Urban rail transit abnormal passenger flow prediction method and system based on artificial intelligence

Abstract

The invention discloses an urban rail transit abnormal passenger flow prediction method and system based on artificial intelligence, and relates to the field of passenger flow prediction. According to the method, the SVM machine learning model is trained and tested by adopting the classified historical abnormal passenger flow data of the station, the historical data of the natural influence factors and the historical data of the social influence factors, so that a relatively accurate classification model is provided for subsequent classification; secondly, collecting real-time data of different types of natural influence factors, real-time data of social influence factors and real-time passenger flow data of different rail transit stations in real time, and continuously updating the final SVM classification model so as to adapt to continuously changed behavioral habits and modes of people; the accuracy of subsequently classifying the predicted data is improved; indirect prediction is carried out on the passenger flow volume data through the predictable natural influence factor data and social influence factor data, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the field of passenger flow prediction, and in particular, relates to an artificial intelligence-based method and system for predicting abnormal passenger flow in urban rail transit. Background Art

[0002] Chinese patent CN108846514B discloses a method for predicting the evacuation demand of rail transit passenger flow under emergencies. The method obtains a sample set of normal short-term traffic demand, uses a hybrid prediction model to perform multi-step prediction, and obtains the prediction results of short-term subway travel demand under normal conditions; establishes a directed graph of the rail transit network; allocates the OD of rail transit travel to obtain an allocation matrix; obtains the calculation formula for the evacuation demand of each station in the entire network during the impact of the event based on the characteristics of the emergency and the prediction results; obtains the spatiotemporal evolution of the evacuation demand through simulation, and obtains the peak retention value.

[0003] At present, the prediction of passenger flow in urban rail transit is generally carried out by collecting passenger flow data at historical moments and fitting these passenger flow data themselves for prediction. However, passenger flow data is often affected by many factors and the data is highly random, which leads to inaccurate results of directly fitting and predicting the passenger flow data. Summary of the invention

[0004] In response to the problems in the related technology, the present invention proposes an artificial intelligence-based urban rail transit abnormal passenger flow prediction method and system to overcome the above-mentioned technical problems existing in the existing related technology.

[0005] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0006] The present invention is an artificial intelligence-based method for predicting abnormal passenger flow in urban rail transit, comprising the following steps:

[0007] S1. The historical data collection module collects multiple groups of different types of historical data on natural influencing factors, historical data on social influencing factors, and historical passenger flow data at different rail transit stations;

[0008] S2, an abnormal passenger flow data extraction module extracts abnormal passenger flow data from the historical passenger flow data to obtain a historical abnormal passenger flow data matrix of the station;

[0009] The classification module classifies the historical abnormal passenger flow data matrix of the site;

[0010] The classification model building module uses the classified historical abnormal passenger flow data of the station, historical data of natural influencing factors, and historical data of social influencing factors to build the final SVM classification model;

[0011] S3, the real-time data collection module collects different types of real-time data of natural influencing factors, real-time data of social influencing factors, and real-time passenger flow data of different rail transit stations;

[0012] The classification model real-time update module updates the final SVM classification model in real time according to the real-time data of natural influencing factors, real-time data of social influencing factors and real-time passenger flow data;

[0013] S4, the prediction module predicts the natural influencing factor data and social influencing factor data at future moments to obtain a future comprehensive influencing factor data matrix;

[0014] S5, the final judgment module uses the final SVM classification model to classify the future comprehensive influencing factor data matrix, and determines whether there will be abnormal passenger flow phenomena at urban rail transit stations at future times based on the classification results;

[0015] Since the influencing factors of the passenger flow of urban rail transit can be generally divided into natural factors and non-natural factors, that is, social factors; therefore, in this scheme, by collecting multiple groups of different types of historical data of natural influencing factors and historical data of social influencing factors, the collected data can cover all aspects of the passenger flow that affects urban rail transit, thereby increasing the accuracy of subsequent predictions; at the same time, the corresponding historical passenger flow data of different rail transit stations is also collected, which provides a basis for the subsequent classification of the historical data of natural influencing factors and the historical data of social influencing factors; secondly, by extracting the abnormal passenger flow data from the collected historical passenger flow data separately, it is convenient to classify the abnormal passenger flow data separately in the subsequent classification, avoiding the interference of normal passenger flow data, and making the classification more efficient and accurate; the classified historical abnormal passenger flow data of the station, the historical data of natural influencing factors and the historical data of social influencing factors are used to train and test the SVM machine learning model, so that the SVM classification model can be used. The corresponding passenger flow category is determined according to the historical data of natural influencing factors and the historical data of social influencing factors, which provides a more accurate classification model for the subsequent classification of the predicted historical data of natural influencing factors and the historical data of social influencing factors; because human behavior models may continue to change with the passage of time, different types of real-time data of natural influencing factors, real-time data of social influencing factors and real-time passenger flow data of different rail transit stations are collected in real time, and the final SVM classification model is continuously updated according to these real-time data to adapt to the constantly changing behavior habits and patterns of people, so that the classification ability of the final SVM classification model can always be maintained at the best, thereby improving the accuracy of the subsequent classification of the predicted data; because the passenger flow data itself has a strong randomness, the passenger flow data cannot be directly fitted and predicted, but the passenger flow data is indirectly predicted through the natural influencing factor data and social influencing factor data that can be predicted, thereby improving the accuracy of the prediction.

