A subway station passenger flow buffer zone risk identification and collaborative management method and system

Abstract

The application discloses a subway station passenger flow buffer zone risk identification and collaborative management method and system, relates to the field of simulation, and comprises the following steps: based on the individual motion simulation software Massmotion of a social force model, a three-dimensional simulation scene of passenger flow buffer zones of each subzone of a subway station is constructed, and subway station passenger flow buffer zone data is acquired; a partition independent component analysis (PIPCA) algorithm is used to obtain dominant variables affecting passenger flow systems of each subzone; according to the dominant variables, the weight and bias parameters of a pre-constructed nonlinear autoregressive neural network (NARX) are optimized through a moss growth algorithm (MGO), and a subway passenger flow buffer zone risk identification model for predicting passenger flow densities of each buffer zone and a bottleneck period is obtained; a target function and a constraint condition are constructed, an optimal control strategy is selected under the constraint condition of meeting passenger flow requirements, and a congestion risk identification method is used to select a control scheme with the lowest total congestion risk as a final implementation scheme. The application efficiently identifies buffer zone congestion risks.

Description

TECHNICAL FIELD

[0001] The application relates to the field of simulation, in particular to a metro station passenger flow buffer risk identification and collaborative management method and system. BACKGROUND

[0002] With the growth of urban rail transit passenger flow, the subway station and transfer node in peak hours face great pressure of passenger flow dredging. The traditional passenger flow control method includes limiting the number of passengers entering the station and guiding the detour, and some transfer stations disperse peak passenger flow by extending the transfer channel, but the long walking distance in off-peak hours may reduce passenger comfort. In order to improve the flexibility of regulation and control, some stations adopt differentiated transfer paths in different operation periods, which may lead to chaotic flow lines and low efficiency. The passenger flow buffer area is surrounded by barriers to guide passengers to detour, which can effectively control the flow. The buffer area is mostly a fixed facility, and the movable fence can be opened in off-peak hours to shorten the walking distance and improve efficiency. The buffer area is preferably set in the non-paid area, and if necessary, it is extended to the paid area, which needs to be considered comprehensively considering the passenger flow, space layout, traffic capacity and transfer convenience. Reasonable transfer distance and open area utilization can optimize the effect of the buffer area and improve the efficiency and service level of the station. SUMMARY

[0003] In view of the lack of dynamic identification and precise collaboration in the regulation and control of the metro passenger flow buffer area, the application provides a metro station passenger flow buffer risk identification and collaborative management method and system. The partition independent component analysis (PIPCA) method is used to identify the dominant variables of passenger flow in each partition, and an updated buffer area information database is constructed. Based on this, a dynamic risk identification algorithm is designed, which uses the moss growth optimization algorithm (MGO) to optimize the nonlinear sub-regression neural network (NARX) model, and efficiently identifies the congestion risk of the buffer area.

[0004] One aspect of the present application provides a subway station passenger flow buffer zone risk identification and collaborative management method, comprising the following steps: step 1: based on the individual motion simulation software Massmotion of the social force model, a three-dimensional simulation scene of the passenger flow buffer zone of each partition of the subway station is constructed, and multi-factor data of the passenger flow buffer zone of the subway station is obtained; step 2: as the number of buffer zones increases, the complexity of the system also shows an upward trend. The independent analysis is carried out on the multiple buffer zones, a partition independent component analysis (PIPCA) technology is used to find the dominant variables affecting the passenger flow system, and the dominant variables affecting each partition passenger flow subsystem are accurately estimated. Step 3: a subway passenger flow buffer zone dynamic risk identification model is obtained by optimizing a nonlinear sub-regression neural network (NARX) identification model with external input through a moss growth algorithm (MGO). Step 4: a model predictive controller is constructed based on the partition independent component analysis (PIPCA) method and the system identification model constructed by the moss growth algorithm (MGO). The controller optimizes the control strategy of the passenger flow buffer zone to cope with the problems of subway high-density crowd management challenges under complex passenger flow, and a congestion risk identification method based on the passenger flow dynamic identification model is used to judge the passenger flow congestion risk of each partition passenger flow subsystem under different schemes.

[0005] The subway station passenger flow buffer zone dynamic risk identification and collaborative management method and system described above, in step 1, the setting scene is a subway station scene with passenger flow buffer zones. In addition, the number and location of the entrances and exits of the station and the number and layout of the key facilities in the station such as stairs, escalators, etc. also affect the traffic capacity and further affect the arrangement of the buffer zones. The three-dimensional simulation scene built should also include the above factors.

[0006] The subway station passenger flow buffer zone dynamic risk identification and collaborative management method and system described above, in step 2, the partition independent component analysis (PIPCA) technology is used to avoid the problem that as the number of buffer zones increases, the complexity of the system also shows an upward trend. The partition independent component analysis (PIPCA) is a data dimensionality reduction analysis technology. The subway station system is divided into multiple independent sub-regions according to the function and passenger flow characteristics. The correlation coefficient matrix and eigenvalues of each factor index are calculated by performing independent component analysis on the passenger flow data of each sub-region, so as to determine the contribution rate of each factor index, and finally extract the key control variables from the numerous variables. The specific content of the proposed PIPCA technology for finding the dominant variables affecting the passenger flow system includes:

[0007] Independent sub-area division: According to the passenger flow density, the risk nodes in the crowd are identified, and on this basis, the passenger flow control can be divided into two links, the first link is the non-paid area in the station passenger flow control, and the second link is the paid area in the passenger flow control. The first link can divide multiple buffer zones in the area before the entrance of the gate according to the number of in-station gate groups. When the in-station passenger flow in the non-paid area is controlled by the first link, if the passenger flow density at the location such as the escalator entrance in the paid area is still too large or the passenger flow lines conflict, the buffer zone can be widened to the paid area, so the second link can divide multiple buffer zones in the paid area of the station hall according to the number of risk nodes.

[0008] Data standardization: The input observable factors of the passenger flow system can record the passenger flow of each in-station, the control scheme of each buffer zone, and the train timetable, and the output data includes the density of each bottleneck area. Assuming that the number of sample data of a certain sub-system is , the number of passenger flow system factor indexes is , , the value of the hth sample in the ith passenger flow system factor index of the jth sub-system is represented, and the standardized matrix is obtained. Passenger flow system factor index: refers to various observable and measurable parameters that affect the passenger flow system of the subway station, including input indexes and output indexes. The input indexes mainly include the passenger flow of each in-station (i.e. the number of passengers entering the subway station from each entrance per unit time), the control scheme of each buffer zone (i.e. different fence layout and its open / close state), and the train timetable (train arrival and departure time and frequency); the output indexes mainly include the passenger flow density of each bottleneck area (i.e. the number of passengers per unit area).

