改进的拟仿射变换算法及其在交通流预测中的应用

基于演化算法，提出了一种改进的拟仿射变换演化算法(quasi-affine transformation evolutionary, QUATRE),以加快收敛速度。该算法将种群分为多个具有不同适应度水平的子群，并随机从高水平子群中选择代表来指导低水平子群的个体进化。受粒子群优化(particle swarm optimization, PSO)算法中惯性权重的启发，提出了一种新的参数更新方法，用来平衡探索与开发。此外，为了避免过早收敛到局部解，算法引入了一种跳跃机制摆脱局部陷阱。通过基准函数验证了该算法的优越性。该算法成功应用于基于神经网络的交通流量预测，其精度优于PSO、灰狼优化算法(grey wolf optimizer, GWO)、鲸鱼优化算法(whale optimization algorithm, WOA)和差分进化(differential evolution, DE)。

Evolutionary algorithms (EAs) have been widely used in the neural network to improve its gradient explosion and vanishing. This paper proposes an improved quasi-affine transformation evolutionary (QUATRE) algorithm to accelerate the convergence rate. The algorithm divides the population into multiple subgroups with different fitness levels and randomly selects exemplars from high-level subgroups to guide the individuals in low-levels. Inspired by the inertia weight of particle swarm optimization (PSO), a new updating method of the parameter is proposed to balance the exploration and exploitation. To avoid premature convergence to the local solutions, a jumping mechanism is brought to escape from local traps. The superiority of the proposed algorithm is proved by the benchmark functions. The algorithm is successfully applied to the traffic flow prediction of neural network, which has better accuracy than PSO, grey wolf optimizer (GWO), whale optimization algorithm (WOA) and differential evolution (DE).