基于随机遗传算法的LM-BP模型的氨氮预测

针对伦河孝感段近几年实际的氨氮含量，提出并建立随机遗传算法的LM-BP模型，并对该地区的氨氮含量数据进行取样、拟合、测试并预测。由于LM-BP网络对其初始权阈值敏感，泛化能力不强，故采用遗传算法（GA）对其初始权阈值进行了优化。为扩展初始种群的覆盖范围得到更优的测试结果，经过多次随机产生初始种群的多次优化，进一步提高了LM-BP网络的泛化能力。实测结果表明：该模型基本能100%拟合，测试误差不超过2%，能够采用该模型对该地区的氨氮含量进行预测，为水质预警预报和水环境规划治理提供科学依据。

Prediction of NH+4-N with random-genetic algorithm based LM-BP network model

Aiming at the actual NH+4-N（ammonia nitrogen） concentration in Xiaogan reach of Lunhe River in the recent years, a random-genetic algorithm based LM-BP Model is put forward and established, and then the data of NH+4-N concentration therein are sampled, fitted, tested and predicted. As the LM-BP neural network is sensitive to its initial threshold with lower generalization capacity, the genetic algorithm is adopted for optimizing the initial threshold. In order to expand the coverage of the initial population for getting better testing result, the generalization capacity of LM-BP network is further improved through optimizing the initial populations randomly and iteratively generated for many times.The actual testing result shows that the data fitting rate of 100% can be got by this model with the testing error not over 2%, and then it can be adopted for predicting the NH+4-N concentration within the region and providing the scientific basis for the early warning and forecasting of water quality and the water environment planning and improvement therein.