[0016] Preferably, the S1 comprises the following steps:

[0017] S11. Set the natural influencing factors of passenger flow and the set of social influencing factors of passenger flow b = {b1,b2,...,b i ,...,b b′}; represents the set i-th natural influencing factor of passenger flow, Indicates the total number of social influencing factors of passenger flow; b i represents the i-th social influencing factor of passenger flow, b′ represents the total number of social influencing factors of passenger flow;

[0018] Re-set the transportation track station set represents the set i-th urban transportation rail station,

[0019] Indicates the total number of set transportation track stations;

[0020] S12. Set the time point set for historical data collection Indicates the set i-th historical data collection time point, Indicates the total number of set historical data collection time points;

[0021] According to the historical data collection time point set Collect the natural influencing factors of passenger flow at each time point The data of each natural influencing factor of passenger flow, the set of social influencing factors of passenger flow b = {b1, b2, ..., b i ,...,b b′} and the data of each social influencing factor of passenger flow and the transportation rail station set The passenger flow data of each transportation track station in is obtained; the historical natural influencing factor data matrix d1, the historical social influencing factor data matrix d2 and the station historical passenger flow data matrix d3 are obtained; d1, d2 and d3 are as follows,

[0022]

[0023] Among them, d 1ij d 2ij and d 3ij They respectively represent the data of the jth natural influencing factor of passenger flow, the data of the jth social influencing factor of passenger flow, and the passenger flow data of the jth urban transportation rail station collected at the i-th historical data collection time point;

[0024] The natural factors affecting passenger flow mainly include climate conditions, such as heavy rain, heavy snow, high temperature, severe cold, etc.; geographical location, such as located in tourist destinations, seaside, mountainous areas, etc.; natural resources, such as hot springs, waterfalls, forests, etc.; natural disasters, such as earthquakes, floods, typhoons, etc.; social factors affecting passenger flow mainly include holidays and special events, such as the Spring Festival, National Day, Labor Day, etc.; and shopping mall openings, etc.; Since passenger flow generally changes over time, the collection of passenger flow data, natural influencing factor data, and social influencing factor data at multiple historical time points provides data support for subsequent predictions of passenger flow data at future times.

[0025] Preferably, S2 comprises the following steps:

[0026] S21. Setting the transportation track site set The abnormal passenger flow data interval of each transportation track station in is obtained, and the abnormal passenger flow data interval set is obtained.

[0027] and They represent the abnormal passenger flow data interval of the i-th transportation track station respectively;

[0028] According to the abnormal passenger flow data interval set

[0029] The abnormal passenger flow data in the passenger flow data matrix d3 of the historical passenger flow data of the station is extracted to obtain the abnormal passenger flow data matrix d′ of the historical passenger flow data of the station; as follows,

[0030]

[0031] Among them, d i ' j represents the passenger flow data of the jth transportation track station at the i-th historical time point when there is abnormal passenger flow data; Indicates the total number of historical time points with abnormal passenger flow data;

[0032] S22, classify the historical abnormal passenger flow data matrix d′ of the site to obtain a first abnormal passenger flow data classification matrix set represents the i-th abnormal passenger flow data classification matrix obtained after the classification operation, and f represents the total number of abnormal passenger flow data classification matrices obtained after the classification operation; as follows,

[0033]

[0034] in, express The passenger flow data of the kth station at the jth historical time point in express The total number of historical time points in the

[0035] Set the first abnormal passenger flow data classification matrix set The classification labels of each abnormal passenger flow data classification matrix in the first abnormal classification label set f′={f1′,f2′,...,f i ′,...,f f ′},f i ′ represents the classification label of the i-th abnormal passenger flow data classification matrix in the first abnormal passenger flow data classification matrix set;

[0036] The site historical passenger flow data matrix d3 is divided into the first abnormal passenger flow data classification matrix set The historical passenger flow data other than the above is used as the normal classification label;

[0037] S23, horizontally merging the historical natural influencing factor data matrix d1 and the historical social influencing factor data matrix d2 to obtain a first historical comprehensive influencing factor data matrix;

[0038] According to the first abnormal classification label set f′={f1′,f2′,...,f i ′,...,f f ′}, the first abnormal passenger flow data classification matrix set And the normal classification label obtains the classification label of each row of data in the first historical comprehensive influencing factor data matrix, and obtains the historical comprehensive influencing factor data classification label set e i Represents the classification label of the comprehensive influencing factor data at the i-th historical time point in the first historical comprehensive influencing factor data matrix;

[0039] S24, constructing an initial SVM classification model, using the historical comprehensive influencing factor data classification label set and the first historical comprehensive influencing factor data matrix to train and test the initial SVM classification model; after the training and testing are completed, the final SVM classification model is obtained;

[0040] By extracting abnormal data from the collected historical passenger flow data and setting classification labels, the complexity of the data is reduced, making subsequent classification easier.

