1. Objectives
- Build a zero-order Sugeno system: constant rule outputs combined by a weighted average.
- Compare Sugeno and Mamdani outputs with response curves and control surfaces.
- Build a first-order Sugeno system whose rule outputs depend on the inputs.
- Explain when to choose Mamdani and when to choose Sugeno.
2. Word Bank
Word | Simple meaning | Real-world example |
|---|---|---|
Sugeno system | A fuzzy system whose rule outputs are numbers or formulas, not fuzzy sets | IF hot AND NOT dry THEN fan = 83 |
Zero-order Sugeno | Each rule's output is a constant | slow = 17, medium = 50, fast = 83 |
First-order Sugeno | Each rule's output is a linear formula of the inputs | fan = 40 + 1.2 x temperature |
Weighted average | Σ (strength x output) / Σ strength | (0.33 x 17 + 0.2 x 83) / 0.53 |
Control surface | Output plotted over every input pair | 3-D 'landscape' of fan speed |
Interpretable | A person can read and understand it | Mamdani rules with fuzzy outputs |
3. Background
Sugeno in one line
Michio Sugeno and Tomohiro Takagi (1985) replaced the fuzzy THEN-sets with crisp outputs. Each rule gives a number z (a constant in a ZERO-ORDER system, a formula like a + b·t + c·h in a FIRST-ORDER system). The final output is the weighted average Σ w·z / Σ w, where w is the rule's firing strength. No clipping, no aggregation, no centroid - so it is much faster and works well inside optimisation and learning (ANFIS, P11).
Same rules, same fire()
The IF parts and firing strengths are exactly those of P09a. Only the THEN part changes. We choose the constants 17, 50 and 83 close to the centroids of the slow, medium and fast sets, so the two systems should behave alike.
Which to choose?
Mamdani: rule outputs are fuzzy words, easy for experts to read - good for advice systems. Sugeno: fast, smooth, easy to tune from data - good for control on small devices and for learning.
Data files for this lab
No data file is needed - the programs make their own data.
4. Time Plan (50 minutes)
Minutes | Activity | Part |
|---|---|---|
0-5 | Background: weighted average; zero- and first-order Sugeno | 1-3 |
5-10 | Guided warm-up: one Sugeno inference; Sugeno vs Mamdani curve | Part A |
10-13 | Predict the output | Part B |
13-17 | Bug hunt | Part C |
17-47 | Graded tasks 1-3 | Part D |
47-50 | Wrap-up; post-lab questions set as homework | Part E |
Part A - Guided Warm-up (type, run, observe)
A1. One zero-order Sugeno inference (30 °C, 40 %)
The same five strengths as P09a; each rule now gives a constant. The answer is their weighted average.
import numpy as np import skfuzzy as fuzz t_x = np.arange(0, 51, 1); h_x = np.arange(0, 101, 1); f_x = np.arange(0, 101, 1) T = {"cold": fuzz.trapmf(t_x, [0, 0, 10, 20]), "comfortable": fuzz.trimf(t_x, [15, 24, 33]), "hot": fuzz.trapmf(t_x, [28, 38, 50, 50])} H = {"dry": fuzz.trapmf(h_x, [0, 0, 20, 50]), "normal": fuzz.trimf(h_x, [30, 50, 70]), "humid": fuzz.trapmf(h_x, [50, 80, 100, 100])} F = {"slow": fuzz.trimf(f_x, [0, 0, 50]), "medium": fuzz.trimf(f_x, [25, 50, 75]), "fast": fuzz.trimf(f_x, [50, 100, 100])} mu = lambda x_axis, mf, value: float(fuzz.interp_membership(x_axis, mf, value)) def fire(t, h): cold, comf, hot = (mu(t_x, T[k], t) for k in ("cold", "comfortable", "hot")) dry = mu(h_x, H["dry"], h) return [("slow", cold), ("slow", min(comf, dry)), ("medium", min(comf, 1 - dry)), ("medium", min(hot, dry)), ("fast", min(hot, 1 - dry))] def mamdani(t, h): agg = np.zeros_like(f_x, dtype=float) for term, w in fire(t, h): agg = np.fmax(agg, np.fmin(F[term], w)) return fuzz.defuzz(f_x, agg, "centroid") Z = {"slow": 17, "medium": 50, "fast": 83} # zero-order: one constant per output def sugeno(t, h): rules = fire(t, h) w = np.array([s for _, s in rules]) z = np.array([Z[term] for term, _ in rules]) return (w * z).sum() / w.sum() # weighted average t, h = 30, 40 for i, (term, w) in enumerate(fire(t, h), 1): print(f"R{i} {term:<6} w = {w:.2f} z = {Z[term]} w x z = {w * Z[term]:.2f}") print(f"Sugeno: {sugeno(t, h):.2f} % Mamdani (P09a): {mamdani(t, h):.2f} %") |
Output R1 slow w = 0.00 z = 17 w x z = 0.00 R2 slow w = 0.33 z = 17 w x z = 5.67 R3 medium w = 0.33 z = 50 w x z = 16.67 R4 medium w = 0.20 z = 50 w x z = 10.00 R5 fast w = 0.20 z = 83 w x z = 16.60 Sugeno: 45.88 % Mamdani (P09a): 45.04 % |
A2. Sugeno and Mamdani side by side
Fan speed against temperature at 70 % humidity for both systems.
