Research On Capital Flow Forecasting Technology Of Internet Financial Platform Based On Deep Learning

With the advancement of technology and the increasing demand for financial institutions,a large number of Internet Finance(ITFIN)companies have sprung up.How to effectively control liquidity risk in the context of huge business data flow is an urgent problem to be solved for the ITFIN companies.As a kind of financial time series,the cash flow on financial platforms has significant high noise and non-stationary characteristics,which somehow brings great difficulties to its prediction.This paper proposes an Attention-based Multitask Gated Recurrent Unit(AM GRU)model in order to solve the problems that existing works cannot adaptively extract effective information in the input sequences and fail to deal with the noise signals well.The experimental results show that this method can effectively improve the noise processing ability of the model,thereby improving the accuracy of the prediction results.At the same time,this paper designs a feature selection method based on the impact factors of cash flow on financial platforms.The feature subset is selected and constructed from three perspectives.The experimental results show that the constructed feature subset can better reflect the trend of the cash flow.Based on the above works,this paper proposes a model fusion method based on time-varying weights.By calculating the time-varying weights,the prediction results of the AM GRU model and the feature subset-based model are merged.It complements the features that deep networks need to learn by exploring complex inter-parameter relationships,reduces the complexity of the model.Further,it is able to solve the overfitting problem caused by the complex models to some extent.