The proliferation of immersive services, including virtual reality/augmented reality, holographic content, and the metaverse, has led to an increase in the complexity of communication networks, and consequently, the complexity of network management. Recently, digital twin network technology, which applies digital twin technology to the field of communication networks, has been predicted to be an effective means of managing complex modern networks. In this paper, a digital twin network data pipeline architecture is proposed that demonstrates an integrated structure for flow within the digital twin network and network modeling from a data perspective. In addition, a network traffic modeling technique using data feature extraction techniques is proposed to realize the digital twin network, which requires the use of massive streaming data. The proposed method utilizes the data generated in the OMNeT++ environment and verifies that the learning time is reduced by approximately 25% depending on the feature extraction interval, while the accuracy remains similar.

A new adaptive neural network (NN) control concept is proposed with proof of stability properties. The NN learns the plant dynamics with online training, and then combines this with background learning from previously recorded data, which can be advantageous to the NN adaptation convergence characteristics. The network adaptation characteristics of the new combined online and background learning adaptive NN is demonstrated through simulations.

The computing in the network (COIN) paradigm is a promising solution that leverages unused network resources to perform tasks to meet computation-demanding applications, such as the metaverse. In this vein, we consider the partial computation offloading problem in the metaverse for multiple subtasks in a COIN environment to minimize energy consumption and delay while dynamically adjusting the offloading policy based on the changing computational resource status. The problem is NP-hard, and we transform it into two subproblems: the task-splitting problem (TSP) on the user side and the task-offloading problem (TOP) on the COIN side. We model the TSP as an ordinal potential game and propose a decentralized algorithm to obtain its Nash equilibrium (NE). Then, we model the TOP as a Markov decision process and propose the double deep Q-network (DDQN) to solve for the optimal offloading policy. Unlike the conventional DDQN algorithm, where intelligent agents sample offloading decisions randomly within a certain probability, the COIN agent explores the NE of the TSP and the deep neural network. Finally, the simulation results reveal that the proposed model approach allows the COIN agent to update its policies and make more informed decisions, leading to improved performance over time compared to the traditional baseline.

Accurately predicting power consumption is essential to ensure a safe power supply. Various technologies have been studied to predict power consumption, but the prediction of power consumption using deep learning models has been quite successful. However, in order to predict power consumption by utilizing deep learning models, it is necessary to find an appropriate set of hyper-parameters. This introduces the problem of complexity and wide search areas. The power consumption field should be accurately predicted in various distributed areas. To this end, a customized consumption prediction deep learning model is needed, which is essential for optimizing the hyper-parameters that are suitable for the environment. However, typical deep learning model users lack the knowledge needed to find the optimal values of parameters. To solve this problem, we propose a method for finding the optimal values of parameters for learning. In addition, the layer parameters of deep learning models are optimized by applying genetic algorithms. In this paper, we propose a hyper-parameter optimization method that solves the time and cost problems that depend on existing methods or experiences. We derive a hyper-parameter optimization plan that solves the existing method or experience-dependent time and cost problems. As a result, the RNN model achieved a 30% and 21% better mean squared error and mean absolute error, respectively, than did the arbitrary deep learning model, and the LSTM model was able to achieve 9% and 5% higher performance.

As electricity supply expands, it is essential for providers to predict and analyze consumer electricity patterns to plan effective electricity supply policies. In general, electricity consumption data take the form of time series data, and to analyze the data, it is first necessary to check if there is no data contamination. For this, the process of verifying that there are no abnormalities in the data is essential. Especially for power data, anomalies are often recorded over multiple time units rather than a single point. In addition, due to various external factors, each set of power consumption data does not have consistent data features, so the importance of pre-clustering is highlighted. In this paper, we propose a method using a CNN model using pre-clustering-based time series images to detect anomalies in time series power usage data. For pre-clustering, the performances were compared using k-means, k-shapes clustering, and SOM algorithms. After pre-clustering, a method using the ARIMA model, a statistical technique for anomaly detection, and a CNN-based model by converting time series data into images compared the methods used. As a result, the pre-clustered data produced higher accuracy anomaly detection results than the non-clustered data, and the CNN-based binary classification model using time series images had higher accuracy than the ARIMA model.