Neural Network Scheduling Decoupling User Equipment Quantity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep reinforcement learning scheduling algorithms face challenges in self-adapting to the varying number of user equipment, leading to performance losses and poor scheduling flexibility.
Innovation Solution
A scheduling method that decouples a deep neural network from the quantity of user equipment by using a shared neural network model across all user equipment, allowing the model to be applied to scenarios with different numbers of user equipment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a deep neural network is designed with input and output neuron sizes depending on the quantity of to-be-scheduled user equipments, then the scheduling algorithm can adapt to different user quantities, but the system causes performance loss and poor scheduling flexibility
Solution Approach 1:
The patent segments the scheduling problem into two independent parts: (1) a user-quantity-independent deep neural network that processes normalized state features, and (2) a separate mapping layer that handles user quantity variations. This segmentation allows the core neural network to maintain fixed architecture while still adapting to different user quantities through the mapping mechanism, thereby resolving the contradiction between adaptability and performance.
Solution Approach 2:
The patent applies parameter changes by introducing a normalization parameter (dividing by maximum user quantity) to transform state features into a fixed-dimensional space. This parameter transformation allows the neural network to process variable user quantities through a fixed architecture, maintaining both adaptability to different user quantities and consistent scheduling performance.
2Adaptability or versatility
If the deep neural network changes with the quantity of to-be-scheduled user equipments, then it can accommodate different user quantities, but the device complexity increases
Solution Approach 1:
The patent creates a universal deep neural network that can handle any user quantity through feature normalization. Instead of designing separate networks for different user quantities, a single fixed-architecture network serves multiple functions across varying user quantities, reducing device complexity while maintaining flexibility.
Solution Approach 2:
The patent transforms the variable user quantity dimension into a normalized fixed dimension by dividing state features by the maximum user quantity. This dimensionality transformation allows the neural network to operate in a fixed-dimensional space regardless of the actual user quantity, simplifying the network architecture while preserving adaptability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A scheduling method and apparatus are provided to resolve a problem in a conventional technology that it is difficult for a scheduling algorithm to self-adapt to a quantity of user equipments. The method is as follows: after processing first state information sets of K to-be-scheduled terminal devices, obtaining second state information sets of the K to-be-scheduled terminal devices, where a second state information set of any to-be-scheduled terminal device includes state information of the any to-be-scheduled terminal device and state association data between the any to-be-scheduled terminal device and another to-be-scheduled terminal device, and a dimension of a second state information set of any to-be-scheduled terminal device is H; and inputting a second state information set of each to-be-scheduled terminal device into a first neural network model, and determine a scheduled weight of each to-be-scheduled terminal device to determine a scheduling result, where the first neural network model is determined based on H. During scheduling, a neural network model is decoupled from a quantity of to-be-scheduled terminal devices, and may be applied to scenarios in which quantities of to-be-scheduled terminal devices are different.