基于改进极限学习机和深度神经网络融合的车辆轨迹长期预测

The Long Term Prediction of Vehicle Trajectory Based on Improved Extreme Learning Machine and Depth Neural Network Fusion

针对轨迹预测中基于物理的运动模型长期预测(超过1 s)不可靠、深度学习预测模型易产生梯度消失或梯度爆炸的问题,提出了一种将改进极限学习机(ELM)和深度神经网络进行信息融合的预测模型。改进的极限学习机以模拟退火粒子群优化(SAPSO)算法为基础,改进的深度神经网络使用修正线性单元(ReLU)函数替换激活函数,并使用均方根反向传播(RMSProp)算法来优化深度神经网络,试验结果表明,提出的预测算法具有较高的预测精度。

For the problem that the long-term prediction of physics-based motion model(more than 1 s)in trajectory prediction is not reliable,and the deep learning prediction model easily produces gradient disappearance or gradient explosion,a prediction model which fuses information for the improved Extreme Learning Machine(ELM)and deep neural network is proposed.An improved extreme learning machine is based on Simulated Annealing Particle Swarm Optimization(SAPSO)algorithm,the improved depth neural network uses the Rectified Linear Unit(ReLU)function replacement activation function,and uses the Root Mean Square Propagation(RMSProp)algorithm to optimize neural network,the test results show that the proposed prediction algorithm has high prediction accuracy.