Solar Panel Generation Estimate AI Model
When it comes to how much electricity solar cells generate, there are many different factors that must be taken into consideration. The amount of sunlight is an obvious one, but what about conditions such as humidity and wind? It can be difficult to account for everything, which means that solar panels are not always placed in the most efficient locations. With AI, solar farm developers can make sure they are making as much usable energy as possible.
Using an idea developed by Neural Designer and data from Kaggle, I created an AI model that takes into account a variety of elements, including:
Distance to solar noon, in radians
Temperature, in Celsius
Wind direction, in degrees
Wind speed, in meters per second
Sky cover, on a five-step scale from 0 to 4
Visibility, in kilometers
Humidity, in percentage
Average wind speed, in meters per second
Average pressure, in mercury inches
With this information, the model estimates how much electricity each solar cell will produce within a three hour time period. It currently has an accuracy rate of 46%, and while low for a model that needs to be 100% accurate, it is relatively high for an estimate.
The above picture shows the code I wrote for the model in Google Colab. I used a TensorFlow sequential model with three layers with 9, 3, and 1 neurons. Additionally, there were 2920 instances in the data set and I used a 80/20 split for the training and test data.
While building this project, one of the major challenges I faced was figuring out the combination of dense amounts and activation shapes that lead to the highest accuracy rate. This was my first time creating an AI model, so I spent a lot of time learning how to write the needed code and make it as accurate as possible.
For my next projects, I want to look into how AI can impact forms of clean energy in addition to solar. Smart energy grids and maintenance are two areas where I have found machine learning could have a positive impact, so I may try to replicate an existing technology surrounding that.
Here’s a link to the actual code on Google Colab
If you want to learn more about this project, check out this article and video I made :)
