PaperlayerView exact revision on arXiv

Source-linked AI summary

A survey of recommender systems for energy efficiency in buildings: Principles, challenges and prospects

Yassine Himeur, Abdullah Alsalemi, Ayman Al-Kababji, Faycal Bensaali, Abbes Amira, Christos Sardianos, George Dimitrakopoulos, Iraklis Varlamis

arXiv:2102.07654v1cs.IR

TL;DR

Rising energy use in buildings and the need to improve energy-saving behavior motivate recommender systems, but their deployment still faces unresolved challenges. This paper reviews and taxonomizes building energy-efficiency recommender systems and critically analyzes their limitations. It traces their evolution from energy-data engines through information-fusion and deep-learning systems to systems adding explanations, visualizations, and time-aware processing.

  • Problem

    The paper identifies open questions about the reliability and efficacy of building energy-saving recommender systems and a lack of review coverage of their challenges and future perspectives.

    Source: §1 · p. 2

  • Method

    The paper critically reviews building energy-efficiency recommender systems, taxonomizes them, and identifies limitations and future directions.

    Source: §6 · p. 23

  • Results

    The review traces three generations, from energy-data-based engines to information-fusion and deep-learning systems, followed by systems adding explanations, visualizations, and time-aware processing.

    Source: §6 · pp. 23–24

  • Takeaways & Limitations

    The paper describes edge computing and edge AI as making real-time recommendations feasible and reports improved recommendation quality, acceptance, and end-user satisfaction.

    Source: §6 · p. 24

    Evaluation faces a lack of unified benchmarking standards and comprehensive datasets.

    Source: §4.2.2 · p. 15

Abstract

from arXiv · show

Recommender systems have significantly developed in recent years in parallel with the witnessed advancements in both internet of things (IoT) and artificial intelligence (AI) technologies. Accordingly, as a consequence of IoT and AI, multiple forms of data are incorporated in these systems, e.g. social, implicit, local and personal information, which can help in improving recommender systems' performance and widen their applicability to traverse different disciplines. On the other side, energy efficiency in the building sector is becoming a hot research topic, in which recommender systems play a major role by promoting energy saving behavior and reducing carbon emissions. However, the deployment of the recommendation frameworks in buildings still needs more investigations to identify the current challenges and issues, where their solutions are the keys to enable the pervasiveness of research findings, and therefore, ensure a large-scale adoption of this technology. Accordingly, this paper presents, to the best of the authors' knowledge, the first timely and comprehensive reference for energy-efficiency recommendation systems through (i) surveying existing recommender systems for energy saving in buildings; (ii) discussing their evolution; (iii) providing an original taxonomy of these systems based on specified criteria, including the nature of the recommender engine, its objective, computing platforms, evaluation metrics and incentive measures; and (iv) conducting an in-depth, critical analysis to identify their limitations and unsolved issues. The derived challenges and areas of future implementation could effectively guide the energy research community to improve the energy-efficiency in buildings and reduce the cost of developed recommender systems-based solutions.

1 Introduction

Rising building energy use and the importance of changing occupants’ behavior motivate recommender systems as a human-centered energy-saving approach. The paper reviews these systems, organizes them in a taxonomy, and analyzes their challenges and future directions.

Motivation

  • The building sector consumes more than 40% of global energy, and its consumption is expected to keep increasing.The cited projections give average annual growth of 1.3% in OECD countries and more than 2% in non-OECD countries from 2018 to 2050.Source: §1 · p. 2

Behavioral change

  • Recommender systems can support energy-saving behavioral change, but sustaining changes in end-user behavior remains a key challenge.The paper identifies training and awareness exercises, incentive recommendations, and feedback assessments as requirements for inducing permanent change.Source: §1 · p. 2

Review scope

  • The paper presents a review of building energy-saving recommender systems and identifies open issues and challenges to their effective development.It frames these systems around human involvement in promoting energy-efficient behavior and reducing carbon emissions.Source: §1 · p. 2

Taxonomy

  • Its taxonomy covers recommendation objectives, recommender methods and algorithms, computing platforms, evaluation metrics, and incentive measures.These factors are discussed in relation to the efficacy of current implementations.Source: §1 · p. 2

Review contribution

  • The paper presents its review framework as the first review of recommender systems for energy efficiency in buildings.The authors describe the review as addressing a previously unreviewed field of energy-saving recommender systems.Source: §1 · p. 2

Analysis and future directions

  • The paper critically analyzes existing frameworks to identify unresolved challenges and difficulties.It also offers future orientations aimed at overcoming existing issues and improving system quality and applicability.Source: §1 · p. 3

2 Methodology

The review uses a narrative literature search to examine building energy-efficiency recommendations and their relationship to intelligent systems and human decision-making. It organizes the literature into three research directions and focuses on their intersection.

