With the rapid development of Industrial Internet of Things (IIoT), the category and quantity of industrial equipment will increase gradually. For the centralized monitoring and management of numerous and multivariate equipment in the intelligent manufacturing process, the equipment categories shall be identified first. However, manual labeling electrical equipment needs high costs. For the purpose of recognizing electrical equipment accurately in manufacturing system, this study adopts the Recurrent Neural Network (RRN) Long Short-Term Memory (LSTM) to build a NonIntrusive Load Monitoring (NILM) system, and combines edge computing to implement parallel computing to practice the effect of power equipment identification. Considering the practical popularity, the fairly priced low frequency Smart Meter is used to collect appliance data. According to proposed optimal adjustment strategy of parameter model, the average random recognition rate can be achieved 88%, the average recognition rate of the continuous data of a single electrical equipment can be achieved 83.6%.
25 April 2019 in IEEE Transactions on Industrial Informatics
Artificial Intelligence (AI) is one of the important technologies for industrial applications but it needs a lot of computing resources and sensing data to support. Therefore, big data transmission is a challenge for current network architectures. In order to have high-performance computing requirements, this study proposes an emerging network architecture that combines edge computing and cloud computing to reduce the transmission of useless data and solve bottleneck problems. Moreover, we define the resource allocation problem about multiple remote radio heads (RRHs) and multiple Base Band Units pools (BBU pools) in the cloud radio access network (C-RAN) for 5G. The Long Short-Term Memory (LSTM) is used to predict dynamic throughput and GA-based Resource Allocation Algorithm (GARAA) is used to optimize resource allocation. The simulation results represented that the proposed mechanism can achieve high resource utilization and reduce power consumption.
This study proposed a children’s digital art ability training system with artificial intelligence-assisted learning (AI-assisted learning), which was designed to achieve the goal of improving children’s drawing ability. AI technology was introduced for outline recognition, hue color matching, and color ratio calculation to machine train students’ cognition of chromatics, and smart glasses were used to view actual augmented reality paintings to enhance the effectiveness of improving elementary school students’ imagination and painting performance through the diversified stimulation of colors. This study adopted the quasi-experimental research method and designs the pre-test and post-test for different groups. The research subjects are the Grade 4 students of an elementary school in Taitung City, Taiwan. The test tools included an imagination test and an evaluation of painting performance ability. The test results of a total of 30 students before and after the experiment included the experimental group that received the children’s digital art ability training system with AI-assisted learning and 30 students in the control group that had not received the teaching were analyzed by covariance. These results were supplemented by the description and interpretation of student feedback, teachers’ reflection notes, and other qualitative data to understand the performance of the students in the experimental group in terms of imagination and painting performance.
The limited battery capacity of sensor nodes (SNs) is one of the challenges for achieving sustainable precision agriculture in 5G. When SNs run out of energy, the data center will not be able to grasp the information of the environment, and the corresponding decisions and actions will not be performed. To solve this problem, this study presents a framework for charging SNs and collecting sensing data using unmanned aerial vehicles. The problems of cluster head (CH) election and path planning are considered at the same time to maximize charging efficiency in a way that the lifetime of SNs can be prolonged. Moreover, ${K}$K-means-based CH and charging position selection algorithm and ACO-based charging path-planning algorithm are proposed to optimize charging planning of the 3-D environment. Simulation results show that the proposed mechanism can minimize the number of CHs and maximize the remaining energy of SNs.
The rise of Distributed Denial of Service (DDoS) attacks on the internet has necessitated the development of robust and efficient detection mechanisms. DDoS attacks continue to present a significant threat, making it imperative to find efficient ways to detect and prevent these attacks promptly. Traditional machine learning approaches raise privacy concerns when handling sensitive data. In response, federated learning has emerged as a promising paradigm, allowing model training across decentralized devices without centralizing data. However, challenges such as the non-IID (Non-Independent and Identically Distributed) problem persist due to data distribution imbalances among devices. In this research, we propose personalized federated learning (PFL) as a solution for detecting DDoS attacks. PFL preserves data privacy by keeping sensitive information localized on individual devices during model training, thus addressing privacy concerns that are inherent in traditional approaches. In this paper, we propose federated learning with DBSCAN clustering (FedDB). By combining personalized training with model aggregation, our approach effectively mitigates the common challenge of non-IID data in federated learning setups. The integration of DBSCAN clustering further enhances our method by effectively handling data distribution imbalances and improving the overall detection accuracy. Results indicate that our proposed model improves performance, achieving relatively consistent accuracy across all clients, demonstrating that our method effectively overcomes the non-IID problem. Evaluation of our approach utilizes the CICDDOS2019 dataset. Through comprehensive experimentation, we demonstrate the efficacy of personalized federated learning in enhancing detection accuracy while safeguarding data privacy and mitigating non-IID concerns.
The rapid evolution of wireless communication systems necessitates advanced handover mechanisms for seamless connectivity and optimal network performance. Traditional algorithms, like 3GPP Event A3, often struggle with fluctuating signal strengths and dynamic user mobility, leading to frequent handovers and suboptimal resource utilization. This study proposes a novel approach combining Federated Learning (FL) and Long Short-Term Memory (LSTM) networks to predict Reference Signal Received Power (RSRP) and the strongest nearby Reference Signal Received Power (RSRP) signals. Our method leverages FL to ensure data privacy and LSTM to capture temporal dependencies in signal data, enhancing prediction accuracy. We develop a dynamic handover algorithm that adapts to real-time conditions, adjusting thresholds based on predicted signal strengths and historical performance. Extensive experiments with real-world data show our dynamic algorithm significantly outperforms the 3GPP Event A3 algorithm, achieving higher prediction accuracy, reducing unnecessary handovers, and improving overall network performance. In conclusion, this study introduces a data-driven, privacy-preserving approach that leverages advanced machine learning techniques, providing a more efficient and reliable handover mechanism for future wireless networks.