Initial Model Parameters Research Articles

Multi-Level Split Federated Learning for Large-Scale AIoT System Based on Smart Cities

In the context of smart cities, the integration of artificial intelligence (AI) and the Internet of Things (IoT) has led to the proliferation of AIoT systems, which handle vast amounts of data to enhance urban infrastructure and services. However, the collaborative training of deep learning models within these systems encounters significant challenges, chiefly due to data privacy concerns and dealing with communication latency from large-scale IoT devices. To address these issues, multi-level split federated learning (multi-level SFL) has been proposed, merging the benefits of split learning (SL) and federated learning (FL). This framework introduces a novel multi-level aggregation architecture that reduces communication delays, enhances scalability, and addresses system and statistical heterogeneity inherent in large AIoT systems with non-IID data distributions. The architecture leverages the Message Queuing Telemetry Transport (MQTT) protocol to cluster IoT devices geographically and employs edge and fog computing layers for initial model parameter aggregation. Simulation experiments validate that the multi-level SFL outperforms traditional SFL by improving model accuracy and convergence speed in large-scale, non-IID environments. This paper delineates the proposed architecture, its workflow, and its advantages in enhancing the robustness and scalability of AIoT systems in smart cities while preserving data privacy.

Clinical and economic effectiveness of using immunotherapeutic drugs in patients with urothelial carcinoma in the healthcare system of the Russian Federation

Aim. To perform pharmacoeconomic evaluation of chemotherapy schemes GC (gemcitabine, cisplatin) and GemCarbo (gemcitabine, carboplatin) in comparison with immunotherapy drugs atezolizumab, pembrolizumab or avelumab in patients with locally advanced or metastatic urothelial carcinoma.Materials and methods. Pharmacoeconomic cost–effectiveness analysis, sensitivity analysis in context of changes of initial model parameters were performed.Results. Literature data analysis allows to make a conclusion of better clinical effectiveness and safety of immunotherapy drugs compared to chemotherapy in patients with urothelial carcinoma. Cost of medications was significantly lower for platinum-based chemotherapy (103,625.61 rubles for GC and 88,733.63 rubles for GemCarbo) compared to a course of immunotherapy (950,092.39 rubles for atezolizumab, 953,340.21 rubles for pembrolizumab, 1,328,999.43 rubles for GC + avelumab). However, the cost of treatment of complications arising during platinum-based chemotherapy was more than 20-fold higher than cost of treatment of immunotherapy complications: 578,853.02 rubles versus 15,336.78– 26,994.52 rubles). Cost–effectiveness analysis favored atezolizumab for which cost–effectiveness ratio was 53,230.69 rubles for 1 month of patient’s life. Atezolizumab had better value than standard 1st line GC chemotherapy by 10,671.80 rubles, as well as immunotherapy courses using pembrolizumab by 9,697.57 rubles and GC + avelumab by 10,824.66 rubles. The highest costs were observed for GemCarbo chemotherapy course: it is 18,522.82 rubles more expensive than atezolizumab course. Sensitivity analysis performed for the cost–effectiveness ratio showed stability of the developed model in regards to increased cost of atezolizumab course up to +18 % and decrease in overall survival with this course up to –15 %.Conclusion. Atezolizumab is a clinically effective and economically justified option for treatment of adults with locally advanced or metastatic urothelial carcinoma and PD-L1 expression ≥5 % in the healthcare system of the Russian Federation.

Model robustness for feedback stabilization of open quantum systems

This paper builds upon and expands the findings presented in Liang et al. (2021). Our focus lies in scenarios where initial states and model parameters remain unknown. In contrast to prior research, we consider multiple measurement channels, addressing the challenge of global exponential stabilization towards a chosen target subspace. Under quantum non-demolition conditions, our analysis establishes essential tools for achieving feedback stabilization. Furthermore, we explore the utilization of a simplified filter as a viable alternative to full state estimator. While the effectiveness of this simplified filter was numerically examined in Cardona et al. (2020), our paper offers a comprehensive proof of its applicability in feedback stabilization. Furthermore, our analysis may serve as an inspiration for demonstrating the validation of the robust feedback stabilization through generic approximate filters.

