Front-end speech processing system with SVM algorithm in rail passenger flow management

Abstract

In order to achieve the goal of dynamically adjusting daily passenger flow to effectively control the overall efficiency of the transportation system, this study constructs a real-time monitoring and prediction system for subway passenger flow based on front-end voice processing technology and support vector machine models. The study first conducted a railway passenger flow analysis, and then used a support vector machine model to construct a preliminary prediction system. In order to achieve global optimization, the study also introduced particle swarm optimization algorithm to construct an optimization prediction model based on PSO-SVM. The results show that the proposed PSO-SVM method has undergone 48 iterations of training, and the predicted values closely match the actual passenger flow curve. The maximum RE error is 2%, and the overall prediction error is 98%. The decision coefficient of PSO-SVM is 0.998932. Therefore, this indicates that it has high performance and feasibility in predicting and controlling passenger flow during peak hours of urban rail transit.

Keywords

Urban rail transit passenger flow support vector machine genetic algorithm particle swarm

1. Introduction

In recent years, major cities in China have been in a stage of market transformation and rapid spatial structure adjustment. Its agglomeration effect is becoming increasingly prominent, with surrounding resources constantly gathering outward, and the concentration of economic factors in the city continues to increase [1, 2]. The urban economy is developing rapidly, but the problems of excessive urban population, excessive carrying capacity, and traffic congestion are becoming increasingly prominent [3]. Traffic congestion not only affects the economic development of cities, but also greatly reduces their economic activities. This wastes a lot of time and energy, increases pollution, and hinders social development and progress [4]. Developing rail transit has become a basic national policy, which can meet transportation needs, save urban resources, and change the urban landscape. Rail transit is the backbone of urban comprehensive transportation system, an important component of urban socio-economic mobility and industrial structure adjustment, and an effective way to promote urban socio-economic operation. Therefore, the development of urban rail transit is imminent [5]. Based on this, the study aims to propose a method based on a front-end speech processing system with SVM algorithm in rail traffic passenger flow management. The study is mainly based on the application of front-end speech processing system and SVM algorithm in rail traffic passenger flow management to analyse its performance.

2. Related works

In recent years, the rapid development of urbanization has led to an increasing number of people relying on public transportation, which has led to an increasing demand for accurate passenger flow forecasting in transportation systems. Accurate prediction of passenger flow is important for optimizing the operation of transportation systems, improving passenger experience, and ensuring the safety of passengers. Therefore, researchers have been exploring various prediction methods and models to improve the accuracy of passenger flow forecasting. Luo et al. [6] proposed a spatiotemporal hash multi graph convolutional network that constructs two types of subgraphs from the perspectives of physical adjacency and semantic similarity to clearly capture the spatiotemporal dependencies between bus stops/routes. The results show that the model can provide route recommendations for congestion perception. Scholars such as He et al. [7] proposed an innovative deep learning method, multi graph convolutional regression neural network, to predict the passenger flow of urban rail transit systems. The experimental results show that the algorithm outperforms the benchmark algorithm in terms of prediction accuracy. Wang et al. [8] proposed a dynamic spatiotemporal hypergraph neural network for passenger flow prediction. In the prediction framework, the main hypergraph is constructed from the topology of the subway system, and then extended with advanced hyperedges discovered from pedestrian travel patterns across multiple time spans. In addition, hypergraph convolution and spatiotemporal blocks are proposed to extract spatiotemporal features and achieve node level prediction. Li et al. [9] focused on forecasting methods based on subway passenger data and proposed practical techniques to improve the accuracy of long-range and short-range forecasts. They emphasized the importance of avoiding oversimplification and ensuring accuracy.

Liu et al. [10] used decision tree based models to model and predict passenger flow in order to better understand the role of different views. Then, the defects and main features of the decision tree based model were analyzed. The analysis results can assist in the architecture design of deep learning networks. Tan and other scholars [11] proposed a new aviation passenger flow prediction model using factor decomposition machines and deep neural networks to capture the complex feature interactions between passenger flow influencing features. The results show that the proposed model can achieve satisfactory prediction results and effectively improve air passenger flow within an acceptable range of deviation from the initial train schedule. Zhang et al. [12] and other scholars proposed a deep learning architecture that combines residual networks, graphical convolutional networks, and short-term and short-term memory for short-term passenger flow prediction. The comparison of prediction accuracy obtained at different time granularities shows that prediction accuracy increases with the increase of time granularity. This indicates that the model can provide insight into short-term passenger flow prediction for subway operators. Yang et al. [13] proposed an improved model based on short-term and short-term memory neural networks to enhance long-term features for predicting explosive passenger flow. The results show that the model overcomes its limitations of insufficient long-term correlation learning caused by time lag. Wu et al. [14] proposed a new scale superposition gradient advancing decision tree model to predict bus passenger flow from multiple sources of data. And a scaling stacking method was designed by introducing a quasi-attention based mechanism. The results show that the model not only exhibits advantages in prediction accuracy and stability, but also can better handle the multicollinearity problem of multi-source data. Du et al. [15] proposed a deep irregular convolutional residual LSTM network model for predicting urban traffic passenger flow and proposed a deep learning framework that integrates irregular convolutional residential networks and LSTM units to learn spatiotemporal feature representations.

