Elevator Dispatch Control Using Neural Networks for Tall Buildings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional elevator control algorithms struggle to efficiently manage trip scheduling in tall buildings with multiple elevators, leading to complex decision-making and increased passenger waiting times.
Innovation Solution
Implementing an artificial neural network trained with reinforcement learning to process sensor data, converting it into control commands for elevator movement and call assignment, using efficient data encoding methods to reduce input size and improve processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional elevator control algorithms are used in tall buildings with multiple elevators, then the control system becomes very complex, but passenger waiting times increase and transport capacity decreases
Solution Approach 1:
The patent replaces conventional rule-based control algorithms with a machine learning model (neural network) that has been trained to optimize elevator dispatching decisions. This substitution transforms the control system from one requiring complex manual rule sets to an intelligent system that automatically learns optimal control strategies, reducing control complexity while improving performance metrics like waiting time and transport capacity.
2Productivity
If machine learning models are used to optimize elevator control decisions, then passenger waiting times decrease and transport capacity increases, but the model requires extensive training data and computational resources
Solution Approach 1:
The patent employs offline training of the machine learning model using simulated elevator operations and historical data before deploying the model to actual elevator systems. This preliminary action allows the model to learn optimal control strategies in advance, so that during real-time operation, the model can make rapid decisions without requiring extensive computational resources or retraining, thus achieving high transport capacity while managing training complexity.
3Measurement precision
If the artificial neural network processes detailed sensor data for each possible control command, then control accuracy improves, but processing time increases
Solution Approach 1:
The patent implements a two-stage processing approach: first, the neural network quickly evaluates a limited set of plausible control commands based on current elevator states and call patterns; second, it selects the best command from this reduced set. This partial action approach maintains control accuracy by still considering multiple options while significantly reducing processing time by not evaluating all possible commands in detail, thus resolving the contradiction between accuracy and speed.
Data Source
AI summary
An elevator control method using an artificial neural network includes: receiving elevator sensor system data input at a controller configured to execute the artificial neural network trained to convert input data into at least one output value indicating a possible elevator control command, the sensor data including a current position of an elevator car with respect to building floors, a car current moving direction and a list of current calls indicating floors where the car should stop; converting the current calls into a set of arrays using the current position and the current moving direction, wherein each array stores variables indicating floors where the car should stop during a trip in one direction to the lowest or highest floor, the number of arrays being a function of the number of trips required for the car to fulfil the current calls; and generating the input data by concatenating the arrays.