M0NARQ | PowerGuard

THE ARBITRAGE CRISIS.

Factories in Gazipur often operate at <30% capacity due to chronic LNG shortages. The alternative—captive diesel generation—costs $0.32/kWh, nearly triple the grid rate of $0.12/kWh. Running a dyeing line on diesel destroys profitability within hours.

Additionally, unmonitored "phantom loads" (machines left idle, inefficient motors) contribute to a 12-15% energy overspend. Without granular data, factory owners cannot identify where the waste is occurring.

$180

SAVED PER LINE/MO

Through 15% energy reduction.

100%

AUDIT READY

Immutable Parquet logs for EU.

Current Solutions Shortfall: Standard smart meters only provide aggregate data (total kWh). They cannot predict spikes or identify specific failing components via power signatures. They are "post-mortem" tools.

ANALYTICS PIPELINE

We use the PZEM-004T split-core sensor ($9) to monitor power quality at 1Hz. The data feeds into a TS2Vec model running on an ESP32 microcontroller.

SOLVING CONCEPT DRIFT (MOMEMTO)
Traditional forecasting (ARIMA) fails in factories because machine behavior changes as parts wear out ("Concept Drift"). We use TS2Vec with a Memory Gate (MOMEMTO) to stabilize learning. This acts as a long-term memory, ensuring the model doesn't over-generalize or "forget" the baseline behavior over weeks of operation.
BAYESIAN CPD
Change Point Detection. We use Bayesian algorithms (Ruptures library) to identify structural shifts in consumption—distinguishing between a normal cycle change and a motor seizing up (anomaly).

EU COMPLIANCE ENGINE

Digital Product Passport (DPP): By 2026, every garment entering the EU must have a digital energy footprint. PowerGuard automatically generates cryptographically signed Parquet files per batch, creating a "Green Certificate" that satisfies CBAM audits.

# ENERGY ANALYTICS & PEAK SHAVING
from ts2vec import TS2Vec
import ruptures as rpt

class PowerMonitor:
    def analyze_stream(self, readings):
        # 1. Vectorize Data Stream (TS2Vec)
        # Handles concept drift via Memory Gate
        vec = self.ts2vec.encode(readings)
        
        # 2. Bayesian Change Point Detection
        # Identify structural shifts (start/stop/fault)
        algo = rpt.Pelt(model="rbf").fit(vec)
        breakpoints = algo.predict(pen=3)
        
        # 3. Tariff Logic (Bangladesh Context)
        # Peak: 17:00 - 23:00 ($0.12/kWh)
        is_peak = 17 <= datetime.now().hour <= 23
        
        # 4. Peak Shaving Logic
        # If spike predicted during peak hours, shed load
        if is_peak and prediction.spike_prob > 0.8:
             return "SHED_LOAD_AC_UNIT_4"
             
        return "MONITOR"

SENSOR	PZEM-004T v3 ($9)
CONTROLLER	ESP32 WROOM ($4)
SAMPLING	1Hz (Voltage, Current, PF)
PROTOCOL	MQTT over TLS 1.3

WIDER APPLICABILITY