Field Spectrum Optimization for Enterprise Planning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optimization algorithms for enterprise planning models are computationally inefficient and prone to converging to local optima, especially when dealing with multi-dimensional problems and coupled decision variables, leading to high computational expense and complexity.
Innovation Solution
The method involves generating a planning function representative of the enterprise model, separating it into independent functions depending on different decision variables, and optimizing each independently to obtain decisions, using a field spectrum optimization approach that decouples decision variables and employs Lagrange multipliers as strategic and constraint factors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If exhaustive search method is used to perform multi-dimensional optimization, then global optimum can be obtained, but computational time becomes extremely slow and not applicable in practical situations
Solution Approach 1:
The patent segments the multi-dimensional optimization problem into multiple one-dimensional optimization problems by decoupling the decision variables through field spectrum separation. Each dimension is optimized independently using Lagrange multipliers, transforming a computationally intensive multi-dimensional search into several simpler one-dimensional optimizations, thereby reducing computational time while maintaining optimization accuracy.
2Productivity
If gradient search methods are used for multi-dimensional optimization, then computational speed improves significantly over exhaustive search, but the methods are still computationally slow and costly
Solution Approach 1:
The patent applies segmentation by decomposing the gradient search process into independent one-dimensional searches along each decision variable axis. By using field spectrum optimization with Lagrange multipliers, the method performs sequential one-dimensional optimizations rather than iterative multi-dimensional gradient computations, significantly reducing the computational burden and time required.
3Productivity
If gradient search methods are used for multi-dimensional optimization, then computational speed improves over exhaustive search, but sensitivity to initial estimates increases and convergence to local optimum may occur
Solution Approach 1:
By segmenting the optimization into independent one-dimensional problems along each field spectrum dimension, the patent eliminates the interdependence between variables that causes gradient methods to converge to local optima. Each one-dimensional optimization can be solved reliably without sensitivity to initial estimates, and the combined solution achieves global optimality.
4Reliability
If simulated annealing or genetic algorithms are used to circumvent convergence to local optimum, then there is a fair chance of finding global optimum, but computational expense increases significantly
Solution Approach 1:
The patent avoids the need for computationally expensive simulated annealing or genetic algorithms by segmenting the optimization problem into independent one-dimensional subproblems. This decoupling ensures convergence to the global optimum through systematic one-dimensional optimization along each field spectrum dimension, achieving reliability without the high computational expense of stochastic global optimization methods.
Data Source
AI summary
In a planning model, a decision variable optimization process (200) generates a planning function (122) describing the planning model, the planning function (122) depending upon a set of decision variables (125). The planning function (122) is separated into independent planning functions, SPi, each of which depend upon different decision variables (125). Each of the independent planning functions, SPi, is independently optimized to obtain decisions for the different decision variables (125), and an outcome is presented that indicates the decisions. The planning function (122) further includes an embedded constraint function that introduces an embedded constraint to weaken the coupling between decision variables (125) in the planning model, thereby reducing an N-dimensional optimization problem into a lower order optimization problem.