^{1}

^{1}

This paper chooses passenger flow data of some stations in China from January 2015 to March 2016, and the time series prediction model of BP neural network for railway passenger flow is established. But because of its slow convergence speed and easily falling into local optimal solution of the problem, we propose to improve the time series model of BP neural network by genetic algorithm to predict railway passenger flow. Experimental results show that the improved method has higher prediction accuracy and better nonlinear fitting ability.

With the Notice of the National Development and Reform Commission of the People’s Republic of China on the reform and improvement of the passenger fare policy for high-speed trains in February 2016, the high-speed train fares will be priced by the railway headquarters [

However, railway passenger flow is affected by many factors, such as the rapid growth of railway passenger flow during the Spring Festival, which leads to the inability of railway capacity to meet customer demand for passenger transport, and also brings great pressure to railway passenger transport organizations. During non-holiday period, some unpopular line occupancy rate is insufficient, resulting in the waste of railway vehicles. Therefore, the prediction of railway passenger flow is of great significance for improving the efficiency of railway passenger transport by helping to set reasonable prices, improving the organization mode of passenger stations, optimizing the allocation of railway vehicle resources and improving the service capacity of passenger transport equipment [

This paper aims to study the application of BP neural network and its improved algorithm in railway passenger flow prediction.

At present, railway travel in China is the first choice of transportation mode for the people. Experts hope to optimize vehicle dispatch and guide the development of the railway industry by analyzing the passenger flow data of a certain area for several consecutive years.

In recent years, Wu Xinhui put forward “by adopting genetic algorithm to optimize parameters of RBF neural network, a better solution to the RBF neural network is easy to fall into local optimum problem, in the railway passenger traffic forecasting, through the comparison shows that GA-RBFNN model to predict the results of the stability and convergence speed, high precision, good maneuverability, the railway enterprise management decision has a good guiding significance [

Based on the characteristics of genetic algorithm, this paper proposes a BP neural network time series algorithm based on GA, and establishes a railway passenger flow prediction model.

In general, the most common BP neural network topology is a three-layer structure (shown in

longer significantly reduced. Finally, the number of hidden layer nodes selected by the test is 6 [

Because the BP algorithm is a faster gradient descent algorithm, it is easy to fall into the local minimum problem and is improved by GA.

Genetic algorithm is a global search algorithm [

According to the above, it has been determined that the BP neural network structure is 4 − 6 − 1, a total of 6 + 1 = 7 thresholds, 4 × 6 + 6 × 1 = 30 weights. Since the genetic algorithm uses real coding, the length of each individual in the population is 30 + 7 = 37.

Random initialization of these individuals according to the real number coding method is equivalent to randomly initializing the initial threshold and weight of a batch of neural networks, and then calculating the fitness values of these

individuals, based on the fitness value, and selecting them one by one. The crossover and mutation operations evolve the population until the number of evolution iterations is reached, and the individuals with the highest fitness within the final population are used as the initial threshold and weight of the optimized BP neural network.

The training set is substituted into GA improved BP neural network training, and the test set is used for testing and simulation prediction.

The experiment selects the passenger flow data of several stations in a certain area of China from January 2015 to March 2016. Firstly, the passenger flow and the passenger load rate are analyzed statistically on the train type, train number, time zone and date, and the following conclusions are obtained:

1) The total number of trains at the beginning of G, D, and K accounts for about 90% of the total number of trains. High-speed trains and trains are the main types of trains (see

2) Except for T01, the average passenger load rate of trains starting with T is less than 20%. Therefore, it is necessary to consider whether the following vehicles can be used efficiently (see

3) The peak period of the flow of people within 24 hours a day is from 1 pm to 5 pm, indicating that the afternoon time is the peak of the flow of people. The relevant departments of the railway can conduct targeted traffic flow during peak hours and increase the number of service personnel at the station (see

4) The passenger load factor in summer is higher than that in spring, autumn and winter. The average daily passenger load factor in summer is over 70%. The main reason is that the summer vacation is the peak period for students to return home and travel, and the transfer of migrant workers can also increase the

passenger load factor to a certain extent. Holidays should also be given as an important factor in the train deployment program (see

In the process of genetic algorithm optimization, the fitness of the best individual in the population changes as shown in

than the actual output compared with the actual output value of the passenger flow, and the prediction result of the BP neural network prediction model improved by the genetic algorithm [

It can be seen from

The two peaks in

The simulation results show that the BP neural network prediction model based on the initial threshold and weight of BP neural network optimized by genetic algorithm is better than the standard BP neural network in predicting railway passenger flow, and the prediction result is stable. Especially for the prediction of certain holidays, it can greatly reduce the prediction error, has strong robustness, can better deal with the problem that BP neural network is easy to fall into local optimum, and improve the convergence speed of the model. Compared with the standard BP neural network prediction model, this model has higher prediction accuracy and better nonlinear fitting ability for railway passenger flow time series.

This work was supported by scientific research fund project in Beijing University of Civil Engineering and Architecture (KYJJ2017035).

The authors declare no conflicts of interest regarding the publication of this paper.

Zhang, J. and Guo, W.H. (2019) Research on Railway Passenger Flow Prediction Method Based on GA Improved BP Neural Network. Journal of Computer and Communications, 7, 283-292. https://doi.org/10.4236/jcc.2019.77023