Hydro-electric Plant Predictive Maintenance
Customer
Swiss Energy Company
Building a predictive maintenance model to detect abnormal measurements earlier, leading to better maintenance planning and cost reduction
Challenge
- Currently no reliable long-term anomaly detection system
- Slow-shifting anomalies are hard to detect with human surveillance as they can take months
- Goal to have a predictive maintenance system to spot these issues early and plan maintenance more effectively
Solution
- Unit8 built a predictive machine learning model for physical variables of interest (e.g., temperature or vibration)
- Define a threshold for values to be throughout the day and sends automatic alerts if a value differs significantly from model predictions
Business impact
- Engineers able to detect abnormal changes in measurements much earlier
- Maintenance can now be planned effectively
- Shorter stops in electricity production leading to lower costs due to plant shutdown
Electricity Price Forecasting
Customer
Energy Company
Developing forecasting model to predict electricity prices and help traders make better decisions resulting in a 25% reduction in deviation compared to previous models in place
Challenge
- Client needs accurate electricity price forecasts for internal trading team
- Goal to improve price forecasting using a fixed train/valid/test reference dataset
- Need for a new probability forecast to be implemented
Solution
- Built backtesting pipeline and test multiple forecasting configurations (multivariate time series)
- Developed forecasting solution which generated probabilistic forecasts with consistent prediction intervals
Business Impact
- Unit8 model significantly improved forecasts compared to previous solutions in place
- Reduced deviation between forecasted and actual prices
- by 25% compared to previous model