Search methodology

  • The authors conducted a narrative review using Scopus searches of titles, abstracts, and keywords for work published from 2000 to 2020.The search terms included recommendation and recommender-system terms alongside energy saving, energy efficiency, buildings, and behaviour.Source: §2 · p. 3

Research directions

  • The Scopus search returned 283 articles, organized into three research directions.These concern building energy efficiency, intelligent systems promoting energy saving, and recommender systems placing humans at the center of energy-efficiency decisions.Source: §2 · p. 3

Recommendation contexts

  • In the recommender-systems literature, recommendations target either building owners’ energy plans and strategies or occupants’ immediate energy-saving actions.Owner recommendations balance energy saving and comfort; occupant actions include turning devices off or shifting their use.Source: §2 · p. 4

Review scope

  • The review focuses on the intersection of building energy efficiency, intelligent systems, and recommendations that involve human decision-making.It examines systems combining intelligent methods and action recommendations to improve efficiency and make conventional buildings smart.Source: §2 · pp. 3–4

3 Recommender systems for energy efficiency in buildings

This section describes energy-efficiency recommender systems for buildings, covering their objectives, methodologies, computing platforms, evaluation metrics, recommendation processes, and related trade-offs. It also outlines metrics for novelty, freshness, and user acceptance, alongside feedback ratings, green currency, and peer comparison as incentives for encouraging energy-saving behavior and tailoring recommendations.

3.1 The objectives of energy efficiency recommender systems

  • Strategy recommenders select building, forecasting, or operational strategies, while action recommenders suggest occupant behaviors that reduce energy use while balancing comfort and efficiency.Actions may assume fixed needs or use occupants’ flexibility to shift activities; end-users retain the final decision.Source: §3.1 · p. 5

3.2 Methodologies and algorithms for energy efficiency recommendations

  • Case-based, collaborative-filtering, and context-aware methods use rules or similar users, preferences, and usage context to tailor building energy-saving recommendations.Case-based systems can match current usage footprints to stored cases, while collaborative filtering identifies similar users; context-aware systems often need longer user interaction to gather data.Source: §3.2 · pp. 6–7
  • Rasch models estimate whether users can implement recommendations; probabilistic relational models infer preferences from user-item transactions, while fusion methods combine multiple data sources or recommenders.Rasch-based systems seek a balance between user behavior and recommendation difficulty; fusion approaches may combine energy, environmental, and preference data.Source: §3.2 · pp. 7–8
  • Deep learning can address collaborative filtering’s cold-start and linear-latent-parameter limitations, while classical optimization recommends or automatically implements energy-saving controls.Optimization approaches model uncertainties such as weather, building parameters, behavior, demand variation, and renewable generation to balance energy savings with thermal comfort.Source: §3.2 · pp. 8–9

3.3 Computing platforms

  • Cloud platforms support resource-intensive processing but raise IoT-data privacy concerns; edge computing supports local processing and privacy, while fog computing has no identified energy-efficiency recommendation application.Hybrid edge-cloud systems distribute processing according to constraints, including using edge devices for smaller datasets and cloud resources for more demanding tasks.Source: §3.3 · pp. 9–10

3.4 Evaluation metrics

  • An experimental study found Pearson and Spearman correlation results quite similar, so using either alone can avoid redundant ranking-accuracy evaluation.The section also identifies rating-error measures such as MAE and RMSE, and information-retrieval measures such as precision, recall, and F1; these can use a rating threshold to classify relevance.Source: §3.4 · p. 10
  • Serendipity measures surprise in accepted relevant actions, while freshness gauges whether recommendations are new; serendipity can conflict with accuracy, so relevance should be checked periodically.For small households with limited energy-saving actions, freshness can be assessed by checking whether the same action was suggested within the previous k hours or days.Source: §3.4 · p. 11
  • Acceptance ratio estimates users’ likelihood of accepting recommendations, helping systems choose between energy-saving suggestions with different acceptance rates.It can be calculated holistically or for a particular suggestion; other evaluation measures include coverage, confidence, trust, utility, risk, and robustness.Source: §3.4 · p. 11

3.5 Incentive measures

  • Feedback ratings capture users’ satisfaction and help adapt recommendations to their preferences, potentially supporting sustained energy reductions.Source: §3.5 · p. 12
  • Green currency can reward energy savings; one system promotes engagement with CO2 credits for recognized savings and achievements.Source: §3.5 · p. 12
  • Social comparison motivates household electricity saving by showing users peer or neighbor energy-use rankings and better-performing profiles.The approach draws on social psychology and uses normalized comparison modules in eco-feedback tools.Source: §3.5 · p. 12

4 Critical analysis and discussion

The discussion reviews how energy-efficiency recommender systems are deployed and the challenges to their reliability and commercialization. Most surveyed systems focus on households, while both action and strategy recommendations receive attention.