[0041] Preferably, the S22 comprises the following steps:

[0042] S221, randomly select f rows of data from the historical abnormal passenger flow data matrix d' of the site as the initial classification center set Represents the randomly selected i-th row of data;

[0043] S222, calculate the historical abnormal passenger flow data matrix d' of the site except the initial classification center set Each row of data outside to the initial classification center set The Euclidean distance of each classification center in the matrix d′ of the historical abnormal passenger flow data of the station is selected as the classification of the row data;

[0044] After the calculation is completed, the initial abnormal passenger flow data classification matrix set is obtained;

[0045] S223, setting a discreteness threshold; calculating the sum of the Euclidean distances from each row of data in each classification matrix in the initial abnormal passenger flow data classification matrix set to the corresponding classification center to obtain an initial Euclidean distance sum value set; adding each Euclidean distance sum value in the initial Euclidean distance sum value set to obtain the initial Euclidean distance sum;

[0046] When the sum of the initial Euclidean distances is less than or equal to the discreteness threshold, the initial abnormal passenger flow data classification matrix set is used as the first abnormal passenger flow data classification matrix set. Otherwise, repeat S221 and S222 until the sum of the initial Euclidean distances is less than or equal to the discreteness threshold, and obtain the first abnormal passenger flow data classification matrix set

[0047] Preferably, the S24 comprises the following steps:

[0048] S241, setting a second training data ratio; and classifying the historical comprehensive influencing factor data into a label set according to the second training data ratio. and the first historical comprehensive influencing factor data matrix to perform data division to obtain a historical comprehensive influencing factor training data classification label set, a historical comprehensive influencing factor test data classification label set, a first historical comprehensive influencing factor training data matrix, and a first historical comprehensive influencing factor test data matrix;

[0049] S242, constructing an initial SVM classification model and setting a second training error threshold; inputting the historical comprehensive influencing factor training data classification label set and the first historical comprehensive influencing factor training data matrix into the initial SVM classification model for training operation; during the training process, when the training error is less than the second training error threshold, stopping the training to obtain a trained SVM classification model;

[0050] S243, setting a first accuracy rate threshold; inputting the historical comprehensive influencing factor test data classification label set and the first historical comprehensive influencing factor test data matrix into the trained SVM classification model for testing to obtain a first accuracy rate;

[0051] When the first accuracy rate is greater than or equal to a first accuracy rate threshold, the trained SVM classification model is used as the final SVM classification model; otherwise, the trained SVM classification model is continuously trained in the manner of S242 until the first accuracy rate is greater than or equal to the first accuracy rate threshold, thereby obtaining a final SVM classification model;

[0052] Preferably, S3 comprises the following steps:

[0053] S31, according to the passenger flow natural influencing factor set The set of social influencing factors of passenger flow b = {b1, b2, ..., b i ,...,b b′} and transportation rail station collection Collect real-time data on natural factors affecting passenger flow, social factors affecting passenger flow, and passenger flow data at sites to obtain a real-time data set of natural factors affecting passenger flow Real-time social influence factor dataset e′2={e′ 21 ,e′ 22 ,...,e′ 2i ,...,e′ 2b′} and real-time site passenger flow data matrix e1′ i , e′ 2i and e3′ i They respectively represent the i-th natural influencing factor data in the real-time passenger flow natural influencing factor data, the i-th social influencing factor data in the real-time passenger flow social influencing factor data, and the passenger flow data of the i-th station in the real-time station passenger flow data;

[0054] S32, using the method of S22 to analyze the real-time site passenger flow data matrix Perform classification to obtain the classification label of the real-time site passenger flow data matrix, which is recorded as the real-time classification label;

[0055] The natural influencing factors of passenger flow are grouped as The set of social influencing factors of passenger flow b = {b1, b2, ..., b i ,...,b b′} Input into the final SVM classification model to obtain the second classification label;

[0056] When the second classification label is different from the real-time classification label, updating the final SVM classification model;

[0057] When the second classification label is the same as the real-time classification label, there is no need to update the final SVM classification model;

[0058] The real-time collected data is input into the final SVM classification model to continuously test the final SVM classification model. When the test does not meet the requirements, it means that the current final SVM classification model cannot meet the changes in current human behavior patterns. Therefore, the final SVM classification model is continuously updated according to the test results, so that the final SVM classification model can always meet the classification requirements and ensure the accuracy of classification.

