes the parameter of seasonal period from a fixed constant to a learnable variable that can take different values at different time points. This allows the model to adapt to complex seasonal patterns with varying periods while keeping the overall model framework relatively simple, thus resolving the contradiction between model simplicity and pattern representation capability.
2Measurement precision
If multiple Holt-Winters models are trained for different seasonal patterns, then seasonal pattern detection accuracy is improved, but computational complexity and model management become cumbersome
Solution Approach 1:
The patent creates a universal model that can handle multiple seasonal patterns with different periods simultaneously. Instead of training separate models for each seasonal pattern, a single model learns to adapt to various seasonal periods through the variable seasonal period parameter, thus improving detection accuracy while avoiding the complexity of managing multiple models.
Solution Approach 2:
The patent merges the functionality of multiple seasonal models into a single unified model. By combining the ability to handle different seasonal periods into one model through learned variable periods, it achieves the detection accuracy of multiple specialized models while eliminating the need to train and manage them separately, thus resolving the contradiction between accuracy and complexity.
3Productivity
If seasonal period is assumed to be fixed, then forecasting computation is simplified, but accuracy in capturing variable seasonal patterns deteriorates
Solution Approach 1:
The patent applies dynamics by allowing the seasonal period to vary and be learned from data rather than fixed beforehand. This enables the model to capture variable seasonal patterns accurately while maintaining computational efficiency through a unified forecasting framework that adapts to different periods automatically, resolving the contradiction between computation simplicity and forecast accuracy.
4Reliability
If seasonal indices are implemented as internal structures within forecasting models, then forecasting capability is maintained, but interpretability and external utilization of seasonal data are limited
Solution Approach 1:
The patent extracts seasonal information from the forecasting model and presents it as separate, interpretable outputs. By learning variable seasonal periods and making them explicit rather than hidden internal structures, the model maintains its forecasting capability while also providing accessible seasonal pattern information that can be interpreted and utilized externally, thus resolving the contradiction between forecasting reliability and data interpretability.
Data Source
AI summary
Techniques for machine-learning of long-term seasonal patterns are disclosed. In some embodiments, a network service receives a set of time-series data that tracks metric values of at least one computing resource over time. Responsive to receiving the time-series data, the network service detects a subset of metric values that are outliers and associated with a plurality of timestamps. The network service maps the plurality of timestamps to one or more encodings of at least one encoding space that defines a plurality of encodings for different seasonal patterns. Based on the mapped encodings, the network service generates a representation of a seasonal pattern. Based on the representation of the seasonal pattern, the network service may perform one or more operations in association with the at least one computing resource.