Urban traffic accident risk prediction method based on a ConvLSTM

Abstract

The invention discloses an urban traffic accident risk prediction method based on a ConvLSTM. The method specifically comprises the following steps: carrying out space grid division on a predicted city; performing statistics on historical traffic accident data according to the same time interval and mapping the historical traffic accident data to a grid to generate a risk matrix changing along with time; dividing the risk matrix sequence into a continuous branch and a periodic branch according to the characteristics of continuity and periodicity of the occurrence of the traffic accident in time; extracting features of external data such as weather, holidays and festivals of the city and quantifying the features, and constructing an external feature vector; and establishing a ConvLSTM-based deep learning model to predict the traffic accident risk of each grid region of the city. According to the invention, modeling is carried out from two aspects of continuity and periodicity of traffic accidents, the ConvLSTM is utilized to extract the time features and the space features of the traffic accident data at the same time, and the external environment features are further fused, so that the accuracy of the prediction result is improved, and the method has certain guiding significance for traffic management and accident early warning.

Description

technical field [0001] The invention belongs to the field of traffic accident prediction, in particular to a method for predicting the risk of urban traffic accidents based on a convolutional long-short-term memory neural network. Background technique [0002] With the rapid development of urbanization and the surge in the number of motor vehicles, the frequency of traffic accidents has increased significantly around the world in the past ten years, becoming a major problem that threatens the safety of human life. In the face of these possible and unexpected risks of traffic accidents, how to use modern traffic big data to predict them has become a research hotspot. Predicting the risk of traffic accidents in various areas of the city can provide early warning for high-risk areas of accidents and help traffic management departments plan effective traffic management solutions. [0003] The traffic accident prediction problem belongs to the time-space sequence prediction prob...

AI Technical Summary

Problems solved by technology

The first type of research mainly uses machine learning methods to model and predict a single road. Machine learning-based methods are often difficult to learn high-dimensional features and nonlinear relationships between multi-source heterogeneous data related to traffic accidents, and only Forecast for a single road

The second type of research mainly uses deep learning methods, such as using Recurrent Neural Network (RNN) and Convolution Neural Network (CNN) to make grid-level predictions for the entire city, but a single network cannot simultaneously learn Temporal and spatial correlation of time-space series data

Images