[0059] Preferably, updating the final SVM classification model in S32 includes the following steps:

[0060] S321, the real-time natural influencing factor data set Real-time social influence factor dataset e′2={e′ 21 ,e′ 22 ,...,e′ 2i ,...,e′ 2b′} and real-time classification labels and historical comprehensive influencing factor data classification label set in S24 and the first historical comprehensive influencing factor data matrix to obtain a combined historical comprehensive influencing factor data classification label set and a combined first historical comprehensive influencing factor data matrix;

[0061] S322, using the merged historical comprehensive influencing factor data classification label set and the merged first historical comprehensive influencing factor data matrix to train and test the final SVM classification model, and update the final SVM classification model;

[0062] Preferably, S4 comprises the following steps:

[0063] S41. Setting future time point set g i ′ represents the set i-th future time point, Indicates the total number of set future time points;

[0064] S42, the future time point set The natural influencing factor data and social influencing factor data at each future time point are predicted to obtain the future comprehensive influencing factor data matrix as follows,

[0065]

[0066] in, represents the data of the jth predictable factor at the i-th future time point;

[0067] By predicting the data of natural influencing factors and social influencing factors at future moments, data basis and support are provided for the subsequent classification and acquisition of passenger flow data at future moments.

[0068] Preferably, the S42 comprises the following steps:

[0069] S421, setting a first training data ratio and a first test data ratio;

[0070] According to the first training data ratio and the first test data ratio, the second historical comprehensive influencing factor data matrix is ​​divided into data to obtain a comprehensive influencing factor training data matrix, a comprehensive influencing factor test data matrix, and a comprehensive influencing factor to-be-predicted data matrix;

[0071] S422, constructing a first initial LSTM neural network model and setting a first training error threshold; inputting the comprehensive influencing factor training data matrix into the first initial LSTM neural network model for training; during the training process, when the training error is less than the first training error threshold, stopping the training to obtain a trained first LSTM neural network model; the weights and bias values ​​of each network node in the trained first LSTM neural network model constitute an initial network node parameter set h={(h 11 ,h 12 ),(h 21 ,h 22 ),...,(h i1 ,h i2 ),...,(h h′1 ,h h′2 )},h i1 and h i2 Respectively represent the weight and bias value of the i-th network node in the trained first LSTM neural network model, and h′ represents the total number of network nodes in the trained first LSTM neural network model;

[0072] S423: Setting the first test accuracy threshold The comprehensive influencing factor test data matrix is ​​input into the trained first LSTM neural network model for testing to obtain the first test accuracy.

[0073] when When the first LSTM neural network model is trained, the first LSTM neural network model is used as the first final LSTM neural network model; otherwise, the first LSTM neural network model is optimized; after the optimization is completed, the first final LSTM neural network model is obtained;

[0074] S424: Input the comprehensive influencing factor to-be-predicted data matrix into the first final LSTM neural network model for prediction operation to obtain the future comprehensive influencing factor data matrix

[0075] The LSTM neural network model has good performance in data fitting and prediction. Therefore, historical training data is used to train the LSTM neural network model, and then historical test data is used to test the trained LSTM neural network model. After the test, the LSTM neural network model has good prediction ability for the data of natural influencing factors of passenger flow and the data of social influencing factors of passenger flow.

[0076] Preferably, in S423, the trained first LSTM neural network model is optimized; after the optimization is completed, obtaining the first final LSTM neural network model includes the following steps:

[0077] S4231, build sparrow population represents the i-th sparrow in the sparrow population, l represents the size of the sparrow population; the maximum number of iterations of the sparrow population is set to l1′, the current number of iterations is set to l2′, and the dimension of the search space is set to 2·h′;

[0078] S4232, setting the value range of the weight value and the value range of the bias value of each network node in the trained first LSTM neural network model, and obtaining the network node weight value interval set And the network node bias value interval set and Respectively represent the lower limit and upper limit of the weight of the i-th network node in the trained first LSTM neural network model; and Respectively represent the lower limit and upper limit of the bias value of the i-th network node in the trained first LSTM neural network model;

[0079] According to the initial network node parameter set h={(h 11 ,h 12 ),(h 21 ,h 22 ),...,(h i1 ,h i2 ),...,(h h′1 ,h h′2 )}、Network node weight value interval set And the network node bias value interval set Set the initial position of each sparrow in the sparrow population and get the initial position matrix as follows,

[0080]

[0081] in, and They respectively represent the weight component and bias value component of the initial position of the i-th sparrow in the sparrow population at the j-th network node; the calculation formulas are as follows:

[0082]

[0083] In the formula, rand ij1 and rand ij2 Respectively for and Two random numbers generated;

