多策略融合学习萤火虫算法在年径流预测中的应用

Application of multi-strategy fusion learning firefly algorithm in annual runoff forecast

为提高萤火虫算法(FA)的收敛速度和求解精度,提出多策略融合学习萤火虫算法(MSFLFA)。该算法为最优萤火虫引入深度学习策略,开展全局搜索,为非最优萤火虫引入随机吸引模型,开展局部开采,兼顾探索和开采平衡,提升收敛速度和寻优精度双目标。为验证MSFLFA性能,基于经典函数测试集进行仿真测试,实验表明,MSFLFA性能优于FA,VSSFA,WSSFA,MFA,CFA,RaFA和ApFA。为扩展MSFLFA应用领域,融合支持向量回归(SVR),建立MSFLFA-SVR径流预测模型。仿真实验表明,与FA-SVR,WSSFA-SVR,MFA-SVR,RaFA-SVR,BP-ANN-SVR和PPR-SVR模型相比,MSFLFA-SVR预测模型具有更高的预测精度。

In order to improve the convergence speed and accuracy of the Firefly Algorithm(FA),a multi-strategy fusion learning firefly algorithm(MSFLFA)is proposed.The algorithm introduces deep learning strategies for optimal fireflies, conducts global search, introduces random attraction models for non-optimal fireflies, conducts local mining, takes into account exploration and mining balance, and improves the dual goals of convergence speed and optimization accuracy.In order to verify the performance of MSFLFA,a simulation test is performed based on the classic function test set.The experiment shows that the performance of MSFLFA is better than FA,VSSFA,WSSFA,MFA,CFA,RaFA and ApFA.In order to expand the application field of MSFLFA,support vector regression(SVR)was integrated, and the MSFLFA-SVR runoff prediction model was established.After the prediction model test, experiments show that the MSFLFA-SVR prediction model has higher prediction accuracy compared with FA-SVR,WSSFA-SVR,MFA-SVR,RaFA-SVR,BP-ANN-SVR and PPR-SVR models.