Imputing missing indoor air quality data via variational convolutional autoencoders: Implications for ventilation management of subway metro systems

Loy-Benitez, Jorge; Heo, SungKu

doi:10.1016/j.buildenv.2020.107135

Cited by 35 publications

(14 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The VAE structure consist of two main components (i) encoder q φ (Z|X) and (ii) decoder p θ (X|Z) as shown in figure 2 . Both the encoder and decoder are multilayered NN with parameters φ and θ respectively [2]. VAE follows the assumption that the input data X is generated by some underlying distribution p(X) which can be represented by the latent variable Z, where Z itself is generated by a distribution p(Z).…”

Section: B Variational Autoencoders (Vae)mentioning

confidence: 99%

“…Low Cost Sensors (LCS) have the potentials to enhance the spatio-temporal resolution of data acquisition for key GHG variables. LCS, however, are prone to diverse issues including bias, drifts, precision degradation, and loss of considerable amount of data due to operational issues [2]. Missing data is a pervasive issue, affecting most real-world datasets including medical records [3], [4], geo-informatics [5], traffic flow [6] and industrial applications [7], [8].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Missing Data Imputation on IoT Sensor Networks: Implications for on-site Sensor Calibration

Okafor¹,

Delaney²

2021

Preprint

View full text Add to dashboard Cite

show abstract

Section: B Variational Autoencoders (Vae)mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Missing Data Imputation on IoT Sensor Networks: Implications for on-site Sensor Calibration

Okafor¹,

Delaney²

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Low Cost Sensors (LCS) can be used to enhance spatial and temporal resolution of data acquisition for key GHG variables to corroborate satellite and remotebased sensing approaches. LCS, however, are prone to various failures including bias, drifts, precision degradation, and loss of considerable amount of data due to operational issues [2]. Missing data is a pervasive issue which occur in most real-world datasets including medical records [3,4], geo-informatics [5], traffic flow [6] and industrial applications [7,8].…”

Section: Introductionmentioning

confidence: 99%

“…Discriminative methods include Multiple Imputation by Chain Equations (MICE) [13], Random Forest-based Imputation (Missforest) [14] and matrix completion [15]. Generative methods consist mostly of techniques based on Deep Learning (DL) e.g Denoising Autoencoders (DAE) [2] [16] and Generative Adversarial Networks(GAN) [17] [18]. GAN learns the latent distribution of a dataset and can generate real samples from a random noise.…”

Section: Introductionmentioning

confidence: 99%

Missing Data Imputation on IoT Sensor Networks: Implications for on-site Sensor Calibration

Okafor¹,

Delaney²

2021

Preprint

View full text Add to dashboard Cite

IoT sensors are gaining more popularity in the environmental monitoring space due to their relatively small size, cost of acquisition and ease of installation and operation. They are becoming increasingly important<br>supplement to traditional monitoring systems, particularly for in-situ based monitoring. However, data collection based on IoT sensors are often plagued with missing values usually occurring as a result of sensor faults, network failures, drifts and other operational issues. Several imputation strategies have been proposed for handling missing values in various application domains. This paper examines the performance of different imputation techniques including Multiple Imputation by Chain Equations (MICE), Random forest based imputation (missForest) and K-Nearest Neighbour (KNN) for handling missing values on sensor networks deployed for the quantification of Green House Gases(GHGs). Two tasks were conducted: first, Ozone (O3) and NO2/O3 concentration data collected using Aeroqual and Cairclip sensors respectively over a six months data collection period were corrupted by removing data intervals at different missing periods (p) where p 2 f1day; 1week; 2weeks; 1monthg and also at random points on the dataset at varying proportion (r) where r 2 f5%; 10%; 30%; 50%; 70%g. The missing data were then filled using the different imputation strategies and their imputation accuracy calculated. Second, the performance of sensor calibration by different regression models including Multi Linear Regression (MLR), Decision Tree (DT), Random Forest (RF) and XGBoost (XGB) trained on the different imputed datasets were evaluated. The analysis showed the MICE technique to outperform the others in imputing the missing values on both the O3 and NO2/O3 datasets when missingness was introduced over periods p. MissForest, however, outperformed the rest when missingness was introduced as randomly occuring point errors. While the analysis demonstrated the effects of missing and imputed data on sensor calibration, experimental results showed that a simple model on the imputed dataset can achieve state of-the-art result on in-situ sensor calibration, improving the data quality of the sensor.

show abstract

“…Even though the related research has already identified autoencoders as a promising technique to address missing values and anomalies in monitoring building data sets [13][14][15][16], some significant research questions are still unaddressed. In particular, the existing studies have often focused on reconstructing a single type of signal or they have been limited by the small amount of available training data and computational power.…”

Section: Introductionmentioning

confidence: 99%

Indoor environment data time-series reconstruction using autoencoder neural networks

Liguori,

Markovic,

Dam

et al. 2020

Preprint

View full text Add to dashboard Cite

As the number of installed meters in buildings increases, there is a growing number of data time-series that could be used to develop data-driven models to support and optimize building operation. However, building data sets are often characterized by errors and missing values, which are considered, by the recent research, among the main limiting factors on the performance of the proposed models. Motivated by the need to address the problem of missing data in building operation, this work presents a data-driven approach to fill these gaps. In this study, three different autoencoder neural networks are trained to reconstruct missing indoor environment data time-series in a data set collected in an office building in Aachen, Germany. The models are applicable for different time-series obtained from room automation, such as indoor air temperature, relative humidity and CO 2 data streams. The results prove that the proposed methods outperform classic numerical approaches and they result in reconstructing the corresponding variables with average RMSEs of 0.42 C, 1.30 % and 78.41 ppm, respectively.

show abstract

Imputing missing indoor air quality data via variational convolutional autoencoders: Implications for ventilation management of subway metro systems

Cited by 35 publications

References 50 publications

Missing Data Imputation on IoT Sensor Networks: Implications for on-site Sensor Calibration

Missing Data Imputation on IoT Sensor Networks: Implications for on-site Sensor Calibration

Missing Data Imputation on IoT Sensor Networks: Implications for on-site Sensor Calibration

Indoor environment data time-series reconstruction using autoencoder neural networks

Contact Info

Product

Resources

About