Identification > Methods of solution of identification problem

To solve the identification problem of blood circulation system we need to choose the optimization method. This choice is one of the most complex scientific problems. This problem demands the carrying out of the whole complex of theoretical and experimental researches connected first of all with finding-out of an opportunity of the effective decision (ambiguity or unambiguity of the decision, speed of the algorithm convergence and etc.). The question of unambiguity of the decision is closely connected with the questions of observability, identifiability and the question of the choice of measuring system.

In the work [9] various methods of the decision of corresponding optimization problem, including search methods. The choice of method for computer realization considered in [25]. In an a priori decision ambiguity of the task of identification as the preferred method seems appropriate to apply statistical ( based on a random
spreading ) algorithm, which is based on a well-known algorithm for a random global search [17].

Method of global random search. Strongly nonlinear character of the optimization problem makes a choice of the algorithm of global random search more preferable for the computer implementation as a method of the computational solution of corresponding optimization problem. Among the variety of algorithms of random search we have chosen the accidental algorithm with local random search [17]. As against the method of local direct search, it is intended for the solution of the problem of search of a global extremum of object functional (estimation criterion).

Algorithm

Step 1. In determining the parameters, according to some initial probability distribution (usually uniform), scatters m random points of the original sample S0 . Step 2. Look over all points of the original sample and from each point perform a random search for global minimum of criterion function identification. Get the resulting sample S1 from the set of global extremum criterion function identification. Step 3. From the resulting sample S1 form a statistical description of the set of solutions of problem identification. This estimate is calculated as the vector of parameters of a sampling medium or as a point of sampling minimum functional criteria for identification. For the resulting sample is also constructed random distribution function parameters, histograms and other standard statistical estimates.

Computer implementation

Consider the most significant features of the software realization of the algorithm. Ease of implementation. Ease of implementation and debugging, reliability, and immunity is well-known properties of random search algorithms. Large amount of computation. The statistical algorithms to achieve acceptable accuracy requires large amounts of sample and the large amounts of computation. Furthermore, in our case, to achieve the periodic motion of the object should be modeling for a sufficiently long period of time, which also requires an increase in computation. Ability to implement parallel computations. Computational procedure for a step 2 algorithm is a finite collection of independent of one another identical computational procedures that can be performed in any order. Known that such algorithms are best suited for implementation on multiprocessor (multicore) computer systems and in homogeneous computing environments.