Garuda - Garba Rujukan Digital

A WireGuard Mesh Overlay Resisting Internet Provider Port Policing Aviv Maghridlo; Nardi Nardi; Marzuki Sinambela
Jurnal Info Sains : Informatika dan Sains Vol. 16 No. 02 (2026): Info sains, 2026
Publisher : SEAN Institute

Show Abstract | Download Original | Original Source | Check in Google Scholar

Overlay networks built on WireGuard can make dispersed devices appear to share one local network, but two obstacles remain: endpoints behind address translation are not directly reachable, and WireGuard's fixed default port is an obvious target for provider traffic discrimination. This study aims to design and empirically evaluate a hybrid overlay that mitigates both. We built WireGuard Manager, a single-binary Windows application combining a hub-and-spoke baseline with an opportunistic direct mesh over an in-process WireGuard data plane and host-orchestrated hole punching, and evaluated it on three physical nodes across two cities and three providers using throughput, latency, and packet-loss measurements. Results show that carrying the tunnel over the default port collapsed throughput to roughly 1–4% of a control port over the identical link (from 7.6–22.1 Mbps to at most 0.33 Mbps), while a randomized stealth port restored it; direct and relayed paths were validated independently through the observed time-to-live, and a live trace captured the automatic failover between them. A secondary result is that a direct path is not always superior to a well-provisioned relay. We conclude that provider port discrimination is a decisive, reproducible factor for such overlays and that a stealth-port strategy is an effective, low-cost mitigation, within the limits of a three-node case study.

A Hybrid GRU-BiLSTM Deep Learning Framework for Solar Radiation Forecasting and Photovoltaic Energy Yield Assessment in Timor Island, Indonesia Yobel Eliezer Mahardika; Agustina Rachmawardani; Marzuki Sinambela
INFOKUM Vol. 14 No. 04 (2026): Infokum 2026
Publisher : Sean Institute

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.58471/infokum.v14i94.3133

Timor Island in East Nusa Tenggara possesses abundant solar energy resources, yet the technical feasibility of solar power plant (PLTS) development in the region has rarely been assessed using data-driven quantitative methods. This study aims to develop and validate a hybrid GRU-BiLSTM deep learning model for short-term solar radiation forecasting, to assess whether this model can be reliably extended into long-horizon autoregressive projection, to estimate the solar radiation potential and photovoltaic energy production across five locations in Timor Island, and to characterize the radiation variability relevant to PLTS design. A hybrid GRU-BiLSTM deep learning model was developed and evaluated for short-term hourly solar radiation forecasting using ERA5 reanalysis data (2015-2025), achieving R² of 0.9507-0.9584 across five locations. Because chained autoregressive projection using this model was found to be unreliable for annual-horizon estimation (R² = -0.53, +46% overestimation), a climatological approach based on 2015-2025 historical averages was applied instead for potential assessment. The results show that the five locations possess high and relatively uniform solar potential, with annual totals ranging from 2,023 to 2,109 kWh/m² (5.54-5.78 kWh/m²/day), corresponding to estimated photovoltaic energy production of 273.1-284.8 kWh/m² per year. Coefficient of variation values (69.6-71.0%) indicate substantial short-term fluctuation despite the locations' consistent seasonal pattern, with the lowest production occurring in June and the highest in October. These findings provide a quantitative basis for PLTS capacity planning and energy storage design in Timor Island, demonstrating that a validated short-term forecasting model can be meaningfully extended into practical renewable energy resource assessment.

