Junaidi, Junaidi (2021) Prediksi dan Deteksi Dini Anomali Pembacaan Data Sensor pH dan Kekeruhan Pada Sistem Pengolahan Air Berbasis SCADA. Masters thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
05111950010006-Master_Thesis.pdf - Accepted Version Download (6MB) | Preview |
Abstract
Anomali pada industri berbasis SCADA (Supervisory Control and Data Acquisition) dapat menjadi masalah jika terjadi terus menerus. Dalam industri pengolahan air, data anomali pada sensor dapat mempengaruhi kualitas air yang diproduksi. Metode prediksi yang akurat dan deteksi anomali yang dapat mendeteksi berbagai anomali sangat diperlukan untuk menunjang pengambilan keputusan.
ARIMA (Autoregressive Integrated Moving Average) adalah model statistik yang banyak digunakan dalam prediksi data time series. Dalam proses pengujian, model harus dapat mengikuti pola sinyal aktual. Kebanyakan studi tentang ARIMA menggunakan sinyal yang diobservasi secara langsung dalam pemodelan dan prediksi. Kekurangan dari metode ini, pola sinyal hasil prediksi dapat menghasilkan garis lurus daripada mengikuti pola sinyal aktual. Hal ini terjadi jika data time series tidak memiliki komponen seasonality yang kuat. Pada deteksi anomali data time series, kebanyakan penelitian fokus pada deteksi anomali data yang telah terjadi dengan mengevaluasi penyimpangan hasil prediksi terhadap data aktual. Kekurangan dari metode ini adalah tidak dapat memprediksi anomali di masa mendatang karena ketergantungan pada data aktual.
Pada penelitian ini, diusulkan metode prediksi ARIMA yang dimodifikasi. Pertama, sinyal yang diamati didekomposisi menjadi komponen trend, seasonal, dan residual. Kemudian, ketiga komponen yang terdekomposisi dimodelkan dan diprediksi secara independen. Terakhir, hasil prediksi ketiga komponen tersebut disusun kembali untuk menghasilkan prediksi sinyal yang diobservasi. Untuk deteksi anomali, diusulkan metode dengan membangun threshold reference dari data historis dengan pendekatan statistik. Statistical threshold memungkinkan deteksi anomali data sensor pada waktu kapan pun.
Berdasarkan hasil uji coba dapat disimpulkan, metode modifikasi ARIMA dapat menurunkan MSE prediksi kekeruhan 90.103% lebih rendah dibandingkan dengan metode peramalan langsung. Sedangkan penurunan MSE pada peramalan pH mencapai 96.465%. Performa terbaik deteksi anomali pada pH memiliki nilai accuracy sebesar 0.811, dan pada kekeruhan mencapai 0.823.
========================================================================================================
Anomalies in SCADA (Supervisory Control and Data Acquisition) -based industries can be a problem if they occur continuously. In the water treatment industry, anomalous data on sensors can affect the quality of produced water. Accurate prediction and anomaly detection methods that could detect various anomalies are needed to support decision making.
ARIMA (Autoregressive Integrated Moving Average) is a statistical model that is widely used in predicting time series data. In the testing process, the model must be able to follow the pattern of actual signal. Most studies about ARIMA used directly the observed signals in modeling and prediction. The disadvantage of this method is that the pattern of predicted signal produces a straight line instead of following the pattern of actual signal. This happens if the time series data does not have strong seasonality. In anomaly detection of time series data, most studies focused on detection of anomalies that have occurred by evaluating the deviation of the predicted values against the actual data. The disadvantage of this method is that it cannot predict future anomalies due to dependency on actual data.
In this study, we propose a modified ARIMA prediction method. First, the observed signal is decomposed into trend, seasonal, and residual components. Then the decomposed components are modeled and predicted independently. Finally, the predicted results of the three components are recomposed to produce the predicted values of the observed signal. For anomaly detection, we propose a method by constructing a threshold reference from historical data with a statistical approach. Statistical threshold allows anomaly detection of sensor data at any index of time.
The experimental result of this study, the modified ARIMA method can reduce the MSE of turbidity predictions down to 90.103% lower than the direct forecasting method. While the decrease of MSE in pH forecasting reaches 96.465%. The most significant performance of anomaly detection on pH has an accuracy value of 0.811, and on turbidity, it reaches 0.823.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | SCADA, time series, trend, seasonal, residual, statistical threshold |
| Subjects: | H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HD Industries. Land use. Labor > HD108 Classification (Theory. Method. Relation to other subjects ) Q Science > QA Mathematics > QA280 Box-Jenkins forecasting |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis |
| Depositing User: | Junaidi Junaidi |
| Date Deposited: | 01 Mar 2021 04:12 |
| Last Modified: | 05 Nov 2024 08:42 |
| URI: | http://repository.its.ac.id/id/eprint/83002 |
Actions (login required)
 |
View Item |