4 Critical analysis and discussion

  • Cloud computing is the most common platform, offering flexible resource scaling but raising privacy and service-availability concerns.The discussion notes possible data exposure or deletion and outages, including denial-of-service attacks.Source: §4 · p. 12
  • Most surveyed systems target household energy-use behavior, while action and strategy recommendations receive nearly equal attention.Other frameworks address commercial buildings, workplaces, hotels, and hospitals.Source: §4 · p. 12
  • The section examines difficulties affecting reliable energy-saving recommendations alongside the market drivers and barriers to commercialization.It frames these as issues for further analysis of energy-efficiency recommender systems.Source: §4 · p. 12

4.1 Limitations and difficulties

The review identifies data sparseness, cold start, limited benchmark datasets, and poor reproducibility as difficulties for energy-saving recommender systems. Figure 4 summarizes the broader limitations and commercialization issues.

4.1 Limitations and difficulties

  • Figure 4 summarizes the main difficulties and commercialization issues affecting energy-saving recommender systems.The review groups the limitations into data sparseness, cold start, benchmark-dataset shortages, and reproducibility problems.Source: §4.1 · p. 12, Figure 4 · p. 14

4.1.1 Data sparseness

  • Sparse ratings for available actions or users can produce unreliable recommendations and reduce consumer satisfaction.Systems often rely on limited demographic information and customer ratings for actions.Source: §4.1.1 · p. 13

4.1.2 Cold start problem

  • New consumers and actions lack ratings, making cold start especially acute early in deployment and limiting collaborative recommendations for new actions.Explicit preferences, user-similarity methods, probabilistic recommendations, and rule-based recommenders are described as possible responses.Source: §4.1.2 · pp. 13–14

4.1.3 Lack of benchmark datasets

  • Energy recommendation systems need datasets from varied building environments, but collection difficulties and missing open repositories hinder evaluation and comparison.The review notes that comparisons can become difficult or impossible without open-access datasets.Source: §4.1.3 · p. 14

4.1.4 Lack of reproducibility of results

  • Absent toolkits make efficacy comparisons hard to reproduce, while researchers have limited guidance on algorithms and baselines.The review calls for common reproducibility definitions, broader experiments, recommender frameworks, experimental platforms, and best-practice guidelines.Source: §4.1.4 · p. 14

4.2 Commercialization of developed solutions

The paper discusses policy and organizational incentives for energy efficiency alongside technical and regulatory barriers to market adoption. It highlights benchmarking, dataset, and compliance challenges.

4.2.1 Market drivers

  • EU climate and energy targets and organizational energy-saving plans are presented as market drivers for efficiency initiatives.The 2030 EU package includes emissions, renewable-energy, and primary-energy goals; firm strategies include reducing energy use and carbon footprints.Source: §4.2.1 · p. 15

4.2.2 Market barriers

  • Technical barriers include inconsistent evaluation standards and missing comprehensive datasets, while readiness, compatibility, and business models also affect a solution’s ability to change its target market.The review describes these factors in the context of energy-saving recommender systems.Source: §4.2.2 · p. 15
  • Regional energy-consumption regulations can complicate commercialization because recommender systems must comply with multiple standards for international recognition.The paper gives Greece’s 2011 building-permit requirements as an example of regional rules.Source: §4.2.2 · p. 15

5 Current challenges and future orientations

Figure 5 summarizes the current challenges and future orientations of energy-efficiency recommender systems addressed in the paper.

5 Current challenges and future orientations

  • Figure 5 brings together challenges and future orientations for energy-efficiency recommender systems.The supplied text does not specify the figure’s individual themes or visual encoding.Source: §5 · p. 15, Figure 5 · p. 16

5.1 Current challenges

The section surveys challenges spanning explainability, personalization, privacy, temporal context, and scalability in energy recommender systems.

