Hong Kong University of Science and Technology
Hong Kong University of Science and Technology Institutional RepositoryNot a member yet
162821 research outputs found
Sort by
Stochastic Compositional Optimization with Compositional Constraints
No full textStochastic compositional optimization (SCO) has attracted considerable attention because of its broad applicability to important real-world problems. However, existing work on SCO typically assumes that the projection within a solution update is straightforward, which is not the case for problem instances where constraints are in the form of expectations, such as empirical conditional value-at-risk constraints. In this paper, we introduce a novel model that integrates single-level expected-value and two-level compositional constraints into the existing SCO framework. Our model has wide applicability to data-driven optimization, fairness optimization, and risk management, including risk-averse optimization and high-moment portfolio selection, and is capable of handling multiple constraints. Additionally, we propose a class of primal-dual algorithms that generate sequences converging to the optimal solution at a rate of 𝒪(𝑁−1/2) under both single-level and two-level compositional expected-value constraints, where N is the iteration counter, thus establishing benchmarks in expected-value-constrained SCO. Numerical experiments show the efficiency of our algorithm over real-world applications
Intelligent Discrimination Between Static and Dynamic Shadows in PV Systems: A CNN-Based Temporal Pattern Recognition Approach
No full textPhotovoltaic (PV) systems suffer significant performance degradation from shadows, yet current monitoring systems cannot distinguish between static shadows requiring physical intervention and temporary dynamic cloud shadows. This misclassification causes unnecessary maintenance dispatches, costing millions annually. This thesis presents an intelligent discrimination framework using convolutional neural networks (CNN) to classify shadow types based on temporal electrical patterns. A MATLAB/Simulink simulation platform was developed to generate realistic PV responses under various shading conditions. The framework models six static shadow patterns (localized blocks, horizontal strips, uniform coverage, gradients, spots) and stochastic cloud movements, producing 12,000 labeled training samples. This addresses the critical challenge of obtaining labeled data for PV fault detection. The proposed CNN architecture processes 10-minute sequences of electrical characteristics sampled at 1 Hz. The optimized three-layer architecture (32→64→128 filters) contains only 285K parameters, enabling edge deployment. Comprehensive validation demonstrates 96.8% classification accuracy, outperforming traditional machine learning by 7.3%. The system maintains >94% accuracy under 20 dB signal-to-noise ratio and >93% with 10% missing data. This research establishes intelligent shadow discrimination as viable for operational PV systems. By accurately distinguishing static faults from dynamic shadows, the framework enables targeted maintenance strategies, reducing operational costs while improving system reliability. The work contributes to sustainable energy development by enhancing PV system efficiency and demonstrates how deep learning can solve practical challenges in renewable energy infrastructure.</p
A linearly convergent distributed heavy-ball GNE seeking algorithm for aggregative games over weight-unbalanced digraphs via finite-time consensus
No full textIn this paper, we study the generalized Nash equilibrium (GNE) seeking problem for aggregative games with affine coupling constraints in a partial-decision information scenario. In this scenario, all players attempt to seek the GNE by exchanging messages with neighbors over strongly connected and weight-unbalanced directed graphs. To overcome the challenges posed by the asymmetric nature of communication matrices, we propose a novel accelerated distributed discrete-time primal–dual algorithm, which integrates finite-time exact ratio consensus (FTERC) and heavy-ball acceleration. Moreover, it is designed to accommodate uncoordinated stepsizes and momentum parameters, enhancing convergence rate. Despite the challenges introduced by uncoordinated elements and an additional momentum term, we present a rigorous analysis demonstrating that the proposed primal–dual algorithm achieves linear convergence to the GNE with constant uncoordinated stepsizes. Finally, we validate our theoretical results with numerical examples on two different communication graphs, demonstrating the effectiveness of the algorithm and how the momentum term enhances acceleration.</p
Historical data-driven self-learning control of battery charging with convex mapping constraints
