Implementasi Chatbot Retrieval-Based Untuk Rekomendasi Parameter PLTS Rooftop Menggunakan TF-IDF Dan Cosine Similarity

Kurniafandi, Rifqi Dwi (2026) Implementasi Chatbot Retrieval-Based Untuk Rekomendasi Parameter PLTS Rooftop Menggunakan TF-IDF Dan Cosine Similarity. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221140-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Pemantauan parameter Pembangkit Listrik Tenaga Surya (PLTS) rooftop diperlukan untuk mengetahui kondisi operasional dan produksi energi sistem. Meskipun data daya, arus, tegangan, energi harian, dan energi bulanan telah tersedia pada dashboard monitoring, pengguna masih harus membuka dashboard, mencari parameter yang dibutuhkan, serta menafsirkan data secara langsung. Selain itu, pertanyaan pengguna dapat disampaikan menggunakan susunan kalimat yang berbeda dan mengandung kesalahan pengetikan, sehingga pencarian berbasis kata kunci sederhana berpotensi menghasilkan pencocokan maksud atau intent yang kurang tepat. Oleh karena itu, diperlukan antarmuka percakapan yang dapat memberikan akses informasi monitoring dan rekomendasi operasional awal secara lebih praktis. Penelitian ini mengimplementasikan chatbot retrieval-based berbasis Telegram untuk menjawab pertanyaan mengenai parameter PLTS rooftop. Sistem dibangun menggunakan n8n dengan mengintegrasikan Home Assistant sebagai sumber data terkini dan MariaDB sebagai penyimpanan data historis. Basis pengetahuan terdiri atas 22 intent dan 158 contoh pertanyaan. Pertanyaan pengguna diproses melalui normalisasi teks, koreksi kesalahan pengetikan, stopword removal, dan tokenisasi. Levenshtein Distance digunakan sebagai metode pendukung untuk mengoreksi kata yang tidak dikenali dengan nilai ambang word similarity sebesar 0,55. Selanjutnya, TF-IDF memberikan bobot pada setiap term, sedangkan Cosine Similarity menentukan contoh pertanyaan dengan tingkat kemiripan tertinggi. Hasil pencocokan diterima apabila nilai kemiripan memenuhi threshold sebesar 0,20. Sistem kemudian menggabungkan hasil tersebut dengan data monitoring untuk menampilkan status parameter, data historis, estimasi Performance Ratio, ekuivalen nilai energi, emisi karbon yang dihindari, serta rekomendasi operasional awal. Pengujian terhadap 110 pertanyaan menghasilkan 90 prediksi intent benar, 17 prediksi salah, dan 3 keluaran tidak valid, sehingga diperoleh akurasi end-to-end sebesar 81,82% dan macro F1-score sebesar 83,88%. Pengujian terhadap 30 pertanyaan yang mengandung kesalahan pengetikan menghasilkan tingkat keberhasilan pengenalan 100%, sedangkan 20 pertanyaan di luar domain berhasil dibatasi oleh sistem. Pengujian terhadap 140 eksekusi menghasilkan rata-rata waktu respons end-to-end sebesar 18,01 detik. Hasil tersebut menunjukkan bahwa chatbot dapat digunakan sebagai conversational human-machine interface untuk membantu pengguna mengakses informasi monitoring dan rekomendasi operasional awal PLTS melalui Telegram. Rekomendasi yang dihasilkan bersifat indikatif karena sistem belum menggunakan data iradiasi matahari, suhu modul, soiling, dan shading.
===================================================================================================================================
Monitoring the parameters of a rooftop photovoltaic (PV) system is required to determine its operating condition and energy production. Although power, current, voltage, daily energy, and monthly energy data are available on the monitoring dashboard, users still need to open the dashboard, locate the required parameter, and interpret the data directly. User questions may also be expressed using different sentence structures and may contain typing errors, causing simple keyword matching to identify the intended meaning inaccurately. Therefore, a conversational interface is required to provide more practical access to monitoring information and initial operational recommendations. This study implements a retrieval-based chatbot on Telegram to answer questions about rooftop PV parameters. The system is developed in n8n by integrating Home Assistant as the current data source and MariaDB as historical data storage. The knowledge base contains 22 intents and 158 sample questions. User questions are processed through text normalization, typing-error correction, stopword removal, and tokenization. Levenshtein Distance is used as a supporting method to correct unrecognized words with a word-similarity threshold of 0.55. TF-IDF then assigns weights to each term, while Cosine Similarity identifies the most similar sample question. A match is accepted when the similarity score reaches the 0.20 threshold. The system combines the retrieval result with monitoring data to present parameter status, historical data, estimated Performance Ratio, energy-value equivalent, avoided carbon emissions, and initial operational recommendations. Testing on 110 questions produced 90 correct intent predictions, 17 incorrect predictions, and 3 invalid outputs, resulting in an end-to-end accuracy of 81.82% and a macro F1-score of 83.88%. All 30 questions containing typing errors were recognized successfully, while 20 out-of-domain questions were restricted by the system. Testing over 140 executions yielded an average end-to-end response time of 18.01 seconds. These results indicate that the chatbot can serve as a conversational human-machine interface that helps users access rooftop PV monitoring information and initial operational recommendations through Telegram. The recommendations remain indicative because the system does not yet use solar irradiance, module temperature, soiling, or shading data.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Chatbot, Retrieval-Based, PLTS Rooftop, TF-IDF, Cosine Similarity, Monitoring, Chatbot, Retrieval-Based, Rooftop PV, TF-IDF, Cosine Similarity, Monitoring.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK1087 Photovoltaic power generation T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms Z Bibliography. Library Science. Information Resources > ZA Information resources > Z699.5 Information storage and retrieval systems
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Rifqi Dwi Kurniafandi
Date Deposited:	10 Aug 2026 02:27
Last Modified:	10 Aug 2026 02:27
URI:	http://repository.its.ac.id/id/eprint/144259

Actions (login required)

View Item