5.1.1 Explainable AI & recommender systems

  • Energy recommender systems have limited literature explaining why particular recommendations are produced, despite broader explainable-AI research and examples in other recommendation domains.The cited energy-related work includes a user-centric visual-analytics approach for forecasting and analyzing prosumer electricity demand.Source: §5.1.1 · p. 17

5.1.2 Psycho-cognitive recommender systems

  • Psycho-cognitive systems combine preference understanding, habit and attitude change detection, choice prediction, and adaptation to changing environments.The survey presents these mechanisms as supporting more sustainable and energy-efficient behavior and awareness.Source: §5.1.2 · p. 17

5.1.3 Privacy preserving recommendations

  • Privacy-preserving recommendation approaches protect consumer feedback by hiding ratings from servers or other consumers, using methods including perturbation, encryption, and local differential privacy.The discussed energy-system design also encrypts sensitive consumption, feedback, ratings, and personal information before server submission.Source: §5.1.3 · pp. 17–18

5.1.4 Time-aware recommendations

  • Energy recommendations give temporal information relatively little weight, although real-time, time-aware advice is especially relevant to changing situations such as pandemic-driven home energy use.The survey identifies contextual information in neural collaborative filtering as a possible approach.Source: §5.1.4 · pp. 18–19

5.1.5 Large-scale recommender systems

  • Connected devices increase data volume and complexity, challenging traditional methods and motivating scalable algorithms and evaluation metrics that account for freshness, diversity, serendipity, and cold start.Discussed responses include parallel or distributed processing, GPUs, and energy-domain data sampling or compression.Source: §5.1.5 · pp. 19–20

5.2 Future orientations

The paper outlines future directions for explainable, visual, non-intrusive, and edge- or fog-based energy recommendations, alongside broader system improvements.

5.2 Future orientations

  • The future-orientations section presents explainability, visualization, non-intrusive monitoring, and edge/fog computing as directions for energy recommender systems.Its proposed visual, NILM, and edge/fog directions address distinct aspects of recommendation presentation, sensing, and computation.Source: §5.2 · p. 20, §5.2.1 · p. 20, §5.2.2 · p. 21

5.2.1 Explainable recommender systems

  • Initial evaluations of hybrid explanations combining persuasion facts with textual explanations report a 20% increase in the acceptance ratio of provided recommendations.The framework pairs explanations of why advice is suggested with statistical facts about its benefits.Source: §5.2.1 · p. 20

5.2.2 Visualization recommender systems

  • Future visualization work should test effects of different displays, develop accessible energy-specific visualizations, connect recommendations to visualized anomalies, and provide timely personalized advice.The section notes increased domestic energy consumption associated with working from home during COVID-19.Source: §5.2.2 · pp. 20–21

5.2.3 Non-intrusive recommender systems

  • Non-intrusive load monitoring could replace individual appliance submetering, but recommender-system use requires assessing the signal fidelity of NILM device fingerprints.The survey identifies potential system-cost reduction from using NILM instead of submetering.Source: §5.2.3 · pp. 21–22

5.2.4 Edge/Fog recommender systems

  • Cloud-to-edge systems can suffer feedback delays from network bandwidth and latency, while computing recommendations on edge devices can support real-time processing and privacy preservation.The survey also notes possible reductions in computational time and cloud-hosting costs from edge or fog deployment.Source: §5.2.4 · pp. 22–23

6 Conclusions

The conclusion reviews recommender systems for building energy efficiency, describes their evolution, and identifies future directions and scope challenges.

6 Conclusions

  • The review develops a taxonomy and critically analyzes system strengths, limitations, challenges, and emerging topics for future improvement.The taxonomy considers recommender components and implementation factors across existing systems.Source: §6 · p. 23
  • The survey distinguishes generations: early systems analyze energy data with collaborative filtering, case-based, PRM, or Rasch-based engines; the second combines energy footprints with other data using information fusion and deep learning.The paper states that this second-generation approach helps generate more accurate recommendations.Source: §6 · p. 23
  • The third generation adds explanations, visualizations, and time-aware processing; the conclusion identifies NILM, edge computing, and privacy preservation as directions for highly automated systems.The paper associates edge computing and edge AI with real-time recommendations and reports improved recommendation quality, acceptance, and user satisfaction.Source: §6 · p. 24
  • Extending recommendations to appliance acquisition adds energy use, manufacturer, price, and other specifications as choice factors, making comprehensive systems more challenging to develop.The conclusion describes appliance acquisition as an extension beyond recommending energy-saving actions.Source: §6 · p. 24

Related papers

Loading 2102.07654v1…