is it possible to compare the result of modelling relationships (e.g rainfall-runoff) using a machine learning model like the Artificial Neural Network (ANN) and that of a deterministic, distributed-parameter, physical process based modeling system like the Precipitation-Runoff Modeling System (PRMS)
Of course you can compare anything you want! Plotting results/predictions from the two models along with actual data is usually a good place to start. You can even produce a hybrid model that interpolates between the to models if there is some reason that appears to work better.
Both, machine learning and deterministic models, allow to describe your process so you could, of course, to compare them and stablish whic of them give a better description comparing with experimental results.
Best wishes,
This is a very good response. am so happy to hear that my idea is doable. thank you very much for your response
@Rick Manner, thank you so much. you just gave me a fantastic idea about creating a Hybrid model. do you have any resources where I can learn how to make hybrid models from two models. I will really love to venture in this direction
Amobichukwu Amanambu, How much control do you have over these two models? Can you start with the PRMS model and "teach" it the ANN model results and then expose the ANN model to addition data to improve the prediction capability or are they both established stagnant models that and longer have "learning" capability?
If the two models are now stagnant models that are no longer evolving in any way you can compare the results of each model vs actual data and attempt to determine the best way to interpolate between the two models to more accurately match the data. I would start with a simple linear interpolation and move on a more complex mathematical interpolation (or extrapolation) I the model prediction errors warrant it..
Another good approach would be to add additional terms and/or adjust existing terms of the current models by including new data in the regression.to help improve the accuracy in regions where the model is known to be weak
thank you very much, Rick Manner, am so happy for this response. it indeed has opened my eyes to see a clearer picture of what to be done. Thank you again
i passed a course called meta heuristic methods the main emphasis of the course was to approve these methods which obtain near optimal solutions such as machine learning methods by deterministic methods
wow @Zohreh Alizadeh, this course must be expository on these methodologies. can i know more?
you can contact Dr Barzinpour via [email protected]
she teaches meta heuristics
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling