Structural Factor Combinations Driving SPKLU Utilisation in Indonesia: A Two-Stage Regression and Residual Diagnostic Approach

Pratama, Kevin Bahari (2026) Structural Factor Combinations Driving SPKLU Utilisation in Indonesia: A Two-Stage Regression and Residual Diagnostic Approach. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5010221057-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

aringan SPKLU di Indonesia telah berkembang pesat, namun utilisasi stasiun masih sangat tidak merata: sekitar setengah dari seluruh SPKLU dengan masa operasi lebih dari 90 hari mencatat throughput harian mendekati nol, sementara sebagian kecil stasiun berkinerja tinggi mendominasi keluaran total jaringan. Penelitian ini menganalisis faktor-faktor struktural dan kontekstual yang menentukan apakah sebuah stasiun digunakan sama sekali dan, di antara yang sudah aktif, seberapa intensif penggunaannya. Menggunakan dataset nasional PLN sebanyak 2.489 stasiun, penelitian ini menerapkan pendekatan dua tahap: regresi logistik biner pada Tahap 1 untuk memodelkan probabilitas aktivasi atas keseluruhan sampel, dan regresi OLS pada Tahap 2 terhadap subset aktif (n = 1.202) dengan log₁₀(kWh/hari) sebagai ukuran intensitas. Kedua tahap menggunakan set prediktor yang sama mencakup jumlah port, komposisi kelas kecepatan pengisian, jarak ke kompetitor terdekat, kapasitas pengisian ratarata per port, pengeluaran tahunan per kapita tingkat kota, rezim akses, wilayah, tipe lokasi, kepemilikan mesin, dan merek charger; Tahap 2 juga menyertakan istilah interaksi antara komposisi charger fast dan ultra fast dengan wilayah serta tipe lokasi. Model Tahap 2 menghasilkan R² = 73,23% dengan uji lack-of-fit yang tidak signifikan. Temuan utama meliputi: kapasitas pengisian rata-rata per port merupakan prediktor tunggal terkuat untuk intensitas throughput; komposisi charger medium secara konsisten menekan baik aktivasi maupun intensitas; aksesibilitas publik merupakan salah satu prediktor kategorikal terkuat di kedua tahap; dan imbal hasil marjinal dari penempatan charger ultra fast tertinggi di Jawa NonMetro, melampaui Jabodetabek. Analisis residual mengidentifikasi 20 stasiun outlier yang diklasifikasikan ke dalam enam kluster faktor-X yang merepresentasikan batas struktural dari model kuantitatif. Temuan-temuan ini diterjemahkan ke dalam matriks keputusan penempatan berbasis koefisien regresi sebagai acuan bagi perencana dalam memprioritaskan tipe lokasi, wilayah, dan konfigurasi kelas kecepatan pada gelombang ekspansi SPKLU berikutnya
======================================================================================================================================
Indonesia's public EV charging network (SPKLU) has grown aggressively, yet utilisation remains critically uneven: roughly half of all stations with more than 90 days of operation record near-zero daily throughput, while a small number of high-performing stations drive the bulk of network output. This thesis investigates the structural and contextual factors that determine both whether an SPKLU station is used at all and, among those that are, how intensively it is used. Using a national operator-level dataset of 2,489 stations drawn from PLN, a two-stage modelling approach is applied: a Stage 1 binary logistic regression on the full sample models the probability of activation (≥ 0.5 kWh/day), while a Stage 2 OLS regression on the active subset (n = 1,202) models log₁₀(kWh/day) as the intensity outcome. Both stages share a common predictor set covering port count, charger speed class composition, distance to the nearest competitor, average charging capacity per port, city-level annual spending, access regime, region, site type, machine ownership, and charger brand; Stage 2 additionally incorporates interaction terms between fast and ultra-fast charger composition and both region and site type. The Stage 2 model achieves R² = 73.23% with a non-significant lack-of-fit test, indicating adequate functional form. Key findings include: average charging capacity per port is the single strongest predictor of throughput intensity; medium charger composition consistently suppresses both activation and intensity; public accessibility is one of the strongest categorical predictors across both stages; and the marginal return to ultra-fast charger deployment is highest in Java Non-Metro, exceeding even the Jakarta Metropolitan Area. A residual diagnostic step flags 20 stations with |standardised residual| > 2.5, classifying their deviations into six X-factor clusters — misclassified access regime, captive demand advantage, market demand mismatch, capacity overhang, temporal ramp-up effect, and raw data inaccuracy — each representing a structural limit of the quantitative model. These findings are translated into a deployment scenario analysis, providing planners and operators with an evidence-based reference for prioritising site type, region, and charger speed configurations in the next wave of SPKLU expansion.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	EV Charging Station, EV Charging Utilisation, Two-Stage Regression, Charger Speed Class, Deployment Strategy, SPKLU, Tingkat Penggunaan Pengisian Daya EV, Regresi Dua Tahap, Kelas Kecepatan Pengisi Daya, Strategi Penempatan
Subjects:	T Technology > TL Motor vehicles. Aeronautics. Astronautics > TL220.5 Battery charging stations (Electric vehicles)
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26201-(S1) Undergraduate Thesis
Depositing User:	Kevin Bahari Pratama
Date Deposited:	01 Aug 2026 05:05
Last Modified:	01 Aug 2026 05:05
URI:	http://repository.its.ac.id/id/eprint/141262

Actions (login required)

View Item