# ... set-up: imports, the P08 sets, fire() and mamdani() from P09a (copy from Part A1) ... Z = {"slow": 17, "medium": 50, "fast": 83} # zero-order: one constant per output def sugeno(t, h): rules = fire(t, h) w = np.array([s for _, s in rules]) z = np.array([Z[term] for term, _ in rules]) return (w * z).sum() / w.sum() # weighted average import matplotlib.pyplot as plt temps = np.arange(0, 51, 1) sug = np.array([sugeno(t, 70) for t in temps]); mam = np.array([mamdani(t, 70) for t in temps]) plt.figure(figsize=(7, 2.6)); plt.plot(temps, mam, label="Mamdani"); plt.plot(temps, sug, "--", label="Sugeno (zero-order)") plt.xlabel("Temperature (°C)"); plt.ylabel("Fan speed (%)"); plt.title("Humidity 70 %"); plt.legend(); plt.grid(alpha=0.3) plt.tight_layout(); plt.savefig("compare.png", dpi=90) d = np.abs(sug - mam) print(f"largest difference {d.max():.2f} % at {temps[d.argmax()]} C; mean difference {d.mean():.2f} %") print("Saved compare.png") |
Output largest difference 2.44 % at 15 C; mean difference 0.68 % Saved compare.png |
Part B - Predict the Output
Write your prediction BEFORE running the code. Then run it and correct yourself in a different colour.
B1.
w = [0.2, 0.6] z = [20, 50] print(sum(a * b for a, b in zip(w, z)) / sum(w)) |
My prediction |
B2.
w = [0.0, 0.3, 0.0] z = [17, 83, 50] print(sum(a * b for a, b in zip(w, z)) / sum(w)) |
My prediction |
B3.
t, h = 36, 70 z_fast = 40 + 1.2 * t # first-order output of the 'fast' rule print(round(z_fast, 1)) |
My prediction |
Part C - Bug Hunt
Each program has ONE bug. Some crash with an error message; others run but give a wrong answer (silent logic error - the dangerous kind). Find the bug, explain it and write the fix.
C1.
# ... set-up: imports, the P08 sets, fire() and mamdani() from P09a (copy from Part A1) ... Z = {"slow": 17, "medium": 50, "fast": 83} rules = fire(5, 50)[1:] # rules 2-5 only w = [s for _, s in rules]; z = [Z[term] for term, _ in rules] print(sum(a * b for a, b in zip(w, z)) / sum(w)) |
Bug, why it happens, and the fixed line(s) |
C2.
# ... set-up: imports, the P08 sets, fire() and mamdani() from P09a (copy from Part A1) ... Z = {"slow": 17, "medium": 50, "fast": 83} rules = fire(30, 40) w = np.array([s for _, s in rules]); z = np.array([Z[term] for term, _ in rules]) print(round((w * z).sum() / len(rules), 2)) |
It runs and prints 9.79 |
Bug, why it happens, and the fixed line(s) |
C3.
