Multi Network Model Research Articles

Optimized Artificial Neural Network Model and Compensator in Model Predictive Control for Anomaly Mitigation

Abstract This paper presents a new artificial neural network (ANN)-based system model that concatenates an optimized artificial neural network (OANN) and a neural network compensator (NNC) in series to capture temporally varying system dynamics caused by slow-paced degradation/anomaly. The OANN comprises a complex, fully connected multilayer perceptron, trained offline using nominal, anomaly free data, and remains unchanged during online operation. Its hyperparameters are selected using genetic algorithm-based meta-optimization. The compact NNC is updated continuously online using collected sensor data to capture the variations in system dynamics, rectify the OANN prediction, and eventually minimize the discrepancy between the OANN-predicted and actual response. The combined OANN–NNC model then reconfigures the model predictive control (MPC) online to alleviate disturbances. Through numerical simulation using an unmanned quadrotor as an example, the proposed model demonstrates salient capabilities to mitigate anomalies introduced to the system while maintaining control performance. We compare the OANN–NNC with other online modeling techniques (adaptive ANN and multinetwork model), showing it outperforms them in reference tracking of altitude control by at least 0.5 m and yaw control by 1 deg. Moreover, its robustness is confirmed by the MPC consistency regardless of anomaly presence, eliminating the need for additional model management during online operation.

Numerical Simulation with Multi-Network Model and Discrete Element Method for Dynamic Structure Change and Cell Performance of All-Solid-State Batteries

Recently, all-solid-state-lithium-ion batteries have attracted attention as next-generation batteries serving as driving sources for electric vehicles (EVs) and hybrid electric vehicles (HEVs). However, it is required to have more power density and more energy density. In order to develop the performance of batteries, High-capacity negative active material (AM) such as Si have been developed, but their use is not easy due to severe expansion during charging and discharging [1]. Although materials with reduced expansion and contraction have been developed [2], it is unclear how much expansion of the AM is allowed in the first place and how much it affects the porous electrode structure. Also, since the performance of batteries depend not only on the material characteristics but also on its electrode structure, it is important to design an optimum electrode structure. Therefore, chasing the state in electrode layer using the numerical computation is a critical measure for the comprehension of phenomenon in the cell. However, in the relatively micro-scale system such as the electrode layer, a slight difference in structure affects the battery performance. However, usual simulations demand the reactive interface area and the tortuosity factor which critically affect the cell performance by reasonableness or approximation, as it might overlook the phenomenon from minute structure of electrode layer. Therefore, our laboratory has devised a multi-network model as a method that directly reflects the transport characteristics in the particle-packed structure [3]. In this study, we apply it to an all-solid-state battery and examine the effect of various structural factors on the expansion and contraction of the AM.This simulation first located AM particles randomly in three-dimensional space. The AM particles are all spherical and overlap is not allowed. Next, assuming the space except active material particles as solid electrolyte (SE) phase and located imaginary spheres with greatest diameter at positions fixed by 4 AM particles or more, which memorize electrolyte information such as ion electric potential at the position. After constructing electrode structure, the AM network and SE network were built as shown in Figure 1 The electronic conduction was calculated in AM network, as the ionic conduction calculated with SE network, while the electrode reaction occurring at interface between AM and SE. Finally, by applying this model to a galvanostatic discharge simulation based on the porous electrode theory [4], the state inside electrode layer was obtained with iterative computation by taking the mass balance and electron balance of each AM particles and SE particles. The structural change in the electrode layer due to the expansion and contraction of the AM was based on the previous research of this laboratory [5], and the discrete element method (DEM) was incorporated into the stress analysis. DEM is a method of calculating a force acting in a normal direction and a tangential direction between particles, assuming a spring representing elasticity and a dashpot representing viscous damping between particles in contact with each other. Using this, the interparticle stress in the electrode layer was calculated, and a dynamic change accompanying expansion and contraction of the AM was considered.With this model, it is possible to examine the effects of various structural factors and the optimal structural design in consideration of the expansion and contraction of the AM, and to provide guidelines for higher power density.AcknowledgmentThis research was supported by Grants-in-Aid for Scientific Research on Innovative Areas, “Science on Interfacial Ion Dynamics for Solid State Ionics Devices” MEXT, Japan FY2019-2023.Reference[1] X. Su et al., Adv. Energy Mater., 4, 1300882 (2014).[2] X. Zhang et al., Electrochemistry, 84 (6), 420 (2016).[3] K. Lin et al., ECS Transaction, 80（10）, 251–258 (2017).[4] G. M. Goldin et al, Electrochimica Acta, 64 , 118-129 (2012).[5] K. Ishikawa et al, the 56th Battery Symposium in Japan, 3E16 (2015). Figure 1

