Question: Can the platform make a profit using the computing power of idle devices available for rent?

My Science Fair Project is about introducing the idea of creating a platform that allows people to share the computing power of spare electronic devices when they’re idle. This platform could connect many of these devices to work on larger tasks such as weather forecasting, space exploration, cryptocurrency calculations, or scientific simulations, at a lower cost and a larger scale than traditional cloud computing. It could also provide extra computing power to users who want it.

Since users of this future platform would have to pay higher electricity bills if they were to rent their own computing power, I’m also going to create a model using code that predicts electricity costs in different areas to find the best times and places to run these devices in a cost-effective way. Using my code, I will get past electricity price data to train an LSTM model and compare predicted prices to the actual ones to determine the accuracy of my model. My future vision is a website or app where users can join and share their idle devices to earn money.

To ensure user privacy, the future platform could only share computing power, not personal data, and each task would run in a secure, isolated environment, so no personal information or files would be accessible to the platform or other users.

 The platform would also ideally monitor and manage the workload, ensuring devices are not overused to the point of damage that they become no longer usable, which turns them into actual electronic waste. A future addition to the platform could allow participants to choose the extent and duration of their device's involvement, which helps balance use and longevity.

This would be pretty similar to sharing economy models like Airbnb and Uber, or cloud services like Google, Youtube and Roblox, but instead of renting out homes and cars, people are renting out unused computing power to people who need it . 

I hope this approach will help make use of idle computiing power and turn underrated devices into a useful resource, where users can make money, contribute to or easily do big computing projects, lessen the strain on the electricity grid, and help the environment.

1. First, I will do some research on LSTM, Bitcoin, Picture Recognition, Convolutional Neural Networks, etc.
2. Secondly, I will create some relatively easy Python code to run my LSTM Model, Bitcoin Price Prediction Model, Picture Recognition Code, Stress Test, and the code that will turn on a computer remotely. 
3. Next, I will conduct experiments by simulating computing power consumption by stress-testing a computer’s CPU%, GPU%, RAM% etc., and compare it to the electricity cost, which will be calculated by measuring parameters of electricity such as voltage, frequency, and current.
4. After that, I will determine the exact timing of the mining schedule using the electricity price predictions, and simulate mining bitcoin with the schedule. 
5. I’m going to be using an LSTM model incorporated with a professional financial analyst library to predict bitcoin prices 1 minute into the future.
6. Afterwards, I’ll try to make stock trading automatic by assuming that a user has $1M to start with and trades the bitcoins when prices are high, and keeps them for later when prices are not high enough.
7. Finally, I will summarize the information I gathered from my research, experiments, and predictions, and confirm my hypotheses.

During my experiment, I noticed that:

• The power wattage was in the range of 17 W to 91W. 17 W was the baseline for Windows to run, and 91W is for when the CPU Usage was 100%, the highest possible computing power used.
• The electricity wattage power or cost was highly correlated to CPU% and GPU% workload. 
• The RAM and Network Card were practically negligible during the stress test.
• From the LSTM Price Prediction experiment, I noticed that for four out of the five cities I trained the model with, the predicted electricity prices each had a high accuracy rate. As displayed in the tables and charts below, electricity prices in London, UK could be predicted within an error tolerance level of 10 cent/KWh from the actual price 96% of the time. Ankara, Turkey had the lowest accuracy rate due to high volatility, with only 29%.
• For the Bitcoin Price Prediction model, after training the model with 70% of the historical data, validating with the 20%, and testing with the last 10%, I gave the actual and predicted prices to the trading automation module. The statistic showed that it had 54.76% of successful trades. In my experiment, to simulate a real-life example, I assumed that if a user invests $1M at the start of the simulation in your account, he or she can finish the next month with a value of $1,029,100. The calculated annual return of income in 12 months was 35.052%
• Using the electricity prices predicted for California in my model, I executed the mining schedule and made trading decisions based on the forecasted bitcoin prices. As you can see from the tables, if you choose to yield with the grid, you can make a profit of $22.7336 in a 24-hours span with just one Tesla Model3. If you want to compete with the grid (though it’s not recommended) you can make about $ 34.0162 per day!
• According to Google, there were 475,592 Tesla vehicles registered in the State of California as of December 31, 2023. Assuming 10% of the vehicles deploy my strategy, these Tesla owners can harvest $1,081,191.8 in one day! If I charge a 10% royalty fee, I could make a daily profit of $108,119.18, from one state alone!
   

It is calculated that such a platform would be profitable because, according to the graphs seen earlier, and my observations, it was predicted that a daily profit of $108,119.18 could be made from one state alone (California in this case), and only using one type of computing (bitcoin mining). Here are some additional things I found out: 

• The CPU Usage, or computing power used, is highly related to the electricity cost.
• Electricity and bitcoin prices can be predicted satisfactorily with this model in some regions with quality data.
• Experiments showed that the electricity supply and demand can be balanced to certain extent through this method (the mining schedule)
• Users could make a decent profit using my trading model, though it could be improved.
• It’s feasible that Tesla owners can use my app to mine bitcoins during off-peak hours and even make a profit when they are sleeping! Profits are larger in regions where price differentials are bigger. (Prices in Canada are lower than in the USA.)

Therefore, my hypotheses are theoretically correct.
 

https://github.com/optuna/optuna
https://github.com/peerchemist/finta
https://www.kaggle.com/aipeli/btcusdt
https://web.stanford.edu/class/cs379c/archive/2018/class_messages_listing/content/Artificial_Neural_Network_Technology_Tutorials/OlahLSTM-NEURAL-NETWORK-TUTORIAL-15 by Stanford.
Sima Siami Namini, Akbar Siami Namin. Forecasting Economics and Financial Time Series: ARIMA vs. LSTM. ArXiv: 1803.06386,2018.
Bao W, Yue J, Rao Y. A deep learning framework for financial time series using stacked auto encoders and long-short term memory. In PLoS ONE 12(7): e0180944, 2017.
Aniruddha Dutta, Saket Kumar and Meheli Basu. A Gated Recurrent Unit Approach to Bitcoin Price Prediction. , 2019.
Bouri, E, Molnár, P, Azzi, G, Roubaud, D. On the hedge and safe haven properties of Bitcoin: Is it really more than a diversifier
https://www.ceicdata.com/zh-hans/turkey/environmental-environmental-policy-taxes-and-transfers-oecd-member-annual/tr-industry-electricity-price-usd-per-kwh
Vasudharini Sridharan, Mingjian Tuo and Xingpeng Li, "Wholesale Electricity Price Forecasting using Integrated Long-term Recurrent Convolutional Network Model",https://arxiv.org/abs/2112.13681.

Thank you to the following people for their support!

•    Ms. Ruzycki, my science fair coordinator
•    My parents, who helped me with my experiment and trifold.
•    CYSF! The website had many helpful tips.
 

Thank you all for your support, encouragement, and belief in my work. This project would not have been possible without you.