基于改进神经网络的环境空气质量预测

Ambient Air Quality Prediction Based on Improved Neural Network

为提高环境空气质量预测的精度,提出一种由改进人工蜂群算法和反向传播神经网络相结合的环境空气质量预测方法(KABC-BP)。对人工蜂群算法中雇佣蜂、跟随蜂的搜索空间提出一种随迭代次数递减的搜索公式,以随机初始化此改进人工蜂群算法的不同初始解作为不同组反向传播神经网络权值,以蜂群算法迭代代替人工神经网络的梯度下降修正迭代,以蜂群个体的对应权值下训练误差倒数作为适应度函数,该改进人工蜂群算法所求全局最优解就是所求反向神经网络最优权值。通过基于改进蜂群算法的反向传播神经网络算法、传统蜂群算法的反向传播神经网络算法(ABC-BP)及反向传播神经网络算法(BPNN)的环境空气质量预测的仿真实验表明,该算法的环境空气质量预测精度是最高的。

In order to improve the accuracy of ambient air quality prediction,we propose a method of ambient air quality prediction basedon improved artificial bee colony algorithm and back propagation neural network （KABC-BP）. In the artificial bee colony algorithm,wegive a search formula with decreasing number of iterations for the search space of employed bees and onlookers. The different initial solu-tions of the improved artificial bee colony algorithm are randomly initialized as the weights of different groups of back propagation neuralnetworks. The gradient iteration of artificial neural network is replaced by iterative algorithm of artificial bee colony algorithm. The re-ciprocal of training errors is used as fitness function under the corresponding weight of colony individuals. The global optimal solution ofthe improved artificial bee colony algorithm is the optimal weight of the back propagation neural network. The simulation of ambient airquality prediction on back propagation neural network algorithm based on improved bee colony algorithm,back propagation neural net-work algorithm based on the traditional bee colony algorithm （ABC-BP） and back propagation neural network algorithm （ BPNN）shows that the method proposed is the highest in the prediction of ambient air quality.