Considering the impact of large transactions in the stock market on the market, this paper aims to reduce the transaction cost of large orders, reduce the market impact and reduce the risk. Therefore, the order splitting algorithm is designed and the optimal order splitting decision is made. This paper first cleans up the data of 10 stocks and leaves the required characteristics, then draws the time series analysis diagram of the trading volume and price of each stock, as well as the cross‐sectional data diagram of different stock trading volume and price at the same time, and finally analyzes the trend change of trading volume and price according to different financial indicators. This paper need to build a time series model to predict the trading volume and price of all stocks in the next 30 ticks. This paper uses LSTM neural network time series and ARIMA exponential smoothing method to predict the future trading volume, compares the prediction errors of the two models, selects the model with small error for the trading volume and price of each stock, and finally obtains the prediction results. The daily trading volume is predicted through xgboost and GBR, and the MSE (mean square error) is used as the evaluation index to select the best model for prediction. The objective function and nonlinear constraints are constructed, and then the weighted average price of twap is compared with the real price. Finally, the Monte Carlo model is used to solve the problem, and then the optimal order splitting strategy is obtained. The strategies and methods adopted in this paper have certain accuracy and practical significance, and can be used as an effective model for market order splitting.