With the rapid expansion of 5G networks, ensuring the security and integrity of data transmission becomes paramount. In this context, anomaly detection plays a critical role in identifying suspicious activities or deviations from normal behavior. This paper presents an approach to anomaly detection in 5G Non-IP Data Delivery (NIDD) datasets using Variational Autoencoders (VAE). We use VAE to build an unsupervised machine learning to enable the detection of unknown and previously unseen threats. The VAE is trained on 5G-NIDD datasets, capturing the intricate patterns and characteristics of normal network behavior. Subsequently, anomalies are detected by measuring the reconstruction error between the input data and the reconstructed output from the VAE. Experimental results demonstrate the efficacy of the VAE, showcasing its ability to accurately detect anomalies in 5G-NIDD datasets. This research contributes to the advancement of anomaly detection techniques in 5G networks, enhancing the security and reliability of data transmission in complex network environments.

This research focuses on the accurate prediction of renewable energy generation in microgrid systems using artificial intelligence (AI) techniques. The study compares and evaluates different AI models, including Artificial Neural Networks (ANN), Fuzzy Logic, and Adaptive Neuro-Fuzzy Inference Systems (ANFIS), for forecasting solar and wind power output. The models are trained and tested using real-world data, and their performance is assessed using metrics such as Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Sum of Squared Errors (SSE). The results demonstrate that the ANN model achieves the lowest RMSE and MAPE values for wind power prediction, while the Fuzzy Logic model performs well in predicting solar power generation. These findings indicate the effectiveness of AI techniques in accurately forecasting renewable energy output in microgrid systems. The proposed approach has implications for optimizing the utilization and integration of renewable energy sources, leading to more efficient and sustainable microgrid operations. Future research directions may involve exploring advanced deep-learning models and incorporating additional environmental factors to further enhance the accuracy and reliability of renewable energy power forecasting in microgrids.

Mobile networking in 4G and 5G remains vulnerable against fake base stations. A fake base station can inject and manipulate the radio resource control (RRC) communication protocol to disable the user equipment’s connectivity. To motivate our research, we empirically show that such a fake base station can cause an indefinite hold of the user equipment’s connectivity using our fake base station prototype against an off-the-shelf phone. To defend against such threat, we design and build an anomaly detection system to detect the fake base station threats. It detects any base station’s deviations from the 4G/5G RRC protocol, which supports both the connectivity provision case (all works well and the user receives connectivity) and the connection-release case (cannot provide connectivity at the time and thus releases connections). Our scheme based on unsupervised machine learning dynamically and automatically controls and sets the detection parameters, which vary with mobility and the communication channel, and utilizes greater information to improve its effectiveness. Using software-defined radios and srsRAN, we implement a prototype of our scheme from sensing to data collection to machine-learning-based detection processing. Our empirical evaluations demonstrate the detection effectiveness and adaptability; i.e., our scheme accurately detects fake base stations deviating from the set protocol in mobile scenarios by adapting its model parameters. Our scheme achieves 100% accuracy in static scenarios against the fake base station threats. If the dynamic control is disabled, i.e., not adapting to mobility and different channel environments, the accuracy drops to 65–76%, but our scheme adjusts the model via dynamic training to recover to 100% accuracy.