No full textThermal conditions significantly influence the battery performance and degradation, and thermal management is vital for the safe, efficient, and reliable operation of battery-powered systems such as grid-tied energy storage and electric vehicles. However, it is challenging to achieve rapid charging while minimizing the thermal impact on battery health, particularly the more critical internal battery temperature is often overlooked. Therefore, an intelligent charging control strategy is essential for effectively managing the thermal effects and enhancing charging efficiency. Given that battery charging is a highly repetitive process throughout the entire life of battery-powered systems, the historical operation data has a significant potential in the design of an effective charging control strategy. This paper proposes a novel historical data-driven self-learning control approach to iteratively optimize the battery charging strategy by applying convex mapping constraints derived from historical state information. This approach introduces a historical state convex mapping constraint, combined with a memory function to quantify the potential contribution of historical system state information and input data to improve the future control performance. The formulated historical data-based constraints and the memory function-enhanced cost function are then integrated into a model predictive control framework to optimize the battery charging current trajectories iteratively. Furthermore, to ensure that the constraints imposed on the battery electrical and thermal states are compatible with the self-learning control framework, a cascading linearized thermoelectric battery model is introduced to characterize the battery dynamics. Particularly, the internal temperature of the battery, which is not directly measurable in practical applications. Extensive simulation studies have been conducted, and the results demonstrate that the proposed control strategy can effectively regulate the internal temperature within a safe range while continuously optimizing the charging efficiency. In addition, the computation time variability is significantly reduced, with the standard deviation being decreased by approximately 80% compared to the standard MPC. The desirable control performance and continuous optimization capability make the proposed control strategy highly applicable to repetitive and complex engineering control problems.</p
Integrating urban analytics in low-noise airspace design for urban air mobility
No full textAircraft noise is a significant constraint in the urban integration of urban air mobility (UAM) system. Recent research has explored various approaches for UAM noise control within the context of urban airspace design, which could facilitate efficient noise-aware flight trajectory planning and enable integrated urban airspace management. However, a notable research gap lies in the lack of an urban analytics component in current noise-aware airspace designs. Urban analytics is essential because UAM operates in complex urban environments where the impact of noise is influenced by residents’ responses to UAM noise emissions. Leveraging modern smart city data, we address this gap through an integrated approach that combines geographic information system techniques, high-quality urban geospatial and spatio-temporal datasets, and an aircraft noise model to create 3D noise-aware no-fly zones. We also introduce a tailored optimal path planning algorithm HFPI-A* that considers both no-fly zones and flight performance constraints. A case study applies the proposed approach to a detailed examination in Hong Kong, revealing the sensitivities of various key factors in the design process and demonstrating their impact on the design outcomes.</p
Pareto optimal regulatory strategies for coupled ridesourcing and taxi markets with impatient passengers
No full textThis study develops a multi-objective bi-level programming model to identify the Pareto optimal combined regulatory strategy that simultaneously accounts for passengers, taxi drivers, ridesourcing vehicle (RSV) drivers, and the transportation network company (TNC). The upper level determines four regulatory controls, including the RSV fleet cap, taxi fare rate, government-guided RSV fare rate, and TNC wage rate floor, while the lower level obtains the steady-state market performance, which is formulated as a fixed-point problem and approximated through iterative agent-based simulations. To solve the model, a multi-objective Bayesian optimization algorithm is developed. Based on the DiDi dataset collected from Hangzhou City in 2018, our experiments demonstrate that no regulatory strategy can simultaneously benefit all stakeholders. If the government considers maximizing vehicle utilization as a secondary criterion, then it should decrease the RSV fleet cap, impose higher fare rates, and allow the TNC to pay lower wages, compared with the benchmark scenario. Furthermore, it is recommended that the government should avoid regulations that primarily favor passengers or the TNC, as our results reveal that such policies could harm other stakeholders and reduce vehicle utilization by up to 11.6%. Finally, if passengers’ impatience is overlooked, taxi drivers may lose 23.3% of potential profits.</p
Trading Strategy Leakage
