Research On The Application Of Cooperative Co-evolution Algorithm In Characteristic K-line Prediction

Financial markets have the key role in the national economy.Both the government and investors are interested in the study of the price behaviors of financial markets.Financial time series modeled by candlesticks can effectively reduce the noise in financial data.According to statistical analysis,after the emergence of some specific single or multiple candlestick combinations,a trend would be shown in the following period of time.These specific combinations of candlesticks can be called as characteristic candlesticks.Characteristic candlesticks is a common kind of technical analysis method,but it is difficult to find such a complex combination relationship only relying on human subjective experience.Using artificial intelligence algorithm,it is more likely to find the intricate relationship among them.The main research contents of this paper include two parts.Firstly,the financial time series data are modeled by characteristic candlesticks,and then the data are analyzed by using traditional SVM in the framework of cooperative coevolution algorithm.The characteristic candlesticks of financial time series data can effectively reduce the noise in financial time series data.Traditional SVM is a kind of fast classifier such that it is suitable for the application in the high-frequency financial transactions.Cooperative evolutionary algorithm has the ability to optimize large-scale models,and can improve the classification accuracy of models.Second,this paper proposes reconstructing training set support vector machines(RTS-SVM)and roulette cooperation coevolution algorithm(R-CC).RTS-SVM is a new support vector machine proposed in this paper for the characteristics of high noise and uneven distribution of financial time series data.After the “soft margin” support vector machine modeling,the original training sample data are divided into three different data sets according to the classification boundary.According to the influence of each type of data set on the accuracy,the data are selected or rejected,and a new training set is reconstructed to improve the classification prediction results.The cooperative evolution algorithm is improved by using roulette strategy.The traditional cooperative coevolution algorithm is to optimize each sub-component step by step.But some sub-components contributing little to the accuracy of the model would also get the same optimization time,which increases the calculation time.The roulette cooperative coevolution algorithm allocates the optimization probability according to the contribution proportion of each sub-component to the model accuracy.The sub-components with more contribution will get more change to be optimized,which can reduce the time complexity and improve the optimization effect of the model.