Geo-Location-Preserving Graph Forecasting for Mobile Traffic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for mobile network traffic decomposition and forecasting oversimplify spatial complexity and fail to incorporate practical demands of mobile operators, leading to irreversible spatial distortions and limited insights into future per-service consumption, which affects resource allocation and inference accuracy.
Innovation Solution
A geo-location preserving mobile network representation is used to convert network elements into a graph representation, employing a graph-based neural network with spatio-temporal concentration and parallel prediction blocks to capture spatial and temporal correlations, while minimizing spatial distortions and reducing trainable parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional deep learning methods (CNN, LSTM) are used to handle mobile traffic decomposition and forecasting, then the model can process traffic data, but spatial complexity is oversimplified and spatial distortions occur during preprocessing
Solution Approach 1:
The patent transforms the spatial representation parameters from regular grid coordinates to graph-based topological relationships, where nodes represent network elements and edges represent spatial relationships. This parameter change allows the model to preserve actual spatial distances and topological structures while maintaining compatibility with deep learning frameworks, thus resolving the contradiction between ease of implementation and spatial accuracy.
Solution Approach 2:
The patent replaces the traditional mechanical grid-based spatial processing mechanism with a graph neural network mechanism that naturally handles irregular spatial relationships. The graph structure substitutes the rigid grid system, allowing flexible representation of mobile network spatial topology without requiring complex preprocessing that introduces distortions.
2Productivity
If coordinates of antennas are mapped onto a regular grid using Hungarian algorithm, then spatio-temporal dependencies can be captured by CNN, but irreversible spatial distortion is introduced
Solution Approach 1:
Instead of forcing network elements into a regular grid structure, the patent inverts the approach by making the graph structure adapt to the actual spatial distribution of network elements. The graph neural network processes data in its native irregular format, eliminating the need for grid mapping and the associated spatial distortions while maintaining processing efficiency through the graph's inherent parallelizability.
3Productivity
If existing mobile traffic forecasting methods are used, then aggregate traffic prediction can be provided, but limited insights into per-service consumption evolution are given
Solution Approach 1:
The patent segments the aggregate traffic data into service-specific components by incorporating service identification features into the graph neural network model. This segmentation allows the model to simultaneously predict overall traffic trends and decompose them into individual service contributions, providing both aggregate forecasting capability and detailed per-service consumption insights without losing granular information.
Solution Approach 2:
The graph neural network model is designed with multi-functionality to simultaneously perform traffic decomposition, pattern recognition, and forecasting across different service types. The universal graph structure can handle diverse service categories while maintaining the ability to provide specific insights for each service, thus achieving both broad forecasting capability and detailed service-level analysis.
4Measurement precision
If deep packet inspection or resource intensive analysis methods are used for traffic decomposition, then accurate per-service prediction can be achieved, but high computational resources are consumed
Solution Approach 1:
The patent extracts only the essential features needed for accurate traffic decomposition and forecasting, such as aggregate traffic metrics, service identifiers, and temporal patterns, while discarding the need for deep packet inspection. The graph neural network processes these extracted features to achieve accurate per-service predictions with significantly reduced computational overhead, thus resolving the contradiction between prediction accuracy and resource consumption.
Data Source
AI summary
A distributed mobile network traffic data decomposition and forecasting computer-implemented method, comprising:using a geo-location preserving mobile network representation. Locations of mobile network elements are received and converted into a graph representation. Relative distances between adjacent network elements are preserved using respective weights on graph edges; input data is received comprising aggregate network traffic data from network elements corresponding to the network elements locations. The aggregate data includes traffic data corresponding to a plurality of services operating over the network. A graph-based neural network based on the geo-location is used, preserving mobile network representation and configured to capture spatial and temporal correlations in the input data, including at least a spatio-temporal concentration block (STCB) and a parallel prediction block (PPB); loss functions train the graph-based neural network using the input data, to provide network per-service traffic forecasting, the loss functions including an operator cost function configured to capture operator costs.