Smart Aquarium IoT System

Updated 20 January 2026

IoT-based smart aquarium systems are integrated, sensor-driven cyber-physical infrastructures that use embedded controllers, wireless sensor networks, and ML algorithms to monitor aquatic environments.
They employ a modular architecture with distinct sensor, edge, and cloud layers to ensure real-time data acquisition, preprocessing, and autonomous actuation using wired and wireless communication protocols.
The integration of machine learning models and rule-based controllers facilitates real-time anomaly detection, precise control of water parameters, and improved operational reliability in diverse aquarium settings.

An IoT-based smart aquarium system is an integrated, sensor-driven cyber-physical infrastructure that utilizes embedded microcontrollers, wired and wireless sensor networks, real-time software pipelines, and machine learning algorithms to autonomously monitor and regulate water quality, fish welfare, and operational efficiency in aquarium or small aquaculture environments. Such systems encompass diverse elements—multivariate sensing, actuator control, edge/cloud computation, anomaly detection, decision support, and user interaction—that collectively enable closed-loop, data-driven aquatic management.

1. System Architecture and Core Hardware Components

A canonical IoT-based smart aquarium is structured around a modular architecture with three primary tiers: sensor layer, edge (processing and actuation) layer, and cloud or application layer (Ayon et al., 13 Jan 2026, Dhinakaran et al., 2023, Rashid et al., 2022, Mozumder et al., 2021).

Sensor Layer:

Key environmental variables are monitored with the following sensor modalities:

Water Parameters: pH electrodes (0–14 range; e.g., SEN0161 or Atlas EZO-pH), digital temperature probes (DS18B20; −55 °C to 125 °C), dissolved oxygen (DO) electrodes (galvanic or optical, e.g., 0–20 mg/L), turbidity sensors (SEN0189; 0–1000 NTU), total dissolved solids (TDS, e.g., SEN0244; 0–1000 ppm), ammonia sensors (e.g., MQ-135 or Vernier ISE), and water level via ultrasonic (HC-SR04).
Atmospheric Parameters: DHT11 or equivalent for air temperature/humidity.
Behavioral/Visual Sensors (where applicable): Cameras for fish behavior, especially in research or advanced feeding setups.

Edge/Processing and Actuation Layer:

Embedded Microcontroller: ESP32, Arduino Uno/MKR/NodeMCU, STM32 (e.g., F103C8 for higher concurrency or camera processing).
Relay Modules and Power Staging: For switching heaters, pumps, feeders, aerators, dosing systems; MOSFET or SSR for DC/AC loads.
Local Display/Interface: LCD via I²C for status, or local webserver for direct browser access.
Onboard computation: Edge pre-processing, anomaly detection, actuation, optional TinyML inference engines.

Network and Communication Layer:

Local Wired: UART/I²C/SPI between sensors/MCU, USB between Arduino & Raspberry Pi.
Wireless: Wi-Fi, ZigBee, LoRaWAN, or BLE for node-network/cloud uplink.
Protocols: Time-stamped JSON/MQTT/HTTP packets, typically at fixed intervals (e.g., every 5 s–1 min).

Cloud/Application Layer:

IoT Platforms: Blynk, Firebase, ThingsBoard, InfluxDB for storage/logging, MQTT brokers for message transport.
User Interfaces: Mobile apps (Android/iOS), dashboards (Grafana, custom), SMS/email/FCM alerting.

The table below summarizes common sensor/actuator mapping for a typical home/small-scale system (Ayon et al., 13 Jan 2026, Islam, 11 Jan 2025, Dhinakaran et al., 2023):

Component	Example Model	Variable(s)
pH Sensor	SEN0161/EZO-pH	pH
Temp Sensor	DS18B20	°C
DO/Turbidity	EZO-DO/SEN0189	mg/L, NTU
TDS/Conductivity	SEN0244/DFROBOT	ppm, µS/cm
Actuator	Relay/MOSFET	Heater/pump/feeder

2. Sensor Data Acquisition and Preprocessing Pipeline

Raw sensor readings are subject to a multi-stage preprocessing pipeline designed to increase data quality and optimize downstream analytics (Ayon et al., 13 Jan 2026, Dhinakaran et al., 2023, Rashid et al., 2022, Hsain et al., 3 Jan 2026, Mozumder et al., 2021):

Acquisition: Synchronized multi-sensor sampling, with time-stamping and local buffering.
Imputation: Statistical estimation (e.g., mean/last-value) to fill missing readings, with sensor fault detection and flagging.
Filtering: Median and/or moving average filters (e.g., N=5) to remove transients/bursts; z-score thresholding (e.g., |z| > 3) for outlier rejection.
Calibration: Linear regression or look-up conversion of analog voltages to physical units (e.g., pH = aV + b).
Normalization: Min-max scaling to [0,1] range using feature-specific safe bounds.
Feature Engineering: Rolling averages/rates (e.g., hourly mean, rate-of-pH change) and composite indices (e.g., temperature × pH).
Data Splitting: Partitioning into training, validation, and testing sets for ML development/evaluation phases.

