Energy big data abnormal cluster detection method based on redundant convolution codec.

Area of Science:

• Energy Systems Engineering
• Data Science
• Machine Learning

Background:

• Current energy big data anomaly detection methods suffer from poor clustering of abnormal data, limiting detection capabilities.
• Scattered distribution of abnormal clusters in multi-energy consumption data poses significant challenges for accurate anomaly identification.

Purpose of the Study:

• To propose a high-energy data anomaly clustering detection method leveraging redundant convolutional encoding.
• To enhance the feature clustering performance and detection accuracy for multi-energy user consumption patterns.
• To develop a robust model for anomaly detection in massive energy big data.

Main Methods:

• Quantitative analysis of multi-energy time series coupling characteristics using Copula functions.
• Utilized redundant convolutional codecs for encoding abnormal energy big data features.
• Employed coupling time capsule layers and fully connected linear regression for feature synthesis and anomaly clustering.

Main Results:

• Achieved excellent feature clustering performance with detection accuracy exceeding 98.7%.
• Demonstrated fast convergence speed and a low error rate below 0.1%.
• Successfully transformed energy time series data into a three-dimensional feature space for comprehensive analysis.

Conclusions:

• The proposed redundant convolutional encoding method significantly improves anomaly detection in energy big data.
• The method offers reliable application value for comprehensive energy systems and massive multi-energy user data analysis.
• The approach effectively captures multi-energy coupling time features for enhanced anomaly clustering.