[0084] S4233: Based on the first test accuracy threshold And the first test accuracy Set the fitness function m of the sparrow population as follows:

[0085]

[0086] In the formula, χ is a positive number, indicating the correction parameter;

[0087] S4234, start iterative operation; in each round of iteration, calculate the fitness value of each sparrow according to the fitness function m of the sparrow population to obtain a fitness value set; select the fitness value with the largest fitness value in the fitness value set and the position of the corresponding individual sparrow as the global optimal fitness value and the global optimal position respectively; update the position of each sparrow according to the global optimal fitness value and the global optimal position;

[0088] S4235. When l2′≥l1′, stop the iteration and get the optimal position set

[0089] m′={(m1′1,m1′2),(m′ 21 ,m′ 22 ),...,(m i ′1,m i ′2),...,(m′ h′1 ,m′ h′2 )},m i ′1 and m i ′2 represents the weight component and bias value component representing the i-th network node in the optimal position set;

[0090] The optimal position set m′={(m1′1,m1′2),(m′ 21 ,m′ 22 ),...,(m i ′1,m i ′2),...,(m′ h′1 ,m′ h′2)} is substituted into the trained first LSTM neural network model to obtain the optimized first LSTM neural network model;

[0091] S4236: Input the comprehensive influencing factor test data matrix in S423 into the optimized first LSTM neural network model for testing to obtain the second test accuracy. when , the optimized first LSTM neural network model is used as the first final LSTM neural network model; otherwise, the iterative operation of S4234 is continued until So far, the first final LSTM neural network model is obtained;

[0092] Preferably, the S5 comprises the following steps:

[0093] S51, the future comprehensive influencing factor data matrix Input into the final SVM classification model for classification operation; after the classification is completed, the future comprehensive influencing factor data matrix is ​​output The classification label corresponding to each row of data in the future classification label set is obtained Represents the data matrix of future comprehensive influencing factors The classification label corresponding to the comprehensive influencing factor data at the i-th future time point in ;

[0094] S52: Based on the future classification label set Determine whether there will be abnormal passenger flow at urban rail transit stations in the future;

[0095] By classifying the comprehensive influencing factor data at the predicted future time, the corresponding passenger flow category can be obtained, so that the passenger flow situation at the future time can be determined, and corresponding measures can be taken in advance to cope with it.

[0096] An artificial intelligence-based urban rail transit abnormal passenger flow prediction system, comprising a historical data collection module, an abnormal passenger flow data extraction module, a classification module, a classification model construction module, a real-time data collection module, a classification model real-time update module, a prediction module and a final judgment module;

[0097] The historical data collection module is used to collect multiple groups of different types of historical data of natural influencing factors, historical data of social influencing factors, and historical passenger flow data of different rail transit stations to obtain a historical natural influencing factor data matrix, a historical social influencing factor data matrix, and a station historical passenger flow data matrix;

[0098] The abnormal passenger flow data extraction module is used to extract the abnormal passenger flow data in the site historical passenger flow data matrix to obtain the site historical abnormal passenger flow data matrix;

[0099] The classification module is used to classify the historical abnormal passenger flow data matrix of the site to obtain a first abnormal passenger flow data classification matrix set;

[0100] The classification model building module is used to construct a final SVM classification model using the first abnormal passenger flow data classification matrix set, the historical natural influencing factor data matrix, and the historical social influencing factor data matrix;

[0101] The real-time data collection module is used to collect multiple groups of different types of real-time data of natural influencing factors, real-time data of social influencing factors, and real-time passenger flow data of different rail transit stations;

[0102] The classification model is used for real-time updating module to update the final SVM classification model in real time according to the real-time data of natural influencing factors, the real-time data of social influencing factors and the real-time passenger flow data;

[0103] The prediction module is used to predict the natural influencing factor data and social influencing factor data at future moments to obtain a future comprehensive influencing factor data matrix;

[0104] The final determination module is used to perform classification operations on the future comprehensive influencing factor data matrix using the final SVM classification model, and determine whether abnormal passenger flow phenomena will occur at urban rail transit stations at future times based on the classification results.

[0105] The present invention has the following beneficial effects:

[0106] 1. In the present invention, the classified historical abnormal passenger flow data of the station, the historical data of natural influencing factors and the historical data of social influencing factors are used to train and test the SVM machine learning model, so that the SVM classification model can determine the corresponding passenger flow category, and provide a more accurate classification model for the subsequent classification of the predicted historical data of natural influencing factors and the historical data of social influencing factors; secondly, different types of real-time data of natural influencing factors, real-time data of social influencing factors and real-time passenger flow data of different rail transit stations are collected in real time, and the final SVM classification model is continuously updated according to these real-time data to adapt to the constantly changing behavior habits and patterns of people, thereby improving the accuracy of the subsequent classification of the predicted data; the passenger flow data is indirectly predicted through the natural influencing factor data and social influencing factor data that can be predicted, thereby improving the accuracy of the prediction.

