Download presentation
Presentation is loading. Please wait.
1
4th April 2005 Colloquium on Predictive Control, Sheffield 1 Nonlinear Model Predictive Control using Automatic Differentiation Yi Cao Cranfield University, UK
2
4th April 2005Colloquium on Predictive Control, Sheffield2 Outline Computation in MPC Dynamic Sensitivity using AD Nonlinear Least Square MPC Error Analysis and Control Evaporator Case Study Performance Comparison Conclusions
3
4th April 2005Colloquium on Predictive Control, Sheffield3 Computation in Predictive Control Predictive control: at t k, calculate OC for t k · t· t k+P, apply only u(t k ), repeat at t k+1 Prediction: online solving ODE Optimization: repeat prediction, sensitivity required. Typically, over 80% time spend on solving ODE + sensitivity
4
4th April 2005Colloquium on Predictive Control, Sheffield4 Current Status Linear MPC successfully used in industry Most systems are nonlinear. NMPC desired. Computation: solving ODE and NLP online. Difficult to get gradient for large ODE systems. Finite difference: inefficient and inaccurate Sensitivity equation: n×m ODE’s Adjoint system: TPB problem Other methods: sequential linearization and orthogonal collocation
5
4th April 2005Colloquium on Predictive Control, Sheffield5 Automatic Differentiation Limitation of finite and symbolic difference Function = sequence of fundamental OP Derivatives of fundamental OP are known Numerically apply chain rules Basic modes: forward and reverse Implementation: operating overloading Source translation
6
4th April 2005Colloquium on Predictive Control, Sheffield6 ODE and Automatic Differentiation z(t)=f(x), x(t)=x 0 +x 1 t+x 2 t 2 +…+x d t d z(t)=z 0 +z 1 t+z 2 t 2 +…+z d t d AD forward: z k = z k (x 0,x 1,…,x k ) AD reverse: z k / x j = z k-j / x 0 = A k-j f’(x)=A 0 +A 1 t+ +A d t d ODE: dx/dt=f(x), dx/dt=z(t), x k+1 =z k /(k+1). x 0 =x(t 0 ), x 1 =z 0 (x 0 ), x 2 =z 1 (x 0,x 1 ), … Sensitivity: B k =dx k /dx 0 =1/k k i=0 A k-i-1 B i, B 0 =I dB/dt=f’(x), B=B 0 +B 1 t+ +B d t d x(t 0 +1)= d i=0 x i, dx(t 0 +1)/dx 0 = d i=0 B i
7
4th April 2005Colloquium on Predictive Control, Sheffield7 Non-autonomous Systems I For control systems: dx/dt=f(x,u) u(t)=u 0 +u 1 t+u 2 t 2 +…+u d t d (x 0, u(t)) → x(t), dx(t)/du k =? Method I: Augmented system: dv 1 /dt=v 2, …, dv d+1 /dt=0, v k+1 (t 0 )=u k, u=v 1 X=[x T v T ] T, dX/dt=F(X) (autonomous) High dimension system, n+md Not suitable for systems with large m
8
4th April 2005Colloquium on Predictive Control, Sheffield8 Non-autonomous Systems II Method II: nonsquare AD Let v=[u 0 T, u 1 T, …, u d T ] T x k+1 = z k (x 0,x 1,…,x k,v)/(k+1) A k =[A kx | A kv ] := [ z k / x 0 | z k / v] B k =[B kx | B kv ] := [dx k /dx 0 | dx k /dv] B k = A k-1 + k-1 j=1 A (k-j-1)x B j, B 0 =[I | 0] x(t 0 +1)= d k=0 x k, dx(t 0 +1)/dv= d k=0 B kv, dx(t 0 +1)/dx(t 0 )= d k=0 B kx
9
4th April 2005Colloquium on Predictive Control, Sheffield9 Nonlinear Least Square MPC Φ=½∑ P k=0 (x(t k )-r k ) T W k (x(t k )-r k ) s.t. dx/dt=f(x,u,d), t [t 0, t P ], x(t 0 ) given, u j =u(t j )=u(t), t [t j, t j+1 ], u j =u M-1, j [M, P-1], L≤u≤V, scale t=t j+1 -t j =1. Nonlinear LS: min L≤U≤V Φ=½E(U) T E(U) Jacobian: J(U)=∂E/∂U Gradient: G(U)=J T (U)E(U) Hessian: H(U)=J T (U)J(U)+Q(U)≈J T (U)J(U)
