TIME SERIES-BASED SPAREPARTS PROCUREMENT RECOMMENDATION MODELLING FOR POWER PLANTS USING MACHINE LEARNING

• Main Author: Mushawir, Ahmad
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86670
• Institution: Institut Teknologi Bandung

To address the challenges of timely and efficient spare parts procurement, PT PLN Nusantara Power (PLN NP) requires an innovative approach to optimize the process and minimize operational risks. This study aims to develop a time-seriesbased recommendation model using machine learning methods, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Prophet, and Recurrent Neural Network (RNN). The model is designed to accurately predict spare part prices, usage quantities, and purchase timing, thereby reducing the risk of power plant trips or derating. The research utilizes historical data from PLN NP’s Enterprise Asset Management (EAM) system, encompassing purchase orders and warehouse transactions from 2012 to 2023. The data underwent preprocessing, including cleaning, aggregation, normalization, and handling missing values. Model evaluation was conducted using metrics such as MAE, RMSE, MAPE, R-squared for price and quantity predictions, and Accuracy, Precision, Recall, and F1-Score for purchase timing predictions. The results demonstrate that LSTM and GRU models outperform Prophet and RNN in terms of accuracy and relevance for procurement recommendations. By implementing this recommendation model, potential procurement delays can be reduced by up to 80%, potentially saving up to 44.8 billion IDR annually. This study contributes significantly to improving PLN NP’s operational efficiency and sustainability while providing valuable insights for spare parts management in the energy generation sector.