Predictive Model Tuning for Time-Series Accuracy and Control Margin
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive models face challenges in optimizing settings such as prediction horizon, explanatory variable selection, and range, requiring trial and error to balance performance and accuracy, which decreases as predictions extend further into the future.
Innovation Solution
An information processing apparatus and method that assists in generating predictive models by calculating prediction accuracy based on selected explanatory variables and positions, allowing for optimization of prediction points and variables through a user-friendly interface and simulation tools.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the prediction point is set further into the future to increase control system performance, then the performance is improved, but the prediction accuracy decreases
Solution Approach 1:
The system calculates prediction accuracy using actual values at selected positions and provides feedback to guide the selection of prediction points and explanatory variables. This feedback mechanism allows optimization of both performance and accuracy by iteratively adjusting settings based on calculated accuracy metrics.
Solution Approach 2:
The system enables changing parameters such as prediction point timing, explanatory variable selection, and explanatory variable section range. By adjusting these parameters and evaluating prediction accuracy at each configuration, the system optimizes the balance between future performance and prediction accuracy.
2Productivity
If trial and error is performed to optimize predictive model settings, then the model performance is improved, but the time and complexity of model generation increases
Solution Approach 1:
The system automatically calculates prediction accuracy and provides guidance for optimization, reducing the need for extensive manual trial and error. The automated accuracy calculation and visual output features enable faster model generation by guiding users toward optimal settings more efficiently.
Solution Approach 2:
The system replaces manual trial-and-error adjustment with automated calculation of prediction accuracy. By using computational methods to evaluate and compare different model configurations, the system reduces the time required for model generation while maintaining or improving performance.
3Measurement precision
If more explanatory variables and positions are selected for the predictive model, then the prediction accuracy is improved, but the device complexity increases
Solution Approach 1:
The system calculates and provides feedback on prediction accuracy for different combinations of explanatory variables and positions. This feedback enables users to identify the minimum necessary complexity required to achieve satisfactory accuracy, avoiding unnecessary model complexity.
Solution Approach 2:
The system allows selective assignment of explanatory variables to specific positions in the explanatory variable section. By enabling local optimization of variable selection at different positions, the system achieves high prediction accuracy with reduced overall model complexity compared to using all variables throughout.
Data Source
AI summary
Provided is an information processing apparatus for generating a predictive model. The predictive model is configured to calculate, at a prediction start point, a value at a prediction point in the future by a predetermined time margin from the prediction start point. The information processing apparatus includes a reception unit configured to receive a setting of the prediction point, a setting of an explanatory variable section corresponding to a range including the prediction start point and a period before the prediction start point, and selection of one or more positions from among a plurality of position candidates in the explanatory variable section, and a calculation unit configured to calculate prediction accuracy of the value at the prediction point using one or more actual values at the selected one or more positions based on time series data of the actual value corresponding to the value calculated by the predictive model.