A unified parameter identification method for nonlinear time-delay systems
Summary (1 min read)
4. Numerical examples.
- For comparison, the authors also solve this problem using the genetic algorithm (GA) in [18] .
- The parameters of GA are: the size of population is 20, the crossover probability is 0.8, the selection rate is 0.9, the mutation probability is 0.01, the number of bits for each individual is 14, and the maximum number of iterations is 1000.
- It takes about 40 minutes for GA to solve this problem, which is more than 20 times longer than the computation time taken by their method.
- Moreover, the cost value obtained by GA is 1.3787 × 10 −9 with corresponding parameter estimates Clearly, the results obtained by their new method are better than those from GA.
- This is not surprising, as their method exploits the gradient of the cost function to achieve fast convergence.
4.2. Example 2.
- The authors use the output trajectory of ( 52 The authors solved this problem using the same Matlab program that was used to solve Example 1.
- The convergence progress of their program is shown in Table 3 for four sets of initial guesses.
- The convergence of the output trajectory for the initial guess τ.
- Moreover, the computation time is much longer than their new method.
- As with Example 1, the authors see that the optimization process converges from all initial guesses to the optimal solution.
5. Conclusion.
- The authors have developed a gradient-based computational method for determining unknown time-delays and unknown parameters in a general nonlinear system.
- This method is unified in the sense that the delays and parameters are determined simultaneously by solving a dynamic optimization problem.
- The numerical simulations in Section 4 demonstrate that this approach is highly effective.
- In particular, it converges quickly even when the initial estimates for the delays and parameters are far away from the optimal values.
Did you find this useful? Give us your feedback
Citations
226 citations
Cites background from "A unified parameter identification ..."
...Gradient formulae for the cost function with respect to the time-delays are derived in [10, 11, 46]....
[...]
...References [10, 11, 46] consider the problem of choosing the delays to minimize the deviation between predicted...
[...]
90 citations
Additional excerpts
...[4, 5, 6, 18, 19]....
[...]
64 citations
Additional excerpts
...It can be shown (see, for example [40, 41]) that the state x(t|u, τ, α) is differentiable with respect to α and ∂x(t|u, τ, α) ∂α = φ(t|u, τ, α), t ∈ (−∞, tf ]....
[...]
53 citations
39 citations
References
4,908 citations
3,206 citations
"A unified parameter identification ..." refers background in this paper
...The problem of estimating the time-delays (and possibly other unknown system parameters) from a given set of experimental data is one of the key problems in the study of time-delay systems [20]....
[...]
238 citations
"A unified parameter identification ..." refers background in this paper
...Time-delay systems arise in many important applications, including medicine [21], chromatography [24], aerospace engineering [5], and chemical reactor control [9]....
[...]
236 citations
"A unified parameter identification ..." refers background or methods in this paper
...However, the computational techniques developed in references [13, 17, 22] are not applicable to Problem (P) because the time-delays in system (1)-(2) are variable....
[...]
...Such cost functions have been considered in [13, 17] for non-delay systems, and in [22] for systems with fixed delays....
[...]
182 citations
"A unified parameter identification ..." refers background or methods in this paper
...One of the limitations of the existing identification methods in [2, 6, 7, 8, 18, 26] is that they are mainly designed for systems with a single input delay and no unknown parameters....
[...]
...In [26], information theory is used to identify time-delays for systems in which each nonlinear term contains at most one unknown delay....
[...]
Related Papers (5)
Frequently Asked Questions (10)
Q2. What is the purpose of this process?
The purpose of this process is to remove harmful cobalt and cadmium ions from a zinc sulphate electrolyte by adding zinc powder to encourage deposition.
Q3. how many ions are in the reaction tank?
V is the volume of the reaction tank (V = 400); Q is the flux of solution (Q = 200); α, β, c, d, are model parameters; and x01 and x 0 2 are the concentrations of cobalt and cadmium ions at the inlet of the reaction tank, respectively (x01 = 6×10−4, x02 = 9×10−3).
Q4. How long does it take for GA to solve this problem?
It takes about 40 minutes for GA to solve this problem, which is more than 20 times longer than the computation time taken by their method.
Q5. What is the way to solve the problem?
In this paper, the authors have developed a gradient-based computational method for determining unknown time-delays and unknown parameters in a general nonlinear system.
Q6. How is the gradient of the cost function obtained?
The gradient of the cost function in this problem is obtained by solving a set of auxiliary delay-differential systems from t = 0 to t = T .
Q7. What grants were awarded to this work?
This work was supported by grants from the Australian Research Council (grant no. DP110100083), the China National Science Fund for Distinguished Young Scholars (grant no. 61025015), and the National Natural Science Foundation of China (grant no. 61174133).
Q8. What is the output trajectory of the problem?
The authors use the output trajectory of (52)-(54) with [τ̂1, τ̂2, τ̂3] = [2.4, 1.8, 1.1] > to generate the observed data for Problem (P).
Q9. What is the simplest way to solve this problem?
Their identification problem is: choose τ and α to minimizeJ(τ, α) = 16∑ l=1 ∣∣y(tl|τ, α)− ŷl∣∣2 = 16∑ l=1 ∣∣x2(tl|τ, α)− x2(tl|τ̂ , α̂)∣∣2 subject to the dynamic system (45)-(47).
Q10. What is the simplest way to solve the problem of a dynamic system?
Consider the dynamic system given below:ẋ1(t) = −2x1(t) + 0.1(1− x1(t− τ1)) exp { 20x2(t)20 + x2(t) } + 0.1x1(t− τ1)x2(t− τ2) + u(t− τ3),(52)ẋ2(t) = −2.5x2(t) + 0.8(1− x1(t− τ1)) exp { 20x2(t)20 + x2(t) } + 0.1x2(t− τ1)x2(t− τ2) + u(t− τ3),(53)with initial condition x1(t) = 1, x2(t) = 1, t ≤ 0. (54)