KR960019946A

KR960019946A - Motor control device using neural network

Info

Publication number: KR960019946A
Application number: KR1019940028790A
Authority: KR
Inventors: 김석우
Original assignee: 이희종; 엘지산전 주식회사
Priority date: 1994-11-03
Filing date: 1994-11-03
Publication date: 1996-06-17
Also published as: KR0152870B1

Abstract

본 발명은 신경회로망을 이용하여 엘리베이터에 적용되는 모터의 회전속도를 제어하는 기술에 관한 것으로, 종래의 비례미분 제어기에 의한 모터제어에 있어서는 모터의 파라미터를 정확하게 알 수 없고, 변화되는 외부환경하에서 고정된 제어 파라미터값을 갖는 비례미분제어기는 항상 잔류 궤적오차가 있어 제어성능면에서 개선점이 요구되고, 더욱이 토오크가 달라지게 되면 모터에 일치하는 제어기를 다시 설계해야 하는 문제점이 있었는 바, 본 발명은 이를 해결하기 위하여, 처음에 일정시간 동안에는 기존의 비례미분 제어시스템으로 모터(3)를 제어하여 모터(3)에 대한 근접한 파라미터를 구하고, 일정시간이 경과된 후 부터 신경회로 제어기(20)를 병렬로 사용함으로써 사용자가 원하는 궤적을 추종하도록 토오크를 생성할 수 있게 하였다.The present invention relates to a technique for controlling the rotational speed of a motor applied to an elevator by using a neural network, and in the conventional motor control by a proportional differential controller, the parameters of the motor cannot be accurately known and fixed under a changed external environment. The proportional derivative controller with the control parameter value has always had a residual trajectory error, which requires improvement in terms of control performance. Moreover, when the torque is changed, there is a problem of redesigning a controller that matches the motor. In order to solve the problem, initially, for a predetermined time, the motor 3 is controlled by an existing proportional differential control system to obtain a close parameter for the motor 3, and after the predetermined time has elapsed, the neural circuit controller 20 is operated in parallel. This allows the torque to be generated to follow the user's desired trajectory.

Description

Motor control device using neural network

본 내용은 요부공개 건이므로 전문내용을 수록하지 않았음As this is a public information case, the full text was not included.

제2도는 본 발명 신경회로망을 이용한 모터 제어장치의 블록도,2 is a block diagram of a motor control apparatus using the neural network of the present invention;

제3도는 입력층에서 은닉층간의 가중치에 의한 은닉층 출력 블록도.3 is a hidden layer output block diagram based on weights between hidden layers in the input layer.

Claims

The input position processing circuit 21 for detecting the rotational position of the motor based on the output of the encoder and scaling the rated torque of the motor 3 using the detected position value as a variable, and the pressure 1 given by the user's control command. And outputs of concealment 1 and concealment 2 from input 2 output from the input value processing circuit 21, and generates output values from the weights between the concealment 1, concealment 2 and the output layer, and the output values and proportional derivatives. The neural circuit 22 that adjusts and outputs an error with the torque output from the controller 2 and the weights adjusted by the neural circuit 22 are processed by concealment 1 and concealment 2, and the motor 3 Torque adjusting circuit 23 for scaling the rated torque of the power supply; and an adder 24 for adding the output torque of the proportional derivative controller 2 and the output of the torque adjusting circuit 23 to supply the motor 3 to the motor 3; Neural Society, characterized in that configured Motor control device using roman.

The position value calculating circuit according to claim 1, wherein the input value processing circuit (21) calculates a rotational position of the motor based on an encoder (61) for detecting the rotation speed of the motor (3) and an output pulse of the encoder (61). And a scaler (63) for scaling the rated torque of the motor (3) using the detection circuit (62) and the output value of the position detection circuit as variables.

The neural circuit 22 multiplexes input 1 and weight * 1-1, input 2 and weight * 2-1, input 1 and weight * 1-2, input 2 and weight * 2-2, respectively. To the outputs of the adders ADD31 and ADD32, and the adders ADD31 and ADD32, respectively. Motor control device using a neural network, characterized in that it comprises a hidden layer output circuit (30) consisting of a limit circuit (31), (32) passing each limited by using an exponential function.

The neural circuit 22 is a multiplexer (MUX41), (MUX42) and the multiplexer (MUX41) for multiplexing concealment 1 and weight ** 1-1, concealment 2 and weight ** 2-1, respectively. And an output layer output circuit 40 including an adder 41 for adding the output of the MUX42 and a limit circuit 41 for restrictively passing the output of the adder 41 by using an exponential function. Motor control device using a neural network, characterized in that.

2. The neural circuit 22 according to claim 1, wherein the neural circuit 22 includes an adder ADD51 for obtaining the difference component between the torque and the adjusted torque, and a limit circuit 51 for passing the output of the adder by using an exponential function. Weight adjustment comprising a multiplexer (MUX51) multiplexing the output and weight ** 1-1 of the limit circuit 51 and a multiplexer (MUX52) multiplexing the output and weight ** 2-1 of the limit circuit 51. Motor control device using a neural network, characterized in that configured to include a circuit (50).

The torque adjusting circuit (22) according to claim 1, wherein the torque adjustment circuit (22) includes multiplexers (MUX71) and (MUX72) for multiplexing the hidden layer 1, the weight layer ** 1-1, the hidden layer 2, and the weight layer ** 2-1, respectively. And a scaler 70 for scaling the rated torque of the motor 3 by using the adder 71 for adding the outputs of the MUX72 and the output of the adder 71 as a variable. Motor control device used.

※ Note: The disclosure is based on the initial application.