Sistem Smart Meter Untuk Deteksi Anomali Kondisi Beban Listrik Berbasis Neural Network

Raharjo, Sudhamono Putra Nyoto (2026) Sistem Smart Meter Untuk Deteksi Anomali Kondisi Beban Listrik Berbasis Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221140-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Pemantauan konsumsi daya kelistrikan secara cerdas menjadi krusial dalam mendeteksi kegagalan operasional dini pada beban induktif seperti motor listrik pompa air dan mesin bor AC. Smart meter konvensional umumnya hanya mampu mencatat akumulasi penggunaan energi tanpa kapabilitas interpretasi anomali fisis secara real-time. Penelitian ini merancang sebuah sistem smart meter terintegrasi kecerdasan buatan berbasis jaringan saraf tiruan Long Short-Term Memory (LSTM) untuk mengklasifikasikan kondisi operasional normal dan anomali. Perangkat keras sirkuit smart meter dibangun menggunakan mikrokontroler ESP32, modul sensor kelistrikan PZEM-004T, dan konverter AC ke DC HLK-10M05. Pipa pra-pemrosesan data menerapkan penormalan skala StandardScaler berbasis Z-score dan pembersihan data pencilan menggunakan metode Interquartile Range (IQR) dengan imputasi median untuk menjaga integritas data sekuensial. Melalui pendekatan sliding window berukuran tiga sampel, array parameter listrik diubah menjadi matriks temporal tiga dimensi yang mencakup parameter tegangan, arus, daya aktif, daya reaktif, dan faktor daya sebagai input model LSTM. Hasil pengujian menunjukkan bahwa model LSTM sangat andal dalam mengenali karakteristik unik tanda tangan elektrikal (electrical signature), seperti kondisi dry running pada pompa air dan kondisi beban kerja dinamis pada mesin bor. Validasi melalui pengujian sampel data real-time menghasilkan performa klasifikasi dengan tingkat kesesuaian prediksi yang sangat presisi terhadap label aktual di setiap kondisi beban. Integrasi framework backend Flask berbasis multithreading mampu melangsungkan proses inferensi model h5 secara real-time dengan kecepatan tinggi, memastikan visualisasi status tanpa latency pada halaman dashboard frontend. Sistem cerdas ini terbukti andal dan berpotensi besar diimplementasikan sebagai sistem proteksi dini motor induksi pada skala domestik maupun industri.
==========================================================================================================================================
Smart monitoring of electrical power consumption is crucial for the early detection of operational failures in inductive loads such as water pump electric motors and AC drilling machines. Conventional smart meters generally only record cumulative energy usage without the capability to interpret physical anomalies in real-time. This study designs an artificial intelligence-integrated smart meter system based on a Long Short-Term Memory (LSTM) neural network to classify normal and anomalous operational conditions. The smart meter circuit hardware is constructed using an ESP32 microcontroller, a PZEM-004T electrical sensor module, and an HLK-10M05 AC-to-DC converter. The data preprocessing pipeline implements Z-score-based StandardScaler normalization and outlier data removal using the Interquartile Range (IQR) method with median imputation to maintain sequential data integrity. Through a sliding window approach with a size of three samples, the electrical parameter array is transformed into a three-dimensional temporal matrix containing voltage, current, active power, reactive power, and power factor as inputs for the LSTM model. The experimental results demonstrate that the LSTM model is highly reliable in recognizing the unique characteristics of electrical signatures, such as dry running conditions on the water pump and dynamic workload conditions on the drilling machine. Validation through real-time sample data testing yields a highly precise classification performance, achieving an exact prediction match against the actual labels across all load conditions. The integration of a multithreading-based Flask backend framework enables real-time inference of the .h5 model with high-speed execution, ensuring latency-free status visualization on the frontend dashboard page. This intelligent system is proven to be reliable and has great potential for implementation as an early protection system for induction motors at both domestic and industrial scales.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Smart meter, Internet of Things, Neural Network, PZEM, ESP32.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK3070 Automatic control T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK351 Electric measurements. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Sudhamono Putra Nyoto Raharjo
Date Deposited:	18 Jul 2026 04:19
Last Modified:	18 Jul 2026 04:19
URI:	http://repository.its.ac.id/id/eprint/135362

Actions (login required)

View Item