Abstract This study proposes a fuzzy neural network-based prediction and optimization method to address the challenge of modeling dynamic stiffness in Stewart platforms. Traditional approaches, such as the Newton-Euler method and finite element analysis, often struggle to capture nonlinear characteristics and multivariate coupling effects under complex conditions. To overcome these limitations, this paper constructs a fuzzy neural network framework that integrates fuzzy logic with neural computation. This model selects drive joint position, velocity, acceleration, torque, and external load as input variables. These inputs are mapped into fuzzy subsets through fuzzification. The fuzzy radial basis function network is designed to simulate the nonlinear relationships between input variables and dynamic stiffness. An error back-propagation algorithm is applied to optimize the network weights, and the structure is refined using cross-validation and grid search. The fuzzy rule base is constructed from both expert knowledge and data-driven insights. Experimental validation is conducted under varying working conditions. This includes load variation and angular velocity changes. The proposed method demonstrates higher accuracy and robustness compared to traditional Newton-Euler, finite element, statistical regression, and reinforcement learning models. The average mean square error under most scenarios is significantly reduced. This paper also highlights the limitations of current fuzzy rule adaptability under unknown disturbances. Future work aims to enhance model generalizability through self-learning mechanisms and simplify computational complexity for real-time applications. Overall, this study contributes a reliable and adaptive approach to improving dynamic stiffness prediction for Stewart platforms, offering insights for broader applications in multi-degree-of-freedom robotic systems.

Abstract In the last few years, hybrid photovoltaic-thermal (PVT) collectors have become an attractive subject of research because of their ability to convert solar radiation into both electrical and thermal energies. Nonlinear relationships among their control variables, such as design parameters, climatic conditions, heat transfer fluid type, and electrical and thermal performances, require advanced modeling methodologies. This review examines the application of machine learning, especially artificial neural networks (ANNs), in photovoltaic-thermal systems. The paper begins with the state of the art in PVT systems, covering types, applications, recent developments, and more. It then presents a detailed analysis of ANN models, including the General Regression Neural Network (GRNN), Elman Neural Network (ENN), Radial Basis Function Network (RBFN), Multilayer Perceptron (MLP), and Adaptive Neuro-Fuzzy Inference Systems (ANFIS). Furthermore, the review highlights the roles that these models have played in enhancing PVT system performance in previous studies and includes a literature analysis to identify research gaps in this field. According to the literature, ANNs are valuable tools for predicting and optimizing the performance of PVT collectors; however, further exploration of alternative ANN models in novel PVT designs, combined with optimization algorithms, is necessary.

Accurately forecasting the demand for air passengers is vital for the aviation industry to formulate appropriate management strategies. Decomposition ensemble learning has attracted much attention from researchers of this problem because it is an effective way to improve forecasting accuracy. In contrast to common ways of generating ensemble forecasts, such as artificial intelligence and linear addition, our study employs the Choquet fuzzy integral. The Choquet integral is effective regardless of the training sample size and it uses a nonadditive fuzzy measure to explain the influence of the inputs on air passenger demand. Data on monthly air passenger flows from major airports in Taiwan were used to assess the effectiveness of the proposed decomposition ensemble models using the Choquet fuzzy integral to generate ensemble forecasts. The results in terms of level and directional forecasting accuracy showed that the proposed models— especially those that integrated smoothing (LOESS) (STL) and radial basis function network with the Choquet integral—significantly outperformed single (non-ensemble) forecasting models and the benchmark models considered.

ABSTRACT This article presents the adaptive sliding mode method for the gun elevation system of the tank operating under model uncertainties and external disturbances. The proposed controller combines optimal control design, the adaptive compensation mechanism using the radial basis function (RBF) neural network, while integrating the sliding mode control (SMC) law to enhance robustness and trajectory tracking accuracy. The RBF network is used to estimate and compensate for unknown nonlinear components and dynamic uncertainties in real time, and the SMC law is incorporated to ensure robustness and force the system output to accurately follow the desired trajectory. The control strategy is synthesized to meet strict performance requirements under complex real‐world operating conditions. Simulation studies conducted in Matlab evaluate the controller's effectiveness. The results demonstrate that the proposed method achieves accurate trajectory tracking, strong disturbance rejection, and improved robustness, confirming its potential for practical military applications.