A Hybrid GRU-BiLSTM Deep Learning Framework for Solar Radiation Forecasting and Photovoltaic Energy Yield Assessment in Timor Island, Indonesia Yobel Eliezer Mahardika; Agustina Rachmawardani; Marzuki Sinambela
Jurnal Info Sains : Informatika dan Sains Vol. 16 No. 02 (2026): Info sains, 2026
Publisher : SEAN Institute

Show Abstract | Download Original | Original Source | Check in Google Scholar

Timor Island in East Nusa Tenggara possesses abundant solar energy resources, yet the technical feasibility of solar power plant (PLTS) development in the region has rarely been assessed using data-driven quantitative methods. This study aims to develop and validate a hybrid GRU-BiLSTM deep learning model for short-term solar radiation forecasting, to assess whether this model can be reliably extended into long-horizon autoregressive projection, to estimate the solar radiation potential and photovoltaic energy production across five locations in Timor Island, and to characterize the radiation variability relevant to PLTS design. A hybrid GRU-BiLSTM deep learning model was developed and evaluated for short-term hourly solar radiation forecasting using ERA5 reanalysis data (2015-2025), achieving R² of 0.9507-0.9584 across five locations. Because chained autoregressive projection using this model was found to be unreliable for annual-horizon estimation (R² = -0.53, +46% overestimation), a climatological approach based on 2015-2025 historical averages was applied instead for potential assessment. The results show that the five locations possess high and relatively uniform solar potential, with annual totals ranging from 2,023 to 2,109 kWh/m² (5.54-5.78 kWh/m²/day), corresponding to estimated photovoltaic energy production of 273.1-284.8 kWh/m² per year. Coefficient of variation values (69.6-71.0%) indicate substantial short-term fluctuation despite the locations' consistent seasonal pattern, with the lowest production occurring in June and the highest in October. These findings provide a quantitative basis for PLTS capacity planning and energy storage design in Timor Island, demonstrating that a validated short-term forecasting model can be meaningfully extended into practical renewable energy resource assessment.

IoT-Enabled Thermal Comfort Monitoring Using THI with XGBoost-Based Short-Term Forecasting Muhammad Afif; Marzuki Sinambela; Dibyo Susanto; Muchamad Rizqy Nugraha; Achmad Fahruddin Rais
INFOKUM Vol. 14 No. 04 (2026): Infokum 2026
Publisher : Sean Institute

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.58471/infokum.v14i03.3149

Rapid urban expansion and climate anomalies have intensified localized heat stress, demanding precise microclimatic tracking and proactive public health measures. Conventional observation networks often lack the spatial density and real-time predictive capabilities required for timely intervention. To address this challenge, this study presents the design, field implementation, metrological validation, and predictive evaluation of an Internet of Things (IoT) monitoring system integrated with eXtreme Gradient Boosting (XGBoost) for short-term Temperature-Humidity Index (THI) forecasting. The physical architecture employs a calibrated DHT22 sensor enclosed within a protective Stevenson screen and connected to an ESP32 processing node, transmitting continuous microclimatic metrics wirelessly to a Firebase cloud database. Sensor calibration against primary national standards confirmed high operational accuracy, yielding expanded uncertainties within -0.14°C to +0.18°C for ambient temperature and -1.21% to +1.76% for relative humidity, fully satisfying World Meteorological Organization (WMO) operational tolerances. Telemetry network stability evaluated under Telecommunications and Internet Protocol Harmonization Over Network (TIPHON) benchmarks demonstrated excellent performance, characterized by a low latency of 88.29 ms, packet jitter of 21.85 ms, and a high data throughput of 80293.43 bps. Furthermore, the cloud-integrated XGBoost regression model achieved exceptional accuracy in forecasting short-term thermal trends, yielding a Mean Absolute Error (MAE) of 0.1580, Root Mean Square Error (RMSE) of 0.2113, Mean Absolute Percentage Error (MAPE) of 0.5525%, and a Coefficient of Determination (R²) of 0.9681 for THI predictions. By combining low-cost, metrologically traceable IoT hardware with high-precision machine learning forecasting, this system provides a reliable and scalable framework to transition urban thermal risk management from reactive tracking to proactive climate adaptation.

Title

Found 4 Documents
Search

Abstract

Abstract

Abstract

Abstract

Title Search

Found 4 Documents Search

Abstract

Abstract

Abstract

Abstract

Title

Found 4 Documents
Search