Network Performance Prediction Research Articles

Performance Evaluation of Graph Neural Network-Based RouteNet Model with Attention Mechanism

Graph representation is recognized as an efficient method for modeling networks, precisely illustrating intricate, dynamic interactions within various entities of networks by representing entities as nodes and their relationships as edges. Leveraging the advantage of the network graph data along with deep learning technologies specialized for analyzing graph data, Graph Neural Networks (GNNs) have revolutionized the field of computer networking by effectively handling structured graph data and enabling precise predictions for various use cases such as performance modeling, routing optimization, and resource allocation. The RouteNet model, utilizing a GNN, has been effectively applied in determining Quality of Service (QoS) parameters for each source-to-destination pair in computer networks. However, a prevalent issue in the current GNN model is their struggle with generalization and capturing the complex relationships and patterns within network data. This research aims to enhance the predictive power of GNN-based models by enhancing the original RouteNet model by incorporating an attention layer into its architecture. A comparative analysis is conducted to evaluate the performance of the Modified RouteNet model against the Original RouteNet model. The effectiveness of the added attention layer has been examined to determine its impact on the overall model performance. The outcomes of this research contribute to advancing GNN-based network performance prediction, addressing the limitations of existing models, and providing reliable frameworks for predicting network delay.

Development of Artificial Intelligence Image Classification Models for Determination of Umbilical Cord Vascular Anomalies.

The goal of this work was to develop robust techniques for the processing and identification of SUA using artificial intelligence (AI) image classification models. Ultrasound images obtained retrospectively were analyzed for blinding, text removal, AI training, and image prediction. After developing and testing text removal methods, a small n-size study (40 images) using fastai/PyTorch to classify umbilical cord images. This data set was expanded to 286 lateral-CFI images that were used to compare: different neural network performance, diagnostic value, and model predictions. AI-Optical Character Recognition method was superior in its ability to remove text from images. The small n-size mixed single umbilical artery determination data set was tested with a pretrained ResNet34 neural network and obtained and error rate average of 0.083 (n = 3). The expanded data set was then tested with several AI models. The majority of the tested networks were able to obtain an average error rate of <0.15 with minimal modifications. The ResNet34-default performed the best with: an image-classification error rate of 0.0175, sensitivity of 1.00, specificity of 0.97, and ability to correctly infer classification. This work provides a robust framework for ultrasound image AI classifications. AI could successfully classify umbilical cord types of ultrasound image study with excellent diagnostic value. Together this study provides a reproducible framework to develop AI-specific ultrasound classification of umbilical cord or other diagnoses to be used in conjunction with physicians for optimal patient care.

Construction of a Computer Network Performance Prediction Optimization Model Based on Machine Learning Algorithms

Computer networks have been widely used in critical infrastructure for businesses, schools, governments, and individuals. The development of computer networks is an irreversible historical trend. This article studies computer network performance prediction optimization models through machine learning, which can improve the intelligence and efficiency of network management. Existing research has made some progress in machine learning based network security and performance optimization. However, good results have not been achieved in real-time performance prediction and optimization. This article uses Long Short Term Memory Network (LSTM) to establish an effective predictive optimization model. Firstly, the data is processed, such as missing values, duplicates, and device time redundancy. Then, features are selected to select the corresponding input features of the model. Finally, the model is optimized to obtain the optimized model. The performance of the LSTM model is compared with the random forest (RF) and support vector machine (SVM) models. Model performance indicators include mean absolute error (MAE), coefficient of determination, F1 score, and model computation time. The experimental results show that the LSTM model has higher accuracy than RF and SVM models in all evaluation metrics. Meanwhile, the LSTM model has significant advantages in computational time compared to the other two models. The LSTM based computer network performance prediction optimization model can be optimized in terms of prediction accuracy, generalization ability, and computational efficiency. It has the ability to effectively process time series data. The LSTM model used for network performance prediction has a minimum average absolute error of 8% and a maximum coefficient of determination of 0.95. It achieves higher F1 scores and faster calculation speed. The results reflect the potential and practicality of the LSTM model in optimizing computer network performance prediction.

