Radial Basis Function Network Research Articles

An adaptive learning strategy for surrogate modeling of high-dimensional functions - Application to unsteady hypersonic flows in chemical nonequilibrium

Many engineering applications rely on the evaluation of expensive, non-linear high-dimensional functions. In this paper, we propose the RONAALP algorithm (Reduced Order Nonlinear Approximation with Adaptive Learning Procedure) to incrementally learn, on-the-fly as the application progresses, a fast and accurate reduced-order surrogate model of a target function. First, a combination of nonlinear auto-encoder, community clustering, and radial basis function networks allows us to learn an efficient and compact surrogate model with limited training data. Secondly, an active learning procedure overcomes any extrapolation issues during the online stage by adapting the surrogate model with high-fidelity evaluations that fall outside its current validity range. This approach results in generalizable, fast, and accurate reduced-order models of high-dimensional functions. The method is demonstrated on three direct numerical simulations of hypersonic flows in chemical nonequilibrium. Accurate simulations of these flows rely on detailed thermochemical gas models that dramatically increase the cost of such calculations. Using RONAALP to learn a reduced-order thermodynamic model surrogate on-the-fly, the cost of such simulations was reduced by up to 75% while maintaining an error of less than 10% on relevant quantities of interest.

Solution of the Forward Problem of ECG using Radial basis Function Network

ECG inverse problem is a mathematical formulation of cardiac electrical activity on the heart surface that can be solved to determine the electric potential on the heart surface. In this paper, a forward problem to ECG will be formulated using a mathematical formulation similar to the ECG forward problem. The forward problem will be solved using radial basis collocation techniques by taking different discretization points of N=9 and N=25 points to predict the solution u(x,t) to the forward problem.

A Target Collision Algorithm of Autonomous Mobile Robot for Two-player Competition

Abstract A variety of robot competitions are held to train and attract students everywhere. This paper studies the algorithm of autonomous robot with two-player competition. The autonomous robots in a two-player competition takes the scoring form of collision and wins first. The impact object is the enemy target. The key algorithms in a two-player competition are target detection, target path decision, and collision target motion control. First, the vehicle target data set was established and enhanced, the lightweight detection algorithm model was screened and trained, the model pruning optimization experiment and the model deployment optimization experiment were carried out, and then the TensorRT framework was used for inter-layer fusion and data quantification processing. After hitting the target path decision algorithm after detecting the target target, it finally integrates the decision in simple and complex situations. The collision target motion control is the scheme of more intelligent and self-adjustable parameters of fuzzy Radial Basis Function (RBF) network Proportional Integral Derivative (PID) controller. Finally, the collision target implementation algorithm is tested in a real environment, and the results show that the task requirements can be met.

Experimental Investigation on collection efficiency of Deep-Sea hydraulic collection parameters

Hydraulic suction collection has been extensively utilized in deep-sea mining due to its high flow rate and efficiency. This paper introduces a novel hydraulic collection model to investigate the impact of various collection parameters on collection efficiency. We analyze the effect of different ore collection parameters on collection efficiency using a radial basis function network and make predictions. The optimal parameter range was obtained by performing combined parameter experiments. The two-parameter combination give better results compared to a single parameter. A theoretical model of collection efficiency is established based on experimental observations and principles of water jet theory. The accuracy of the model is confirmed through experimental validation, and the research findings can provide valuable insights for the ore collection process in deep-sea mining.

Design and development of different adaptive MPPT controllers for renewable energy systems: a comprehensive analysis

