Deep stock clustering method based on auto-encoder and affinity propagation algorithm

The invention discloses a deep stock clustering method based on an auto-encoder and an affinity propagation algorithm, and is applied to the technical field of stock prediction. Comprising the steps of obtaining historical closing price time sequence data of a stock; performing data preprocessing on the obtained stock historical closing price time series data to obtain preprocessed stock historical closing price time series data; performing feature extraction and dimensionality reduction on the preprocessed historical closing price time series data of the stock by using a pre-trained auto-encoder to obtain low-dimensional and effective representation of the historical closing price time series data of the stock; the stock clustering layer is used for clustering the low-dimensional and effective representation of the historical closing price time series data of the stock to obtain a stock clustering result; and constructing a same-trend stock association relation graph based on a stock clustering result, and o