Picked from the software, topic, and level of “MPPT Solar Panel Fuzzy-Neural Class #5: ANFIS / Fuzzy-Neural Network for Tracking Global Maximum Power Point”.
At most one email a week. Unfollow anytime with one click.
Transcript
The full narration of this video. The current sentence is highlighted while the video plays; click any sentence to jump there.
Show transcript
Introduction
Here is the final result, ANVIS, a combination of fuzzy logic and neural networks, which we wrote from scratch and trained with 3,000 shadow patterns. From the irradiation of the three modules, ANVIS directly guesses at which voltage the global peak lies, with an average error of about 4 volts. Let's get started. Hello, welcome to episode five of the fuzzy-neural MPPT solar panel series. Local algorithms get stuck because they only look at the slope around them. The idea now is to let a model learn the relationship between shadow patterns and the location of the global peak. The model we use is ANVIS, Adaptive Neurofuzzy Inference System from the 1990s, which has fuzzy rules but parameters trained like a neural network.
Step 1: Five ANFIS layers
Create ANVIS.py file ANVIS has 5 layers. Layer 1 is fuzzification with Gaussian membership functions, 3 inputs, each with 3 Gaussians, with centers and widths that can be trained. Layer 2 calculates the strength of each rule as the multiplication of membership degrees, 3 to the power of 3, there are 27 rules. Layer 3 normalizes the rule strength. Layer 4 and 5, each rule has a linear output from the input, Sugeno type 1 model, then all are summed with normalized weights. The training function uses hybrid learning.
Step 2: Hybrid learning
Step forward. With fixed premise parameters, the output of ANVIS is linear to the consequent parameters, So that the 108 parameters can be calculated simultaneously with list squares. Step back. Gala is propagated back to update the centers and widths of Gaussian with gradient descent. Width is kept minimal at 0.05 to avoid being too sharp. Create a file named data underscore mppt.py
Step 3: Shadow pattern data
We generate 3000 random irradiation patterns between 100 and 1000 per square meter, 20% of them without shadows. For each pattern, the area curve is calculated and the global peak voltage is recorded as the target. One important tip, the order of series modules does not affect the curve, so irradiation is ordered from the largest, so that ANVIS only needs to learn one order. Create a file named train underscore anvis.py and run it.
Step 4: Training ANFIS
3000 patterns are completed in about 1 second, with global peak voltages ranging from 24 to 85 volts. There are 27 rules, 108 consequent parameters, and 18 premise parameters. Training for 100 epochs only takes a few seconds. On the test data, the average error is around 4.4 volts, and about 70% of the error predictions are below 5 volts. The model is saved to ANVIS underscore MPPT.NPZ.
Step 5: Training curve & prediction
Let's take a look at the graph. The training curve shows RMSI already low since the first IPO because the list cuares directly found the best consequent, then gradient premises only slightly improved it. The right graph compares guesses with actual values. Data clusters in three regions, according to three possible peaks. Guesses don't have to be exact, just fall on the correct peak slope, then the local algorithm completes it.
Done!
Done. We have built ANVIS Sugeno Order 1 from scratch, understood its five layers, trained it with hybrid learning, and estimated global peak voltage from shadow patterns. In the final episode, we combined ANVIS with FASI and compared all MPPT methods. Currently working on a thesis, dissertation, or research in the field of solar energy and renewable energy. Cerdas Inovasi Digital helps implement solar panel simulations, from 1 diode model, partial shedding, MPPT perturb and observe, and incremental conductance, fuzzy logic, until ANVIS, as well as opening a Python programming course for electrical engineering. Want to master the entire series, from solar panel models up to comparisons of all MPPT methods? All 6 episodes, complete with step-by-step PDF guides and source code, are available for premium tutorial subscribers. Subscribe starting at Rp69.000 per month, save 28% with the 12-month package, or pay once Rp999.000 for lifetime access. Want just one class? Buy the single class with lifetime access. Subscribe now on the premium page cerdasinovasidigital.com. Contact our WhatsApp, or visit cerdasinovasidigital.com. Thank you for watching, until next episode. Thank you for watching,
Ask the class material
Type a question or keyword; we find the part of this episode that explains it and jump the video there.
My notes
Notes are saved with the current video time and are only visible to you.
Questions & answers
Stuck on a step? Ask here; the instructor and other learners can answer.
No questions yet.
Your instructor
Cerdas Inovasi Digital Team
Software developers, BIM & 3D engineers, and AI practitioners
53 video classes1,290 episodes
We build software, BIM models, games and AI automations for clients every day, and turn that hands-on work into step-by-step classes: every step is shown on screen and comes with a PDF guide you can keep.