The growing demand for energy and the environmental impacts of fossil fuels have driven the search for sustainable alternatives such as biodiesel. Castor oil stands out as a promising non-edible feedstock but requires optimization strategies to overcome challenges in its conversion to biodiesel. This study developed a predictive model to determine the optimal parameters for homogeneous alkaline or acid transesterification of castor oil, aiming to maximize fatty acid methyl ester (FAME) yield. A dataset of 406 operating conditions from the literature was used to train and evaluate six models: Multilayer Perceptron with logistic sigmoid activation (MLP-logsig), hyperbolic tangent activation (MLP-tansig), Radial Basis Function network (RBF), hybrid RBF + MLP, Random Forest (RF), and Adaptive Neuro-Fuzzy Inference System (ANFIS). The MLP-tansig achieved the best performance in training, validation, and testing (R > 0.98). However, when combined with a Genetic Algorithm (GA), it generated infeasible parameters. Conversely, the RBF + GA combination yielded results consistent with the literature: molar ratio 19.35:1, alkaline catalyst 1.13% w/w, temperature 50 °C, reaction time 70 min, and stirring speed 548.32 rpm, achieving 100% FAME yield. This approach reduces the need for extensive experimental testing, offering a cost- and time-efficient solution for optimizing biodiesel production.

A parametric machine learning tool with many applications is the radial basis function (RBF) network, which has been incorporated into various classification and regression problems. A key component of these networks is their radial functions. These networks acquire adaptive capabilities through a technique that consists of two stages. The centers and variances are computed in the first stage, and in the second stage, which involves solving a linear system of equations, the external weights for the radial functions are adjusted. Nevertheless, in numerous instances, this training approach has led to decreased performance, either because of instability in arithmetic computations or due to the method’s difficulty in escaping local minima of the error function. In this manuscript, a three-stage method is suggested to address the above problems. In the first phase, an initial estimation of the value ranges for the machine learning model parameters is performed. During the second phase, the network parameters are fine-tuned within the intervals determined in the first phase. Finally, in the third phase of the proposed method, a local optimization technique is applied to achieve the final adjustment of the network parameters. The proposed method was evaluated on several machine learning models from the related literature, as well as compared with the original RBF training approach. This methodhas been successfully applied to a wide range of related problems reported in recent studies. Also, a comparison was made in terms of classification and regression error. It should be noted that although the proposed methodology had very good results in the above measurements, it requires significant computational execution time due to the use of three phases of processing and adaptation of the network parameters.

Objectives: The main objective of this study was to develop a novel and intelligent framework by integrating the Generative Adversarial Network (GAN) based data augmentation technique with ensembled Radial Basis Functions Networks (RBFNNs). The goal was to address the challenges of scarcity of training data and false positive cases and efficiently detect zero-day intrusion attacks in real-time with higher precision and accuracy. Methods: This study has proposed a novel framework for developing an IDS model, which was designed by integrating an ensemble of Radial Basis Function Neural Networks (RBFNNs) with Generative Adversarial Network (GAN) based data augmentation. This model was trained and tested with popular benchmark real-world datasets, augmented with GAN-based synthetic data, and outperformed traditional classification models. Findings: The proposed model achieved an accuracy of 98.7% and under 1.2% false positive rate (FPR) for zero-day attack detection, and demonstrated superior performance, as compared to standalone classifiers like Random Forest, SVM, CNN, and standard RBFNN. Novelty: The results suggested that ensembled learning along with generative data augmentation can help in building adaptable, intelligent and resilient IDS systems. Keywords: Zero-day attack, Cloud Security, Resilient IDS, Ensemble Learning, GAN Data Augmentation

