The combined sewer contributes to improvement in environment of polluting point sources and non-point sources in urban areas, but the conventional diversion facility is low in accuracy of sewage flow rate control and has difficulty in reliably managing wastewater and rainwater, and therefore it is a big problem to improve the function of the sewage flow rate control. This study aims to put the diversion facility improvement theory into practical use on an extension of the knowledge obtained through the conventional hydraulic engineering theory and actual results shown in related reference documents and so on, based on the past study results⁵⁾ which enabled sewage flow rate control with high accuracy. The study pointed the way to a basic technology of realizing this improvement theory by a hydraulic design under the diversion facility plan condition of a project case and achieving practical use by a structural design. Reliable sewage flow rate management is a key technology capable of effectively coping with the attainment of the purpose through utilization of stock in a confluence improvement project and new sewer vision measures, and it is expected in the future to verify this new technology in test construction.

View full abstract

 This paper describes an application of neural networks for forecasting the flow rate upper district of dams for hydropower plants. The forecasting of recession characteristics of the river flow after rainfalls is important with respect to system operation and dam management. We present a method for improving the precision of forecasting flow rate upper district of dams by utilizing steady-state estimation and recession time constant of the river flow. A case study was carried out on the upper district of the Yahagi River in Central Japan. It is found from our investigations that the forecasting accuracy is improved to 18.6% from 25.8% with a forecasted error of the total amount of river flow by using steady-state estimation.

View full abstract

 To improve the accuracy and reliability of the real-time flood prediction, we developed the real-time river stage prediction model, using the hybrid deep neural network and physically based distributed rainfall-runoff model.
 The main component of the hybrid model was 4-layer feed-forward artificial neural network. As the training method of the network, we applied the deep learning technique to improve the ability of network expression. To optimize the network weight, the stochastic gradient descent method based on the back propagation method was used. As a pre-training method, the denoising autoencoder was used. By using the predicted flow of the rainfall-runoff model as the input data of the neural network, we integrated two models into the hybrid model. The input data of the hybrid model were upstream water level, hourly change in water level, and estimated hourly change of catchment storage. The output is the change in water level at the prediction point. The prediction procedure of the hybrid model is as follows; first, calculate the downstream flow by the distributed model, and then estimate the hourly change of catchment's storage form the observed rainfall and calculated flow. The estimated change of catchment's storage is used as the input of the ANN model, and finally the ANN model can be calculated. In the training phase, hourly change of catchment storage can be calculated by the observed rainfall and flow data.
 The developed model was applied to the one catchment of the OOYODO River, one of the first-grade rivers in Japan. The result of the hybrid model outperformed the ANN model and distributed model. Especially in the biggest flood event, performance of the hybrid model was significant improvement.

View full abstract