Conference PaperPDF Available

A Recurrent Neural Network for Multisensory Non-Intrusive Load Monitoring on a Raspberry Pi

Authors:
Alon Harell at Simon Fraser University
Alon Harell
Stephen Makonin at Simon Fraser University
Stephen Makonin
Ivan V. Bajic at Simon Fraser University
Ivan V. Bajic
Download full-text PDFDownload full-text PDF
Read full-text
Download citation

Abstract

Understanding how appliances consume power is important for energy conservation. Non-intrusive load monitoring (NILM) helps meeting energy conservation goals by inferring individual appliance power usage from a single measurement. By using additional, readily available, sensor information such as weather data, it is possible to improve the accuracy of NILM. In this paper we present an example case of disaggregating two appliances using a recurrent neural network, and show how the use of multisensory data (weather and power) as an input improves performance. We demonstrate the system on a Raspberry Pi -- a cost-efficient platform with limited computational capabilities.
MMSP_20
18.pdf
Content uploaded by Stephen Makonin
Author content
All content in this area was uploaded by Stephen Makonin on Jul 06, 2018
Content may be subject to copyright.
MMSP_20
18.pdf
Content uploaded by Stephen Makonin
Author content
All content in this area was uploaded by Stephen Makonin on Jul 05, 2018
Content may be subject to copyright.
Accepted at the IEEE 20th International Workshop on Multimedia Signal Processing (MMSP 2018).
A Recurrent Neural Network for Multisensory
Non-Intrusive Load Monitoring on a Raspberry Pi
Alon Harell, Stephen Makonin, and Ivan V. Baji´
c
School of Engineering Science, Simon Fraser University, Burnaby, Canada
Email: aharell@sfu.ca, smakonin@sfu.ca, ibajic@ensc.sfu.ca
Abstract—Understanding how appliances consume power is
important for energy conservation. Non-intrusive load monitoring
(NILM) helps meeting energy conservation goals by inferring
individual appliance power usage from a single measurement.
By using additional, readily available, sensor information such
as weather data, it is possible to improve the accuracy of NILM.
In this paper we present an example case of disaggregating
two appliances using a recurrent neural network, and show
how the use of multisensory data (weather and power) as
an input improves performance. We demonstrate the system
on a Raspberry Pi – a cost-efficient platform with limited
computational capabilities.
Index Terms—NILM, disaggregation, RNN, LSTM, weather
I. PROTO TY PE DE SC RI PT IO N
Non-intrusive load monitoring (NILM) [?] traditionally uses
the aggregate power meter as the only sensor. Multisensory
NILM can help improve the accuracy of appliance usage de-
tection. In order to demonstrate the usefulness of multisensory
data for NILM, we compare the performance of two neural
networks on readings from two sub-meters taken from the
AMPds2 [?] dataset. Both networks are hybrid recurrent neural
networks, which output both an estimated total power and a
simplified ON/OFF state classification, for each appliance. A
detailed schematic of the architecture is presented in Fig. ??.
One network will use only the total power as an input,
while the other will also incorporate temperature readings.
Training, validation, and testing were all performed on a subset
of the data from the AMPds2 [?] dataset. In AMPds2, power
readings are taken from a central meter as well as 20 sub-
meters in one house over two years. In order to simplify
the problem, we take only two of the sub-meters (HPE and
OFE) and manually add them to create the aggregated power
signal. These two sub-meters were selected because they have
opposite correlation with the temperature measurement while
almost no correlation with each other. Training was done on
the first 600 days of measurement (500 for training, 100 for
validation) and testing was performed on the following 100
days. Training was limited to 100 epochs or until validation
loss plateaued. After training, the weights of both networks
were loaded on to a Raspberry Pi 3 computer (model B V1.2),
which runs each network and performs power disaggregation
faster than real-time, presenting dynamically updating graphs
of both appliances and their comparison to ground truth data.
This work was funded in part by IC-IMPACTS.
Having both the classification and total power estimate at
the output allows us to compare the two networks using both
the F-score and Estimation Accuracy [?]. As seen in Table ??,
the inclusion of temperature in the input data improves the
performance in both metrics, especially for OFE.