[0009] Correlation coefficient: the correlation coefficient of different factor indexes is calculated, and the calculation formula is: ; wherein is the correlation coefficient of the ith factor index and the jth factor index, is the value of the ith factor index in the kth sample, is the average value of the ith factor index, n is the number of samples, and i and j are the factor index numbers.

[0010] Correlation coefficient matrix: according to the correlation coefficient, the correlation coefficient matrix R of different passenger flow system factor indexes is constructed: Eigenvalue: the eigenvalue of the correlation coefficient matrix R is solved according to .

[0011] Contribution rate of factor index: according to the eigenvalue of each passenger flow system factor index in the correlation coefficient matrix R, the contribution rate of each passenger flow system factor index is calculated, and the formula is: ​​​The contribution rates are summed according to the obtained contribution rates, and when the contribution rate reaches 85% or more, it is considered that the components participating in the calculation have covered most of the effective information in the original data sample, and then these components are selected as the main influencing factors of the partition subsystem.

[0012] The key control variables are extracted from the set of observable passenger flow related variables of the subway station by the partition independent component analysis (PIPCA) technical method, and the principal components obtained after dimensionality reduction are used as the external input of each partition subsystem.

[0013] The above-mentioned subway station passenger flow buffer zone dynamic risk identification and collaborative management method and system, in step 3, the NARX model of the multi-layer neural network has the optimal model prediction effect. The model allows the neurons in the network layer to be interconnected and feedback connected through the introduction of the feedback mechanism of the recurrent neural network, thereby realizing the strong memory ability of the time series data. The multi-layer neural network NARX model structure is: ; wherein y and u represent the output and input of the system respectively, and represent the time delay of the output and input respectively, is the nonlinear mapping of the input and output.

[0014] The above-mentioned subway station passenger flow buffer zone dynamic risk identification and collaborative management method and system, the construction steps of the system identification model include: the definition of the neural network input and output layer structure in the multi-layer neural network NARX model structure. This architecture combining recurrent neural network and multilayer perceptron makes NARX model show significant advantages in processing nonlinear dynamic system prediction problems.

[0015] The output of the first hidden layer node of the neural network is defined as: ; wherein, is the activation function of the hidden layer, is the time delay order of the input layer, is the time delay order of the output layer, is the weight between the first hidden layer node and the first time delay node of the input signal, the first time delay output of the input signal is , is the weight between the first hidden layer node and the first time delay node of the output signal, is the output of the first time delay node, The threshold value of hidden layer nodes.

[0016] The neural network The output corresponding to the output node is defined as: ;in, For the output nodes and The weights between hidden layer nodes, For the The threshold of the output nodes, is the number of nodes in the output layer. The output nodes represent the predicted passenger flow density of each buffer zone and bottleneck area, which is the final prediction result of the model.

[0017] The aforementioned method and system for dynamic risk identification and collaborative management of subway station passenger flow buffer zones includes the following steps for constructing a system identification model: applying the MGO algorithm to optimize the weights and bias parameters in the NARX neural network, gradually adjusting the weights and bias parameters in the NARX network to reduce overfitting caused by inaccurate parameters. The MGO algorithm (Moss Growth Algorithm) is a bio-inspired optimization algorithm based on moss growth characteristics that simulates the natural process of wind propagation and growth of moss spores. In this application, the MGO algorithm is used to optimize the weights and bias parameters of the NARX neural network. By simulating wind direction guidance, stable and turbulent propagation, and dual-propagation search mechanisms, it avoids falling into local optimal solutions, effectively improving the model's generalization ability and reducing overfitting. The NARX neural network (Nonlinear Autoregressive Exogenous Neural Network) is a recurrent neural network structure specifically designed for time series prediction. By introducing the feedback mechanism of the recurrent neural network, it allows for interconnection and feedback connections between neurons within the network layer, achieving strong memory for time series data. In this application, the NARX network takes the dominant variable obtained in step 2 as input and predicts the future passenger flow density change trend in each buffer zone and bottleneck area by learning the time dependency of historical passenger flow data.

[0018] Initialization: Randomly initialize a set of moss individuals, each representing a set of possible solutions for the NARX network weights and bias parameters. Set parameters such as population size and maximum number of iterations.

[0019] Individual fitness: Calculate individual fitness. For each moss individual, i.e., each set of NARX network weights and biases, train them in the NARX model and calculate their fitness according to the fitness function. The fitness function is defined as: ;in, denote the mean square error of the model prediction in open-loop and closed-loop modes, respectively, MSE, Mean Squared Error: The mean squared error of the model in single-step and multi-step prediction, respectively. Open-loop and closed-loop mode: In NARX neural networks, these two modes refer to different ways the network processes time series data. Open-loop mode: The network uses the true historical output values as input for the current prediction, without forming a feedback loop. In this application, this means that the model uses the actually observed historical passenger flow density data to predict the passenger flow density at the next time. This mode is usually used in the network training stage, as it does not accumulate prediction errors, and the training is more stable. Closed-loop mode: The network uses its own previous predicted output as input for the current prediction, forming a feedback loop. In practical applications, this means that the model uses its own previously predicted passenger flow density values to predict the passenger flow density further in the future. This mode better reflects the performance of the model in practical applications, especially when making multi-step predictions.

[0020] Determine the wind direction: Take the best individual in the population as the reference, and set its dimensional value as the threshold, thereby dividing the population into two subsets. Compare the number of individuals in the subsets by the following formula, and select the subset with more individuals as : ; where represents the number of individuals in the subset.

[0021] In the MGO algorithm, each moss individual represents a vector composed of all the weight and bias parameters of the NARX network. Assuming there are D parameters to be optimized, each individual is a D-dimensional vector. The specific process of population division is as follows: find the individual with the highest fitness in the current population (the best individual ); for each dimension j (j = 1, 2,..., D), do the following: take the j-th dimensional parameter value of the best individual as the threshold ; traverse all individuals in the population, if the j-th dimensional parameter value of an individual is less than or equal to , then it is assigned to subset , otherwise it is assigned to subset ; compare the number of individuals in and , and select the subset with more individuals as the main subset ; this division method helps the algorithm identify potential promising areas in the parameter space, providing a basis for subsequent simulation of wind direction, thereby guiding the population to move towards a better solution space and avoiding falling into local optimal solutions.

[0022] After several divisions, the final subset is: ; where represents the j-th​​ a random number, denotes the number of divisions.