Adaptive rock-paper-scissors game enhances eco-evolutionary performance at cost of dynamic stability

The rock-paper-scissors (RPS) game is a classic model for exploring the performance of how multiple strategies interact and evolve over time. The classic RPS game assumes a fixed benefit and cost for each strategy against another one when two players meet, while its evolutionary game considers the frequency dynamics of the three strategies with each's fitness influenced by its net payoff. This may not reflect the complexity of real-world scenarios as strategies can co-adapt with each other. We introduce an adaptive RPS game that captures the dynamics of strategy densities with trait-mediated payoffs, and the adaptive dynamics of coevolving traits via incremental mutations leading to adaptively evolving payoffs. Results show that the adaptive RPS game approaches a steady state of strategy density, if any, faster than the evolutionary RPS game. The stable coexistence of all strategies in the evolutionary game can be easily destabilized in the adaptive game. Strategies that are allowed to adaptively change their traits also performed better and achieved greater strategy densities than those fixed strategies in the adaptive game. The coevolving strategies in the adaptive RPS game exhibit complex and diverse attractors in the trait space, sensitive to both initial conditions and model parameters, but exhibiting positive payoffs with greater benefits than costs. These findings highlight how adaptive games enhance strategy performance by sacrificing eco-evolutionary dynamic stability.

Application of model updating to a large‐scale hybrid simulation

AbstractModel updating can enhance hybrid simulation by utilizing the experimental data from the physically tested substructure to update the parameters of like‐components in the numerical substructure throughout the test, improving the overall accuracy and reducing the extent of the experimental setup. Identifying and updating parameters can be challenging, especially when coupling between degrees of freedom (DOF) must be considered or the specimen experiences loading scenarios which result in newly observed behavior. To explore the performance of model updating under these challenging conditions, a large‐scale hybrid simulation was conducted using a model of a major toll bridge with seismic isolation lead rubber bearings (LRB). One LRB is physically tested considering axial, shear, and rotational loading, while the remainder of the bearings are simulated and updated with a phenomenological model within the numerical substructure. A weighted adaptive constrained unscented Kalman filter is applied as the online model updating algorithm. The study explored the effect of learning over different loading patterns, the selection of initial model parameters, and the selection of the physically tested substructure. The improvement of numerical model hysteresis performance accuracy of the force prediction demonstrates the benefits of model updating in large‐scale hybrid simulation.

Node fault diagnosis algorithm for wireless sensor networks based on BN and WSN

Wireless sensor networks, as an emerging information exchange technology, have been widely applied in many fields. However, nodes tend to become damaged in harsh and complex environmental conditions. In order to effectively diagnose node faults, a Bayesian model-based node fault diagnosis model was proposed. Firstly, a comprehensive analysis was conducted into the operative principles of wireless sensor systems, whereby fault-related features were then extrapolated. A Bayesian diagnostic model was constructed using the maximum likelihood method with sufficient sample features, and a joint tree model was introduced for node diagnosis. Due to the insufficient accuracy of Bayesian models in processing small sample data, a constrained maximum entropy method was proposed as the prediction module of the model. The use of small sample data to obtain the initial model parameters leads to improved performance and accuracy of the model. During parameter learning tests, the limited maximum entropy model outperformed the other two learning models on a smaller dataset of 35 with a distance value of 2.65. In node fault diagnosis, the diagnostic time of the three models was compared, and the average diagnostic time of the proposed diagnostic model was 41.2 seconds. In the node diagnosis accuracy test, the proposed model has the highest node fault diagnosis accuracy, with an average diagnosis accuracy of 0.946, which is superior to the other two models. In summary, the node fault diagnosis model based on Bayesian model proposed in this study has important research significance and practical application value in wireless sensor networks. By improving the reliability and maintenance efficiency of the network, this model provides strong support for the development and application of wireless sensor networks.

