Wind turbine availability is becoming more crucial as the world’s electricity supply becomes increasingly dependent on them. The performance of wind energy conversion systems (WECS) is influenced by the control strategies, particularly through the incorporation of maximum power point tracking (MPPT) algorithms, which enable the maximum available power to be tracked. Unexpected faults in various system components could easily disrupt the designed controller’s performance. Therefore, research into fault-tolerant control with rapid fault detection, isolation, and successful control reconfiguration would be highly beneficial for WECS operation. To guarantee trajectory tracking, this study focuses on designing a control that uses the optimal torque algorithm to maximize the power from the available wind. Then, an active actuator fault-tolerant control is presented using the fault detection and estimation results offered by the observer. Therefore, the aim is to determine an additive control law to compensate for the fault effects.
Early lane change is a safety–critical driving action in autonomous vehicles based on several road accident records. Therefore, reliable prediction of such movements earlier is important for comfortable and safe operation of the autonomous driving system. Various existing methods depend on identifying driving actions that have already started, instead of forecasting maneuvers in advance. Moreover, traditional approaches did not evaluate the vital timings of driving actions that can actually provide more helpful data for decision-making in autonomous vehicles. In this research, Evolutionary Dual Self Attention Network (EDSANet) based early lane change prediction is presented for autonomous vehicles. Initially, the simulated video is obtained as an input, and thereafter, frames are extracted from the simulated video. After that, road layout information is extracted using Gated Fusion Attention Network (GFANet). Then, vehicle position is tracked using Deep Affinity Network (DAN). Lastly, the extracted road layout and tracked vehicle position are subsequently given as inputs to EDSANet for early lane change prediction. EDSANet is modelled by incorporating Evolutionary Attention-based Long Short-Term Memory (EA-LSTM) with Dual Self-Attention Network (DSANet) using harmonic analysis. The model predicts one of three maneuver classes, namely left lane change, right lane change, or straight, as output. Additionally, EDSANet has achieved loss, normalized Mean Bias Error (MBE), normalized Mean Squared Error (MSE), and normalized Root Mean Squared Error (RMSE) of 0.093, 0.533, 0.220, and 0.470, as well as Nash–Sutcliffe Efficiency (NSE) and R2 score of 0.681 and 0.803.
Accurate home energy forecasting is important for effective energy management in distribution systems. However, aggregated load forecasting cannot capture individual appliance’s usage patterns. In centralised forecasting, home energy consumption data has to be transferred to a central server. This may expose sensitive information, such as occupancy patterns and lifestyle habits. Therefore, this study proposes a privacy-preserving framework that integrates fully homomorphic encryption, federated learning, and multi-head attention sequential long short-term memory. The fully homomorphic encryption-federated learning method addresses the user data privacy concerns and the multi-head attention sequential long short-term memory minimizes the forecasting error. The proposed model exhibits a low average absolute error for electric vehicles, dryer, well pump, and water heater when compared with the recurrent and convolutional models. The proposed model achieved superior R^2 values, outperforming baseline models for the well pump (87.40
Multimodal biometric systems have gained increasing attention due to their ability to improve the reliability and security of user authentication. This study presents FI-VeT (Feature Fusion-based Iris-Finger Vein Vision Transformer), a lightweight multimodal biometric framework that combines iris and finger vein modalities through feature-level fusion using dual ViT-Tiny encoders. The iris images are preprocessed using segmentation, CLAHE enhancement, morphological operations, and normalization, while the finger vein images undergo Mini-ROI extraction, CLAHE enhancement, and normalization to enhance discriminative vascular patterns. The extracted modality-specific representations are fused and classified using a compact multilayer perceptron architecture. The proposed model was evaluated using a multimodal dataset constructed from the CASIA-Iris-Thousand and FV-USM datasets under controlled acquisition conditions. Since no publicly available paired iris and finger vein dataset exists, the study aligned 492 iris classes with 492 finger vein classes to enable multimodal feature-level fusion and evaluation. Experimental results showed that FI-VeT achieved an accuracy of 99.66
In Content-Based Image Retrieval (CBIR), a major problem for using deep learning for the retrieval process is the need for well-annotated training resources, which are costly and take more time for the annotation process. To overcome these issues, an advanced deep learning-based image retrieval model is implemented in this research work. The major contribution of this research is the development of a deep learning-based image retrieval model capable of retrieving visually similar images based on their content. The images required for the CBIR process are collected from standard databases. Next, the pre-processing step is conducted, where the collected input images undergo normalization process to standardize the intensity and quality of the input images. The pre-processed images are fed into the Attention-guided Query-Driven Vision Transformer with Graph Feature Fusion (AQ-ViTGFF), which effectively focuses on the most significant features within the image. To enhance feature presentation, the Graph Feature Fusion (GFF) module integrates these features to produce more comprehensive features. The major novelties of this proposed framework are the development of an Adaptive Learnable Attention Network (ALAN)-based similarity checking for the CBIR model, where the fused features are fed into an ALAN model to check the similarity between the query and reference images. The reliability of the suggested ALAN method is improved by optimizing the parameters of Learnable Attention Network (LAN) using the Leader-Driven Revolution Optimization Algorithm (LROA). Experimental results are presented to demonstrate the designed approach’s reliability in the image retrieval process. The precision rate of the developed LROA-ALAN-based image retrieval model is 90.83
