Research On Risk Evaluation Model Of Overseas Mining Investment Based On Deep Learning

In recent years,with the continuous investment in overseas mining markets,correctly identifying the risk level of investment target countries can reduce the investment risk.However,in the actual risk assessment,there are some problems,such as the subjective weights of experts constitute the evaluation model,and the lack of characteristic data for establishing the risk level prediction model.Therefore,this paper takes more scientific risk assessment as the starting point,introduces the idea of deep learning,and improves the accuracy of risk level identification.The main work includes the following aspects:First of all,in order to solve the problem of small sample of risk assessment characteristic data.In this paper,on the basis of Wasserstein gan gradient penalty(WGAN-GP),an improved Wasserstein gan gradient penalty based on difference(WGAN-GPD)is proposed to solve the shortage of risk assessment feature data and improve the convergence speed of the discriminator.By comparing the training process of WGAN-GPD and WGAN-GP,and comparing the synthetic minority oversampling technique(SMOTE),the results show that the loss function convergence speed and stability of WGAN-GPD discriminator are better than that of WGAN-GP,and from the point of view of data distribution and statistical information,WGAN-GPD can better generate samples that accord with the original data distribution.Secondly,in order to improve the accuracy of risk level identification,this paper proposes two risk level prediction models.Firstly,based on the original risk characteristic data,the FA-SSA-SVM risk level prediction model is constructed by optimizing the selection of penalty parameter C and kernel function parameter g in Support Vector Machine(SVM)by Firefly Algorithm(FA)and Sparrow Search Algorithm(SSA).Compared with other swarm intelligence optimization algorithms,the improved SVM has fewer iterations and higher accuracy.Secondly,based on the risk characteristic data enhanced by WGAN-GPD,by combining Denoising Autoencoders(DAE)and Deep Neural Networks(DNN),and introducing the random deactivation technology Dropout to prevent over-fitting and adopting Adam optimizer,the risk grade prediction model of DAE-DNN-DA is constructed.Experiments show that the dimension reduction effect of DAE is better than the traditional dimension reduction algorithm in dimension and accuracy.Experiment compares FA-SSA-SVM and DAE-DNN-DA from accuracy,training time and application range.It shows that FA-SSA-SVM is suitable for solving the nonlinear risk level prediction under limited samples,and DAE-DNN-DA is more suitable for the risk level prediction of large data scale.Finally,this paper develops a prototype system of risk assessment,including index classification module,evaluation index module,evaluation scheme module,evaluation task module and risk assessment module,which provides support for risk assessment.