Forming a portfolio of different stocks instead of buying a particular type of stock can reduce the potential loss of investing in the stock market. Although forming a portfolio based solely on past data is the main theme of various researches in this field, considering a portfolio of different stocks regardless of their future return can reduce the profits of investment. The aim of this paper is to introduce a new two-phase approach to forming an optimal portfolio using the predicted stock trend pat-tern. In the first phase, we use the Hurst exponent as a filter to identify stable stocks and then, we use a meta-heuristic algorithm such as the support vector regression algorithm to predict stable stock price trends. In the next phase, according to the predicted price trend of each stock having a positive return, we start arranging the portfolio based on the type of stock and the percentage of allocated capacity of the total portfolio to that stock. To this end, we use the multi-objective particle swarm optimization algorithm to determine the optimal portfolios as well as the optimal weights corresponding to each stock. The sample, which was selected using the systematic removal method, consists of active firms listed on the Tehran Stock Ex-change from 2018 to 2020. Experimental results, obtained from a portfolio based on the prediction of stock price trends, indicate that our suggested approach outperforms the retrospective approaches in approximating the actual efficient frontier of the problem, in terms of both diversity and convergence. پرونده مقاله

In this paper, a two layer recurrent neural network (RNN) is shown for solving nonsmooth pseudoconvex optimization . First it is proved that the equilibrium point of the proposed neural network (NN) is equivalent to the optimal solution of the orginal optimization problem. Then, it is proved that the state of the proposed neural network is stable in the sense of Lyapunov, and convergent to an exact optimal solution of the original optimization. Finally two examples are given to illustrate the effectiveness of the proposed neural network. پرونده مقاله

The multi-parametric programming (mp-P) is designed to minimize the number of unnecessary calculations to obtain the optimal solution under uncertainty, and since we widely encounter that kind of problem in mathematical models, its importance is increased. Although mp-P under uncertainty in objective function coefficients (OFC) and right-hand sides of constraints (RHS) has been highly considered and numerous methods have been proposed to solve them so far, uncertainty in the coefficient matrix (i.e., left-hand side (LHS) uncertainty) has been less considered. In this work, a new method for solving multi-parametric mixed integer linear programming (mp-MILP) problems under simultaneous uncertainty OFC, RHS, and LHS is presented. The method consists of two stages which in the first step, using tightening McCormick relaxation, the boundaries of the bilinear terms in the original mp-MILP problem are improved, the approximate model of the problem is obtained based on the improved boundaries of the first stage, and finally, an algorithm is presented to solve these kinds of problems. The efficiency of the proposed algorithm is investigated via different examples and the number of required calculations for solving the problem in different partitioning factors is compared. Also, model predictive control (MPC) using mp-P is designed for an example of an urban traffic network to examine the practical application of the proposed algorithm. پرونده مقاله