[0023] The simulated wind direction blows from the region to The wind direction is determined by calculating the average distance of individuals in from to smooth the path of individuals approaching , wherein , denotes the total number of individuals in , denotes the distance of an individual from , denotes the set of distances of individuals within from . denotes the pth individual in the subset after division by threshold in the jth dimension.

[0024] Spore dispersal search: the MGO algorithm updates the spore position by the following formula , and mainly simulates the characteristics of spore dispersal and propagation with wind in the exploration stage.

[0025] ; wherein, and denote the propagation distances under stable and turbulent flow, respectively, and the stable propagation distance simulates the directional propagation of moss spores under stable airflow conditions, representing the exploitation ability of the algorithm. In this application, this mechanism guides the algorithm to approach known better parameter combinations, and adjusts the moving range by wind force E and random factor , ensuring that the algorithm can finely adjust the NARX network parameters and improve the prediction accuracy. The turbulent propagation distance simulates the random disturbance propagation of moss spores under turbulent airflow conditions, representing the exploration ability of the algorithm. During the model optimization process, this mechanism enables the algorithm to explore a wider parameter space by introducing randomness, avoiding falling into local optimal solutions, and helping to discover potential better NARX network parameter combinations, improving the model's adaptability to different passenger flow conditions.

[0026] and The significant difference between and enables individuals to randomly select compensation, avoiding the problem of slow early convergence or inability to converge later caused by fixed step size, thereby maintaining the diversity of the population. The specific expression is ; wherein, denotes the wind force, and decreases as the number of iterations increases. express The total number of individuals The proportion of total individuals. 、 and is a random number in the range (0, 1), and is a constant parameter. Indicates the current number of calculations. Indicates the maximum number of iterations.

[0027] Double propagation search: The MGO algorithm adopts a double propagation search strategy during its development phase. The double propagation search strategy includes a social learning mechanism and a wind direction guidance mechanism. The social learning mechanism means that individuals not only learn from the global optimal solution, but also refer to the position information of other individuals, so that the NARX network parameter optimization process can comprehensively consider multiple possible optimization directions. The wind direction guidance mechanism means using group trend information to optimize the NARX network parameters. Guided search, combining the current position and wind factors, forms a directional search path in the parameter space, accelerating the convergence speed of the NARX model parameter optimization process while maintaining appropriate randomness to avoid premature convergence.

[0028] The position of a new moss individual is determined by the following formula, ;in, Indicates the A new individual, express The particles. represents the current optimal individual, is the wind vector. is a random number in the range of (0, 1), is a constant parameter. , is a random vector in the range (0, 1). , is a random number in the range of (0, 1), For wind power.

[0029] During the simulated moss growth process, the position updates of individual mosses correspond to adjustments to the NARX network weights and bias parameters. Through a natural selection process, these weights and biases are gradually optimized, approaching the global optimal solution. When the maximum number of iterations is reached, the globally optimal moss individual position, representing the optimal weights and biases of the NARX network, is returned.

[0030] In the above-mentioned method and system for dynamic risk identification and collaborative management of a subway station passenger flow buffer zone, the performance evaluation of the identification model includes: evaluating the performance of the identification model through the following mean square error and correlation coefficient indicators.

[0031] The deviation between the true value and the predicted value is quantified by the following formula of mean square error, the smaller the value is, the higher the prediction accuracy of the model is, .

[0032] In addition, the goodness of fit of the model to the data is also measured by the following formula of correlation coefficient R, the closer the value is to 1, the better the fitting performance of the model is, ; wherein, is the total number of samples, is the true value and the predicted value, is the mean value, is the mean value.

[0033] The above subway station passenger flow buffer zone dynamic risk identification and collaborative management method and system, in the step 4, the construction step of the model predictive control system, includes:

[0034] The system identification model is used as the prediction model of the predictive controller, and the target function and the constraint condition are defined according to the passenger flow demand; a new type of model predictive controller is proposed, and the target function and the constraint condition are defined by the following formula. In the cost function, the density output error value is an important part in the design process of the cost function of the model predictive controller, which is represented by the first term on the right side of the formula. The second term on the right side of the formula can make the control command of the barrier relatively stable, avoid the sudden gathering or poor evacuation of passenger flow caused by the too large difference of the command change, and maintain the smooth flow of passenger flow in the buffer zone. In order to reduce system oscillation, prolong equipment life, improve passenger experience and enhance system robustness, we limit the frequency of command adjustment of the facility to ensure the efficient, stable and reliable flow of passenger flow, which is represented by the third term on the right side of the formula.

[0035] ;

[0036] wherein, and respectively represent the set value and the predicted output value of the crowd density at the k+i time step; the parameters and define the future time range, in which the output error is minimized, Nu is the number of steps ahead for minimizing the command increment. In a model predictive controller, it represents the range of time steps over which changes in the control action need to be optimized. In subway station passenger flow buffer control systems, this parameter limits the number of future time steps within which changes in the control scheme are considered, while outside this range, the control is assumed to remain unchanged. Specifically, when calculating the objective function, the controller only optimizes the control command increment Δu(k+i-1|k) for the next Nu time steps to reduce computational complexity and ensure the stability of the control strategy. Smaller Nu values ​​can reduce the computational burden, avoid overly frequent switching of control schemes, and improve the operational feasibility of the system; larger Nu values ​​can provide more forward-looking control and optimize the control strategy over a longer time frame, but may increase computational complexity and cause the control strategy to be overly sensitive.

[0037] 、 、 They are the output error of crowd density, the difference of input instruction of buffer scheme, and the weighted coefficient of instruction change frequency. is an indicator variable that indicates whether a change in the input instruction occurs at the kth time step; at the same time, the sequence Records the time interval from the last input change to the current time step; is a positive weight coefficient used to adjust the intensity of the penalty; and is a very small positive value used to avoid the denominator being zero.

[0038] According to the control strength, input Normalization is required and calculated using the following formula: ;in, is the normalized value of the input variable, are the minimum and maximum values ​​of the variable, respectively. Indicates different control schemes The set of natural numbers in order of strength 's mapping.

[0039] The above-mentioned method and system for dynamic risk identification and collaborative management of passenger flow buffer zones in subway stations can effectively control the density of people at nodes by introducing buffer zones. However, the control effects of different layout schemes in daily operation management need to be evaluated. A congestion risk identification method based on the passenger flow dynamic identification model is proposed. The total congestion risk Defined as:

[0040] ;

[0041] in, Indicates the cumulative value of risk over a period of time, represents the congestion risk value per unit time, denotes the unit time, denotes the current passenger density, is the maximum density threshold, is the time when the density value first exceeds the threshold, is the time when it first falls to the threshold or below.

