CN110688982B - Intelligent rail transit time control method based on target detection technology and ACO-BP algorithm - Google Patents

Abstract

The invention requests to protect an intelligent rail transit time control method based on a target detection technology and an ACO-BP algorithm. Firstly, counting the daily pedestrian flow of each station of a third line of the track traffic for Chongqing by using a target detection technology, and counting to obtain the waiting number and the actual number of passengers getting on the train; secondly, taking the number of waiting people and the number of actual people who actually get on the train as input, and obtaining a relation weight of the influence of the pedestrian volume and the arrival time interval on the train carrying capacity based on the BP neural network; and further, taking the current pedestrian flow in each station as an input quantity, taking the relation weight as an influence parameter, transmitting the influence parameter into an ant colony algorithm to establish a mathematical model, and obtaining the maximum carrying capacity by continuously changing the departure interval. Meanwhile, the error between the expected carrying capacity and the actual carrying capacity is used as input, and the updating of the weight value and the optimization result are realized based on the learning function of a BP (Back propagation) neural network. The invention acquires the train time interval of the maximum carrying capacity by collecting the environmental information of the light rail station in real time and continuously updating the algorithm.

Description

Intelligent rail transit time control method based on target detection technology and ACO-BP algorithm

technical field

The invention belongs to the field of urban rail transit control, and specifically uses the current human flow and historical data in each station to match an optimal traffic plan through an algorithm. The system is an intelligent rail transit time control system based on target detection technology and ACO-BP algorithm.

Background technique

With the continuous expansion of the city, the number of cars has soared, and the roads are very congested during high commuting hours and key areas. Under the circumstance of huge traffic demand, urban rail transit is more and more popular because of its convenient, safe, fast, and high punctuality rate. But now the light rail is facing the same congestion problem. In some business districts, airports, stations and during peak hours, the subway and subway stations have a huge flow of people, which leads to long waiting times for passengers and is very crowded, and it is also prone to safety accidents.

At present, most subway stations use separate timings during peak and low-peak hours. For example, Wang Shouqin [1] took Guangzhou Metro Line 3 as an example. Monte Carlo modeling and simulation was used to obtain the optimal departure time interval between peak hours and off-peak hours. However, this method can only obtain certain simulation conditions. The optimal departure time interval cannot completely solve the problem of congestion; Literature [2] also pointed out that in order to ensure the safety of passengers and convenient transportation, it is necessary to set the stopping time of different subway stations in different periods according to the actual situation; Wang Lin [3], Zhang Tianyu [4] took Nanjing Metro Line 2, Beijing Metro Changping Line and Shanghai Urban Rail Transit Jinshan Line as examples, and simplified the problem into a mathematical model to optimize the timetable. The above algorithms are basically modeled by optimizing the timetable to alleviate the problem of congestion. The present invention will set the stop time of different subway stations in different periods by analyzing the actual situation, so as to realize an automatic adjustment.

In recent years, in order to solve the problem of congestion in subway stations, various regions have invested a lot of research here to find solutions. Liu Tao[5] and others put forward the concept of "paying for congestion" to reduce the equilibrium income level in the case of passenger flow, and improve the congestion degree by increasing the fare during peak hours. This method is effective to a certain extent, and can make some People who are not in a hurry do not take the subway during peak hours, but the effect needs to be further improved for the time period when going to school and after school. Based on this, taking the dynamic angle as the starting point, Wei Huayong [6] proposed an urban rail transit signal control system based on fuzzy control algorithm. The design method uses the variable structure PID fuzzy neural network control method to improve the control rules of urban rail transit signals, and carries out the hardware design of the urban rail transit signal control system in an embedded environment; Chen Zhijie [7] takes passenger flow as the main research line, based on A F C card swiping data, from the estimation of the destination of incoming passengers, the estimation of the parameters of the passengers entering and leaving the station and the transfer process, the estimation of the train and station load, and the control of the passenger flow to relieve congestion. This method is mainly to control the passenger flow, which can play a role in slowing down The congestion of subways and subway platforms can not relieve the congestion of subway stations and make passengers wait longer, which is contrary to the convenience and speed of subways.

