Table of Contents (Volume #10, Issue #4)

         Cover Page (Volume #10, Issue #4)

         Editorial (Volume #10, Issue #4)

Paper #1

Using PER-DQN for Intelligent Control of Dynamic Warehousing and Sorting Systems

Xizhuo Han and Chuanchuan Teng

Abstract: This paper aims to address the lag problem of traditional control strategies caused by order fluctuations and equipment coordination complexity in dynamic warehousing and sorting systems, specifically addressing the limitations of the Deep Q-Network (DQN) algorithm, notably its poor sample efficiency and instability in dynamic environments. It uses PER-DQN (Prioritized Experience Replay-DQN) to build an efficient and stable intelligent control method, which enhances timing adaptability through dynamic replanning and a time-varying reward mechanism to solve existing problems. First, a 10-dimensional state vector is constructed, covering real-time orders and robot status information, and a dynamic environment model is established using the discrete grid division method to achieve a comprehensive perception of the system's operating status. Subsequently, a structured action space was designed that includes task assignment priorities and robot path directions to support multi-dimensional collaborative control. Then, a multi-objective composite reward function is proposed to comprehensively consider the task completion efficiency, dynamically adjust the optimization direction through time-varying weights, and accelerate the convergence of the policy network. Finally, a dual-network training architecture combining the PER mechanism and Double DQN is adopted to improve the utilization of key experience and suppress Q-value overestimation. An online learning mechanism is applied to achieve incremental strategy updates, ensuring that the system can maintain stable performance in sudden order scenarios. Experiments show that PER-DQN has an average delay of 14.35 seconds for small-batch orders, 24.21 seconds for mixed orders, and 34.82 seconds for large-batch orders. PER-DQN also has a path conflict rate of only 1.2% and a sorting accuracy of 99.1%. In the robustness test, PER-DQN's device utilization increases by 25.4% and 27.8% compared to DQN under device failure and high load pressure conditions, respectively. This study verifies the efficiency and robustness of PER-DQN, which improves task scheduling accuracy and resource coordination capabilities through prioritized experience replay and structured action space design.

Keywords: Deep Q-Network; Intelligent Control; Multi-objective Composite Reward Function; Prioritized Experience Replay; Warehousing and Sorting System.

Download Full Text

Paper #2

AGI-inspired UAV Cluster Communication Self-Organization Framework: Multi-Agent Learning and Network Optimization

Fan Zhou

Abstract: Unmanned Aerial Vehicle (UAV) cluster communication networks play a vital role in Space-Air-Ground Integrated Networks (SAGIN), offering flexible deployment and high mobility for emergency communications, disaster relief, and edge computing. However, traditional approaches face significant limitations in dynamic environments, often focusing on single objectives, lacking adaptive capabilities, and failing to achieve global coordination across multiple UAVs. To address these challenges, this paper presents Artificial General Intelligence-inspired Multi-Agent Knowledge Transfer (AGI-MAKT), a novel self-organizing framework for UAV cluster communication networks. Our framework overcomes existing limitations through several key innovations: a multi-modal perception system that enables comprehensive environmental understanding by fusing heterogeneous data sources; a hierarchical multi-agent learning architecture that reduces decision-making complexity while enhancing coordination; an adaptive multi-objective optimization approach that dynamically balances coverage, energy efficiency, and throughput based on contextual requirements; and efficient knowledge transfer mechanisms that accelerate learning in new environments. Comprehensive experiments demonstrate that AGI-MAKT consistently outperforms representative baselines across the evaluated performance metrics while maintaining strong robustness under challenging conditions. This research establishes a foundation for next-generation aerial networks capable of self-organization and adaptation in complex, resource-constrained environments with limited human supervision.

Keywords: Adaptive Multi-Objective Optimization; Artificial General Intelligence; Knowledge Transfer; Multi-Agent Learning; Self-Organization; Space-Air-Ground Integrated Networks; UAV Communication Networks.

Download Full Text

Paper #3

Small Cell Integration into Urban Furniture for Smart City Development

Mohammed Sh. Ali, Mohammed Othman Hasan, Rana H. A. Zubo and Raed A. Abd-Alhameed