[0107] 2. In the present invention, by continuously iterating the classification center, each comprehensive data is classified according to the selected classification center so that the overall discreteness of the classified data is minimized, reflecting that the classification effect is best at this time, thereby obtaining the best classification.

[0108] 3. In the present invention, the sparrow optimization algorithm is used to perform multiple iterative optimizations on the weights and bias values ​​of each network node in the LSTM neural network model that does not meet the test requirements. By randomly generating multiple initial values ​​of weights and bias values, the solution space of the sparrow optimization algorithm is enriched, and the optimal weights and bias values ​​can be more easily searched, so that the LSTM neural network model can meet the test requirements.

[0109] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0110] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.

[0111] Figure 1 The present invention is a schematic diagram of a process of predicting future passenger flow by an abnormal passenger flow prediction system for urban rail transit based on artificial intelligence. DETAILED DESCRIPTION

[0112] The following will be combined with the drawings in the embodiments of the invention to clearly and completely describe the technical solutions in the embodiments of the invention. Obviously, the described embodiments are only part of the embodiments of the invention, not all of the embodiments. Based on the embodiments in the invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the invention.

[0113] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0114] Embodiment 1

[0115] This embodiment is a method for predicting abnormal passenger flow in urban rail transit based on artificial intelligence, comprising the following steps:

[0116] S1. The historical data collection module collects multiple groups of different types of historical data on natural influencing factors, historical data on social influencing factors, and historical passenger flow data at different rail transit stations;

[0117] The S1 comprises the following steps:

[0118] S11. Set the natural influencing factors of passenger flow and the set of social influencing factors of passenger flow b = {b1,b2,...,b i ,...,b b′}; represents the set i-th natural influencing factor of passenger flow, Indicates the total number of social influencing factors of passenger flow; b i represents the i-th social influencing factor of passenger flow, b′ represents the total number of social influencing factors of passenger flow;

[0119] Re-set the transportation track station set represents the set i-th urban transportation rail station,

[0120] Indicates the total number of set transportation track stations;

[0121] S12. Set the time point set for historical data collection Indicates the set i-th historical data collection time point, Indicates the total number of set historical data collection time points;

[0122] According to the historical data collection time point set Collect the natural influencing factors of passenger flow at each time point The data of each natural influencing factor of passenger flow, the set of social influencing factors of passenger flow b = {b1, b2, ..., b i ,...,b b′} and the data of each social influencing factor of passenger flow and the transportation rail station set The passenger flow data of each transportation track station in is obtained; the historical natural influencing factor data matrix d1, the historical social influencing factor data matrix d2 and the station historical passenger flow data matrix d3 are obtained; d1, d2 and d3 are as follows,

[0123]

[0124] Among them, d 1ij d 2ij and d 3ijThey respectively represent the data of the jth natural influencing factor of passenger flow, the data of the jth social influencing factor of passenger flow, and the passenger flow data of the jth urban transportation rail station collected at the i-th historical data collection time point;

[0125] S2, an abnormal passenger flow data extraction module extracts abnormal passenger flow data from the historical passenger flow data to obtain a historical abnormal passenger flow data matrix of the station;

[0126] The classification module classifies the historical abnormal passenger flow data matrix of the site;

[0127] The classification model building module uses the classified historical abnormal passenger flow data of the station, historical data of natural influencing factors, and historical data of social influencing factors to build the final SVM classification model;

[0128] The S2 comprises the following steps:

[0129] S21. Setting the transportation track site set The abnormal passenger flow data interval of each transportation track station in is obtained, and the abnormal passenger flow data interval set is obtained.

[0130] and They represent the abnormal passenger flow data interval of the i-th transportation track station respectively;

[0131] According to the abnormal passenger flow data interval set

[0132] The abnormal passenger flow data in the passenger flow data matrix d3 of the historical passenger flow data of the station is extracted to obtain the abnormal passenger flow data matrix d′ of the historical passenger flow data of the station; as follows,

[0133]

[0134] Among them, d i ' j represents the passenger flow data of the jth transportation track station at the i-th historical time point when there is abnormal passenger flow data; Indicates the total number of historical time points with abnormal passenger flow data;

[0135] S22, classify the historical abnormal passenger flow data matrix d′ of the site to obtain a first abnormal passenger flow data classification matrix set represents the i-th abnormal passenger flow data classification matrix obtained after the classification operation, and f represents the total number of abnormal passenger flow data classification matrices obtained after the classification operation; as follows,

[0136]

[0137] in, express The passenger flow data of the kth station at the jth historical time point in express The total number of historical time points in the

