A novel least squares twin support vector regression method is proposed based on the robust L₁-norm distance to alleviate the negative effect of traffic data with outliers. Although there is some known work for the short-term traffic flow prediction problems, their efficacy depends heavily on the collected traffic data, which are often affected by various external factors (e.g. weather, traffic jam or accident), leading to errors and missing data. This makes it difficult to pick an effective method that accurately predicts the traffic state. As a contribution of this paper, an iterative algorithm is designed to solve the non-smooth L₁-norm terms of our method; its convergence also proved. Further, a comprehensive traffic flow indicator system based on speed, traffic flow, occupancy and ample degree is utilized in this paper. We also extend the proposed method to a nonlinear version by hybridizing the polynomial kernel and radial basis function kernel, where the weight coefficient of hybrid kernel is determined by the change tendency of traffic data. To promote the prediction performance, the parameters of our nonlinear method are optimized by adaptive fruit fly optimization algorithm. Extensive experiments on real traffic data are performed to evaluate our model. The results indicate that the newly constructed model yields better prediction performance and robustness than other models in various experimental settings.

提出了一种基于鲁棒L _{1 -}范数距离的最小二乘双支持向量回归方法，以减轻交通数据与异常值的负面影响。虽然对于短期交通流量预测问题有一些已知的工作，但它们的有效性在很大程度上取决于收集的交通数据，这些数据往往受到各种外部因素（例如天气、交通拥堵或事故）的影响，导致错误和数据丢失. 这使得很难选择一种有效的方法来准确预测交通状况。作为本文的贡献，设计了一种迭代算法来解决非光滑 L ₁- 我们方法的规范条款；它的收敛性也得到了证明。此外，本文还利用了基于速度、交通流量、占用率和充足度的综合交通流量指标系统。我们还通过混合多项式核和径向基函数核将所提出的方法扩展到非线性版本，其中混合核的权重系数由交通数据的变化趋势决定。为了提高预测性能，我们的非线性方法的参数通过自适应果蝇优化算法进行了优化。对真实交通数据进行了大量实验来评估我们的模型。结果表明，在各种实验设置中，新构建的模型比其他模型具有更好的预测性能和鲁棒性。