基于最优梯度自适应优化算法的交通流预测

短时交通流预测在交通控制中起着基础的作用。建立了一类不需要选取初始值、带有动态参数的指数平滑模型。以预测误差平方和SSE最小为目标，构造了优选并自动生成最佳参数，使平滑模型得以优化的最速下降算法，增强了模型对时间序列的适应能力。较好地解决了指数平滑预测中，平滑参数靠检验确定且为静态、平滑初值难以确定并导致预测偏差等问题。通过比较上海浦东的实测数据和其它预测算法，验证了该算法的有效性和实用性。

Traffic Flow Prediction Based on Best Grads Self-adaptive Optimized Algorithm

The forecast for short-time traffic flow is the foundation of the traffic control. This paper sets up one new class of exponential smooth model with dynamic smooth parameter without for selecting the initial parameter. Aimed at the square sum of error (SSE), it constructs the algorithm to iterate and selects an optimal parameter for optimizing the new model, which enhances the adaptability of the model. Such problems, i.e., the parameter is static and determined only by one’s experiences, and smoothing initial value isn’t easy to determine and leads to a deviation easily, are resolved well. Compared with Shanghai Pudong New Area’s real data, it compares this model with other forecast methods to validate the efficiency and practicability of this algorithm.