To sum up, based on the target detection technology, considering that the ant colony algorithm has the characteristics of information distribution, dynamics, randomness and asynchrony in network routing, combined with the BP (back propagation) neural network algorithm, the present invention proposes a Intelligent rail transit time control method based on target detection technology and ACO-BP algorithm to alleviate the congestion problem of subway stations.

references

[1] Wang Shouqin. Research on the departure time interval of Guangzhou Metro Line 3 based on Monte Carlo modeling and simulation [D]. South China University of Technology: Wang Shouqin, 2010.

[2] Li Sirang, Luo Xuyang, Fu Rao. Analysis of the Best Stop Time of Beijing Subway During Peak Period——Taking the South Exit Station of Huixin West Street of Beijing Subway Line 10 as an Example [J]. Famous Teachers Online, 2016, (08): 11- 13.

[3] Wang Lin. Research on optimization of urban rail transit timetable based on passenger flow demand [D]. Southeast University: Wang Lin, 2015.

[4] Zhang Tianyu. Optimization model and algorithm of urban rail transit train schedule based on fairness and efficiency under dynamic passenger flow demand [D]. Beijing Jiaotong University: Zhang Tianyu, 2018.

[5] Liu Tao, Yao Hongchao. Research on Time-of-use Pricing Strategy of Beijing Subway [J]. Enterprise Reform and Management, 2017, (13): 207-215.

[6] Wei Huayong. Design and Analysis of Urban Rail Transit Signal Control System Based on Fuzzy Control Algorithm [J]. Journal of Xuchang University, 2018, (06): 73-76.

[7] Chen Zhijie. Real-time estimation model and control method of urban rail transit line load [D]. Beijing Jiaotong University: Chen Zhijie, 2018.

SUMMARY OF THE INVENTION

The present invention aims to solve the above problems of the prior art. An intelligent rail transit time control method based on target detection technology and ACO-BP algorithm is proposed to maximize the passenger volume of rail transit. The technical scheme of the present invention is as follows:

An intelligent rail transit time control method based on target detection technology and ACO-BP algorithm, comprising the following steps:

First, considering the unique detection performance of deep learning in the field of machine vision, the Faster-RCNN target detection method is used to extract features from images through deep learning technology, and then capture images of vehicles and pedestrians, and count the daily flow of people at each station of the track line. The number of people waiting for each shift and the actual number of vehicles are obtained by statistics; the number of cars and the interval time between vehicles are calculated in real time, and then passed into the BP neural network;

Secondly, taking the number of people waiting, the actual number of trains, and the actual departure interval of the vehicle as the input, based on the BP neural network, the weight of the relationship between the number of waiting people and the time interval of train departure to the actual number of trains is obtained;

Finally, take the current flow of people in each station as the input, and the relationship weight as the influence parameter is input to the ant colony algorithm to establish a mathematical model, and the maximum carrying capacity is obtained by continuously changing the departure interval. The error is input, based on the learning function of BP neural network, the update of weights is realized and the result is optimized;

作为传递函数The step of finding the relationship weight value by the BP neural network specifically includes: using a nonlinear transformation function

as a transfer function

Its activation function is:

w _ik learning weight, B is a constant, B _k is the offset to be learned, y _k represents the output, and I represents the number of input data;

In the initial establishment process, the weights are initialized to W ₁ and W ₂ , and the gradient descent principle is used in the weight update process;

At this time, the error between the output of the algorithm and the actual situation is expressed by the following formula

where i is the input element, k is the implicit element, j is the output element, X is the input quantity, Y is the output quantity, S _j is the actual carrying quantity, and Y _j is the maximum carrying quantity; O represents the number of outputs, corresponding to the input number of data;

According to the principle of the gradient descent method, the correction value of the weight is proportional to the error function as:

make:

W _kj represents the weight to be learned in the network, a represents a constant, and E represents an error function;

but:

Y _k represents the output of the current hidden layer, δ _k is the gradient change, and H represents the number of input data of the current hidden layer;

For bias B _j

make:

but:

ΔB _j = δ _kj (7)

From the above formula, the weight W and bias B of the back propagation from the output layer node j to the hidden layer can be obtained as follows:

B _j =B _j -a*δ _kj (9)

make:

The weight W and bias B of the back propagation from the hidden layer node k to the input layer are obtained as follows:

B _k =B _k -b* _δik (12)

b represents a constant, so as to realize the update of the weights from the hidden layer to the output layer

Among them, the output value of the hidden layer of the BP neural network is

Y _k =f(y _k ) (14)

The output value of the final output layer is

Y _j =f(y _j ) (16)

