Neural Network Approximation Research Articles

A decoupling estimation method for baseline load of photovoltaic user clusters based on multidimensional abnormal data restoration

Abstract With the widespread installation of distributed photovoltaics, their fluctuation characteristics with weather changes have posed significant challenges to baseline load estimation. In addition, due to measurement and collection anomalies, it is easy to cause continuous missing and abnormal data collection. To address the issue of low baseline estimation accuracy caused by distributed photovoltaic fluctuation characteristics and abnormal data collection, this paper proposes a photovoltaic user cluster baseline load decoupling estimation method based on multi-dimensional abnormal data repair. Firstly, a multidimensional anomaly data repair system is established based on extreme learning machines and multi-step prediction. Secondly, using the repaired data for baseline estimation, photovoltaic users and non-photovoltaic users are decoupled and separated. Long short-term memory neural network estimation models are used for non-photovoltaic users, and baseline load estimation methods based on weather type analysis are used for photovoltaic users. Finally, the baseline results are accumulated. The simulation results show that compared with the direct estimation method of baseline load, the accuracy of decoupling estimation has improved by 33.7%.

Robust neuro-adaptive command-filtered back-stepping fault-tolerant control of satellite using composite learning

This study proposes an adaptive neural command-filtered backstepping control system for precise and fast attitude control of a satellite considering different uncertain dynamics. Mission success crucially depends on robust attitude control resilient to model uncertainties, unmodeled dynamics, external disturbances, and actuator faults. The proposed approach synergistically combines a neural network for handling model uncertainties, unmodeled dynamics, and actuator faults, with a disturbance observer for compensating external disturbances and neural network estimation errors. The command-filtered backstepping technique avoids the explosion of complexity inherent to traditional backstepping, while integrating integral action into the design results in the elimination of the steady-state tracking error. Besides, a composite learning method optimizes the update laws for the neural network and disturbance observer weights, enhancing control performance. Despite the presence of uncertainties, the closed-loop system stability is guaranteed by the Lyapunov stability theorem. Simulation results demonstrate the proposed controller’s ability to handle severe actuator faults, unmodeled dynamics, and measurement noise without requiring explicit fault detection and isolation schemes.

The Degree of Best Downsampled Convolutional Neural Network Approximation in Terms of Orlicz Modulus of Smoothness

The Degree of Best Downsampled Convolutional Neural Network Approximation in Terms of Orlicz Modulus of Smoothness

Optimizing stock market volatility predictions based on the SMVF-ANP approach

The stock market is considered one of the most complicated financial systems, comprising several components or inventories, whose prices vary substantially over time. Bursaries include revealing market tendencies over time. All investors in the stock market aim to maximize profits and reduce the associated risks. Stock currency forecasts are a significant financial concern that is being handled even more closely. In recent years, various neural network and hybrid models have surpassed classic linear and non-linear techniques to produce reliable prediction outcomes. This study investigates the efficacy of dynamic and effective stock market forecasting using neural network models. This study examines market transmission mechanisms and assesses the predicted links between multiple financial and economic factors. A stock market volatility and artificial network prediction (SMVF–ANP) approach is presented. The models analyzed included a multi-layer perceptron (MLP), dynamic artificial neural network, and generalized autoregressive conditional heteroscedasticity to extract additional input variables. The results reveal that the trade strategies led by the classification models yield superior risk-adjusted returns compared to the buy-and-hold approach and those led by the neural network and linear regression model-level estimate predictions. The numerical results show that the proposed SMVF-ANP method achieved a high performance ratio of 94.1%, an enhanced prediction ratio of 98.4%, a high stock market volatility rate of 96.7%, a reduced mean square percentage error ratio of 16.3%, a probability rate of 32.7%, an increased efficiency rate of 96.9%, and an accuracy ratio of 97.2% compared to other methods.

Disturbance observer-based finite-time adaptive neural control scheme of DFIG-wind turbine

This paper introduces a novel disturbance observer-based finite-time adaptive neural control approach to optimize wind power conversion in a doubly fed induction generator-based wind turbines (DFIG-WT). This control strategy offers appealing features including rapid finite-time convergence, both transient and steady-state performance enhancements, and robustness against external disturbances and inherent model uncertainties. The control strategy integrates the neural networks estimation capability with the interesting proprieties of the finite-time control method to achieve efficient wind power conversion. Closed-loop finite-time stability is conducted using the finite-time Lyapunov stability concept of nonlinear systems. The developed control strategy’s effectiveness is confirmed through numerical simulation.

