US 12,230,134 B2
Systems and methods for public transit arrival time prediction
Soumen Pachal, Chennai (IN); Nancy Bhutani, New Delhi (IN); and Avinash Achar, Chennai (IN)
Assigned to Tata Consultancy Services Limited, Mumbai (IN)
Filed by Tata Consultancy Services Limited, Mumbai (IN)
Filed on Dec. 15, 2022, as Appl. No. 18/066,392.
Claims priority of application No. 202221042121 (IN), filed on Jul. 22, 2022.
Prior Publication US 2024/0029562 A1, Jan. 25, 2024
Int. Cl. G08G 1/123 (2006.01); G01C 21/34 (2006.01)
CPC G08G 1/123 (2013.01) [G01C 21/3492 (2013.01)] 15 Claims
OG exemplary drawing
 
1. A processor implemented method comprising:
obtaining, via one or more hardware processors, a route information pertaining to one or more trips associated to one or more vehicles;
segmenting, via the one or more hardware processors, the route information into a plurality of sections, wherein each of the plurality of sections comprises travel time of the one or more trips, and wherein the travel time comprises running time and dwell time;
grouping, via the one or more hardware processors, for each trip (i) travel time of one or more previous sections with reference to a current section from the plurality of sections, and (ii) travel time of one or more previous sections of a historical trip from a previous week, to obtain a grouped trip data;
generating, via an encoder comprised in an encoder-decoder model, a context vector using the grouped trip data, wherein the encoder-decoder model further comprises a decoder and one or more bidirectional layers at the decoder;
predicting, by the decoder via the one or more hardware processors, a travel time for one or more subsequent sections, for one or more trips based on one or more exogenous inputs further comprising (i) travel time and entry time of the historical trip from the previous week, (ii) travel time and entry time of a closest previous trip, (iii) the context vector, (iv) a current position of one or more vehicles, and (v) the current time, wherein a final arrival time is obtained based on the predicted travel time of the one or more subsequent sections; and
capturing, via the decoder comprised in an encoder-decoder model, an upstream propagation of one or more congestions that originate downstream in the one or more subsequent sections,
wherein the bidirectional layers at the decoder, capture, for a given section, influence of past congestions in time across the one or more subsequent sections propagating backward in space,
wherein in a bi-directional setting, an additional state-vector custom character is available and f2 is mapped, with an update custom character=custom character with reverse information flow, a state at time-step i, hi, is a concatenation custom character, a state-update ht==f1 (ht−1, ut) happens for each timestep with a state information flowing from left to right.