HOW TO APPLY ROULETTE WHEEL SELECTION IN CASE OF MINIMIZATION PROBLEM FOR LARGE POPULATION

My GA problem is a minimization problem where i need to choose that chromosome from the population (X) whose features (A)are close to the input(desired) features(B). The objective function (Ob_function) is sqrt((A(i)-B)^2).; I=1:1000 ; size of X say= 1000 x 500; A=1000x4; B=1x4; Ob_function=1x1000; What i am doing is : 1. calculating ob_function 2. apply Roulette wheel on the ojective function and arrange population according to roulette wheel selection 3. uniform crossover on population 4. mutation 5. then repeat step 1 until objective function is zero. my problem : error varies after every iteration but it has not reached even close to zero, In certain cases, the objective function was near to zero but after applying GA the difference increases... I think the problem is in selection of population ...please help