A typical workflow, as pseudocode (Ayon et al., 13 Jan 2026):

loop every 5 seconds:
    read raw_sensors[]
    for each sensor i:
        calibrated[i] = lin_regression(i, raw_sensors[i])
        smoothed[i] = moving_average(i, calibrated[i])
        if smoothed[i] < L[i] or smoothed[i] > U[i]:
            if now() − lastAlert[i] > cooldown:
                sendAlert(i, smoothed[i])
                lastAlert[i] = now()
    updateLocalLCD(smoothed[])
    publishToBlynk(smoothed[])
    handleFeedPumpCommands()

3. Machine Learning and Rule-Based Decision Modules

Intelligent control in smart aquaria leverages both classical ML and lightweight (TinyML) neural models for real-time water quality prediction, anomaly detection, and actuator control (Dhinakaran et al., 2023, Rashid et al., 2022, Mozumder et al., 2021, Hsain et al., 3 Jan 2026, Islam, 11 Jan 2025).

ML Model Architectures (as provided):

Random Forests (RF):
- Used for predicting and optimizing water temperature and pH setpoints, ensemble of decision trees trained on historical sensor data.
Support Vector Machines (SVM):
- Early warning for disease/parasite detection using behavioral and environmental features.
Gradient Boosting Machines (GBM):
- Dynamic optimization of feeding schedule based on environmental and behavioral data.
Neural Networks (NN):
- Feed-forward architectures (e.g., 5 hidden layers, ReLU activations, Adam optimizer) for direct mapping of sensor vector to actuator control (e.g., on/off decisions for pumps, dosing, feeding).
- TinyML CNN for resource-constrained edge deployment: e.g., Conv1D (16/32 filters), pooling, Dense(64)→Dense(6) (for 6-parameter forecasting), 8-bit quantization, ∼24 KB flash, inference ∼50 ms on Arduino (Hsain et al., 3 Jan 2026).

Rule-Based/Hybrid Logic:

Thresholding: Each variable R has bounds (L, U); anomaly if R<L or R>U.
PID/Fuzzy Controllers:
- PID form:
$u(t) = K_p e(t) + K_i \int_{0}^{t} e(\tau)d\tau + K_d \frac{de}{dt}$

with $e(t) = x_0 - x(t)$ (setpoint tracking), for closed-loop control of temperature (heater/cooler), DO (aerator), turbidity (filter pump), pH (acid/base dosing) (Han et al., 2018, Hossam et al., 2024). - Optional fuzzy layer for nonlinear dynamics and disturbance rejection (7-level membership, centroid defuzzification).
Decision Trees/Random Forests: Tree models trained on multi-parametric labeled datasets for water condition (“good” vs. “bad”) or direct fish-species suitability, e.g., Gini/entropy splitting, no post-pruning, ∼79–94% accuracy observed on test sets (Mozumder et al., 2021, Islam, 11 Jan 2025).

4. Control, Alerting, and Actuation Logic

Actuator Control:

Automated switching of relays/MOSFETs for heaters, pumps, filters, aerators, feeders based on decision module outputs.
Servo-based feeders (e.g., SG90) with quantized dispensation cycles (e.g., one full sweep ≈ 0.5 g).

Alert and Notification Systems:

Intelligent cooldown logic for alert suppression (default interval e.g., 600 s), to mitigate notification fatigue during sustained or transient excursions.
Alerts delivered via SMS, email, mobile push (FCM, Blynk, Twilio, etc.), and local HMI (LED or LCD) indicators.
Manual override interfaces (mobile app, desktop dashboard, local tactile controls).

User Interfaces:

Mobile apps (typically Android/iOS): Real-time dashboard, control toggles, push notifications, and historical trends.
Desktop/web dashboards (e.g., Grafana, WPF/.NET): Multi-pane real-time status, visual charts, setpoint configuration, manual/automatic control toggles, CSV export.

