Traffic Prediction Using Concatenated Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing traffic prediction technologies often fail to accurately predict short-term traffic volumes, relying on additional information and having limitations in model expression capabilities, which increases latency and device complexity.
Innovation Solution
An electronic device and communication system that obtain first and second traffic feature vectors from past traffic data, generate a concatenation feature vector by combining these vectors, and use this feature vector to predict subsequent traffic data, thereby performing dynamic spectrum sharing in heterogeneous networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RNN-based traffic prediction models use long traffic data as input, then prediction accuracy is improved, but inference time increases
Solution Approach 1:
The patent segments the traffic data into multiple time intervals and extracts feature vectors from each segment. Instead of feeding the entire long traffic data sequence to the RNN model, only the extracted feature vectors from segmented intervals are used as input, reducing the computational burden while maintaining prediction accuracy.
Solution Approach 2:
The patent extracts key feature vectors from the traffic data at different time intervals using an autoencoder model. These extracted feature vectors capture the essential characteristics of traffic patterns without requiring the full raw data, thereby reducing inference time while preserving prediction accuracy.
2Measurement precision
If additional information (application type, statistical measurement) is obtained and processed, then prediction accuracy is improved, but latency and device complexity increase
Solution Approach 1:
The patent employs an autoencoder model that automatically extracts feature vectors from raw traffic data without requiring external additional information such as application types or statistical measurements. The model serves itself by learning representative features directly from the traffic data, eliminating the need for complex data collection and processing of supplementary information.
3Device complexity
If linear models (ARIMA, GPR) or Bayesian nonlinear models are used, then model simplicity is maintained, but model expression capabilities are insufficient
Solution Approach 1:
The patent replaces traditional linear models (ARIMA, GPR) and Bayesian nonlinear models with a deep neural network-based RNN model that uses autoencoder for feature extraction. This substitution enables the system to capture complex non-linear traffic patterns and temporal dependencies that simpler models cannot represent, significantly improving model expression capabilities.
Solution Approach 2:
The patent creates a composite prediction system combining an autoencoder model for feature extraction and an RNN model for temporal pattern recognition. This composite architecture integrates the strengths of both components: the autoencoder captures spatial features from traffic data while the RNN processes temporal dependencies, achieving superior overall performance.
4Duration of action of moving object
If long-term traffic prediction in units of hours is performed, then comprehensive network planning is enabled, but short-term traffic prediction in units of milliseconds cannot be achieved
Solution Approach 1:
The patent makes the prediction system dynamic by adjusting the input feature vectors and prediction parameters according to the required time horizon. For short-term prediction, the system uses recent feature vectors with smaller time intervals, while for long-term prediction, it incorporates historical features with larger intervals, enabling flexible adaptation to different prediction requirements without sacrificing speed or accuracy.
Data Source
AI summary
An electronic device according to various embodiments comprises: a memory including instructions; and a processor that is electrically connected to the memory and executes the instructions. When the instructions are executed by the processor, the processor may: acquire a first traffic feature vector and a second traffic feature vector on the basis of traffic data generated in a wireless network device during a plurality of time intervals; generate a combined feature vector by combining the first traffic feature vector and the second traffic feature vector; and acquire predicted traffic data, to be generated in the wireless network device, on the basis of the combined feature vector. Various other embodiments are possible.