Open AccessResearch on Prediction of the Cash Usage in Banks Based on LSTM of Improved Grey Wolf Optimizer
Author(s) -
Jingfeng Rong,
Di Wang
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1769/1/012031
Subject(s) - artificial neural network , computer science , artificial intelligence , mean squared error , data mining , series (stratigraphy) , machine learning , layer (electronics) , genetic algorithm , mean squared prediction error , cash , algorithm , mathematics , statistics , finance , paleontology , economics , biology , chemistry , organic chemistry
In the real production and operation, it is impossible to predict the amount of cash in daily use. Therefore, the prediction model of improved LSTM neural network is proposed to cope with the problem for preparing excessive cash. Hence, the improved Grey Wolf Optimizer is most effective in searching for the optimal solution by optimizing the impact factors of Grey Wolf Optimizer. Combining the improved Grey Wolf Optimizer with LSTM neural network, the neural networking learning rate parameters are set reasonably by optimizing the algorithm to reduce the impact of inappropriate parameters on the prediction results of either over-fitting or under-fitting. What’s more, the neural network topological structure, weighing the number of LSTM network layers and the number of the neural units in each layer, determines the neural network’s description of data. If the network topology is too simple, the prediction results may not be enough to describe the real data. However, if it is too complex, it will not only waste the computing resources, but also make the prediction results over-fitting with only good description of training data. Therefore, it avoids the problem of large errors in predicting results caused by the parameters of neural network and realizes the prediction the daily cash usage. Finally, the test is completed on the data of a sub-branch network of bank with mean square error (MSE) 0.016. Compared with the traditional time series model ARAM and the unimproved LSTM, the improved LSTM predicts cash usage more accurately and efficiently.