In the current world, data analysis is carried out with a large volume. Meanwhile, in order to more accurately analysis and achieve newer methods, many studies have been done with innovative approaches in various fields, including artificial intelligence. One of the methods available in this area is artificial neural networks for data processing and analysis iired by the nervous system of the human brain, and there are used in some cases, such as image processing, topic modeling and pattern classification problems. There are many ways to develop and upgrade artificial neural networks, which can be referred to using fuzzy logic that is used in this research. In this research, type 2 fuzzy Boltzmann machinne and type 1 and 2 Fuzzy Deep belief Network have been experimented on the English and Persian handwritten datasets. These methods, by adding insufficiency to energy-based artificial neural networks, improve their performance in more complex problems, as well as modeling them with a greater number of variables, and applying these methods have increased classification accuracy, so that the type 2 fuzzy deep belief network has more than 96% accuracy for classifying the Persian handwritten datasets, which is suitable according to the research done in this area Key Words : Energy-base Artificial Neural Networks, Restricted Boltzmann Machine, Deep Belief Network, Fuzzy Numbers, Contrastive Divergence.