Mobile Network Traffic Prediction Using Bi-LSTM and Adaptive Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for predicting mobile network traffic struggle with complexity, scalability, and accuracy, leading to resource consumption, latency, and inaccurate long-term forecasts due to the heterogeneous and nonlinear nature of mobile network traffic.
Innovation Solution
A prediction system utilizing a Bi-LSTM model trained with an adaptive waterwheel plant optimization model to process time series data, enhancing accuracy and reducing prediction errors by minimizing MAPE, MSE, and RMSE.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current prediction techniques are used, then prediction capability is provided, but accuracy is poor due to heterogeneous and nonlinear traffic patterns
Solution Approach 1:
The patent combines multiple prediction models (ARIMA, exponential smoothing, neural networks) into a hybrid ensemble system that leverages the strengths of each individual model to accurately capture heterogeneous and nonlinear traffic patterns that single models cannot handle effectively
Solution Approach 2:
The system dynamically adjusts model parameters and weights based on changing traffic conditions and patterns, allowing the prediction system to adapt to non-stationary mobile network traffic characteristics and improve accuracy over time
2Measurement precision
If complex prediction models are deployed, then prediction capability is enhanced, but resource consumption increases
Solution Approach 1:
The patent divides the prediction task into multiple independent model components that can be executed separately and in parallel, reducing the computational burden on single resources and enabling efficient distribution across available computing infrastructure
Solution Approach 2:
The system uses lightweight model copies and approximations for rapid predictions, reserving full-complexity models only when necessary, thereby reducing overall computational resource consumption while maintaining prediction accuracy
3Device complexity
If traditional prediction methods are used, then simplicity is maintained, but latency increases due to inefficiency
Solution Approach 1:
The system performs preliminary data preprocessing, feature extraction, and model warm-up operations in advance, so that when predictions are needed, the computational work is already partially completed, reducing actual prediction latency
Solution Approach 2:
The patent replaces traditional mechanical computation methods with optimized algorithms and data structures that reduce computational complexity and execution time, achieving faster predictions without proportionally increasing system complexity
4Measurement precision
If prediction systems are scaled up, then coverage is improved, but scalability is limited by current architecture
Solution Approach 1:
The patent designs a universal prediction framework that can handle multiple traffic types, network conditions, and deployment scenarios through a common architecture, enabling scalable expansion without requiring separate specialized systems for each use case
Data Source
AI summary
A device may receive time series data associated with a mobile network, and may process the time series data, with a bidirectional-long short-term memory (Bi-LSTM) model, to generate a preliminary mobile network traffic prediction. The device may apply an adaptive waterwheel plant optimization model, to the preliminary mobile network traffic prediction, to train the Bi-LSTM model and to generate a trained Bi-LSTM model. The device may process the preliminary mobile network traffic prediction, with the trained Bi-LSTM model, to generate a final mobile network traffic prediction, and may perform one or more actions based on the final mobile network traffic prediction.