CoChat

Optimising Deep Learning at the Edge for Accurate Hourly Air Quality Prediction

Open access

The DOI matches its Crossref record; 41 of the 41 references it cites were found in a registry or an index.

Abstract

Accurate air quality monitoring requires processing of multi-dimensional, multi-location sensor data, which has previously been considered in centralised machine learning models. These are often unsuitable for resource-constrained edge devices. In this article, we address this challenge by: (1) designing a novel hybrid deep learning model for hourly PM2.5 pollutant prediction; (2) optimising the obtained model for edge devices; and (3) examining model performance running on the edge devices in terms of both accuracy and latency. The hybrid deep learning model in this work comprises a 1D Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) to predict hourly PM2.5 concentration. The results show that our proposed model outperforms other deep learning models, evaluated by calculating RMSE and MAE errors. The proposed model was optimised for edge devices, the Raspberry Pi 3 Model B+ (RPi3B+) and Raspberry Pi 4 Model B (RPi4B). This optimised model reduced file size to a quarter of the original, with further size reduction achieved by implementing different post-training quantisation. In total, 8272 hourly samples were continuously fed to the edge device, with the RPi4B executing the model twice as fast as the RPi3B+ in all quantisation modes. Full-integer quantisation produced the lowest execution time, with latencies of 2.19 s and 4.73 s for RPi4B and RPi3B+, respectively.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.3390/s21041064

Cites

41 of 41 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 The feature detector of 1D CNN slides over time-series data.
Figure 1 The feature detector of 1D CNN slides over time-series data.CC BY · PubMed Central
Figure 2 An LSTM cell structure.
Figure 2 An LSTM cell structure.CC BY · PubMed Central
Figure 3 Post-training optimisation methods provided by TensorFlow [46].
Figure 3 Post-training optimisation methods provided by TensorFlow [46].CC BY · PubMed Central
Figure 4 PM2.5 concentration gathered by Node 1 (Aotizhongxin monitoring site). This figure depicts the whole period of PM2.5 concentration, starting from 1 March 2013 to 28 February 2017.
Figure 4 PM2.5 concentration gathered by Node 1 (Aotizhongxin monitoring site). This figure depicts the whole period of PM2.5 concentration, starting from 1 March 2013 to 28 February 2017.CC BY · PubMed Central
Figure 5 Autocorrelation coefficients for PM2.5 concentration with different time lags.
Figure 5 Autocorrelation coefficients for PM2.5 concentration with different time lags.CC BY · PubMed Central
Figure 6 The proposed hybrid CNN-LSTM model. In this model, there are two parallel inputs (INPUT-1 and INPUT-2), one coverts the local node data (processed by CNN-1), and the other contains spatiotemporal dependency of PM2.5 data (processed by CNN-2).
Figure 6 The proposed hybrid CNN-LSTM model. In this model, there are two parallel inputs (INPUT-1 and INPUT-2), one coverts the local node data (processed by CNN-1), and the other contains spatiotemporal dependency of PM2.5 data (processed by CNN-2).CC BY · PubMed Central
Figure 7 Illustration of spatiotemporal consideration for predicting the value of PM2.5 concentration at Node 1. Nodes 2–12 send their PM2.5 data to Node 1. Node 1 uses these PM2.5 data as the second input of the proposed model (INPUT-2 in Figure 6), whereas the local data at Node 1 are used as the first input (INPUT-1 in Figure 6). This technique also applies to all other nodes.
Figure 7 Illustration of spatiotemporal consideration for predicting the value of PM2.5 concentration at Node 1. Nodes 2–12 send their PM2.5 data to Node 1. Node 1 uses these PM2.5 data as the second input of the proposed model (INPUT-2 in Figure 6), whereas the local data at Node 1 are used as the first input (INPUT-1 in Figure 6). This technique also applies to all other nodes.CC BY · PubMed Central
Figure 8 Details of data processing in the proposed deep learning model.
Figure 8 Details of data processing in the proposed deep learning model.CC BY · PubMed Central
Figure 9 Evaluation scenario used in this work.
Figure 9 Evaluation scenario used in this work.CC BY · PubMed Central
Figure 10 Boxplot of the prediction deviation of all investigated models at Node 1.
Figure 10 Boxplot of the prediction deviation of all investigated models at Node 1.CC BY · PubMed Central
Figure 11 Line plot of real and predicted PM2.5 data at Node 1.
Figure 11 Line plot of real and predicted PM2.5 data at Node 1.CC BY · PubMed Central
Figure 12 Scatter plots of real and model predicted values of PM2.5 at all nodes.
Figure 12 Scatter plots of real and model predicted values of PM2.5 at all nodes.CC BY · PubMed Central
Figure 13 TensorFlow Lite model size comparison.
Figure 13 TensorFlow Lite model size comparison.CC BY · PubMed Central
Figure 14 Comparison of TensorFlow Lite execution time for test data.
Figure 14 Comparison of TensorFlow Lite execution time for test data.CC BY · PubMed Central
Figure 15 Boxplot of prediction deviation resulted from each TFLite model. The deviation is calculated by subtracting the real values from the predicted values of each TFLite model.
Figure 15 Boxplot of prediction deviation resulted from each TFLite model. The deviation is calculated by subtracting the real values from the predicted values of each TFLite model.CC BY · PubMed Central
Figure 16 Scatter plot of the prediction data obtained by TensorFlow and TensorFlow Lite models.
Figure 16 Scatter plot of the prediction data obtained by TensorFlow and TensorFlow Lite models.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 41 references

