跳至主要内容

Week 3 Progress

 This is the third week of the project. Now that the main part of the project is complete, we have added the GRU method.

simple demostration of GRU algorithm



Now our project has two algorithms: LSTM and GRU. Both methods achieve acceptable prediction results.

In addition, we added a visualization of the predicted results to make it more intuitive to compare the predicted results with the actual results.

Finally, we adjusted the training parameters slightly based on the results, then we got the test prediction result as follows:



Prediction Result for LSTM:

Prediction Result for GRU:

Note that the prediction result may vary depends on training time and sets of data used.

The data set is divided into two parts. One part of the data will be used to train our algorithm (30%), and the other part of the function will be provided to the trained algorithm to predict battery life (70%).





评论

发表评论

此博客中的热门博文

Week 2 Progress

 T his is the second week of the project.   This week we did a lot of programming and a lot of expansions. First of all, we abandoned the bi-LSTM work and used LSTM to complete the project, because bi-LSTM had twice the amount of training as LSTM, and time cost would be a big problem. Then we finished the basic part of the LSTM program. Simple Demostration of LSTM algorithm Besides finishing the LSTM part, we also add two features into our program: · Data processing part. In this part, the program reads the data from the database and processes it in a format that is easy for the main program to read. · Data Visualization Part. In this part, the program will plot the read data for visual viewing.
发表评论
阅读全文

Poster (final version)

 
发表评论
阅读全文

Week 4 Progress

  This is the fourth week of the project.   The main body of work has been completed. The purpose of this week is to make the final adjustment and prepare the bench inspection. First we made some final adjustments to the program to make it as accurate as possible. ·For example, we found that the deviation of data in a certain interval of the data set was too large, which would lead to the deviation of the whole prediction result when generating the prediction result. As shown in the figure, the initial data circled in the figure is obviously "error data", which in some cases eventually leads to inadequate fitting of our prediction results. Therefore, during training, we need to remove the data with large errors in the original data set to ensure the correctness of our training. Second, we finished the poster over the weekend. Finally, we prepared for the bench inspection. Our bench inspection speech is divided into three parts: Part 1: Brief Introduction By Li Yuantong. Part ...
发表评论
阅读全文