In summary, researchers have proposed various methods and models for predicting passenger flow in transportation systems. These methods take into account the time-space correlations between multiple sites, analyze the factors that affect passenger flow, and use neural networks and clustering models to improve prediction accuracy. These efforts have helped optimize the operation of transportation systems, improve passenger experience, and ensure the safety of passengers. It can be seen through the research of domestic and international scholars based on rail traffic passenger flow management methods that there are more studies on rail traffic passenger flow management methods. However, in urban rail traffic flow management, there are few researches using SVM method. This paper focuses on the combination of support vector machine and PSO to achieve effective control of rail transit passenger flow, so that the front-end speech processing system can perform its passenger flow. This study focuses on the fusion of SVM algorithm and PSO algorithm for rail passenger flow management, so that the front-end speech processing system can broadcast the passenger flow, thus improving the recognition rate compared to a single algorithm and playing a crucial role in relieving the passenger flow of urban railways during peak periods.

Figure 1.

Factors affecting passenger flow.

3. Model building based on front-end speech processing system and SVM algorithm in rail passenger flow management

3.1 Rail passenger flow analysis in rail passenger flow management

Since subway passenger flow is a complex and dynamic system, it is affected by many factors. The goal of this article is to manage it effectively. Build a reasonable passenger flow management model, it is necessary to analyze it, find out its basic changing laws, and propose corresponding forecasting methods. The generation, development and change of rail transit passenger flow, as well as the comprehensive effect of various factors such as fares and rail transit services [16], are shown in Fig. 1.

Urban development and land use are both strongly related to the increase in population. Urban function planning can guide the land development in the surrounding area, thereby promoting the spatial distribution of the city and attracting the population effectively. The size of the urban population has a great influence on the traffic flow of the city, and the economic development of the city mainly depends on the number of trips of the residents. Because the railway originally had only a few lines, it has a relatively low level of urban coverage and is not very attractive to the city’s passenger flow. As the size and coverage of the rail network increases, so does its attractiveness to passengers, and the final passenger flow situation depends on the dynamic balance between transport demand and transport supply. In the event of insufficient rail supply, i.e. insufficient capacity, this will lead to longer intervals between trains and a shift to other modes of transport, thus constraining the growth of passenger traffic on the metro. In the case of special weather, holidays, etc. This affects passenger flow across the rail network. Because of New Year’s Day, rainy June, and middle school exams, the number of tourists in February every year decreases. The daily flow change refers to the rail traffic flow in each period of the day, as can be seen in Fig. 2, the change in daily commuter flow of the Chengdu metro will occur during the commuter morning peak 07:00–09:00 and the commuter evening peak 17:00–19:00 during the day, with the morning and evening peak hour inbound traffic accounting for 39% of the total daily inbound traffic [17].

Figure 2.

Incoming passenger flow and hourly coefficient of inbound passenger flow in a certain working day of Chengdu Metro.

Weekly traffic variation refers to the flow of passengers for each day of the week, as commuting to work and school results in differences between weekday and weekend traffic. The distribution of traffic during all-day periods on weekdays is characterised by a double-peak or single-peak pattern, while at double-peak stations the distribution of traffic is more dispersed, with morning peak hours often later than on weekdays. At stations with a high concentration of commuter and school-going passengers, weekday traffic is relatively high and falls significantly on weekends, while at stations near scenic and commercial centres, traffic flows increase more rapidly on weekends than on weekdays, as shown in Fig. 3.

Figure 3.

Passenger flow distribution within a week.