Deep learning of free boundary and Stefan problems

Free boundary problems appear naturally in numerous areas of mathematics, science and engineering. These problems present a great computational challenge because they necessitate numerical methods that can yield an accurate approximation of free boundaries and complex dynamic interfaces. In this work, we propose a multi-network model based on physics-informed neural networks to tackle a general class of forward and inverse free boundary problems called Stefan problems. Specifically, we approximate the unknown solution as well as any moving boundaries by two deep neural networks. Besides, we formulate a new type of inverse Stefan problems that aim to reconstruct the solution and free boundaries directly from sparse and noisy measurements. We demonstrate the effectiveness of our approach in a series of benchmarks spanning different types of Stefan problems, and illustrate how the proposed framework can accurately recover solutions of partial differential equations with moving boundaries and dynamic interfaces. All code and data accompanying this manuscript are publicly available at https://github.com/PredictiveIntelligenceLab/DeepStefan.

A multinetwork inelastic model for the hysteretic response during cyclic loading of pig and rat skin

In this paper, we develop a 2-dimensional membrane model for the inelastic response of nominally leathery materials applicable to many tissues, particularly skin. These materials, while nominally elastic, show substantial and complex hysteretic response under cyclic loading (usually referred to as the Mullins-type softening). Furthermore, they exhibit both rate dependent behavior and permanent residual strain upon unloading. The proposed model is based on the notion of a multinetwork model composed of a background material and possibly many families of fibers with a superposed response. The rate dependent inelasticity and permanent strain are modeled through fiber slippage. We show the efficacy of the model by comparing with experiments in rat and pig skin under various cyclic loading conditions. The parameters of the modal are reasonably easy to measure directly from experiments. The key ingredients of the fiber part of the response are (1) a simple two-slope elastic response that accounts for the rapid stiffening response of the fibers, with smooth transition between the slack and stiff response representing the fiber recruitment (2) a non affine deformation of the fibers to account for the fiber slippage and recovery (3) a rapidly hardening response to account for the memory of the skin regarding its previous maximum stretched state. The entire framework is written in rate form and is easily implemented with an elementary integration scheme. The model is capable of simulating not only the overall behavior of the skin but internal minor hysteresis as well.

Modeling temporal and spatial changes during hydrolytic degradation and erosion in biodegradable polymers

This paper presents a computational model in describing the spatial and temporal changes in the chemical and physical characteristics of biodegradable polymer solids during hydrolytic degradation and erosion processes. A set of coupled governing differential equations are formulated to account for the diffusion of water molecules into the polymers, scission kinetics from a hydrolytic process, formation of monomers, and diffusion of the soluble monomers out of the polymer body. A multi-network model is adopted for capturing the hydrolytic scission, in which a long polymeric chain is being converted to monomers and byproducts. As the monomers leave the polymer body, surface and bulk erosion mechanisms are described and changes in the shape and size of the polymer body also take place. The governing differential equations are solved using a finite difference approach. It is noted that hydrolytic degradation induces temporal and spatial changes in the chemical and physical properties of polymers; and thus, the polymers that are initially homogeneous with regards to the above properties become heterogeneous due to the spatial changes in their macromolecular networks during biodegradation. The responses obtained from the model are compared with available experimental data of PLGA biodegradable polymer.

Towards Robust and Accurate Detection of Abnormalities in Musculoskeletal Radiographs with a Multi-Network Model.

This study proposes a novel multi-network architecture consisting of a multi-scale convolution neural network (MSCNN) with fully connected graph convolution network (GCN), named MSCNN-GCN, for the detection of musculoskeletal abnormalities via musculoskeletal radiographs. To obtain both detailed and contextual information for a better description of the characteristics of the radiographs, the designed MSCNN contains three subnetwork sequences (three different scales). It maintains high resolution in each sub-network, while fusing features with different resolutions. A GCN structure was employed to demonstrate global structure information of the images. Furthermore, both the outputs of MSCNN and GCN were fused through the concat of the two feature vectors from them, thus making the novel framework more discriminative. The effectiveness of this model was verified by comparing the performance of radiologists and three popular CNN models (DenseNet169, CapsNet, and MSCNN) with three evaluation metrics (Accuracy, F1 score, and Kappa score) using the MURA dataset (a large dataset of bone X-rays). Experimental results showed that the proposed framework not only reached the highest accuracy, but also demonstrated top scores on both F1 metric and kappa metric. This indicates that the proposed model achieves high accuracy and strong robustness in musculoskeletal radiographs, which presents strong potential for a feasible scheme with intelligent medical cases.

Network Functional Architecture and Aberrant Functional Connectivity in Post-Traumatic Stress Disorder: A Clinical Application of Network Convergence.

