1 of 15

IoT and AI-Based System for Forescasting Particulate Matter in Panama

UNIVERSIDAD TECNOLÓGICA DE PANAMÁ

ENG. JOSÉ COLLADO

PH.D CRISTIAN PINZÓN

PH.D EDWIN COLLADO

PH.D YESSICA SAÉZ

2 of 15

Current situation

https://apps.who.int/iris/bitstream/handle/10665/280113/PMC6357572.pdf?sequence=1&isAllowed=y

Nearly 99% of the world's population breathes polluted air

3 of 15

It is estimated that Panama's population will increase by approximately 200,000 every five (5) years from 2020 to 2050

Behavior of air pollution in Panama from 2013 to 2017 shows that emissions of air pollutants are mainly produced by cars.

Approximately 17% of deaths in Panama are caused by diseases that might be related to air pollution

Current situation

https://www.inec.gob.pa/publicaciones/Default.aspx

4 of 15

Implementation of ICTs to monitor pollutants and generate indicators that allow solving problems related to environmental pollution and health in Panama.

Conceptual model

The monitoring system is made up of:

Monitoring stations.
Data storage unit.
Data visualization and analysis platform.

Project ITE18-R2-011: Monitoring network based on Internet of Things (IoT) for the generation of air pollution indicators in Panama

https://www.laccei.org/LACCEI2021-VirtualEdition/full_papers/FP160.pdf

5 of 15

This work proposes to include an architecture based on Fog and Cloud Computing to improve the monitoring system developed in ITE18-R2-011.

Proposed architecture

Now in this work we proposes to include an architecture based on Fog and Cloud Computing with a prediction model to improve the monitoring system developed in ITE18-R2-011.

The advantages of using fog computing with cloud computing are:

· Lower latency (leidensi): in the propagation and transmission of information reducing the response time due to the proximity to end-user devices.

· Real-time interactions: By placing computing resources at the edge of the network, very close to end users and devices, improving the quality and efficiency of services with faster and more responsive services.

· Geographic Distribution: It allows placing services close to the end devices improving connectivity and availability even in situations where connectivity to the cloud may be intermittent or unreliable.

· Heterogeneity: Enables the integration of diverse heterogeneous devices allowing them to operate and collaborate without problems, and ensures that they can interact and coexist efficiently, regardless of their individual capabilities or technical specifications.

· Interoperability: Fog offers a wide range of services can be distributed diverse users and applications across disparate systems.

The image shows a diagram of the model, which is divided into 4 parts: the sensing layer, the fog layer, the cloud layer, and the mobile layer.

First we have the sensing layer, which is responsible for capturing data on different air pollutants and atmospheric conditions, such as CO, NO2, O3, SO2, PM2.5 and PM10, together with environmental factors such as temperature, humidity and wind parameters.

This layer will use the environmental monitoring stations previously developed and implemented in the research project ITE18-R2-011, which saves on hardware costs.

Next we have the fog layer. This layer serves as an intermediary to transmit the information generated and requested by the other layers.

Here, the data obtained from the sensing layer reaches the Mobile API Gateway and goes through the data preprocessing module before being sent to the cloud layer to remove or fix erroneous data.

The fog layer will be developed in Python using Rassberry PI 4 and placed near the monitoring stations to avoid latency issues.

In the cloud layer, a microservice architecture is applied, which consists of modular and independent services, each with its own database, designed for specific tasks.

Within the Cloud Layer, we have the following microservices:

User Authentication: This microservice will oversees authenticating users who want to access the system.
Notification Service: This microservice will oversees the results generated by the predictive module and sends notifications to the corresponding users based of their location when harmful levels of air pollution are detected.
Monitoring Service: This microservice will oversee requests to the predictive module to monitor air pollution levels.
Air Pollution Data Service: This microservice will oversee the collection and management of the data measured in the sensing layer and its storage in its corresponding database.
User Context Service: This microservice will oversee the collection of information about the user's location.
Prediction Module: This microservice will use historical air pollution data to make predictions of air pollution levels.

Mobile layer provides end-users with a mobile application to access air quality data in real time.

6 of 15

CNN-LSTM model

7 of 15

Dataset

80% train

10% validation

10% test

Data preparation

Data structura for training

Monitoring station

8 of 15

Model training

Hyperparameters used:

Number of neurons in the convolutional layers: [64,64], [64,32], [32,32]
Number of neurons in the recurrent layer: 50, 100
Learning Rate: 0.001
Batch Size: 256

The combination of convolutional layers [64,64] units and a BiLSTM layer of 50 units was selected.

9 of 15

Model validation

By adjusting the learning rate and batch size, from 0.001 to 0.0001 and from 256 to 512 respectively, fluctuations during training were reduced and the R2 performance improved from 0.55 to 0.65.

Next, we proceeded to compare the performance of our model with other time series prediction techniques.

10 of 15

Models comparation

11 of 15

PM_2.5 Forecast| 24 hours

12 of 15

Benefits of the proposal

High availability: Each microservice is independent of the other, so that if one fails, it will only affect the functionalities related to that microservice.

Scalability: Each microservice can be developed in different languages or technologies, allowing for the extension, updating, and incorporation of new components.

Functionality optimization: Each microservice must be defined and optimized for a single service rather than the entire application.

An intelligent system based on neural networks was designed, integrated into a microservices architecture that facilitates the incorporation of artificial intelligence models into an ecosystem for real-time data management.

13 of 15

Conclusions

The proposed model provides an efficient and novel tool in Panama to generate information about air pollution that helps public and private institutions.

This architecture is composed of four parts: the sensing layer, the fog layer, the cloud layer, and the mobile layer.

The evaluation of the proposed model for predicting 24-hour particulate matter concentrations demonstrated its ability to identify patterns and generate useful result.

The proposed model provides an efficient and novel tool in Panama to generate information about air pollution that helps public and private institutions, especially those related to environmental protection and healthcare.

This architecture is composed of four parts: the sensing layer, which is responsible for data collection; the fog layer, which serves as an intermediary to transmit the information generated and requested in the other layers; the cloud layer, which is composed of several microservices for processing and data analysis for the assessment and detection of harmful levels of air pollution; and the mobile layer, which allows end-users to access the system through a mobile application.

Although the proposed model for predicting PM over 24 hours manages to identify patterns and offers useful results, it has a lag in its predictions at certain times of the day, which is its main weakness.

14 of 15

Future work

Develop and implement the proposed model in our recent research project SENACYT FID23-078: SIMA: IoT and AI-based air pollution monitoring systems in Panama.

Validate the model and evaluate its performance using more data.

Share this tool wih public and private institutions in Panama.

15 of 15

THANK YOU

Eng. José Collado

jose.collado@utp.ac.pa

Research Group: Telecommunications Engineering and Intelligent Systems Applied to Society (ITSIAS)