PURPOSE: To judge the optimum car including the estimating function of situation changes by using many learning samples mated with responding cars when landing calls occur and the system state data of a group at that time as teach signals to learn a neural net.

CONSTITUTION: When a landing call H1 in the ascending direction occurs at the eighth floor, the data indicating the state of a group and considered to be required for allocation such as the car positions and operation directions of cars A-C at this time are stored. A landing call H2 in the descending direction newly occurs at the seventh floor some time later, the landing call H2 is allocated to the car A, and the landing call H1 is changed and allocated to the car B. When the car A is stopped in response to the landing call H2, the system state data of the group stored when the call occurs and the car A are paired and registered as a pair of learning samples. Similar actions are taken for the car B. When the number of samples reaches a prescribed number, many learning samples are used as teaching signals to learn a neural net for call allocation.