Set Of Neural Networks Research Articles

Command Filtered Neuroadaptive Fault-Tolerant Control for Nonlinear Systems With Input Saturation and Unknown Control Direction.

This article studies the tracking control of a class of nonlinear systems with input saturation, subject to nonaffine faults and unknown control direction. A fault-tolerant command filtered control (CFC) method based on adaptive neural networks (NNs) is proposed for this kind of nonlinear system. First, the combination of CFC and error compensation overcomes the "explosion of complexity" issue and alleviates the impact of filter errors. Then, a set of radial basis function NNs is constructed to approximate the unknown nonlinear items containing the nonaffine fault function. Additionally, the issue of unknown control direction in the system is effectively resolved by using Nussbaum gain technology. It is proven that the designed controller can ensure that all signals in the closed-loop system are bounded and convergent, and the upper bound of the absolute value of system tracking error is given. Finally, three comparative simulation results are illustrated to show the effectiveness of the proposed method.

Traffic Signal Control With Adaptive Online-Learning Scheme Using Multiple-Model Neural Networks.

This article proposes a new traffic signal control algorithm to deal with unknown-traffic-system uncertainties and reduce delays in vehicle travel time. Unknown-traffic-system dynamics are approximated using a recurrent neural network (NN). To accurately identify the traffic system model, an online-learning scheme is developed to switch among a set of candidate NNs (i.e., multiple-model NNs) based on their estimation errors. Then, a bank of optimal signal-timing controllers is designed based on the online identification of the traffic system. Simulation studies have been carried out for the obtained control strategies using multiple-model NNs, and the desired results have been obtained. Moreover, compared with the widely used actuated traffic signal control schemes, it is shown that the proposed method can reduce vehicle travel delays and improve traffic system robustness.

Disturbance observer-based adaptive neural control for underactuated surface vehicle with constraint of input saturation

In this paper, a robust adaptive neural control algorithm is provided to solve the problem of path-following control for underactuated surface vehicle (USV) subject to input saturation, in the presence of model uncertainties and unknown marine environmental disturbance.In the algorithm, through the application of neural network (NN), an adaptive disturbance observer (ADO) is proposed to observe the unknown disturbance, which does not need the precise information of the upper bound of the disturbance, and the model uncertainty can be appropriated simultaneously. Specially, since the ADO and control law share the same set of NN, the number of adaptive parameters of the controller can be greatly reduced. After that, dynamic surface control (DSC) method is used to avoid repeated derivative in the process of back-stepping, which solve the “calculation explosion” problem. An auxiliary system is designed to compensate errors caused by input saturation, which solve the input saturation problem of actuators. Through Lyapunov stability analysis, it is proved that all signals of the closed-loop system are semi-globally ultimately uniformly bounded (SGUUB). Finally, some experiments are conducted to demonstrate the effectiveness of the algorithm.

Cascading and Ensemble Techniques in Deep Learning

In this study, we explore the integration of cascading and ensemble techniques in Deep Learning (DL) to improve prediction accuracy on diabetes data. The primary approach involves creating multiple Neural Networks (NNs), each predicting the outcome independently, and then feeding these initial predictions into another set of NN. Our exploration starts from an initial preliminary study and extends to various ensemble techniques including bagging, stacking, and finally cascading. The cascading ensemble involves training a second layer of models on the predictions of the first. This cascading structure, combined with ensemble voting for the final prediction, aims to exploit the strengths of multiple models while mitigating their individual weaknesses. Our results demonstrate significant improvement in prediction accuracy, providing a compelling case for the potential utility of these techniques in healthcare applications, specifically for prediction of diabetes where we achieve compelling model accuracy of 91.5% on the test set on a particular challenging dataset, where we compare thoroughly against many other methodologies.

Biomimetic Force and Impedance Adaptation Based on Broad Learning System in Stable and Unstable Tasks: Creating an Incremental and Explainable Neural Network With Functional Linkage

This article presents a novel biomimetic force and impedance adaption framework based on the broad learning system (BLS) for robot control in stable and unstable environments. Different from iterative learning control, the adaptation process is realized by a neural network (NN)-based framework, similar to a BLS, to realize a varying learning rate for feedforward force and impedance factors. The connections of NN layers and settings of the feature nodes are related to the human motor control and learning principle that is described as a relationship among feedforward force, impedance, reflex and position errors, and so on to make the NN explainable. Some comparative simulations are created and tested in five force fields to verify advantages of the proposed framework in terms of force and trajectory tracking efficiency and accuracy, robust responses to different force situations, and continuity of force application in a mixed stable and unstable environment. Finally, an experiment is conducted to verify effectiveness of the proposed method.

