Neural Network ADMM Optimization for Dynamic Objective Functions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optimization methods, such as the Alternating Direction Method of Multipliers (ADMM), are challenging to apply in systems where the objective function is difficult to define or dynamically changes, making it hard to manage complex systems and subsystem interactions effectively.
Innovation Solution
A system and method that utilize a network of nodes with calculation processing units, function value calculation units, and control units to calculate optimal determination variables using a neural network constructed through learning processing, allowing for dynamic adaptation to changes in the objective function by transforming the optimization problem into a dual problem and repeatedly updating variables until predetermined conditions are met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the Alternating Direction Method of Multipliers (ADMM) is used to solve optimization problems, then optimal determination variables can be calculated for systems with defined objective functions, but the method cannot be applied when objective functions are difficult to define or dynamically change
Solution Approach 1:
The patent transforms the static ADMM method into a dynamic system by introducing a neural network that can adaptively learn and update objective functions in real-time. The system dynamically adjusts the objective function based on changing system conditions, allowing the optimization method to handle both static and dynamic optimization problems effectively.
Solution Approach 2:
The patent introduces a neural network as an intermediary between the optimization problem and the ADMM solver. This neural network learns the objective function from system data and provides it to the ADMM algorithm, enabling the method to handle cases where the objective function is difficult to define or changes dynamically.
2Adaptability or versatility
If the objective function is redefined to accommodate dynamic changes in system behavior, then the system can adapt to new conditions, but the complexity of defining and updating the objective function increases
Solution Approach 1:
The patent implements a self-service mechanism where the neural network automatically learns and updates the objective function from system data without requiring manual intervention. The system autonomously adapts to changing conditions by training the neural network on new data, eliminating the need for experts to manually redefine objective functions.
Solution Approach 2:
The patent replaces the manual mechanical process of defining objective functions with an automated intelligent system. The neural network automatically learns the objective function from data, substituting the need for manual mathematical modeling and reducing the complexity associated with defining and updating objective functions.
3Ease of operation
If traditional optimization methods are applied to complex systems with multiple interacting subsystems, then control variables can be determined, but the methods fail when subsystem interactions are unknown or dynamically changing
Solution Approach 1:
The patent performs preliminary action by training the neural network to learn subsystem interactions and the objective function before the actual optimization process. This pre-learning phase captures the complex relationships between subsystems, enabling the system to handle dynamic interactions without requiring real-time analysis of these complex relationships.
Data Source
AI summary
An optimization system has a plurality of nodes, each including: a calculation processing unit that calculates an optimum value of a determination variable representing a parameter that controls the nodes by using the ADMM; and a function value calculation unit that receives an input of a value of the determination variable and calculates the value of the objective function based on a calculation model constructed by a learning processing. An arbitrary value of the determination variable is input to the function value calculation unit, and the calculation processing unit substitutes the value of the objective function calculated by the function value calculation unit and the arbitrary value of the determination variable into a second optimization problem by dual transformation of the first optimization problem, thereby repeatedly executing a processing of calculating a value of the dual variable until the value of the dual variable satisfies a predetermined condition.