TABLE I
F1-SC OR E AND ES TI MATIO N ACC URAC Y RES ULTS
Input Sub-meter F1-Score Est Acc
HPE 0.9997 0.966
Power OFE 0.790 0.535
Overall 0.865 0.912
HPE 0.9996 0.976
Power + Temperature OFE 0.847 0.688
Overall 0.903 0.939
Regression
Linear
Dense2outputs
Input
LSTM1
128Nodes
LSTM2
128Nodes
Classification
Sigmoid
Dense2outputs
Output
Fig. 1. Network architecture
Copyright © 2018 IEEE. The original publication is available for download at ieeexplore.ieee.org.
... These cases are marked with a subscript m in the 'Output' column of Table 2. Where multi-task learning is done on different modes, the corresponding outputs are jointed with an '&' in Table 2: [35,63,97,106] trained networks on both on/off and active power data. Ref. [117] used both active and reactive power of an appliance as target, and [59] used three targets, i.e., the aggregate power (P total ), the appliance power (P app ), and the difference between the two (P rest ). ...
... The following authors report results from a comparison of input features [22,36,89,94,106,107,117]. Ref. [117] was the first DNN-NILM approach we are aware of that used multiple input features. ...
... Refs. [89,94,106] all worked with the AMPds dataset. This data set contains measurements from a single house, thus all results stem from seen evaluation scenarios. ...
Review on Deep Neural Networks Applied to Low-Frequency NILM
Article
Full-text available
  • Apr 2021
This paper reviews non-intrusive load monitoring (NILM) approaches that employ deep neural networks to disaggregate appliances from low frequency data, i.e., data with sampling rates lower than the AC base frequency. The overall purpose of this review is, firstly, to gain an overview on the state of the research up to November 2020, and secondly, to identify worthwhile open research topics. Accordingly, we first review the many degrees of freedom of these approaches, what has already been done in the literature, and compile the main characteristics of the reviewed publications in an extensive overview table. The second part of the paper discusses selected aspects of the literature and corresponding research gaps. In particular, we do a performance comparison with respect to reported mean absolute error (MAE) and F1-scores and observe different recurring elements in the best performing approaches, namely data sampling intervals below 10 s, a large field of view, the usage of generative adversarial network (GAN) losses, multi-task learning, and post-processing. Subsequently, multiple input features, multi-task learning, and related research gaps are discussed, the need for comparative studies is highlighted, and finally, missing elements for a successful deployment of NILM approaches based on deep neural networks are pointed out. We conclude the review with an outlook on possible future scenarios.
View
... The authors combined a dAE and a LSTM network, to reconstruct the appliance specific signal and then to identify in which appliance this signal belongs to. Furthermore, several RNN-based approaches have been published [157,158,159,160,161,162,163,164] for state detection or load disaggregation while in some occasions RNN models also deployed for real-time disaggregation [165,166] in residential buildings. Though, until [167] was published, there wasn't a study that had employed RNN for industrial energy disaggregation. ...
NILM Applications: Literature review of learning approaches, recent developments and challenges
Article
Full-text available
  • Feb 2022
  • ENERG BUILDINGS
This paper presents a critical approach to the non-intrusive load monitoring (NILM) problem, by thoroughly reviewing the experimental framework of both legacy and state-of-the-art studies. Some of the most widely used NILM datasets are presented and their characteristics, such as sampling rate and measurements availability are presented and correlated with the performance of NILM algorithms. Feature engineering approaches are analyzed, comparing the hand-made with the automatic feature extraction process, in terms of complexity and efficiency. The evolution of the learning approaches through time is presented, making an effort to assess the contribution of the latest state-of-the-art deep learning models to the problem. Performance evaluation methods and evaluation metrics are demonstrated and it is attempted to define the necessary requirements for the conduction of fair evaluation across different methods and datasets. NILM limitations are highlighted and future research directions are suggested.
View
... WaveNILM requires only 512 new parameters for each new input signal, and only 3072 parameters for each added output (load). This is particularly important when exploring additional external measurements that have been shown to be useful for disaggregation such as weather data [19], or time of day [20]. Reactive power has been shown to be of particular benefit to NILM, providing most of the performance benefit, with apparent power and current only adding marginal improvement. ...