Hybrid-Layers Neural Network Architectures for Modeling the Self-Interference in Full-Duplex Systems

Full-duplex (FD) systems have been introduced to provide high data rates for beyond fifth-generation wireless networks through simultaneous transmission of information over the same frequency resources. However, the operation of FD systems is practically limited by self-interference (SI), and efficient SI cancelers are sought to make the FD systems realizable. Typically, polynomial-based cancelers are employed to mitigate the SI; nevertheless, they suffer from high complexity. This article proposes two novel hybrid-layers neural network (NN) architectures to cancel the SI with low complexity. The first architecture is referred to as hybrid-convolutional recurrent NN (HCRNN), whereas the second is termed as hybrid-convolutional recurrent dense NN (HCRDNN).In the HCRNN, a convolutional layer is employed to extract the input data features using a reduced network scale. Moreover, a recurrent layer is then applied to assist in learning the temporal behavior of the input signal from the localized feature map of the convolutional layer. In the HCRDNN, an additional dense layer is exploited to add another degree of freedom for adapting the NN settings in order to achieve the best compromise between the cancellation performance and computational complexity. The complexity analysis of the proposed NN architectures is provided, and the optimum settings for their training are selected. The simulation results demonstrate that the proposed HCRNN and HCRDNN-based cancelers attain the same cancellation of the polynomial and the state-of-the-art NN-based cancelers with an astounding computational complexity reduction. Furthermore, the proposed cancelers show high design flexibility for hardware implementation, depending on the system demands.

An AI-Based Methodology for the Automatic Classification of a Multiclass Ebook Collection Using Information From the Tables of Contents

Book recommendation to support professors and students in the identification of relevant sources is of significant importance for both universities and digital libraries and, hence, motivates the development of a recommendation system. This paper aims at automatically classifying a multiclass corpus that was created from ebooks from the Springer collection, which is available through the Hellenic Academic Libraries’ subscription, by utilizing an unsupervised neural network (NN) (self-organizing maps, SOM) and two deep neural network (DNN) architectures, namely, a long short-term memory (LSTM) and a convolutional neural network (CNN) combined with a LSTM(CNN+LSTM) under various configuration scenarios. The vector construction leverages information that was extracted from the table of contents (ToC) of each book using the TF-IDF weighting scheme (for the first case) and the Keras tokenizer (for the second). Extensive experiments were conducted using various configurations of preprocessing steps, NN set up and vector and vocabulary sizes to assess their impact on the classifier’s performance. Furthermore, we show that majority voting is more suitable for selecting the dominant label for a specified node. The experimental analysis showed the feasibility of developing a recommendation system for supporting professors and students in the identification of related sources based on a detailed thematic description (e.g., abstract or table of contents of a book) rather than a few keywords. In the conducted experiments, the subsystem that utilized the DNN (LSTM) performed the best, with F1-scores of 67% for the 26 categories and 80% for the 5 general categories, whereas SOM realizes F1-scores of less than 5% in both cases.

Improved decentralized finite-time formation control of underactuated USVs via a novel disturbance observer

This note investigates the decentralized finite-time formation control of underactuated unmanned surface vessels (USVs) in the presence of model uncertainty and environmental disturbance. In the algorithm, a novel adaptive finite-time disturbance observer (AFTDO) is incorporated into the proposed control strategy that enhances its robustness to the environmental disturbance. By fusion of neural network (NN) and minimal learning parameterization (MLP) techniques, the AFTDO is constructed without priori information about the ship model and the upper bound of disturbance. On the basis of undirected graph, the decentralized control law is developed with local information from neighboring USVs. Specially, since the AFTDO and the decentralized control law share the same set of NN, the whole algorithm is with merits of concise form and less adaptive parameters. Such design contributes to smaller computation load and facilitates the implementation of the algorithm in ocean engineering. The semi global finite-time uniformly boundedness (SGFTUB) of the closed-loop system is proved by the Lyapunov theory. Numerical simulations and comprehensive comparisons are conducted to demonstrate the remarkable performance and superiority of the proposed algorithm.

