基于改进杂草算法优化的神经网络模型在径流预报中的应用

针对BP神经网络在径流预报中易陷入局部最优解的缺陷及智能优化算法的优势,引入改进的杂草算法优化神经网络权值和阈值,将传统的杂草算法个体以正态分布空间扩散的方式改进为混合种群多种分布的方式产生子代个体。以金沙江流域中长期径流预报为例,将改进杂草算法优化的神经网络模型的径流预报结果与传统的BP神经网络和基于遗传算法优化的神经网络模型的预报结果进行对比。结果表明,改进杂草算法优化的神经网络应用到金沙江流域的径流预报精度较高,模型收敛更快,结果更加稳定,在实际预测中合理可行,具有一定的应用优势。研究成果为径流预报提供了新思路。

Application of Neural Network Combined with Improved Invasive Weed Optimization Algorithm in Runoff Forecasting

In view of the defects that the convergence of BP neural network is easy to fall into the local extreme value and the advantages of intelligent optimization algorithm, the improved invasive weed optimization algorithm was introduced to optimize the weights and thresholds of neural networks. The offspring individual is produced by the diffusion of normal distribution space to substitute the normal distribution in the traditional invasive weed optimization algorithm. The mid-and long-term runoff forecasting of Jinshajiang river basin was taken as an example. The prediction results of neural network model optimized by improved weed algorithm were compared with the results of traditional BP neural network and neural network model optimized by genetic algorithm. The results of neural network model optimized by improved weed algorithm in Jinshajiang river basin has better accuracy, convergence rate and stability. The model is reasonable and feasible in actual prediction, and has certain application advantages. The research provides new ideas for runoff forecasting.