Abstarct: Modeling and forecasting the prices of various types of stocks in financial markets with very high sales and a very large impact on the economic growth of countries and organizations, are the challenging goals of researchers, investment companies and reputable banks. With the growth of artificial intelligence and especially deep learning, the performance of the feature learning task was performed through networks designed to do so. In this dissertation, a model of neural networks using CNN and LSTM architectures and inspired by multi-filter neural network (MFNN) is presented, which differs from other models. Different methods are predicted for Feature extraction and production, a new method, specifically for feature extraction in nominal financial series samples and forecasting market price trends. In this method, from different types of indicators, the strategies produced have been produced, which has been used in the production of features. In the following, by introducing a new method in targeting financial data and comparing it with other methods, we will complete the pre-processing steps on the financial time series data. Also, in the continuation of the proposed model, a combination of both CNN and LSTM neural network architectures is used, in order to obtain information about different spaces and the market. We apply our method to predict the most important and largest financial market in the world, the foreign exchange market. The results showed that the proposed model performed better than traditional machine learning models, statistical models and single construction (convolution, RNN and LSTM). (Networks are visible in terms of accuracy, usefulness and stability).