Application of the neural network computing technology for calculating the interval-index characteristics of a minimally redundant modular code

The article is devoted to the problem of creation of high-speed neural networks (NN) for calculation of interval-index characteristics of a minimally redundant modular code. The functional base of the proposed solution is an advanced class of neural networks of a final ring. These neural networks perform position-modular code transformations of scalable numbers using a modified reduction technology. A developed neural network has a uniform parallel structure, easy to implement and requires the time expenditures of the order (3[log2b]+ [log2k]+6tsum close to the lower theoretical estimate. Here b and k is the average bit capacity and the number of modules respectively; t sum is the duration of the two-place operation of adding integers. The refusal from a normalization of the numbers of the modular code leads to a reduction of the required set of NN of the finite ring on the (k – 1) component. At the same time, the abnormal configuration of minimally redundant modular coding requires an average k-fold increase in the interval index module (relative to the rest of the bases of the modular number system). It leads to an adequate increase in hardware expenses on this module. Besides, the transition from normalized to unregulated coding reduces the level of homogeneity of the structure of the NN for calculating intervalindex characteristics. The possibility of reducing the structural complexity of the proposed NN by using abnormal intervalindex characteristics is investigated.

Robust adaptive formation control of underactuated autonomous surface vessels based on MLP and DOB

This paper investigates the leader–follower formation problem of multiple underactuated autonomous surface vessels in the presence of model uncertainties and environmental disturbances. Specially, the formation is defined in the body-fixed coordinates of the leader vessel and velocities of the leader are unavailable to followers. A novel robust adaptive formation control scheme based on the minimal learning parameter (MLP) algorithm and the disturbance observer (DOB) is presented. To address related formation configurations and unknown velocities of the leader, adaptive programming of the virtual vessel is introduced. By the neural networks (NNs) technique, the DOB is constructed and the formation controller is developed with different MLP-based adaptive laws. Under the proposed controller, it is shown that the desired formation can be achieved only with the position and yaw angle of the leader. And formation errors are guaranteed to be semiglobal uniformly ultimately bounded. Compared with existing results, the NNs-based DOB can compensate disturbances effectively with less model information. Meanwhile, the formation controller and the DOB can share the same set of NNs with smaller computational effort, where only two parameters need to be learned online for each of them. Simulations and comparison results are provided to illustrate the effectiveness of theoretical results.

Approximate N-Player Nonzero-Sum Game Solution for an Uncertain Continuous Nonlinear System.

An approximate online equilibrium solution is developed for an N -player nonzero-sum game subject to continuous-time nonlinear unknown dynamics and an infinite horizon quadratic cost. A novel actor-critic-identifier structure is used, wherein a robust dynamic neural network is used to asymptotically identify the uncertain system with additive disturbances, and a set of critic and actor NNs are used to approximate the value functions and equilibrium policies, respectively. The weight update laws for the actor neural networks (NNs) are generated using a gradient-descent method, and the critic NNs are generated by least square regression, which are both based on the modified Bellman error that is independent of the system dynamics. A Lyapunov-based stability analysis shows that uniformly ultimately bounded tracking is achieved, and a convergence analysis demonstrates that the approximate control policies converge to a neighborhood of the optimal solutions. The actor, critic, and identifier structures are implemented in real time continuously and simultaneously. Simulations on two and three player games illustrate the performance of the developed method.

Gait Pattern Adaptation for an Active Lower-Limb Orthosis Based on Neural Networks

This work deals with neural network (NN)-based gait pattern adaptation algorithms for an active lower-limb orthosis. Stable trajectories with different walking speeds are generated during an optimization process considering the zero-moment point (ZMP) criterion and the inverse dynamic of the orthosis–patient model. Additionally, a set of NNs is used to decrease the time-consuming analytical computation of the model and ZMP. The first NN approximates the inverse dynamics including the ZMP computation, while the second NN works in the optimization procedure, giving an adapted desired trajectory according to orthosis–patient interaction. This trajectory adaptation is added directly to the trajectory generator, also reproduced by a set of NNs. With this strategy, it is possible to adapt the trajectory during the walking cycle in an on-line procedure, instead of changing the trajectory parameter after each step. The dynamic model of the actual exoskeleton, with interaction forces included, is used to generate simulation results. Also, an experimental test is performed with an active ankle–foot orthosis, where the dynamic variables of this joint are replaced in the simulator by actual values provided by the device. It is shown that the final adapted trajectory follows the patient intention of increasing the walking speed, so changing the gait pattern.

