Self-Learning Network Control Policy Generation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for generating network control policies are limited by their scalability and complexity, requiring high computational costs and manual configuration, making them inefficient for dynamic network adjustments and unable to adapt to changing network conditions.

Innovation Solution

A self-learning method and apparatus for generating network control policies that utilize a central control server, prediction server, and training server to predict network status and adjust policies in real-time, optimizing data flow routing and improving network control efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional optimization methods (linear planning or integer planning) are used to generate network control policy, then a network control policy can be obtained, but the calculation complexity increases and scalability becomes weak as network elements, service types, and traffic increase

Engineering Contradiction:
Improvenetwork control policy generationVSAvoidcalculation complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces conventional mechanical optimization methods (linear planning, integer planning) with a deep reinforcement learning model. The neural network learns optimal routing policies through self-training on historical data, substituting complex mathematical optimization with a trained intelligent agent that can generalize to new scenarios without re-optimization.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system performs preliminary action by pre-training the reinforcement learning model on historical network data offline. This allows the model to learn optimal policies in advance, so when deployed online, it can make rapid decisions without performing complex real-time optimization calculations.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If offline optimization is used to generate network control policy, then a control policy can be obtained, but dynamic instant policy adjustment cannot be implemented

Engineering Contradiction:
Improvecontrol policy generationVSAvoiddynamic adjustment capability
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent transforms the static offline optimization approach into a dynamic system using reinforcement learning. The trained model can adapt to changing network conditions in real-time by processing current state inputs and generating updated routing policies, enabling dynamic instant adjustment while maintaining policy quality.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system implements feedback by continuously monitoring network state and using the reinforcement learning model to adjust policies based on current conditions. The model learns from historical feedback (routing outcomes) and adapts its policy generation based on real-time network state observations.

Inventive Principle:
Principle #23Feedback

3Reliability

If optimization problem is solved again when network status changes, then updated control policy can be obtained, but the process becomes hysteretic and requires large amount of manual configuration

Engineering Contradiction:
Improvecontrol policy updateVSAvoidpolicy update time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent replaces the time-consuming manual re-optimization process with a pre-trained reinforcement learning model. When network status changes, the model instantly processes the new state through neural network inference (milliseconds) rather than performing full optimization calculations, eliminating both time loss and manual configuration requirements.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system achieves self-service by automatically adapting to network changes without human intervention. The reinforcement learning model autonomously processes new network states and generates updated policies, eliminating the need for manual model reconfiguration that would otherwise be required when network topology or traffic patterns change.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3297211B1Method and apparatus for generating network control policy, and network controller
Publication Date: 2021.03.17 HUAWEI TECH CO LTD
  • EP3297211B1 patent drawingFigure 1
  • EP3297211B1 patent drawingFigure 2
  • EP3297211B1 patent drawingFigure 3

AI summary

The present invention provides a method for generating a network control policy, including: obtaining network status data; estimating network status information according to the network status data, and predicting next-moment status information of the network according to the network status information; receiving a data flow currently entering the network, and preprocessing the data flow, to obtain a preprocessing result of the data flow; generating a control action by using a current network control policy and according to the network status information, the next-moment status information of the network, and the preprocessing result of the data flow; controlling the network according to the control action, and obtaining feedback information of the network controlled according to the control action; and adjusting the current network control policy in real time according to the feedback information, to obtain a next-moment network control policy. The network control policy is adaptively adjusted in real time, to improve network control efficiency.