A dual adaptive control theory inspired by Hebbian associative learning. The resultant dual adaptive control scheme is shown to achieve superior tracking performance. Questions of biological realism are important for the sake of physiological validity.
A dual adaptive control theory inspired by Hebbian associative learning. The resultant dual adaptive control scheme is shown to achieve superior tracking performance. Questions of biological realism are important for the sake of physiological validity.
A dual adaptive control theory inspired by Hebbian associative learning. The resultant dual adaptive control scheme is shown to achieve superior tracking performance. Questions of biological realism are important for the sake of physiological validity.
=500 (a) 0 20 40 60 80 100 ~0.015 ~0.01 ~0.005 0 0.005 0.01 0.015 Time [s] T r a c k i n g
e r r o r
e
=1000 (b) 0 20 40 60 80 100 ~0.015 ~0.01 ~0.005 0 0.005 0.01 0.015 Time [s] T r a c k i n g
e r r o r
e
=1500 (c) Fig. 2. Simulation for Example 1. 2(a), 2(b) and 2(c) are about the tracking error for , and . These show that the tracking error for conveges faster than those of and . Noting that error equations (13) and (7) have essentially the same structure, we obtain the following convergence rule for DAC, by direct applicating of Lemma 2. Theorem 1. (Convergence rule for DAC with given / ) For any positive j 0 and / 0, if the DAC system (13) all conditions in Lemma 2, then (13) is exponentially stable. The advantages of introducing of error feedback into MRAC will be shown via the following example. Example 2. Consider the system in Example 1. In order to improve the speed of tracking, we introduce error feedback. Figure 4 shows the tracking errors for error feedback gain / = 100 (pink line) and / = 0 (black line) under adaptation gain j = 1000. It can be seen that the tracking error converges much faster with error feedback than without. 4508 e d e e e r u 0 y p y m W 1 W 0 W 2 Plant G + + _ Fig. 3. Dual Adaptive Control 0 20 40 60 80 100 ~0.015 ~0.01 ~0.005 0 0.005 0.01 0.015 time [s] t r a c k i n g
e r r o r
e
k 0 =0 k 0 =100 Fig. 4. Tracking error curves for and under adaptation gain . Compared with the tracking error without error feedback gain (black line), the tracking error using error feedback converges much faster. Remark 1. Advantages of introducing error feedback into MRAC: (1) It can improve the convergence rate, which cannot be achieved by simply increasing adaptation gain j; (2) It can reduce overshoot, as shown in Example 2. B. Robustness analysis Adaptive systems are known to be sensitive to unknown disturbances or uncertainties in the system structure. Such non-robust behavior may be exacerbated by over-adaptation with excessive adaptation gains. Therefore, robustness of adaptive control has received much attention in recent years (see [9], [10], [15], [16], [33] and the references therein). In this subsection, we shall examine the robustness of DAC, which will be accessed by subjecting the system to unknown disturbance input, unknown system dynamics and high- frequency reference input. 1) Unknown disturbance input. For simplicity, we study plants with disturbance input: ` j ¡ = ÷o ¡ j ¡ ÷ d ÷ / ¡ n (14) where d is an unknown nonzero constant, which can be regarded as a constant disturbance input. Furthermore, the corresponding model reference is ` j n = ÷o n j n ÷ / n v(t) (15) where v(t) is reference input, o n and / n are known constant parameters, and o n is required to be strictly positive. Without loss of generality / n can be chosen strictly positive. In such case, the dual adaptive controller becomes n = ´ o : (t)v ÷ ´ o ¸ (t)j ¡ ÷ ´ o J ÷/ c (16) where ´ o : , ´o } and ´o J are variable feedback gains. The adaptive law is ` ´o : = ÷:qn(/ ¡ )jcv ` ´o ¸ = ÷:qn(/ ¡ )jcj ¡ ` ´o J = ÷:qn(/ ¡ )jc. (Note when / = 0, controller (16) with the above adaptive law is equivalent to MRAC [29]). It is obvious that the controller above is robust for any constant disturbance input d. For high-order systems, j ¡ = / ¡ (:)n ÷ / } (:)d, where d is unknown constant and / ¡ (:) satisfying same conditions as before. Then the corresponding control law still has the form of (11), but there / = / / ´o J i T i T T , . = v j ¡ 1 · T · T T . Furthermore, / and . still satisfy (3), where A = dioq ` ` ... ` n with ` I 0, i = 1, 2, ..., 2n ÷ 1. 