CRS4 TECHNOLOGY CATALOGUE

MICE

Smart Energy Consumption Monitoring

A scalable solution for analyzing residential energy consumption

Background

The growing penetration of intermittent renewable energy sources makes active management of residential electricity demand essential.

Flexible loads—such as programmable appliances and heating and cooling systems—represent a strategic resource for grid balance, but identifying them and quantifying their flexibility typically requires dedicated sensors or costly sub-metering.

Second-generation smart meters, now widely deployed, offer an alternative: the aggregated consumption data they measure contains sufficient information to estimate the contribution of individual flexible loads, without the need for additional infrastructure.

However, there is currently no scalable solution that leverages this data in a non-intrusive manner, is adaptable to diverse consumer profiles, and operates with minimal user interaction.

What does MICE offer?

The application is an easily scalable solution for analyzing residential energy consumption, which transforms existing smart meter data into high-value insights for more efficient and sustainable energy demand management.

Thanks to a non-invasive approach based on artificial intelligence, it is able to identify and quantify flexible loads without the need for new infrastructure and with minimal user interaction.

Specifically, the system analyzes aggregated consumption data from smart meters to identify and quantify residential flexible loads, following two complementary approaches:

for programmable appliances (washing machines, dishwashers), it uses convolutional neural networks that are pre-trained and fine-tuned through minimal user interaction: just a few activation labels—obtainable via a simple app—are sufficient to adapt the model to the specific user. The required data has minute-by-minute resolution, compatible with second-generation meters.
for thermal loads (heat pumps), it adopts a fully unsupervised approach: a dual-head neural network with causal feature separation generates probability distributions for thermal consumption, refined using unlabeled aggregated data from the target user base. The calibrated probabilistic outputs support automated load control systems with explicit comfort level management.

Innovative features

Estimation of residential demand flexibility using only data from smart meters, without additional sensors or sub-metering
For heating loads: a fully unsupervised approach, with no labels at any level—neither during training nor during adaptation to a specific user—with probabilistic outputs calibrated for comfort risk management
Causal separation of features (temporal → base load; meteorological → thermal load) that improves signal identifiability in the absence of supervision
For programmable appliances: adaptation to new users with minimal user interaction (3–7 labeled activations), without sub-metering and using data at one-minute resolution
Competitive performance compared to state-of-the-art supervised models in cross-domain transfer scenarios

Potential users

Energy companies and utilities (demand response programs, dynamic pricing, customer profiling
Flexibility aggregators and dispatch services market operators
Providers of smart home platforms and building management systems
Energy service companies — ESCOs (data-driven energy optimization)
Residential consumers and prosumers (visibility and control over their flexible consumption)

Sectors impacted

Energy markets and dispatch services (estimation and aggregation of residential flexibility)
Residential energy management (granular monitoring and optimization of flexible loads)
Smart grids and demand response (identification and activation of distributed flexibility resources)
Sustainability and decarbonization (supporting the integration of renewables through active demand management)
IoT and home automation (analytical layer for consumption-aware automation systems)

Economic and social value

Implementation of the demand-response system → monetization of flexibility
Zero-label / zero-hardware → large-scale scalability
Use of existing infrastructure → high return on investment
Reduction in grid costs → added value for utilities
Environmental sustainability

Additional resources and information

Luca Massidda, Marino Marrocu, Simone Manca. Non-Intrusive Load Disaggregation by Convolutional Neural Network and Multilabel Classification. Applied Sciences, 10(4), 1454, 2020. https://doi.org/10.3390/app10041454
Marco Manolo Manca, Luca Massidda. Deep Learning Based Non-Intrusive Load Monitoring with Low Resolution Data from Smart Meters. Communications in Applied and Industrial Mathematics, 13(1), 39–56, 2022. https://doi.org/10.2478/caim-2022-0004
Luca Massidda, Marino Marrocu. A Bayesian Approach to Unsupervised, Non-Intrusive Load Disaggregation. Sensors, 22(12), 4481, 2022. https://doi.org/10.3390/s22124481
Luca Massidda, Marino Marrocu. Total and Thermal Load Forecasting in Residential Communities through Probabilistic Methods and Causal Machine Learning. Applied Energy, 351, 121783, 2023. https://doi.org/10.1016/j.apenergy.2023.121783
Gabriella Pusceddu, Simone Manca, Luca Massidda. Fine-Tuning Non-Intrusive Load Monitoring Model Through User Interaction: A Practical Approach to Appliance Recognition with Limited Labeled Data. Applied Energy, 391, 125943, 2025. https://doi.org/10.1016/j.apenergy.2025.125943
Luca Massidda, Marino Marrocu. Beyond Labels: Bayesian and Causal Heat Pump Disaggregation from Unlabeled Smart Meter Data. Applied Energy, 412, 127684, 2026. https://doi.org/10.1016/j.apenergy.2026.127684

Status

Type of innovation:
incremental

Product type:
software

Emerging technologies adopted:
Artificial Intelligence (AI) / Machine Learning; Transfer Learning; Smart meter analytics; Bayesian inference; High Performance Computing.

Stage of technological
development:
TRL 4-5

Limitations:

Appliances: Requires data at one-minute intervals, which can be obtained using a powerline communication analyzer for smart meters or a simple pulse counter for all modern meters;
Heating loads: Validated for climates where heating is the dominant factor; transferability to climates where cooling is the dominant factor has not yet been verified; performance degrades when the heating component of the aggregate load is low.

Intellectual property characteristics:
proprietary code

Other partners:
No

Keywords

NILM
Energy disaggregation
Deep learnin
Smart meters
Transfer learning
Energy analytics platforms
Predictive maintenance systems

Contacts

Luca Massidda
– industrial collaborations
– pilot projects and demonstrators
– technology transfer (licensing)

For information:
valorisation@crs4.it

Date

Last update:
25/03/2026