Machine learning segment routing for multiple traffic matrices

In some embodiments, a method may be provided, the method comprising: receiving a first traffic matrix; receiving information about a link associated with each segment of the network; determining a total amount of segment streams using at least one non-linear deflection parameter applied to the traffic requirements of the first traffic matrix; determining a link flow for each link using the total amount of segment flows and a second input of the machine learning model; determining a link utilization rate of each link by using the link flow and the capacity of each link; by adjusting at least a value of at least one non-linear deflection parameter, a gradient descent method is used to learn a minimum value of a maximum amount of link utilization over the link by the machine learning model. Related systems, methods, and articles of manufacture are also disclosed. 在一些实施例中，可提供一种方法，该方法包括：接收第一业务矩阵；接收关于与网络的每个分段相关联的链路的信息；使用应用于第一业务矩阵的业务需求的至少一个非线性偏转参数，确定分段流的总量；使用分段流的总量、以及机器学习模型的第二输入，确定每个链路的链路流；使用链路流、以及每个链路的容量，确定每个链路的链路