Smart Energy-Meter Reading & Power-Theft Anomaly Logger (ESP32, Python)

Smart Energy-Meter Reading & Theft-Anomaly Logger (SEM-TAL) is a buildable metering unit for a hostel block, a department building or a row of rented shops — the places where an institution's electricity bill usually goes wrong. A clamp-on current transformer (SCT-013-000) and a ZMPT101B voltage-sensing module feed an ESP32, which computes RMS voltage, RMS current, real power and running kWh every second, shows them on a 16x2 I2C LCD, and writes a 1-minute interval record to a microSD card. Over Wi-Fi, the node pushes the same records to a small Python (Flask + SQLite) dashboard that plots the daily load curve and lists every anomaly the firmware raised.

Three anomaly rules run on the device itself: (1) overload — current above the sanctioned limit for more than a few minutes; (2) tamper / direct hooking — load current detected while the voltage channel says the supply at the meter is effectively disconnected, the classic signature of a bypass; (3) off-hours surge — a night-time load profile far outside the learned baseline. Every flag carries an NTP-synced timestamp.

C firmware runs on the ESP32 (Arduino framework), which the VTU 2022 scheme teaches in the IoT lab; Python and Scilab/MATLAB handle the dashboard, calibration regression and load-profile analysis. Flask and SQLite are industry-standard extras beyond the syllabus, chosen because they keep the server small enough to explain end-to-end in the viva.

Hostels, department buildings and rented shops are billed from a single monthly meter reading, with no record of when energy was actually used. As a result, overloads from geysers and heaters go unnoticed until the breaker trips, suspected power theft or meter bypass cannot be evidenced or ruled out because there is no timestamped data, and disputes between owners and tenants degenerate into arguments. Commercial smart meters and DISCOM interval data are not accessible to a college or a small landlord.

There is a need for a low-cost, safe, buildable energy-monitoring node that measures single-phase consumption at 1-minute resolution, stores it locally so no data is lost, flags overload and tamper signatures as they happen, and presents the load profile and anomaly history on a simple dashboard — giving the warden or owner the same kind of evidence a DISCOM interval meter would, at a student-project price.

Abstract

Institutional and rental consumers in India rarely see their own consumption until the monthly bill arrives. This project builds a single-phase metering node that measures voltage, current, real power and energy, stores 1-minute interval records with timestamps, and raises on-device anomaly flags for overload, tamper and off-hours surges. A Flask + SQLite dashboard visualises the load curve and anomaly history. The unit was calibrated against a commercial clamp meter and a plug-in kWh meter and trialled for two weeks on a college building feeder.

Introduction

Hostels and small commercial rows are billed on one static meter that someone reads once a month. Between readings, nobody knows whether tonight's load was 2 kW or 8 kW, why the bill jumped last March, or whether a tenant has hooked a line directly off the pole. Interval metering answers all three — which is why distribution companies are moving to smart meters — but a commercial three-phase smart meter is out of a student's budget, and a hostel warden cannot request DISCOM interval data. A ₹2,300 open-hardware node fills the gap for study and demonstration.

Existing System

• One electromechanical or basic static meter per building, read manually once a month; no interval data.
• Theft and tampering surface only during a flying-squad inspection or after an abnormally high bill — weeks after the event.
• Overloads (geysers, heaters, motors) trip the breaker repeatedly with no record of when.
• Limitations: no timestamps, no load profile, no evidence trail, no early warning; disputes between tenants and owners are settled by argument, not data.

Proposed System

• Sensing: non-invasive SCT-013-000 clamp CT (20 A working range) plus ZMPT101B mains-voltage module, so the circuit never touches mains directly.
• Computation: the ESP32 samples both channels, computes true RMS and real power (mean of the v·i product, so power factor is measured, not assumed), and accumulates kWh.
• Anomaly engine on-device: sanctioned-limit overload, current-with-no-voltage (tamper/direct hooking) and off-hours baseline breach, each debounced to suppress false alarms.
• Storage and connectivity: every 1-minute interval goes to microSD first (source of truth), then in batches over Wi-Fi to the dashboard; missed Wi-Fi periods are back-filled from the card.
• Dashboard: Flask + SQLite plots voltage/power curves, marks anomalies on the timeline and exports CSV for the report.