[0042] When the index value is zero, it indicates that there is no risk of passenger flow congestion under the scheme; when the index value is positive, the numerical value is positively correlated with the degree of congestion risk, that is, the larger the value, the higher the risk of congestion that may be caused after the implementation of the scheme. Through the index, different independent partition subsystems are evaluated for implementing different intensity schemes.

[0043] Another aspect of the present application also provides a subway station passenger flow buffer risk identification and collaborative management system, characterized in that it comprises: a three-dimensional simulation module, an individual motion simulation software Massmotion based on a social force model, which constructs a three-dimensional simulation scene of the passenger flow buffer of each partition of the subway station and obtains passenger flow buffer data of the subway station; a data processing module, which processes passenger flow data of each partition by using a partition independent component analysis (PIPCA) algorithm according to the passenger flow buffer data of the subway station, and obtains dominant variables affecting the passenger flow system of each partition; a risk identification module, which obtains a subway passenger flow buffer risk identification model for predicting passenger flow density of each buffer and bottleneck area by optimizing weights and bias parameters of a pre-constructed nonlinear autoregressive neural network (NARX) through a moss growth algorithm (MGO) according to the dominant variables; a control strategy module, which constructs an objective function and constraint conditions according to the subway passenger flow buffer risk identification model, and selects an optimal control strategy; and a risk assessment module, which performs passenger flow congestion risk assessment on the selected optimal control strategy according to the subway passenger flow buffer risk identification model by using a congestion risk identification method, and selects a control scheme with the lowest total congestion risk as the final implementation scheme.

[0044] Compared with the prior art, the present application has the following advantages:

[0045] The application determines the dominant variables of each partition passenger flow subsystem by analyzing the buffer regulation scheme under the background of subway station passenger flow, combining the partition independent component analysis (PIPCA) method proposed under the buffer scene, and establishes an updated subway station passenger flow buffer information database. By designing a new type of buffer dynamic risk identification algorithm, using the moss growth optimization algorithm (MGO) to optimize the initial weight and threshold value of the NARX identification model, the accurate identification of the passenger flow congestion risk in the buffer area is realized, which has high identification accuracy and strong adaptability. At the same time, an improved prediction controller method with superior prediction performance is proposed, which can dynamically optimize the buffer design scheme according to the risk change, and improves the flexibility and accuracy of passenger flow regulation. Compared with the traditional buffer management method which relies on experience and manual adjustment, the regulation effect is improved, the personnel flow rate is optimized, and the stability and safety of system regulation are enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0046] The application will be further described in the form of exemplary embodiments, which will be described in detail through the drawings; these embodiments are not limiting, and in these embodiments, the same numbers represent the same structures, wherein:

[0047] Figure 1 A subway station passenger flow buffer dynamic risk identification and collaborative management method and system process schematic diagram in an embodiment of the application;

[0048] Figure 2 A subway station passenger flow buffer longitudinal S-shaped fence control scheme schematic diagram in an embodiment of the application;

[0049] Figure 3 A subway station passenger flow buffer transverse S-shaped fence control scheme schematic diagram in an embodiment of the application;

[0050] Figure 4 A dominant variable identification schematic diagram by a partition independent component analysis (PIPCA) method in an embodiment of the application;

[0051] Figure 5 A moss growth algorithm (MGO) optimized NARX multilayer neural network schematic diagram in an embodiment of the application;

[0052] Figure 6 A prediction control system structure schematic diagram in an embodiment of the application;

[0053] Figure 7 The prediction comparison results of the system under two modes of screening dominant index and single input factor index based on the PIPCA technology in an embodiment of the application are shown;

[0054] Figure 8The prediction comparison results of system two under the two modes of screening dominant indexes and single input factor indexes based on the PIPCA technology in the embodiments of the present application are shown;

[0055] Figure 9 The prediction comparison results of system three under the two modes of screening dominant indexes and single input factor indexes based on the PIPCA technology in the embodiments of the present application are shown;

[0056] Figure 10 The prediction comparison results of system four under the two modes of screening dominant indexes and single input factor indexes based on the PIPCA technology in the embodiments of the present application are shown;

[0057] Figure 11 The prediction comparison results of system five under the two modes of screening dominant indexes and single input factor indexes based on the PIPCA technology in the embodiments of the present application are shown;

[0058] Figure 12 The prediction comparison results of system six under the two modes of screening dominant indexes and single input factor indexes based on the PIPCA technology in the embodiments of the present application are shown;

[0059] Figure 13 The optimal individual fitness value of the method in the embodiments of the present application;

[0060] Figure 14 The optimal individual fitness value of the traditional algorithm in the embodiments of the present application;

[0061] Figure 15 The prediction results of system one identified by the PIPCA-GWO identification method of the method in the embodiments of the present application to the NARX model;

[0062] Figure 16 The prediction results of system two identified by the PIPCA-GWO identification method of the method in the embodiments of the present application to the NARX model;

[0063] Figure 17 The prediction results of system three identified by the PIPCA-GWO identification method of the method in the embodiments of the present application to the NARX model;

[0064] Figure 18 The prediction results of system four identified by the PIPCA-GWO identification method of the method in the embodiments of the present application to the NARX model;

[0065] Figure 19 The prediction results of system five identified by the PIPCA-GWO identification method of the method in the embodiments of the present application to the NARX model;

[0066] Figure 20The sixth prediction result of the system identified by the PIPCA-GWO identification method in the method of the present application in the embodiments of the present application to the NARX model;

[0067] Figure 21 The output results of the system under the two schemes of the improved predictive control and the conventional predictive control in the method of the present application in the embodiments of the present application are compared;

[0068] Figure 22 The duration of the control input action of the system under the two schemes of the improved predictive control and the conventional predictive control in the method of the present application in the embodiments of the present application;

[0069] Figure 23 The shortest instruction time of the system under the two schemes of the improved predictive control and the conventional predictive control in the method of the present application in the embodiments of the present application is compared;

[0070] Figure 24 The number of instruction frequencies of the system under the two schemes of the improved predictive control and the conventional predictive control in the method of the present application in the embodiments of the present application is compared;

[0071] Figure 25 The subway station passenger flow density map of the buffer-free control in the method of the present application in the embodiments of the present application;

[0072] Figure 26 The subway station passenger flow density map of the PIPCA-GWO-NARX model control in the method of the present application in the embodiments of the present application. DETAILED DESCRIPTION

[0073] The method and system provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0074] Figure 1 A subway station passenger flow buffer dynamic risk identification and collaborative management method and system schematic diagram to which the present application can be applied is shown. The method is described in detail as follows:

[0075] Step S101: Constructing a three-dimensional simulation scene of passenger flow buffer zones of each subarea of a subway station

[0076] An individual motion simulation software based on a social force model can be used to build a three-dimensional simulation scene of a subway station, including scene elements such as entrances and exits, station hall layers, platform layers, and passenger flow buffers, to simulate the actual scene. It should be noted that in the present embodiment, the simulation environment is the entire subway station, and the passenger flow includes in-and-out station passenger flow, boarding and alighting passenger flow, and all station passers-by. Further reference Figure 2 and 3 Different fence control scheme examples of the passenger flow buffer provided by the present application, the device embodiment corresponds to the method of the present application, and the device can be applied to various electronic devices.