As of now, all over the world is focusing on the Electric Vehicle (EV) technology because its features are low environmental pollution, less maitainence cost required, high robustness, and good dynamic response. Also, the EVs work continuously until the input fuel is provided to the fuel stack. Here, a Proton Exchange Membrane Fuel Cell (PEMFC) is used as an input source to the electric vehicle system because of its merits fast startup, and quick response. However, the PEMFC gives nonlinear voltage versus current characteristics. As a result, the extraction of maximum power from the fuel stack is very difficult. The main aim of this work is to study different Maximum Power Point Tracking Techniques (MPPT) for the DC-DC converter-fed PEMFC system. The studied MPPT controllers are Adjusted Step Value of Perturb & Observe (ASV with P&O), Adaptive Step Size with Incremental Conductance (ASS with IC), Radial Basis Functional Network (RBFN), Incremental Step-Fuzzy Logic Controller (IS with FLC), Continuous Step Variation based Particle Swarm Optimization (CSV with PSO), and Adaptive Step Value-Cuckoo Search Algorithm (ASV with CSA). The selected MPPT controllers’ comprehensive study has been in terms of maximum power extraction, tracking speed of the MPP, settling time of the fuel stack output voltage, oscillations across the MPP, and design complexity. From the comprehensive performance results of the hybrid MPPT controllers, the ASV with CSA technique gives superior performance when equated to the other MPPT controllers.

Investigation of interfacial bond strength of circular CFST columns based on ordinary and inorganic polymers through advanced machine learning methods

Concrete-filled steel tube (CFST) is a combination of materials, in which the interfacial bond-slip behavior is the prerequisite to reflect the synergistic loading of external steel tube and core concrete. The inorganic polymer concrete within the concrete-filled steel tube (IPCFST) compensates for the drying shrinkage of the concrete and enhances the adhesion at the interface between the steel tube and the concrete. However, the current code is too simple for the bond strength of CFSTs, and the existing theoretical calculation methods are not universally applicable. This paper aims to develop a practical artificial neural network tool for predicting the interfacial bond strength of ordinary CFSTs and IPCFSTs. An efficient prediction model for the bond strength of CFST interfaces was developed using a radial basis function network (RBFNN), with concrete strength, steel tube size, bond length and other factors as the main parameters. Bond performance test data was collected for a sample database of 322 circular CFST columns, including 56 ordinary CFSTs and 263 IPCFSTs. The results show that the Artificial Neural Network (ANN) algorithm can effectively address the bond strength problem of CFST structures and improve the accuracy and density of prediction in the comparative analyses with the measured values and the results of the semi-theoretical formulations. Based on the prediction results of ANN model, a sensitivity analysis of the parameters using Garson's algorithm shows that the diameter-to-thickness ratio and concrete strength significantly affect bond strength for both ordinary CFSTs and IPCFSTs. This study provides a more reliable prediction method and technical support for the engineering design of CFST structures.

Data-driven control of wind turbine under online power strategy via deep learning and reinforcement learning

This study proposes a data-driven wind turbine (WT) model predictive control (MPC) enhanced by a deep-learning (DL) radial basis function network (RBFN) and a reinforcement-learning (RL) deep Q-learning network (DQN). The RBFN provides comprehensive aerodynamic predictions, including thrust, torque, and power. Besides, the MPC linearization relies on the RBFN prediction to estimate force sensitivities. The DQN achieves an online power strategy (OPS) that solves the 2-degree-of-freedom (2-DOF) optimization of rotor speed and pitch angle, which can actively adjust power capture to meet different power requirements. The DQN adopts a novel bisection algorithm with a first-in-first-out (FIFO) queue for high-precision 2-DOF results. The MPC coordinates the permanent magnet synchronous generator (PMSG) and pitch servo, considering shaft rotation and tower movement. Compared with the maximum power point tracking (MPPT) and power reference point tracking (PRPT) based controls, the proposed RBFN-DQN-MPC reduces power fluctuation and ensures constant output. This study also compares the DQN with the categorical DQN (C51), which indicates that the DQN is more effective in the 2-DOF optimization. Hence, WTs enhanced by the DL-RL-MPC are intelligent and reliable for flexible wind generation.

