Time Series Clustering for Prediction Accuracy and Tuning Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods face challenges in predicting outputs for large numbers of time series data, especially when dealing with biases and future events, leading to impractical model tuning and reduced accuracy.
Innovation Solution
A system and method that generate predicted outputs by obtaining datasets, applying algorithms such as piecewise linear regression and deep learning, weighting the time series, and simulating perturbations to provide confidence ranges for predictions, allowing for scalable and accurate forecasting across multiple time series.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate models are generated and tuned for each time series, then prediction accuracy is improved, but device complexity and time consumption increase significantly when the number of time series is large
Solution Approach 1:
The patent merges multiple time series into clusters based on similarity metrics, then applies a single predictive model to each cluster rather than individual models to each time series. This clustering approach maintains prediction accuracy while significantly reducing model complexity and tuning requirements when dealing with large numbers of time series.
Solution Approach 2:
The patent creates universal predictive models that can be applied across multiple time series within a cluster. These models are designed to handle diverse time series data through standardized processes, reducing the need for customizing separate models for each individual time series while maintaining effectiveness across different data sets.
2Measurement precision
If separate models are tuned for each time series, then prediction accuracy is improved, but loss of time increases due to impractical model tuning when the number of time series is large
Solution Approach 1:
By clustering similar time series together and applying a single model to each cluster, the patent dramatically reduces the total number of models that need to be tuned and maintained. This approach preserves prediction accuracy for similar time series while reducing the time investment required for model development and tuning.
Solution Approach 2:
The patent performs preliminary clustering of time series before model application, grouping similar sequences together in advance. This preliminary organization enables the use of fewer, more generalizable models, reducing both the time and computational resources needed for subsequent model tuning and deployment.
3Device complexity
If a single model is applied to all time series, then device complexity is reduced, but measurement precision deteriorates due to inability to account for unique characteristics of individual time series
Solution Approach 1:
The patent applies local quality by tailoring predictive models to specific clusters of time series based on their shared characteristics. Rather than using a completely generic model for all data or highly customized models for each individual series, the system creates locally-optimized models for each cluster, balancing generalization with specificity to maintain accuracy while reducing complexity.
Data Source
AI summary
Aspects of the subject disclosure may include, for example, obtaining a plurality of historical inputs, obtaining a plurality of historical outputs, applying a piecewise linear regression, deep learning algorithm to at least the plurality of historical inputs and the plurality of historical outputs to generate a plurality of predicted inputs, applying a plurality of weightings to the plurality of predicted inputs to generate a plurality of predicted weighted inputs, and applying at least one simulation to the plurality of predicted weighted inputs to generate a plurality of predicted weighted outputs. Other embodiments are disclosed.