2) Fast-adaptation instability. For convenience of discus- sion, we will our analysis to the following example of plant with unknown dynamics: ` a = ÷a ÷ /n ÷ n j ` n = ÷n ÷ 2n j = a (17) where / 1,2 is a constant parameter, and j is a small positive number which corresponds to the time constant of an un-modeled state n. The control objective is to track the output a n of a reference model ` a n = ÷a n ÷ v. (18) For j = 0, the MRAC law n = /v, ` / = ÷jcv, c = a ÷ a n (19) guarantees that c ÷0 as t ÷· and all signals are bounded for any bounded input v. For 1 j 0, the closed-loop plant is: ` c = ÷c ÷ /n ÷ (/ ÷1)v j ` n = ÷n ÷ 2/v ` / = ÷jvc. (20) For large j the above MRAC system may become unstable. In particular, for any v = con:tont the system is linear and time invariant. Hence, it can be shown by using the Routh- Hurwitz criterion that (20) is stable iff the adaptation gain j jv < 1 j 1 ÷ j 2/ ÷j - 1 2/j (21) where the approximation in the above inequality is from j << 1. 4509 Now suppose that in addition to the above MRAC law there is a negative error feedback with a gain / 1. The two equations in (20) becomes ` c = ÷(1 ÷/ )c ÷ /n÷(/ ÷ 1)v j ` n = ÷n ÷ 2/v ÷2/ c. (22) Using the Routh-Hurwitz criterion as before it can be shown that the dual adaptive control system is stable iff jv < 1 j (1 ÷ / ÷ 2// )[1 ÷ (1 ÷/ )j[ 2/ ÷(1 ÷/ )j . (23) For / = 0 the stability condition is identical to (21). For b << / < ÷1 ÷ 2/,j, however, the right-hand side of (23) tends to / ,j 1,2/j. This result shows that the dual adaptive control system is considerably more robust than MRAC. V. CONCLUSION Combining HGAC and MRAC, we have formulated DAC which outperforms either of the controller implemented alone. DAC scheme has proved superior in terms of conver- gence rate and robustness properties, which are demonstrated with several simulation results in this paper. Furthermore, DAC also shows to be promising in dealing with a class of nonlinear systems, which is omitted since the space limited. REFERENCES [1] B. R. Dworkin, Learning and Physiological Regulation, University of Chicago Press, Chicago, 1993. [2] D. O. Hebb, The Organization of Behavior, Wiley, New York, 1949. [3] T. H. Brown, P. F. Chapman, E. W. Kairiss and C. L. Keenan, ¨Long- term synaptic potentiation,¨ Science, vol. 242, pp. 724-728, 1988. [4] D. E. Rumelhart, G. E. Hinton and R. J. William, ¨Learning internal representations by error propagation.¨, In Rumelhard, D.E., McClel- land, J. L., and the PDP Research Group, editors, Paralled Distributed Processing. Explorations in the Microstructure of Cognition. Volume 1. Foundations, pp. 318-362. The MIT Press, Cambridge, MA. [5] P. J. Werbos, ¨Neural networks for control and system ¨ Proceedings of the 28th Conference on Decision and Control, FL, USA, vol. 1, pp. 260-265, 1989. [6] J. V. Shah and C.-S. Poon, ¨Linear independence of internal represen- tations in multilayer perceptrons,¨ IEEE Trans. Neural Network, vol. 10, pp. 10-18, 1999. [7] B. D. O. Anderson, R. R. Bitmead, C. R. Johnson, Jr., P. V. Kokotovi´ c, R. L. Kosut, I. M. Y. Mareels, L. Praly, and B. D. Riedle, Stability of adaptive systems. passivity and averaging systems, ser. MIT Press Series in Signal Processing, Optimization, and Control, 8. Cambridge, MA: MIT Press, 1986. [8] K. Astr¨ om and B. Wittenmark, Adaptive Control, 2nd ed. MA : Addison Wesley, 1995. [9] G. Bartolini, A. Ferrara, and A. A. Stotsky, Robustness and perfor- mance of an indirect adaptive control scheme in presence of bounded disturbances,¨ IEEE Trans. Automat. Control, vol. 44, no. 4, pp. 789- 793, 1999. [10] M. Cadic and J. W. Polderman, Strong robustness in adaptive con- trol,¨ in Nonlinear and adaptive control (Shef 2001), ser. Lecture Notes in Control and Inform. Sci. Berlin: Springer, 2003, vol. 281, pp. 45-54. [11] M. K. Ciliz