Layered Mobile Network KPI Prediction for Scalable UE Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional methods for predicting user end (UE) device-level key performance indicators (KPIs) in large-scale mobile networks are not scalable due to limited compute resources and high costs, and they struggle to provide predictions for newly deployed UEs without historical data.
Innovation Solution
A scalable, layered architecture that uses a hybrid approach to predict KPIs by profiling and clustering UE behaviors per cell, fusing cell-level and per-UE predictions, and employing a multilayered model with short-term and long-term predictors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If per UE machine learning/artificial intelligence models are used for KPI prediction, then prediction accuracy is improved, but scalability deteriorates due to limited compute resources and high development/maintenance costs
Solution Approach 1:
The patent segments the KPI prediction task into two distinct layers: a cell-level prediction layer that handles aggregate traffic predictions for entire cells, and a user-level prediction layer that handles individual UE predictions. This segmentation allows the system to leverage both ML models for accuracy while maintaining scalability through the aggregate-level processing that requires fewer computational resources per user.
Solution Approach 2:
The patent merges cell-level aggregate KPI predictions with user-level individual predictions to produce final KPI estimates. By combining these two prediction layers, the system achieves both scalability (through cell-level aggregation) and accuracy (through user-level ML predictions), resolving the contradiction between these two requirements.
2Measurement precision
If per UE machine learning/artificial intelligence models are used for KPI prediction, then prediction accuracy is improved, but development and maintenance costs increase
Solution Approach 1:
The patent segments the prediction architecture into cell-level and user-level components, where the cell-level aggregate models are simpler and less expensive to develop and maintain compared to full per-UE ML models. This segmentation reduces overall development and maintenance costs while preserving prediction accuracy through the complementary user-level predictions.
3Ease of operation
If traditional KPI prediction methods are used, then resource allocation can be optimized, but predictions cannot be provided for newly deployed UEs without historical KPI values
Solution Approach 1:
The patent performs preliminary cell-level aggregate KPI predictions that are independent of individual UE historical data. For newly deployed UEs without historical KPI values, the cell-level predictions provide baseline estimates that enable immediate resource allocation decisions, while user-level predictions are gradually refined as historical data becomes available.
Data Source
AI summary
Aspects of the subject disclosure may include, for example, receiving a request from a mobile network entity for a key performance indicator (KPI) prediction over a portion of a mobile network, and obtaining a group of identifiers associated with the mobile network entity. Further embodiments can include obtaining a group of KPIs associated with the mobile network entity based on the group of identifiers, and determining a KPI prediction associated with the mobile network entity based on the group of KPIs. Additional embodiments can include allocating a group of network resources to the mobile network entity based on the KPI prediction. Other embodiments are disclosed.