The final result is the actual carrying capacity of the train at each station, and y _j represents the intermediate output value;

The current flow of people in each station is used as the input, and the relationship weight is input as an influence parameter to the ant colony algorithm to establish a mathematical model, and the maximum carrying capacity is obtained by continuously changing the departure interval. The error is the input, and based on the learning function of the BP neural network, the weights are updated and the results are optimized, including:

Taking the current flow of people in each site as the input, and the relationship weight as the influencing parameter, it is input to the ant colony algorithm to establish a mathematical model. The basic framework of the ant colony algorithm here is to randomly assign the people to the car for the current flow of people. number of people, and leave the corresponding pheromone, and use these pheromone in the next iteration process. For different number of people getting on the bus, let it choose whether it is the maximum load with a specific probability distribution. The probability distribution is as follows:

表示当前人数量上车的概率，η表示一个启发值，τ为信息素浓度，J_k(r)表示不同上车人数的集合，δ和β表示τ和η的比重，δ越大，τ比重越大，反之亦然，通常设定δ＝1，β＝6；in

Indicates the probability of the current number of people getting on the bus, η represents a heuristic value, τ is the pheromone concentration, J _k (r) represents the set of different number of people getting on the bus, δ and β represent the proportion of τ and η, the larger the δ, the greater the proportion of τ The larger, and vice versa, usually set δ=1, β=6;

Adjust the relationship between the number of carriages when the train departs and the passenger flow of the entire line throughout the day, determine the optimal number of carriages, and finally use the ACO algorithm to intelligently match, and finally obtain the optimal value, obtain the optimal train time interval and the number of carriages, and make the train single Maximum sub-carriage.

Further, the BP neural network has three layers, including the input layer, namely the coding layer, the hidden layer and the output layer. The input layer has 3 neurons, the hidden layer has 6 neurons, and the output layer has 2 neurons. . The advantages and beneficial effects of the present invention are as follows:

(1) First, the deep learning technology Faster-RCNN is used to perform target detection, which can ensure the accuracy of detection and obtain accurate actual flow of people in different time periods; It can analyze and calculate according to historical data information, and dynamically adjust the running time of each station of the train. Compared with traditional control methods, it is more flexible and has stronger real-time performance.

(2) The algorithm based on the BP neural network can continuously learn the incoming data, and can quickly adapt to the sudden change of the current passenger flow in the light rail station by continuously updating the weight value, and timely feedback to the train, which has strong adaptability and achieves improvement. carrying capacity purpose.

(3) During the road operation, the intelligent detection systems in each light rail station can operate independently, with a small amount of calculation. The distributed storage and parallel collaborative processing of the ACO-BP algorithm can promote global optimization, produce superposition effects, and expand the optimization effect.

Through deep learning target detection technology, combined with traditional BP network and ant colony algorithm, scheduling is realized.

Description of drawings

Fig. 1 is the logical framework model of the preferred embodiment provided by the present invention

Figure 2: BP neural network structure

Figure 3: Simulated station train and pedestrian entrance

Figure 4: Simulated pedestrian waiting for a train

Figure 5: Train carrying effect

Detailed ways

The technical solutions in the embodiments of the present invention will be described clearly and in detail below with reference to the accompanying drawings in the embodiments of the present invention. The described embodiments are only some of the embodiments of the invention.

The technical scheme that the present invention solves the above-mentioned technical problems is:

At present, the traditional bus time interval setting is still used to solve the number of light rail carriages and the departure time interval. Although the light rail setting will be optimized to a certain extent according to its own actual situation, it is difficult to get the best results without the support of big data. Optimization Results. The present invention proposes an intelligent vehicle scheduling model based on the ACO-BP intelligent algorithm and target detection technology. The system collects data such as the flow of people and the number of passengers on each station of Chongqing Light Rail Line 3, and uses the AOC-BP algorithm at the same time. , combined with the actual number of train carriages and the vehicle interval time of the day to obtain the optimal configuration plan, which has the characteristics of strong feasibility and high efficiency.