Posttraumatic stress disorder (PTSD) is associated with disrupted functional connectivity in multiple neural networks. Multinetwork models of PTSD hypothesize that aberrant regional connectivity emerges from broad network-level disruptions. However, few studies have tested how characteristics of network-level organization influence regional functional connectivity in PTSD. This gap in knowledge impacts both our understanding of the pathophysiology of the disorder and the development of network-targeted PTSD treatments. We acquired resting-state imaging from a naturalistic sample of patients with PTSD (n = 42) and healthy controls (n = 42). Group differences in functional connectivity were identified using region of interest analyses and estimations of within- and between neural network activity; PTSD patients demonstrated reduced amygdala-orbitofrontal connectivity and increased default mode network (DMN) connectivity compared with controls. We then used convergence-a novel measure representing the capacity for functional integration-to test whether differences in functional architecture underlie connectivity signatures of PTSD. This approach found that reduced frontoparietal network (FPN) convergence was associated with reduced amygdala-orbitofrontal connectivity. Furthermore, in controls only, increased DMN convergence was associated with reduced DMN-to-salience network connectivity, and increased FPN convergence was associated with reduced FPN-to-ventral attention network connectivity. These results suggest that FPN functional architecture may underlie insufficiencies in top-down control in PTSD, with results broadly supporting the notion that networks' functional architecture influences the breakdown of normative functional relationships in PTSD. This work also indicates the potential of convergence to be applied to clinical populations in future research studies.

Prediction of construction cost index based on multi variable grey neural network model

The construction cost index can reflect the construction market price changes on the project cost. Due to unsuitable way of management and the consideration of data security, there is the possibility of researchable sample data missing, or the possibility of not enough data. This way many classical prediction models cannot function well in practice. The study compared numerous models. According to the characteristics of poor sample data and numerous factors for construction engineering, the grey model and BP neural network were selected to establish the prediction model. This study built material and machinery shift as independent variables, the construction cost as the dependent variable to establish the grey neural network model. In empirical analysis, data processing, training and simulation were used by MATLAB, the results concluded: compared with other conventional models, the combined forecasting model for poor data is more effective.

Prediction of construction cost index based on multi variable grey neural network model

Prediction of construction cost index based on multi variable grey neural network model

Disease gene prioritization by integrating tissue-specific molecular networks using a robust multi-network model.

BackgroundAccurately prioritizing candidate disease genes is an important and challenging problem. Various network-based methods have been developed to predict potential disease genes by utilizing the disease similarity network and molecular networks such as protein interaction or gene co-expression networks. Although successful, a common limitation of the existing methods is that they assume all diseases share the same molecular network and a single generic molecular network is used to predict candidate genes for all diseases. However, different diseases tend to manifest in different tissues, and the molecular networks in different tissues are usually different. An ideal method should be able to incorporate tissue-specific molecular networks for different diseases.ResultsIn this paper, we develop a robust and flexible method to integrate tissue-specific molecular networks for disease gene prioritization. Our method allows each disease to have its own tissue-specific network(s). We formulate the problem of candidate gene prioritization as an optimization problem based on network propagation. When there are multiple tissue-specific networks available for a disease, our method can automatically infer the relative importance of each tissue-specific network. Thus it is robust to the noisy and incomplete network data. To solve the optimization problem, we develop fast algorithms which have linear time complexities in the number of nodes in the molecular networks. We also provide rigorous theoretical foundations for our algorithms in terms of their optimality and convergence properties. Extensive experimental results show that our method can significantly improve the accuracy of candidate gene prioritization compared with the state-of-the-art methods.ConclusionsIn our experiments, we compare our methods with 7 popular network-based disease gene prioritization algorithms on diseases from Online Mendelian Inheritance in Man (OMIM) database. The experimental results demonstrate that our methods recover true associations more accurately than other methods in terms of AUC values, and the performance differences are significant (with paired t-test p-values less than 0.05). This validates the importance to integrate tissue-specific molecular networks for studying disease gene prioritization and show the superiority of our network models and ranking algorithms toward this purpose. The source code and datasets are available at http://nijingchao.github.io/CRstar/.Electronic supplementary materialThe online version of this article (doi:10.1186/s12859-016-1317-x) contains supplementary material, which is available to authorized users.

