Uncovering Residential PV-EV Co-Adoption from Smart-Meter Data: Load Archetypes and Detection for Demand-Side Planning

Jack Zheng; Hao Wang

cs.LGSep 23, 2026

Uncovering Residential PV-EV Co-Adoption from Smart-Meter Data: Load Archetypes and Detection for Demand-Side Planning

Authors: Jack Zheng, Hao Wang

Organizations: Department of Data Science and AI, Faculty of Information Technology, Monash University, Melbourne, Victoria, Australia · Monash Energy Institute, Monash University, Melbourne, Victoria, Australia

Abstract

The increasing adoption of electric vehicles (EVs) and rooftop photovoltaic (PV) systems is reshaping residential electricity demand and creating new challenges for demand-side management (DSM), tariff design, and low-voltage network planning. Much of the existing literature examines EV charging or PV generation in isolation, leaving the behavioral dynamics of household co-adoption less understood. We develop an integrated, two-part workflow to analyze advanced metering infrastructure (AMI) data. A discovery component applies dynamic time warping (DTW) k-means with DTW barycenter averaging to cluster daily import or export profiles into interpretable behavioral archetypes, while a predictive component trains a bidirectional long short-term memory (BiLSTM) model on 21-day windows and benchmarks it against tabular baselines for PV/EV activity detection. The EV activity labels are inferred from charging-like load signatures because charger measurements are unavailable. Using half-hourly AusNet residential data from Victoria, Australia, the clustering uncovers distinct patterns across PV-only, EV-only, co-adoption, and neither cohorts; for co-adopters, a midday-centered weekday export archetype accounts for approximately 50% of days. At validation-tuned thresholds, both BiLSTM and XGBoost achieve strong discrimination. BiLSTM obtains 0.991 for the area under the receiver operating characteristic curve (AUROC), 0.906 for macro-F1, and the highest recall on the most difficult class (0.836 for EV-only recall). Tree-based baselines remain competitive. Performance remains stable across plausible labeling rules (macro-F1: 0.894--0.914) and strictly forward temporal splits (macro-F1: 0.894--0.906).

Figures & tables

Research theme	Representative studies	Focus and relevance to this study
Residential load-shape clustering	[ 33 , 11 , 43 ]	Shape- and feature-based segmentation of daily profiles, including EV-user clustering, supporting the descriptive archetype analysis.
Behind-the-meter PV inference	[ 22 , 24 , 23 ]	Estimate PV size and orientation and disaggregate PV and other DERs from net load, providing baselines for the PV side of detection.
EV detection from AMI	[ 21 , 39 , 17 , 25 ]	Supervised, event-based, and training-free EV detection for demand-response targeting, providing baselines for the EV side of detection.
EV charging behavior and flexibility	[ 26 , 32 , 44 , 29 ]	Characterize charging timing, flexibility, and distribution-network impacts that inform the activity-labeling rules.
PV–EV co-adoption	[ 20 , 34 , 31 , 3 ]	Document behavioral change, incentives, and barriers under joint ownership, providing the empirical basis for interpreting co-adoption archetypes.
DSM, time-of-use tariffs, and managed charging	[ 8 , 6 , 14 , 27 , 19 , 41 , 5 ]	Quantify peak-shifting benefits, solar-aligned charging, and synchronization risks that motivate cohort-aware targeting.

Table 1: Summary of the reviewed literature by research theme, with representative studies and their relation to the present work.

Figure 1: Standardized DTW

k

-means daily-profile archetypes across the four adoption cohorts. Rows identify cohort, clustering channel, and day type; columns identify the panel-local clusters 1–5. Orange lines show export centroids for the PV-associated cohorts, and blue lines show import centroids for the EV-only and neither-adopter cohorts. Shading shows the pointwise interquartile range of the standardized profiles assigned to each cluster.

	Cluster sizes
	1	2	3	4	5	Total
PV-only weekday	92	504	289	108	274	1,267
PV-only weekend	117	107	51	29	244	548
EV-only weekday	451	479	241	196	299	1,666
EV-only weekend	120	160	160	108	165	713
Co-adoption weekday	563	1,654	254	524	337	3,332

Table 2: Cluster sizes for behavioral archetypes (

k=5

for all groups).

Model	Accuracy	Precision	Recall	F1	AUROC
BiLSTM	0.926	$0.907$	$0.906$	$0.906$	0.991
XGBoost	$0.926$	0.908	0.908	0.908	$0.990$
Random forest	$0.910$	$0.895$	$0.890$	$0.892$	$0.986$
Logistic regression	$0.903$	$0.880$	$0.881$	$0.880$	$0.980$

Table 3: Comparison of model performance (mean over three runs).

Model	Class	Precision	Recall	F1 score	Sample size
BiLSTM sequence	PV-only	$0.981$	$0.965$	$0.973$	4,359
	EV-only	$0.828$	$0.836$	$0.832$	1,539
	Co-adoption	$0.920$	$0.958$	$0.939$	1,897
	Neither	$0.871$	$0.862$	$0.866$	1,935
XGBoost	PV-only	$0.969$	$0.970$	$0.970$	4,359
	EV-only	$0.861$	$0.820$	$0.840$	1,539

Table 4: Comparison of per-class classification metrics across models.

Setting (label rule)	F1	AUROC	Accuracy	PV threshold	EV threshold
Baseline	$0.904$	$0.991$	$0.924$	$0.30$	$0.70$
EV window: $\geq 4$ active days	$0.914$	$0.992$	$0.934$	$0.30$	$0.65$
Higher contrast gate	$0.911$	$0.992$	$0.931$	$0.30$	$0.55$
Looser energy floor (35th percentile)	$0.909$	$0.991$	$0.927$	$0.35$	$0.55$
Other variants (range)	0.894–0.905	0.988–0.991	0.915–0.925	0.30–0.70	0.55–0.70

Table 5: BiLSTM performance under alternative label-generation rules (same test split). Note that other variants include alternative EV-active-day thresholds, energy floors, and contrast gates; full definitions are provided in the text.

