Aiming at the problem that traditional methods can not effectively predict multi-noise and nonlinear time series, focusing on multi-scale features fusion, a prediction method based on tree structure Long Short-Term Memory (LSTM) neural network was proposed and verified. First of all, the core methods of realizing the prediction were proposed, and the internal advantages of the methods were analyzed. Secondly, the prediction model based on tree structure LSTM neural network was constructed. Finally, the model was verified based on the international gold spot transaction data of the last ten years. The results show that the prediction accuracy is nearly 10 percentage points higher than the minimum success rate, and the availability of the methods is proved.

For the problems of the traditional method of data record-replay, such as packet format association, close corporation with the controlled application, and low transmission efficiency, a new record-replay method based on data interception and cheating injection was proposed. Firstly, the network data packet was automatically intercepted through the service provider interface technique of Winsock 2. Secondly, the problem of the data sharing and high speed data access was solved by using the memory-mapped file technique. Finally, the saved data packet was intercepted into the user program by the data read operation motivated by the fake messages. The practical application shows that the new method is suitable for the distributed simulation and simulated training system for its merits such as the avoidance of network packet transmission, no necessity for corporation with the controlled application, irrelevance of the packet format, smooth recurrence with ten times the speed.