[0077] Step S102: using principal component analysis to extract dominant variables of each sub-region, accurately identifying key factors affecting the passenger flow subsystem.

[0078] The above step S102 further includes Figure 4 The steps S201, S202, S203 and S204 in the schematic diagram are performed by the partition independent component analysis (PIPCA) method shown in the figure, and the steps are specifically:

[0079] Step S201: independent sub-region division.

[0080] First, identify the congestion risk nodes according to the passenger flow density, and on this basis, the passenger flow control can be divided into two links, the first link is the non-paid area in-station passenger flow control, and the second link is the paid area passenger flow control. The first link can divide multiple buffer zones in the area before the entrance of the gate according to the number of in-station gate groups. When the non-paid area in-station passenger flow is controlled by the first link, if the passenger flow density at positions such as escalator entrances in the paid area is still too large or the passenger flow lines are in conflict, the buffer zone can be widened to the paid area, so the second link can divide multiple buffer zones in the paid area of the station hall according to the number of risk nodes.

[0081] Step S202: data standardization.

[0082] The input observable factors of the passenger flow system can record the passenger flow of each in-station, the control scheme of each buffer zone, and the train timetable, and the output data includes the density of each bottleneck area. Assuming that the number of sample data of a partition subsystem is , the number of passenger flow system factor indexes is , represents the value of the hth sample in the partition subsystem. th passenger flow system factor index, and the standardized matrix is obtained as .

[0083] Step S203: correlation coefficient and correlation coefficient matrix.

[0084] The correlation coefficient of different factor indexes is calculated, and the calculation formula is: ;

[0085] According to the correlation coefficient, the correlation coefficient matrix R m×m of different passenger flow system factor indexes is constructed: ;

[0086] wherein is the correlation coefficient of the ith factor index and the jth factor index, is the value of the ith factor index in the kth sample, and is the average value of the i-th factor index, n is the sample number, and i and j are the factor index numbers.

[0087] Step S204: Contribution rate of the factor index.

[0088] According to Solve the eigenvalues of the correlation coefficient matrix R According to the eigenvalues of the correlation coefficient matrix R of different factor indexes, the contribution rate of each passenger flow system factor index is calculated, and the formula is:

[0089] According to the contribution rate obtained, the sorted contribution rate is summed up. When the contribution rate reaches more than 85%, it is considered that the components participating in the calculation have covered most of the effective information in the original data sample, and then these components are selected as the main influencing factors of the partition subsystem.

[0090] Step S103: Combine the MGO algorithm to optimize the parameters to improve the precision of the NARX model of the multi-layer neural network and suppress overfitting.

[0091] The NARX model of the multi-layer neural network has the optimal model prediction effect. The model introduces the feedback mechanism of the recurrent neural network, allows the neurons in the network layer to be interconnected and feedback connected, and thus realizes strong memory ability for time series data. The structure of the multi-layer neural network NARX model is: ; wherein y and u respectively represent the output and input of the system, and represent the time delay of the output and input respectively, is the nonlinear mapping of the input and output.

[0092] In the multi-layer neural network NARX model, the output of the first hidden layer node of the neural network is defined as: ; wherein is the hidden layer activation function, is the time delay order of the input layer, is the time delay order of the output layer, is the weight between the first hidden layer node and the first time delay node of the input signal, the first time delay output of the input signal , is the weight between the first hidden layer node and the first time delay node of the output signal, is the output of the first time delay node, is the first​ The threshold value of hidden layer nodes.

[0093] The neural network The output corresponding to the output node is defined as: ;in, For the output nodes and The weights between hidden layer nodes, For the The threshold of the output nodes, is the number of nodes in the output layer.

[0094] In order to determine and optimize the weight and bias parameters in the NARX neural network, the MGO algorithm is used to gradually adjust the weight and bias parameters in the NARX network. The MGO algorithm includes Figure 5 Steps S301, S302, S303, S304, and S305 in the schematic diagram of the moss growth algorithm (MGO) optimizing the NARX multi-layer neural network are specifically as follows:

[0095] Step S301: Initialize the number of populations.

[0096] Randomly initialize a set of moss individuals, each representing a set of possible solutions for the NARX network weights and bias parameters. Set parameters such as population size and maximum number of iterations.

[0097] Step S302: Calculate individual fitness.

[0098] Each moss individual, i.e., each set of NARX network weights and biases, is trained in the NARX model, and its fitness is calculated according to the fitness function. Figure 8 This is the fitness value comparison result in this application example. The fitness function is defined as:

[0099]

[0100] in, 、 denote the mean square error of the model prediction in open-loop and closed-loop modes, respectively, 、 They represent the mean square error of the model in single-step prediction and multi-step prediction, respectively.

[0101] Step S303: Determine the wind direction.

[0102] By population The best individual in As a benchmark, The dimension value is used as the threshold to divide the population into and Two subsets. Use the following formula to compare the number of subset individuals and select the subset with more individuals as :

[0103] ;

[0104] in, Represents the number of individuals in the subset.

[0105] After multiple partitions, the final subset is:

[0106] ;

[0107] in, Indicates the A random number, Indicates the number of divisions.

[0108] Simulated wind direction blowing from area to , by calculating Individuals and The average distance to determine wind direction , to smooth individual approximation The path, where , express The total number of individuals in Represents an individual relative to distance, express Intrapersonal relative to The distance set.

[0109] Step S304: Spore spreading search.

[0110] The MGO algorithm updates the spore position using the following formula: , and in the exploration stage, it mainly simulated the characteristics of spores spreading and spreading with the wind.

[0111] ;in, and denote the propagation distances under steady and turbulent conditions, respectively. and The significant difference in enables individuals to randomly select compensation, avoiding the problem of slow convergence in the early stage or failure to converge in the later stage caused by a fixed step size, thereby maintaining the diversity of the population. The specific expression is ;in, Represents the wind force and weakens as the number of iterations increases. express The total number of individuals The proportion of total individuals. 、 and is a random number in the range of (0, 1), and is a constant parameter. represents the current number of calculations, represents the maximum number of iterations.

[0112] Step S305: Double propagation search.

