Neural Network Elevator Dispatching via Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing elevator control algorithms become complex and inefficient in tall buildings with multiple elevators, as they rely on conventional rules-based systems rather than advanced machine-learning models.
Innovation Solution
A computer-implemented method for training multiple artificial neural networks to assign calls to elevator cars, where each neural network is trained using reinforcement learning with reward values to optimize call assignment efficiency, promoting competition and collaboration between networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional rules-based control algorithms are used in tall buildings with multiple elevators, then the control logic becomes very complex, but the system can still manage basic elevator operations
Solution Approach 1:
The patent replaces conventional rules-based control algorithms with machine-learning models (artificial neural networks) that learn optimal control strategies through reinforcement learning. This substitution transforms the complex rule-based mechanical control system into an adaptive intelligent system that automatically optimizes elevator dispatching without requiring complex predefined rules
Solution Approach 2:
The patent changes the control approach from static rules to dynamic parameter optimization by training neural networks with adjustable weights and parameters. The system learns optimal control parameters through reinforcement learning, allowing adaptive adjustment of dispatching strategies based on real-time building traffic patterns and elevator states
2Productivity
If machine-learning models are used to control elevators in tall buildings, then control decisions are improved, but the training and implementation become more complex
Solution Approach 1:
The patent creates multiple artificial neural networks that serve as copies or agents, each representing a different elevator or control perspective. These multiple agents compete and collaborate to optimize overall system performance, distributing the computational complexity across multiple simpler networks rather than requiring one extremely complex centralized model
Solution Approach 2:
The patent segments the elevator control problem into multiple independent neural network agents, each responsible for specific aspects of control. This segmentation allows parallel training and evaluation of multiple strategies simultaneously, reducing the complexity burden on any single model while improving overall control efficiency
3Loss of time
If reinforcement learning is used to train artificial neural networks for elevator control, then future control decisions are optimized, but waiting times during training may increase
Solution Approach 1:
The patent performs preliminary training of multiple neural network agents in a simulated environment before deploying them to real elevators. This preliminary action allows the system to learn optimal control strategies offline, so that when deployed, the networks immediately provide optimized control decisions without causing delays in actual passenger service
Solution Approach 2:
The patent uses a simulation environment as an intermediary between the training process and real elevator operations. The simulation acts as a mediator that absorbs the training time and computational requirements, allowing real elevators to benefit from optimized control without experiencing training-related delays or performance degradation
Data Source
AI summary
A method for training neural networks to assign calls to elevator cars simulates an environment in which first and second cars move between building floors in reaction to calls indicating desired floors, each simulation including steps: determining a current state of the environment including a current position of each car, a list of current calls and a new call; inputting first and second input data encoding at least a part of the current state into respective first and second neural networks each configured to convert the input data into output values indicating a probability and/or tendency for the cars to be assigned to the new call; determining a selected car using the output values; assigning the new call to the selected car, and determining reward values quantifying a usefulness of the assignment; training the neural networks using past simulation reward values to increase the usefulness of future assignments.