5G Slice and Cell Resource Prediction with Unified ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current 5G networking systems lack efficient methods for predicting network usage and resource allocation across multiple slices or cells, leading to inefficiencies in resource management due to scalability and maintainability issues, particularly with dynamic activation and deactivation of network slices and cells.
Innovation Solution
A cloud-based, model-efficient AI/ML framework that trains a common prediction model for multiple network slices or cells, using clustering and in-window standardization to handle multiple key performance indicators, enabling rapid adaptation to new slices or cells with reduced latency and resource optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If individual prediction models are trained for each network slice or cell, then prediction accuracy is improved, but system complexity and maintenance burden increase significantly
Solution Approach 1:
The patent merges multiple individual prediction models into a single unified prediction model that can handle multiple network slices and cells simultaneously. This consolidation reduces system complexity while maintaining prediction accuracy through shared model architecture and centralized training on aggregated time-series data from all network elements.
Solution Approach 2:
The unified prediction model is designed with universal functionality to serve multiple network slices and cells through a single model instance. The model accepts time-series data from different network elements as input and generates predictions for various KPIs, eliminating the need for separate specialized models for each network element.
2Reliability
If individual prediction models are maintained for each network slice or cell, then prediction reliability is improved, but ease of operation and maintenance deteriorate
Solution Approach 1:
By combining multiple prediction models into one unified model, the system enables centralized maintenance and updates. Operators can maintain a single model instance, apply updates uniformly across all network slices and cells, and reduce operational overhead while preserving prediction reliability through consistent modeling approaches.
3Measurement precision
If prediction models are trained on complete historical datasets, then prediction accuracy is improved, but latency increases for newly deployed network slices or cells
Solution Approach 1:
The system performs preliminary training of the unified prediction model using aggregated historical time-series data from all network slices and cells before deployment. This pre-training establishes a robust baseline model that can immediately provide predictions for newly deployed network elements without requiring extensive data collection and training time, thus reducing latency while maintaining accuracy.
4Stability of the object's composition
If resource allocation is statically configured, then system stability is improved, but adaptability to changing network demands deteriorates
Solution Approach 1:
The system implements feedback mechanisms where prediction outputs from the unified model are used to dynamically adjust resource allocation decisions. The model continuously monitors network performance and demand patterns, providing feedback that enables adaptive resource allocation while maintaining system stability through controlled adjustment based on reliable predictions.
Data Source
Figure 1A
Figure 1B
Figure 1C
AI summary
A machine learning system is trained to predict resource usage by cells or network slices of a mobile network. For example, a computing system obtains respective datasets for the cells or network slices. Each dataset comprises time steps and respective values for a performance metric of the corresponding one of the cells or network slices. The computing system groups, based on a clustering algorithm applied to (1) the datasets, or (2) the cells or network slices, the datasets into clusters of datasets. The computing system applies, to a subset of most-recent time steps and corresponding values of each dataset of a first cluster of the clusters, a transformation to obtain a set of time steps and corresponding standardized values with which a machine learning system is trained to generate predicted values at future time steps of the datasets of the first cluster.