非线性检验及预测在污水处理厂评价中的应用

为了避免污水处理厂规模盲目扩大造成的投资效率低下的现象发生,科学地预测合理的用水量必不可少。基于用水量的实际历史数据,利用BDS检验、Box-Pierce检验和Box-Ljung检验以及非线性检验,如代替数据检验Surrogate date test、Hinich双谱检验、White人工神经网络检验来选择时间序列重构预测模型。根据实际用水量情况,比较各种不同重构模型预测误差,包括线性AR模型以及随机森林、随机梯度Boosting、支持向量、人工神经网络和自适应样条等。结果表明,有着非线性关系的人工神经网络误差最小,符合检验结果。

Application of Nonlinearity Test and Prediction in Assessment of Sewage Disposal Plants

Scientific prediction of reasonable water consumption is inevitable to avoid blind expansion in sewage disposal plants with low efficiency of investment.Historical data were collected.Independent tests such as BDS,Box-Pierce and Box-Ljung tests and nonlinearity tests including surrogate data,Hinich＇s bispectrum and White＇s artificial neuron network tests were applied jointly.The reconstruction prediction model is selected through these tests.The prediction errors of AR,random forest,stochastic gradient boosting,support vector,artificial neuron network and multivariate adaptive regression splines were calculated based on real consumption.The results show that artificial neuron network with nonlinear relation exhibits the minimal error,which accords with the conclusion of all tests.