A Cloud-Based Adaptive ConvolutionalNetwork for Real-Time Error Detection in Energy Meter Systems

Abstract

To solve the problem of real-time measurement error detection of intelligent electric meter, this paper proposes a cloud edge collaborative online detection framework based on CNN-GRU mixed depth model and entropy weight scoring mechanism. The system extracts the local variation characteristics by one-dimensional convolution, and uses two-way GRU to model the time-dependent error evolution. Meanwhile, the dynamic threshold score function is introduced to realize the self-adaptive judgment of error alarm. The special data sets R-Elec, I-Load and S-ErrGen covering residential, industrial and rural scenes are constructed, and pre-processed through standardization, sliding window segmentation and statistical cleaning. The experimental results show that the average identification accuracy of the system in the three types of data is 97.6% and the average response time is 0.62 seconds, which is significantly superior to the performance of traditional models (such as RF, SVM, LSTM) in false alarm control and adaptability. The ablation experiments validated the key role of multi-source feature fusion and entropy scoring mechanisms in performance improvement. The research shows that the fusion of lightweight depth model and cloud edge architecture can achieve efficient, low delay and extensible intelligent meter error online detection, and has the practical engineering deployment value.