initial import
Browse files
README.md
CHANGED
@@ -1,3 +1,21 @@
|
|
1 |
---
|
2 |
license: cc-by-nc-4.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: cc-by-nc-4.0
|
3 |
---
|
4 |
+
### AI Energy Forecast using LTSM
|
5 |
+
|
6 |
+
It basically takes some smartmeter data (5 cols, > 12mil. instances, cols: id, device_name, property, value, timestamp) and creates a custom forecast based on selected window.
|
7 |
+
The file is available in .py and .ipynb format, so you can choose according to your preferences.
|
8 |
+
|
9 |
+
Please notice that once you load up the smartmeter data, there are inputs created on the timestamp col like wd_input (the weekday of the timestamp), as well as a cos(inus) and sin(us)
|
10 |
+
time inputs, giving the model the ability to keep track of the daytime of each instance. Finally, the inputs are merged to an input df, standardized and differenced.
|
11 |
+
After that, some functions are used to give the user the ability to use time windows from the data. Based on these, the model generates forecasts.
|
12 |
+
|
13 |
+
![Model](https://github.com/databloom-ai/LLM-LTSM/blob/main/energy-forcast/model.png?raw=true)
|
14 |
+
|
15 |
+
The first models created are a simple baseline model, used for evaluating the performance of the later on built LTSM model. The baseline model simply shifts the values by t=1. Hence,
|
16 |
+
there is no t=0 and each timestamp uses the value from t-1.
|
17 |
+
Finally, there's the 2-layer plain vanilla LTSM. After 11 epochs, I reached a loss of 10.86 which is rather mediocre. However, the main idea here is to build a basic forecasting model
|
18 |
+
for which this seems appropriate.
|
19 |
+
|
20 |
+
|
21 |
+
![LTSM](https://github.com/databloom-ai/LLM-LTSM/blob/main/energy-forcast/LTSM.png?raw=true)
|