A nonlinear viscoelastic constitutive model for polymeric solids based on multiple natural configuration theory

In this paper we carry out an experimental study and develop and corroborate a model for describing the response of viscoelastic solids undergoing microstructural changes. We adopt the theory of multiple natural configurations (multi network model) and a single integral constitutive representation to incorporate the effect of microstructural changes on the time-dependent macroscopic deformations of polymers. In formulating the model, we assume that the polymers have two natural configurations, which are different stress-free configurations associated with the original and new networks. The new network is formed when the polymers are subjected to relatively large deformation, leading to the breaking of the junctions in the original network, and is irrecoverable. While the responses of the two networks from their respective natural configuration is elastic, the process of microstructural changes of the original network is dissipative and hence the response of the two networks together is not elastic. Experimental data on polyoxymethylene (POM) under uniaxial tensile quasi-static ramp and creep-recovery histories are used to examine the applicability and efficacy of the model. During testing, deformations in the loading and lateral directions are recorded, which allow for determining various stress-strain measures. We also discuss different mathematical functions that can be considered in describing responses of the polymers based on the recorded stress-strain measures. The model is shown to be capable of describing three dimensional responses of POM polymers under various loading histories.

Self-recovering caddisfly silk: energy dissipating, Ca(2+)-dependent, double dynamic network fibers.

Single fibers of the sticky underwater larval silk of the casemaker caddisfly (H. occidentalis) are viscoelastic, display large strain cycle hysteresis, and self-recover 99% of their initial stiffness and strength within 120 min. Mechanical response to cyclical strains suggested viscoelasticity is due to two independent, self-recovering Ca(2+)-crosslinked networks. The networks display distinct pH dependence. The first network is attributed to Ca(2+)-stabilized phosphoserine motifs in H-fibroin, the second to Ca(2+) complexed carboxylate groups in the N-terminus of H-fibroin and a PEVK-like protein. These assignments were corroborated by IR spectroscopy. The results are consolidated into a multi-network model in which reversible rupture of the Ca(2+)-crosslinked domains at a critical stress results in pseudo-plastic deformation. Slow refolding of the domains results in nearly full recovery of fiber length, stiffness, and strength. The fiber toughening, energy dissipation, and recovery mechanisms, are highly adaptive for the high energy aquatic environment of caddisfly larvae.

A simple model for myocardial changes in a failing heart

In a simplified setting, a multi-network model for remodeling in the left ventricle (LV) is developed that can mimic various pathologies of the heart. The model is an extension of the simple model introduced by Nardinocchi and Teresi [9], Nardinocchi et al. [10,11] that results in an algebraic relation for LV pressure–volume–contraction. We considered two networks, the original tissue and a new tissue, each of which has its own volume fraction, stress-free reference configuration, elastic properties, and contractility. This is used to explore the consequences of microstructural changes in the muscle tissue on LV function in terms of the pressure–volume loop during a single cardiac cycle. Special attention is paid to the stroke volume, which is directly related to cardiac output, and changes in LV wall stress caused by various disease states, including wall thinning (dilated cardiomyopathy), wall thickening (hypertrophic cardiomyopathy), contractility degradation, and stiffness changes (scarring). Various scenarios are considered that are of clinical relevance, and the extent and nature of remodeling that could lead to heart failure are identified.

The Research of Multi-Network Switching Model for Power Communication Network Based on Analytic Hierarchy Process

To further improve the quality of service of communication business for power distribution and utilization, and to meet the needs of communication network for smart power distribution and utilization, we propose a multi-network switching model based on AHP; First, this model analyzes the performance of LTE, WCDMA and WiFi, to determine the integrated weights of optional network by solving the relative weight of rules and index, then select a network which is based on guarantee the performance index of QoS; In the process of transmission of power distribution and utilization business, take the transmission efficiency as the optimization target, the network of which weight is larger is priority to transfer power distribution and utilization business; Simulation results show that the model can improve the quality of service of communication business for power distribution and utilization.

Genetics as a tool for the dissociation of mental operations over the course of development

In recent years it has become possible to differentiate separable aspects of attention and to characterize the anatomical structure and dynamic states of their underlying networks. When individual differences in the structure and dynamics of these networks are used as dependent measures in associations with individual genetic variation, it becomes possible to assign cellular and molecular changes that occur over the course of normal development to specific aspects of network structure and function. In this way, a more granular understanding of the physiology of neural networks can be obtained. Here we review a translational research strategy focused on how genetic variation contributes to the normal development of attentional function. We seek to use genetic information to help construct a multinode, multinetwork model that can explain, in part, individual differences in the development of attention over the course of development.

Qualitative network models and genome-wide expression data define carbon/nitrogen-responsive molecular machines in Arabidopsis

Qualitative network models and genome-wide expression data define carbon/nitrogen-responsive molecular machines in Arabidopsis and indicate that regulation by carbon/nitrogen metabolites occurs at multiple levels.

Interworking among UUCP Information Servers

Interworking of information servers on UUCP connectivity has been optimized by incorporating suitable protocols. A novel scheme has been proposed which uses a centrally located server node. Also, a multi server network model has been designed, developed and discussed from its implementation view point for performance gain. Implementation of the binary splitting tree algorithm for contention resolution has been explored under the assumption that feedback about channel status in UUCP network is made available through private (dedicated) channels.