No full textModern machine learning (ML) technology has shifted information extraction from simple, direct patterns to complex, deep relationships. I study how ML affects the trade-offs of information disclosure via a novel channel: the reverse-engineering of underlying strategies. First, I provide empirical evidence for this channel using a rare natural experiment in the asset management industry, where an unexpected disclosure of trades facilitates statistical learning of complex relationships—but not of simple patterns. Second, I conduct ML-based counterfactual analysis and show that disclosure frequency—by increasing the volume of training data—is central to facilitating reverse-engineering. Finally, I develop a stylized dynamic model featuring strategic interaction between a fund manager and an outsider, and demonstrate that ML raises disclosure costs (outweighing its benefits) by prompting strategic deviation to deter strategy leakage. Overall, my findings highlight that more frequent disclosure requirements can unintentionally reduce market efficiency due to reverse-engineering
Repeated-Root Cyclic Codes with Optimal Parameters or Best Parameters Known
No full textCyclic codes are the most studied subclass of linear codes and widely used in data storage and communication systems. Many cyclic codes have optimal parameters or the best parameters known. They are divided into simple-root cyclic codes and repeated-root cyclic codes. Although there are a huge number of references on cyclic codes, few of them are on repeated-root cyclic codes. Hence, repeated-root cyclic codes are rarely studied. There are a few families of distance-optimal repeated-root binary and p-ary cyclic codes for odd prime p in the literature. However, it is open whether there exists an infinite family of distance-optimal repeated-root cyclic codes over Fq for each even q ≥ 4. In this paper, three infinite families of distance-optimal repeated-root cyclic codes with minimum distance 3 or 4 are constructed; two other infinite families of repeated-root cyclic codes with minimum distance 3 or 4 are developed; seven infinite families of repeated-root cyclic codes with minimum distance 6 or 8 or 10 are presented; and two infinite families of repeated-root binary cyclic codes with parameters [2n, k, d ≥ (n - 1)/ log2 n], where n = 2m - 1 and k ≥ n, are constructed. In addition, 26 repeated-root cyclic codes of length up to 254 over Fq for q ϵ 2, 4, 8 with optimal parameters or best parameters known are obtained in this paper. The results of this paper show that repeated-root cyclic codes could be very attractive and are worth of further investigation.</p
Tunable Hetero-Intercalated 2D Superlattice for Frustrated Kondo Triangles
No full textWith the successes of van der Waals heterostructures in unveiling exotic quantum states and enabling applications, direct fabrication of 2D superlattices beyond bulk limits is appealing. We realize an unprecedented 2D Kondo superlattice, Ta-NbS2, consisting of two coupled metallic 1H-NbS2 monolayers intercalated with growth-controlled periodic Ta sublattices. One-step CVD enables tunable periodicities, yielding three Ta-moment motifs: bulk-like simple triangular geometry of (Formula presented.) a× (Formula presented.) a (Type I), 2D spin-frustrated 4a×4a (Type II), and 2D saturated (Formula presented.) a× (Formula presented.) a (Type III). Beyond the logarithmic resistance upturn at Kondo temperature (TK), we observe anomalous Hall effect evolution from single-impurity to coherent Kondo regimes, reflecting collective Ta-Kondo sublattice interactions. Carrier density modulation tunes the balance between Kondo screening and RKKY interactions, establishing 2D bilayer Ta-NbS2 as an ultra-tunable 2D Kondo superlattice for correlated quantum phenomena, with potential applications in spintronics and quantum devices.</p
Far-UVC photolysis of chlorine dioxide for micropollutant abatement in water
No full textChlorine dioxide (ClO2) outperforms free chlorine by minimizing halogenated disinfection byproducts. Upon UV photolysis, ClO2 undergoes photodecomposition, generating reactive species that can degrade micropollutants. This study investigated ClO2 photodecay and radical production under 222 nm UV irradiation, with 365 nm (the absorption peak of ClO2) as a reference. The fluence rate-based photodecay rate constant at 222 nm was 5.69-fold lower than at 365 nm, attributed to the 4.37-fold lower molar absorption coefficient of ClO2 at 222 nm compared to 365 nm, not quantum yield differences. Hydroxyl radical (HO•) production was 3.9-fold lower at 222 nm than at 365 nm, while chlorine atom (Cl•) production was 4.0-fold higher. Nitrate significantly enhanced HO• production during UV222 photolysis but had minimal impact under UVA365, leading to elevated HO• levels in secondary effluent and surface water compared with buffered deionized water. Using the experimentally determined radical concentrations, pseudo first-order degradation rate constants of seven micropollutants were predicted for UV222/ClO2 process. UV222 photolysis of ClO2 produced chlorite and chlorate, though at lower concentrations than UVA365. These results enhance understanding of ClO2 photochemistry and provide insights for selecting optimal UV radiation sources for ClO2 photodecay and micropollutant abatement in water treatment.</p