Accurate estimation of scour depth around bridge piers remains a challenging task due to the complex interaction between flow hydraulics and sediment dynamics; however, it is vital to safeguard bridge stability and reduce economic and human losses. Empirical relations are insufficient to satisfactorily simulate this very complex phenomenon. This study proposes intelligent models for the estimation of scour depth around bridge piers using three different modelling approaches, namely Multilayer Perceptron Artificial Neural Networks (MLP-ANN), Radial Basis Function Networks (RBFN), and Multigene Genetic Programming (MGGP). In addition, dimensional analysis was used to reach dimensionless quantities, simplifying the complex relations and also improving the accuracy of the models. The ANN model achieved the highest accuracy, with an R² value of 0.94, an RMSE of 0.032, and a WI of 0.97, indicating an excellent alignment with the observed data. The MGGP model yielded an R² of 0.91, demonstrating balanced performance in multiple statistical metrics. In contrast, the Basis Radial Function (BRF) model, although robust, employed a more conservative estimation approach, with an R² of 0.86, and exhibited limited sensitivity to extreme values. The results of the sensitivity analysis revealed that the bedload transport rate, dimensionless time, and depth-to-width ratio are critical parameters in scour depth calculations, which tangibly confirms the ability of ANNs and dimensional analysis to improve anti-scouring in designs and maintenance, and reduce the failure risk of structures. The findings highlight the effectiveness of machine learning in enhancing hydraulic prediction and improving the resilience of bridge design.

Abstract This study investigates the application of established open-source machine learning tools, specifically CatBoost, XGBoost, LightGBM, and TensorFlow, which are based on Forest and Radial Basis Function Networks, to predict and analyze the structural behavior of concrete arch dams. Utilizing the Karun-I dam as a case study, the research assesses the performance of various machine learning frameworks. The results demonstrate that Random Forest-based methods achieve superior prediction accuracy and computational efficiency in comparison to Radial Basis Function Networks. Additionally, the analysis emphasizes the critical influence of lake levels as the primary factor impacting dam displacement, as revealed through feature importance evaluation. Overall, this research underscores the promising potential of machine learning in enhancing structural health monitoring for large dams, offering significant insights that contribute to the improvement of safety measures and operational efficiency in dam management.

Sustainable environmental education: Some machine learning algorithms in the classification of sustainable environmental attitudes.

Directional drilling often encounters challenges such as eccentric annulus conditions caused by the weight of the drill string and oscillations, compounded by gravity-induced cuttings accumulation that obstructs flow and impedes drilling processes due to inefficient hole cleaning. This study focuses on addressing these issues by developing machine learning (ML) models to predict cuttings concentration (CA) in eccentric deviated wells, aiming to enhance predictive accuracy and optimize hole-cleaning operations. The research employs multiple ML algorithms including back propagation neural network (BPNN), radial basis function network (RBFN), and support vector machine (SVM). Models are trained using comprehensive field data from six deviated wells in the Gulf of Suez, Egypt, with inputs comprising rheological properties, drilling operation parameters, cutting transport velocity ratio (VTR), and carrying capacity index (CCI). The models undergo rigorous validation to ensure robustness and accuracy, employing both internal validation techniques to avoid overfitting and extensive testing across varying degrees of eccentricity. The developed RBFN model demonstrated superior performance compared to existing empirical and fuzzy logic models, achieving a relation coefficient (R) of 0.993 and an average absolute error (AAE) of 1.18 at an eccentricity degree (ε) of 0.5. In further validation within neighboring test wells, the RBFN model accurately predicted CA across different eccentricities, showing high reliability with R-values of 0.984, 0.978 and 0.971 and AAE-values of 1.1, 1.4 and 1.7 for = 0, 0.4 and 0.8, respectively. Sensitivity analyses confirmed the critical influence of VTR and CCI, with their impact most pronounced at the highest eccentricity tested. This study presents a significant advancement in drilling technology by integrating advanced ML methodologies to improve the monitoring and optimization of hole-cleaning efficiency in deviated wells. The novel application of these sophisticated models offers a promising solution to real-time challenges in drilling operations, enhancing efficiency and reducing operational risks associated with eccentric deviated wells. Incorporating ML models into routine drilling operations can potentially transform standard practices, making this approach a valuable asset in the field of petroleum engineering.