Neural network-based hybrid modeling approach incorporating Bayesian optimization with industrial soft sensor application

Hybrid modeling combines physical and data-driven models to improve the performance of industrial soft sensors. However, simplified physical assumptions and extensive parameter optimization increase the complexity of the model and decrease its robustness and accuracy. In this study, a novel neural network-based approach for hybrid modeling was developed, based on Runge-Kutta (RK) and Bayesian hyperparameter optimization. First, the recursive solution equation for the quality variable was derived using the master equation, which included the black-box term and RK method. Second, the numerical solution of the black-box term was inverted using the derived equation as a label for the training samples, and hyperparameters were optimized using Bayesian optimization. Finally, during the prediction phase, quality variables at subsequent times were derived by solving a recursive equation containing neural network approximations. The superiority of the proposed soft sensing method was assessed through studies on a continuous stirred tank reactor (CSTR) case and a simulated penicillin case. The experimental findings demonstrate that the proposed method is better than existing methods in terms of robustness and accuracy.

Utilization of genetic algorithm in tuning the hyper-parameters of hybrid NN-based side-slip angle estimators

This paper proposes a solution to enhance and compare different neural network (NN)-based side-slip angle estimators. The feed-forward neural networks (FFNNs), recurrent neural networks, long short-term memory units (LSTMs), and gated recurrent units are investigated. However, there is a lack in the selection criteria of the architectures’ hyper-parameters. Therefore, the genetic algorithm is integrated with the NN-based estimators to find the optimal hyper-parameters for the studied architectures. The tuned hyper-parameters in this work include the number of neurons, number of layers, activation function, optimizer type, and learning rate. The objective function of the optimization problem is minimizing the root-mean-square error (RMSE) on multiple testing data. The optimal models are further included in the design of a hybrid NN estimator with Kalman filter. In the hybrid estimators, the optimal NN estimators are used as virtual sensors to correct the prediction of the side-slip angle resulting from the mathematical lateral vehicle model. Eventually, the performance of the best selected model is evaluated in terms of different metrics; mean RMSE, mean error variance, mean training time, and mean estimation time. LSTMs are found to achieve the lowest mean RMSE while being tested on highly generalized data yielding the highest training and estimation time. However, FFNNs achieve the lowest RMSE while being tested on low generalized data and the lowest training and estimation time. Meanwhile, it is observed that the hybrid estimators achieved lower RMSE with great enhancement compared to the non-hybridized ones proving the effectiveness of the proposed approach and increasing the side-slip estimation generalization ability in unknown environments with high uncertainties, which are not covered by the training dataset for the NNs estimators.

Adaptive neural event-dependent intermittent fault-tolerant control of reaction–diffusion multi-agent systems

This article employs an event-dependent intermittent fault-tolerant control approach to address the adaptive consensus of reaction–diffusion multi-agent systems under directed topologies in both leaderless and leader-following networks. Initially, an event-dependent intermittent control mechanism is proposed to determine the work/rest time, relying on three partitions of the non-negative real region and the formulated Lyapunov function. Subsequently, under the introduced control mechanism, an adaptive consensus protocol is established, incorporating the updating law for adaptive coupling strength and adaptive weight estimation. Utilizing the neural network estimation theorem for fault estimation, the article presents explicit criteria to ensure the attainment of leaderless or leader-following consensus in reaction–diffusion multi-agent systems. Finally, two numerical simulation experiments are conducted to validate the robustness and effectiveness of the derived theorem.

Physics-informed neural network estimation of material properties in soft tissue nonlinear biomechanical models

AbstractThe development of biophysical models for clinical applications is rapidly advancing in the research community, thanks to their predictive nature and their ability to assist the interpretation of clinical data. However, high-resolution and accurate multi-physics computational models are computationally expensive and their personalisation involves fine calibration of a large number of parameters, which may be space-dependent, challenging their clinical translation. In this work, we propose a new approach, which relies on the combination of physics-informed neural networks (PINNs) with three-dimensional soft tissue nonlinear biomechanical models, capable of reconstructing displacement fields and estimating heterogeneous patient-specific biophysical properties and secondary variables such as stresses and strains. The proposed learning algorithm encodes information from a limited amount of displacement and, in some cases, strain data, that can be routinely acquired in the clinical setting, and combines it with the physics of the problem, represented by a mathematical model based on partial differential equations, to regularise the problem and improve its convergence properties. Several benchmarks are presented to show the accuracy and robustness of the proposed method with respect to noise and model uncertainty and its great potential to enable the effective identification of patient-specific, heterogeneous physical properties, e.g. tissue stiffness properties. In particular, we demonstrate the capability of PINNs to detect the presence, location and severity of scar tissue, which is beneficial to develop personalised simulation models for disease diagnosis, especially for cardiac applications.

Smooth Logistic Real and Complex, Ordinary and Fractional Neural Network Approximations over Infinite Domains

In this work, we study the univariate quantitative smooth approximations, including both real and complex and ordinary and fractional approximations, under different functions. The approximators presented here are neural network operators activated by Richard’s curve, a parametrized form of logistic sigmoid function. All domains used are obtained from the whole real line. The neural network operators used here are of the quasi-interpolation type: basic ones, Kantorovich-type ones, and those of the quadrature type. We provide pointwise and uniform approximations with rates. We finish with their applications.

