Non-technical losses detection in energy consumption focusing on energy recovery and explainability

Coma-Puig, Bernat; Carmona, Josep

doi:10.1007/s10994-021-06051-1

Cited by 12 publications

(5 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Based on the revealed dependencies on the input data, the behavior of the model is predicted. Regressions were used in [27][28][29][30][31]; correlation analysis was used in [32][33][34][35][36][37][38][39].…”

Section: Methods Of Identification and Assessment Of Nontechnical Los...mentioning

confidence: 99%

Data-Driven Machine Learning Methods for Nontechnical Losses of Electrical Energy Detection: A State-of-the-Art Review

Pazderin,

Kamalov,

Gubin

et al. 2023

Energies

View full text Add to dashboard Cite

Nontechnical losses of electrical energy (NTLEE) have been a persistent issue in both the Russian and global electric power industries since the end of the 20th century. Every year, these losses result in tens of billions of dollars in damages. Promptly identifying unscrupulous consumers can prevent the onset of NTLEE sources, substantially reduce the amount of NTLEE and economic damages to network grids, and generally improve the economic climate. The contemporary advancements in machine learning and artificial intelligence facilitate the identification of NTLEE sources through anomaly detection in energy consumption data. This article aims to analyze the current efficacy of computational methods in locating, detecting, and identifying nontechnical losses and their origins, highlighting the application of neural network technologies. Our research indicates that nearly half of the recent studies on identifying NTLEE sources (41%) employ neural networks. The most utilized tools are convolutional networks and autoencoders, the latter being recognized for their high-speed performance. This paper discusses the main metrics and criteria for assessing the effectiveness of NTLEE identification utilized in training and testing phases. Additionally, it explores the sources of initial data, their composition, and their impact on the outcomes of various algorithms.

show abstract

Section: Methods Of Identification and Assessment Of Nontechnical Los...mentioning

confidence: 99%

Data-Driven Machine Learning Methods for Nontechnical Losses of Electrical Energy Detection: A State-of-the-Art Review

Pazderin,

Kamalov,

Gubin

et al. 2023

Energies

View full text Add to dashboard Cite

show abstract

“…We highlight the work of Lundberg et al [15] that proposes an implementation of the Shapley Values in healthcare where the explanatory method is used to prevent hypoxaemia during surgery, and the work of Meacham et al [25], where explainability is used for analysis of patience re-admittance. In Coma-Puig and Carmona [26], Shapley Values are leveraged in order to validate the correctness of the predictive approach in a utility company. The SHAP implementation of the Shapley values has the strong theoretical foundation of the original game theory approach, with the advantage of providing offline explanations that are consistent with the online explanations.…”

Section: Explanation Of Machine-learning Modelsmentioning

confidence: 99%

An Explainable Decision Support System for Predictive Process Analytics

Galanti¹,

Leoni²,

Monaro³

et al. 2022

Preprint

View full text Add to dashboard Cite

Predictive Process Analytics is becoming an essential aid for organizations, providing online operational support of their processes. However, process stakeholders need to be provided with an explanation of the reasons why a given process execution is predicted to behave in a certain way. Otherwise, they will be unlikely to trust the predictive monitoring technology and, hence, adopt it. This paper proposes a predictive analytics framework that is also equipped with explanation capabilities based on the game theory of Shapley Values. The framework has been implemented in the IBM Process Mining suite and commercialized for business users. The framework has been tested on real-life event data to assess the quality of the predictions and the corresponding evaluations. In particular, a user evaluation has been performed in order to understand if the explanations provided by the system were intelligible to process stakeholders.

show abstract

“…This allowed us to implement different solutions that would improve our system overall. However, as we explain in our work [34], there exist several biases and other data related problems that cannot be treated with a general approach, since it affects specific domains. In other words, the observational data causes different specific biases in each domain that it is difficult to treat at once.…”

Section: Mitigating the Existing Problems For Each Model Builtmentioning

confidence: 99%

“…Different aspects of this work can be read in Chapters 3.4 and 3.5. [34] fully introduces the Shapley values in our system. As we explain in this work, the use of Shapley values made us understand well the shortcomings and problems of our system, and helped us to correct them by implementing several solutions (e.g.…”

Section: Our Published Workmentioning

confidence: 99%

“…• Depth: The big difference in terms of depth is that Feature Importance provides a superficial and modular explanation of the influence of each variable on the predictions, while LIME and SHAP offer deeper explanations at the instance level. Therefore, as we have seen in our use case, Feature Importance can be interesting because of its speed in getting a first sanity check of the model, but its superficiality would not allow us to implement the double-checking methods exemplified in [31] (that uses LIME) or [35] (that uses Shapley values).…”

