Forecast Covid-19 daily cases based on multivariate Bi-direction Long Short Term Memory

Covid-19 is highly contagious and can infect many people in a short time. Many people have built models to predict novel Coronavirus infections. The paper establishes a Bi-direction Long Short Term Memory model to predict the daily number of Covid-19 cases in China. The models are divided into univariate and univariate models with noise and multivariate models with noise. Noise is used to simulate data errors. The loss function value of the univariate model without error reached 0.000460 after 4600 epochs, and the loss function value of the multivariate model came to 0.000648 after 2460 epochs without noise. When noise is added, the loss function of the univariate model rises faster. With an error greater than or equal to 2 percent, the multivariate model shows a lower loss than the univariate model. This shows that the multivariate model predicts the number of Covid-19 patients with noise data better.