Deep reinforcement learning-based dynamic multi-beam power allocation for GEO-LEO co-existing satellites

This paper first formulates a novel long-term beam power allocation (BPA) problem to tackle the harmful co-linear interference issue in the geostationary earth orbit (GEO) and low earth orbit (LEO) co-existing satellite system. This BPA problem intends to optimize the long-term weighted sum rate of the LEO system while ensuring that GEO user’s received interference from the LEO satellite system is lower than a pre-fixed threshold. To solve it in a real-time manner, a deep reinforcement learning (DRL) framework based on the proximal policy optimization (PPO) algorithm is proposed, named as drlBPA. In addition, for the existing most relevant baseline, the fractional optimization (FO)-based BPA scheme, on the one hand, this paper improves it via a greedy strategy to fully exploit time resource. On the other hand, to further reduce the computational complexity stemming from its iterative solving procedure, a deep neural network approximation scheme is also developed. Simulation results demonstrate that (i) The trained DRL model of the proposed drlBPA scheme has good convergence and generality performance. (ii) Compared with the three FO-based benchmarks, the drlBPA scheme not only achieves the highest throughput of the LEO system within a significantly reduced computation time, but also yields the best system stability.

Neural-Network-Based Predefined-Time Adaptive Consensus in Nonlinear Multi-Agent Systems With Switching Topologies.

A predefined-time adaptive consensus control strategy is developed for a class of multi-agent systems containing unknown nonlinearity. The unknown dynamics and switching topologies are simultaneously considered to adapt to actual scenarios. The time required for tracking error convergence can be easily adjusted using the proposed time-varying decay functions. An efficient method is proposed to determine the expected convergence time. Subsequently, the predefined time is adjustable by regulating the parameters of the time-varying functions (TVFs). The neural network (NN) approximation technique is used to address the issue of unknown nonlinear dynamics through predefined-time consensus control. The Lyapunov stability theory testifies that the predefined-time tracking error signals are bounded and convergent. The feasibility and effectiveness of the proposed predefined-time consensus control scheme are demonstrated through the simulation results.

Event-triggered stochastic finite-time tracking control of robot manipulator with uncertain disturbance neural network estimation

Event-triggered stochastic finite-time tracking control of robot manipulator with uncertain disturbance neural network estimation

Unknown input observer based neuro-adaptive fault-tolerant control for vehicle platoons with sensor fault and output quantization

Sensor fault and output quantization are common issues acting on vehicle platoon, and they may lead to performance deterioration, instability and even insecurity of the platoon. Therefore, this paper investigates the fault-tolerant control (FTC) problem of vehicle platoons with regard to the above two issues. Considering the probabilistic sensor fault and quantized measurement signals, an unknown input observer (UIO) based fault detection algorithm with adaptive threshold is developed for sensor health status monitoring. Then, an augmented vehicle platoon model is constructed by introducing a low-pass output filter, and a robust UIO is established for state reconstruction. Based on the above results, a fault-tolerant control scheme is exploited by employing the back-stepping control method and adaptive radial basis function neural network (RBF NN) approximation technique, which is proved to be capable of achieving the time-domain string stability (TSS) of vehicle platoons in the presence of sensor fault and output quantization. Simulation results demonstrate the effectiveness of the proposed algorithms.

Adaptive position control using backstepping technique for the leader‐follower multiple quadrotor unmanned aerial vehicle formation

SummaryThis article addresses the position control issue of multi‐quadrotor unmanned aerial vehicle (QUAV) formation. Concerning the translational dynamic of a multi‐QUAV system, on the one hand, it is an under‐actuation dynamic; on the other hand, it does not satisfy the matching condition. These features will cause inevitable thorny in the formation position control design. Furthermore, because of the state coupling problem, the formation control of multi‐QUAV system is more challenging and knotty than the control of single QUAV system. To achieve this control, both backstepping technique and neural network (NN) approximation strategy are combined by introducing an intermediary control, where NN is employed to compensate the system uncertainty. However, since the traditional adaptive NN control methods need to train a large number of adaptive parameters for the high approximation accuracy, it will cause the heavy computing burden if traditional adaptive method is used for the QUAV formation control. The proposed adaptive NN strategy in this paper only requires training a scalar adaptive parameter, which is generated from the norm of NN weight vector or matrix, thereby significantly reducing computational burden. Finally, according to Lyapunov stability proof and computer simulation, it is demonstrated that the control tasks can be successfully accomplished.