FuXi: a cascade machine learning forecasting system for 15-day global weather forecast

Over the past few years, the rapid development of machine learning (ML) models for weather forecasting has led to state-of-the-art ML models that have superior performance compared to the European Centre for Medium-Range Weather Forecasts (ECMWF)’s high-resolution forecast (HRES), which is widely considered as the world’s best physics-based weather forecasting system. Specifically, ML models have outperformed HRES in 10-day forecasts with a spatial resolution of 0.25∘. However, the challenge remains in mitigating the accumulation of forecast errors for longer effective forecasts, such as achieving comparable performance to the ECMWF ensemble in 15-day forecasts. Despite various efforts to reduce accumulation errors, such as implementing autoregressive multi-time step loss, relying on a single model has been found to be insufficient for achieving optimal performance in both short and long lead times. Therefore, we present FuXi, a cascaded ML weather forecasting system that provides 15-day global forecasts at a temporal resolution of 6 hours and a spatial resolution of 0.25∘. FuXi is developed using 39 years of the ECMWF ERA5 reanalysis dataset. The performance evaluation demonstrates that FuXi has forecast performance comparable to ECMWF ensemble mean (EM) in 15-day forecasts. FuXi surpasses the skillful forecast lead time achieved by ECMWF HRES by extending the lead time for Z500 from 9.25 to 10.5 days and for T2M from 10 to 14.5 days. Moreover, the FuXi ensemble is created by perturbing initial conditions and model parameters, enabling it to provide forecast uncertainty and demonstrating promising results when compared to the ECMWF ensemble.

Experimental study and numerical modeling of the thermo-hydro-mechanical process in unsaturated soil under freeze-thaw cycles

Hydropower resources have been developed to the cold regions in the west of China recently, wherein the dam will be subjected to high frequency of repeated freeze-thaw action during the construction process in winter, and thus will cause safety and stability problems of these projects. According to the actual construction situations of Lianghekou hydropower station located in Sichuan Province, one-side freezing and two-side thawing tests of unsaturated core-wall soil samples in a closed system are carried out, and the evolution of temperature, water and displacement as well as their interaction are tested and analyzed. The results show that with the increasing of freeze-thaw cycles, water migration results in a decrease of water content in the unfrozen zone and an increase in the frozen zone, and the water concentration area is formed in the soil column. Simultaneously, a deeper freezing depth accompanied with a shorter thawing duration after 10 freeze-thaw cycles is observed through the temperature field results, whereas a steady fluctuation of the deformation field is showed after the 7th freeze-thaw cycle. Moreover, based on the test results and exiting researches, a thermo-hydro-mechanical coupling model for unsaturated soils subjected to cyclic freeze-thaw action is proposed, in which both the permeability and compressibility are inextricably related to the void ratio, and vary in different calculated areas. Additionally, the effects of number of cycles on initial conditions and model parameters are also considered. Comparisons with the experimental data demonstrate the validity and adaptability of the proposed model in describing the freeze-thaw process of unsaturated soil. Through the combination of experiments and numerical simulation, the freeze-thaw process can be more clearly understood, and related engineering problems can be predicted with the help of the mathematical model in this paper.

Model-free adaptive control of average coolant temperature in a novel marine small lead–bismuth-cooled reactor

Marine lead–bismuth-cooled nuclear reactors are sensitive to the marine environment, facing motion-induced changes like heeling and rolling. These changes induce heavy oscillatory processes, and general controllers cannot totally meet the power output requirements from reactors. To address these challenges, this paper proposes a model-free adaptive control (MFAC) method. Initially, an initial model is established based on mechanism analysis, incorporating neutron kinetics, fuel thermal dynamics, and core thermal dynamics of the reactor while considering a dynamic marine environment. Subsequently, a model-free adaptive controller with online identification and adaptive capability is designed, employing a projection algorithm to continuously identify initial model parameters to match the harsh marine condition and correct the control law based on the input and output data. Finally, through simulation experiments, the effectiveness of the MFAC controller is demonstrated, showcasing superior adaptive capabilities and robustness, enabling rapid and precise tracking of the desired setpoint for average coolant temperature, even under fluctuating marine conditions.

2D magnetotelluric inversion based on ResNet

In this study, a deep learning algorithm was applied to two-dimensional magnetotelluric (MT) data inversion. Compared with the traditional linear iterative inversion methods, the MT inversion method based on convolutional neural networks (CNN) does not rely on the selection of the initial model parameters and does not fall into the local optima. Although the CNN inversion models can provide a clear electrical interface division, their inversion results may remain prone to abrupt electrical interfaces as opposed to the actual underground electrical situation. To solve this issue, a neural network with a residual network architecture (ResNet-50) was constructed in this study. With the apparent resistivity and phase pseudo-section data as the inputs and with the resistivity parameters of the geoelectric model as the training labels, the modified ResNet-50 model was trained end-to-end for producing samples according to the corresponding production strategy of the study area. Through experiments, the training of the ResNet-50 with the dice loss function effectively solved the issue of over-segmentation of the electrical interface by the cross-entropy function, avoided its abrupt inversion, and overcame the computational inefficiency of the traditional iterative methods. The proposed algorithm was validated against MT data measured from a geothermal field prospect in Huanggang, Hubei Province, which showed that the deep learning method has opened up broad prospects in the field of MT data inversion.