Based on the passenger flow pattern, the short-term passenger flow of the subway was predicted and reasonable conclusions were drawn. Firstly, its randomness. Due to factors such as land use conditions, population size, road network structure, and transportation capacity around the city, the subway passenger flow at each time period has a certain degree of randomness. The second characteristic is periodicity. The passenger flow of rail transit has significant cyclical characteristics. Without considering unexpected events such as holidays, the passenger flow during each time period shows periodic changes within a week. Within a day, due to differences in passenger flow time characteristics, the passenger flow continues to change during each time period [9].

3.2 Prediction model building based on front-end speech processing system with SVM algorithm

Forecasting is to make predictions about the future condition and development trend of something based on the analysis of existing objective things. In order to obtain the correct passenger flow, forecast and organize it effectively, the current passenger flow forecasting methods are studied. Make reasonable forecasts for passenger flow. Construction of the corresponding forecast model. Support Vector Machines (SVM) models have a solid theoretical foundation and can be well adapted to complex environments. Firstly, it can be used for smaller samples, and the performance of the model is better when the initial data is small. This method ensures the good performance of the model itself and provides continuous approximate estimates for estimation and prediction. The SVM model can achieve the best results even under small sample conditions. Therefore, the support vector machine model can be applied to short-term passenger flow prediction of urban rail transit, and the selection of SVM models has certain application value. The support vector machine model is a very practical method. For the recommended set of sample data $\{(x_{i},y_{i})|i=1,2,\ldots,k\}$ , (where $x_{i}$ is the input value and $y_{i}$ is the output value), let it obey the function $y=f(x)$ . Firstly, the function $g(x)=(w\cdot x)+b$ is fitted to the sample data set and the distance between the function $f$ and the function $g$ is minimised, i.e. the loss function $R(f,g)=\int{L(f,g)}dx$ , and based on the structural risk minimisation principle, the study gives a formula for the non-linear support vector [18], as shown in Eq. (1).

$\displaystyle\left\{{\begin{array}[]{l}\min\left[\frac{1}{2}\left\|w\right\|^{% 2}+C\sum\limits_{i=1}^{k}{(\xi_{i}+\xi_{i}^{\ast})}\right],\\ y_{i}-(w\cdot x)-b\leqslant\varepsilon+\xi_{i},\\ (w\cdot x)+b-y_{i}\leqslant\varepsilon+\xi_{i}^{\ast},\\ \xi_{i},\xi_{i}^{\ast}\geqslant 0,\\ \end{array}}\right.$ (1)

In Eq. (1), $x$ represents the instance feature vector, used as input to the training model, which is the input sample sequence of subway passenger flow. $w$ represents weight, which refers to the variables corresponding to each input feature. These variables are used to adjust the prediction results of the model. Each variable has a specific numerical value, representing the importance of that feature in the model’s prediction results. $\varepsilon>0$ is the accuracy of its fit; $\xi_{i}$ is set above the target beyond the $\varepsilon$ part; $\xi_{i}^{\ast}$ is set below the target beyond the $\varepsilon$ part; and the constant $C>0$ represents the penalty factor. $b$ represents biases, which are used to adjust the position of decision boundaries. $y$ represents the instance labels, used to indicate the category to which each instance belongs. $\lambda$ represents the penalty factor, used to adjust the model complexity; $C$ represents the regularization parameter, used to control the complexity of the model; $\alpha$ represents the Lagrange multiplier, used to represent the effect of each sample on the results. $\xi$ represents the error tolerance, used to indicate the maximum proportion of classification errors allowed; The Lagrangian method is used to transform the problem into Eq. (2).

$\displaystyle\left\{{\begin{array}[]{l}\max\left[-\frac{1}{2}\sum\limits_{i,j=% 1}^{k}{(a_{i}-a_{i}^{\ast})}(a_{j}-a_{j}^{\ast})(x_{i}\cdot x_{j})-\varepsilon% \sum\limits_{i=1}^{k}{(a_{i}+a_{i}^{\ast})}+\sum\limits_{i=1}^{k}{y_{i}(a_{i}^% {\ast}-a_{i})}\right],\\ \sum\limits_{i=1}^{k}{(a_{i}-a_{i}^{\ast})}=0,\\ a_{i},a_{i}^{\ast}\in[0,C]\\ \end{array}}\right.$ (2)

In Eq. (2), $a_{i}$ and $a_{i}^{\ast}$ are Lagrange factors, which are transformed into a higher dimensional space using the kernel function $K(x_{i},x_{j})$ to find the regression function, as shown in Eq. (3).