Two-Way Neural Network Performance Prediction Model Based on Knowledge Evolution and Individual Similarity

Two-Way Neural Network Performance Prediction Model Based on Knowledge Evolution and Individual Similarity

ML-based SDN performance prediction

Software-defined networking (SDN), a new type of network architecture with the advantages of programmability and centralized management, has become a promising solution for managing and optimizing network traffic in modern data centers. However, designing efficient SDN controllers and applications requires a deep understanding of their network performance characteristics. In this work, we implement a machine learning-based method for SDN performance prediction. Our method uses supervised learning to build a training model based on a set of publicly available real network traffic datasets and then uses the model to predict future network performance metrics, such as RTT, S2C, and C2C. Our method is evaluated in two different SDN distributed deployment structures, demonstrating its effectiveness in network performance prediction. We observed that XGBoost achieves the lowest error in most of the cases in terms of MAE, RMSE and MAPE, and feature selection through PCA fails to further improve the prediction performance of XGBoost.

SLAPP: Subgraph-level attention-based performance prediction for deep learning models

The intricacy of the Deep Learning (DL) landscape, brimming with a variety of models, applications, and platforms, poses considerable challenges for the optimal design, optimization, or selection of suitable DL models. One promising avenue to address this challenge is the development of accurate performance prediction methods. However, existing methods reveal critical limitations. Operator-level methods, proficient at predicting the performance of individual operators, often neglect broader graph features, which results in inaccuracies in full network performance predictions. On the contrary, graph-level methods excel in overall network prediction by leveraging these graph features but lack the ability to predict the performance of individual operators. To bridge these gaps, we propose SLAPP, a novel subgraph-level performance prediction method. Central to SLAPP is an innovative variant of Graph Neural Networks (GNNs) that we developed, named the Edge Aware Graph Attention Network (EAGAT). This specially designed GNN enables superior encoding of both node and edge features. Through this approach, SLAPP effectively captures both graph and operator features, thereby providing precise performance predictions for individual operators and entire networks. Moreover, we introduce a mixed loss design with dynamic weight adjustment to reconcile the predictive accuracy between individual operators and entire networks. In our experimental evaluation, SLAPP consistently outperforms traditional approaches in prediction accuracy, including the ability to handle unseen models effectively. Moreover, when compared to existing research, our method demonstrates a superior predictive performance across multiple DL models.

Machine learning enhancement of a digital twin for wavelength division multiplexing network performance prediction leveraging quality of transmission parameter refinement

Digital twins capable of quality of transmission (QoT) estimation and prediction are powerful tools for efficient design and operation of optical networks. In this paper, we employ machine learning techniques to enhance both the QoT estimation and prediction capabilities of an optical network digital twin. By leveraging a method to infer or refine the unknown lumped loss distributions and amplifier gain spectra for accurate characterization of the current optical network state, the accuracy of estimation and prediction is substantially improved. For QoT prediction, we further develop a novel neural-network architecture for erbium-doped fiber amplifiers that, after factory training on a single device and with fully loaded configurations only, generalizes to partially loaded configurations seen after deployment and to different gain and tilt settings and other physical devices of the same type. We combine refined parameters and a novel method that predicts the difference in per-channel power when individual services are added or removed or multiple services are lost due to a fiber cut. The impact of error amplification due to cascading of individual components’ models is shown to be reduced, yielding predictions that are substantially more accurate than simply cascaded predictions. As the network ages, the digital twin can be updated by retraining while using only information available in the current network state.

A Survey on Virtual Network Functions for Media Streaming: Solutions and Future Challenges

Media services must ensure an enhanced user’s perceived quality during content playback to attract and retain audiences, especially while the streams are distributed remotely via networks. Thus, media streaming services rely heavily on good and predictable network performance when delivered to a large number of people. Furthermore, as the quality of media content gets high, the network performance demands are also increasing, and meeting them is challenging. Network functions devoted to improving media streaming services become essential to cope with the high dynamics of network performance and user mobility. Furthermore, new networking paradigms and architectures under the 5G networks umbrella are bringing new possibilities to deploy smart network functions, which monitor the media streaming services through live and objective metrics and boost them in real-time. This survey overviews the state-of-the-art technologies and solutions proposed to apply new network functions for enhancing media streaming.

Performance analysis of a two-level polling control system based on LSTM and attention mechanism for wireless sensor networks.