Abstract: Fifth-generation networks in mid-band spectrum need dense small-cell infrastructure. The binding constraint is site cost, not radio performance. Urban furniture removes part of it: street lighting columns, bus shelters and traffic signals already offer mounting height, mains power and, increasingly, fibre. This paper asks which of those assets an operator with a fixed capital budget should equip. We formulate a budget-aware greedy site-selection algorithm over a discretised service area and evaluate it against seven alternatives: random selection, a snapped lattice, k-means clustering, cost-unaware greedy maximum coverage, simulated annealing, a genetic algorithm and binary particle swarm optimisation. The coverage objective is interference-aware over a constrained, heterogeneously priced candidate set, a combination the surveyed studies do not make. Every selected cell that does not serve a test point interferes there, and propagation follows the Third Generation Partnership Project Urban Micro street-canyon model with line-of-sight and non-line-of-sight states and per-link shadow fading. Demonstrated empirically rather than assumed, the objective is then neither monotone nor submodular, so the classical constant-factor guarantee does not apply and the method is a heuristic. Across fifty paired Monte-Carlo realisations under matched expenditure it exceeds all six remaining baselines at every threshold, by margins whose confidence intervals exclude zero after correction, and it matches simulated annealing at the tightest threshold on about one thirteenth of the computation while trailing it by 0.54 and 0.32 percentage points at the looser two. Under a common budget it comes within 1.2 percentage points of the best baseline on a third to just over a half of the capital. On real urban-furniture inventories extracted from OpenStreetMap for three Swiss city centres, that matched-spend ordering holds and the advantage over cost-unaware selection is largest on the densest inventory, but under a common budget cost-unaware selection is ahead in all three cities. Runs at 28 and 39 GHz show the interference-driven stopping point disappearing as cells shrink, and a probabilistic reading of coverage matches the threshold metric within one percentage point while showing the reliably covered area to be far smaller. We release the simulator and the extracted inventories.

Keywords: Budget-constrained optimisation; Interference modelling; Network densification; Small cell deployment; Street furniture; Submodular maximisation; Urban infrastructure planning.

Download Full Text

Paper #4

Scalable AI-Based Framework for Anomaly Detection and MQTT Traffic Optimisation in IoT Systems

Emran Aljarrah, Bilal Bataineh, Saleem Alzoubi, Md Nasiruzzaman and Mahdi H. Miraz

Abstract: Internet of Things (IoT) networks are becoming increasingly prevalent, but facing serious challenges with regards to data integrity and communication security, due to resource constraint IoT devices being used in large distributed networks. This paper proposes a scalable AI-based anomaly detection framework and efficient traffic optimisation for Message Queuing Telemetry Transport (MQTT) messages sent through IoT systems. Both real-time data collected from the IOT nodes as well as the CICIDS2017 dataset have been utilised for this research. Due to biases and inaccuracies inherent in the data, a Conditional Generative Adversarial Network (CGAN) model was used for pre-processing the data. A novel selective feature process has also been presented, which first performs Principal Component Analysis (PCA), Genetic Algorithm and Univariate Statistical Test, and then perform dimensionality reduction using the Extreme Learning Machine (PCA-GA-UST-ELM) model. The traffic optimisation process for the overall IoT network uses the integration of a Deep Neural Network (DNN) model and an MQTT parsing engine, to facilitate processing of network traffic through MQTT. The anomalous threats presented in the IoT nodes were classified using the Bidirectional Gated Recurrent Unit (Bi-GRU) model to improve threats detection accuracy of the proposed DNN models. Finally, the routing process of messages has been optimised using a tail hybridisation method of GTLPSO (Game Theory, Linear Programming, Stochastic Problems). The performance of the framework has been evaluated in terms of accuracy, precision, latency, Prediction time and throughput, for varying numbers of epochs and IoT nodes.

Keywords: Anomaly detection; CICIDS2017 dataset; IoT systems; MQTT; Scalable AI.

Download Full Text

Paper #5

Automatic Brain Tumor Detection and Segmentation from MRI Images

Zainab Ghayyib Abdul Hasan

Abstract: A brain tumor is an abnormal growth of tissue in the central spine or brain that can cause the brain to become more pressured and interfere with normal brain function. Early detection of a brain tumor is extremely challenging for medical professionals. MRI images can be greatly impacted by noise and other environmental disruptions. As a result, diagnosing the tumor and its causes becomes challenging for medical professionals. So, we present a methodology to detect a brain tumor and segment it from medical images. According to the studies, the region growing algorithm initializes the seed point either manually or semi-manually, which has an impact on the segmentation outcome. Therefore, we suggested an improved region growing algorithm for the automatic seed point initialization. This paper presents an automated and integrated framework for the highly accurate detection and segmentation of brain tumors. The suggested methodology starts with image pre-processing to enhance image quality and reduce noise. The next step involves using the skull stripping developed algorithm, which is specifically made for precise brain mask extraction. The generated mask is crucial in automatically determining the tumor region's seed point, which removes the need for human intervention and reduces the variability of the results. Lastly, the entire tumor region is segmented using a region-growing algorithm that starts with the identified seed. Experimental results demonstrate that our proposed technique achieves robust performance and accurate results in detecting and segmenting brain tumor boundaries in MRI images, making it a promising tool for supporting clinical decision-making.

Keywords: Brain MRI Image; Brain Tumor; Detection; Morphological Filtering; Region Growing; Segmentation; Skull Stripping.

Download Full Text