针对运用BP神经网络模型来编制水利定额存在收敛慢、精度低、稳定性差的缺陷,提出利用粒子群算法(PSO)来优化BP神经网络初始权值阈值的模型,优化模型结合了粒子群全局搜索能力和BP网络的局部探优能力.在运用MATLAB对算法模型进行编程中,首先确定模型的关键参数和开展数据的预处理,其次利用标准粒子群算法优化BP神经网络的初始连接权值阈值,最后将优化的连接值带入BP模型训练并预测,实验中连续运行了50次模型.结果表明:BP模型的双输出预测精度分别为11.13%和8.41%,有10次未达到目标精度;PSO-BP模型的双输出预测精度分别为5.65%和5.44%,全部达到目标精度.因而得出结论, PSO-BP模型比单纯BP神经网络的预测精度和稳定性更好,更适合用来指导水利定额的编制工作.%Slow convergence,low accuracy and poor stability are the shortcomings when using BP neural network model to pre-pare water conservancy quota. It is proposed that the initial weight threshold of BP neural network should be optimized by particle swarm optimization,which combines the global search ability of particle swarm and the local exploration ability of BP network. During the model programming using MATLAB algorithm,key parameters are determined first as well as the pre-processing of raw data;then the initial connection value of BP neural network is optimized using standardized particle swarm optimization algo-rithm;lastly,the optimized value is introduced into the model for training and predicting. Each model was run fifty times contin-uously. The results show that the dual output prediction accuracy of the BP model are 11.13% and 8.41%,10 times of which fail to reach the expected accuracy;the dual output prediction accuracy of PSO-BP model are 5.65% and 5.44% with the ex-pected accuracy achieved meeting the objectives. It is concluded that PSO-BP model performs better than BP neural network in prediction accuracy and stability,and thus more suitable for guiding the preparation of water quotas.
展开▼
