LSTM identification and modeling

3 views (last 30 days)
Hello, I have a plant identified in LSTM. I have created normalized datasets, made the network architecture, (numHiddenUnits, numHiddenUnits, layers, maxEpochs, miniBatchSize, options...) Now, based on the weights of the matrices W, R and b of the trained model, I want to make the mathematical model of this plant. State space or its transfer function(s). How would this be done?
Thank you.
Hola,
tengo una planta identificada en LSTM. He creado dataset normalizados, realizado la arquitectura de la red, (numHiddenUnits, numHiddenUnits, layers,maxEpochs, miniBatchSize, options...)
Ahora, a partir de los pesos de las matrices W, R y b del modelo entrenado quiero hacer el modelo matemático de esta planta. Espacio de estados o su(s) funcione(s) de transferencia. ¿Cómo se haría esto?
Gracias.

Accepted Answer

praguna manvi
praguna manvi on 30 Sep 2024
I understand that you are attempting to estimate a state-space model from a trained LSTM model. This task can be challenging for direct translation since LSTMs are complex, inherently nonlinear, and dynamic. They typically do not conform to simple empirical equations that represent the learned relationships.
However, it is possible to capture the behavior of the LSTM model and estimate an approximate “ssm” model using the “estimate” function, as detailed in this documentation:
Hope this helps!
  2 Comments
Jesus Mª Juarez Ferreras
Good morning, if I have a LSTM network trained like this:
[net,info] = trainnet(XTrain,TTrain,layers, “mse”,options);
What data do I put in the following formula?
EstMdl = estimate(Mdl,Y,params0)
Thank you.
Buenos días, si tengo una red LSTM entrenada así:
[net,info] = trainnet(XTrain,TTrain,layers,"mse",options);
¿Qué datos pongo en la siguiente fórmula?
EstMdl = estimate(Mdl,Y,params0)
Gracias.

Sign in to comment.

More Answers (1)

Jesus Mª Juarez Ferreras
Good morning, if I have a LSTM network trained like this:
[net,info] = trainnet(XTrain,TTrain,layers, “mse”,options);
What data do I put in the following formula?
EstMdl = estimate(Mdl,Y,params0)
Thank you.
Buenos días, si tengo una red LSTM entrenada así:
[net,info] = trainnet(XTrain,TTrain,layers,"mse",options);
¿Qué datos pongo en la siguiente fórmula?
EstMdl = estimate(Mdl,Y,params0)
Gracias.

Categories

Find more on Matrix Computations in Help Center and File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!