The light rail train operation and deployment model based on target detection technology and ACO-BP algorithm is characterized in that the target detection technology is used to conduct real-time statistics on the number of passengers waiting and boarding at each light rail station. Secondly, by inputting the historical data of Rail Transit Line 3 into the BP neural network, the relationship weights of the influence of the station traffic and the departure time interval on the actual number of trains are obtained. The current station traffic information is used as the input, and the relationship weights are used as the influencing parameters. , get the optimal train arrival time interval, and at the same time pass the data information of the number of people who fail to get on the train to the ant colony algorithm to get the optimal number of carriages setting matching scheme, and realize the reasonable configuration of the departure interval time and the number of carriages of light rail trains, so that Rail transit capacity is efficiently and rationally maximized. The corresponding operation principle is shown in Figure 1. At the same time, the BP (Backpropagation) neural network can compare and analyze the theoretical carrying capacity and the actual carrying capacity, realize the update of the relationship weight and optimize the algorithm.

1. Target detection

First of all, through the target detection technology Faster-RCNN, through the deep learning technology to extract the features of the image, and then capture the images of moving vehicles and pedestrians, to achieve the purpose of target detection, to achieve the purpose of target detection.

2. BP neural network to find relationship weights

The BP (back propagation) neural network model is an intelligent algorithm that can obtain the optimal solution by continuously learning the input data and updating the weights.

Taking the Longtousi Light Rail Station of Chongqing Light Rail Line 3 as an example, taking the current number of people waiting for trains in the station, the actual number of trains and the actual departure interval of vehicles as the input quantities, and finally the number of people waiting for trains and the time interval between train departures and the actual number of trains are obtained respectively. The relationship weights are shown in Figure 2.

Therefore, we set the number of input layer nodes to 3, the number of output layer nodes to 2, and the set number of hidden layers to 6, and use the single-layer neural network model.

作为传递函数Using nonlinear transformation functions

as a transfer function

Its activation function is:

Where i is the input element, k is the hidden element, j is the output element, X is the input quantity, and Y is the output quantity.

In the initial establishment process, we initialize the weights to W ₁ and W ₂ , and use the gradient descent principle in the weight update process.

At this time, the error between the output of the algorithm and the actual situation is expressed by the following formula

S _j is the actual carrying capacity of the intersection, and Y _j is the maximum carrying capacity;

According to the principle of the gradient descent method, the correction value of the weight is proportional to the error function as:

make:

but:

For bias B _j

make:

but:

ΔB _j = δ _kj (7)

From the above formula, the weight W and bias B of the back propagation from the output layer node j to the hidden layer can be obtained as follows:

B _j =B _j -a*δ _kj (9)

make:

The weight W and bias B of the back propagation from the hidden layer node k to the input layer are obtained as follows:

B _k =B _k -b* _δik (12)

So as to realize the update of the weights from the hidden layer to the output layer

Among them, the output value of the hidden layer of the BP neural network is

Y _k = f(y _k ) (where f is the activation function) (14)

The output value of the final output layer is

Y _j =f(y _j ) (16)

The final result is the actual capacity of the train at each station.

3. Intelligent optimization of ant colony algorithm

The ACO-BP intelligent algorithm is used to solve the problem of the interval time between trains arriving at the station and the number of carriages. Because the same light rail station under different time lines changes in roughly the same amount of people throughout the day. The traditional ant colony algorithm cannot obtain the global optimal solution with reference to historical information. Through the introduction of the BP (back propagation) neural network algorithm, the weights W ₁ , W ₂ of the relationship between the train running time and the historical data information and the flow of people to the actual number of passengers in the train can be referred to, and the actual number of passengers waiting in the station is used as the input condition. , the relationship weight is the influence parameter to calculate the arrival time interval that makes the train carrying the highest efficiency, and the theoretical maximum carrying capacity is obtained.

At the same time, we regard the train carrying capacity and the number of train cars as a proportional relationship. Each car can carry 60 people. The target detection technology is used to detect the number of passengers who fail to get on the train at each station, and the data of the entire line is integrated to simulate the increase. Or reduce the influence of the carriages on the carrying efficiency of the entire line, to adjust the relationship between the number of carriages when the train departs and the passenger flow of the entire line throughout the day, and to determine the optimal number of carriages. Finally, the ACO algorithm is used for intelligent matching to obtain the optimal train time interval and the number of carriages, so as to maximize the single carrying capacity of the train. At the same time, the error E generated between the theoretical carrying capacity and the actual carrying capacity is transmitted to the neural network to further update the weights and optimize the algorithm.