Generalized [formula omitted]-sparse algorithms for constructing fault tolerant RBF networks

In the construction process of radial basis function (RBF) networks, two common crucial issues arise: the selection of RBF centers and the effective utilization of the given source without encountering the overfitting problem. Another important issue is the fault tolerant capability. That is, when noise or faults exist in a trained network, it is crucial that the network’s performance does not undergo significant deterioration or decrease. However, without employing a fault tolerant procedure, a trained RBF network may exhibit significantly poor performance. Unfortunately, most existing algorithms are unable to simultaneously address all of the aforementioned issues. This paper proposes fault tolerant training algorithms that can simultaneously select RBF nodes and train RBF output weights. Additionally, our algorithms can directly control the number of RBF nodes in an explicit manner, eliminating the need for a time-consuming procedure to tune the regularization parameter and achieve the target RBF network size. Based on simulation results, our algorithms demonstrate improved test set performance when more RBF nodes are used, effectively utilizing the given source without encountering the overfitting problem. This paper first defines a fault tolerant objective function, which includes a term to suppress the effects of weight faults and weight noise. This term also prevents the issue of overfitting, resulting in better test set performance when more RBF nodes are utilized. With the defined objective function, the training process is designed to solve a generalized M-sparse problem by incorporating an ℓ0-norm constraint. The ℓ0-norm constraint allows us to directly and explicitly control the number of RBF nodes. To address the generalized M-sparse problem, we introduce the noise-resistant iterative hard thresholding (NR-IHT) algorithm. The convergence properties of the NR-IHT algorithm are subsequently discussed theoretically. To further enhance performance, we incorporate the momentum concept into the NR-IHT algorithm, referring to the modified version as “NR-IHT-Mom”. Simulation results show that both the NR-IHT algorithm and the NR-IHT-Mom algorithm outperform several state-of-the-art comparison algorithms.

Early detection of abiotic stress in plants through SNARE proteins using hybrid feature fusion model.

Agriculture is the main source of livelihood for most of the population across the globe. Plants are often considered life savers for humanity, having evolved complex adaptations to cope with adverse environmental conditions. Protecting agricultural produce from devastating conditions such as stress is essential for the sustainable development of the nation. Plants respond to various environmental stressors such as drought, salinity, heat, cold, etc. Abiotic stress can significantly impact crop yield and development posing a major threat to agriculture. SNARE proteins play a major role in pathological processes as they are vital proteins in the life sciences. These proteins act as key players in stress responses. Feature extraction is essential for visualizing the underlying structure of the SNARE proteins in analyzing the root cause of abiotic stress in plants. To address this issue, we developed a hybrid model to capture the hidden structures of the SNAREs. A feature fusion technique has been devised by combining the potential strengths of convolutional neural networks (CNN) with a high dimensional radial basis function (RBF) network. Additionally, we employ a bi-directional long short-term memory (Bi-LSTM) network to classify the presence of SNARE proteins. Our feature fusion model successfully identified abiotic stress in plants with an accuracy of 74.6%. When compared with various existing frameworks, our model demonstrates superior classification results.

A classifier based on mixed radial basis function network and combinatorial optimization model for medical diseases diagnosis

A classifier based on mixed radial basis function network and combinatorial optimization model for medical diseases diagnosis

Startup drift compensation of RLG based on monotone constrained RBF neural network

Serious startup drift of the Ring Laser Gyroscope (RLG) is observed during cold startup process, which will dramatically degrade the performances of the corresponding Inertial Navigation System (INS). In this paper, correlation analysis method, which analyzes the relationship between the startup drift of the RLG and the temperature change, is used to determine the significant temperature-related terms during gyroscope startup. Based on the significant temperature-related terms and the startup time length, a startup drift compensation model for RLG based on monotonicity-constrained Radial Basis Function (RBF) neural network is proposed and validated. Compared with the raw RLG data without compensation, the standard deviation of the RLG output with the proposed constrained RBF network model is decreased by more than 46%, and the peak-to-peak value is decreased by more than 35%. Compared with the traditional multiple regression model, the standard deviation and peak-to-peak value of the RLG output are decreased by more than 10% and 6%, respectively. Compared with the common RBF network model, the standard deviation and peak-to-peak value of the RLG output are decreased by more than 8% and 3%, respectively. Navigation experiments also validate the effectiveness of the compensation model.