Quantification of forest carbon flux and stock uncertainties under climate change and their use in regionally explicit decision making: Case study in Finland

Uncertainties are essential, yet often neglected, information for evaluating the reliability in forest carbon balance projections used in national and regional policy planning. We analysed uncertainties in the forest net biome exchange (NBE) and carbon stocks under multiple management and climate scenarios with a process-based ecosystem model. Sampled forest initial state values, model parameters, harvest levels and global climate models (GCMs) served as inputs in Monte Carlo simulations, which covered forests of the 18 regions of mainland Finland over the period 2015–2050. Under individual scenarios, the results revealed time- and region-dependent variability in the magnitude of uncertainty and mean values of the NBE projections. The main sources of uncertainty varied with time, by region and by the amount of harvested wood. Combinations of uncertainties in the representative concentration pathways scenarios, GCMs, forest initial values and model parameters were the main sources of uncertainty at the beginning, while the harvest scenarios dominated by the end of the simulation period, combined with GCMs and climate scenarios especially in the north. Our regionally explicit uncertainty analysis was found a useful approach to reveal the variability in the regional potentials to reach a policy related, future target level of NBE, which is important information when planning realistic and regionally fair national policy actions.

2D and 3D modelling of Molucca Sea double subduction zone based on primary waves travel time seismic tomography inversion

Molucca Sea is one of the areas with a high level of seismicity in Indonesia due to its location between three plates. The activities of them forming a quite unique subduction zone. The Eurasian Plate (Sangihe Microplate) move to the East and The Philippine Sea Plate (Halmahera Microplate) continues to Westward caused The Molucca Sea Microplate to be pressed, and subducts in two opposite directions modelat the same time forming an inverted “U”. This research used 3663 hypocenters and 9 BMKG stations in the last 5 years (2016 - 2021) with a magnitude of 3 - 8 Mw and a depth of 3 - 630 kilometers. Seismic inversion of travel time tomography is used to produce 2D models of the subduction zone around The Molucca Sea. The distribution of high ΔVp values is ranging from 3.5% to 4% is associated with a solid medium, like The Sangihe Plate, Halmahera Plate, and The Molucca Sea Plate subduction slab. While the low ΔVp is ranging from 1.8% to 2% is associated with the destruction zone and the presence of thermal fluids such as magma or partial melting (the presence of volcanoes). The 3D model is based on the Gaussian Process Regression principle using magnitude and depth data as initial model parameters. Plotting of the thrust faults at the front of the two arcs, namely The West Sangihe Thrust Fault and The East Halmahera Thrust Fault. It can be seen that the subduction under The Sangihe Microplate is deeper (627.2 kilometers) has the average subduction angle of 45.3° compared to the subduction under The Halmahera Microplate (280.0 kilometers) has the average subduction angle of 35.8°

Design of Electric Supercharger Compressor and Its Performance Optimization

The performance of the centrifugal compressor, which is the main component of the electric supercharger, significantly impacts the engine’s dynamics, economy, emissions, and responsiveness. The purpose of this paper is to enhance the aerodynamic performance of the centrifugal compressor of the electric supercharger for the two-stroke engine by optimizing the design of its impeller and diffuser parameters. The paper employs the numerical simulation method and applies the Spalart–Allmaras turbulence model to solve the RANS equations to analyze the impact of impeller-related parameters on the centrifugal compressor’s performance. Subsequently, the paper optimizes the initial model parameters based on the simulation results and confirms its performance through an experiment. The findings indicate that enhancing the isentropic efficiency and pressure ratio of the compressor can be achieved by increasing the number of blades on the impeller, selecting an appropriate blade backward angle, and increasing the relative outlet width. After optimization, the compressor’s efficiency can achieve 0.842, the pressure ratio can reach 1.49 with a working margin of 22%, and the efficiency is enhanced by 1.4%, while the pressure ratio is increased by 1.8% compared to the pre-optimization state. Moreover, the optimized model is experimentally validated to meet the design requirements.