This paper proposes a Recurrent Non-linear Autoregressive Network with Exogenous inputs (RNARXNet) for Solar Power Forecasting (SPF) using time series data. Initially, input time series data is acquired from the Solar Power Generation dataset. Next, technical indicators such as Stochastic Oscillator (STOCH), Aroon (AR), Williams % R (WillR), Time Series Forecast (TSF), Moving Average Convergence Divergence (MACD), and Rainbow Moving Average are extracted. Afterwards, feature selection is done based on Hamming distance. Lastly, SPF is carried out utilizing RNARXNet which is newly modelled by integrating the Recurrent Radial Basis Function network (RRBFN) with a Non-linear Autoregressive Network with Exogenous inputs (NARX). The model obtained a normalized Mean Absolute Error (MAE) of 0.160, normalized Mean Square Error (MSE) of 0.352, normalized R 2 of 0.933, and normalized Root Mean Square Error (RMSE) of 0.594. These values show that RNARXNet is reliable and effective for precise SPF.

Cloud Computing (CC) has gained huge attention from both public and private organisations as it offers flexibility and pay-per-use-based services for users. The privacy and security of the users are affected due to the open and distributed nature of the cloud. The CC poses several risks due to the rise of intrusion attacks and recognising them is a crucial aspect in the internet world. Thus, the Intrusion Detection System (IDS) is the most widely used technique for detecting attacks in the cloud. Hence, this research proposes Dense_Deep Stacked AutoencoderNet (Dense_DeepSANet) for detecting intrusion. Initially, the cloud simulation is done and from the specified dataset, the acquisition of input data is made. Subsequently, the [Formula: see text]-score normalisation is used for normalising the data in the pre-processing step. Additionally, feature fusion is performed using Radial Basis Function Networks (RBF) with Lorentzian similarity to merge similar data. After that, the oversampling method concerning Synthetic Minority Over-sampling Technique (SMOTE) is used for augmenting the data to avoid overfitting. Finally, intrusion detection is done based on Dense_DeepSANet. Moreover, Dense_DeepSANet is modelled by combining the Deep Stacked Autoencoder (DSA) and Dense Convolutional Network (DenseNet). The datasets used for testing the devised approach are the ISCX NSL-KDD dataset, KDD Cup 1999 dataset and Cloud Intrusion Detection Dataset (CIDD). The experimental result shows that Dense_DeepSANet achieved superior results in terms of offering accuracy, Precision, Recall and [Formula: see text]1-score with values of 90.60%, 91.40%, 92.30% and 91.90% for the KDD Cup 1999 dataset.

Adaptive RBF network based sliding mode controller with novel switching term for hovering deep sea mining vehicle trajectory tracking

Radial basis function neural networks (RBFNNs) are widely applied due to their rapid modeling capabilities and efficient learning performance. However, when dealing with high-dimensional data, RBFNNs encounter two critical limitations: the hidden layer responses using Gaussian kernels suffer from ineffective activation and numeric underflow; and the estimation of output layer weights typically involves tedious parameter tuning and inefficient loading of high-dimensional feature matrices. To overcome these challenges, we first propose a dimensionality-adaptive Gaussian kernel function (DAGKF) equipped with a novel width adjustment mechanism that flexibly mitigates the numerical difficulties inherent in high-dimensional spaces. Moreover, to avoid processing entire feature matrices simultaneously, we introduce a multioutput coordinate descent (MOCD) algorithm that enables parallel computation across multioutput systems. Building upon MOCD, we further develop the joint residual MOCD (JRMOCD) algorithm, which incorporates a joint residual criterion for more effective weight estimation. The convergence of the JRMOCD algorithm is rigorously proven. Extensive experiments demonstrate the superior performance of the proposed methods, particularly in high-dimensional settings.