$\displaystyle f(x)=(w\cdot x)+b=\sum\limits_{i=1}^{k}{(a_{i}^{\ast}-a_{i})K(x_% {i}\cdot x_{j})+b^{\ast}}$ (3)

When using support vector machines, the choice of kernel function is critical in the calculation. The commonly used kernel functions are linear kernel function, polynomial kernel function, radial basis kernel function and sigma kernel function. These are shown in Eqs (4)–(7).

$\displaystyle K(x,x^{\prime})=x\cdot x^{\prime}$ (4) $\displaystyle K(x,x^{\prime})=(x\cdot x^{\prime}+c)^{d},c\geqslant 0$ (5) $\displaystyle K(x,x^{\prime})=\exp(-\|{x-x^{\prime}}\|^{2}/\sigma^{2})$ (6) $\displaystyle K(x,x^{\prime})=\tan(\gamma(x\cdot x^{\prime})+c)$ (7)

In Eqs (4)–(7), the meaning of $c$ represents a normal number greater than zero, $\gamma$ represents the scaling parameter of the kernel function, used to adjust the degree of influence of the kernel function. $K$ represents the kernel function, used to map the input instances into a higher dimensional space. When choosing a kernel function, it is important to start from a practical problem. Current research shows that the radial basis kernel function can achieve a non-linear mapping between the input and higher dimensions, thus solving non-linear problems better, so it is extremely widely used in SVM models. According to the short-term passenger flow characteristics of SVM, the radial basis kernel function based on support vector machine is selected as the kernel function of support vector machine. The linear $\varepsilon$ insensitive linear loss function is usually chosen to measure its value [19], and its mathematical expression is shown in Eq. (8).

$\displaystyle L_{\varepsilon}(|{y-f(x,a)}|)=|{y-f(x,a)}|_{\varepsilon}=\max(0,% |{y-f(x,a)}|-\varepsilon)$ (8)

In Eq. (8), the size of the $\varepsilon$ insensitive loss function determines the number of support vectors. As the $\varepsilon$ insensitive loss function increases, the accuracy of the model decreases. Conversely, as the $\varepsilon$ insensitive loss function decreases, increasing the number of support vectors can improve the prediction accuracy of the model, but it also increases the computation time, so a suitable $\varepsilon$ should be selected to ensure the speed and accuracy of the support vector machine. There are many methods for determining parameters, among which the cross determination method includes dividing the sample information into K groups. Test the remaining samples, then train and validate them repeatedly until all samples become test samples. This results in higher computational costs, higher sample requirements, and difficulty in practical application. The empirical method, on the other hand, is determined by an a priori equation between the parameters and the samples, which must be derived after several trials.

3.3 Research on improved SVM algorithms for forecasting methods in rail passenger flow management

While the traditional SVM algorithm is prone to local optimisation when performing parameter optimisation, the particle swarm algorithm (PSO) can, to a certain extent, avoid getting stuck in a local extremum problem and thus obtain an optimal combination of parameters. PSO solves the optimisation problem by simulating the foraging behaviour of birds, with a fitness function in the particle swarm that measures the merit of the particles in the overall solution space [20]. So in the whole particle swarm in the solution space there are two optimal extremes, one for the particle itself $p_{i}=(p_{i1},p_{i2},\ldots,p_{iD})$ and the other for the current particle swarm $p_{g}=(p_{g1},p_{g2},\ldots,p_{gD})$ . After these two extremes are found, according to Eqs (9) and (10), by comparing the adaptability of the optimal particle with the optimal solution, the velocity and position of the particle are updated in real time.

$\displaystyle v_{id}^{t+1}=v_{id}^{t}+c_{1}r_{1}(p_{id}^{t}-x_{id}^{t})+c_{2}r% _{2}(p_{gd}^{t}-x_{id}^{t})$ (9) $\displaystyle x_{id}^{t+1}=x_{id}^{t}+v_{id}^{t+1}$ (10)

In Eqs (9) and (10), $d=1,2,\ldots,D$ , $i=1,2,\ldots,n$ , and $n$ are the sizes of their populations, $t$ the current number of iterations of evolution, $c_{1}$ and $c_{2}$ represent the constant coefficients, and $r_{1}$ and $r_{2}$ represent the number of randoms distributed in $[0,1]$ . $v_{id}^{t}$ represents the velocity of the i-th particle at moment t. $p_{id}^{t}$ represents the optimal position of the ith particle in the history, generally the particle with the optimal fitness value. The study uses the PSO method to optimise the parameters of the SVM model, i.e. for an improved support vector machine model (PSO-SVM), to make it more compatible with the actual situation. On this basis, the short-term passenger flow is solved by using the PSO-SVM method.

