2024 - 2026 MPPT Algorithm Software

Requirements

• Faster + more accurate than previous

  • Handles sudden environment changes

• Withdraws less power or equal

Plan

• Experiment with deep learning

Challenges

• Gathering partially shaded data from simulation to train on

• Finding the power withdrawal from each algorithmic run

Design Decisions

• Reinforcement learning

  • Similar to FLC (Fuzzy Logic Control) but a step higher

  • Pros

    • Able to find the maximum power under multiple local maximums

      • PandO will fluctuate between local maximums

    • Less steps required to find the peak

      • Less time between steps to stabilize

      • Less number of times to boost

      • Less ripple percentage

    • Finding the peak when the environment changes

      • Finding the best action in sudden environmental changes

      • PSO (Particle Swarm Optimization) and other algorithms struggle to find the peak during sudden changes

  • Cons

    • Amount of power withdrawal

• Tuning existing algorithms using ML

  • Weights on existing algorithms can be better

Diary