The rapid advancement of key technologies such as Artificial Intelligence (AI), the Internet of Things (IoT), and edge-cloud computing has significantly accelerated the transformation toward smart industries across various domains, including finance, manufacturing, and healthcare. Edge and cloud computing offer low-cost, scalable, and on-demand computational resources, enabling service providers to deliver intelligent data analytics and real-time insights to end-users. However, despite their potential, the practical adoption of these technologies faces critical challenges, particularly concerning data privacy and security. AI models, especially in distributed environments, may inadvertently retain and leak sensitive training data, exposing users to privacy risks in the event of malicious attacks. To address these challenges, this study proposes a privacy-preserving, service-oriented microservice architecture tailored for intelligent Industrial IoT (IIoT) applications. The architecture integrates Differential Privacy (DP) mechanisms into the machine learning pipeline to safeguard sensitive information. It supports both centralised and distributed deployments, promoting flexible, scalable, and secure analytics. We developed and evaluated differentially private models, including Radial Basis Function Networks (RBFNs), across a range of privacy budgets (varepsilon), using both real-world and synthetic IoT datasets. Experimental evaluations using RBFNs demonstrate that the framework maintains high predictive accuracy (up to 96.72%) with acceptable privacy guarantees for budgets varepsilon ge 0.5. Furthermore, the microservice-based deployment achieves an average latency reduction of 28.4% compared to monolithic baselines. These results confirm the effectiveness and practicality of the proposed architecture in delivering privacy-preserving, efficient, and scalable intelligence for IIoT environments. Additionally, the microservice-based design enhanced computational efficiency and reduced latency through dynamic service orchestration. This research demonstrates the feasibility of deploying robust, privacy-conscious AI services in IIoT environments, paving the way for secure, intelligent, and scalable industrial systems.

Electric rope shovels play a critical role in open-pit mining, where their automation and operational efficiency directly affect productivity. This paper presents a LiDAR-based relative positioning method to determine the spatial relationship between the ERS and mining trucks. The method utilizes dynamic DBSCAN for noise removal and RANSAC for truck edge detection, enabling robust and accurate localization. Leveraging this positioning data, a time-optimal trajectory planning strategy is proposed specifically for autonomous swing motion during the unloading process. The planner incorporates velocity and acceleration constraints to ensure smooth and efficient movement, while obstacle avoidance mechanisms are introduced to enhance safety in constrained excavation environments. To execute the planned trajectory with high precision, a neural network-based sliding-mode controller is designed. An adaptive RBF network is integrated to improve adaptability to model uncertainties and external disturbances. Experimental results on a scaled-down prototype validate the effectiveness of the proposed positioning, planning, and control strategies in enabling accurate and autonomous swing operation for efficient unloading.

Improved Adaptive Neural Control for USV-UAV via Gaussian Filter and Sigmoid RBF Networks

Constructing RBF network based on interference robust projected gradient

Objectives: To compare time series prediction-based deep learning techniques with Numerical Weather Prediction (NWP) to identify the optimal model for weather forecasting based on accuracy and consistency. Methods: The study utilizes a weather dataset collected from Kaggle, containing meteorological parameters such as temperature, humidity, and wind speed. Deep learning models such as the Radial Basis Function Network, the Convolutional Neural Network, the Recurrent Neural Network, and the Long Short-Term Memory have been developed and trained using the dataset for time-series forecasting. The models have been trained and evaluated using standard performance metrics, MAE, MSE, RMSE, and R2. Their predictions have been compared with a traditional Numerical Weather Prediction model. Findings: An evaluation of NWP and the LSTM model demonstrates that the LSTM provides significantly better accuracy in weather predictions. The mean absolute error (MAE) for NWP model is 1.198, while the LSTM achieves a lower MAE of 0.734, indicating that the forecasts from the LSTM model are closer to the actual values on average. Correspondingly, the Mean Squared Error (MSE) decreases from 2.255 in the NWP model to 1.237 in the LSTM, and the Root Mean Squared Error (RMSE) reduces from 1.502 to 1.112, indicating a decrease in prediction errors. Most notably, the R2 (coefficient of determination) improves from 0.975 in the NWP model to 0.985 in the LSTM, showing that the LSTM model accounts for 98.5% of the variability in the weather data, compared to 97.5% with the NWP model. These results illustrate that the LSTM model surpasses the traditional NWP approach in terms of accuracy and reliability for weather forecasting. Novelty: The research offers a comparative evaluation of deep learning models, the CNN, RNN, LSTM, and RBFN, for weather forecasting using both numerical climate data and satellite images, offering an inclusive approach rarely addressed in existing studies. Keywords: CNN, RNN, LSTM, RBFN, NWP, Weather Forecasting