# ... set-up: imports, the P08 sets, fire() and mamdani() from P09a (copy from Part A1) ... t, h = 30, 40 comf, hot = mu(t_x, T["comfortable"], t), mu(t_x, T["hot"], t) dry = mu(h_x, H["dry"], h) w = [max(comf, dry), max(comf, 1 - dry), max(hot, dry), max(hot, 1 - dry)] # R2-R5 (AND) z = [17, 50, 50, 83] print(round(sum(a * b for a, b in zip(w, z)) / sum(w), 2)) |
It runs and prints 55.5 |
Bug, why it happens, and the fixed line(s) |
Part D - Graded Tasks (10 marks)
Task | Level | Title | Marks |
|---|---|---|---|
T1 | Easy | Zero-order Sugeno for the five days | 3 |
T2 | Medium | Two control surfaces | 3 |
T3 | Challenge | A first-order Sugeno system | 4 |
Total | 10 |
Task 1 (Easy, 3 marks): Zero-order Sugeno for the five days
Write sugeno(t, h) with Z = {slow: 17, medium: 50, fast: 83} and the weighted average. Print Sugeno and Mamdani speeds for the five P07b days (30, 40), (34, 65), (38, 75), (41, 55), (36, 90) and the difference. Why do they differ most on day 2?
Sample run (your output must match):
Output day input Sugeno Mamdani diff 1 30 C, 40 % 45.88 45.04 +0.83 2 34 C, 65 % 83.00 81.43 +1.57 3 38 C, 75 % 83.00 83.33 -0.33 4 41 C, 55 % 83.00 83.33 -0.33 5 36 C, 90 % 83.00 82.78 +0.22 |
Your code (or attach the file) |
Task 2 (Medium, 3 marks): Two control surfaces
Compute both systems on a grid (temperature 0-50 step 2, humidity 0-100 step 5) and draw the two surfaces side by side (two 3-D subplots, save task2.png). Print the largest difference and the input where it happens, and the mean difference.
Sample run (your output must match):
Output largest difference 7.01 % at 18 C, 45 % mean difference 0.78 % Saved task2.png |
Your code (or attach the file) |
Task 3 (Challenge, 4 marks): A first-order Sugeno system
Make the rule outputs depend on the inputs: slow z = 15; medium z = 30 + t; fast z = 50 + 0.8·t + 0.1·h (more humid -> faster). Write sugeno1(t, h). Print the five days, and plot fan speed against temperature at humidity 40 % and 90 % for the first-order system (save task3.png). What can a first-order system do that the zero-order one cannot?
Sample run (your output must match):
Output day 1: 30 C, 40 % -> 49.31 % day 2: 34 C, 65 % -> 83.70 % day 3: 38 C, 75 % -> 87.90 % day 4: 41 C, 55 % -> 88.30 % day 5: 36 C, 90 % -> 87.80 % at 45 C: humidity 40 % -> 85.0 %, humidity 90 % -> 95.0 % Saved task3.png |
Your code (or attach the file) |
Part E - Post-Lab Questions (finish as homework if time runs out)
E1. Write the Sugeno output formula and explain each symbol.
Your answer |
E2. Give two differences between Mamdani and Sugeno systems.
Your answer |
E3. Why did dividing by the number of rules give 9.79 % instead of 45.88 %?
Your answer |
Marking Rubric (applied to Part D)
Criterion | Full | Partial | None |
|---|---|---|---|
Code runs without errors | Runs fully | Small fixable error | Does not run |
Correct weighted average and first-order formulas | All correct | One slip | Several wrong |
Tables, surfaces and curves | All present | Some missing | None |
Explanations (day 2, surfaces, first-order) | Clear and correct | Partly correct | Missing |
Submission
- Make a folder RollNo_P09b with task1.py, task2.py, task3.py (or P09b.ipynb), task2.png and task3.png.
- Upload the zipped folder to the LMS / Google Classroom by the end of the day.
- Copy Task 1 and Task 3 code and output into your Lab Book (Practical 09b page).
- Keep sugeno1() - ANFIS (P11) learns first-order Sugeno formulas from data.