Research on forecast and recommendation technology of taxi passengers based on time-varying Markov decision process

To solve the problems of unloading rate caused by blind passenger search of taxis, the hotspot recommendation strategy of taxi passengers was proposed. The proposed strategy could optimize the process of matching passengers to the greatest extent to increase the efficiency of passenger search. Based on the historical trajectory data of taxis and the time series characteristics of hotspot passenger information, a segment prediction method was proposed based on recurrent neural network (SPBR) and a passenger recommendation model was proposed based on time-varying Markov decision process (TMDP). Experimental results show that the RMSE predicted by SPBR algorithm is 67.6%, 71.1% and 64.5% lower than the SVR, CART and BPNN algorithms. The expected return of taxis based on the TMDP algorithm is 35.9% higher than historical expectations. © 2021, Editorial Board of Journal on Communications. All right reserved.