Background. Organic water pollution has complex, nonlinear, nonstationary, and self-similar dynamics that conventional forecasting models often fail to represent. This study develops an automated hybrid framework combining multifractal analysis and deep learning to improve Biochemical Oxygen Demand (BOD) forecasting and support early risk identification. Materials and Methods. A retrospective BOD time series from a river monitoring station was processed through an automated data and modeling pipeline. Sliding windows extracted localized features and captured long-range correlations. Multifractal Detrended Fluctuation Analysis (MFDFA) estimated generalized Hurst exponents and singularity spectra for each window. A fractional-exponential approximation converted the continuous spectra into discrete coefficients while preserving their structural characteristics. These multifractal descriptors were combined with statistical measures to form the feature space. A Long Short-Term Memory (LSTM) network was trained using time-series cross-validation and hyperparameter search. Results were compared with Autoregressive Integrated Moving Average (ARIMA) and Autoregressive Fractionally Integrated Moving Average (ARFIMA) benchmarks. Results and Discussion. The hybrid MFDFA-LSTM framework produced more accurate and stable forecasts than the benchmark models across short-term and extended horizons. Fractional-exponential descriptors supplied information about multi-scale fluctuations, persistence, spectrum width, and asymmetry, enabling the LSTM to represent nonlinear pollution dynamics more effectively. Hyperparameter optimization balanced temporal memory, model complexity, and generalization, reducing root mean square error (RMSE) and maintaining stable predictive performance. Conclusion. The study validates an automated system for water-quality forecasting. Its innovation is transforming continuous multifractal spectra into informative discrete features through fractional-exponential approximation. Combined with LSTM modeling, this framework improves forecast reliability and provides environmental authorities with an early-warning and decision-support tool for proactive water-quality management.
Background. The shift to online businesses has driven unprecedented growth in traffic density and data volumes. Engineering scalable high-load systems involves challenges like systemic latency, cascading failures from tight component coupling, I/O boundaries and CAP theorem trade-offs. Flawed architectural patterns cause performance degradation and financial losses. Therefore, the goal of this study is to analyze and systematize architectural solutions to minimize mentioned infrastructure challenges and ensure linear scalability of high-load applications. Materials and Methods. This paper presents a theoretical study on the architectural design of high-load distributed systems based on Apache Kafka. The research synthesizes a comparative analysis of academic literature with the author's practical development experience. Results and Discussion. A theoretical analysis reveals that synchronous inter-service communication based on HTTP/gRPC protocols leads to reduced efficiency in high-load environments due to cascading delays, tight coupling and thread pool exhaustion risks. The study substantiates that transitioning to an asynchronous Publish/Subscribe with a Pull-based model via Apache Kafka provides significant architectural advantages for extreme horizontal scalability. It provides a stable backpressure mechanism, prevents computational node overloading, minimizes CPU overhead through batch aggregation and simplifies broker architecture by shifting data offset persistence to the consumer. The paper systemically analyzes specific Apache Kafka engineering bottlenecks and proposes effective mitigation methods. Conclusion. The paper analyzes architectural solutions for high-load systems based on the Apache Kafka platform, aimed at increasing the efficiency of stream data processing. The suggested approaches minimize the systemic risks of asynchronous communication, ensuring protection against event chronology violation in distributed transactions, data skew and anomalous message duplication.
Background. Sheet metal bending is a key manufacturing operation in which springback — the partial elastic recovery of the workpiece after unloading — causes deviations from the target geometry and must be predicted accurately. Repeated finite element simulations are computationally expensive for wide parametric studies, motivating fast surrogate models. This work compares surrogate modeling approaches for elastic-plastic sheet bending and evaluates the diagnostic value of a physics consistency check for detecting physically implausible neural surrogates, independently of whether physics regularization improves raw predictive accuracy. Methods. A dataset of 39 568 finite element simulations was generated in COMSOL Multiphysics 6.2 by varying sheet thickness, punch displacement, and Young’s modulus of a 6111-T4 aluminum alloy blank. Five classical surrogate methods and three neural network models (a pure data-driven baseline, a Hollomon-regularized model with correct parameters, and a control model with intentionally incorrect parameters) were compared over 5 random data splits. A sensitivity sweep varied the control model's hardening exponent between the true and the incorrect value to test how sharply the diagnostic responds. Results. Classical non-parametric methods slightly outperform the neural surrogates on the dense smooth parameter grid: the random forest achieves a coefficient of determination of R² > 0.9999, compared to R² ≈ 0.999 for the neural models. Differences in raw accuracy among the three neural models are statistically detectable across seeds (paired t-test) but tiny in magnitude (ΔR² < 0.1 percentage point). The physics consistency metric, in contrast, separates the models by orders of magnitude: R² = 0.999 for the correct model versus R² ≈ −2.9 for the incorrect one. The sensitivity sweep reveals a sharp threshold rather than gradual degradation: consistency remains near-perfect while the hardening exponent error stays small, then collapses once the exponent deviates substantially from its true value. Conclusion. Standard accuracy metrics are insufficient to detect physically inconsistent neural surrogates, and the underlying diagnostic signal is thresholded rather than gradual. The physics consistency metric provides an independent diagnostic criterion that complements standard accuracy measures.