Research works printed but not linked: 7

  1. Hinton, G., Vinyals, O., and Dean, J. (2015). Distilling the Knowledge in a Neural Network. arXiv.
  2. Xu, J., Chen, L., Lv, M., Zhan, C., Chen, S., and Chang, J. (2021). HighAir: A Hierarchical Graph Neural Network-Based Air Quality Forecasting Method. arXiv.
  3. Zhang, Q., Lam, J.C., Li, V.O., and Han, Y. (2020). Deep-AIR: A Hybrid CNN-LSTM Framework forFine-Grained Air Pollution Forecast. arXiv.
  4. Chen, H.C., Putra, K.T., and Chun-WeiLin, J. (2021). A Novel Prediction Approach for Exploring PM2.5 Spatiotemporal Propagation Based on Convolutional Recursive Neural Networks. arXiv.
  5. Wu, H., Judd, P., Zhang, X., Isaev, M., and Micikevicius, P. (2020). Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation. arXiv.
  6. Li, J., and Alvarez, R. (2021). On the quantization of recurrent neural networks. arXiv.
  7. Abadi, M., Agarwal, A., Barham, P., Brevdo, E., Chen, Z., Citro, C., Corrado, G.S., Davis, A., Dean, J., and Devin, M. (2016). TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems. arXiv.

Cited by

Metadata

Authors
Wardana, I. N. K., Gardner, J. W., & Fahmy, S. A.
Affiliations
University of Warwick (GB)
Corresponding author
Wardana, I Nyoman Kusuma
Venue
Sensors
Publisher
MDPI AG
Published
2021-02-04
Volume
21
Issue
4
Pages
1064
ISSN
1424-8220 (ISSN-L)
OpenAlex
W3126841703
Cited by
61
References
41
Citations by year
2026: 10 · 2025: 8 · 2024: 14 · 2023: 16 · 2022: 11 · 2021: 2
FWCI
3.48
Type
journal article
Language
en
Open access
Locations
  • Sensors (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Warwick Research Archive Portal (University of Warwick) (acceptedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • King Abdullah University of Science and Technology Repository (King Abdullah University of Science and Technology) (submittedVersion, other-oa)
Funders
Lembaga Pengelola Dana Pendidikan (Ref: S-1027/LPDP.4/2019)
Topics
  • Physical Sciences › Environmental Science › Environmental Engineering › Air Quality Monitoring and Forecasting
  • Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
Keywords
air quality forecasting, PM2.5, edge computing, CNN-LSTM hybrid model, model compression, Raspberry Pi, inference latency, sensor networks

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.