Photovoltaic Anomaly Detection via Neighbor Data Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting anomalies in photovoltaic installations are unreliable due to high false alarm rates, as they rely on deterministic models that fail to account for meteorological hazards and installation-specific parameters, leading to prolonged production losses and high costs associated with on-site technician visits.
Innovation Solution
A method that compares electricity production values of a monitored installation with those of neighboring installations to account for common random variations, using a learning-based approach to establish an estimation function and detect anomalies, thereby improving reliability and reducing false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deterministic models are used to compare measured electricity production values with theoretical values, then anomaly detection can be performed, but the false alarm rate becomes very high and reliability deteriorates
Solution Approach 1:
The patent creates a virtual copy of the photovoltaic installation by training a neural network model on historical data from the monitored installation and neighboring installations. This digital twin replicates the production patterns and allows comparison without relying on deterministic models, thereby reducing false alarms while maintaining detection capability
Solution Approach 2:
The patent introduces neighboring photovoltaic installations as intermediaries to mediate the anomaly detection process. By training the neural network on data from multiple installations in the same geographical area, the system captures common environmental variations and uses this learned behavior as a reference, filtering out false anomalies caused by weather and other shared factors
2Loss of information
If technicians are sent to photovoltaic installation sites to identify disturbance causes, then accurate diagnosis can be achieved, but costs increase significantly due to geographical distribution
Solution Approach 1:
The patent enables the photovoltaic installation monitoring system to perform self-diagnosis by automatically analyzing production data patterns and identifying probable causes of anomalies. The neural network model processes measured values and generates diagnostic information without requiring human intervention, allowing the system to serve itself in terms of monitoring and initial troubleshooting
Solution Approach 2:
The patent replaces the mechanical system of physical technician visits with an automated information processing system. The neural network model and data analysis algorithms substitute for human technicians in the initial diagnosis phase, eliminating the need for physical travel to remote installations while maintaining diagnostic capability
3Ease of operation
If monitoring of photovoltaic installations is performed remotely, then technician visits are reduced, but the ability to detect and identify disturbances deteriorates
Solution Approach 1:
The patent merges data from multiple neighboring photovoltaic installations into a unified training dataset for the neural network model. By combining information from installations in the same geographical area, the system creates a more robust reference model that captures regional patterns, enabling effective remote monitoring without sacrificing detection sensitivity
Solution Approach 2:
The patent transitions from single-installation monitoring to multi-installation comparative analysis by adding the dimension of spatial correlation. The neural network model processes data across multiple installations simultaneously, learning from the collective behavior of neighboring systems to improve remote detection accuracy without requiring physical presence at each site
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A method for remotely detecting an anomaly that affects the operation of a photovoltaic installation for producing electricity, referred to as the monitored installation (1), comprises at least one step during which electricity production values that are measured for the monitored installation are compared with estimated values that are deduced from measurements carried out for several other photovoltaic installations for producing electricity which neighbor (2-10) the monitored installation.