Training/ validation cutoff	F1	AUROC	Accuracy	PV threshold	EV threshold
31 Mar / 30 Jun	$0.898$	$0.990$	$0.920$	$0.30$	$0.50$
30 Jun / 30 Sep	$0.894$	$0.989$	$0.914$	$0.30$	$0.60$
30 Sep / 31 Dec	$0.906$	$0.991$	$0.930$	$0.30$	$0.50$

Table 6: BiLSTM model performance across temporal data splits.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Cohort	Cluster	Daily import (kWh)	Peak import (kW)	Evening import (kWh)	Evening peak (kW)	Midday import (kWh)	Midday export (kWh)	Peak-to- average ratio
PV-only	1	15.87	2.75	6.02	2.01	2.89	1.23	5.11
PV-only	2	9.40	1.97	4.16	1.54	0.26	11.21	5.73
PV-only	3	12.41	2.33	4.33	1.65	1.38	4.95	5.61
PV-only	4	15.89	2.65	4.96	1.67	2.84	0.64	5.15
PV-only	5	12.10	2.66	4.86	1.87	1.20	5.74	5.69
EV-only	1	18.90	3.31	8.05	2.91	3.33	–	4.45

Appendix

Table 7: Weekday raw-energy means by daily archetype. A dash denotes a cohort without an export stream.

Cohort	Cluster	Daily import (kWh)	Peak import (kW)	Evening import (kWh)	Evening peak (kW)	Midday import (kWh)	Midday export (kWh)	Peak-to- average ratio
PV-only	1	12.25	2.54	4.81	1.73	1.50	5.72	5.41
PV-only	2	13.90	2.60	5.01	1.70	1.48	6.12	6.04
PV-only	3	24.87	3.93	7.80	2.63	5.77	1.55	3.91
PV-only	4	15.97	2.75	4.86	1.46	2.84	0.77	6.67
PV-only	5	8.15	1.65	3.67	1.33	0.23	12.35	5.44
EV-only	1	16.26	2.62	4.64	1.57	3.64	–	4.50

Appendix

Table 8: Weekend raw-energy means by daily archetype. A dash denotes a cohort without an export stream.

Explore similar work

CROCS: A Two-Stage Clustering Framework for Behaviour-Centric Consumer Segmentation with Smart Meter Data
Jan 15, 2026Luke W. Yerbury, Ricardo J. G. B. Campello, G. C. Livingston +2Demand Response Consumption
A Grid-Aware Agent-Based Model for Analyzing Electric Vehicle Charging Systems
Apr 30, 2026Khalil Al-Rahman Youssefi, Marija Gojkovic, Walter Stefanutti +2Electric Vehicles Power Systems
Lakeside Labs, Lakeside Park B04b, Klagenfurt, Austria · Alpen-Adria-Universität, Universitätsstraße 65–67, Klagenfurt, Austria · Silicon Austria Labs, Power Electronics division, High Tech Campus Villach - Europastraße 12, Austria
Federated Learning for Early Prediction of EV Charging Demand
May 6, 2026Vasilis Perifanis, Foteini Nikolaidou, Nikolaos Pavlidis +2Electric Vehicles Federated Learning
Indigma Innovations

Jan 15, 2026, stat.MLJ/K move · Enter open · S save

CROCS: A Two-Stage Clustering Framework for Behaviour-Centric Consumer Segmentation with Smart Meter Data

Luke W. Yerbury, Ricardo J. G. B. Campello, G. C. Livingston, Mark Goldsworthy+1

Demand Response Consumption Clustering Graph Representations

Open paper AskPDF arXiv

With grid operators confronting rising uncertainty from renewable integration and a broader push toward electrification, Demand-Side Management (DSM) -- particularly Demand Response (DR) -- has attracted significant attention as a cost-effective mechanism for balancing modern electricity systems. Unprecedented volumes of consumption data from a continuing global deployment of smart meters enable consumer segmentation based on real usage behaviours, promising to inform the design of more effective DSM and DR programs. However, existing clustering-based segmentation methods insufficiently reflect the behavioural diversity of consumers, often relying on rigid temporal alignment, and faltering in the presence of anomalies, missing data, or large-scale deployments. To address these challenges, we propose a novel two-stage clustering framework -- Clustered Representations Optimising Consumer Segmentation (CROCS). In the first stage, each consumer's daily load profiles are clustered independently to form a Representative Load Set (RLS), providing a compact summary of their typical diurnal consumption behaviours. In the second stage, consumers are clustered using the Weighted Sum of Minimum Distances (WSMD), a novel set-to-set measure that compares RLSs by accounting for both the prevalence and similarity of those behaviours. Finally, community detection on the WSMD-induced graph reveals higher-order prototypes that embody the shared diurnal behaviours defining consumer groups, enhancing the interpretability of the resulting clusters. Extensive experiments on both synthetic and real Australian smart meter datasets demonstrate that CROCS captures intra-consumer variability, uncovers both synchronous and asynchronous behavioural similarities, and remains robust to anomalies and missing data, while scaling efficiently through natural parallelisation. These results...

Abstract

Explore similar work

CROCS: A Two-Stage Clustering Framework for Behaviour-Centric Consumer Segmentation with Smart Meter Data

A Grid-Aware Agent-Based Model for Analyzing Electric Vehicle Charging Systems

Federated Learning for Early Prediction of EV Charging Demand