and A. Cezayirli, Increased transient performance for the adaptive control of feedback linearizable systems using multiple models,¨ Internat. J. Control, vol. 79, no. 10, pp. 1205-1215, 2006. [12] A. Ilchmann, -based adaptive control of dynamical systems: a survey,¨ IMA J. Math. Control Inform., vol. 8, no. 4, pp. 321-366, 1991. [13] A. Ilchmann and D. H. Owens, Threshold switching functions in high-gain adaptive control,¨ IMA J. Math. Control Inform., vol. 8, no. 4, pp. 409-429, 1991. [14] P. A. Ioannou and P. V. Kokotovi´ c, Instability analysis and improve- ment of robustness of adaptive control,¨ Automatica J. IFAC, vol. 20, no. 5, pp. 583-594, 1984. [15] P. Ioannou and J. Sun, Robust adaptive control. Prentice Hall, Upper Saddle River, NJ, 1996. [16] P. Ioannou and A. Datta, Robust adaptive control: design, analysis and robustness bounds,¨ in Foundations of adaptive control (Urbana, IL, 1990), ser. Lecture Notes in Control and Inform. Sci. Berlin: Springer, 1991, vol. 160, pp. 71-152. [17] H. Kaufman, I. Barkana, and K. Sobel, Direct adaptive control algorithms, 2nd ed., ser. Communications and Control Engineering Series. New York: Springer-Verlag, 1998, theory and applications, Case studies contributed by David S. Bayard and Gregory W. Neat. [18] H. K. Khalil and A. Saberi, Adaptive stabilization of a class of nonlinear systems using high-gain feedback,¨ IEEE Trans. Automat. Control, vol. 32, no. 11, pp. 1031-1035, 1987. [19] I. Mareels and J. W. Polderman, Adaptive systems. An introduction. Boston, MA: Birkh¨ auser Boston Inc., 1996. [20] R. Marino and P. Tomei, Adaptive observers with arbitrary exponen- tial rate of convergence for nonlinear systems,¨ IEEE Trans. Automat. Control, vol. 40, no. 7, pp. 1300-1304, 1995. [21] --, Nonlinear control design. geometric, adaptive, and robust. Prentice Hall International (UK) Limited, 1995. [22] A. S. Morse, A three-dimensional universal controller for the adaptive stabilization of any strictly proper minimum-phase system with relative degree not exceeding two,¨ IEEE Trans. Automat. Control, vol. 30, no. 12, pp. 1188-1191, 1985. [23] D. Mudgett and A. Morse, A smooth, high-gain adaptive stabilizer for linear systems with known relative degree,¨ in Proc. Automatic Control Conf., Kyoto, Japan, 1989, pp. 2315-2320. [24] K. Narendra and A. Annaswamy, Stable Adaptive Systems, 2nd ed., ser. Applied Mathematical Sciences. Englewood Cliffs, NJ: Prenctice Hall, 1989. [25] K. S. Narendra and J. Balakrishnan, Improving transient response of adaptive control systems using multiple models and switching,¨ IEEE Trans. Automat. Control, vol. 39, no. 9, pp. 1861-1866, 1994. [26] R. D. Nussbaum, Some remarks on a conjecture in parameter adaptive control,¨ Systems Control Lett., vol. 3, no. 5, pp. 243-246, 1983. [27] I. Polderman, JW. Mareels, Rob of high gain adaptive control,¨ in Proceedings of the 6th European Control Conference, Porto, Portugal, 2001, pp. 2428-3333. [28] --, Two scale high gain adaptive control,¨ Int. J. Adapt. Control Signal Process, vol. 18, pp. 393-402, 2004. [29] J.-J. E. Slotine and W. Li, Applied nonlinear control. Prentice Hall: Englewood Cliffs, New Jersey, 1991. [30] A. Stotsky, Lyapunov design for convergence rate improvement in adaptive control,¨ Int. J. Control, vol. 57, no. 2, pp. 501-504, 1993. [31] J. Sun, A model reference adaptive control scheme for im- proved transient performance,¨ IEEE Trans. Automat. Control, vol. 38, no. 8, pp. 1255-1259, 1993. [32] J. C. Willems and C. I. Byrnes, Global adaptive stabilization in the absence of information on the sign of the high frequency gain,¨ in Analysis and optimization of systems, Part 1 (Nice, 1984), ser. Lecture Notes in Control and Inform. Sci. Berlin: Springer, 1984, vol. 62, pp. 49-57. [33] X. Xie and J. Li, A robustness analysis of discrete-time direct model reference adaptive control,¨ Int. J. Control, vol. 79, no. 10, pp. 1196- 1204, 2006. [34] B. E. Ydstie, Transient performance and robustness of direct adaptive control,¨ IEEE Trans. Automat. Control, vol. 37, no. 8, part I, pp. 1091-1105, 1992. 4510