A novel approach for identifying customer groups for personalized demand-side management services using household socio-demographic data

Hanguan Wen, Xiufeng Liu, Ming Yang, Bo Lei, Xu Cheng, Zhe ChenEnergy 2023 · DOI

TL;DR

This paper introduces a novel data-driven approach to personalize Demand-Side Management (DSM) services by linking household socio-demographic data with energy consumption patterns. It leverages a robust clustering algorithm and feature selection techniques to predict DSM group membership, even when smart meter data is unavailable, offering a more targeted and efficient approach to energy management.

What Problem Does This Paper Address?

The core challenge addressed is the difficulty in providing personalized Demand-Side Management (DSM) services, especially for new customers or in situations where smart meter data is scarce or privacy-restricted. Traditional DSM relies heavily on detailed energy consumption data, which isn't always accessible. This paper aims to bridge this gap by demonstrating how socio-demographic information, combined with a novel analytical framework, can effectively predict a household's likely DSM group and thus enable tailored energy efficiency and demand response programs from the outset.

Methods

The authors propose a four-step methodology:

  1. Data Preparation: Raw energy consumption and socio-demographic data are collected and preprocessed. Weekly consumption patterns are used to mitigate daily variations.
  2. Clustering Consumption Patterns: A Robust-learning Fuzzy C-Means (RL-FCM) algorithm is employed. This is a key innovation as it automatically determines the optimal number of clusters (customer groups) without manual intervention, making it more robust to outliers and irregular data than traditional Fuzzy C-Means.
  3. Feature Selection: For each identified cluster, symmetric uncertainty is used to measure the correlation between socio-demographic features and the consumption patterns. This is followed by Pareto analysis to select the most relevant features for each cluster, ensuring that only the most impactful socio-demographic indicators are used.
  4. Prediction Model: A Long Short-Term Memory (LSTM) neural network is trained for each cluster. These models use the selected socio-demographic features as input to predict the probability of a household belonging to a specific DSM group.

Datasets

The study utilizes a dataset from the Commission for Energy Regulation (CER) in Ireland. This dataset comprises:

  • Energy Consumption Data: 536 semi-hourly daily consumption records for 4232 residential households. This was processed into 50 weeks of weekly data for 1000 households.
  • Socio-demographic Data: Information gathered from pre- and post-trial surveys for these households, covering 12 key features.

The paper does not introduce new benchmark datasets but uses this real-world dataset for evaluation.

Key Insights and Ideas

  • Socio-demographics as a powerful proxy: The research strongly suggests that socio-demographic information, when analyzed through a robust framework, can be a highly effective predictor of energy consumption behavior, even in the absence of detailed smart meter data.
  • RL-FCM's advantage: The RL-FCM algorithm's ability to automatically determine the optimal number of clusters is a significant practical advantage, reducing the complexity and subjectivity often associated with clustering.
  • Feature importance varies by cluster: The study highlights that the most influential socio-demographic features for predicting consumption patterns are not universal. For instance, "number of adult residents" (F5) was a top predictor for most clusters, but "number of bedrooms" (F10) was key for others, underscoring the need for cluster-specific feature selection.
  • Personalized DSM is feasible: The framework demonstrates a clear path towards highly personalized DSM strategies, moving beyond generic recommendations to tailored interventions based on predicted behavior.

Key Results

  • The proposed model achieved a substantial reduction in average Root Mean Square Error (RMSE) compared to three baseline models: 52.53% improvement over Baseline 1 (LSTM with full features), 85.59% over Baseline 2 (ELM with selected features), and 86.08% over Baseline 3 (ELM with full features).
  • The model demonstrated strong prediction accuracy across eight identified customer clusters, with the red dashed line (proposed model) in radar plots closely matching the ground truth for four test households.
  • The feature selection process identified key socio-demographic indicators for each cluster, with "number of adult residents" (F5) and "number of bedrooms" (F10) frequently appearing as the most influential.

Limitations and Future Work

  • Data availability and quality: The authors acknowledge that obtaining and maintaining accurate socio-demographic data can be challenging. Future work could explore alternative data collection methods.
  • Stability of characteristics: The assumption that socio-demographic characteristics remain stable over time might not always hold true. More robust methods are needed to handle dynamic changes.
  • Feedback effects of DSM: The current model does not explicitly account for how DSM interventions might influence household behavior and consumption patterns. Incorporating feedback mechanisms is a suggested area for future research.
  • Hypothetical queries: The reliance on hypothetical queries based on socio-demographic data might not always reflect actual energy usage patterns. However, the authors argue the approach remains valuable due to its focus on relative feature importance and its adaptability to new data.
  • Future work directions: The authors suggest exploring seasonal fluctuations, holiday patterns, tiered electricity tariffs, and privacy-preserving techniques like federated learning. They also emphasize the need to consider technical and environmental constraints that influence energy consumption.

Summary updated

Abstract

Demand-side management (DSM) is crucial to smart energy systems. This paper presents a data-driven approach for understanding the relationship between energy consumption patterns and household characteristics to better provide DSM services. The proposed method uses a robust learning fuzzy c-Means clustering algorithm to automatically generate the optimal number of customer groups for DSM, and then uses symmetric uncertainty techniques to identify the identified load patterns and socio-demographic characteristics as the features for training a deep learning model. The model obtained can be used to predict the possibility of DSM group membership for a given household. This approach can be applied even in situations where smart meter data are not available, such as when new customers are added to the system or when residents change, or due to privacy concerns. The proposed model is evaluated comprehensively, including prediction accuracy, comparison with other baselines, and case studies for DSM. The results demonstrate the usefulness of weekly energy consumption data and associated household socio-demographic information for distinguishing between different consumer groups, the effectiveness of the proposed model, and the potential for targeted DSM strategies such as time-of-use pricing, energy efficiency measures, and demand response programs.
BibTeX
@inproceedings{wen2023novel,
  author = {Hanguan Wen and Xiufeng Liu and Ming Yang and Bo Lei and Xu Cheng and Zhe Chen},
  title = {{A novel approach for identifying customer groups for personalized demand-side management services using household socio-demographic data}},
  booktitle = {Energy},
  year = {2023},
  doi = {10.1016/j.energy.2023.129593},
}
Figure