Two-Stage Clustering for Power Consumption Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large-scale information processing systems face inefficiencies in power consumption prediction due to the use of general clustering algorithms, which result in low prediction accuracy and questionable training data quality, leading to potential power consumption exceeding thresholds.
Innovation Solution
The proposed solution involves a two-stage clustering approach where samples are first classified into favorable and unfavorable clusters based on correlation variance and average criteria, with further subdivision of unfavorable clusters to generate high-quality training data for a power consumption prediction model using machine learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If general clustering algorithms are used to reduce training data size, then the training data size is reduced, but the prediction accuracy deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the training data clustering process into two distinct stages: first clustering to reduce data size, and second clustering to refine cluster quality. This two-stage approach allows the system to benefit from data reduction while maintaining or improving prediction accuracy through the quality enhancement provided by the second clustering stage.
Solution Approach 2:
The patent changes the clustering parameters dynamically by adjusting cluster quality thresholds and data sampling rates between the two clustering stages. The first clustering uses initial parameters to reduce data size, while the second clustering modifies these parameters to enhance cluster quality, ensuring that the final training data meets the required accuracy standards.
2Measurement precision
If all collected samples are used as training data, then the prediction accuracy is improved, but the processing time and computational cost increase
Solution Approach 1:
The patent extracts only the most representative and high-quality samples from the complete dataset through the two-stage clustering process. Instead of using all collected samples, the system identifies and extracts key samples that best represent the underlying patterns, thereby maintaining prediction accuracy while significantly reducing the amount of data that requires processing.
Solution Approach 2:
The patent performs preliminary clustering and quality assessment before the actual model training. By pre-processing the data to identify and select high-quality training samples in advance, the system prepares an optimized dataset that reduces computational burden during training while ensuring the model learns from the most informative samples.
3Productivity
If first clustering is performed to reduce training data size, then the processing efficiency is improved, but the training data quality deteriorates
Solution Approach 1:
The patent segments the data processing workflow into two distinct clustering phases. The first clustering phase focuses on reducing data size to improve processing efficiency, while the second clustering phase specifically addresses quality enhancement. This segmentation allows each phase to optimize for its specific goal without compromising the other.
Solution Approach 2:
The patent implements a feedback mechanism where the output of the first clustering serves as input to the second clustering, which evaluates and refines the cluster quality. The second clustering uses quality metrics to assess the results of the first clustering and applies corrective refinements, ensuring that the final training data meets quality standards despite the initial reduction in data size.
Data Source
AI summary
First clustering is performed on a plurality of samples each including time-series measurement values of power consumption to thereby generate a plurality of first clusters. The plurality of first clusters are each classified as a second cluster satisfying a determination condition or a third cluster that does not satisfy the determination condition. The determination condition includes at least one of a first criterion in which the variance of correlation values between samples is less than a first threshold and a second criterion in which the average of the correlation values exceeds a second threshold. Second clustering is performed on samples included in the third cluster to divide the third cluster into a plurality of fourth clusters. Training data for use in generation of a model for predicting power consumption is generated based on the second cluster and at least one of the fourth clusters.