[0113] The development stage of the MGO algorithm adopts a double propagation search strategy, and the position of the new moss individual is determined by the following formula.

[0114] ; wherein, represents the th new individual, represents the th particle in . represents the current optimal individual, is the wind vector. is a random number in the range of (0, 1), is a constant parameter. , is a random vector in the range of (0, 1). , is a random number in the range of (0, 1), is the wind force.

[0115] In the process of simulating the growth of moss, the position update of the moss individual corresponds to the adjustment of the NARX network weight and bias parameters. Through the natural selection process, the weight and bias parameters of the NARX network are gradually optimized, and the global optimal solution is approached. When the maximum number of iterations is reached, the global optimal moss individual position, that is, the optimal weight and bias of the NARX network, is returned.

[0116] We use the mean square error (MSE) and the correlation coefficient R to evaluate the performance of the identification model. The mean square error quantifies the deviation between the true value and the predicted value, and the smaller the value, the higher the prediction accuracy of the model. The correlation coefficient R measures the goodness of fit of the model to the data, and the closer the value is to 1, the better the fitting performance of the model. The mean square error (MSE) and the correlation coefficient R are expressed as follows: ; ; wherein, is the total number of samples, is the true value and the predicted value, is the mean, is the mean.

[0117] The present example uses a conventional method based on the PIPCA technology, a PIPCA-GWO method and a conventional method without PIPCA technology to identify the NARX model respectively, and verifies the effectiveness of the prediction model of the method of the present application, and the specific effects are shown in Table 1 as follows:

[0118] Table 1 Comparison of results of different system identification models

[0119]

[0120] Step S104: Constructing a prediction controller.

[0121] The system identification model is used as the prediction model of the prediction controller, and the target function and the constraint condition are defined according to the traffic flow demand; a new type of model prediction controller is proposed, and the target function and the constraint condition are defined by the following formula.

[0122]

[0123] and respectively represent the set value and the predicted output value of the crowd density at the k+i time step. Parameters and define the future time range; is the number of steps in advance of the minimum instruction increment; are the weighted coefficients of the crowd density output error, the buffer scheme input instruction difference and the instruction change frequency respectively; is an indicator variable, representing whether the input instruction has changed at the k time step; the sequence records the time interval from the last input change to the current time step; is a positive weight coefficient for adjusting the intensity of the penalty; and is a very small positive value;

[0124] represents the input value of the system at the k time, and represents the input value of the system at the k+1 time; represents the input value of the system at the k+i time; represents the input value of the system at the k+i-1 time; are the minimum value and the maximum value of the corresponding variable respectively; represents the maximum control strategy change amplitude allowed at the k+i time; represents the k-1 sequence;

[0125] According to the control intensity, input ​​​​Normalization is needed, which is calculated by the following formula. ; wherein, is the normalized value of the input variable, is the minimum and maximum value of the variable, respectively, denotes different control schemes a mapping of the set of natural numbers in order of strength.

[0126] In the cost function, the density output error value is an important part of the cost function design process of the model predictive controller, which is represented by the first term on the right side of the equation. The second term on the right side of the equation can make the control command of the barrier relatively smooth, avoiding sudden gathering or poor dispersion of passenger flow caused by too large difference in command change, and maintaining the smooth flow of passenger flow in the buffer zone. In order to reduce system oscillation, prolong equipment life, improve passenger experience and enhance system robustness, we limit the frequency of command adjustment of the facility to ensure efficient, stable and reliable passenger flow, which is represented by the third term on the right side of the equation.

[0127] Step S105: Construct congestion risk index to evaluate the control effect of different schemes.

[0128] The introduction of buffer zone can realize effective control of node crowd density, but the control effect of different layout schemes in daily operation and management needs to be evaluated. A congestion risk identification method based on passenger flow dynamic identification model is proposed, and the total congestion risk is defined as:

[0129] ; wherein, denotes the cumulative value of risk in a period of time, denotes the congestion risk value per unit time, denotes the unit time, denotes the current passenger density, is the maximum density threshold, is the time when the density value first exceeds the threshold, is the time when it first falls below the threshold.

[0130] When the index value is zero, it indicates that there is no risk of passenger flow congestion under this scheme; when the index value is positive, the size of its value is positively correlated with the degree of congestion risk, that is, The larger the value, the higher the risk of congestion that may occur after the implementation of the scheme. Through this index, different intensity schemes of different independent partition subsystems are evaluated.

[0131] ​In some optional implementations in the present application, the construction step of the model predictive control system comprises: deploying the controlled object by using the identification model; receiving the adjustment instruction from the predictive controller according to the reference target, and the output of the controlled object is transmitted to the predictive controller. Referring to Figure 2 and Figure 3 , the predictive control system structure in the embodiments of the present application is shown in FIG. 1. Figure 6 , Figure 6 is the predictive control system structure in the embodiments of the present application.

[0132] Figure 7 The predictive comparison results of the same conventional NARX model identification method but based on two kinds of ways of screening the dominant input factor index and the single input factor index respectively by using the PICPA technology are shown in FIG. 2, wherein Figure 7 to Figure 12 The predictive comparison results of systems one to six in the embodiments of the present application based on two kinds of ways of screening the dominant index and the single input factor index by using the PICPA technology are shown in FIG. 3. The simulation results show that the PICPA-NARX model can more accurately capture the passenger flow dynamic characteristics and significantly reduce the prediction error compared with the conventional NARX model, verifying the effectiveness of the PICPA-NARX model in complex system modeling.

[0133] Figure 13 and Figure 14 The fitness value comparison results of the new fitness index method and the conventional method are shown in FIG. 4, Figure 13 is the fitness value of the present application, Figure 14 is the fitness value of the conventional algorithm. It can be seen that the new fitness function improves the robustness and comprehensive prediction ability of the NARX model in the complex passenger flow system by fusing multiple prediction error indicators, and performs better than the conventional single-step MSE optimization method.

[0134] Figure 15 to Figure 20 The predictive results of the NARX model identification by using the PICPA-GWO method of the present application are shown in FIG. 5, wherein Figure 15 is the predictive result of system one, Figure 16 is the predictive result of system two, Figure 17 is the predictive result of system three, Figure 18 is the predictive result of system four, Figure 19 is the predictive result of system five, Figure 20 is the predictive result of system six. It can be seen that the new identification model proposed in the present application can more accurately capture the dynamic characteristics of the system and effectively handle the coupling relationship between variables, thereby showing higher precision and robustness in complex dynamic system modeling.

