Deep Reinforcement Learning for the Detection of Abnormal Data in Smart Meters.

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Smart grids generate vast amounts of power data, posing significant security management challenges.
• Processing large-scale power data using artificial intelligence is a critical research area.
• Early warning detection for smart meter data anomalies is essential for grid stability.

Purpose of the Study:

• To propose an advanced abnormal data detection network for smart meters.
• To enhance the accuracy and efficiency of anomaly detection in smart grid data.
• To address the security management difficulties arising from rapidly growing power data.

Main Methods:

• Developed a Deep Reinforcement Learning (DRL) network comprising a main and target network.
• Employed the greedy policy algorithm and Q-learning for optimal policy calculation.
• Integrated fuzzy c-means for predicting future state information to improve DRL computational accuracy.

Main Results:

• The proposed DRL model demonstrated improved accuracy in detecting meter data anomalies.
• Experimental results show superior performance compared to traditional smart meter data anomaly detection methods.
• The integration of fuzzy c-means enhanced the overall computational accuracy of the DRL model.

Conclusions:

• The Deep Reinforcement Learning network effectively detects abnormal data in smart meters.
• This method offers a significant improvement over existing anomaly detection techniques.
• The study contributes to enhanced security management in smart grids through accurate data analysis.