How to do a forecast with FPM-ANN hybrid model in Matlab

Hi, I need to utilize a hybrid of First Principle model and ANN for forecasting values. How can I implement the FPM-ANN hybrid in MATLAB? The FPM is a simple PDE developed in GAMS.

I'm hoping to use FPM to optimise the ANN weight and bias. I already have a NARX netowrk with 10 hidden neurons and 6 time delays which gives a good network fit. There's 10 variables and 89 timestpes in my data.

Thank you!

Answer 1

[Dhruv]

To implement the FPM-ANN hybrid in MATLAB, follow these steps:

1. Firstly, you need to export the data and load it into MATLAB with the help of readtable function, to read the data from a CSV file.
2. Next, try creating an initial ANN model in MATLAB. Use the narxnet function to create a NARX network with 10 hidden neurons and 6 time delays.
3. Then, use the trainlm function to train the ANN model on the input data. This function uses the Levenberg-Marquardt algorithm to train the network.
4. After training the ANN, use the FPM to optimize the ANN weight and bias. To do this, define the FPM in MATLAB using the pdepe function. This function solves partial differential equations using the method of lines.
5. Following which try using the fmincon function to minimize the error between the FPM solution and the ANN output. This function uses a constrained optimization algorithm to find the minimum error.
6. Finally, update the weights and biases of the ANN using the optimized values obtained from the FPM. This will improve the accuracy of the ANN predictions.

Overall, the steps involved in implementing a FPM-ANN hybrid in MATLAB are: load the data, create and train the ANN, define and solve the FPM, optimize the ANN using the FPM, and update the weights and biases of the ANN.