[0135] Figure 21The output results of the system under the two schemes of improved predictive control and conventional predictive control are shown. It can be seen that the conventional predictive control strategy has higher density tracking accuracy, and the output fluctuation closely follows the target value, and the tracking performance is excellent. Although the improved predictive control strategy has local tracking error, the overall trend is consistent with the conventional predictive control, and successfully controls the density within the preset threshold, taking into account the regulation effect and control action smoothness, meeting the demand of crowd density control.

[0136] Figure 22 The duration of the control input action of the system under the two schemes of improved predictive control and conventional predictive control is shown. Each color block in the figure represents the duration of one control action. It can be seen that the improved predictive control strategy greatly reduces the control action frequency and amplitude compared with the conventional predictive control, and takes into account the density control and energy optimization, showing higher practical application value.

[0137] Figure 23 And Figure 24 The comparison results of the control input indicators of the system under the two schemes of improved predictive control and conventional predictive control are shown, Figure 23 is the shortest instruction time, Figure 24 is the number of instruction frequencies. It can be seen that the shortest instruction time under the conventional predictive control is generally lower than 60 s, while the improved predictive control is more than 180 s, showing obvious advantages. In terms of control operability, the improved predictive control strategy reduces the instruction frequency of each partition by 33.3% to 50% compared with the conventional predictive control strategy, verifying its effectiveness and practical application value in the buffer zone passenger flow regulation system.

[0138] Figure 25 And Figure 26 The passenger flow density diagrams of the subway station with and without buffer zone control are shown, Figure 25 is the passenger flow density diagram of the subway station without buffer zone control, Figure 26 is the passenger flow density diagram of the subway station obtained by the PICPA-GWO-NARX model under low-intensity control of the buffer zone. It can be seen that subsystem 3 successfully realizes the congestion risk control under the lowest control intensity, so the buffer zone adopts the lowest intensity fixed control mode, and through the coordinated regulation of other partitions, ensures that the density of each bottleneck area meets the standard.

[0139] The purpose of the embodiment is to provide a subway station passenger flow buffer zone dynamic risk identification and collaborative management method and system. The control device is an electronic device, including at least one processor, and at least one memory and bus connected with the processor, the processor, the memory complete communication through the bus, the processor is used to call the program instruction in the memory, in order to execute the subway station passenger flow buffer zone dynamic risk identification and collaborative management method and system of the application.

[0140] The above description of the application and its embodiments is illustrative, and not restrictive. Many variations of the application can become apparent to those of ordinary skill in the art upon review of the foregoing description. The only true limitation on the scope of the application is set forth in the appended claims and equivalents thereof. The appended claims are intended to cover all such variations as fall within the scope of the application. The claims' scope is not limited to the specific embodiments described herein, but extends to and includes all alternative embodiments, modifications, and equivalents thereof. The words "comprise," "comprising," "include," "including," and "includes" are not limited to only the elements or steps listed. The words "a," "an," and "the" include both singular and plural referents. The terms "first," "second," and "third" are used to denote names of elements and are not intended to otherwise limit the elements.

Claims

1. A subway station passenger flow buffer zone risk identification and collaborative management method, characterized in that, The method comprises the following steps: S1, constructing a three-dimensional simulation scene of the passenger flow buffer zone of the subway station by using the individual motion simulation software Massmotion based on the social force model, and obtaining the passenger flow buffer zone data of the subway station; S2, processing the passenger flow data of each subzone by using the independent component analysis (PIPCA) algorithm according to the passenger flow buffer zone data of the subway station, and obtaining the dominant variables affecting the passenger flow system of each subzone; S3, taking the dominant variables as inputs, optimizing the weights and bias parameters of the pre-constructed nonlinear autoregressive neural network (NARX) by using the moss growth algorithm (MGO), and obtaining the subway passenger flow buffer zone risk identification model for predicting the passenger flow density of each buffer zone and the bottleneck period; S4, constructing a target function and constraint conditions according to the subway passenger flow buffer zone risk identification model, and selecting the optimal control strategy under the constraint condition of meeting the passenger flow demand; S5, evaluating the passenger flow congestion risk under different control strategies selected in step S4 according to the subway passenger flow buffer zone risk identification model by using the congestion risk identification method, and selecting the control scheme with the lowest total congestion risk as the final implementation scheme. The moss growth algorithm (MGO) is used to optimize the weights and bias parameters of the pre-constructed nonlinear autoregressive neural network (NARX), and the method comprises the following steps: A group of moss individuals are randomly initialized as the initial population, and each moss individual represents the weight of the nonlinear sub-regression neural network NARX 、 and , and the bias parameters and A set of feasible solutions, and set the population size and maximum number of iterations; calculating a fitness of each moss individual, wherein, , respectively represent the mean square error of the model prediction in open loop and closed loop mode, , respectively represent the mean square error of the model in one-step prediction and multi-step prediction; The individual with the highest fitness in the population is the optimal individual ; The optimal individual is taken as the reference, for each dimension j in the population, the jth dimension value of the optimal individual is taken as the threshold, the population is divided into and two subsets; selecting as the main subset the subset with the largest number of individuals : wherein, denotes the number of individuals of the subset; After multiple divisions, the final subset is obtained : wherein, represents the jth random number, represents the division number; represents the pth individual in the subset after division according to the threshold value in the jth dimension; after division according to the threshold value in the jth dimension; According to the final subset , simulated wind direction from the final subset The region blows towards the optimal individual , by calculating the final subset Average individuals and optimal individuals The average distance to determine wind direction : ,in, express The total number of individuals in Represents an individual relative to distance, express Intrapersonal relative to The distance set of ; In the exploration phase, the wind direction is exploited updating the position of each moss individual in the population , simulating the process of spore dispersal by the wind; In the development phase, the double propagation search strategy is adopted to determine the new moss individual position ; When a preset maximum iteration number is reached, output the global optimal moss individual position as the weight of the nonlinear subregression neural network NARX , and and bias parameters and ; wherein, in the exploration phase, the wind direction updating the position of each moss individual in the population simulating the process of spore dispersal by wind: ， wherein, denotes the ith individual; and denotes the propagation distance under stability and turbulence, respectively: ； Wherein, E represents wind force: , represents the current number of calculations, represents the maximum number of iterations; , and is a random number in the range of 0 to 1, and are constant parameters; represents the proportion of the total number of individuals in the total number of individuals. The step S4 of selecting the optimal control strategy comprises the following steps: Constructing a target function and constraint conditions according to the subway passenger flow buffer zone risk identification model; ； wherein, and respectively represent the set value and the predicted output value of the crowd density at the k+i time step; the parameters and define the future time range; is the number of steps ahead for which the increment minimization is instructed; , , are respectively the crowd density output error, the buffer scheme input instruction difference value, and the weighted coefficient of the instruction change frequency; is an indicator variable representing whether a change in the input instruction has occurred at the k time step; the sequence records the time interval from the last input change to the current time step; is a positive weight coefficient for adjusting the intensity of the penalty; and is a very small positive value; represents the input value of the system at the kth time instant, represents the input value of the system at the k+1th time instant; represents the input value of the system at the k+i time instant; represents the input value of the system at the k+i-1th time instant; are respectively the minimum and maximum values of the corresponding variable; represents the maximum control policy variation amplitude allowed at the k+i time instant; represents the k-1th sequence; Minimizing the target function under the constraint condition to select the optimal control strategy from the buffer zone control schemes.

