The fitting method of the joint probability distribution of wind speed and direction was enhanced by incorporating a deep neural network. This approach allows for a more sophisticated analysis of the relationship between wind speed and direction, leading to more accurate predictions. In this paper. An advanced neural network model was introduced to address issues such as under-fitting or over-fitting. Under-fitting occurs when a model is too simple to capture the underlying patterns in the data. At the same time, over-fitting occurs when a model is too complex and starts to learn the noise in the data rather than the actual patterns. The new model aimed to balance higher complexity while minimizing over-fitting, resulting in more reliable and accurate predictions. The testing of measurement data across various districts demonstrated that the improved neural network model successfully trained and fitted wind probability distributions in multiple districts simultaneously. This suggests that the enhanced model possesses attributes of high accuracy and broad applicability, making it a valuable tool for analyzing wind patterns and making informed decisions in various geographical locations.