This paper is based on some assumptions validated using real data and the most used Forex trading platform Meta Trader 4. First, we assume that any reasonable and relatively simple models, reflecting some trading hypotheses, can be profitable for a certain period. Having a sufficient set of such models optimized for selected currency pairs, we can use the model that provides the greatest profit in the current trading period. The second proposal is to use a fuzzy multiple-criteria approach at the training optimization stage using historical data in order to overcome or reduce the negative effect of overfitting. Here for the first time, the problem of fuzzy multiple-criteria optimization of trading was formulated and solved based on the output parameters of the developed simple single-criteria crisp models. This provides significantly greater profit than that obtained using single-criteria crisp models. The third proposal is to use the hierarchical structure of fuzzy local criteria to solve the multiple-criteria problem. It is shown that this additionally provides a significant increase in profit. The profitability and riskless of the developed trading models are studied based on real quotations of currency pairs USDJPY, EURUSD and AUDUSD using H4 timeframe.