Estimation Program for Power Exchange Contract Price Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in accurately predicting contract prices in power exchange markets due to poor prediction accuracy of demand and supply curves for future time slots.
Innovation Solution
A computer-readable recording medium stores an estimation program that approximates demand and supply curves using linear functions, generates machine learning models based on these approximations, and estimates the intersection point between the curves in a future period using the trained models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to predict contract prices, then the prediction accuracy is poor, but using complex non-linear models would increase computational complexity and training difficulty
Solution Approach 1:
The patent divides the curve approximation task into multiple linear segments. Instead of using a single complex non-linear model, the demand and supply curves are approximated by multiple linear functions (segments) that are stitched together. This segmentation approach reduces the complexity of individual models while maintaining overall prediction accuracy through piecewise linear representation.
Solution Approach 2:
The patent employs dynamic curve approximation where the linear functions are fitted adaptively based on historical data patterns. The model dynamically adjusts the linear approximation parameters (slope and intercept) for different time periods and market conditions, allowing the system to capture non-linear relationships through multiple simple linear models rather than one complex static model.
2Measurement precision
If machine learning models are trained using historical data, then prediction accuracy improves, but the training process requires significant computational resources and time
Solution Approach 1:
The patent uses simple linear functions as the building blocks for curve approximation. These linear models are computationally inexpensive compared to complex non-linear or deep learning models. By composing multiple simple linear segments, the system achieves accurate curve representation without requiring intensive computational resources for training and inference.
Solution Approach 2:
The patent transforms the training problem by changing parameters from raw historical data directly into linear function parameters (slope and intercept). Instead of training complex models on raw data, the system fits linear parameters to historical contract price data, significantly reducing computational complexity while maintaining the ability to capture market dynamics.
3Device complexity
If the demand and supply curves are approximated by linear functions, then the calculation becomes simpler, but the accuracy of curve representation may be reduced
Solution Approach 1:
The patent applies segmentation by dividing the curve into multiple linear segments rather than using a single linear function. Each segment represents a portion of the demand or supply curve with its own linear parameters. This piecewise linear approach allows the model to capture non-linear characteristics through the arrangement and transition of multiple simple linear segments, balancing calculation simplicity with representation accuracy.
Solution Approach 2:
The patent adds a dimensional aspect by introducing multiple linear functions with different parameters (slope, intercept) for different time periods and market conditions. Instead of a single linear approximation, the system uses a multi-dimensional space of linear models that can adapt to various scenarios, enhancing curve representation accuracy while maintaining computational simplicity.
Data Source
AI summary
A non-transitory computer-readable recording medium stores an estimation program for causing a computer to execute a process including: approximating a first curve and a second curve that has one intersection point with the first curve in a first period to a first approximation expression and a second approximation expression that are linear functions that pass through the intersection point, respectively; generating a machine learning model by respectively performing training by using data of slopes and intercepts of the first approximation expression and the second approximation expression; and estimating an intersection point between the first curve and the second curve in a second period that corresponds to a period after the first period by using the trained machine learning model.