Feasibility Study

• Technical: every stage is individually testable with instruments the department owns (DMM, clamp meter, plug-in kWh meter). OpenEnergyMonitor documents the exact burden and bias design used. The firmware uses well-documented Arduino-core APIs only.
• Economic: the bill of materials is about ₹2,300 per node, one-time, with no subscription; the dashboard runs on an existing laptop.
• Operational: installation is a clamp-and-go job for a licensed electrician — no rewiring, no meter-seal disturbance. Wardens read a browser page; training takes minutes. The main risk (working near 230 V AC) is contained by design: the CT clamps outside the insulation and the low-voltage side is powered by an isolated phone charger.

In scope

• Single-phase 230 V AC feeders up to 20 A load, one metering node plus one dashboard.
• Non-invasive CT-clamp measurement; the prototype does not replace or open the utility meter.
• On-device anomaly rules (overload, tamper signature, off-hours surge) and interval logging to microSD + dashboard.
• Calibration against department instruments (DMM, clamp meter, plug-in kWh meter) with a published error table.

Out of scope (v1)

• Three-phase feeders, CTs above 20 A, and revenue-grade accuracy classes (IS/IEC 62053-21 type testing is a specialist lab activity).
• Billing, tariff computation and any integration with BESCOM or other discom systems — the prototype is evidence, not a billing device.
• Prepaid/vending mode and automatic disconnection via contactor.
• Battery powering of the node itself (it runs from an isolated 5 V phone charger).

The project follows the V-model, because every design stage on the left arm has a matching verification stage on the right arm — the discipline a measuring instrument demands.

Testing approach: unit testing with instrument comparisons at each measurement stage; integration tests for SD retry and batch upload; black-box rule tests where each anomaly is deliberately provoked (a heater above the limit, a load running while the voltage channel is unplugged) and the flag, timestamp and debounce behaviour are checked against a written test-case table (ID, input, expected, actual, result); and a two-week acceptance trial whose data feeds the load-profile analysis in Scilab.

SEM-TAL is a classic IoT edge node plus a thin server. The sensing front end converts the CT's secondary current to a measurable voltage with a burden resistor and biases it to mid-rail so the ADC sees the full AC waveform; the ZMPT101B module scales mains voltage down to a safe level for the second ADC channel. The ESP32 firmware samples both channels at roughly 2 kHz over a 100 ms window (an integer number of 50 Hz cycles), computes RMS current, RMS voltage and the mean of the instantaneous v·i product (real power), accumulates Wh, evaluates the three anomaly rules, and writes one record per minute.

Storage is deliberately layered: microSD is the source of truth, Wi-Fi upload is opportunistic, and the dashboard can import a CSV straight off the card, so a viva demo never depends on college Wi-Fi.

flowchart TD
  MAINS["230 V AC single-phase feeder"] -->|SCT-013 clamp, non-invasive| CT["SCT-013-000 current transformer"]
  CT --> COND["Burden + bias network (33 ohm burden, 2 x 470k bias)"]
  COND --> ADC1["ESP32 ADC1 GPIO34"]
  MAINS -->|ZMPT101B module| ADC2["ESP32 ADC1 GPIO35"]
  MCU["ESP32: C firmware (sampling, RMS, kWh, anomaly rules)"]
  ADC1 --> MCU
  ADC2 --> MCU
  MCU --> LCD["16x2 I2C LCD (V, I, W, kWh pages)"]
  MCU --> SD["microSD: 1-min interval records (SPI)"]
  MCU -->|Wi-Fi, HTTP POST batch| DASH["Flask + SQLite dashboard on laptop"]
  DASH --> CSV["CSV export + anomaly timeline"]
  MCU --> NTP["NTP time sync"]

The dashboard keeps three SQLite tables (devices, intervals, anomalies), renders the load curve with an open-source JS charting library, and exports the same CSV the report uses. Firmware and server communicate with a documented 12-field JSON batch format, so either side can be replaced independently — useful when the team splits work four ways.

• Prepaid-style display: map accumulated kWh to the local tariff and show rupees on the LCD, with a UPI-based top-up alert as an extension.
• MQTT + an open-source IoT platform (e.g., ThingsBoard) instead of direct HTTP, for multi-building deployments.
• Three-phase and CT-ratio scaling for a full building panel, with per-phase anomaly rules.
• Better anomaly detection: replace fixed thresholds with rolling-baseline statistics (per-hour mean ± k·σ) and simple pattern classification trained on the team's own logged data.
• LoRa backhaul for feeders outside Wi-Fi coverage.
• DISCOM-style interval reports — 96-slot daily profiles — as a teaching artefact showing how utilities study loss pockets.