Abstract:Forecasting temporal processes such as virus spreading in epidemics often requires more than just observed time-series data, especially at the beginning of a wave when data is limited. Traditional methods employ mechanistic models like the SIR family, which make strong assumptions about the underlying spreading process, often represented as a small set of compact differential equations. Data-driven methods such as deep neural networks make no such assumptions and can capture the generative process in more detail, but fail in long-term forecasting due to data limitations. We propose a new hybrid method called MP-PINN (Multi-Phase Physics-Informed Neural Network) to overcome the limitations of these two major approaches. MP-PINN instils the spreading mechanism into a neural network, enabling the mechanism to update in phases over time, reflecting the dynamics of the epidemics due to policy interventions. Experiments on COVID-19 waves demonstrate that MP-PINN achieves superior performance over pure data-driven or model-driven approaches for both short-term and long-term forecasting.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to predict the spread of epidemics (such as COVID - 19) more accurately, especially in the case of limited data. Traditional methods either rely on mechanical models (such as the SIR model), which have strong assumptions about the spread process but are difficult to reflect complex changes in reality; or rely on data - driven methods (such as deep neural networks), which can capture more details but perform poorly in long - term predictions due to insufficient data. To solve these problems, the author proposes a new hybrid method - MP - PINN (Multi - Phase Physics - Informed Neural Network). This method combines the advantages of the Physics - Informed Neural Network (PINN) and multi - phase modeling, enabling the model to adjust parameters according to the actual situation at different stages, so as to better adapt to the dynamic changes of epidemic spread. ### Specific Problems and Solutions 1. **Challenges in short - term and long - term predictions**: - Traditional model - driven methods (such as the SIR model) perform well in long - term predictions, but are not flexible enough in the short - term due to overly rigid assumptions. - Data - driven methods (such as deep learning) perform well in the short - term, but fail in long - term predictions due to a lack of understanding of the mechanism. 2. **Complexity of epidemic spread**: - Epidemic spread is affected by multiple factors, such as policy interventions, changes in public behavior, etc. Traditional single - parameter models are difficult to capture these dynamic changes. 3. **Situations with limited data**: - In the early stage of an epidemic, data is very limited, so a method that can use limited data and combine prior knowledge is needed for prediction. ### Innovations of MP - PINN - **Multi - phase modeling**: MP - PINN divides the entire epidemic spread process into multiple phases, and each phase uses different SIR parameters to reflect the spread characteristics of different periods. - **Physics - Informed Neural Network (PINN)**: By embedding physical laws (such as the differential equations of the SIR model) into the neural network, it is ensured that the model not only fits the data but also conforms to the known epidemiological laws. - **Dynamically adjust parameters**: At different stages, the model can dynamically adjust key parameters such as the infection rate \(\beta\) and the recovery rate \(\gamma\) to better adapt to changes in the real situation. ### Experimental Results The author conducted experiments on the COVID - 19 data of 21 regions in Italy. The results show that MP - PINN outperforms the traditional SIR model, pure data - driven methods (MLP), and single - stage PINN in both short - term and long - term predictions. Specifically: - **Short - term prediction (within 30 days)**: MP - PINN and SP - PINN perform similarly and both are better than SIR and MLP. - **Long - term prediction (more than 30 days)**: MP - PINN is significantly better than other methods, especially more accurate in capturing long - term trends. ### Conclusion MP - PINN successfully solves the limitations of traditional methods in short - term and long - term predictions by combining physical information and multi - phase modeling, providing a more flexible and accurate epidemic prediction tool. This shows that in the face of complex epidemic spread, combining domain knowledge and data - driven methods can significantly improve the accuracy of prediction.

MP-PINN: A Multi-Phase Physics-Informed Neural Network for Epidemic Forecasting

Approaching epidemiological dynamics of COVID-19 with physics-informed neural networks

A Physics-Informed Neural Network approach for compartmental epidemiological models

Euler iteration augmented physics-informed neural networks for time-varying parameter estimation of the epidemic compartmental model

Physics-Informed Neural Networks Integrating Compartmental Model for Analyzing COVID-19 Transmission Dynamics.

EINNs: Epidemiologically-Informed Neural Networks

Bayesian Physics-Informed Neural Networks for real-world nonlinear dynamical systems

Epi-DNNs: Epidemiological Priors Informed Deep Neural Networks for Modeling COVID-19 Dynamics.

A Comparative Study on Deep Learning Models for COVID-19 Forecast

NeuralCODE: Neural Compartmental Ordinary Differential Equations Model with AutoML for Interpretable Epidemic Forecasting

A Novel Multi-scale Spatiotemporal Graph Neural Network for Epidemic Prediction.

MSGNN: Multi-scale Spatio-temporal Graph Neural Network for Epidemic Forecasting

Graph Neural Ordinary Differential Equations for Epidemic Forecasting

Neural parameter calibration and uncertainty quantification for epidemic forecasting

Deep Dynamic Epidemiological Modelling for COVID-19 Forecasting in Multi-level Districts

A data-driven epidemic model with human mobility and vaccination protection for COVID-19 prediction

Temporal Multiresolution Graph Neural Networks For Epidemic Prediction

Epicasting: An Ensemble Wavelet Neural Network (EWNet) for Forecasting Epidemics

Epidemiology-Aware Neural ODE with Continuous Disease Transmission Graph

Forecasting infectious disease prevalence with associated uncertainty using neural networks