MEWMA Control Chart for Monitoring Clean Water Quality by PDAM Production in Surabaya Based on Residual of Generative Adversarial Network

Indarsanto, Raditya Widi (2024) MEWMA Control Chart for Monitoring Clean Water Quality by PDAM Production in Surabaya Based on Residual of Generative Adversarial Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003201114-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (3MB) | Request a copy

Abstract

Perusahaan Daerah Air Minum (PDAM) Surya Sembada Surabaya merupakan badan usaha milik daerah yang fokus pada penyediaan dan produksi air bersih di kota Surabaya. Salah satu instalasinya berada di kawasan Ngagel Surabaya yakni Ngagel II yang termasuk dalam proses produksi air bersih perseroan. Penelitian ini menganalisis karakteristik kualitas air khususnya pH, kekeruhan, dan KMnO4. Ketiga karakteristik tersebut mempunyai kondisi kualitas yang baik dilihat dari nilai rata-rata pada fase I dan fase II, dimana nilai tersebut berada dalam batas spesifikasi yang ditentukan. Kemudian ketiga karakteristik tersebut saling berhubungan dengan menunjukkan adanya autokorelasi. Autokorelasi dapat menyebabkan kesalahan dalam pengambilan keputusan. Maka langkah yang perlu dilakukan untuk mengatasi hal tersebut adalah dengan mencari dan membangun model baru yang tidak mengandung autokorelasi yaitu model Generative Adversarial Network (GAN). Penentuan arsitektur GAN dan hyperparameter terbaik didasarkan pada eksperimen berulang dengan tujuan meminimalkan nilai MSE, RMSE, dan MAE. Untuk input GAN digunakan dua pendekatan yaitu data asli dan data acak yang berasal dari distribusi normal (Latent Space), sehingga keluaran model GAN berupa nilai residual yang membantu mengurangi autokorelasi. Analisis dilakukan dengan menggunakan peta kendali MEWMA fase I dari hasil sisa pendekatan pertama, dengan λ yang ditetapkan sebesar 0,4. Peta kendali ditemukan terkendali secara statistik setelah sebelumnya menghadapi situasi di luar kendali. Data aktual telah berhasil meminimalkan autokorelasi dan data fase I mewakili proses di bawah kendali statistik. Jadi pada analisis fase II menggunakan λ yang sama dan terdapat beberapa observasi yang berada di luar batas kendali, artinya proses pada Tahap II tidak terkendali secara statistik. Sedangkan pada MEWMA Tahap I hasil sisa pendekatan kedua dengan λ yang ditetapkan sebesar 0,4 diperoleh terkendali secara statistik setelah penanganan situasi di luar kendali sebelumnya dan Tahap II menggunakan λ yang sama, masih terdapat beberapa observasi yang berada di luar batas kendali, artinya Fase II dengan pendekatan GAN kedua tidak terkontrol secara statistik. Analisis menggunakan diagram kendali MEWMA berbasis residual GAN menunjukkan bahwa metode ini dapat mengurangi autokorelasi pada data. Namun, dalam hal komputasi, Generative Adversarial Networks cenderung kompleks dan memerlukan banyak proses adversarial untuk melatih model guna mendapatkan hasil prediksi data terbaik.
==============================================================================================================================
Perusahaan Daerah Air Minum (PDAM) Surya Sembada Surabaya is a regionally owned business entity that focuses on providing and producing clean water in the city of Surabaya. One of the installations is in the Ngagel area of Surabaya, namely Ngagel II which is included in the company's clean water production process. This research analyzes water quality characteristics, especially pH, turbidity, and KMnO4. These three characteristics have good quality conditions seen from the average values in Phase I and Phase II, where these values are within the specified specification limits. Then these three characteristics are interconnected by showing autocorrelation. Autocorrelation can cause errors in decision making. So, the step that needs to be taken to overcome this is to look for and build a new model that does not contain autocorrelation, namely Generative Adversarial Network (GAN) model. Determining the best GAN architecture and hyperparameters is based on repeated experiments with the purpose of minimizing the MSE, RMSE, and MAE values. Two approaches are used for GAN input, namely original data and random data originating from a normal distribution (Latent Space), so that the output of the GAN model is a residual value which helps reduce autocorrelation. The analysis was carried out using the MEWMA control chart in Phase I of the residual results from the first approach, with the λ set at 0.4. The control chart is found to be in control statistically after dealing with out-of-control situations previously. The actual data has successfully minimized autocorrelation and Phase I data represents a process under statistical control. So, in the Phase II analysis using the same λ and there are several observations that were outside the control limits, means that process in Phase II was not statistically controlled. Meanwhile, in MEWMA Phase I of residual results from the second approach with λ set at 0.4 obtained statistically controlled after handling the previous out of control situation and Phase II used the same λ, there are still several observations that were outside control limits, means that Phase II with the GAN second approach is not statistically controlled. Analysis using the GAN residual-based MEWMA control chart shows that this method can reduce autocorrelation in the data. However, in terms of computing, Generative Adversarial Networks tend to be complex and require many adversarial processes to train the model to get the best predicted data results.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	PDAM, Forecasting, Generative Adversarial Network, Control Chart, MEWMA
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Raditya Widi Indarsanto
Date Deposited:	19 Aug 2024 08:11
Last Modified:	19 Aug 2024 08:11
URI:	http://repository.its.ac.id/id/eprint/115102

Actions (login required)

View Item