Comparing and Demonstrating

After both the models were trained, the performance of the trained shallow neural network and LSTM model was evaluated using the following decision matrix shown in Table 1 below

Table 1: Evaluation of the Best LSTM and Shallow Neural Network Models Trained and Tested using Foreign Data

The error values shown above were extracted from the best performing LSTM and shallow neural network models. It can be seen that the LSTM model performs better in terms of how accurately it plots the forecasted capacitance values to the actual values for different training and testing sets. Note that not only did the best performing LSTM model outweigh the performance of the shallow neural network, it is also important to observe how the models performed when compared to the worst results obtained form both the trained models

Table 2: Evaluation of the Worst LSTM and Shallow Neural Network Models Trained and Tested using Foreign Data

As seen form the RMSE recorded values for the worst performing LSTM and shallow neural network models, when foreign data was used meaning a set of data that both the models were not trained with but were tested with, the LSTM model still performed way better as the error value was slightly far from 0 but not further then the shallow neural network. For the ideal applications of this algorithm in the industries, it can be deemed that the LSTM model can be a better option as it handles random data better then the countering model tested in this project.

An important note made from the comparison matrices shown in tables 1 and 2, was that among the best and worst performing models of LSTM and shallow neural network, the LSTM model performed better in both cases. These performing models were compared for when they were trained with a different set of data and tested using a different set. The significance of this comparison was that it proved that the LSTM model would be a better choice for real life industrial application as the capacitance values would be unknown to the model and very different to the values with which it was trained with and the very ability of the model to still accurately plot the capacitance curve showing the degrading nature of the supercapacitor with minimum errors would deem it more superior.

In Conclusion 

                                         

                                             LSTM

LSTM was chosen as the more superior model.

To watch the full demonstration of the codes and simulations carried out on Matlab, please use the following link: 

https://drive.google.com/file/d/1d3u0wL-NAJio4F_V0ifrflVWTNPRepA3/view?usp=sharing 

    ------------------------------------------- Thank You For Following ----------------------------------

Edited by Shahil and Henal

S11172483@student.usp.ac.fj 

S11085370@student.usp.ac.fj