A new area-specific bio-optical algorithm for the Bay of Biscay and assessment of its potential for SeaWiFS and MODIS/Aqua data merging

Based on a feed-forward and error-back-propagated neural network (NN), a new bio-optical algorithm is developed for the Bay of Biscay. It is designed as a set of NNs individually dedicated to the retrieval of the phytoplankton chlorophyll (chl), and total suspended matter (tsm) from Sea-viewing Wide Field-of-View Sensor (SeaWiFS) and Moderate Resolution Imaging Spectroradiometer (MODIS) Aqua data. The retrieved versus in situ measured concentrations of chl and tsm correlation coefficients for chl proved to be ∼0.8 (SeaWiFS) and 0.72 (MODIS), and for tsm 0.71 (SeaWiFS) and 0.74 (MODIS). The developed NN-based bio-optical algorithms are employed to assess the compatibility of SeaWiFS and MODIS data on chl and tsm in the coastal zone of the Bay of Biscay (case 2 waters). The value of the ratio between the concentration of chl and tsm derived from the same-day SeaWiFS and MODIS data (the overflight time difference, Δt is ≤2.5 hours) has in most cases values of approximately 1, however, in specific cases it varies appreciably. These results indicate that, unlike the reportedly very successful cases of merging of SeaWiFS and MODIS data on chl in open ocean waters (case 1 waters), the merging of chl (and tsm) data from these sensors collected over case 2 waters needs to be supervised at a level of a few pixels. At the same time, when averaged over the entire coastal zone of the Bay of Biscay, the retrieved monthly mean chl and tsm concentrations from SeaWiFS and MODIS practically coincide throughout the years (2002–2004) of contemporaneous operation of these two satellite sensors. Thus, even in the case of such dynamic and optically complex case 2 waters that are inherent in the Bay of Biscay, the potentials for ocean colour data merging are very good. The merging efficiency is assessed and illustrated via documenting the spatio-temporal dynamics of bottom sediment re-suspension in the bay occurring in winter – the season of heaviest cloudiness over the bay.

Retrieving forest stand parameters from SAR backscatter data using a neural network trained by a canopy backscatter model

It was possible to retrieve the stand mean dbh (tree trunk diameter at breast height) and stand density from the Jet Propulsion Laboratory (JPL) Airborne Synthetic Aperture Radar (AIRSAR) backscatter data by using threelayered perceptron neural networks (NNs). Two sets of NNs were trained by the Santa Barbara microwave canopy backscatter model. One set of the trained NNs was used to retrieve the stand mean dbh, and the other to retrieve the stand density. Each set of the NNs consisted of seven individual NNs for all possible combinations of one, two, and three radar wavelengths. Ground and multiplewavelength AIRSAR backscatter data from two ponderosa pine forest stands near Mt. Shasta, California (U.S.A.) were used to evaluate the accuracy of the retrievals. The r.m.s. and relative errors of the retrieval for stand mean dbh were 6.1cm and 15.6 per cent for one stand (St2), and 3.1cm and 6.7 per cent for the other stand (St11). The r.m.s. and relative errors of the retrieval for stand density were 71.2 treesha-1 and 23.0 per cent for St2, and 49.7 treesha-1 and 21.3 per cent for St11.

Optimization of HV electrode systems by neural networks using a new learning method

To avoid a large number of iterations, optimization of electrode shapes has been done by artificial neural networks (NN). Two practical examples have been considered, an axisymmetric single-phase GIS bus termination and an axisymmetric transformer shield ring. The shape of the electrodes has been taken as quarter-ellipse or half-ellipse because an ellipse has more flexibility than a circle. For NN, the so-called resilient propagation algorithm, learning faster than the standard back-propagation algorithm, has been employed. The training sets as well as the test sets of NN have been prepared by charge simulation method.

Improving model accuracy using optimal linear combinations of trained neural networks

Neural network (NN) based modeling often requires trying multiple networks with different architectures and training parameters in order to achieve an acceptable model accuracy. Typically, only one of the trained networks is selected as "best" and the rest are discarded. The authors propose using optimal linear combinations (OLC's) of the corresponding outputs on a set of NN's as an alternative to using a single network. Modeling accuracy is measured by mean squared error (MSE) with respect to the distribution of random inputs. Optimality is defined by minimizing the MSE, with the resultant combination referred to as MSE-OLC. The authors formulate the MSE-OLC problem for trained NN's and derive two closed-form expressions for the optimal combination-weights. An example that illustrates significant improvement in model accuracy as a result of using MSE-OLC's of the trained networks is included.