Deep learning for predicting 16S rRNA gene copy number

Culture-independent 16S rRNA gene metabarcoding is a commonly used method for microbiome profiling. To achieve more quantitative cell fraction estimates, it is important to account for the 16S rRNA gene copy number (hereafter 16S GCN) of different community members. Currently, there are several bioinformatic tools available to estimate the 16S GCN values, either based on taxonomy assignment or phylogeny. Here we present a novel approach ANNA16, Artificial Neural Network Approximator for 16S rRNA gene copy number, a deep learning-based method that estimates the 16S GCN values directly from the 16S gene sequence strings. Based on 27,579 16S rRNA gene sequences and gene copy number data from the rrnDB database, we show that ANNA16 outperforms the commonly used 16S GCN prediction algorithms. Interestingly, Shapley Additive exPlanations (SHAP) shows that ANNA16 can identify unexpected informative positions in 16S rRNA gene sequences without any prior phylogenetic knowledge, which suggests potential applications beyond 16S GCN prediction.

Cognitive analytics enabled responsible artificial intelligence for business model innovation: A multilayer perceptron neural networks estimation

Cognitive analytics employs and analyses complex and heterogeneous data sources generating deeper insights that mimic the natural intelligence of the human brain. Cognitive analytics-enabled Artificial Intelligence (AI) that promotes Business Model Innovation (BMI) for the efficiency of the healthcare system is a nascent and undertheorized domain. Within the healthcare management systems, stakeholders’ engagement with AI, particularly with responsible AI, to optimize BMI and improve business performance is bounded by several caveats. Using the Technology Acceptance Model (TAM) and Social Network Theory (SNT) as our conceptual foci, we empirically examine through the Multilayer Perceptron Neural Network the extent to which responsible AI leads to Business Model Innovation (BMI) through the stakeholders’ engagement. Our contributions are novel which demonstrate that cognitive analytics-enabled responsible AI is central to innovation, and healthcare stakeholders exhibit a robust propensity to reorientate and innovate their existing BMI to achieve improved business performance. It has significant implications for innovation, AI and cognitive analytics literature.

Inf-sup neural networks for high-dimensional elliptic PDE problems

Solving high dimensional partial differential equations (PDEs) has historically posed a considerable challenge when utilizing conventional numerical methods, such as those involving domain meshes. Recent advancements in the field have seen the emergence of neural PDE solvers, leveraging deep networks to effectively tackle high dimensional PDE problems. This study introduces Inf-SupNet, a model-based unsupervised learning approach designed to acquire solutions for a specific category of elliptic PDEs. The fundamental concept behind Inf-SupNet involves incorporating the inf-sup formulation of the underlying PDE into the loss function. The analysis reveals that the global solution error can be bounded by the sum of three distinct errors: the numerical integration error, the duality gap of the loss function (training error), and the neural network approximation error for functions within Sobolev spaces. To validate the efficacy of the proposed method, numerical experiments conducted in high dimensions demonstrate its stability and accuracy across various boundary conditions, as well as for both semi-linear and nonlinear PDEs.

Adaptive finite-time neural control of nonstrict-feedback nonlinear systems with input dead-zone and output hysteresis

This paper explores the adaptive finite-time neural control issue for nonlinear systems with input dead zone and output hysteresis in nonstrict-feedback form. The unknown functions are estimated by employing the radial basis function neural networks (RBFNN) approach. A systematic adaptive finite-time control method is introduced using the backstepping technique and neural network approximation properties. The stability of the system is also examined by using semi-global practical finite-time stability theory. The established control approach guarantees the boundedness of all signals within the closed-loop system, enabling the system output to accurately follow the desired signal within a finite time framework while maintaining a small and bounded tracking error. Finally, simulation results are shown to demonstrate the efficacy of the suggested strategy.

State estimation and structure identification of complex networks with time‐varying delays and model uncertainty based on probabilistic graph neural network

SummaryIn this paper, the state estimation problem and structure identification problem are investigated for an array of complex networks with time‐varying delays and model uncertainty. Model uncertainty is generated by uncertain connections between nodes in complex networks. Combining probabilistic graph model and graph neural network, a probabilistic graph neural network is proposed to deal with the uncertainty, and the multiple linear terms of the errors between nodes are estimated. The purpose of the addressed state estimation problem is to design a state estimator such that, in the presence of the model uncertainty, the estimation error to converge to zero can be guaranteed and the explicit expression of the estimator parameters is given. Based on a Lyapunov function, the parameters of the estimator and the updating laws of the weights of probabilistic graph neural network can be obtained by the solutions to linear matrix inequalities. The purpose of the addressed structure identification problem is to infer a definite graph structure model based on Bayes' theorem in the presence of a stable probabilistic graph neural network and state estimator. Finally, an illustrative example is provided to demonstrate the feasibility and effectiveness of the developed state estimation and structure identification scheme.