A continuous-time exhaustive-limited (K = 2) two-level polling control system is proposed to address the needs of increasing network scale, service volume and network performance prediction in the Internet of Things (IoT) and the Long Short-Term Memory (LSTM) network and an attention mechanism is used for its predictive analysis. First, the central site uses the exhaustive service policy and the common site uses the Limited K = 2 service policy to establish a continuous-time exhaustive-limited (K = 2) two-level polling control system. Second, the exact expressions for the average queue length, average delay and cycle period are derived using probability generating functions and Markov chains and the MATLAB simulation experiment. Finally, the LSTM neural network and an attention mechanism model is constructed for prediction. The experimental results show that the theoretical and simulated values basically match, verifying the rationality of the theoretical analysis. Not only does it differentiate priorities to ensure that the central site receives a quality service and to ensure fairness to the common site, but it also improves performance by 7.3 and 12.2%, respectively, compared with the one-level exhaustive service and the one-level limited K = 2 service; compared with the two-level gated- exhaustive service model, the central site length and delay of this model are smaller than the length and delay of the gated- exhaustive service, indicating a higher priority for this model. Compared with the exhaustive-limited K = 1 two-level model, it increases the number of information packets sent at once and has better latency performance, providing a stable and reliable guarantee for wireless network services with high latency requirements. Following on from this, a fast evaluation method is proposed: Neural network prediction, which can accurately predict system performance as the system size increases and simplify calculations.

Evaluation and Prediction of Train Communication Network Performance

The performance of train communication network (TCN) is one of the most important factors to ensure the safety of train operation, and there is a lack of research for the TCN performance currently. To master the state of TCN, a combined method to evaluate and predict the TCN performance is proposed in the paper. First, we select appropriate evaluation indexes and determine their comprehensive weights. Then, the evaluation model of TCN performance is established based on a multidimensional normal cloud model combined with weights. According to the evaluation results, the artificial neural network (ANN) is used to predict the TCN performance to realize the real-time evaluation. Finally, the effectiveness of the evaluation and prediction model of TCN performance is verified by the simulation result, which is obtained from the simulation platform of TCN. The results show that service quality for TCN has been effectively improved, which provides a theoretical reference for evaluation and prediction of TCN performance.

V-Band Channel Modeling, Throughput Measurements, and Coverage Prediction for Indoor Residential Environments

With the increased resolution and frame rates of video recordings, in combination with the current evolution towards video-on-demand streaming services and the user expecting ubiquitous wireless connectivity, it is necessary to design wireless communication systems that allow high-rate data transfer. The large bandwidths that are available in the mmWave frequency band allow such high data rates. In this paper, we provide an experimental and simulated indoor residential radio channel model at V-band frequencies and perform packet error rate and throughput measurements at 60 GHz using IEEE 802.11ad transceivers. We compare the path loss and throughput measurements to simulations using a network performance prediction tool. The path loss measurement results using an omnidirectional transmit antenna correspond well to generic indoor mmWave channel models. Double-directional path loss measurements show that generic models underestimate path loss of non-Line-of-Sight (NLOS) links. A ray-launching algorithm is designed and validated, and used for IEEE 802.11ad throughput estimation based on link budget calculations. The link budget underestimates the achieved throughput, when comparing to adaptive-rate MCS selection in a commercial transceiver, based on the measured signal-to-noise ratio. Packet error rate measurements confirm that, even for NLOS links, throughputs exceeding 1 Gbps are possible.

Predictive Capacity Planning for Mobile Networks—ML Supported Prediction of Network Performance and User Experience Evolution

Network performance prediction is crucial for enabling agile capacity planning in mobile networks. One of the key problems is predicting evolution of spectral efficiency in growing network load conditions. The main factor driving network performance and spectral efficiency is reportedly the Channel Quality Indicator (CQI). In this paper, the performance of different Machine Learning (ML) models were examined, and XGBoost was selected as the best performing model. Furthermore, to improve modeling accuracy, several features were introduced (operating frequency band, Physical Resource Block (PRB) utilization in surrounding cells, number of surrounding cells within a radius, heavy data factor and higher order modulation usage). The impact of these features on CQI prediction were examined.

Identification and Analysis of a Unique Cell Selection Phenomenon in Public Unlicensed Cellular Networks Through Machine Learning

