遗传规划自适应建模的 JAVA 实现及在股票价格预测中的应用

摘要

Aiming at the way to combine the genetic programming method with time series effectively,we built the genetic programming-based adaptive time series model,and implemented it by Java language assisted algorithm.Compared with previous genetic programming algorithm,it uses the average value method to improve the generation manner of initial population,makes the mutation probability decrease along with the increase of evolution algebra.Moreover,the idea of parallel computing is introduced,in the core part of the algorithm,the calculation of fitness value,the operations of individual copy,the crossover and the mutation are carried out from the thread granularity.Based on the actual operation results,the CPU operates clearly in multi-core mode,the algorithm can take full advantages of multi-processor and multi-core in computation and this improves the efficiency of operation.Applying the genetic programming model to the prediction of stocks prices in China’s securities market,we compared the prediction results with that of the artificial neural network optimised by genetic algorithm and the traditional genetic programming,results showed that the prediction precision of the improved genetic programming was higher,and could more intuitively express the relationship between input and output.%针对如何将遗传规划方法与时间序列有效结合，构建基于遗传规划的时间序列自适应模型，并通过 Java 语言辅助算法的实现。相比之前遗传规划算法，采用平均值法改进初始群体的生成方式，使变异概率随着进化代数的增加而递减，且加入并行计算思想。在算法流程的核心计算环节，将适应度值的计算、个体的复制、交叉、变异操作都从线程的粒度来进行，基于实际运行效果来看，CPU 多核运行明显，算法能够充分利用多处理器、多核进行计算，提高了运行效率。将改进的遗传规划模型应用于我国股票市场上股票价格的预测，将预测结果与经遗传算法优化的神经网络方法和传统遗传规划方法进行比较，结果证明改进遗传规划方法的预测精度更高，且能够更直观地表达输入与输出之间的关系。