自适应粒子群神经网络交通流预测模型
A Traffic Forecasting Model Using Adaptive Particle Swarm Optimization Trained Neural Network

针对传统神经网络预测模型预测结果准确性低且存在大量无效迭代的问题,提出了自适应权重粒子群神经网络交通流预测(PSOA？？NN)模型。首先根据待预测点的上下游观测点数和历史数据,随机初始化若干组模型参数并计算每组参数对应粒子的适应度;然后采用改进的sigmoid函数替代原有模型中的固定惯性权重,并根据其中适应度变好的粒子更新粒子速度和位置,一直迭代到粒子适应度小于预设值为止;最后将满足条件粒子对应的模型参数应用到神经网络模型,根据实时交通流数据预测出15 min后的数据。仿真结果表明,使用PSOA？？NN模型,可使得在同等预测误差范围内收敛速度提升0.6～1.7倍。
A novel traffic forecasting model using particle swarm optimized neural network with adaptive weights (PSOA？？NN) is proposed to address the issue of low accuracy prediction of traditional neural network forecasting model with a lot of invalid iterations in prediction process. Several groups of model parameters are initialized casually according to the observing point numbers of up and down streams and historical data in a neighbourhood of the forecast location, and the particle fitness of corresponding parameters in each group is calculated. Then, an improved sigmoid function is used to replace the fixed inertia weights in original model, speeds and locations are updated for the particles with improved fitness, and the iteration continues until the fitnesses of particles are less than a preset value. Finally, the model parameters that satisfies particle requirement are applied to the neural network model, and the traffic data 15 min later is predicted according to real time traffic flow data. Simulation results show that the convergent speed of the PSOA？？NN model increases by about 0.6 to 1.7 times within the identical deviation range