[关键词]
[摘要]
调蓄水位与其影响因素之间存在着复杂的非线性关系,针对BP神经网络模型的局限性,选取泵站开启时间差、起调水位、入流量、出流量作为主要影响因素,建立一种基于相关向量机(relevance vector machine,RVM)的调水工程调蓄水位预测模型。通过实例应用表明在相同样本情况下与BP神经网络模型预测结果相比,RVM预测模型均方根误差和平均绝对误差均小于BP神经网络预测模型的预测结果,说明在调水工程调蓄水位的预测中,RVM预测模型具有精度高、离散性小等优点,为调水工程调蓄水位的预测提供了一条新途径。
[Key word]
[Abstract]
China has serious water shortage problems,and the regional distribution of water resources is very unbalanced.The water diversion project may solve the water shortage problem and uneven spatial distribution,but also affect the water situation in the receiving area, making the local water source and the pump station reservoir interrelated and interdependent.It is of great significance for improving the efficiency and reducing the cost of water transfer to master the changing process and influencing factors of water storage level in water transfer projects.As traditional methods are difficult to analyze the action mechanism of multiple influencing factors,several scholars begin to apply the intelligent method based on the "black box" processing of mapping relationship to analyze the reducing water level under the influence of multiple factors and start to use intelligent algorithms such as artificial intelligence algorithm,BP neural network and correlation vector machine to seek the relationship between the regulating and storing water level and various influencing factors. Relevance vector machine (RVM) is a sparse probability model based on Sparse Bayes,which has many advantages such as its kernel functions without the restriction of Mercer′s conditions,and automatically determined relevance vectors.Many factors are influencing the change of the water level before the pump,and there is a complex nonlinear relationship with the water level before the pump.Given the limitations of the BP neural network model,through the analysis and comparison of a large number of experimental data,the selection of pump stations open time,adjust the water level,flow,the analysisof the existing data collection,sorting out of the water diversion scheme fitting data set and build learning samples and fitting training, to regulate water level and forecasting model.The RVM model predicted results are compared with BP neural network prediction model. By selecting reasonable kernel function and kernel width parameters,the prediction results of RVM model are obtained,the application of example shows that the root mean squared error and mean absolute error for RVM prediction model are smaller than BP neural network prediction model using the same sample,prediction accuracy is higher.However,due to the insufficient number of learning samples,the RVM prediction model is not as good as the BP neural network prediction model in reflecting the trend and amplitude in the water level prediction,but the BP neural network prediction model has greater discretization in the result prediction because the main reason is that the model has low accuracy in predicting the samples. Aimed at the complex nonlinear relationship between the water level and its influencing factors,a prediction model of water level for water transfer project based on RVM is established.The model has high precision and low dispersion,which provides a new way for the prediction of water levels for water transfer projects.The prediction results show that the BP neural network prediction can reflect the trend and amplitude,but the RVM prediction model has the advantages of higher accuracy,shorter time consumption,and better generalization ability.In practical application,the advantages of the two different models can be played,the BP neural network model is used to predict the overall trend of the water level,and the RVM model is used to predict the corresponding sample value,to provide more reliable reference for engineers and technicians.
[中图分类号]
[基金项目]