Method for approximation of optimal control for nonlinear discrete time systems
원문보기
IPC분류정보
국가/구분
United States(US) Patent
등록
국제특허분류(IPC7판)
G06E-001/00
G06E-003/00
G06F-015/18
G06G-007/00
G05B-013/02
G06N-003/00
출원번호
US-0701222
(2010-02-05)
등록번호
US-8538901
(2013-09-17)
발명자
/ 주소
Prokhorov, Danil V.
출원인 / 주소
Toyota Motor Engineering & Manufacturing North America, Inc.
대리인 / 주소
Gifford, Krass, Sprinkle, Anderson & Citkowski, P.C.
인용정보
피인용 횟수 :
0인용 특허 :
8
초록▼
A method for approximation of optimal control for a nonlinear discrete time system in which the state variables are first obtained from a system model. Control sequences are then iteratively generated for the network to optimize control variables for the network and in which the value for each contr
A method for approximation of optimal control for a nonlinear discrete time system in which the state variables are first obtained from a system model. Control sequences are then iteratively generated for the network to optimize control variables for the network and in which the value for each control variable is independent of the other control variables. Following optimization of the control variables, the control variables are then mapped onto a recurrent neural network utilizing conventional training methods.
대표청구항▼
1. A method for approximation of optimal control for nonlinear discrete time systems comprising the steps of: obtaining state variables from a system model,iteratively generating control sequences to optimize desired control variables for subsequent training a recurrent neural network, the value of
1. A method for approximation of optimal control for nonlinear discrete time systems comprising the steps of: obtaining state variables from a system model,iteratively generating control sequences to optimize desired control variables for subsequent training a recurrent neural network, the value of each said control variable being independent of the other control variables,mapping said control variables onto the recurrent neural network implemented by a processor and the neural network has the following form, x(n+1)=f(x(n),u(n)),y(n+1)=h(x(n+1)),where: n=time increment,u=control variable,x=state variable,y=output vector,f and h are differentiable functions, andgenerating an optimal control sequence with the following form, x(n+1)=A(n)x(n)+B(n)u(n),λT(n)=ΔxT(n)Q(n)+λT(n+1)A(n), andδJ(n)/δu(n)=ΔuT(n)R(n)+λT(n+1)B(n)=0,where: A(n)=fx(x(n), u(n)) is a state Jacobian of f,B(n)=fu(x(n), u(n)) is a control Jacobian of f,λT(n)=δJ(n)/δx(n) is an adjoint state vector,Q(n) is a weighting matrix, andR(n) is a weighting matrix. 2. The method as defined in claim 1 wherein in said generating step all control variables are fixed when sequencing forward in time and wherein all state variables are fixed when sequencing backwards in time. 3. The method as defined in claim 1 wherein said mapping step comprises the step of training said recurrent neural network using said optimized control variables. 4. The method as defined in claim 1 wherein the optimal control sequence further comprises: u(n)=ud(n)−R−1(n)BT(n)λ(n+1),x(n+1)=A(n)x(n)−B(n)R−1(n)BT(n)λ(n+1)+B(n)ud(n), andλT(n)=Q(n)x(n)+AT(n)λ(n+1)−Q(n)xd(n), where:ud(n)=target control sequence, andXd(n)=target state sequence. 5. The method as defined in claim 4 wherein the optimal control sequence further comprises: λ(n)=P(n)x(n)+s(n),x(n+1)=M(n)A(n)x(n)+v(n),where: P(N)=Q(N),s(N)=−Q(N)xd(N),M(n)=(I+B(n)R−1(n)BT(n)P(n+1))−1,I=identity matrix, andv(n)=M(n)B(n)(ud(n)−R−1(n)BT(n) s(n+1)). 6. The method as defined in claim 5 wherein the optimal control sequence further comprises: P(n)=Q(n)+AT(n)P(n+1)M(n)A(n),s(n)=AT(n)(I−P(n+1)M(n)B(n)R−1(n)BT(n))×s(n+1)+AT(n)P(n+1)M(n)B(n)ud(n)−Q(n)xd(n). 7. The method as defined in claim 6 wherein the step of iteratively generating control sequences further comprises: Uk+1=Uk+γkŪk,Uk+1=Uk+diag{Γk}Ūk,where: Uk=[uT(0) uT (1) . . . uT(N−1)],Ūk=[ūT(0) ūT(1) . . . ūT(N−1)],γk=step size, andΓk=N×N diagonal matrix. 8. The method as defined in claim 7 wherein in said generating step all control variables are fixed when sequencing forward in time and wherein all state variables are fixed when sequencing backwards in time. 9. The method as defined in claim 7 wherein said mapping step comprises the step of training said recurrent neural network using said optimized control variables.
연구과제 타임라인
LOADING...
LOADING...
LOADING...
LOADING...
LOADING...
이 특허에 인용된 특허 (8)
Werbos Paul J., 3-brain architecture for an intelligent decision and control system.
Sekine Masatoshi (Kanagawa CA JPX) Chang Eric (Beverly Hills CA), Heuristic control system employing expert system, neural network and training pattern generating and controlling system.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.