The time resolution and prediction accuracy of the power generated by building-integrated photovoltaics are important for managing electricity demand and formulating a strategy to trade power with the grid. This study presents a novel approach to improve short-term hourly photovoltaic power output predictions using feature engineering and machine learning. Feature selection measured the importance score of input features by using a model-based variable importance. It verified that the normative sky index in the weather forecasted data had the least importance as a predictor for hourly prediction of photovoltaic power output. Six different machine-learning algorithms were assessed to select an appropriate model for the hourly power output prediction with onsite weather forecast data. The recurrent neural network outperformed five other models, including artificial neural networks, support vector machines, classification and regression trees, chi-square automatic interaction detection, and random forests, in terms of its ability to predict photovoltaic power output at an hourly and daily resolution for 64 tested days. Feature engineering was then used to apply dropout observation to the normative sky index from the training and prediction process, which improved the hourly prediction performance. In particular, the prediction accuracy for overcast days improved by 20% compared to the original weather dataset used without dropout observation. The results show that feature engineering effectively improves the short-term predictions of photovoltaic power output in buildings with a simple weather forecasting service.

With the rapid industrialization and urbanization, suburban areas have been developed to accommodate the sudden demand of the population. However, recent problems such as low fertility and aging induces urban shrinkage by reducing the urban population and the economy in old areas around the suburbs. As urban shrinkage causes inequality among residents in terms of the opportunities to access public services, the enhancement of the accessibility of public services is crucial to achieve inclusive growth. This paper proposes a framework for supplying public services based on the transit-oriented development (TOD) concept with geographic information system (GIS) analysis technique. A total of 24 indices, 4 criteria for 6 public services, are measured and weighted by the entropy method to find vulnerable residential buildings with a poor environment in Jung-nang district, Seoul. With a spatial analysis based on this weight value of residential buildings and the TOD concept, old commercial buildings are selected as candidate buildings for public services. According to the derived results, one candidate building as a public service can improve the environment of 3% to 8% of vulnerable residential buildings. The proposed decision-making methods can provide a valuable reference for selecting the location of public services by computational analysis with GIS.

An effective control of air handling unit (AHU) systems is crucial not only for managing the energy consumption of buildings but ensuring indoor thermal comfort for occupants. Although the initial control schema of AHU is appropriate at installation and testing, it is frequently necessary to adjust the control variables due to the changing thermal response of the building envelope and space usage. This paper presents a novel optimization process for the control parameters of old AHU systems in existing commercial buildings without system downtime and massive operational data. First, calibrating the building and system simulator with limited system operation data and unknown building parameters can provide identical responses to the system operation with the Hooke–Jeeves algorithm during the cooling season. The deep deterministic policy gradient algorithm is employed to determine the optimal control parameters for the valve opening position of the cooling coil within less than three hours of training based on the calibrated simulator. By using actual implementations with the developed optimal control variables for an old AHU in a real building, the proposed auto-tuned PID control in the simulator and with machine learning improves thermal environments with a steady room temperature (23.5 ± 0.5 °C) by 97% in occupied periods. It is also proved that this can reduce cooling energy consumption by up to 13.71% on a daily average. The successful AHU controller can improve not only the stability of AHU systems but the efficiency of a building’s energy use and indoor thermal comfort.

Photovoltaics are methods used to generate electricity by using solar cells, which convert natural energy from the sun. This generation makes use of unlimited natural energy. However, this generation is irregular because they depend on weather occurrences. For this reason, there is a need to improve their economic efficiency through accurate predictions and reducing their uncertainty. Most researches were conducted to predict photovoltaic generation with various machine learning and deep learning methods that have complicated structures and over-fitted performances. As improving the performance, this paper explores the probabilistic approach to improve the prediction of the photovoltaic rate of power output per hour. This research conducted a variable correlation analysis with output values and a specific EM algorithm (expectation and maximization) made from 6054 observations. A comparison was made between the performance of the EM algorithm with five different machine learning algorithms. The EM algorithm exhibited the best performance compared to other algorithms with an average of 0.75 accuracies. Notably, there is the benefit of performance, stability, the goodness of fit, lightness, and avoiding overfitting issues using the EM algorithm. According to the results, the EM algorithm improves photovoltaic power output prediction with simple weather forecasting services.

Accurate player tracking in American football typically relies on multi-camera or sensor-based systems. However, high costs limit their use to elite leagues. While monocular video analysis offers a scalable alternative, it remains challenging due to dense formations, frequent contact, and prolonged occlusions, which lead to identity fragmentation in standard tracking pipelines. We propose CBY (CLIP–ByteTrack–YOLO), a monocular tracking framework tailored to the dense-contact nature of American football. The system integrates a football-finetuned YOLO11 detector with ByteTrack for association. We incorporate a CLIP-based module to assign semantic attributes to trajectories, enabling automatic team and referee classification without manual calibration. Furthermore, we apply domain-specific constraints to minimize spurious associations and stabilize identity tracking during occlusions. Experiments on 35 broadcast videos (263,897 detections) show that domain-adapted detection improves tracking viability, raising MOTA from −9.16% to 58.96%. The CBY pipeline achieves identity preservation with 74.0% team classification accuracy using a single camera setup. These results indicate that domain-adapted detection, rather than association complexity, is the primary determinant of tracking reliability under dense occlusion.

Multi-object tracking (MOT) research has predominantly focused on improving data association—designing better cost functions, motion models, and re-identification features—under the implicit assumption that detection is a solved upstream problem. We challenge this assumption for a class of scenarios we term dense-contact regimes, where spatial collapse among targets renders association ill-posed regardless of its sophistication. Using American football as a representative case study, we provide empirical evidence across five tracking backbones (ByteTrack, DeepSORT, BoT-SORT, StrongSORT, Deep OC-SORT) that domain-adapted detection yields up to a +68-point MOTA improvement, while varying association strategy produces only marginal differences. This finding identifies a regime shift in MOT: under dense contact, tracking is not association-limited but detection-limited. To further stabilize identity under extreme occlusion, we introduce a lightweight semantic regularization mechanism that leverages vision–language embeddings (CLIP) as constraints on identity consistency rather than standalone classifiers, achieving 74.0% team classification accuracy without manual calibration. Ablation analysis confirms that detection and semantic modules can be decoupled and independently optimized without cross-interference, establishing a practical design principle for composite tracking pipelines. Experiments on 35 broadcast videos (263,897 detections) from Korean amateur American football validate these findings. Our results have implications beyond sports analytics to any domain featuring crowded, visually homogeneous targets under persistent occlusion.