PLTP Lahendong adalah pembangkit listrik terbesar yang mensuplai hingga 60% kebutuhan listrik di Sulawesi Utara. Pada tanggal 15 Desember 2020 PLTP Lahendong unit 1 mengalami trip secara tiba-tiba. Dari hasil investigasi, ditemukan bahwa turbin mengalami bowing dimana turbin tidak lagi lurus sesuai standar dikarenakan kegagalan maintenance. Karena pentingnya PLTP Lahendong ini dalam memenuhi kebutuhan listrik, diperlukan maintenance yang optimal. Predictive maintenance merupakan metode yang efektif dimana mampu memprediksi kerusakan yang akan terjadi berdasarkan historical data pada unit seperti data vibrasi menggunakan pemrograman dan machine learning model time series regression. Metodologi penelitian difokuskan kepada analisa dan pemrograman menggunakan bahasa pemrograman python dimulai dari import dataset, EDA (Exploratory Data Analysis), data preparation, train test, modelling, model evaluation, hyperparameter tuning, dan hasil prediksi. Model machine learning yang diuji adalah linear regression, decision tree regression, random forest regression, dan xgboost regression. Hasil yang diperoleh dari predictive maintenance menggunakan model machine learning sangat akurat dengan tingkat akurasi prediksi mencapai 96.97 % - 97.14 %. Machine learning sangat efektif digunakan dalam predictive maintenance karena tingkat akurasi prediksinya yang tinggi. Model random forest regression merupakan model yang akurasinya paling tinggi bahkan tidak perlu dilakukan hyperparameter tuning untuk meningkatkan akurasinya.

---

PLTP Lahendong is the largest power plant that supplies up to 60% of electricity needs in North Sulawesi. On December 15, 2020 PLTP Lahendong unit 1 experienced a sudden trip. From the results of the investigation, it was found that the turbine was bowing where the turbine was no longer straight according to the standard due to maintenance failure. Because of the importance of the Lahendong PLTP in meeting electricity needs, optimal maintenance is needed. Predictive maintenance is an effective method which is able to predict the damage that will occur based on historical data on the unit such as vibration data using programming and machine learning model time series regression. The research methodology focuses on analysis and programming using the python programming language starting from import datasets, EDA (Exploratory Data Analysis), data preparation, train tests, modeling, model evaluation, hyperparameter tuning, and prediction results. The machine learning models tested are linear regression, decision tree regression, random forest regression, and xgboost regression. The results obtained from predictive maintenance using machine learning models are very accurate with prediction accuracy levels reaching 96.97% - 97.14%. Machine learning is very effective in predictive maintenance because of its high predictive accuracy. The random forest regression model is the model with the highest accuracy, even if hyperparameter tuning is not necessary to increase its accuracy.