基于佳点萤火虫算法与BP神经网络并行集成学习的旱情预测模型  

Drought prediction model based on GPSGSO-BPNN parallel ensemble learning algorithm

作　　者：李敬明[1,2] 倪志伟 朱旭辉[1] 许莹[3] LI Jingming1,2,NI Zhiwei1,ZHU Xuhui1,XU Ying 3(1. School of Management, Hefei University of Technology, Hefei 230001, China; 2. School of Management Science and Engineering, Anhui University of Finance and Economics, Bengbu 233030, China; 3. Key Laboratory of Atmospheric Science and Satellite Remote Sensing of Anhui Province, Meteorological Science Institute of Anhui Province, Hefei 230031, Chin)

机构地区：[1]合肥工业大学管理学院,合肥230009 [2]安徽财经大学管理科学与工程学院,蚌埠233030 [3]安徽省气象科学研究所安徽省大气象科学与卫星遥感重点实验室,合肥230031

出　　处：《系统工程理论与实践》2018年第5期1343-1353,共11页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金重大研究计划培育项目(91546108);国家自然科学基金重大项目(71490725);国家自然科学青年基金项目(71601061);安徽省教育厅重点自然科学项目(KJ2016A308)~~

摘　　要：针对传统BP神经网络在旱情预测的实际应用中随机初始权值和阈值导致网络学习速度慢、易陷入局部解以及计算精度低等缺陷,提出一种基于数论佳点集萤火虫（good point set glowworm swarm optimization,GPSGSO）算法与BP神经网络（back propagation neural network,BPNN）并行集成学习算法（GPSGSO-BPNN）的旱情预测模型.首先,借鉴数论中佳点集理论构造初始均匀分布的萤火虫种群,并引入惯性权重函数动态修正移动步长,生成基于数论佳点集理论萤火虫算法,并从理论上分析算法的有效性;其次,将GPSGSO算法与BPNN相结合构建并行集成学习算法,实现两种算法的并行交互集成.最后,将并行集成学习算法应用于农业干旱灾害预测中,构建基于GPSGSO-BPNN并行集成学习算法的旱情预测模型.通过8个Benchmark函数验证了GPSGSO算法在收敛速度、计算精度及稳定性等方面的有效性.同时,以皖北农业干旱气象数据作为仿真数据,实验结果表明GPSGSO-BPNN算法在计算速度、精度及稳定性方面较传统BPNN、GSO-BPNN及GA-BPNN等算法有较明显的优势,提高了旱情等级预测的准确性.Aiming at defects of slow learning speed, trapped in local solution and inaccurate operating result of traditional BP neural network in the practical application of drought prediction, a drought prediction model based on parallel ensemble learning algorithm of good point set glowworm swarm optimization algorithm（GPSGSO） and back propagation neural network（BPNN） is proposed. Firstly, a new kind of improved glowworm swarm algorithm based on good point set theory and inertia weight function is constructed, and the validity of the algorithm is analyzed theoretically. Secondly, GPSGSO algorithm and BPNN are combined to construct parallel ensemble learning algorithm. GPSGSO is used to optimize the weight and threshold of BPNN, and the ensemble strategy is carried out for the best weights and thresholds. Finally, the parallel ensemble learning algorithm is applied to the prediction of agricultural drought disaster, which can accurately determine the drought level. The effectiveness of the GPSGSO algorithm in terms of convergence speed, accuracy and stability is verified by 8 Benchmark functions. At the same time, agricultural meteorological data of Northern Anhui Province is used to simulate validate experiment, the experimental results show that the algorithm has obvious advantages over the traditional BPNN, GSO-BPNN and GA-BPNN algorithm in terms of convergence speed, operation accuracy and stability. Therefore, the drought prediction model based on GPSGSO-BPNN parallel learning algorithm can effectively improve the accuracy of agricultural drought prediction.

关 键 词：佳点集萤火虫算法 BP神经网络 并行集成学习 旱情预测模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]