Research on Error Correction Algorithm for Intelligent Energy Metering System Based on Deep Learning

Abstract

Smart energy metering systems enable energy operators to allocate energy efficiently. However, these metering systems are not exempt from hardware impracticalities, environmental effect, and communication errors that sometimes produce, as a result, potentially erroneous data. For these reasons, this research proposes an approach to deep learning (DL) error mitigation framework (C-ESA-LSTMAE), incorporating various factors into the training model focused upon the object of the Chaotic-Enriched Seahorse Algorithm-Mutated LSTM Autoencoder. This research utilizes energy consumption logs consisting of voltage, current, frequency, and power from the Kaggle smart meter dataset. The preprocessing of the dataset took many forms, to include identifying missing value, chats data with outliers, identifying duplicated values, Min-Max normalizing, and applying a Discrete Fourier transform (DFT) to all time-domain features as well frequency-domain features to best leverage use of features for purposes of anomaly detection and anomaly mitigation. The C-ESA-LSTMAE proposed framework also made a comparison to various baseline contenders and other prior traditional error correction methods, as well tested with a standard LSTM Autoencoder for the sake of comparison. The results were significant and lend a lot of promise as to the proposed C-ESA-LSTMAE approach and framework. C-ESA-LSTMAE produced errors considerably lower than the baseline (MAPE = 0.01413, RMSE = 0.02011, MAE = 0.01242, MSE = 0.2642), and in comparison to the baseline's, C-ESA-LSTMAE produced errors that were 18-25% lower in different scenarios (with comparable error levels) claiming it was able to increase the reliability and accuracy of smart meter data. Paired t-tests using IBM SPSS 26.0 were conducted between the baseline models and the proposed C-ESA-LSTMAE framework on the same dataset splitsThe C-ESA-LSTMAE framework represents a considerable opportunity for developing a real-world error abatement solution for Active Power distribution networks, which maxim IZES efficiencies in operational costs AND, ultimately, the auditing of energy consumption.