Figure 4.

PSO-SVM algorithm flow chart.

As shown in Fig. 4, the input of the specimen data and the output of the specimen data are first decided. Then, based on this, the samples are pre-processed, a suitable kernel function is selected, a support vector machine model is constructed, and then the parameters are optimised using the PSO optimisation algorithm to continuously update the state of the particles using the accuracy of sampling as the adaptive degree until the final destination is reached. Finally, the regression model is adjusted for the parameters and is predicted and analysed. Where the pre-processing is done using i.e. data normalisation, the normalisation formula is shown in Eq. (11).

$\displaystyle X_{i}^{\prime}=\frac{X_{i}-X_{\min}}{X_{\max}-X_{\min}}$ (11)

In Eq. (11), $X_{i}^{\prime}$ is the normalised value of the data, $x_{i}$ is the sample data; $X_{\max}$ is the maximum value of the sample data and $X_{\min}$ is the minimum sampled data. From the above results, it can be seen that the genetic algorithm is a stochastic search optimisation method that is not limited by a specific model, which can iterate and update chromosomes to obtain the best parameter solution and shows good performance in global search. In this process, each adaptation function needs to be calculated and the individuals in the network are selected according to certain probability values, which are shown in Eq. (12).

$\displaystyle P_{i}=f(i)\left/\sum\limits_{i=1}^{n}{f(i)}\right.$ (12)

In Eq. (12), $f(i)$ is the fitness of individual i. The penalty factor $C$ , the proportion of the insensitive loss function $\varepsilon$ and the kernel parameters in the support vector machine are taken and optimised. The improved GA-SVM model has the following steps, as shown in Fig. 5.

Figure 5.

GA-SVM algorithm flow chart.

As shown in Fig. 5, GA-SVM is a classifier that combines Genetic Algorithm and Support Vector Machine to solve data classification problems. The following is the step-by-step description of the GA-SVM improvement algorithm: first, data preparation: collect and sort the data into training and test sets; second, feature selection: pre-process the data through feature selection techniques to reduce noise and redundant features and improve the performance and efficiency of the classifier; third, GA parameter setting: set the parameters of the GA algorithm, including population size, crossover rate, variation rate, etc. These parameters will affect the performance of the GA algorithm. These parameters will affect the search space and search efficiency of the GA algorithm; fourth, initialize the population: generate the initial population using random numbers, where each individual represents a set of parameters of the SVM model, such as penalty factor and kernel function type; fifth, evaluate the fitness: apply the SVM model of each individual to the training set and calculate its prediction accuracy as the fitness value; sixth, select operation: use the selection operator (e.g. roulette selection) to select a subset of individuals with high fitness from the population as parents of the next generation population; seventh, crossover operation: use the crossover operator (e.g. single-point crossover) to crossover the individuals of the parents to generate new offspring individuals; eighth, mutation operation: use the mutation operator (e.g. random mutation) to mutate the individuals of the offspring to increase the diversity of the population; ninth. Evaluation of fitness: the SVM model of the newly generated offspring individuals is applied to the training set and its prediction accuracy is calculated as the fitness value; tenth, selection of the best individual: the individual with the highest fitness among all individuals is selected as the final SVM model; eleventh, model testing: the final SVM model is applied to the test set and its prediction accuracy and other performance metrics are calculated; twelfth, optimisation of Parameters: The parameters of the SVM model are adjusted based on the test results to further optimize the performance of the classifier. Based on this, in this paper, mean absolute error, mean square error, root mean square error, etc. are used as evaluation indicators, and their expressions are as follows Eqs (13)–(15).

$\displaystyle\textit{MAE}=\frac{1}{{n}^{\prime}}\sum\limits_{i=1}^{n}{|y_{i}}-% \hat{y}_{i}|$ (13) $\displaystyle\textit{MSE}=\frac{1}{{n}^{\prime}}\sum\limits_{i=1}^{n}{(y_{i}-% \hat{y}_{i})^{2}}$ (14) $\displaystyle\textit{RMSE}=\sqrt{\frac{1}{{n}^{\prime}}\sum\limits_{i=1}^{n}{(% y_{i}-\hat{y}_{i}})^{2}}$ (15)

