基于遗传-卷积神经网络算法的废水处理预测模型研究

A Prediction Model Based on Genetic Algorithm(GA)and Convolutional Neural Network(CNN)in Wastewater Treatment System

在废水生物处理过程建立出水COD与出水SS的预测模型中,针对卷积神经网络在设计时没有规律遵循并很难保证网络最优化的问题,提出了一种基于遗传算法降维的卷积神经网络优化方法｡本文将遗传算法(GA)与卷积神经网络(CNN)耦合起来形成一种新颖的混合算法--GA-CNN算法,并将该算法与CNN算法和BP神经网络的预测效果进行对比｡仿真结果表明,对于出水COD的浓度预测,GA-CNN的预测性能相比于CNN提升了13.66%,相比于BP提升了19.40%,其中GA-CNN算法的最优预测效果如下:均方根误差(RMSE)为3.5303,平均绝对百分比误差(MAPE)为3.92%,决定系数(R^(2))为0.7195。对于出水SS的浓度预测,GA-CNN的预测性能相比于CNN提升了9.26%,相比于BP提升了13.43%,其中GA-CNN算法的最优预测效果如下:均方根误差(RMSE)为0.5883,平均绝对百分比误差为1.99%,决定系数(R^(2))为0.6770。

In this model that predicts effluent COD and SS in the processes of biological wastewater treatment,in order to solve the problem that a convolutional neural network does not follow rules regularly and it is difficult to guarantee the network to optimize,a convolutional neural network optimization method based on genetic algorithm is proposed.In this paper,the genetic algorithm(GA)and convolutional neural network(CNN)are combined to form a new hybrid algorithm-GA-CNN algorithm,and this new algorithm is compared with the CNN algorithm and BP neural networks on the aspect of prediction performance.The simulation results are as follows.For the concentration prediction of CODeff,the prediction performance of GA-CNN is 13.66%higher than that of CNN,which is 19.40%higher than BP,the prediction result of GA-CNN algorithm shows that the minimum root mean square error(RMSE)is 3.5305,the maximum correlation coefficient(R^(2))is 0.7195,and the minimum mean absolute percentage errors(MAPE)is 3.92%;for the concentration prediction of SSeff,the prediction performance of GA-CNN is 9.26%higher than that of CNN,which is 13.43%higher than BP,the prediction result of GA-CNN algorithm shows that the minimum root mean square error(RMSE)is 0.5883,the maximum correlation coefficient(R^(2))is 0.6770,and the minimum mean absolute percentage errors(MAPE)is 1.99%。