Research on the Prediction of Emergency Response Time for Public Emergencies Based on Long and Short-Term Memory Network Models

Abstract

Efficient emergency response to public emergencies is key to minimizing social losses. This study proposes an ARIMA-LSTM combined measurement model. By employing an optimal weighted combination prediction method, the model integrates the linear trend analysis capabilities of the ARIMA model with the nonlinear time-series dependency learning advantages of the long short-term memory (LSTM) network. This approach extracts the linear and nonlinear features from public emergency response time prediction data, addressing the issue of low simulation accuracy in standalone ARIMA and LSTM models. The experimental results show that the prediction performance of the BP neural network, single ARIMA model, and LSTM model are all inferior to the ARIMA-LSTM combination model. When applying the model in this paper to predict the emergency response time for public emergencies in multiple regions, the average emergency response time is only 2.3 hours, and the positive rate of suspected events and outbreak events in early warning signals is less than 5%, indicating that the model in this paper has a low false alarm rate and high practicality. This provides a reliable path for improving the efficiency of emergency response departments and reducing social losses.