Machine learning analysis of pressure fluctuations in a gas-solid fluidized bed

Pressure fluctuations in a gas-solid fluidized bed involve much information of the dynamic system. To uncover the value and significance of the pressure fluctuations time series, two meaningful tasks, i.e., the fluidization regime classification and the future state prediction, are investigated using Machine Learning (ML) algorithms, including Back Propagation neural network (BP), Convolutional Neural Network (CNN), Long Short-Term Memory network (LSTM), Radial Basis Function network (RBF), Random Forest (RF) and Support Vector Machine (SVM). Findings indicate that the six ML methods rank their performance for the classification task from high to low as: BP ∼ RF > SVM > CNN ∼ LSTM ∼ RBF. For the one-step-ahead future state prediction, the accuracy goes from high to low as: BP ∼ RBF > CNN ∼ LSTM > SVM ∼ RF, whilst for the multistep-ahead prediction the LSTM approach shows the best performance. Additionally, the sampling frequency of pressure time series has significant effects on both tasks. Overall, this study demonstrates the capability of machine learning methods for the value analysis of nonlinear time series and the better operation of gas-solid fluidization systems.

Safe dynamic sparse training of modified RBF networks for joint feature selection and classification

Developing an efficient embedded feature selection method for both binary and multiclass classification problems is a fundamental topic to be further studied. In this paper, the conventional radial basis function (RBF) network is modified for joint feature selection and classification. By utilizing the anisotropic Gaussian basis function, an individual weight parameter is assigned to each feature for feature weighting. The L1-regularized loss function is designed by combining the softmax cross entropy loss with L1 regularization terms imposing on feature and output weights. A specialized active-set limited-memory projected Quasi-Newton (SAL-PQN) algorithm is proposed to minimize the L1-regularized loss function, which can optimize all the adjustable parameters and zero out the redundant feature and output weights simultaneously. Furthermore, a safe dynamic pruning strategy is integrated into the SAL-PQN algorithm for dynamic sparse training of the modified RBF network, which continuously excludes the dynamically zeroed-out feature weights from the training process. It is theoretically assured of yielding the same gradient updates as SAL-PQN, thus the relevant computation of the zeroed-out feature weights in the forward and backward propagations can be safely eliminated. The experimental results demonstrate the effectiveness and efficiency of the proposed methods for joint feature selection and classification, and illustrate the practical utility of the inherent feature weighting capability.

Neural network-based distributed intelligent supervisory control for multi-agent systems with unknown input powers

In this paper, a synchronisation problem is studied for a class of nonlinear leader-following systems. Based on adaptive control with some algebra lemmas, we propose a distributed control scheme. This scheme ensures each follower to asymptotically synchronise with the leader. Compared with the existing works where system input powers are assumed one, the input powers considered in this paper are unknown but larger than one. Based on Graph theory, Lyapunov theory and radial basis function (RBF) networks, we design a distributed adaptive supervisory control method. It is proven that the consensus among the leader and followers are achieved and all the signals including tracking errors asymptotically converge to a small neighbourhood of the origin. Finally, simulation results are displayed to demonstrate the effectiveness of control scheme.

Multi-response Optimization of GTAW Process Parameters in Terms of Energy Efficiency and Quality

This work optimizes the current (I), voltage (V), flow rate (F), and arc gap (G) of the gas tungsten arc welding (GTAW) of the Ti40A titanium alloy to decrease the heat input (HI) and improve the ultimate tensile strength (TS) and micro-hardness (MH). The radial basis function network (RBFN) was utilized to present performance measures, while weighted principal component analysis (WPCA) and an adaptive non-dominated sorting genetic algorithm II (ANSGA-II) were applied to estimate the weights and generate optimal points. The evaluation via an area-based method of ranking (EAMR) was employed to determine the best solution. The results indicated that the optimal I, V, F, and G are 89 A, 23 V, 20 L/min, and 1.5 mm, respectively. The improvements in the TS and MH were 1.2 % and 19.8 %, respectively, while the HI was saved by 18.4 %. The RBFN models provided acceptable accuracy for prediction purposes. The ANSGA-II provides better optimality than the conventional NSGA-II. The HI, TS, and MH of the practical GTAW Ti40A could be enhanced using optimality. The optimization method could be utilized to deal with optimization problems for not only other GTAW operations but also other machining processes.

