dhikarachman/ml_service

NILM v9 — CNN-BiLSTM Multi-label

Model Non-Intrusive Load Monitoring (NILM) untuk mendeteksi perangkat listrik dari data PZEM (tegangan, arus, daya, PF, frekuensi).

Perangkat yang dideteksi

Label	Nama
charger_hp	Charger HP
hair_dryer	Hair Dryer
kipas	Kipas Angin
laptop	Laptop

File model

File	Deskripsi
best_nilm_model.keras	Model CNN-BiLSTM (TensorFlow/Keras)
meta_nilm.json	Metadata, scaler, threshold, label mapping
scaler_nilm.pkl	Scikit-learn scaler (backup)
nilm_inference.py	Skrip inferensi standalone

Penggunaan cepat

from nilm_inference import NilmInference

infer = NilmInference(
    model_path="best_nilm_model.keras",
    meta_path="meta_nilm.json",
)
# raw: dict dengan voltage, current, power, power_factor, frequency
result = infer.predict_single(raw)
print(result)

Input fitur (8 kolom)

voltage, current, power, power_factor, frequency, apparent_power, reactive_power, power_ratio

Window size: 30 sampel (stride 3).

ML service (Flask)

Kode layanan API ada di folder asli ml_service/ (app.py, nilm_v9_predictor.py, thingsboard_client.py).