In Eqs (13)–(15), the mean absolute error MAE (Mean Absolute Error), mean absolute percentage error MAPE (Mean Absolute Percentage Error), mean square error MSE (Mean Square Error), root mean square error RMSE (Root Mean Square Error) were selected as evaluation indicators as evaluation indicators. Among them, the $\hat{y}_{i}$ is the predicted value of $i$ , $y_{i}$ is the true value of $i$ , $\bar{y}$ is the mean value of the series, and ${n}^{\prime}$ is the number of predicted points. After the forecast of the passenger flow, in order to maintain the passenger flow of the venue, the method of voice broadcasting is adopted to control the passenger flow. The voice announcement architecture used in this study consists of three modules: monitoring and collecting front-end, central management platform and intelligent voice platform. The main role of the monitoring front-end is to monitor and collect signals, mainly for frequency identification and recording. With the central monitoring platform as the control core, including equipment management, business management and artificial intelligence recognition, the system has various applications such as speech recognition, semantic understanding, speech recognition and speech translation.

4. Performance analysis of front-end speech processing system with SVM algorithm in rail passenger flow management

4.1 Analysis of comparison results of accuracy of different passenger flow prediction algorithms

The study selected the daily and monthly passenger flow at Chunxi Road Station of Chengdu Metro as well as Southwest University of Finance and Economics Station as the subjects of the study. For the daily passenger flow, 18 time points from 6:00–24:00 on 20 May 2021 were taken, while for the monthly passenger flow analysis, the passenger flow data during September was selected and pre-processed to normalise the monthly data, and the before and after pre-processing comparison is shown in Fig. 6.

Figure 6.

Comparison of passenger flow data before and after normalization.

From Fig. 6a, it can be seen that the value range of daily passenger flow is between 14000 and 31000, with a large data span. The distribution of passenger flow during the observation month shows certain fluctuations, with fluctuations ranging from 1000 to 10000 on alternate days, indicating significant fluctuations. Therefore, normalization is necessary. As shown in Fig. 6b, after normalizing its passenger flow data in September, the obtained data image is smoother, and the standardized data sets are concentrated in [0, 1], which makes its stability as well as feasibility higher after importing it into the model, making the passenger flow data input to the model more convincing. After importing their data into the established model as input data, it can be seen from Fig. 7 that the SVM model constructed by the study reached the predetermined error after 184 training sessions, the GA-SVM model reached the predetermined accuracy after 148 training sessions, while the PSO-SVM model reached the prediction accuracy after 48 training sessions, it can be seen that the improved SVM method, namely PSO-SVM, has higher learning efficiency and lower error rate.

Figure 7.

Training effect comparison chart.

From Fig. 7, it can be seen that the root mean square error curves of the three passenger flow prediction algorithms, including SVM, GA-SVM, and PSO-SVM, all show a downward trend and can converge to 10^{- 4}. However, the PSO-SVM prediction algorithm proposed by the research institute has a faster descent speed and smaller fluctuations during the descent process, indicating that the convergence performance of the algorithm is better.

4.2 Analysis of the practical application effect of passenger flow prediction model

The hourly passenger flow at for 18 time points between 6:00 and 24:00 on 20 May 2021 was used as an example to conduct a simulation analysis using the model developed by the study, and thus to determine the accuracy of its model predictions. In Fig. 8, its horizontal coordinate represents the 18 moments, i.e. between 6:00 and 24:00, while the vertical axis is the incoming and outgoing passenger flow. TRUE is the actual number of incoming and outgoing passengers on that day. The study used a modified GA-SVM method to fit the passenger flow expectations.

Figure 8.

Comparison of the effects of different algorithms in predicting passenger flow at the time of day.

As can be seen from Fig. 8, the prediction results obtained by the improved PSO-SVM algorithm and the GA-SVM model proposed by the study are in good agreement with the actual passenger flow curve, but its passenger flow prediction from 8:00–9:00 is 225 passengers. While its actual passenger flow is 997 passengers, with a large prediction error, while the deviation of passenger flow prediction in other time periods is also more obvious, so its model In summary, The results show that the PSO-SVM method has higher prediction accuracy than the first two methods, and has achieved good results in engineering practice.

Figure 9.

The relative error of the test sample of inbound passenger flow in the morning peak.

As shown in Fig. 9, The RE ratio of the SVM model is compared with the PSO-SVM and GA-SVM models to predict the passenger flow of Chunxi Road Station. The results show that the support vector machine has the largest prediction error with a maximum deviation of 7%. The maximum error of the PSO-SVM model is 2%, and the overall prediction deviation is below 98%, which shows that the PSO-SVM prediction results show that the PSO-SVM can better predict the passenger flow of Chunxi Road in September. There are also some good prophecies.

