# Appendix D: Core Algorithm Description and Key Parameters for STM32 MCU
## Target Controller
- MCU Model: STM32F103C8T6
- RAM Capacity: 20 KB
- Flash Capacity: 64 KB
- Core Frequency: 72 MHz

## Core Algorithm Architecture
1. Input Layer
   - Input variables: agricultural season, time period, farmer type, horizontal irradiance
   - Sampling interval: 1 hour
   - Quantization bit width: 16-bit fixed point

2. Lightweight Fuzzy Logic Inference Engine
   - Membership function type: Triangular
   - Core fuzzy rules: 25 rules (see Appendix B)
   - Defuzzification method: Weighted average
   - Rule coverage rate: 95.3%

3. Adaptive Correction Module
   - Algorithm: Recursive least squares (RLS) with 7-day sliding window
   - Function: Dynamically adjust model weight according to real-time load change
   - Correction period: ≤ 24 hours
   - No extra RAM overhead

4. Human-Centred Control Execution Layer
   - Operation step limit: ≤ 3 steps
   - Command response time: ≤ 2 seconds
   - Misoperation auto-correction: Supported
   - Fault tolerance mechanism: Enabled

## Performance Metrics on STM32
- Single-step prediction time: 186 ms
- Peak RAM usage: 27.4 KB
- Average prediction MAPE: 10.8%
- Average prediction R²: 0.932

## Key System Parameters
- PV array capacity: 5 kWp
- Energy storage capacity: 10 kWh
- System voltage: DC 48 V
- Charge/discharge rate: 0.5 C
- SOC operating range: 20% – 80%
- Conservative SOC lower limit: 22%
- Voltage deviation allowance: ±7%
- Total harmonic distortion (THD): ≤5%

## Memory Optimization Methods
- Floating-point quantization to 16-bit fixed point
- Membership function lookup table instead of real-time calculation
- Redundant input variable pruning
- Static memory allocation only (no dynamic malloc)

## Safety and Human-Centred Constraints
- Max operation steps per task: 3
- Max cognitive load design: K = 35 (NASA-TLX)
- Auto-correction rate for misoperations: ≥ 98%
- Real-time performance: Meet rural low-cost controller requirements