Time-Series Factor Selection for More Accurate Prediction Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting values of objects fail to accurately consider interaction effects between factors and often select irrelevant factors, leading to decreased explanatory power due to reliance on static relations rather than time series changes.
Innovation Solution
An electronic apparatus and method that identifies significant factors by analyzing time series changes in values, using indices like up/down, extremum, and precedence prediction to determine the probability and accuracy of factors, grouping factors based on their relation to the prediction object, and selecting factors with representativeness to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional factor selection methods (forward selection, backward elimination, principal component analysis) are used, then the prediction model can be built with selected factors, but the interaction effects between factors are not considered and irrelevant factors may be selected leading to decreased explanatory power
Solution Approach 1:
The patent transforms static factor selection into a dynamic process by analyzing time-series changes of factors. Instead of selecting factors based on static relationships, the system identifies factors whose time-series patterns (increasing/decreasing trends, extrema points) dynamically correspond to the prediction object's reference values, thereby capturing temporal interaction effects that static methods miss.
Solution Approach 2:
The patent changes the selection criteria from static parameter values to dynamic parameter changes over time. By focusing on how factors change (increasing/decreasing trends, rate of change) rather than their absolute values, the system identifies factors whose temporal patterns correspond to the prediction object, capturing interaction effects through parameter transformations.
2Adaptability or versatility
If more factors are selected to improve prediction coverage, then the model includes more variables, but irrelevant factors are included causing decreased explanatory power
Solution Approach 1:
The patent implements a feedback mechanism where the time-series changes of selected factors are compared against the actual reference values of the prediction object. Factors are selected based on whether their temporal patterns (trends, extrema) provide useful feedback about the prediction object's behavior, allowing the system to maintain adaptability while ensuring each selected factor contributes meaningfully to explanatory power.
Solution Approach 2:
The patent performs preliminary analysis of factor time-series patterns before final selection. By pre-identifying factors whose historical changes correspond to reference value changes, the system prepares a refined set of candidate factors that are more likely to be relevant, thus maintaining versatility while improving explanatory power through preliminary filtering.
Data Source
AI summary
An electronic apparatus and a control method thereof are provided. The electronic apparatus may include an interface; and a processor configured to obtain, via the interface, information related to values, which occur in time series, of a plurality of factors regarding a prediction object, identify, based on the information related to the values of the plurality of factors, at least one factor, from among the plurality of factors, having a time series change of values that corresponds to a time series change of reference values of the prediction object, and output information related to a predicted value of the prediction object based on the time series change of the values of the at least one factor.