Research On Taxi Booking Habit Evaluation And Strategy Optimization Based On Artificial Intelligence Algorithm

Since the 20 th century,residents’ travel demand has been growing,and taxis play a crucial role in residents’ daily commuting.How to optimize taxi drivers’ decision making on taking orders has gradually become a research hotspot in the field of urban transportation.In recent years,reinforcement learning algorithm has made remarkable progress and has been applied in various fields,including taxi industry.In the taxi field,reinforcement learning algorithms have been used to optimize taxi dispatch systems,passenger allocation and pricing strategies.In addition,reinforcement learning algorithms are also used to improve passenger allocation strategies by learning passenger preferences and travel patterns,and to optimize pricing strategies by considering factors such as passenger demand,traffic congestion and driver availability.However,the application of reinforcement learning algorithm in the optimization of taxi drivers’ order taking strategy focuses more on predicting or guiding taxi drivers’ subsequent order taking behavior,which often leads to the algorithm’s excessive pursuit of maximizing target benefits and ignoring taxi drivers’ own order taking habits.Based on this,this paper analyzes taxi drivers’ ordering habits by virtue of Advantage Actor Critic algorithm in reinforcement learning.Subsequently,based on the analysis results and with the help of the Generative Adversarial Imitation Learning algorithm in imitation learning,the order receiving strategies of high-income expert taxi drivers are simulated and applied to the order receiving models of taxi drivers.Aiming at the optimization problem of urban taxi drivers’ order taking strategy,this study aims to improve the efficiency and income level of middle-and low-income taxi drivers’ order taking strategy,and explores the solution method of urban taxi drivers’ order taking optimization problem from the direction of artificial intelligence.In this paper,historical taxi orders in New York City in 2016 and 2021 were preprocessed using the python programming language data processing library to obtain the required order data.Then analyze the relationship between order time,order start and finish point and order duration and taxi driver income from the processed historical order data.Based on the above correlation analysis,an improved analysis model of taxi drivers’ ordering habit based on Advantage Actor-Critic(A2C)algorithm is further proposed in this paper.This model is based on the idea of interactive learning between the agent and the environment in the reinforcement learning mechanism.The agent is a taxi driver who has the ability and specific habits of taking orders,while the environment is a large number of historical data of taxi orders.By designing state space,action space and reward function,the taxi driver’s order receiving process is transformed into Markov decision process.Different from traditional habit analysis methods,based on the Critic network’s evaluation value of different order receiving habits in the model,this paper analyzed the influence degree of each order receiving habit on the revenue efficiency of taxi drivers at different time periods.On the basis of the above evaluation of ordering habits,the driver’s ordering habits were combined with the environment state vector,converted into feature vectors and added into the research on the optimization of taxi driver ordering strategies,and the optimization model of taxi driver ordering strategies based on the generated adversarial imitation learning algorithm was constructed.The model is iteratively optimized by imitating and learning the ordering strategy of expert taxi drivers.Meanwhile,the behavioral cloning algorithm(BC)and maximum entropy inverse reinforcement learning algorithm(MAXEnt)are also used to construct the comparison model.The test comparison results show that the strategy optimization model based on GAIL algorithm has higher imitation accuracy and learning efficiency.Moreover,the model has a high similarity with expert data in income level,order duration and passenger point distribution.Therefore,it can be concluded that the model has a certain practical guiding significance for optimizing the ordering strategy of middle and low income taxi drivers.