Arman Process Journal (APJ)

Abstract The explosive growth of Internet of Things (IoT) devices in home environments, coupled with the inherent limitations of these devices in terms of computing power, memory, and energy consumption, has increased the level of cyberattacks to an unprecedented level. Traditional home networks lack adequate security mechanisms to deal with emerging IoT-specific threats. This paper presents a comprehensive and multi-layered solution to enhance security in IoT-based home networks. The proposed approach, titled "SecHome-IoT", is composed of three main layers: (1) a deep learning-based anomaly detection layer (automated preprocessing and 1D convolutional neural network with long-term short-term memory), (2) a secure virtualization layer based on lightweight microservices (using hardware containers and critical path isolation), and (3) Dynamic and adaptive policy management layer (using adaptive-neural fuzzy inference system). Best hardware-software simulation on real-world datasets CICIDS2017, Bot-IoT and UNSW-NB15 along with implementation on Raspberry Pi platform and OpenWrt smart switches shows that the proposed method is superior to the previous ones (such as IoT-IDSA and Deep-STM) with an attack detection rate of 98.6% and a false positive rate of 1.4%. It has an improvement of 25% in security and performance metrics. This paper provides a roadmap for the security of IoT home networks by providing a comprehensive threat analysis, real-world and similar implementations.

Abstract Background and Objectives: With the advancement of vehicular networks and the increasing demand for accurate and timely data, challenges such as data retrieval delays and security concerns have garnered significant attention. Traditional cloud-based storage methods are unable to meet temporal and security requirements due to the considerable distance between vehicles and servers. Although edge computing offers a solution for reducing latency, it requires improvements due to limitations in storage and distributed management. Previous research has primarily focused on one aspect of optimization (reducing delay or enhancing security), with less attention paid to combining these two objectives. This paper aims to propose an innovative hybrid optimization model using deep learning and blockchain that considers both security and delay reduction. This is achieved by optimizing caching locations, storage, retrieval processes, and storing critical information on the blockchain, ensuring a scalable and flexible model adaptable to traffic changes and user demands.

Methods: The study population and sample include mobile vehicular networks (VANETs), considering edge servers and vehicular nodes. An LSTM model was used to predict traffic patterns and data popularity, while blockchain with a Proof of Authority (PoA) consensus mechanism and smart contracts was employed for secure data storage. Performance was evaluated based on delay, security, and scalability metrics, and compared with existing methods such as Tabu Search, CCS-AGP, and Random Caching in terms of delay and security.

Findings: The proposed model significantly reduced delay (by 10% to 30% compared to baseline methods). The use of blockchain introduced only an 8% additional delay while elevating security to a "very high" level. The system demonstrated stability and scalability under increasing numbers of nodes and data volume. Simulation results indicated that the combination of deep learning and blockchain achieves an optimal balance between performance and security.

Conclusion: The proposed model, integrating deep learning and blockchain, not only reduces delay but also ensures data security and integrity. This framework can serve as a foundation for developing intelligent systems in domains such as the Internet of Things (IoT), smart cities, and next-generation transportation.

Abstract Background and Objectives: Wind turbines as one of the means of producing electrical energy from renewable and clean energies have been the focus of many researchers. The discussion of turbine control in order to produce more power and its economical use against fossil fuels has challenged different control methods.

Methods: In the current research, the purpose of using intelligent fuzzy controllers is to improve the output power and stabilize it when necessary due to its robustness. For this purpose, the induction generator with two-way feeding and variable wind was modeled first, then phase controllers will be designed to separately control active and reactive powers, reduce interference and uncertainty effects. that we used the particle swarm algorithm and the best rules and fuzzy parameters of the intelligent fuzzy system based on deep learning to create the rule and interference system to better performance.

Findings: The comparison of the simulation results of intelligent fuzzy and PI controllers shows the better performance and efficiency of the fuzzy controller in terms of more stability, steady state error and less settling time than the PI controller used in the system. The performance accuracy of the fuzzy controller based on deep learning due to rule extraction and optimal PSO design using random forest algorithm for this system is suitable according to the obtained outputs and the system is controlled in less than 0.4 seconds.

Conclusion: Our integrated and hybrid algorithm shows the good performance due to accuracy and precision parameters, applying the deep learning in order to select the effective parameters on system design for rule extraction in fuzzy and create the decision making in PSO leads the novel way to approach the results.

Abstract In this article, a method is proposed for diagnosing children with Attention Deficit Hyperactivity Disorder (ADHD) using deep learning concepts and analyzing the correlation between segmented areas of functional brain images. The proposed method includes preprocessing medical images to remove distorted, noisy, incomplete, and problematic images, generating new functional brain images using autoencoders, which are applicable in artificial intelligence and deep learning, to assist in better analysis of medical images and address the limitations of medical image availability. It also involves segmenting images and creating separate networks to enhance the diagnostic accuracy of the condition and calculate the correlation between regions, ultimately leading to an effective diagnosis of ADHD in children. The conclusion of this study indicates that this combined method has a high capability for timely and accurate diagnosis of this condition in children and can serve as an effective tool in the field of child neurology and psychiatry.