Sentinel

BEHOLD THE HOLY NUMBER 0.00321
Ye ik ik the title is kindda lame but heres the thing: I Spent the last 2-3 hours working on Evaluating the model and Trying to Find a Good System to Find the Anamolies threshold so no anomalies are missed but we dont have alarms blaring all the time. So I (ngl it was from help with gemini) came up with an algo to find the threshold .

So i have Got the whole Thing planned out, I have my Main Model with the Threshold I have set and the model is fed the data, it tries to recontruct it and then It Calculates the Error the Model Made. If this Error is more than the Threshold , then Tier 2 is engaged and its another model which predicts whether its Real or False.

Also If anyone’s wondering like we have climate change, population change and all in the real world, then won’t the Model just be unusable after some time??

So heres the thing, the Model is Not static, it will be continuously retrained on the Data collected over time , this Makes it More Precise and adaptable.

Now Im DONE with the Model and Just working on the GUI for the person to interact with the Model in real time Simulated City.

**MODEL is TRAINEDDD!!!!!
So this is a MASSIVE devlog cuz i forgot TO check my time and wasted like 5 hours …… but heres wat I did:

Completed the Model training
Made a Basic Website Tho I still wanna make the website even better by using a Scroll Animation
Completed the Problem Video
Tested the Model and its AWESOME.
Did some Work on a GUI to showcase the Model, Im thinking like I make a Section in the website where theres a simulated City and the user can mess around with the Sim Parameters to test the Model and all.

SO below you can see the Model Tested on two Test Datasets -> The Entire Anomalous Dataset and the Hourly Anomaly injection Dataset where i just took the Clean Dataset and injected some fake anomaly hours. Also the Only change in the Simulation was a single number change by .15 and the Model was still able to Class it as an anomaly!

Data Simulation Complete!!!

So I finally Completed one of the Harder parts of the Project which was getting the frickin data to Train a Model On.

So Heres wat I did :

I First Extracted some data from the IMD - Indian Weather website using Chatgpt to Compile it into usable Python Data Types.
Now these were in the form of Parameters for the Simulation like the Rainfall Chances, Types, Human Consumption (hourly) etc.

Then Came the Actual Environment Making Which is now Done!!!

So Now Wat i wanna Do is Extract Features from the data Cuz the Model Can’t be trained on Raw Time Series Data. Then Train the Actual Model.

Also Btw Wait for a Awesome Full Intro Vid for the Project!!

So I Spent like 2 Hours Getting a basic Simulation to work, then I Binged like 2-3 YT vids about Anamoly detection models and made a Pretty small one to train.

After like 3-4 Hours of training, the Model didnt Make any progress its still a Baseline random Model. Now Since this was a very small model which had to somehow find anamolies in a Time Series Data that too in Multiple variables I just figured its way too small.

So Heres my Plan:
Complete the Entire Simulation code.
Binge more Videos
Build and train a Better Model.
Also in the Future Make a Gui to see the Simulation Interactively.

I Don’t have much to show rn except some cool lil pic I Took of the rain.

Virtual ENV Data Extraction Stage Complete

So I Spent 2 Hours Collecting Data Regarding the Indian Climate from the IMD website and Used AI to Compile it Into a Useable Form.

So Basically In a Nutshell here’s what Im Planning :
1) I collect the Data regarding the Rain, and Human Water Usage W.R.T to India (Could have been any country but I thought India Would Be easier to verify cuz I live in India). –> This is Done.
2) I use this Data To Make a Virtual Env which Simulate the Water Entering the Drainage Pipes. Basically the Data Collected is Used to Realistically Generate a Day and How Much chances that Rain can occur and how much rain and How much water Used by Humans etc. in that day
3) Since the activity of the Env Is based on The same Background Info, the Anomaly Detection Model learns How the Water Level in the Drain Pipe Works in the env. Then I introduce a Blockage in Pipes, Then Use this to Test the Model.

Also, If you have read this far, thank you 😉

Y’all Ready for This??
So here is what Sentinel is about :

So basically as you know we have drainage pipes which with time develop blockages due to various reasons. These blockages right now have NO way of being detected until they cause waterlogging and cause infrastructural damage and are costly to repair. But these blockages are actually NOT UNDETECTABLE.
So here is the thing -> when the blockages start forming, the water level in the drains acts abnormally. These abnormalities can be detected by using an Anomaly Detection Model .So my entire project is based on this idea.

Here is a 3D animation I spent way too much time on.