这是老师放出来的题解：
Let’s first focus on predicting the range of possible rates of traffic at the end of the highway (past mile N). To do this, we start with a large possible range [a,b] (initially set to [−999999999,+999999999]) and narrow / modify it as we scan through the different highway components from miles 1…N. Every time we see a sensor reading directly from the highway, this clips the possible range [a,b] to the range given by the sensor. Every time we see an on-ramp with range [a′,b′], the new range of possible traffic flows is [a+a′,b+b′]. Similarly, when we see an off-ramp with range [a′,b′], the new range of possible traffic flow values is [a−b′,b−a′] (after this update, we set the lower end of the range to zero if it goes negative, since we can’t have a negative rate of traffic flow).
Predicting the range of possible initial flows is similar and essentially symmetric, where we scan backwards and keep track of a working range [a,b] that is narrowed / modified appropriately by each highway feature.
翻译后：
让我们首先关注预测高速公路末端（N英里）可能的交通量范围。为此，我们从一个较大的可能范围[a，b]开始（最初设置为[-999999999，+999999999]），并在扫描1…N英里的不同高速公路组件时缩小/修改它。每次我们看到传感器直接从高速公路读取数据时，这会将可能的范围[a，b]剪辑到传感器给定的范围。每当我们看到范围为[a′，b′的入口匝道时，可能的交通流的新范围为[a+a′，b+b′。类似地，当我们看到范围为[a′，b′的出口匝道时，可能的交通流值的新范围为[a-b′，b-a′（在这次更新之后，如果范围的下限变为负值，我们将其设置为零，因为我们不能有负的交通流速率）。
预测可能的初始流量范围是相似的，并且基本上是对称的，我们向后扫描并跟踪一个工作范围[a，b]，该工作范围被每个公路特征适当地缩小/修改。