[0138] Set the first abnormal passenger flow data classification matrix set The classification labels of each abnormal passenger flow data classification matrix in the first abnormal classification label set f′={f1′,f2′,...,f i ′,...,f f ′},f i ′ represents the classification label of the i-th abnormal passenger flow data classification matrix in the first abnormal passenger flow data classification matrix set;

[0139] The site historical passenger flow data matrix d3 is divided into the first abnormal passenger flow data classification matrix set The historical passenger flow data other than the above is used as the normal classification label;

[0140] The S22 comprises the following steps:

[0141] S221, randomly select f rows of data from the historical abnormal passenger flow data matrix d' of the site as the initial classification center set Represents the randomly selected i-th row of data;

[0142] S222, calculate the historical abnormal passenger flow data matrix d' of the site except the initial classification center set Each row of data outside to the initial classification center set The Euclidean distance of each classification center in the matrix d′ of the historical abnormal passenger flow data of the station is selected as the classification of the row data;

[0143] After the calculation is completed, the initial abnormal passenger flow data classification matrix set is obtained;

[0144] S223, setting a discreteness threshold; calculating the sum of the Euclidean distances from each row of data in each classification matrix in the initial abnormal passenger flow data classification matrix set to the corresponding classification center to obtain an initial Euclidean distance sum value set; adding each Euclidean distance sum value in the initial Euclidean distance sum value set to obtain the initial Euclidean distance sum;

[0145] When the sum of the initial Euclidean distances is less than or equal to the discreteness threshold, the initial abnormal passenger flow data classification matrix set is used as the first abnormal passenger flow data classification matrix set. Otherwise, repeat S221 and S222 until the sum of the initial Euclidean distances is less than or equal to the discreteness threshold, and obtain the first abnormal passenger flow data classification matrix set

[0146] S23, horizontally merging the historical natural influencing factor data matrix d1 and the historical social influencing factor data matrix d2 to obtain a first historical comprehensive influencing factor data matrix;

[0147] According to the first abnormal classification label set f′={f1′,f2′,...,f i ′,...,f f ′}, the first abnormal passenger flow data classification matrix set And the normal classification label obtains the classification label of each row of data in the first historical comprehensive influencing factor data matrix, and obtains the historical comprehensive influencing factor data classification label set e i Represents the classification label of the comprehensive influencing factor data at the i-th historical time point in the first historical comprehensive influencing factor data matrix;

[0148] S24, constructing an initial SVM classification model, using the historical comprehensive influencing factor data classification label set and the first historical comprehensive influencing factor data matrix to train and test the initial SVM classification model; after the training and testing are completed, the final SVM classification model is obtained;

[0149] S3, the real-time data collection module collects different types of real-time data of natural influencing factors, real-time data of social influencing factors, and real-time passenger flow data of different rail transit stations;

[0150] The classification model real-time update module updates the final SVM classification model in real time according to the real-time data of natural influencing factors, real-time data of social influencing factors and real-time passenger flow data;

[0151] The S3 comprises the following steps:

[0152] S31, according to the passenger flow natural influencing factor set The set of social influencing factors of passenger flow b = {b1, b2, ..., b i ,...,b b′} and transportation rail station collection Collect real-time data on natural factors affecting passenger flow, social factors affecting passenger flow, and passenger flow data at sites to obtain a real-time data set of natural factors affecting passenger flow Real-time social influence factor dataset e′2={e′ 21 ,e′ 22 ,...,e′2i ,...,e′ 2b′} and real-time site passenger flow data matrix e1′ i , e′ 2i and e3′ i They respectively represent the i-th natural influencing factor data in the real-time passenger flow natural influencing factor data, the i-th social influencing factor data in the real-time passenger flow social influencing factor data, and the passenger flow data of the i-th station in the real-time station passenger flow data;

[0153] S32, using the method of S22 to analyze the real-time site passenger flow data matrix Perform classification to obtain the classification label of the real-time site passenger flow data matrix, which is recorded as the real-time classification label;

[0154] The natural influencing factors of passenger flow are grouped as The set of social influencing factors of passenger flow b = {b1, b2, ..., b i ,...,b b′} Input into the final SVM classification model to obtain the second classification label;

[0155] When the second classification label is different from the real-time classification label, updating the final SVM classification model;

[0156] When the second classification label is the same as the real-time classification label, there is no need to update the final SVM classification model;

[0157] S4, the prediction module predicts the natural influencing factor data and social influencing factor data at future moments to obtain a future comprehensive influencing factor data matrix;

[0158] The S4 comprises the following steps:

[0159] S41. Setting future time point set g i ′ represents the set i-th future time point, Indicates the total number of set future time points;

[0160] S42, the future time point set The natural influencing factor data and social influencing factor data at each future time point are predicted to obtain the future comprehensive influencing factor data matrix as follows,

[0161]

[0162] in, represents the data of the jth predictable factor at the i-th future time point;

[0163] The S42 comprises the following steps:

