From fight against COVID-19 to Zero Downtime

Surprisingly, over one year has passed already since the beginning of the COVID-19 pandemic, a situation that completely changed our lives: the way we interact, our habits, and even our thoughts. We are fighting this virus in many ways. Social distancing, vaccines, and medication are the most common, but many other methods have emerged, some of which come from data science and AI.

Back in March 2020, at a time when the Artificial Intelligence Team of the State Health Department of Ceará in Brazil started to face one of the greatest challenges of the modern age, the COVID-19 pandemic, the number of infections was increasing alarmingly and we had no training base or ready-to-use models to help predict its impact.

At the time, I was leading the team and we were in charge of coming up with a short-term strategy to provide information to the State Government for the purchase of ventilators and the creation of ICU beds. One of the few data available was the number of cases in the city of Wuhan in China, the first city to enter a lock-down to prevent the spread of the infection.

To overcome this challenge we used a late 1950’s model named Kalman Filter, a recursive algorithm that uses time-series measurement with statistical noise to generate estimates. The algorithm is adaptive and it does not require a lot of historical/training data. The early use of Kalman Filter happened in the Cold War between the Soviet Bloc and the North American Treaty Organization [2].

The model was used in the Apollo project to control the trajectory between Moon and Earth. When the Apollo 11 lunar, piloted by Neil Armstrong and a software application, landed on the Sea of Tranquility, the Kalman filter ensured that real-time tracking data from radar surveillance on Earth matched the sensors on board exactly [3].

Kalman Filters is also used by self-driving cars to estimate their localization in one environment. Sensors in automobiles are used to detect other cars, pedestrians, and obstacles. Knowing the location of these objects is essential in the decision-making process of autonomous vehicles in order to navigate without collisions [4].

The Kalman Filter uses a cycle of four steps [5]:

1. Make a prediction based on some previous known values

2. Obtain the measurement of the real state

3. Update the prediction based on errors

4. Repeat

Figure 1: Kalman Filter algorithm

The Kalman Filter algorithm is updated with new observations every day, and a forecast can be generated for the next day. The problem was to accurately estimate the number of cases in a month’s time with a filter that updates itself on a daily basis.

Just as the filter is used in autonomous cars to collect signals from different sensors, the proposal was to join the “signal” of the local COVID-19 cases from Ceara with the Wuhan case curve. Figure 2 shows the influence of Kalman Filter on predictions of Ceara COVID-19 cases (Local COVID Cases) based on curve of COVID-19 cases from other country.

Figure 2: The influence of Kalman Filter on predictions of Ceara COVID-19 cases (Local COVID Cases) based on curve of COVID-19 cases from other country.

The proposal was applied and brought surprising results. Below are two clippings from a local newspaper, with a day less from when the state reached the number of cases predicted by the Kalman filter.

Figure 3: Clipping from a local newspaper in 16th of April

Figure 4: Clipping from a local newspaper in 22nd of April

The proposal was published in the soft computing journal entitled Modeling the progression of COVID-19 deaths using Kalman Filter and AutoML (https://link.springer.com/article/10.1007/s00500-020-05503-5). We show that by using the Kalman filter and AutoML models, we can achieve very high accuracy in predicting COVID-19 cases. Kalman Filter approach needs data from other countries/cities to feed the model, which makes it possible to use the approach but with the risk that if the behavior of the country’s epidemic curve used to feed the model has very different characteristics from the region where we want to get the death curve can lead to a high margin of error.

The difference between the approach presented and the one commonly used in other studies is that the use of Kalman Filter allows a long-term prediction at the beginning of the epidemic period using data from other countries/regions [6].

In November of 2020 I moved to Stratio, a new challenge and opportunity to use the previous models created to fight against COVID-19, but now in another context, supporting the journey to move vehicles towards a Zero Downtime future.

Vehicle failures occur continuously on a global scale every day, and naturally affect the most intensively operated equipment, which is usually trucks and buses. These are the means of transportation that most of the world’s population rely on, as well as the means of transportation that provide food to our supermarkets, medicines for our pharmacies and almost everything that our lives and economies depend on. And now, more than ever, the failure of vehicles could delay the delivery of medicines, vaccines, and other vital life-saving supplies in this COVID-19 pandemic.

Here at Stratio, the main challenge is detecting vehicle failures before they occur. The same Kalman Filter used to model COVID-19 dynamics is used to combine multiple signals from the Stratio Box to provide better quality information. Better data then typically translates into better predictive models, which is our ultimate goal. This is an essential need for us, in particular for battery-electric vehicles. We use sensor fusion to enhance our models for battery electric vehicles by using knowledge of the battery pack’s structure, chemistry, its sensors, and Kalman Filters.

The result is more precise state-of-health and remaining useful life models for batteries.

Therefore, we can demonstratively use the same method in two absolutely distinct contexts: epidemic processes and in-vehicle physical processes. So, in this way, artificial intelligence (AI) benefits governments, citizens, and businesses, including fighting Covid-19, empowering resilience and improving green, sustainable growth and ensuring zero downtime in the delivery of medicines and vaccines.

[1] Ersin Soken, H., Sakai, S. I., Asamura, K., Nakamura, Y., & Takashima, T. (2017). Spin parameters and nonlinear kalman filtering for spinning spacecraft attitude estimation. Advances in the Astronautical Sciences, 160(February), 2615–2629.

[2] Grewal, M. S., & Andrews, A. P. (2010). Applications of Kalman Filtering in Aerospace 1960 to the Present. IEEE Control Systems, 30(3), 69–78. https://doi.org/10.1109/MCS.2010.936465

[3] “How an Inventor You’ve Probably Never Heard of Shaped the Modern World”, MIT Technology Review, 2021. [Online]. Available: https://www.technologyreview.com/2016/09/05/157723/how-an-inventor-youve-probably-never-heard-of-shaped-the-modern-world/. [Accessed: 21- Feb- 2021]

[4] Lin, M., Yoon, J., & Kim, B. (2020). Self-driving car location estimation based on a particle-aided unscented kalman filter. Sensors (Switzerland), 20(9). https://doi.org/10.3390/s20092544

[5] “Kalman Filter: Predict, Measure, Update, Repeat.”, Medium, 2021. [Online]. Available: https://medium.com/@tjosh.owoyemi/kalman-filter-predict-measure-update-repeat-20a5e618be66. [Accessed: 21- Feb- 2021]

[6] Han, T., Gois, F. N. B., Oliveira, R., Prates, L. R., & Porto, M. M. de A. (2021). Modeling the progression of COVID-19 deaths using Kalman Filter and AutoML. Soft Computing, 5. https://doi.org/10.1007/s00500-020-05503-5