At the moment we have developed a set of random genetic optimization strategies and a corresponding software implementation. The tests have been performed that show the algorithm is highly capable and surpasses the common search methods.
The importance of theoretical and applied results is that using a genetic approach we have partially solved the local and global search incompatibility problem for poorly determined high dimensionality models.
It is well known that the basic features of optimization algorithms are:
- Versatility: a principal possibility of applying it to a wide range of problems like:
- smooth continuous optimization problems;
- integer optimization problems;
- discrete optimization problems;
- mixed optimization problems (a mixture of the first three types).
- Finding the global solution;
- Maximum number of variables being optimized;
- Number of model (goal function) calls;
- Optimization accuracy by the parameters and by the goal;
- Computational complexity of the optimization algorithm itself (time required for performing the computational procedures).
The proposed algorithm simultaneously yields high quality by all the above listed features; by each feature it is significantly better than the existing optimization methods and it is suitable for all listed types oа problems.
Another important feature is that the algorithm gives stable optimal solutions for partially undetermined models (when for some parameters’ value the object’s model does not exist or the process in such a state is not defined yet or the model cannot be computed like division by zero, negative radicand, undefined arcos, arctg trigonometric functions, etc.)
The proposed random genetic algorithm solves all problems in a reliable and accurate way.
A set of tests is provided to prove it
(download TEST_exe.rar TEST_exe.rar). The algorithm’s performance can be tested with any arbitrary problem.
