Clustered Mobility Prediction Model for 5G User Terminals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for predicting user terminal mobility in 5G networks face challenges in accuracy, especially for terminals with unusual movement patterns, and require continuous learning to maintain accuracy, leading to increased computational resource consumption.
Innovation Solution
A prediction model is trained using a neural network-based approach, grouping user terminals into clusters based on similarity in mobility data, including previous movement patterns, day of the week, and time information, to predict base stations for serving user terminals' movement positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained using only previous movement paths to predict user terminal position, then prediction accuracy is high for terminals following general movement patterns, but prediction accuracy is low for terminals with unusual movement patterns
Solution Approach 1:
The patent segments user terminals into different clusters based on their movement patterns. By dividing the homogeneous training approach into heterogeneous clustered training, the system can specialize different models for different movement pattern types, thereby improving accuracy for both general and unusual patterns simultaneously.
Solution Approach 2:
The patent applies local quality by training separate prediction models for different clusters with distinct movement characteristics. Each cluster receives customized training data and model parameters tailored to its specific movement patterns, allowing the system to optimize prediction accuracy locally for each pattern type rather than using a one-size-fits-all approach.
2Reliability
If prediction results are provided to each user terminal using the machine learning model continuously, then mobility prediction is maintained for all terminals, but computational resource consumption increases
Solution Approach 1:
The patent segments the continuous prediction task into cluster-based batch processing. Instead of continuously predicting for each individual terminal, the system processes terminals in clusters periodically, reducing the frequency and intensity of computational operations while maintaining reliable mobility predictions through the clustered model architecture.
Solution Approach 2:
The patent merges multiple user terminals into clusters for joint processing. By combining prediction tasks for multiple terminals within the same cluster, the system reduces redundant computations and optimizes resource utilization, thereby maintaining prediction reliability while lowering overall computational energy consumption.
3Measurement precision
If newly generated data is learned continuously to maintain machine learning model accuracy, then model accuracy is maintained, but training time and computational costs increase
Solution Approach 1:
The patent segments the training process into cluster-specific training tasks. Instead of continuously retraining a single global model with all new data, the system periodically trains separate models for different clusters using only the relevant new data for each cluster, significantly reducing training time and computational overhead while maintaining accuracy for each movement pattern type.
Data Source
AI summary
A server according to an embodiment can collect mobility data of user terminals and group the user terminals into clusters based on the similarity the similarity of the mobility data. The server applies identification information of the clusters to which the grouped user terminals belong and previous movement paths of the grouped user terminals to a neural network-based prediction model, thus making it possible to train a prediction model to predict a base station serving moving locations of the user terminals belonging to each cluster.