5. Experimental Validation and Performance Metrics

Multiple research prototypes have demonstrated system reliability and performance in practical settings (Ayon et al., 13 Jan 2026, Mozumder et al., 2021, Rashid et al., 2022, Hsain et al., 3 Jan 2026):

Sensing Fidelity: Sensor accuracy typically above 90% (e.g., pH accuracy 91%, temp 96.5%, turbidity 92.5% over 200 trials (Ayon et al., 13 Jan 2026)).
System Responsiveness: Response/latency from anomaly to actuation typically sub-2 s; end-to-end cloud latency in the 1–5 s range.
Operational Reliability: Actuator success rates (servo feeders, pumps) above 97%.
ML Metrics: ML-based DO class prediction at 77% accuracy, random-forest water condition prediction at 79–94%, TinyML MAE ≈ 0.15 (pH), 0.3 °C (temp), >90% anomaly detection precision/recall observed (Rashid et al., 2022, Mozumder et al., 2021, Hsain et al., 3 Jan 2026).
Power and Scalability: Embedded edge platforms consume ∼150 mW on average; battery/supply options include LiPo+solar for off-grid; multi-tank extension by daisy-chaining nodes.
Maintenance and Safety: Periodic calibration of pH, DO, and TDS sensors required; use of waterproof (IP65/IP67) enclosures and low-voltage actuators recommended.

6. Adaptation and Scalability in Diverse Aquatic Contexts

Commercial-scale fish farms, research testbeds, and home aquaria each require system tailoring (Dhinakaran et al., 2023, Ayon et al., 13 Jan 2026, Mozumder et al., 2021, Islam, 11 Jan 2025, Hossam et al., 2024):

Hardware Scaling: High-density fish farms may use more robust sensors, distributed wireless sensor networks (star or mesh), and frequent multi-point sampling. Home/small-scale systems can employ compact (DS18B20, micro-pH, TDS) sensors and simpler Wi-Fi/BLE connectivity.
ML/Computation Scaling: Commercial deployments offload heavier ML to the cloud/edge server; home systems may deploy reduced NN models or rule-based logic on the MCU (e.g., TinyML on ESP32).
Control/Interface:
- Aquarium adaptation: Replace industrial actuators by miniature pumps, submersible heaters, auto-feeders.
- User interface: App-based live readout, manual override, push notification when parameters breach preset thresholds.
Network Topology:
- WSN (ZigBee, LoRa) favored for distributed or outdoor large-scale aquaculture.
- Wi-Fi/BLE preferred for home aquaria due to coverage, UI compatibility, and ease of integration.

Adaptation options may include additional measures for edge inference (TinyML), redundant power (UPS/battery/solar), periodic cloud fallback to local-only control, and simplified control logic suitable for non-farm settings.

7. Research Challenges, Open Issues, and Future Directions

Data Quality and Robustness: Sensor drift, noise, and missing data remain persistent issues, as addressed with multimodal filtering, imputation, and outlier handling (Rashid et al., 2022, Hsain et al., 3 Jan 2026).
Scalability and Interoperability: Modular, loosely coupled design (service-oriented API/MQTT/REST) supports integration of additional tanks or cross-vendor devices (Teixeira et al., 2021).
Edge Intelligence and Energy Optimization: TinyML on edge nodes demonstrates significant gains in anomaly prediction latency/power, but model compression/pruning is mandatory to fit low-SRAM MCUs (Hsain et al., 3 Jan 2026).
Control Algorithm Expansion: Movement from simple thresholding to PID/fuzzy or model-predictive control frameworks increases stability and adaptive performance (Han et al., 2018, Hossam et al., 2024).
Ethical and Security Considerations: Privacy (environmental only), encrypted transports, access controls, and safety (fail-safes, manual override) are standard practices (Hsain et al., 3 Jan 2026).
Extension to Precision Feeding and Computer Vision: Integration of visual analytics (YOLOv8-based keypoint + depth estimation for fish size/biomass, feed control) expands application domains into precision aquaculture (Hossam et al., 2024).
Experimental Gaps: Most studies report proof-of-concept or pilot-scale deployments; comprehensive longitudinal performance analyses and greater dataset accumulation for ML generalizability are needed.

References:

(Ayon et al., 13 Jan 2026) "An IoT-Enabled Smart Aquarium System for Real-Time Water Quality Monitoring and Automated Feeding"
(Dhinakaran et al., 2023) "IoT-Based Environmental Control System for Fish Farms with Sensor Integration and Machine Learning Decision Support"
(Rashid et al., 2022) "IoT based Smart Water Quality Prediction for Biofloc Aquaculture"
(Mozumder et al., 2021) "Smart IoT-Biofloc water management system using Decision regression tree"
(Islam, 11 Jan 2025) "Prediction Model of Aqua Fisheries Using IoT Devices"
(Han et al., 2018) "An intelligent household greenhouse system design based on Internet of Things"
(Hsain et al., 3 Jan 2026) "Tiny Machine Learning for Real-Time Aquaculture Monitoring: A Case Study in Morocco"
(Hossam et al., 2024) "Precision Aquaculture: An Integrated Computer Vision and IoT Approach for Optimized Tilapia Feeding"
(Teixeira et al., 2021) "Towards Precision Aquaculture: A High Performance, Cost-effective IoT approach"
(Agrawal et al., 2023) "Cyber Physical Aquaponic System (CyPhA): a CPS Testbed"