AI-2200 System
This system is designed for larger buildings with three to eight cars in the elevator group. It suits buildings with dynamic traffic conditions throughout the day and peak carrying times.
Design RoomElevatorsGPS-IIIAI
Series GPS-III Passenger Elevators
Advanced AI Supervisory Control
Mitsubishi's AI Supervisory Control is the key factor in creating an ideal
elevator system with optimum user service. Two basic systems are available,
and between them they offer a wide range of special functions to suit the
needs of any type of building.
AI-21 System
This system is designed for small or medium size buildings with two to four
cars in the elevator group.
AI-2200 System
This system is designed for larger buildings with three to eight cars in the
elevator group. It suits buildings with dynamic traffic conditions throughout
the day and peak carrying times.
Expert System and Fuzzy Logic
The brain of the AI Supervisory Control employs an intelligent Expert
System that utilizes the practical knowledge and experience of elevator
group control experts. This information is stored in the system's memory as
a "Knowledge Database." Drawing from this database, various traffic
conditions are monitored and analyzed applying IF-THEN decision rules to
maximize the effectiveness of each elevator operation. Mitsubishi Electric
has applied fuzzy logic in a manner that enables the elevator control system
to make decisions using fragmentary and fuzzy logic intelligence concepts.
For example, using its "intelligence" and "common sense," the system can
determine whether or not potential car assignments will result in longer
waiting times for calls in the near future or cause elevator congestion. The
assessment results are applied to determine the car assignments in order to
improve overall service.
Configuration of AI-2200 System
Psychological Waiting Time Evaluation
This evaluation function is Mitsubishi Electric technology that originates from
the psychological thought patterns of a passenger waiting for an elevator:
the irritation of a passenger waiting for elevator arrival is proportional to the
square of the actual waiting time. Elevator assignments to hall calls are
performed on the basis of evaluation results. In addition to forecasted
waiting time, such factors as probability of being bypassed for a hall call,
probable time required for traveling after car assignment, current car load
and many others are applied in the evaluation function owing to its
coefficient diversity. Car assignments to hall calls are made as a sum of all
factors.
Strategic Overall Assignment
Combining all of the building traffic conditions, the system forecasts where
future service will be needed and assigns cars accordingly. This greatly
reduces the average overall waiting time and provides optimum service to
passengers throughout the building. Once all car and hall calls have been
serviced, the system forecasts where the next calls for service will arise and
accordingly assigns the cars so that the waiting time for future passengers is
also reduced. (Strategic Overall Spotting)
Dynamic Rule-set Optimizer
Elevator control (car allocation) performed using ideal rule-set
Using Neural Networks technology, this system predicts elevator traffic a
few minutes later. According to the predicted traffic, a high speed RISC
(Reduced Instruction Set Computer) runs real-time simulations and selects
the optimal rule-set. In this way, elevator control (car allocation) is
performed using the ideal rule-set.
Example of rule-set selection with real-time simulation
The diagram below shows an example of rule-set selection for morning up
peak time. The ideal rule-set will be selected every few minutes, according
to traffic conditions in the building.
Distinction of Traffic Flow with Neural Networks
Operation pattern (accompanied with car allocation or parking function) is
one of the major phases in the supervisory control for the efficient operation.
This responds to the daily traffic fluctuations in the building, which is mainly
observed in the morning, lunchtime and evening. The 
AI-2200
system utilizes neural networks to recognize the distinctive patterns of traffic
flows in real time with highly precision, that results in optimizing the
selection and cancellation of the patterns. Traffic flows in each zone are
distinguished by basic traffic data as the number of passengers in boarding
and exiting (estimated by measuring car loads). This system adopts a
learning module performed with training data for each neural network, that is
responsible to a temporary traffic flow, traffic fluctuations in weekly. These
training data for the learning module are compiled from a vast pool of data
gathered through the actual traffic measurings.
Car Allocation Tuning
The AI-2200 system applies a refined algorithm to improve the
average waiting times at each floor in the building, which controls the
number of elevator cars allocated or parked to the crowded floors during
peak periods in incoming, outgoing and lunchtime traffic. The algorithm
covers the situations of elevator services, the conditions of elevator
operations, etc., as well as the degree of traffic density or flows. The tuning
process is described as follows:
Step 1: The initial number is set at the starting of crowd.
Step 2: Gathering the data of each elevator operation.
Step 3: And then, the initial allocation number is tuned to be increased or
decreased according to the fuzzy rules.
Destination Oriented Prediction System
Press the destination floor, and the service elevator is indicated immediately
When a passenger presses a destination floor button on the newly
developed hall operating panel installed in the lobby and another busy floor,
the code letter for the service elevator immediately appears next to the
destination floor button. The passenger knows at a glance which elevator to
board.
Because the Destination Oriented Prediction System analyzes passengers
destinations from congested floor to reduce passenger travel times and
minimize congestion, the system greatly improves traffic efficiency
particularly in peak times. In addition, since the passengers destination floor
is registered automatically, theres no need to press the floor button again
after boarding the elevator.
Immediate Prediction Indication
Once a passenger has registered a hall call, the ideal car to respond is
selected, and the hall lantern lights and a chime sounds once to indicate
which door will open. As the car approaches, the lantern begins to flash and
the chime sounds twice. This system provides a highly reliable prediction of
car arrival and reduces passenger irritation.
HomeDesign RoomElevatorsGPS-IIIAI
Copyright(c) 2002 Mitsubishi Electric Corporation. All Rights Reserved.