Cellular operators deploy 4G License Assisted Access (LAA) and 5G NR-U base stations in the unlicensed spectrum, to enhance the overall network capacity. This work highlights a unique Physical Cell Id (PCI) related phenomenon observed in the public LAA operator deployments. Notably, the licensed and unlicensed carriers of a device may have the same PCI or different PCIs. The phenomenon is triggered by the combined effect of unlicensed deployment architectures and cell selection mechanisms. Consequently, the phenomenon will intensify manifold in 5G NR-U whose public deployment will soon commence. Unfortunately, the impact of this phenomenon on coexistence network performance is unexplored. It is also desirable to accurately identify the PCI scenarios at the device for improved cell selection and network performance. However, the data imbalance makes the classification problem challenging. This work addresses these problems through the following approach. Operator data from three LAA cellular providers is gathered and analyzed using machine learning algorithms. The impact of the phenomenon on LTE, LAA, and Wi-Fi components is demonstrated in three steps: First, the variation in network performance prediction accuracy in the PCI scenarios is examined. Second, the efficacy of numerosity reduction techniques used in data-driven cell selection is evaluated in both PCI scenarios. The third step entails a comparison of operator data analysis with network measurements. On-site experiments are conducted at the same PCI and different PCI sites to study the differences in real-time network performance. A controlled LTE-WiFi coexistence environment is created and multiple traffic categories are considered. Finally, a class-weight-based solution is proposed for PCI scenario identification. F-score of 0.75 and AUC-ROC of 0.84 is achieved for LAA, with a minimalist feature-set of SINR and Throughput.

Prediction of Optimal Conditions of Hydrogenation Reaction Using the Likelihood Ranking Approach.

The selection of experimental conditions leading to a reasonable yield is an important and essential element for the automated development of a synthesis plan and the subsequent synthesis of the target compound. The classical QSPR approach, requiring one-to-one correspondence between chemical structure and a target property, can be used for optimal reaction conditions prediction only on a limited scale when only one condition component (e.g., catalyst or solvent) is considered. However, a particular reaction can proceed under several different conditions. In this paper, we describe the Likelihood Ranking Model representing an artificial neural network that outputs a list of different conditions ranked according to their suitability to a given chemical transformation. Benchmarking calculations demonstrated that our model outperformed some popular approaches to the theoretical assessment of reaction conditions, such as k Nearest Neighbors, and a recurrent artificial neural network performance prediction of condition components (reagents, solvents, catalysts, and temperature). The ability of the Likelihood Ranking model trained on a hydrogenation reactions dataset, (~42,000 reactions) from Reaxys® database, to propose conditions that led to the desired product was validated experimentally on a set of three reactions with rich selectivity issues.

Estimating entry counts and ATFM regulations during adverse weather conditions using machine learning

In recent years, convective weather has been the cause of significant delays in the European airspace. With climate experts anticipating the frequency and intensity of convective weather to increase in the future, it is necessary to find solutions that mitigate the impact of convective weather events on the airspace system. Analysis of historical air traffic and weather data will provide valuable insight on how to deal with disruptive convective events in the future. We propose a methodology for processing and integrating historic traffic and weather data to enable the use of machine learning algorithms to predict network performance during adverse weather. In this paper we develop regression and classification supervised learning algorithms to predict airspace performance characteristics such as entry count, number of flights impacted by weather regulations, and if a weather regulation is active. Examples using data from the Maastricht Upper Area Control Centre are presented with varying levels of predictive performance by the machine learning algorithms. Data sources include Demand Data Repository from EUROCONTROL and the Rapid Developing Thunderstorm product from EUMETSAT.

On Performance of Ultra-Dense Neighborhood Small Cell Networks in Urban Scenarios

The neighborhood small cell (NSC) network, which employs indoor small base stations (SBSs) to serve both indoor and outdoor users, is considered a cost-effective solution to address the increasing wireless traffic demand. In this letter, a novel evaluation scheme is proposed to predict the performance of a pure NSC network. A network model composed of indoor ultra-dense small cells (UDS) with coordinated multi-point (CoMP) transmissions is employed to address the maximum achievable performance gain in a given neighborhood. A closed-form analytical model is proposed to approximately compute the signal strength received by a specific outdoor user from the SBSs in a complex-shaped building and verified by the Monte Carlo simulations. The performance prediction of the ultra-dense NSC network is demonstrated in two practical urban scenarios. The proposed evaluation scheme facilitates the tractability of NSC network performance predictions in practical scenarios, which can be employed in future NSC embedded neighborhoods design.

Performance analysis and prediction of asymmetric two-level priority polling system based on BP neural network

Concerning the needs of multi-service and network performance prediction in the Internet of Things (IoT), we propose an asymmetric two-priority polling control system model, and use the neural network algorithm to predict and analyze its performance. Firstly, the mathematical model of the system in the continuous time state is established by using the embedded Markov chain theory and the probability generating function. Meanwhile, the characteristics like the average queue length and average cycle of the system are accurately analyzed, and verified in simulation experiments. Subsequently, a three-layer multi-input single-output backpropagation (BP) network model is constructed to predict the performance of the polling system. The results show that the model can not only distinguish multi-business tasks, but also ensure the system delay. BP neural network prediction algorithm can accurately predict the performance of the system, which has a guiding significance for its performance evaluation, and provides a new method for the research of the polling system.