The above embodiments should be understood as only for illustrating the present invention and not for limiting the protection scope of the present invention. After reading the contents of the description of the present invention, the skilled person can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims (2)

1. an intelligent rail transit time control method based on target detection technology and ACO-BP algorithm, is characterized in that, comprises the following steps: First, considering the unique detection performance of deep learning in the field of machine vision, the Faster-RCNN target detection method is used to extract features from images through deep learning technology, and then capture images of vehicles and pedestrians, and count the daily flow of people at each station of the track line. The number of people waiting for each shift and the actual number of vehicles are obtained by statistics; the number of cars and the interval time between vehicles are calculated in real time, and then passed into the BP neural network; Secondly, taking the number of people waiting, the actual number of trains, and the actual departure interval of the vehicle as the input, based on the BP neural network, the weight of the relationship between the number of waiting people and the time interval of train departure to the actual number of trains is obtained; Finally, take the current flow of people in each station as the input, and the relationship weight as the influence parameter is input to the ant colony algorithm to establish a mathematical model, and the maximum carrying capacity is obtained by continuously changing the departure interval. The error is input, based on the learning function of BP neural network, the update of weights is realized and the result is optimized; The step of finding the relationship weight value by the BP neural network specifically includes: using a nonlinear transformation function

as a transfer function Its activation function is:

w _ik learning weight, B is a constant, B _k is the offset to be learned, y _k represents the output, and I represents the number of input data; In the initial establishment process, the weights are initialized to W ₁ and W ₂ , and the gradient descent principle is used in the weight update process; At this time, the error between the output of the algorithm and the actual situation is expressed by the following formula

where i is the input element, k is the implicit element, j is the output element, X is the input quantity, Y is the output quantity, S _j is the actual carrying quantity, and Y _j is the maximum carrying quantity; O represents the number of outputs, corresponding to the input number of data; According to the principle of the gradient descent method, the correction value of the weight is proportional to the error function as:

make: W _kj represents the weight to be learned in the network, a represents a constant, and E represents an error function;

but:

Y _k represents the output of the current hidden layer, δ _k is the gradient change, and H represents the number of input data of the current hidden layer; For bias B _j

make:

but: ΔB _j = δ _kj (7) From the above formula, the weight W and bias B of the back propagation from the output layer node j to the hidden layer can be obtained as follows:

B _j =B _j -a*δ _kj (9) make:

The weight W and bias B of the back propagation from the hidden layer node k to the input layer are obtained as follows:

B _k =B _k -b* _δik (12) b represents a constant, so as to realize the update of the weights from the hidden layer to the output layer Among them, the output value of the hidden layer of the BP neural network is

Y _k =f(y _k ) (14) The output value of the final output layer is

Y _j =f(y _j ) (16) The final result is the actual carrying capacity of the train at each station, and y _j represents the intermediate output value; The current flow of people in each station is used as the input, and the relationship weight is input as an influence parameter to the ant colony algorithm to establish a mathematical model, and the maximum carrying capacity is obtained by continuously changing the departure interval. The error is the input, and based on the learning function of the BP neural network, the weights are updated and the results are optimized, including: Taking the current flow of people in each site as the input, and the relationship weight as the influencing parameter, it is input to the ant colony algorithm to establish a mathematical model. The basic framework of the ant colony algorithm here is to randomly assign the people to the car for the current flow of people. number of people, and leave the corresponding pheromone, and use these pheromone in the next iteration process. For different number of people getting on the bus, let it choose whether it is the maximum load with a specific probability distribution. The probability distribution is as follows:

in

Indicates the probability of the current number of people getting on the bus, η represents a heuristic value, τ is the pheromone concentration, J _k (r) represents the set of different number of people getting on the bus, δ and β represent the proportion of τ and η, the larger the δ, the greater the proportion of τ The larger, and vice versa, usually set δ=1, β=6; Adjust the relationship between the number of carriages when the train departs and the passenger flow of the entire line throughout the day, determine the optimal number of carriages, and finally use the ACO algorithm to intelligently match, and finally obtain the optimal value, obtain the optimal train time interval and the number of carriages, and make the train single Maximum sub-carriage. 2. a kind of intelligent rail transit time control method based on target detection technology and ACO-BP algorithm according to claim 1, is characterized in that, described BP neural network has three layers in all, including input layer namely coding layer, hidden layer The input layer has 3 neurons, the hidden layer has 6 neurons, and the output layer has 2 neurons.

Images