US 12,243,421 B2
Method and arrangement for predicting switching times of a signal group of a signal installation for controlling a flow of traffic
Harald Frank, Höhenkirchen-Siegertsbrunn (DE); Michel Tokic, Tettnang (DE); and Marc Christian Weber, Munich (DE)
Assigned to Yunex GmbH, Munich (DE)
Appl. No. 17/781,642
Filed by Yunex GmbH, Munich (DE)
PCT Filed Nov. 10, 2020, PCT No. PCT/EP2020/081561
§ 371(c)(1), (2) Date Jun. 1, 2022,
PCT Pub. No. WO2021/110370, PCT Pub. Date Jun. 10, 2021.
Claims priority of application No. 10 2019 218 903.2 (DE), filed on Dec. 4, 2019.
Prior Publication US 2022/0415170 A1, Dec. 29, 2022
Int. Cl. G08G 1/081 (2006.01); G06N 3/045 (2023.01)
CPC G08G 1/081 (2013.01) [G06N 3/045 (2023.01)] 13 Claims
OG exemplary drawing
 
1. A method for predicting switching times of a signal group of a signal installation for controlling a flow of traffic, wherein
real state information, which influences the switching times of the signal group, is received at a fixed clock rate, and
a computing unit predicts at least one switching time of the signal group at the fixed clock rate by using artificial intelligence to which the real state information is supplied,
wherein
the real state information is supplied to a first neural network as input signals,
the first neural network computes estimated state information as a replacement for real state information or parts of the real state information that is/are not received on time or is/are received erroneously for a current time at the fixed clock rate, and
the first neural network outputs the estimated state information for the current time to the artificial intelligence at the fixed clock rate, the estimated state information replacing the real state information or the parts of the real state information that was/were not received on time or was/were received erroneously,
or wherein
the artificial intelligence is a third neural network to which the real state information or estimated state information is supplied as input signals,
the third neural network computes estimated state information as a replacement for real state information or parts of the real state information that is/are not received on time or is/are received erroneously for a current time at the fixed clock rate, and
the third neural network uses the estimated state information for the current time at the fixed clock rate in order to predict the at least one switching time of the signal group, the estimated state information replacing the real state information or the parts of the real state information that was/were not received on time or was/were received erroneously,
wherein the first neural network or the third neural network estimates state information for a current time, wherein it is then checked whether the current state information for the current time was received on time, completely and correctly, and wherein the real state information ( . . . y<d>t−3, y<d>t−2, y<d>t−1, y<d>t) is then replaced with the estimated state information if the real state information was not received on time or was received erroneously.