System and method for stochastic simulation of nonlinear dynamic systems with a high degree of freedom for soft computing applications

• US 20040039555A1
• Filed: 07/30/2002
• Published: 02/26/2004
• Est. Priority Date: 07/30/2002
• Status: Abandoned Application

• Current Assignee: Yamaha Hatsudoki Kabushiki Kaisha (Yamaha Corporation)
• Original Assignee: Yamaha Hatsudoki Kabushiki Kaisha (Yamaha Corporation)
• Inventors: Ulyanov, Sergei V., Panfilov, Sergei
• Application Number:
  US10/209,636
• Publication Number:
  US 20040039555A1
• Time in Patent Office: Days
• US Class Current: 703/2
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: G06F 17/10 Complex mathematical operations function generation by table look-up G06F1/03; evaluation of elementary functions by calculation G06F7/544

Abstract

A system and method for efficient stochastic simulation of dynamic systems is described. Since analytic solutions cannot usually be found for stochastic differential equations, complete analysis requires numerical simulations. These simulations are most commonly done with first-order Euler-type algorithm. The efficiency of these algorithms is improved by removing algebraic loops in the simulation. An algebraic loop occurs when an output variable of the system of equations is also in an input variable to one or more of the equations describing the system. In one embodiment, the algebraic loops are removed by formulating a simulation wherein an output variable that gives rise to an algebraic loop is integrated to produce an integrated output. The integrated output is later provided to a differentiator to reconstruct the output variable as needed.

Claims

1. An efficient method for numerical integration for use in simulation of non-linear differential equations with essential non-linearities including higher-order derivatives:

comprising;

providing one or more input variables to a system of equations;

computing one or more outputs from said system of equations using said input variables;

integrating at least one selected output to produce an integrated output;

differentiating said integrated output to produce a reconstructed selected output; and

providing said reconstructed selected output as a next input to said system of equations.

2. A method for stochastic simulation of non-linear differential equations with non-linearities including higher order derivatives, comprising:

defining a system of non-linear differential equations having an algebraic loop wherein an output variable of at least one equation is also an input to said at least one equation, said output variable corresponding to an n-th derivative of a quantity represented by said output variable;

defining a simulation system that removes said algebraic loop by;

integrating said output variable to produce an integrated output variable, said integrated output variable corresponding to an (n−

1)-th derivative of said quantity represented by said output variable;

providing said integrated output variable to an input of said at least one equation; and

differentiating said integrated output variable and providing an output of said integration to an input of said at least one equation; and

using an Euler-type method to numerically evaluate said simulation system.

3. The method of claim 2, further comprising:

providing one or more inputs to a control system;

computing a control output from said one or more inputs to said control system; and

providing said control output to at least one onput of said system of equations.

4. The method of claim 2, further comprising:

generating a control signal using a controller, said controller receiving input from a first information signal, said first information signal comprising at least one variable from said system of equations;

computing an entropy from said information signal;

computing a teaching signal using said entropy;

teaching said controller using said teaching signal.

5. The method of claim 4, further comprising:

using said teaching signal to train a neural network.

6. The method of claim 5, wherein said teaching signal is computed by a genetic analyzer.

7. The method of claim 6, wherein a fitness function of said genetic analyzer is based on said information signal.

8. The method of claim 6, wherein a fitness function of said genetic analyzer is configured to reduce an entropy of said first information signal.

9. The method of claim 5 wherein said neural network is a fuzzy neural network.

10. The method of claim 5, wherein said neural network is a fuzzy neural network trained by said teaching signal.

11. The method of claim 5, wherein computing said teaching signal comprises running a genetic analyzer having a fitness function that reduces an entropy of said system of equations.

12. A simulation system for simulating control of a plant described as a system of non-linear differential equations, comprising:

a plant simulation module configured to compute one more plant outputs from a system of equations based on one or more plant variables, wherein output variables from said system of equations that also appear as inputs to said system of equations are first integrated and then differentiated before being provided as inputs to said system of equations;

means for generating a teaching signal by computing said teaching signal to produce control that reduces an entropy of said plant;

means for generating a gain schedule as directed by said teaching signal; and

control means to generate a control signal using at least one of said plant variables and said gain schedule.

13. The control system of claim 12, wherein said means for generating a gain schedule comprises a genetic analyzer.

14. An apparatus for simulating control of a plant described as a system of non-linear differential equations, comprising:

a plant simulation module configured to compute one more plant outputs from a system of equations based on one or more plant variables, wherein output variables from said system of equations that also appear as inputs to said system of equations are first integrated and then differentiated before being provided as inputs to said system of equations; and

a multiplexer configured to provide said inputs to said system of equations according to a simulation algorithm.

