Diyar Altinses

Accurate heat demand forecasting and control are critical for optimizing the operational efficiency of thermal energy systems, including power plants and district heating networks. This work presents a receding horizon control framework for optimizing heat production in district energy systems under real-world operational conditions. The proposed controller minimizes production costs while considering electricity market prices, which significantly affect the economic performance of heat generation units. The system coordinates multiple heterogeneous generation assets under mixed-integer programming constraints. To support the controller, we develop a forecasting framework that combines transformer and gated recurrent unit (GRU) architectures to improve prediction accuracy. This hybrid model utilizes multivariate input features, including ambient temperature and historical thermal output, to generate robust and probabilistic forecasts. The Transformer captures long-range dependencies in the data, while the GRU models temporal sequences and cyclical behavior. Experimental results on real-world data demonstrate that this method can reduce total operational costs by 16.3% and enhance resource efficiency in thermal energy production, providing a scalable solution for smart energy management.

This study aims to develop a scalable methodology for identifying optimal locations of micro-hubs and intermediate hubs within an intermodal drone delivery network. It addresses the need for novel facility location models tailored to the unique constraints of drone logistics, especially in continuous geographic spaces without predefined candidate sites. The research adopts a hybrid approach combining density-based clustering (DBSCAN) with an uncapacitated facility location model. DBSCAN is employed as a preprocessing tool to identify initial micro-hub candidate locations based on customer distribution. These are refined by a mathematical optimization model that assigns customers to the nearest feasible facility within drone range while penalizing unassigned customers. To quantify the impact of spatial discretization, a discretization loss analysis is conducted, assessing the deviation between continuous customer locations and centroid-based candidate hubs. For customers beyond drone range, intermediate hubs are introduced heuristically at fixed intervals, while a GIS-based relocation step ensures all facilities are placed in viable land areas, avoiding restricted or unusable zones, with displacement statistics reported to quantify real-world feasibility impacts. The methodology is applied in a case study involving two logistics companies in Germany. Results show that the proposed approach ensures complete customer coverage via drones, both with and without intermediate hubs, while maintaining operational feasibility and reducing complexity. Benchmark comparisons against optimal formulations, range sensitivity analysis, and scalability experiments demonstrate that the framework achieves near-optimal performance while scaling efficiently to larger urban instances. A final simulation confirms the efficiency and practicality of the designed delivery network. This work contributes a scalable decision-support tool for drone logistics planning, bridging clustering heuristics, mathematical optimization, and feasibility validation, and providing a foundation for future extensions incorporating capacity, routing, and sustainability considerations.

Reliable non-destructive hardness measurement, such as using magnetic properties, is a critical requirement in industrial quality assurance and structural health monitoring, where destructive testing is impractical or economically infeasible. Despite its importance, extracting physically meaningful and predictive features from frequency-domain magnetic signals remains challenging, particularly in scenarios where annotated measurement data are scarce and expensive to obtain. In this study, the challenge of extracting physically meaningful features from limited data is addressed by proposing a complex-valued autoencoder. This architecture is designed to jointly encode magnitude and frequency information into a compact latent representation, thereby preserving phase-sensitive structural dependencies that are typically lost in standard approaches. The method was evaluated on a real-world dataset of high-strength alloys with sparse annotations. It is demonstrated that the proposed complex-valued representations yield significantly improved regression performance, characterized by lower prediction error and reduced variance compared to a classical real-valued baseline. These results indicate that leveraging the algebraic structure of complex-valued networks offers a superior approach for non-destructive testing in low-data regimes, enabling more stable and accurate industrial inspection systems.