Neural net solutions to fuzzy linear programming (Q1304321): Difference between revisions

The authors show the possibility of employing neural networks to solve the following fuzzy linear programming problem (FLPP): \[ \text{Max} (\overline{Z})= \overline{C_1}x_1+ \overline{C_2}x_2, \] \[ \text{subject to }\overline{A_{11}} x_1+ \overline{A_{12}} x_2\leq \overline{B_1}, \quad \overline{A_{21}} x_1+ \overline{A_{22}} x_2\leq \overline{B_2}, \quad x_1,x_2\geq 0, \] where \(\overline{C_j}\), \(\overline{A_{ij}}\), \(\overline{B_i}\), for \(i,j\in \{1,2\}\), are the triangular fuzzy numbers. The optimal value of the objective function of FLPP is called max fuzzy set for FLPP. It is defined as a fuzzy number \(\overline{Z}\), such that if crisp values of all parameters of the above model are selected from the proper \(\alpha\)-levels, then the optimal value of the objective function of such crisp model belongs to the \(\alpha\)-level of \(\overline{Z}\). Similarly, the joined solution is a fuzzy vector \(\overline{X}\) in \(R^2\), such that if crisp values of all parameters are selected from their \(\alpha\)-levels, then the optimal solution \((x_1, x_2)\) obtained with these values belongs to the \(\alpha\)-level of \(\overline{X}\). Two neural networks are proposed, the first one to approximate the max fuzzy set \(\overline{Z}\) and the second one to approximate the joined solution \(\overline{X}\). Proposed networks have sign constraints on their weights, so the special method have to be used in training (method based on the evolutionary programming is described in appendix). Presented results of some computational experiments show that solutions obtained with proposed neural networks approximate well the solutions of considered problem. In conclusions the authors inform, that their future research will be concerned with the design the neural networks for a general (bigger) fuzzy linear programming problem.