LSTM neural network AEB system control method based on driver data

Abstract

The invention provides an LSTM neural network AEB system control method based on driver data, and the method comprises the steps of: acquiring and classifying the data of a driver during emergency braking in real vehicle driving, wherein the data comprise environment data and vehicle driver output data, and carrying out classification according to three types of working conditions: front vehicle resting, front vehicle slow speed and front vehicle emergency braking; inputting the classified data into an LSTM neural network model for training; wherein the input data comprises N1 time sequences of each parameter by taking the current moment as a center, the output data comprises N2+1 time sequences of each parameter at the current moment and N2 moments in the future, and outputting the parameter at the current moment as a prediction result; calculating a root-mean-square error between the prediction result and the driving data, and finishing training when the root-mean-square error is smaller than a set threshold value; applying the trained LSTM neural network model to actual driving, and outputting braking pressure for emergency braking. According to the invention, the anthropomorphism and comfort of the control system can be improved.

Description

technical field [0001] The invention belongs to the technical field of intelligent driving assistance, and in particular relates to a driver data-based LSTM neural network AEB system control method. Background technique [0002] In recent years, various evaluation agencies have incorporated the Automatic Emergency Braking System (AEB) into the active safety evaluation procedures. These vehicle evaluation regulations have promoted the development of automotive active safety to a certain extent. Among them, the AEB system has become hotspots. The AEB system can avoid traffic accidents such as rear-end collisions by means of hierarchical early warning or automatic braking in extremely dangerous situations. It is mainly composed of an environmental perception module, a control unit and an actuator. At present, the most common research on AEB system is the control algorithm based on rule design. This algorithm mainly depends on the artificial exhaustion of various working condit...

AI Technical Summary

Problems solved by technology

The most common way to train RNNs is backpropagation through time, but vanishing gradients often lead to parameter capture with short-term dependencies, while information from earlier time steps will gradually decay; conversely, gradient explosions can also occur, causing errors to vary with time. Each time step increases dramatically

Images