Prediction with Back-Propagation and Linear Regression

Prediction of the power of the turbine of a hydro-electrical plant, using the following algorithms:

• Back-Propagation (BP), implemented by the student
• Multiple Linear Regression (MLR), using free software

• File: turbine.txt
• Columns: 4 variables, 1 value t - predict
  • Variables:
    • Height above sea level
    • Fall
    • Net fall
    • Flux
  • Prediction:
    • Power of the turbine
• Patterns: 451 patterns
  • Training and Validation (and cross-validation): the first 401 patterns
  • Test: the remaining 50 patterns

• Try different values of the training parameters and select those with the best results:
  • Architecture of the neural network
  • Type of scaling of the data
  • Initial range of weights and thresholds
  • Learning rate and Momentum
  • Batched/online
  • Number of training epochs