Inspiration
Power plants burn the same fuel regardless of whether conditions are optimal. We kept asking: what if a plant could know it was leaving energy on the table, and fix it instantly? That question became EcoOptimizer AI.
What It Does
EcoOptimizer AI predicts energy output using real-time environmental conditions (ambient temperature, pressure, humidity, and exhaust vacuum) and runs an optimization sweep to find the exact vacuum setting that maximizes power generation for those conditions.
The result: a live dashboard that tells operators what to adjust, right now, to extract more megawatts from the same combustion cycle.
How We Built It
We trained a Random Forest Regressor on 9,000+ rows of Combined Cycle Power Plant (CCPP) data. The model learns the nonlinear relationship between environmental inputs and energy output \( PE \), approximating:
$$\hat{PE} = f(AT, V, AP, RH)$$
To optimize, we sweep exhaust vacuum \( V \) across its operational range while holding current conditions fixed, then surface the peak:
$$V^* = \arg\max_{V} \hat{PE}(AT, V, AP, RH)$$
The frontend is built in Streamlit with Plotly for interactive visualization.
Challenges
- Getting meaningful optimization signal from a model trained on observational data rather than controlled experiments
- Designing a UI that communicates ML output clearly to a non-technical operator
- Ensuring the app degrades gracefully without the real dataset
What We Learned
That the biggest wins in sustainability are often not about new technology. They are about using existing infrastructure more intelligently.
Built With
- ccpp
- numpy
- pandas
- plotly
- python
- random-forest-regressor
- scikit-learn
- streamlit
- uci
Log in or sign up for Devpost to join the conversation.