BP神经网络在水厂预测混凝投药量中的应用研究

Application of BP neural network for prediction coagulation dosage of waterworks

建立了一种基于三层结构BP神经网络的混凝投药量前馈控制模型,采用烧杯试验数据进行了仿真验证,同时建立了传统的线性回归混凝投药量前馈控制模型,并采用两种模型基于同一样本数据进行仿真。从投药预测值-实际值的对比图和均方根误差等可以看出,BP模型优于回归模型,它通过学习可以根据原水水质进行投药量的有效预测,有一定的自适应性,实用性较强,但也存在一定的局限性,对某些水质的投药预测值还存在一定误差。

A coagulant dosage feedforward control model based on three-tier structure of BP neural network was established, and a simulation test was conducted by jar-test data. Traditional linear regression model was also simulated for comparison, which was simulated based on the same sample data. It was found from the comparison chart of the predicted dosage and the actual value and root-mean-square error that the BP model were superior to the linear regression model. Through the investigation, the prediction of effective dosage could be completed according to the raw water quality, and it had certain self-adaptive, strong practicality, but it also had some limitations, and the prediction errors of dosage still existed under some water quality conditions.