Table 1

The calculation results of each model evaluation index

Error	PSO-SVM	GA-SVM	SVM	Graph convolution spatiotemporal correlation algorithm	Dynamic adaptive graph algorithm
Mean absolute error MAE	1.01024	1.82613	7.15654	1.26024	1.29024
Mean Squared Error MSE	0.98313	1.23852	8.56133	1.30313	1.43313
Coefficient of determination $R^{2}$	99.8932	89.1561	75.6162	98.0332	97.8832

It can be seen from Table 1 that the prediction model of the established PSO-SVM has a value of 0.998932, of which the prediction mode of MAE and PSO-SVM is 1.01024%. The mean square error values of the graph convolution spatiotemporal correlation algorithm and the dynamic adaptive graph algorithm are both higher than those of the PSO-SVM algorithm proposed by the research institute, with improvements of 0.32 and 0.45, respectively. In addition, in terms of determination coefficients, both algorithms are lower than the prediction model proposed by the research institute, with reductions of 1.86 and 2.01, respectively [21, 22]. Compared with traditional support vector machine and support vector machine model, PSO-SVM has higher prediction accuracy and stronger prediction ability for passengers. The research results show that the PSO-SVM prediction has a high accuracy rate and has a good effect on the forecast of inbound passengers. The combined PSO-SVM algorithm used in this study can effectively compensate for the pain points of both algorithms and has extremely high performance, with all evaluation indicators remaining at the best level. Figure 10 gives a comparison between the algorithm proposed in this study and two other literature algorithms by counting 147 synchronizations and comparing their energy consumption.

Figure 10.

Variation in energy consumption and number of data sets.

As shown in Fig. 10, the algorithm proposed in the study consumes less energy than the other two algorithms, which are 38.52% and 60.97% lower than the other two algorithms, respectively. The algorithm proposed in the study requires much less energy than the other two methods because it uses the PSO-SVM algorithm for multiple training, so in the 2nd stage of the synchronisation process, the algorithm in the literature [17] requires a The algorithm of [17] therefore requires a bi-directional packet exchange between the benchmark node and the regular node during phase 2 of the synchronisation process in order to achieve a higher accuracy rate. The algorithm of [16] has to increase the number of linear fits in order to increase the accuracy, which causes a large amount of energy loss, so in summary, the algorithm proposed in the study is applied well and the accuracy achieves the expected results.

5. Conclusion

China’s railway construction started late, but its development speed is very fast. The passenger flow prediction of the railway passenger dedicated line is an important part of the railway passenger transport business. Therefore, this paper proposes a passenger dedicated passenger dedicated line based on the railway passenger dedicated line, rational scheduling and assessment of the comprehensive service level of the railway network. The study proposes a front-end speech processing system with SVM algorithm in the passenger flow management model of rail transport, and the particle swarm algorithm (PSO) as well as genetic algorithm (GA) are optimised successively to the support vector machine algorithm (SVM). The study shows that the passenger flow data is smoother after normalised pre-processing, and then the study compares the three algorithms, and the results show that while the PSO- SVM model was trained 48 times to achieve the prediction accuracy, which is significantly better than the other two algorithms, indicating that its algorithm performance is good and the training efficiency is extremely high. The overall prediction error is low and the prediction accuracy reaches over 98%, the research results show that the decision factor of support vector machine is 0.998932, which has similar characteristics with support vector machine and GA-SVM. The results show that the PSO-SVM and PSO-SVM hybrid algorithms have good correlation and can effectively solve the traffic congestion problem. Considering the cyclical nature of passenger flows, additional base data can be considered in subsequent work to observe and improve the performance of the model, and to study the applicability of the method for forecasting rail cross-sections, line passenger flows, etc. In conclusion, this study proposes a novel method that combines front-end voice processing technology and SVM algorithm to achieve real-time monitoring and forecasting of subway passenger flow, providing an effective way to control the flow of the system and improve its overall efficiency. The PSO-SVM method shows high performance and feasibility, with a maximum RE error of 2% and an overall prediction error of 98%. The proposed method provides a reference for future research in this area, which could potentially benefit the prediction and control of passenger flow in urban railways during peak periods. The study used an improved support vector machine model to predict subway passenger flow under normal conditions, but did not study sudden passenger flow situations such as major holidays and large-scale events. Therefore, accurate prediction of sudden passenger flow is expected to be analyzed in the future.