An Integrated Simulation of Multiple-Pass U-10Mo Alloy Hot Rolling and Static Recrystallization

AbstractTo achieve a desired microstructure and minimize the thickness variation in rolled foils, researchers must understand the effects of foil fabrication process variables on microstructure evolution. We developed an integrated simulation of deformation and recrystallization that employs the finite element method (FEM) and the kinetic Monte Carlo (KMC) Potts model, respectively, to investigate microstructure evolution during multiple-pass hot rolling and heat treatment in polycrystalline U-10Mo fuel. Scanning electron microscopy and electron backscatter diffraction images of microstructures were directly used as input in FEM calculation of deformation, and the calculated strains were used to determine the driving force of nucleation and growth of recrystallized grains in the Potts model. Grain structures predicted by the Potts model were used to update the grain structure and material properties for FEM. Simulation alternated between FEM and the Potts model to simulate grain structure evolution during multiple rolling and heat treatments. The initial model parameters were determined by benchmarking the recrystallization kinetics against experimental data. Then, the model was applied to predict the grain structure evolution. Results showed that our model can capture the coupling between deformation and recrystallization and can quantitatively reproduce the observed U-10Mo recrystallization and grain growth kinetics. The simulation results demonstrated that the developed model can predict U-10Mo grain structures as a function of initial microstructure and foil fabrication parameters.

Decoupling Control of Six-Pole Hybrid Magnetic Bearing Based on LM Neural Network Inverse System Optimized by Improved Differential Evolution Algorithm

In order to solve the problems of strong coupling and nonlinearity of the hybrid magnetic bearings for flywheel energy storage, a novel decoupling control combining Levenberg Marquardt neural network inverse system (LMNNIS) with improved differential evolution (IDE) algorithm is proposed for the six-pole radial-axial hybrid magnetic bearing (HMB). Firstly, the structure and principle of the HMB are introduced, and the mathematical model of its radial-axial suspension forces is deduced. The IDE algorithm is used to optimize the initial model parameters of LMNNIS, and the displacement and current data are collected and normalized to input into LMNNIS for training, so as to realize the decoupling control between the suspension forces of the HMB. Then, the simulation system is constructed, and the simulation experiments of the rotor floating and anti-interference are carried out, which show the convergence speed and anti-interference performance of IDE algorithm optimized LMNNIS are better than the traditional neural network inverse system. Finally, the verification experiments are carried out on the basis of the experimental platform, which prove the feasibility and reliability of the proposed method.

Evolutionary games and spatial periodicity

Spatial interactions are considered an important factor influencing a variety of evolutionary processes that take place in structured populations. It still remains an open problem to fully understand evolutionary game dynamics on networks except for certain limiting scenarios such as weak selection. Here we study the evolutionary dynamics of spatial games under strong selection where strategy evolution of individuals becomes deterministic in a fashion of winners taking all. We show that the long term behavior of the evolutionary process eventually converges to a particular basin of attraction, which is either a periodic cycle or a single fixed state depending on specific initial conditions and model parameters. In particular, we find that symmetric starting configurations can induce an exceedingly long transient phase encompassing a large number of aesthetic spatial patterns including the prominent kaleidoscopic cooperation. Our finding holds for any population structure and a broad class of finite games beyond the Prisoner’s Dilemma. Our work offers insights into understanding evolutionary dynamics of spatially extended systems ubiquitous in biology and ecology.

