Using unmanned aerial vehicles (UAVs) in modern agriculture faces the problem of overcoming obstacles. One of the most widespread types of these obstacles on the agricultural land is powerlines and aerial communication lines, which are crucial for agriculture itself as well as for other economic activities, and thus cannot be removed from the agricultural lands. The initial subtask in overcoming poles is the detection of such objects. Recent neural network detection architectures, such as YOLO, have shown promising results in general object detection tasks, however, the results of comparative studies of YOLO architectures in a specific task of pole detection are not presented in scientific literature. In this work, we present results of a comparative study of a set of YOLO architectures’ performance on a custom dataset of powerlines and aerial communication lines poles on the agricultural land obtained using a UAV. The dataset consists of 3508 images with 1691 wooden poles and 1750 concrete poles. We consider five recent YOLO architectures from v8 to v12. Comparative analysis of the considered architectures has shown that YOLOv11 achieved the best performance in average according to recall (0.765), precision (0.798), mAP@50 (0.809) and mAP@50–90 (0.484) metrics. These results, along with the least required computational resources (6.5 GFLOPS), make YOLOv11 the most appropriate architecture for pole detection on the agricultural land.
Financial markets are strongly nonstationary, making short-horizon foreign exchange decision support sensitive to estimation windows, feature design, and transaction costs. This paper proposes a transparent multiwindow ensemble framework in which interpretable multiregression experts use rolling-statistic features computed over different lookback windows. Time-scale diversity is combined with a supervisory layer that converts recent cost-aware utility scores into expert weights and selects execution thresholds on a meta window. The utility criterion is based on net pips after transaction costs and includes penalties for drawdown and turnover, while the execution rule incorporates volatility gating and minimum holding constraints. The framework is evaluated on synchronized one-minute quotes for 16 major currency pairs using a nonoverlapping walk-forward protocol. The benchmark set includes lag-based machine-learning models, LSTM, Transformer, Echo State Network, Bayesian model averaging, dynamic model averaging, stacking, and alternative multiexpert aggregation rules. In the main fixed-horizon experiment with τ=15 minutes, the proposed utility-weighted ensemble achieves the highest mean Sharpe ratio of 1.085 and a mean net profit of 312.3 pips. The results support utility-calibrated time-scale diversification as an auditable approach to FX decision support under nonstationarity.
Financial time series in volatile markets often exhibit non-stationary behavior and signatures of stochastic chaos, challenging traditional forecasting methods based on stationarity assumptions. In this paper, we introduce a novel multi-expert forecasting system (MES) that leverages ensemble machine learning techniques—including bagging, boosting, and stacking—to enhance prediction accuracy and support robust risk management decisions. The proposed framework integrates diverse “weak learner” models, ranging from linear extrapolation and multidimensional regression to sentiment-based text analytics, into a unified decision-making architecture. Each expert is designed to capture distinct aspects of the underlying market dynamics, while the supervisory module aggregates their outputs using adaptive weighting schemes that account for evolving error characteristics. Empirical evaluations using high-frequency currency data, notably for the EUR/USD pair, demonstrate that the ensemble approach significantly improves forecast reliability, as evidenced by higher winning probabilities and better net trading results compared to individual forecasting models. These findings contribute both to the theoretical understanding of ensemble forecasting under chaotic market conditions and to its practical application in financial risk management, offering a reproducible methodology for managing uncertainty in highly dynamic environments.
This paper addresses the challenge of online forecasting for inherently unstable technological processes—such as chaotic gas- and hydrodynamic flows—by leveraging metric-based machine-learning algorithms. Traditional models, including the Wold decomposition and stationary Gaussian noise assumptions, fail to capture the complex interplay of deterministic chaos and stochastic fluctuations present in turbulent environments. To overcome these limitations, the authors propose a precedent-based forecasting framework in which the current state is compared, via a chosen similarity metric, against historical segments ("analogs") drawn from a high-dimensional retrospective database. Forecasts are then generated by extrapolating the observed evolution following the most similar past situations, with accuracy improvements achieved through multi-analog averaging and similarity-weighted ensembles. Empirical validation on real-world oil-refining process data demonstrates that the metric approach can maintain relative prediction errors within 2–3% over one-hour horizons. The paper further discusses the extension to multivariate series using Mahalanobis and information-theoretic metrics, highlighting the method’s adaptability to nonstationary, heteroskedastic data. This work illustrates a promising, data-driven alternative to classical statistical and deterministic chaos models for real-time control of complex industrial processes.
Purpose of research. Evaluation of the effectiveness of the UAV automatic landing system on a mobile platform using an infrared beacon based on criteria for landing accuracy and maneuver success at various altitudes. Methods. Modeling the process of movement of a complex object (UAV) in the Gazebo environment using the ROS ecosystem. The positioning of the UAV is based on a mathematical model of an infrared beacon consisting of four pairs of emitters. The landing algorithm includes adaptive PID controllers for the X and Y coordinates and a logo polynomial controller to ensure the descent of the UAV along the Z axis. Results. The UAV landing was tested 50 times from heights of 5 m, 10 m and 15 m. At a height of 5 m, the landing time was 9.04 seconds (0.504 sec deviation), the error was 0.18 m (0.035 m deviation), the success rate was 100 %. At 10 m, the time increased to 19.17 seconds (1.78 sec deviation), the error was 0.19 m (0.036 m deviation), the success rate remained 100 %. At 15 m, the time increased to 40.45 seconds (5.502 seconds deviation), the error was 0.21 m (0.046 m deviation), the data distribution became wider, outliers appeared, the success rate decreased to 92 %, which is due to signal losses, their attenuation and the need to correct the trajectory. Increasing the height of the landing process testing is impractical due to a decrease in the probability of a successful landing. Conclusion. The study showed that the infrared beacon system works effectively for landing UAVs on a mobile platform at altitudes up to 10 m, providing the necessary stability and accuracy. At altitudes above 10 m, problems arise with loss of signals, increased landing time and errors, which require improvements to ensure the reliability of landing.
