Systems and methods related to traffic flow prediction in a wireless network are disclosed. In one embodiment, a computer-implemented method comprises collecting training data comprising Internet Protocol (IP) addresses extracted from packets for traffic flows in a wireless network and one or more actual traffic type related parameters for each of the traffic flows. The method further comprises training heavy-hitter IP address encodings based on the extracted IP addresses and encoding the extracted IP addresses using the trained heavy-hitter IP address encodings. The method further comprises training a traffic type predictor of a traffic flow predictor based on the encoded IP addresses and the one or more actual traffic type related parameters for each of the traffic flows, where the traffic type predictor is a learning model that maps encoded IP addresses to one or more predicted traffic type related parameters.

Systems and methods related to traffic flow prediction in a wireless network are disclosed. In one embodiment, a computer-implemented method comprises collecting training data comprising Internet Protocol (IP) addresses extracted from packets for traffic flows in a wireless network and one or more actual traffic type related parameters for each of the traffic flows. The method further comprises training heavy-hitter IP address encodings based on the extracted IP addresses and encoding the extracted IP addresses using the trained heavy-hitter IP address encodings. The method further comprises training a traffic type predictor of a traffic flow predictor based on the encoded IP addresses and the one or more actual traffic type related parameters for each of the traffic flows, where the traffic type predictor is a learning model that maps encoded IP addresses to one or more predicted traffic type related parameters.

A method performed by a first radio node for handling a beam pair with a second radio node is provided. The first radio node receives a first information from one or more other radio nodes. The first information comprises a number of quality values related to a number of beam pairs. The first radio node predicts a time to failure for a first beam pair. The first radio node then decides whether or not there is enough time until the predicted time to failure, for performing a beam pair switch from the first beam pair to a second beam pair. When there is enough time, the first radio node switches to the second beam pair before the predicted time to failure. When there is not enough time, the first radio node prepares an upcoming beam pair failure.

A method performed by a first radio node for handling a beam pair with a second radio node is provided. The first radio node receives a first information from one or more other radio nodes. The first information comprises a number of quality values related to a number of beam pairs. The first radio node predicts a time to failure for a first beam pair. The first radio node then decides whether or not there is enough time until the predicted time to failure, for performing a beam pair switch from the first beam pair to a second beam pair. When there is enough time, the first radio node switches to the second beam pair before the predicted time to failure. When there is not enough time, the first radio node prepares an upcoming beam pair failure.

A method for controlling deployment of network configuration changes includes receiving, by centralized network management system executed by a processor and memory, configuration change instructions to alter a configuration of a network; computing, by the centralized network management system, a weighted impact of the configuration change instructions; determining, by the centralized network management system, whether the weighted impact of the configuration change instructions exceeds a threshold impact level; and in response to determining that the weighted impact does not exceed the threshold impact level, executing the configuration change instructions.

A method includes obtaining input features based on network traffic received during a time window. The method also includes generating multiple network service type predictions about the network traffic during the time window using a machine learning (ML) classification system operating on the input features. The method also includes storing the multiple network service type predictions in different time steps in a first-in first-out (FIFO) buffer and generating decisions about a presence of each of multiple service types in the network traffic using a voting algorithm. The method also includes reducing fluctuations in the generated decisions using a logic-based stabilizer module to generate a final network service type decision.

Techniques for monitoring network connectivity using machine learning are provided. A plurality of historical connectivity records is received, and a first machine learning model type, of a plurality of machine learning model types, is selected based on the plurality of historical connectivity records. A machine learning model, of the first machine learning model type, is trained based on the plurality of historical connectivity records, where the machine learning model learns to generate forecasted connectivity records based on the training.

There is provided a method and system for communication network management. There is provided an active TE architecture and procedure that rely on the epistemic uncertainty obtained from traffic forecasting models. According to embodiments, the traffic forecasting models can predict the mean of the network traffic demand and can extract one or more of the features relating epistemic uncertainty and the aleatoric uncertainty. According to embodiments, the epistemic uncertainty is used to vary the sampling frequency of network statistics in TE applications, for specific times or specific flows. A time-window can be used to predict network traffic can be varied (e.g. increased or decreased) to adjust the epistemic uncertainty.

There is provided a method and system for communication network management. There is provided an active TE architecture and procedure that rely on the epistemic uncertainty obtained from traffic forecasting models. According to embodiments, the traffic forecasting models can predict the mean of the network traffic demand and can extract one or more of the features relating epistemic uncertainty and the aleatoric uncertainty. According to embodiments, the epistemic uncertainty is used to vary the sampling frequency of network statistics in TE applications, for specific times or specific flows. A time-window can be used to predict network traffic can be varied (e.g. increased or decreased) to adjust the epistemic uncertainty.

A method and a system for identifying radio access network (RAN) coverage at geographic locations includes generating a grid layer of real-time RAN coverage based on a key performance indicator (KPI). Generating a grid layer of predicted RAN coverage. Generating a viewport representation corresponding to a geographic location supported by the RAN. Superimposing the real-time RAN coverage grid layer, the predicted RAN coverage grid layer, and the viewport representation into a unified coverage representation. Determining RAN availability at a selected geographic location based on the unified coverage representation.