10
4th April 2005Colloquium on Predictive Control, Sheffield10 ODE and Jacobian using AD Efficient algorithm requires efficient J(U) Difficult: E(U) is nonlinear dynamic J i,j =W i ½ dx(t i )/du j-1 for i≥j, otherwise, J i,j =0 Algorithm: for k=0:P-1, x 0 =x(t k ) Forward AD: x i, i=1,…,d, → x(t k+1 ) Reverse AD: A ix, A iu Accumulate: B ix, B iu, → B u (k)= B iu, B x (k)= B ix J (k+1)j =W k+1 ½ B x (k)…B x (j)B u (j-1), j=1,…,k+1 K=k+1
11
4th April 2005Colloquium on Predictive Control, Sheffield11 NLS MPC using AD Collect information: x, d, r, etc. Nonlinear LS to give a guess u Solve ODE and calculate J Update u and check convergence Implement the first move
12
4th April 2005Colloquium on Predictive Control, Sheffield12 Error Analysis Taylor coefficients by AD is accurate. x(t k )= x k has truncation error, e k (local). e k will propagated to k+1, …, P (global). Local error controllable by order and step Global error depend on sensitivity dx k1 /dx k Remainder: k ≈C(h/r) k+1 Convergence radius: r ≈ r k =|x k-1 |/|x k | k-1 = k (r/h)= k +|x k | → k =|x k |/(r/h-1)
13
4th April 2005Colloquium on Predictive Control, Sheffield13 Error Control Tolerance < d Increase order, d or decrease step, h? Decrease h by h/c (c>1): = d (1/c) d+1 c=( d / ) 1/(d+1), increase op by factor c Increase d to d+p (p>0): = d (h/r) p p=ln( / )/ln(h/r), increase op by (1+p/d) 2 c<(1+p/d) 2 decrease h, otherwise increase d
14
4th April 2005Colloquium on Predictive Control, Sheffield14 Case Study Evaporator process Evaporator process Evaporator process 3 measurable states: L2, X2 and P2 3 manipulates: 0≤F2≤4, 0≤P100,F200≤400 Set point change:X2 from 25% to 15% P2 from 50.5 kPa to 70 kPa Disturbance: F1, X1, T1 and T200 20% All disturbance unmeasured. T=1 min, M=5 min, P=10 min, W=[100,1,1]
15
4th April 2005Colloquium on Predictive Control, Sheffield15 Evaporation Process
16
4th April 2005Colloquium on Predictive Control, Sheffield16 Simulation Results
17
4th April 2005Colloquium on Predictive Control, Sheffield17 Performance Comparison CVODES, a state-of-the-art solver for dynamic sensitivity. Simultaneously solves ODE and sensitivity Two approaches: full & partial integration. Three approaches programmed in C Tested on Windows XP P-IV 2.5GHz Solve evaporator ODE + sensitivity using input generated by NMPC.
18
4th April 2005Colloquium on Predictive Control, Sheffield18 Accuracy and Efficiency Taylor AD CVODES P CVODES F ToldtimeerrortimeerrortimeError 1e-44.0087e-5.0224e-4.8124e-3 1e-66.0097e-8.0429e-51.648e-5 1e-88.0114e-11.0632e-62.362e-6 1e-1111.0135e-13.1146e-94.563e-11
19
4th April 2005Colloquium on Predictive Control, Sheffield19 Conclusions AD can play an important role to improve nonlinear model predictive control Efficient algorithm to integrate ODE at the same time to calculate sensitivity Error analysis and control algorithm Efficiency validated via comparison with state-of-the-art software. Satisfactory performance with Evaporator study
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.