If it is necessary to find multiple optima or search for a global optimum there are two methods available to do this, non-random initialization and enforced diversity. If a good estimate of the solution is available, it makes no sense to randomly initialise a population, seeding around this solution is a much better practice. In large and complicated but regular spaces (i.e. locally continuous), one or two passes can be made to heuristically search for good areas in which to perform the optimisation. An example would be to place chromosomes on a regular grid, assess their fitness and then assign a number of search chromosomes to each area based on relative fitness.
One method of ensuring a good spread of chromosomes throughout the space is to test the diversity of the population and reject similar chromosomes. This method is not widely adopted since good measures of diversity are difficult to formulate. This is a problem on which further work is now being done.