Background. Qualitative investigation of the electrical and optical characteristics of optoelectronic components is an essential part of designing, developing, and testing optoelectronic systems. For this purpose, an automated system for measuring the characteristics of optoelectronic devices was developed based on an Arduino UNO microcontroller, featuring programmable voltage control via digital potentiometers and data transfer to a computer for visualization. Additionally, the system was optimized to enhance its performance. Materials and Methods. A fully functional automated system for measuring the voltage–current and optical characteristics of optoelectronic devices has been developed and implemented based on the Arduino UNO microcontroller. The hardware part of the system is built using MCP4131 digital potentiometers and an integrated GY-302 light sensor. The desktop application was developed in Python, utilizing the Tkinter, pySerial, Matplotlib, and Threading libraries. Adaptive system optimization was carried out using the finite difference method to compute the corresponding local derivatives. Results and Discussion. Experimental investigation of the optimized system's performance demonstrated a significant increase in efficiency while maintaining high-quality characteristics. The developed adaptive optimization system showed a stable acceleration of 2.4-4.3 times, depending on the type of component. During the experiments, no cases of missing critical regions or significant distortion of characteristic curves were observed. This confirms the reliability of the developed zone-classification system and the correctness of the selected adaptive algorithm parameters. Conclusion. The results demonstrate the feasibility of creating an efficient and accessible system for studying the characteristics of optoelectronic devices based on modern microcontroller platforms. The developed adaptive optimization system proved effective under real operating conditions, making it a valuable tool for both educational and practical applications in the research and development of optoelectronic devices.
Background. Cloud-oriented Internet of Things (IoT) platforms integrate embedded sensing/actuation devices, edge nodes, and cloud services. Within these distributed cyber-physical architectures, cyber-attacks and operational faults frequently co-propagate, critically degrading availability, data integrity, and physical safety. Achieving fault tolerance and cyber resilience under severe embedded constraints remains a fundamental scientific challenge. Materials and Methods. This review targets device–edge–cloud deployments constrained by limited energy, computational capacity, memory, and intermittent connectivity. It consolidates systems engineering guidance from ISO/IEC 30141, NIST SP 800-160, ETSI EN 303 645, NISTIR 8259A, and secure update frameworks (e.g., SUIT and TUF). We construct a joint taxonomy of attacks and failures, systematize resilience patterns for identity, observability, and recovery, and integrate current research on sensor degradation monitoring. The theoretical framework introduces formal mathematical metrics for resilience loss, Markovian state transitions, and dynamic evidence budgeting. Results. The review synthesizes a generalized cyber-resilience model in which fault tolerance and cybersecurity are treated as a single adaptive observe-detect-decide-recover loop. The model is derived from the reviewed literature and combines hierarchical instrumentation, provenance-aware lifecycle control, evidence budgeting, risk-adaptive attestation, and latency-aware edge orchestration. Key gaps identified include the lack of unified metrics linking cybersecurity and dependability properties, fleet-scale attestation challenges, and resource-aware detection on constrained microcontroller endpoints. Conclusion. Fault-tolerant cyber-resilience should be considered as an adaptive observe-detect-decide-recover control loop optimized for embedded resource bounds. The review summarizes a literature-based analytical model that can serve as a basis for future experimental validation of risk-adaptive remote attestation and locality-aware orchestration in constrained IoT environments.