15. The apparatus of claim 14, wherein said simulation algorithm is a first-order Euler algorithm.

16. The apparatus of claim 14, wherein said simulation algorithm is a Runge-Kutta algorithm.

17. The apparatus of claim 14, further comprising:

an analyzer for generating a teaching signal by computing said teaching signal to produce control that reduces an entropy of said plant;

a fuzzy logic classifier module for generating a gain schedule as directed by said teaching signal; and

a control module to generate a control signal using at least one of said plant variables and said gain schedule.

18. The control system of claim 17, wherein said means for generating a gain schedule comprises a genetic analyzer.

19. A self-organizing method for simulating control of a nonlinear plant described by one or more differential equations, comprising:

obtaining a difference between a time differentiation (dS_u/dt) of the entropy of a plant and a time differentiation (dS_c/dt) of the entropy provided to the plant from a low-level controller that controls the plant;

evolving a control rule by evolution in a genetic algorithm, said genetic algorithm using said difference as a fitness function;

removing algebraic loops from said simulation by integrating outputs from said system of equations that also appear as inputs to said system of equations to produce integrated outputs, differentiating said integrated outputs to produce reconstructed inputs, providing said reconstructed inputs to said system of equations, simulating operating of said non-linear plant by computing new inputs for said system of equations from previous outputs of said system of equations according to the method of Euler or the Runge-Kutta method.

20. The method of claim 19, further comprising:

analyzing one or more nonlinear operation characteristics of said physical plant by using a Lyapunov function; and

correcting said control rule based on an evolution.

21. The method of claim 19, further comprising:

evolving a control rule relative to a variable of said low-level controller by using a genetic algorithm, said genetic algorithm using fitness function that reduces a difference between a time differentiation of an entropy of said plant (dS_u/dt) and a time differentiation (dS_c/dt) of an entropy provided to said plant from said low-level controller; and

correcting a variable of said low-level controller based on said evolved control rule.

22. A control apparatus adapted to control a non-linear plant, comprising:

a simulator configured to use a system of non-linear differential equations to simulate operation of a non-linear plant according to the method of Euler, wherein outputs from said system of equations that also appear as inputs to said system of equations are first integrated and then differentiated before being provided as inputs to said system of equations;

an entropy calculator that calculates an entropy production amount based on a difference between a time differentiation of entropy of said plant (dS_u/dt) and a time differentiation (dS_c/dt) of an entropy provided to said plant from a low-level controller that controls said plant;

a genetic algorithm module that obtains an adaptation function in which said difference is minimized; and

a fuzzy logic classifier configured to determine a fuzzy rule by using a learning process, said fuzzy logic controller configured to use an output from said genetic algorithm as a teaching signal, said fuzzy logic controller further configured to form a control rule that sets a variable gain of said controller by following said fuzzy rule.

23. The apparatus of claim 22, wherein said fuzzy logic classifier comprises:

a fuzzy neural network configured to form a look-up table for said fuzzy rule by using said learning process; and

a fuzzy controller configured to generating a variable gain schedule for said controller that controls said plant.

24. The apparatus of claim 22, wherein said low-level controller is a linear controller.

25. The apparatus of claim 22, wherein said low-level controller is a PID controller.

26. An apparatus for simulation of non-linear differential equations with non-linearities including higher order derivatives, comprising:

an equation module for computing a system of non-linear differential equations wherein an output variable of at least one equation is also an input to said at least one equation, said output variable corresponding to an n-th derivative of a quantity represented by said output variable;

an integrator module configured to integrate said output variable to produce an integrated output variable, said integrated output variable corresponding to an (n-1)-th derivative of said quantity represented by said output variable;

a differentiator module configured to differentiate said integrated output variable to reconstruct said output variable as a reconstructed output variable; and

a multiplexer configured to receive said integrated output vairable and said reconstructed output variable and to compute new inputs for said equation module according to a solution method.

27. The apparatus of claim 26, wherein said solution method is an Euler method.

28. The apparatus of claim 26, wherein said solution method is a Runge-Kutta method.

29. The apparatus of claim 26, wherein said system of non-linear differential equations describes a unicycle.

30. The apparatus of claim 26, wherein said system of non-linear differential equations comprises a simulation model of a unicycle.

31. The apparatus of claim 26, wherein said system of non-linear differential equations comprises a simulation model of a suspension system.

32. The apparatus of claim 26, wherein said system of non-linear differential equations comprises a simulation model of a suspension system in the presence of a stochastic road signal.