Section: Comparison In Our Ntl Detection Systemmentioning

confidence: 99%

See 1 more Smart Citation

Human-aware application of data science techniques

Coma Puig

View full text Add to dashboard Cite

In recent years there has been an increase in the use of artificial intelligence and other data-based techniques to automate decision-making in companies, and discover new knowledge in research. In many cases, all this has been performed using very complex algorithms (so-called black-box algorithms), which are capable of detecting very complex patterns, but unfortunately remain nearly uninterpretable. Recently, many researchers and regulatory institutions have begun to raise awareness of their use. On the one hand, the subjects who depend on these decisions are increasingly questioning their use, as they may be victims of biases or erroneous predictions. On the other hand, companies and institutions that use these algorithms want to understand what their algorithm does, extract new knowledge, and prevent errors and improve their predictions in general. All this has meant that researchers have started to focus on the interpretability of their algorithms (for example, through explainable algorithms), and regulatory institutions have started to regulate the use of the data to ensure ethical aspects such as accountability or fairness. This thesis brings together three data science projects in which black-box predictive machine learning has been implemented to make predictions: - The development of an NTL detection system for an international utility company from Spain (Naturgy). We combine a black-box algorithm and an explanatory algorithm to guarantee our system's accuracy, transparency, and robustness. Moreover, we focus our efforts on empowering the stakeholder to play an active role in the model training process. - A collaboration with the University of Padova to provide explainability to a Deep Learning-based KPI system currently implemented by the MyInvenio company. - A collaboration between the author of the thesis and the Universitat de Barcelona to implement an AI solution (a black-box algorithm combined with an explanatory algorithm) to a social science problem. The unique characteristics of each project allow us to offer in this thesis a comprehensive analysis of the challenges and problems that exist in order to achieve a fair, transparent, unbiased and generalizable use of data in a data science project. With the feedback arising from the research carried out to provide satisfactory solutions to these three projects, we aim to: - Understand the reasons why a prediction model can be regarded as unfair or untruthful, making the model not generalisable, and the consequences from a technical point of view in terms of low accuracy of the model, but also how this can affect us as a society. - Determine and correct (or at least mitigate) the situations that cause the problems in terms of robustness and fairness of our data. - Assess the difference between the interpretable algorithms and black-box algorithms. Also, evaluate how well the explanatory algorithms can explain the predictions made by the predictive algorithms. - Highlight what the stakeholder's role in guaranteeing a robust model is and how to convert a data-driven approach to solve a predictive problem into a data-informed approach, where the data patterns and the human knowledge are combined to maximize profit. En els últims anys s'ha produït un augment de l'ús de la intel·ligència artificial i altres tècniques basades en dades per automatitzar la presa de decisions en les empreses, i descobrir nous coneixements en la recerca. En molts casos, tot això s'ha realitzat utilitzant algorismes molt complexos (anomenats algorismes de caixa negra), que són capaços de detectar patrons molt complexos, però, per desgràcia, continuen sent gairebé ininterpretables. Recentment, molts investigadors i institucions reguladores han començat a conscienciar sobre el seu ús. D'una banda, els subjectes que depenen d'aquestes decisions estan qüestionant cada vegada més el seu ús, ja que poden ser víctimes de prejudicis o prediccions errònies. D'altra banda, les empreses i institucions que utilitzen aquests algoritmes volen entendre el que fa el seu algorisme, extreure nous coneixements i prevenir errors i millorar les seves prediccions en general. Tot això ha fet que els investigadors hagin començat a centrar-se en la interpretació dels seus algorismes (per exemple, mitjançant algorismes explicables), i les institucions reguladores han començat a regular l'ús de les dades per garantir aspectes ètics com la rendició de comptes o la justícia. Aquesta tesi reuneix tres projectes de ciència de dades en els quals s'ha implementat aprenentatge automàtic amb algorismes de caixa negra per fer prediccions: - El desenvolupament d'un sistema de detecció de NTL (Non-Technical Losses, pèrdues d'energia no tècniques) per a una empresa internacional del sector de l'energia d'Espanya (Naturgy). Aquest sistema combina un algorisme de caixa negra i un algorisme explicatiu per garantir la precisió, la transparència i la robustesa del nostre sistema. A més, centrem els nostres esforços en la capacitació dels treballadors de l'empresa (els "stakeholders") per a exercir un paper actiu en el procés de formació dels models. - Una col·laboració amb la Universitat de Padova per proporcionar l'explicabilitat a un sistema KPI basat en Deep Learning actualment implementat per l'empresa MyInvenio. - Una col·laboració de l'autor de la tesi amb la Universitat de Barcelona per implementar una solució d'AI (un algorisme de caixa negra combinat amb un algorisme explicatiu) a un problema de ciències socials. Les característiques úniques de cada projecte ens permeten oferir en aquesta tesi una anàlisi exhaustiva dels reptes i problemes que existeixen per a aconseguir un ús just, transparent, imparcial i generalitzable de les dades en un projecte de ciència de dades. Amb el feedback obtingut de la recerca realitzada per a oferir solucions satisfactòries a aquests tres projectes, el nostre objectiu és: - Entendre les raons per les quals un model de predicció pot considerar-se injust o poc fiable, fent que el model no sigui generalitzable, i les conseqüències des d'un punt de vista tècnic en termes de baixa precisió del model, però també com pot afectar-nos com a societat. - Determinar i corregir (o almenys mitigar) les situacions que causen els problemes en termes de robustesa i imparcialitat de les nostres dades. - Avaluar la diferència entre els algorismes interpretables i els algorismes de caixa negra. A més, avaluar com els algorismes explicatius poden explicar les prediccions fetes pels algorismes predictius. - Ressaltar el paper de les parts interessades ("Stakeholders") per a garantir un model robust i com convertir un enfocament únicament basat en les dades per resoldre un problema predictiu en un enfocament basat en les dades però complementat amb altres coneixements, on els patrons de dades i el coneixement humà es combinen per maximitzar els beneficis.

show abstract

Non-technical losses detection in energy consumption focusing on energy recovery and explainability

Cited by 12 publications

References 43 publications

Data-Driven Machine Learning Methods for Nontechnical Losses of Electrical Energy Detection: A State-of-the-Art Review

Data-Driven Machine Learning Methods for Nontechnical Losses of Electrical Energy Detection: A State-of-the-Art Review

An Explainable Decision Support System for Predictive Process Analytics

Human-aware application of data science techniques

Contact Info

Product

Resources

About