Meta-Reinforcement Learning-Based Transferable Scheduling Strategy for Energy Management

In Home Energy Management System (HEMS), the scheduling of energy storage equipment and shiftable loads has been widely studied to reduce home energy costs. However, existing data-driven methods can hardly ensure the transferability amongst different tasks, such as customers with diverse preferences, appliances, and fluctuations of renewable energy in different seasons. This paper designs a transferable scheduling strategy for HEMS with different tasks utilizing a Meta-Reinforcement Learning (Meta-RL) framework, which can alleviate data dependence and massive training time for other data-driven methods. Specifically, a more practical and complete demand response scenario of HEMS is considered in the proposed Meta-RL framework, where customers with distinct electricity preferences, as well as fluctuating renewable energy in different seasons are taken into consideration. An inner level and an outer level are integrated in the proposed Meta-RL-based transferable scheduling strategy, where the inner and the outer level ensure the learning speed and appropriate initial model parameters, respectively. Moreover, Long Short-Term Memory (LSTM) is presented to extract the features from historical actions and rewards, which can overcome the challenges brought by the uncertainties of renewable energy and the customers’ loads, and enhance the robustness of scheduling strategies. A set of experiments conducted on practical data of Australia’s electricity network verify the performance of the transferable scheduling strategy.

Stability and asymptotic analysis for instationary gas transport via relative energy estimates

We consider the transport of gas in long pipes and pipeline networks for which the dynamics are dominated by friction at the pipe walls. The governing equations can be formulated as an abstract dissipative Hamiltonian system which allows us to derive perturbation bounds via relative energy estimates using a problem adapted nonlinear analysis. As particular consequences of these results, we are able to prove stability estimates with respect to initial conditions and model parameters and we conduct a quantitative asymptotic analysis in the high friction limit. Our results are established first for the flow in a single pipe and we then extend our analysis to pipe networks in the spirit of energy-based port-Hamiltonian modelling.

Fitting and validation of an agent-based model for COVID-19 case forecasting in workplaces and universities.

COVID-19 forecasting models have been critical in guiding decision-making on surveillance testing, social distancing, and vaccination requirements. Beyond influencing public health policies, an accurate COVID-19 forecasting model can impact community spread by enabling employers and university leaders to adapt worksite policies and practices to contain or mitigate outbreaks. While many such models have been developed for COVID-19 forecasting at the national, state, county, or city level, only a few models have been developed for workplaces and universities. Furthermore, COVID-19 forecasting models have rarely been validated against real COVID-19 case data. Here we present the systematic parameter fitting and validation of an agent-based compartment model for the forecasting of daily COVID-19 cases in single-site workplaces and universities with real-world data. Our approaches include manual fitting, where initial model parameters are chosen based on historical data, and automated fitting, where parameters are chosen based on candidate case trajectory simulations that result in best fit to prevalence estimation data. We use a 14-day fitting window and validate our approaches on 7- and 14-day testing windows with real COVID-19 case data from one employer. Our manual and automated fitting approaches accurately predicted COVID-19 case trends and outperformed the baseline model (no parameter fitting) across multiple scenarios, including a rising case trajectory (RMSLE values: 2.627 for baseline, 0.562 for manual fitting, 0.399 for automated fitting) and a decreasing case trajectory (RMSLE values: 1.155 for baseline, 0.537 for manual fitting, 0.778 for automated fitting). Our COVID-19 case forecasting model allows decision-makers at workplaces and universities to proactively respond to case trend forecasts, mitigate outbreaks, and promote safety.

Design of Predefined Time Convergent Sliding Mode Control for a Nonlinear PMLM Position System

The significant role for a contemporary control algorithm in the position control of a permanent magnet linear motor (PMLM) system is highlighted by the rigorous standards for accuracy in many modern industrial and robotics applications. A robust predefined time convergent sliding mode controller (PreDSMC) is designed for the high precision position tracking of a permanent magnet linear motor (PMLM) system with external disturbance, and its convergence time is independent of the system’s initial value and model parameters. We verified theoretically that the performance function conditions are satisfied, the motor speed is steady and constrained, and the motor position tracking error converges to zero within the prescribed time. First, we designed a sliding mode (SM) surface with predetermined time convergence, which mathematically demonstrates that the tracking error converges to zero within the predefined time and shows that the position tracking accuracy is higher. Secondly, we developed a PreDSMC law that is independent of initial state and based on the predefined time convergence Lyapunov stability criterion. Finally, to prove the accuracy and higher precision of the proposed PreDSMC, comparative numerical simulations are performed for PMLM with compound disturbances. Simulation findings show that the suggested robust predefined control method considerably reduces the impacts of friction and external disturbances; consequently, it may increase the control performance when compared to the typical proportional integral derivative (PID) controller, the nonsingular fast terminal SMC, and the linear SMC.