A Study Of Fund Allocation In Investment Models Based On Multi-objective Evolutionary Algorithm And Deep Learning

Quantitative investment uses computer algorithms to analysis financial data automatically and intelligently and invest the money in each market.So fund managers have its own investment robots which are created and developed by themselves.In order to cope with changeable market,they often have multiple non-homogeneous investment robots.As a result,in order to gain a predetermined goal how to allocate the limited funds to different investment robots become a thorny problem.However,the traditional portfolio areas research is how to combine different investments such as stocks,futures,bonds to achieve a stable target.It didn't solve the above issues.Therefore,the research in this article is now facing a new problem in the field of financial technology.In practice,the research on traditional portfolio model still exist the following problems:1.The limitations of mean-variance model.Coupled with a great change of the global stock markets have been a deterministic method can't keep up with the change in time and can't make the adjustment of the adaptive evolution.2.The constrained problems of multi-objective optimization algorithm.Multi-objective optimization algorithm is applied to the actual scene which is tend to have a range of restrictions.The traditional multi-objective genetic algorithm is unconstrained which leads to huge computation redundancy.3.The selection of optimization targets and multi-objective balance problems.The assumption of using variance or standard deviation to measure the risk is that investors make the negative and positive earnings the same weight.It doesn't make sense in fact.4.The multiphase dynamic asset allocation problem.In the actual investment activities fund manager will adjust the portfolio and optimize the allocation of investment dynamically.The proportion of investors need to constantly adjust their capital ratio to achieve its investment objectives.Although the single phase of the case study of fuzzy portfolio selection problem has made considerable achievements.The questions about multiphase dynamic fund allocating has not been solved systematically.5.High-dimension calculating problem.In practical applications,the fund managers typically have hundreds of thousands of investment robots,and to allocate funds in the multiple investment model requires a lot of computing resources.To solve these problems,this paper put forward a model based on multi-objective genetic algorithm and deep learning oriented to solve multiple funds allocation problems of the investment robots.First,this paper puts forward the following solutions:1.CMOEA/D-FA algorithm solve the problem of single phase of capital allocation.The algorithm is on the basis of MOEA/D algorithm,combined with the actual problem of capital allocation algorithm,adding initialization activation function,restrictive constraints,difference evolution,penalty function,and other optimal algorithms.By optimizing the combination of the objective of this part,CVar and information entropy are added to the model.The combinatorial optimization algorithm of target selection provides a more practical solution,and test them respectively.2.The second one is the multiphase dynamic evolutionary algorithms.We put forward fixed cycle and adaptive cycle creatively,which are the classification of research period optimization phase.This part does the research on the multiphase dynamic environment based on the above research.And it also includes the corresponding optimization and the analysis of the performance of multiple algorithm.This paper then puts forward the final pareto set selection algorithm based on fuzzy theory which can obtain efficient solution in different environment.Some experience shows later.3.The solution which is using the DAE to reduce the dimension of investment portfolio of robot sequences is proposed.