Combining simulated annealing and RBF networks for accurate volumetric fraction determination of two-phase flows

The importance of flow monitoring in the oil industry has expanded due to the global need for fossil fuels. This has led to the emergence of a new subset of the flowmeter market. The goal of this study is to use a Radial Basis Function (RBF) neural network developed through Simulated Annealing (SA) to pick features of the signals generated by gamma-based flowmeters in order to determined volumetric fractions. The volumetric detection system presented in this article consists of a137Cs isotope as gamma emitter, two NaI detectors for collecting the photons, and a glass pipe in between them. Monte Carlo N-Particle (MCNP) was used to model the above-mentioned geometry. Fifteen wavelet, frequency, and time characteristics were extracted from the raw data captured by both detectors. First, the SA optimization algorithm was used to identify the suitable attributes. Five useful features were presented as a consequence of this procedure, and they were fed into the RBF network in order to estimate volumetric percentages. This study is innovative in that it combines the RBF neural network with the SA algorithm to pick effective features. The outcome is as follows: (1) Outlining five appropriate characteristics for use in determining percentages of volume. (2) Predicting the volume fraction of materials in two-phase flow with a root mean square error of less than 0.22. (3) By recognizing suitable inputs using the method based on the SA algorithm, the artificial neural network can determine the target output with less computational load.

Anti-windup design for supercavitating vehicle based on sliding mode control combined with RBF network

During the longitudinal motion of a supercavitating vehicle, the stability control problem is complicated because of the nonlinear planing force on the tail part. The dynamic model of a supercavitating vehicle in longitude plane is nonlinear, simultaneously, the control instructions of a supercavitating vehicle may exceed the physical limits of an actuator. Therefore, designing a longitudinal stability control system for a supercavitating vehicle, not only the treatment of nonlinear planing force, but also the physical constraints of the actuator should be considered. For the longitudinal motion model of supercavitating vehicle, a cascade model is proposed, which decomposes the longitudinal motion of supercavitating vehicle into two subsystems. Sliding mode control based on RBF neural network compensation is adopted in the controller design process, and RBF neural network is exploited to approach the deviation caused by actuator saturation. The proposed control method can effectively compensate the performance degradation caused by control variable saturation, and has strong robustness.

Numerical analysis of uncertain frequency of rotating composites using NURBS-based isogeometric HSDT approach: Rim-driven configuration

This study delves into the impact of material and fabrication uncertainties on the natural frequencies of rim-driven rotating structures. Utilizing a higher-order shear deformation theory (HSDT), an isogeometric analysis (IGA) model predicts these frequencies. Sensitivity analysis via a radial basis function network (RBFN) model, validated with Monte Carlo simulation, highlights parameter influence. Parametric investigations unveil sensitivity patterns and influential parameters. Employing a stochastic approach, the study compares the effects of random material properties on the free vibration response of rotating and non-rotating plates. Additionally, the RBFN-based surrogate model assesses reliability parameters concerning resonance failure in rim-driven rotating composite structures.

Improved Glowworm Swarm Optimization for Parkinson’s Disease Prediction Based on Radial Basis Functions Networks

Parkinson’s disease is caused by a disruption in the chemical products that enables the communication between brain cells. The brain’s dopamine cells are responsible for movement control, adaptability, and fluidity. Parkinson’s motor symptoms manifest when 60–80% of these cells are damaged due to insufficient dopamine. Researchers are working to find a way to identify the non-motor symptoms that manifest early detection in the disease to stop the disease’s progression because it is believed that the disease starts many years before the motor symptoms. This research presents Parkinson’s disease diagnosis based on deep learning. Processes for feature selection and classification encompass the suggested diagnosis technique. The proposed model searches for the best subset of characteristics using the Improved Glowworm Swarm Optimization (IGSO) algorithm. Radial Basis Functions Networks (RBFN) classifiers evaluate the chosen features. The suggested model is tested using datasets from Parkinson’s Handwriting samples and Parkinson’s Speech and voice with various sound recordings. With an accuracy of about 95.78%, the suggested algorithm forecasts Parkinson’s disease using the VoicePD dataset more precisely.

Research on load excitation identification method of multi-connected air conditioning compressor based on RBF network with multi-strategy fusion SSA

Research on load excitation identification method of multi-connected air conditioning compressor based on RBF network with multi-strategy fusion SSA