膜计算粒子群算法改进极限学习机的水肥预测模型研究

Research on water and fertilizer prediction model based on extreme learning machine optimized by membrane computing and particle swarm optimization

针对灌溉系统水肥预测计算复杂、预测精度不高、实时性不强等问题,提出膜计算粒子群算法改进极限学习机(MCPSO-ELM)的水肥预测模型。为提高极限学习机的泛化能力和预测精度,引入粒子群算法与膜计算进行优化,利用粒子群算法的高效率搜索能力与膜计算的平行计算优势,大幅度提高收敛速度和搜索精度,不断迭代优化ELM网络的连接权值和阈值以提高预测精度,种群多样性有效解决全局搜索和局部寻优之间的平衡。建立PSO-ELM、WPSO-ELM、IPSO-ELM和MCPSO-ELM四个模型进行对比试验,MCPSO-ELM模型的预测误差小于30 m 3/hm^(2),MAPE为2.0%,预测曲线与水肥实际用量曲线最为接近,预测性能明显优于其他模型。本文提出的MCPSO-ELM能够获得更高的预测精度、更好的预测效率和稳定性,可以为智能灌溉系统提供可靠参考。

In order to solve various problems of water and fertilizer prediction in irrigation system,such as complex calculation,low prediction accuracy and weak real-time capability,a water and fertilizer prediction model based on extreme learning machine optimized by membrane computing and particle swarm optimization was proposed.The efficient search capability of particle swarm optimization coupled with the parallel computation of membrane computation greatly improved the convergence rate and search precision of the algorithm.This also enhanced iterative optimization of the connection weights and threshold of ELM network to improve the prediction accuracy whereby the balance between global search and local optimization was solved by population diversity.Four prediction models(PSO-ELM,WPSO-ELM,IPSO-ELM and MCPSO-ELM)were established for comparative experimental analysis.The absolute error of prediction data for MCPSO-ELM was less than 30 m 3/hm^(2),and was 2.0%for MAPE.It was observed that the prediction curve was the closest to the actual water and fertilizer consumption curve,signifying that prediction performance was obviously better than other models.The MCPSO-ELM can achieve higher prediction accuracy,better prediction efficiency and stability,and can provide reliable reference for intelligent irrigation system.