Elevator Call Allocation Using ML-Guided Iterative Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Heuristic optimization algorithms, such as genetic algorithms or ant colony optimization, used in elevator systems for call allocation can be time-consuming, often resulting in non-optimal solutions due to their iterative nature, which is undesirable in time-sensitive elevator systems.
Innovation Solution
A method combining a machine learning module, an iterative module, and an allocation decision storage, where existing or predicted calls are processed to generate allocation decisions, with the machine learning module being trained using iterative module outputs until predetermined criteria are met, allowing for faster and potentially better allocation decisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If heuristic optimization algorithms (genetic algorithms or ant colony optimization) are used to generate elevator allocation decisions, then allocation decisions can be made, but execution time becomes very long when long population and great number of iterations are used
Solution Approach 1:
The system performs preliminary training of the machine learning module using historical call data and iterative optimization results before actual operation. This pre-computation phase stores learned patterns and optimization strategies, enabling fast allocation decisions during runtime without repeating lengthy iterative processes.
Solution Approach 2:
A machine learning module is introduced as an intermediary between the input call data and the iterative optimization module. The ML module pre-processes calls and predicts optimal allocations, providing high-quality initial solutions to the iterative module, thereby reducing the number of iterations needed and overall execution time.
2Measurement precision
If iterative algorithms are used with long population and great number of iterations, then better allocation decisions may be found, but the execution time becomes very much time
Solution Approach 1:
The system performs preliminary training of the machine learning module using historical call data and iterative optimization results before actual operation. This pre-computation phase stores learned patterns and optimization strategies, enabling fast allocation decisions during runtime without repeating lengthy iterative processes.
Solution Approach 2:
The system uses feedback from iterative optimization results to continuously train and improve the machine learning module. The iterative module's output is fed back to refine the ML module's predictions, creating a learning loop that improves decision quality over time while maintaining fast execution speed.
3Productivity
If machine learning module is used to process existing calls, then faster allocation decisions are provided, but the machine learning module requires training time and computational resources
Solution Approach 1:
The system merges the machine learning module with the iterative optimization module into a hybrid architecture. The ML module handles fast pattern recognition while the iterative module provides optimization refinement, combining the advantages of both approaches into a unified system that manages complexity through functional integration.
Solution Approach 2:
The system performs preliminary training of the machine learning module using historical call data and iterative optimization results before actual operation. This pre-computation phase stores learned patterns and optimization strategies, enabling fast allocation decisions during runtime without repeating lengthy iterative processes.
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
According to an aspect, there is provided a method and an apparatus (200) for determining an allocation decision for at least one elevator. In the solution existing calls are used in an elevator system as a first input in a machine learning module (202). The first input is processed with the machine learning module (202) to provide a first output comprising a first allocation decision. The first output is then used as a second input in an iterative module (204). The second input is processed with the iterative module (204) to provide a second output comprising a second allocation decision. The second allocation decision is provided to an elevator control module (212) and to an allocation decision storage (206) for further machine learning module training.