Footnotes

Data availability statement

The data used to support the findings of this study are within the article.

References

Hou

Dong

Gao

Nicholson

Chen

Roberts

. Energy-saving metro train timetable rescheduling model considering ATO profiles and dynamic passenger flow. IEEE T Intell Transp. 2019; 20(7): 2774-2785.

Yang

Zheng

Zhao

Chen

. Rail potential and stray current mitigation for urban rail transit with multiple trains under multiple conditions. IEEE T Transp Electr. 2021; 8(2): 1684-1694.

Xia

Wang

Kong

Zhang

Wang

. Ranking station importance with human mobility patterns using subway network datasets. IEEE T Intell Transp. 2019; 21(7): 2840-2852.

Jing

Guo

Wang

Chen

. Short-term prediction of urban rail transit passenger flow in external passenger transport hub based on LSTM-LGB-DRS. IEEE T Intell Transp. 2020; 22(7): 4611-4621.

Wang

Tang

Yang

. Event-triggered predictive control for automatic train regulation and passenger flow in metro rail systems. IEEE T Intell Transp. 2020; 23(3): 1782-1795.

Luo

Zhao

You

Liu

. Spatiotemporal hashing multigraph convolutional network for service-level passenger flow forecasting in bus transit systems. IEEE Internet Things. 2021; 9(9): 6803-6815.

Zhu

Tsui

. Multi-graph convolutional-recurrent neural network (MGC-RNN) for short-term forecasting of transit passenger flow. IEEE T Intell Transp. 2022; 23(10): 18155-18174.

Wang

Zhang

Wei

Piao

Yin

. Metro passenger flow prediction via dynamic hypergraph convolution networks. IEEE T Intell Transp. 2021; 22(12): 7891-7903.

Yan

Zhang

. Long-short term spatiotemporal tensor prediction for passenger flow profile. IEEE Robot Autom Let. 2020; 5(4): 5010-5017.

10.

Liu

Lyu

Liu

. Automatic feature engineering for bus passenger flow prediction based on modular convolutional neural network. IEEE T Intell Transp. 2020; 22(4): 2349-2358.

11.

Tan

Wang

Zheng

, et al. Improving synchronization in high-speed railway and air intermodality: Integrated train timetable rescheduling and passenger flow forecasting. IEEE T Intell Transp. 2022; 23(3): 2651-2667.

12.

Zhang

Chen

Cui

Guo

Zhu

. Deep learning architecture for short-term passenger flow forecasting in urban rail transit. IEEE T Intell Transp. 2020; 22(11): 7004-7014.

13.

Yang

Chen

Yang

Zhao

. Urban rail transit passenger flow forecast based on LSTM with enhanced long-term features. IET Intell Transp Sy. 2019; 13(10): 1475-1482.

14.

Xia

Jin

. Predicting bus passenger flow and prioritizing influential factors using multi-source data: Scaled stacking gradient boosting decision trees. IEEE T Intell Transp. 2020; 22(4): 2510-2523.

15.

Peng

Wang

Gong

Liu

. Deep irregular convolutional residual LSTM for urban traffic passenger flows prediction. IEEE T Intell Transp. 2019; 21(3): 972-985.

16.

Bešinović

De Donato

Flammini

. Artificial intelligence in railway transport: Taxonomy, regulations, and applications. IEEE T Intell Transp. 2021; 23(9): 14011-14024.

17.

Jiang

Xiao

Liu

. Passenger flow prediction for new line using region dividing and fuzzy boundary processing. IEEE T Fuzzy Syst. 2019; 27(5): 994-1007.

18.

Chen

Wang

. Subway passenger flow prediction for special events using smart card data. IEEE T Intell Transp. 2019; 21(3): 1109-1120.

19.

Zhu

Zhang

Liu

Cai

. Speed profile tracking by an adaptive controller for subway train based on neural network and PID algorithm. IEEE T Veh Technol. 2020; 69(10): 10656-10667.

20.

Wang

Ren

Takyi-Aninakwa

Jin

Fernandez

. A critical review of improved deep convolutional neural network for multi-timescale state prediction of lithium-ion batteries. Energies. 2022; 15(14): 5053.

21.

Ali

Zhu

Zakarya

. Exploiting dynamic spatio-temporal graph convolutional neural networks for citywide traffic flows prediction. Neural Networks. 2022; 145: 233-247.

22.

Huang

Xiao

. Multi-view clustering based on low-rank representation and adaptive graph learning. Neural Process Lett. 2022; 54(1): 265-283.