Hard problem generation for multidimensional knapsack problem

We developed generators that produce challenging Multidimensional Knapsack Problem instances. Our approaches use independently exponential distributions over a wide range to generate the constraint coefficients, and the corresponding average for each variable is used to calculate directly correlated coefficients in the objective function, RHS values are a percentage of the sum of constraint coefficients. We present a comparative table with the average performance of the most important generators reported in the literature and our generators over a wide range of parameters and instances in the OR library.