busi732 / Team-2-2024Fall

This is a project repository for Team 2 members in BUSI 732 Quantitative Research, 2024Fall
0 stars 3 forks source link

Model Research #10

Open ajhaller opened 6 days ago

rr-85 commented 9 hours ago

The merged dataset appears to be extensive, containing columns on fault occurrences, weather conditions, turbine status, and various operational parameters. Given this, here are some data science approaches for optimizing predictive maintenance:

Random Forest Regression (Predictive Modeling)

Purpose: Use this model to predict the likelihood and timing of faults based on historical data. Pros: Handles mixed data types and provides feature importance, highlighting which parameters influence fault occurrence most. Cons: Computationally intensive; may require tuning to prevent overfitting. Fit with Data: Random Forest models perform well on time-series or event-driven data like fault records, making it useful for interpreting the relationship between weather, operational conditions, and fault occurrences. Gradient Boosting Machines (GBM)

Purpose: Optimize the timing of maintenance actions by predicting fault risks and prioritizing which faults to address. Pros: GBMs excel in capturing complex patterns; can be used to create probabilistic predictions for fault likelihood. Cons: Requires tuning and may overfit on smaller datasets. Fit with Data: Effective for datasets with multiple features influencing outcomes, such as energy production and fault metrics.

Markov Decision Process (MDP) for Maintenance Scheduling Optimization

Purpose: To create a strategy that minimizes maintenance costs while maximizing uptime, considering the variable cost of in-house vs. external maintenance. Pros: Considers sequential decision-making; can optimize costs by balancing preventative maintenance and repair. Cons: Requires assumptions about state transitions and fault frequencies. Fit with Data: MDP is suitable for data with state-based transitions, like turbine operational states and maintenance options.

Reinforcement Learning (RL) for Preventative vs. Reactive Maintenance Strategy

Purpose: Develop a dynamic policy to determine whether to conduct preventative maintenance based on evolving conditions. Pros: RL adapts to new data, allowing the model to optimize decisions based on ongoing operations. Cons: Needs significant data and computational resources; may be complex to set up. Fit with Data: RL fits well with continuous decision-making processes in variable environments, like adjusting for fluctuating weather and production needs