Decentralized State Control for Real-Time Load Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for controlling large-scale systems, such as data centers and energy grids, face challenges in efficiently managing load balancing and unit commitment in real-time, especially with varying server performance and external disturbances, leading to inefficiencies and energy wastage.
Innovation Solution
A decentralized adaptive control system using state control devices that acquire and transmit evaluation functions to manage the activation and shutdown of function blocks based on convex evaluation functions, ensuring optimal load distribution and robustness against disturbances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If offline scheduling based on deterministic mathematical programming is used, then optimal load allocation can be achieved under fixed scenarios, but the system cannot handle unexpected external disturbances or dynamic changes in real-time
Solution Approach 1:
The patent transforms the static offline scheduling approach into a dynamic real-time control system. Each function block continuously adjusts its operation state based on real-time evaluation of convex functions and feedback from neighboring blocks, enabling the system to adapt to changing conditions and external disturbances while maintaining optimal load allocation.
Solution Approach 2:
The patent implements a feedback mechanism where each function block evaluates its own state and the states of neighboring blocks using convex evaluation functions. This feedback loop allows the system to continuously monitor performance and adjust load allocation in real-time, resolving the contradiction between optimized allocation and adaptability to disturbances.
2Quantity of substance
If the system scale increases with more function blocks, then system capability and coverage improve, but control complexity and difficulty of real-time optimization increase significantly
Solution Approach 1:
The patent divides the large-scale system into autonomous function blocks that each independently evaluate their own state and make local decisions. This segmentation reduces control complexity by eliminating the need for centralized optimization, as each block only needs to consider its own convex evaluation function and exchange information with immediate neighbors rather than processing entire system state.
Solution Approach 2:
Each function block autonomously evaluates its operation state using its own convex evaluation function and independently determines optimal load allocation without requiring external control. This self-service capability allows the system to scale to large numbers of function blocks while maintaining manageable control complexity through decentralized autonomous decision-making.
3Ease of operation
If uniform control policy is applied to all function blocks, then implementation is simple, but the system cannot optimally handle mixed-machine systems with different performance characteristics
Solution Approach 1:
The patent assigns individual convex evaluation functions to each function block, allowing each block to have customized control characteristics tailored to its specific performance capabilities. This local quality approach enables optimal handling of mixed-machine systems while maintaining mathematical tractability through the convexity property, which ensures efficient optimization without requiring complex uniform policies.
4Adaptability or versatility
If real-time control is implemented to handle unexpected disturbances, then system adaptability improves, but computational burden and difficulty of solving optimization problems increase
Solution Approach 1:
The patent utilizes the mathematical property of convex functions to transform the real-time optimization problem into a computationally tractable form. By ensuring each function block's evaluation function is convex, the system can efficiently solve optimization problems in real-time using convex optimization techniques, avoiding the computational intractability of general non-convex optimization while maintaining real-time adaptability to disturbances.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A state control device is arranged for each function block. When acquiring from a function block (Bi) a state (λ) representing a state of the function block (Bi), a state control device (Ci) corresponding to the function block (Bi) acquires a value (Ui) and a value (Vi) based on an evaluation function (fi) convex upward or downward depending on the performance of the function block (Bi), and transmits the value (Vi) to a state control device (Cj). When receiving from the state control device (Cj) a value (Wj) which the state control device (Cj) has acquired similarly to the value (Vi), the function block (Bi) is controlled based on the value (Ui) and the value (Wj).