View
... WaveNILM requires only 512 new parameters for each new input signal, and only 3072 parameters for each added output (load). This is particularly important when exploring additional external measurements that have been shown to be useful for disaggregation such as weather data [19], or time of day [20]. Reactive power has been shown to be of particular benefit to NILM, providing most of the performance benefit, with apparent power and current only adding marginal improvement. ...
Wavenilm: A causal neural network for power disaggregation from the complex power signal
Preprint
Full-text available
  • Feb 2019
Non-intrusive load monitoring (NILM) helps meet energy conservation goals by estimating individual appliance power usage from a single aggregate measurement. Deep neural networks have become increasingly popular in attempting to solve NILM problems; however, many of them are not causal which is important for real-time application. We present a causal 1-D convolutional neural network inspired by WaveNet for NILM on low-frequency data. We also study using various components of the complex power signal for NILM, and demonstrate that using all four components available in a popular NILM dataset (current, active power, reactive power, and apparent power) we achieve faster convergence and higher performance than state-of-the-art results for the same dataset. Code now available at: https://github.com/picagrad/WaveNILM
View
A Study of the Effects of Appliance Energy Signatures on Different Neural Network Types in Non-Intrusive Load Monitoring
Article
  • Jan 2023
Access to appliance-level energy measurements enable end users to take conservative actions. The availability of energy data and smart energy meters has recently increased the popularity of Non-Intrusive Load Monitoring (NILM). Both end users and utilities can estimate appliance-level energy consumption data using NILM without installing appliance-level sensors. Neural Network (NN)-based NILM models have shown promising results compared to other energy disaggregation approaches. The estimation accuracy depends on the NN type and architecture selected for the NILM application. The varying characteristics and operating parameters of different types of appliances make a substantial impact on NILM performance. Therefore, it is essential to study the effects of appliance signatures on different NN architectures to select the optimum NN type and efficient energy disaggregation. This study comprehensively investigates the impact of various parameters specific to household environments and appliances on disaggregation performance. The NN models used in this study were modified and improved for comparison on a common platform. Subsequently, a rigorous comparison was performed using disaggregation performance indices, the training time of NNs, and computational complexity. The results show that the disaggregation performance depends on the types of NN, network hyperparameters, appliance activation signatures, and the availability of training data volume. The findings of this study demonstrate that the Convolutional Neural Network (CNN) outperforms other NNs for NILM applications. Additionally, the specific characteristics of individual appliances have a notable impact on disaggregation performance.
View
Electricity, water, and natural gas consumption of a residential house in Canada from 2012 to 2014
Article
Full-text available
  • Jun 2016
With the cost of consuming resources increasing (both economically and ecologically), homeowners need to find ways to curb consumption. The Almanac of Minutely Power dataset Version 2 (AMPds2) has been released to help computational sustainability researchers, power and energy engineers, building scientists and technologists, utility companies, and eco-feedback researchers test their models, systems, algorithms, or prototypes on real house data. In the vast majority of cases, real-world datasets lead to more accurate models and algorithms. AMPds2 is the first dataset to capture all three main types of consumption (electricity, water, and natural gas) over a long period of time (2 years) and provide 11 measurement characteristics for electricity. No other such datasets from Canada exist. Each meter has 730 days of captured data. We also include environmental and utility billing data for cost analysis. AMPds2 data has been pre-cleaned to provide for consistent and comparable accuracy results amongst different researchers and machine learning algorithms.
View
Nonintrusive Load Monitoring (NILM) Performance Evaluation
Article
Full-text available
  • Dec 2014
Nonintrusive load monitoring (NILM), sometimes referred to as load disaggregation, is the process of determining what loads or appliances are running in a house from analysis of the power signal of the whole-house power meter. As the popularity of NILM grows, we find there is no consistent way researchers are measuring and reporting accuracies. In this short communication, we present a unified approach that would allow for consistent accuracy testing.
View
Nonintrusive appliance load monitoring
Article
  • Jan 1993
A nonintrusive appliance load monitor that determines the energy consumption of individual appliances turning on and off in an electric load, based on detailed analysis of the current and voltage of the total load, as measured at the interface to the power source is described. The theory and current practice of nonintrusive appliance load monitoring are discussed, including goals, applications, load models, appliance signatures, algorithms, prototypes field-test results, current research directions, and the advantages and disadvantages of this approach relative to intrusive monitoring
View