Accelerated Monte Carlo analysis of flow-based system reliability through artificial neural network-based surrogate models

Conventional Monte Carlo simulation-based methods for seismic risk assessment of water networks often require excessive computational time costs due to the hydraulic analysis. In this study, an Artificial Neural Network-based surrogate model was proposed to efficiently evaluate the flow-based system reliability of water distribution networks. The surrogate model was constructed with appropriate training parameters through trial-and-error procedures. Furthermore, a deep neural network with hidden layers and neurons was composed for the high-dimensional network. For network training, the input of the neural network was defined as the damage states of the k-dimensional network facilities, and the output was defined as the network system performance. To generate training data, random sampling was performed between earthquake magnitudes of 5.0 and 7.5, and hydraulic analyses were conducted to evaluate network performance. For a hydraulic simulation, EPANET-based MATLAB code was developed, and a pressure-driven analysis approach was adopted to represent an unsteady-state network. To demonstrate the constructed surrogate model, the actual water distribution network of A-city, South Korea, was adopted, and the network map was reconstructed from the geographic information system data. The surrogate model was able to predict network performance within a 3% relative error at trained epicenters in drastically reduced time. In addition, the accuracy of the surrogate model was estimated to within 3% relative error (5% for network performance lower than 0.2) at different epicenters to verify the robustness of the epicenter location. Therefore, it is concluded that ANN-based surrogate model can be utilized as an alternative model for efficient seismic risk assessment to within 5% of relative error.

Study on prediction model of diesel engine with regulated two-stage turbocharging system based on hybrid genetic algorithm-particle swarm optimization method at different altitudes

ABSTRACT Aiming at the problem of complicated mechanism modeling for regulated two-stage turbocharging (RTST) system of diesel engine, a hybrid genetic algorithm–particle swarm optimization (GA-PSO) method was applied to optimize back propagation neural network (BPNN) performance prediction model of diesel engine. First, using a high-altitude (low-pressure) simulation test bench for diesel engines, a full-load performance test of RTST diesel engine at different altitudes was carried out, and the regulation rules of diesel engine intake characteristics and two-stage turbocharger (TST) matching characteristics with altitude were obtained. Based on the test data, the BPNN is used to establish the boost pressure prediction model of the RTST system of the diesel engine. For the problem of slow convergence and low prediction accuracy of the BPNN, adaptive improvement measures such as momentum and learning rate are introduced, and the GA-PSO optimizes the weights and thresholds of BPNN. The performance comparison shows that hybrid GA-PSO is superior to particle swarm optimization (PSO) and genetic algorithm (GA) in terms of evolution speed and convergence accuracy. Genetic algorithm–particle swarm optimization–back propagation neural network (GA-PSO-BPNN) model has higher prediction accuracy and higher accuracy than particle swarm optimization–back propagation neural network (PSO-BPNN) model and genetic algorithm–back propagation neural network (GA-BPNN) model. Compared with the original model, the root-mean-square error of the model is reduced by 50%, the mean error percentage (MEP) is reduced by 38%, and the sum of the squared errors is reduced by 79%. The performance verification of the GA-PSO-BPNN model shows that the MEP of the model prediction results is 5.5%, and the GA-PSO-BPNN model has higher prediction accuracy.

Effects of network structure on the mechanical and thermal responses of liquid crystal elastomers

Thermoresponsive liquid crystal elastomers (LCEs) have a high potential to be used for actuation applications. There has been a substantial amount of literature on synthesis of different LCE networks and their corresponding performance. However, much of the prior work focuses on the experimental aspect of the effects of mesogenic species, crosslinkers, and spacers on the thermal and mechanical response of LCE. Here we have built on these prior studies, and expanded understanding of LCE work capacity and thermal properties to the molecular and network structures by comparing the experimental results to the theoretically predicted values based on a random walk model derived from classical rubber elasticity. A previously developed two stage thiol-acrylate LCE chemistry was used as the model system. On the basis of increasing the chain entropy, we varied crosslinker concentration, crosslinker functionality, and liquid crystal mesogen length and showed that average molecular weight between crosslinks and molecular weight of the Kuhn segment play important roles in controlling the work capacity. The rubber elastic model predicted network performance agreed reasonably well with the experimental results.