Elevator Call Allocation via Machine Learning and Iterative Refinement
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 decisions 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 and apparatus that combine a machine learning module, an iterative module, and an allocation decision storage to process existing and predicted calls, using a machine learning module to provide initial allocation decisions that are refined by an iterative module, with the iterative module's outputs used to train the machine learning module until predetermined criteria are met, enabling faster and better allocation decisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If heuristic optimization algorithms are used for elevator allocation decisions, then allocation quality can be improved, but execution time increases significantly
Solution Approach 1:
The system performs preliminary training of the machine learning module offline using historical allocation data and iterative optimization results. This pre-computed knowledge is then applied during real-time operation, allowing fast inference without executing time-consuming iterative algorithms during actual elevator dispatch. The preliminary action transfers optimization quality to the runtime phase while eliminating computational delays.
Solution Approach 2:
The machine learning module acts as an intermediary between the iterative optimization algorithm and the real-time control system. It learns optimal allocation patterns from the iterative module during training, then independently generates allocation decisions during operation. This intermediary transfers the computational burden to an offline training phase while enabling fast online decisions.
2Manufacturing precision
If iterative optimization algorithms are used with long population and many iterations, then allocation decision quality improves, but the system becomes too slow for practical use
Solution Approach 1:
The system performs exhaustive iterative optimization during the offline training phase to teach the machine learning module optimal allocation strategies. Once trained, the model generates decisions instantly during runtime without requiring additional iterative computations. This separates the quality-achieving iteration from the speed-critical execution phase.
Solution Approach 2:
The machine learning module creates a copy of the optimal allocation knowledge learned from extensive iterative optimization. Instead of re-executing the iterative algorithm during real-time operation, the system uses the learned model which replicates the decision-making capability at fraction of the computational cost, enabling both high quality and fast response.
3Productivity
If machine learning module is used alone for allocation decisions, then execution speed improves, but allocation decision quality may be insufficient
Solution Approach 1:
The machine learning module is preliminarily trained using allocation decisions generated by iterative optimization algorithms. This training phase imbues the fast model with high-quality decision patterns, allowing it to achieve both speed and accuracy during real-time operation without requiring the iterative algorithm to run during dispatch.
Solution Approach 2:
The system uses feedback from iterative optimization results to continuously improve the machine learning module through retraining. The iterative module evaluates allocation quality and provides training data that refines the neural network's decision-making capability, ensuring the fast model maintains high allocation quality over time.
Data Source
AI summary
A method for determining an allocation decision for at least one elevator includes using an existing calls in an elevator system as a first input in a machine learning module, processing the first input with the machine learning module to provide a first output comprising a first allocation decision, using the first output as a second input in an iterative module, processing the second input with the iterative module to provide a second ouput comprising a second allocation decision, and providing the second allocation decision to an elevator control module and to an allocation decision storage for further machine learning module training.