Distribution Network Loss Detection Using Autoencoder Anomaly Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning algorithms for detecting non-technical energy losses in electrical distribution networks face challenges due to selection bias and the need for extensive, costly human inspections to create training datasets, which are not representative of the overall utility set.
Innovation Solution
Employing an autoencoder, a type of unsupervised neural network, trained on a diverse dataset of electrical parameters and consumption trends to identify non-technical losses by reconstructing input data, using records with known absence of non-technical losses to validate the model, and comparing input-output data for anomaly detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised machine learning algorithms are used to detect non-technical energy losses, then detection capability is improved, but training data collection requires extensive human inspections which are time-consuming and expensive
Solution Approach 1:
The system uses unsupervised learning algorithms that automatically analyze consumption patterns and electrical parameters without requiring human-labeled training data. The algorithm self-trains by identifying anomalies in the data itself, eliminating the need for time-consuming manual inspections to create training datasets.
Solution Approach 2:
Instead of using supervised learning that requires pre-labeled training data (traditional approach), the patent inverts the approach by using unsupervised learning that detects anomalies without prior labeling. The system inverts the dependency from human-inspected training data to algorithm-driven automatic detection.
2Measurement precision
If supervised machine learning algorithms are used to detect non-technical energy losses, then detection capability is improved, but training data collection requires extensive human inspections which are costly
Solution Approach 1:
The unsupervised learning algorithm performs self-training by automatically identifying patterns and anomalies in the consumption data without requiring human inspectors to label training samples, thereby eliminating the costly manual inspection process while maintaining high detection capability.
Solution Approach 2:
The patent replaces the mechanical human inspection process with an automated computational system that uses unsupervised learning algorithms to detect non-technical losses, substituting expensive manual labor with cost-effective automated data analysis.
3Measurement precision
If training datasets are created using human inspections, then detection accuracy is improved, but selection bias occurs making the training data not representative of the overall utility set
Solution Approach 1:
The patent inverts the traditional supervised learning approach by using unsupervised learning that does not rely on human-inspected training data. This inversion eliminates selection bias because the algorithm learns from all available data without human pre-selection, improving both accuracy and data representativeness simultaneously.
Solution Approach 2:
The patent changes the fundamental parameter of training data labeling from human-inspected (supervised) to unlabeled (unsupervised). This parameter change transforms the training process to use all available data uniformly, eliminating selection bias while maintaining detection accuracy through anomaly detection capabilities.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
Method for determining losses due to non-physical and non-systematic effects of electrical energy in an electrical distribution network, which network comprises a plurality of stations for measuring time series of different electrical parameters which characterize the conditions of withdrawal of the electrical energy in the various stations and the time trend of the quantity of electrical energy withdrawn measured at the various stations, i.e. the consumption of electrical energy at said measuring stations and wherein at least said data relating to one or more of said electrical parameters and/or the time trend of the quantity of electrical energy withdrawn from said network in correspondence with said one or more measuring stations, are analyzed to determine whether any loss of energy is due to a systematic physical factor or to an anomaly resulting from incidental malfunctions of the network or from abusive withdrawals of electricity, said analysis being performed by means of a machine learning algorithm trained to classify said one or more measured data as relating to losses due to systematic physical effects or losses due to malfunctions of the network or to actions of abusive energy withdrawals, and wherein said algorithm consists of a so-called autoencoder, which is trained on a dataset comprising at least said one or more electrical parameters and the data relating to the measurements of the trends of the electrical energy withdrawn from the network or consumption.