2. The subway station passenger flow buffer zone risk identification and collaborative management method according to claim 1, wherein: The subway station passenger flow buffer zone data comprises input data and output data; The input data comprises the passenger flow of each entrance, the train schedule of each train, and the buffer zone control scheme; The output data comprises the passenger flow density of each buffer zone and the bottleneck period; The buffer zone control scheme refers to a target area surrounded by multiple barriers, and different passenger flow management and control layout schemes are formed by opening and closing the barriers.

3. The subway station passenger flow buffer zone risk identification and collaborative management method according to claim 2, wherein: The step S2 of obtaining the dominant variables affecting the passenger flow system of each subzone comprises the following steps: Dividing the subway station into multiple subsystems according to the three-dimensional simulation scene of the subway station; Identifying the congestion risk nodes of each subsystem according to the subway station passenger flow buffer zone data; The metro station passenger flow buffer data is standardized to obtain a standardized matrix X of each subsystem, wherein, the matrix element represents a value of the bth passenger flow system factor index of the hth sample in the subsystem, and the number of passenger flow system factor indexes is m. According to the standardized matrix X, the correlation coefficients without the factor indicators are calculated ; wherein, is a correlation coefficient of the i-th factor index and the j-th factor index, is a value of the i-th factor index in the k-th sample, is an average value of the i-th factor index, n is a number of samples, and i and j are positive integers. According to the correlation coefficient , a correlation coefficient matrix of different passenger flow system factor indexes is constructed : ; Solving for eigenvalues of a correlation matrix ;​ According to eigenvalues of the correlation coefficient matrix of different passenger flow system factor indexes , the contribution rate of each passenger flow system factor index is calculated : wherein, is the ith eigenvalue; According to the contribution rate The passenger flow system factor index with the cumulative contribution rate greater than the threshold value is selected as the dominant variable of each subarea passenger flow system. Dividing the passenger flow control into two links according to the location of the congestion risk nodes: the first link is the entrance passenger flow control in the non-payment area, and the second link is the passenger flow control in the payment area.

4. The subway station passenger flow buffer zone risk identification and collaborative management method according to claim 2, wherein: S31, taking the dominant variable as input, constructing a nonlinear subregression neural network NARX: wherein y represents the output of the system, and u represents the input of the system; and respectively represent the time delay of the output and the input, is the nonlinear mapping of the input and output; S32, based on the nonlinear sub-regression neural network NARX, defines the hidden layer node output of the neural network : ,in, is the hidden layer activation function, is the delay order of the input layer, is the delay order of the output layer, For the The hidden layer nodes and the input signal The weights between the delay nodes, the input signal No. The step delay output is , For the The hidden layer nodes and the output signal The weight between the delay nodes, For the The output of the step delay node, For the The threshold of hidden layer nodes; S33, according to , define the output node corresponding output : , wherein, is the weight between the first output node and the first hidden layer node, is the threshold value of the first output node, is the number of output layer nodes; S34, using the moss growing algorithm MGO to optimize the weight of the nonlinear sub-regression neural network NARX 、 and and bias parameters and ; S35, using the optimized weights , and and bias parameters and updating the nonlinear subregression neural network NARX and evaluating the performance of the nonlinear subregression neural network NARX; The step S3 of obtaining the subway passenger flow buffer zone risk identification model for predicting the passenger flow density of each buffer zone and the bottleneck period comprises the following steps: S36, selecting the nonlinear autoregressive neural network (NARX) with the best performance as the subway passenger flow buffer zone risk identification model. In the development phase, a twin propagation search strategy is adopted to determine the new moss individual position : ， in, Indicates the A new individual, express The particles; represents the current optimal individual, is the wind vector; is a random number between 0 and 1, is a constant parameter; For evaluation Whether the particles inside are utilized, is a random vector in the range 0 to 1; , is a random number between 0 and 1, For wind power; Represents the current optimal individual The value in the jth dimension.

5. The subway station passenger flow buffer zone risk identification and collaborative management method according to claim 1, wherein:

6. The subway station passenger flow buffer zone risk identification and collaborative management method according to claim 1, wherein: S5, select the control scheme with the lowest total congestion risk as the final implementation scheme, including: According to the subway passenger flow buffer risk identification model, the passenger flow density data of each control strategy is obtained by predicting the different control strategies selected in step S4; Based on the acquired passenger flow density data, define total congestion risk : ； wherein, represents the cumulative value of risk over a period of time, represents the congestion risk value per unit of time, represents a unit of time, represents the current passenger density, is the maximum density threshold value, is the time at which the density value first exceeds the threshold value, is the time at which it first falls back to the threshold value or below; Comparing the total congestion risk of different control strategies , selecting the control scheme with the lowest total congestion risk as the final implementation scheme.

7. A system based on the risk identification and collaborative management method of the passenger flow buffer zone of the subway station according to any one of claims 1 to 6, characterized in that, Including: The three-dimensional simulation module is based on the individual motion simulation software Massmotion of the social force model. The three-dimensional simulation scene of the passenger flow buffer zone of each partition of the subway station is constructed, and the subway station passenger flow buffer zone data is obtained. The data processing module processes the passenger flow data of each partition by using the partition independent component analysis (PIPCA) algorithm according to the subway station passenger flow buffer zone data, and obtains the dominant variables affecting the passenger flow system of each partition. The risk identification module obtains the subway passenger flow buffer risk identification model for predicting the passenger flow density of each buffer zone and bottleneck zone by optimizing the weight and bias parameters of the pre-constructed nonlinear autoregressive neural network (NARX) through the moss growth algorithm (MGO) according to the dominant variables. The control strategy module constructs the objective function and constraint condition according to the subway passenger flow buffer risk identification model, and selects the optimal control strategy. The risk assessment module uses the congestion risk identification method to evaluate the passenger flow congestion risk of the optimal control strategy selected according to the subway passenger flow buffer risk identification model, and selects the control scheme with the lowest total congestion risk as the final implementation scheme.

Images