how can I predict streamflow discharge?

Hi there,

I have a project of streamflow prediction. I have two series of recorded data include precipitation and discharge. Both data have been recorded every 15 min for about 28 years. I want to find the relationship between these series by using Neural Networks tool in MATLAb. I also want to predict the flow discharge for next 10 years by using the obtained function and future rainfall data. Another target is to train the data with two algorithms and compare their results. I would really appreciate it if you can help me and give me an example about this issue.

Kind regards, Matin

Answer 1

[Greg Heath]

1. Standardize the data to have zero-error/unit-variance.

2. Remove or modify outliers

3. Find the statistically significant lags of the target auto-correlation function and the target/input cross-correlation function

4. Subsample the data without losing the main characteristics of the plots.

5. For each subsample use a selection of significant delays to design multiple narx networks for a range of hidden nodes Hmin:dH:Hmax (outer loop). For each H, design Ntrials nets (inner loop) with different initial weights.

6. Choose one or more of the successful nets with the lowest number of hidden nodes.

7. Useful searches

 help narx
 help narxnet
 doc narxnet

 greg narxnet Ntrials

Hope this helps.

Thank you for formally accepting my answer

Greg