[0164] S421, setting a first training data ratio and a first test data ratio;

[0165] According to the first training data ratio and the first test data ratio, the second historical comprehensive influencing factor data matrix is ​​divided into data to obtain a comprehensive influencing factor training data matrix, a comprehensive influencing factor test data matrix, and a comprehensive influencing factor to-be-predicted data matrix;

[0166] S422, constructing a first initial LSTM neural network model and setting a first training error threshold; inputting the comprehensive influencing factor training data matrix into the first initial LSTM neural network model for training; during the training process, when the training error is less than the first training error threshold, stopping the training to obtain a trained first LSTM neural network model; the weights and bias values ​​of each network node in the trained first LSTM neural network model constitute an initial network node parameter set h={(h 11 ,h 12 ),(h 21 ,h 22 ),...,(h i1 ,h i2 ),...,(h h′1 ,h h′2 )},h i1 and h i2 Respectively represent the weight and bias value of the i-th network node in the trained first LSTM neural network model, and h′ represents the total number of network nodes in the trained first LSTM neural network model;

[0167] S423: Setting the first test accuracy threshold The comprehensive influencing factor test data matrix is ​​input into the trained first LSTM neural network model for testing to obtain the first test accuracy.

[0168] when When the first LSTM neural network model is trained, the first LSTM neural network model is used as the first final LSTM neural network model; otherwise, the first LSTM neural network model is optimized; after the optimization is completed, the first final LSTM neural network model is obtained;

[0169] In S423, the trained first LSTM neural network model is optimized; after the optimization is completed, obtaining the first final LSTM neural network model includes the following steps:

[0170] S4231, build sparrow population represents the i-th sparrow in the sparrow population, l represents the size of the sparrow population; the maximum number of iterations of the sparrow population is set to l1′, the current number of iterations is set to l2′, and the dimension of the search space is set to 2·h′;

[0171] S4232, setting the value range of the weight value and the value range of the bias value of each network node in the trained first LSTM neural network model, and obtaining the network node weight value interval set And the network node bias value interval set and Respectively represent the lower limit and upper limit of the weight of the i-th network node in the trained first LSTM neural network model; and Respectively represent the lower limit and upper limit of the bias value of the i-th network node in the trained first LSTM neural network model;

[0172] According to the initial network node parameter set h={(h 11 ,h 12 ),(h 21 ,h 22 ),...,(h i1 ,h i2 ),...,(h h′1 ,h h′2 )}、Network node weight value interval set And the network node bias value interval set Set the initial position of each sparrow in the sparrow population and get the initial position matrix as follows,

[0173]

[0174] in, and They respectively represent the weight component and bias value component of the initial position of the i-th sparrow in the sparrow population at the j-th network node; the calculation formulas are as follows:

[0175]

[0176] In the formula, rand ij1 and rand ij2 Respectively for and Two random numbers generated;

[0177] S4233: Based on the first test accuracy threshold And the first test accuracy Set the fitness function m of the sparrow population as follows:

[0178]

[0179] In the formula, χ is a positive number, indicating the correction parameter;

[0180] S4234, start iterative operation; in each round of iteration, calculate the fitness value of each sparrow according to the fitness function m of the sparrow population to obtain a fitness value set; select the fitness value with the largest fitness value in the fitness value set and the position of the corresponding individual sparrow as the global optimal fitness value and the global optimal position respectively; update the position of each sparrow according to the global optimal fitness value and the global optimal position;

[0181] S4235. When l2′≥l1′, stop the iteration and get the optimal position set

[0182] m′={(m1′1,m1′2),(m′ 21 ,m′ 22 ),...,(m i ′1,m i ′2),...,(m′ h′1 ,m′ h′2 )},m i ′1 and m i ′2 represents the weight component and bias value component representing the i-th network node in the optimal position set;

[0183] The optimal position set m′={(m1′1,m1′2),(m′ 21 ,m′ 22 ),...,(m i ′1,m i ′2),...,(m′ h′1 ,m′ h′2 )} is substituted into the trained first LSTM neural network model to obtain the optimized first LSTM neural network model;

[0184] S4236: Input the comprehensive influencing factor test data matrix in S423 into the optimized first LSTM neural network model for testing to obtain the second test accuracy. when , the optimized first LSTM neural network model is used as the first final LSTM neural network model; otherwise, the iterative operation of S4234 is continued until So far, the first final LSTM neural network model is obtained;

[0185] S424: Input the comprehensive influencing factor to-be-predicted data matrix into the first final LSTM neural network model for prediction operation to obtain the future comprehensive influencing factor data matrix

[0186] S5, the final judgment module uses the final SVM classification model to classify the future comprehensive influencing factor data matrix, and determines whether there will be abnormal passenger flow phenomena at urban rail transit stations at future times based on the classification results;

[0187] The S5 comprises the following steps:

[0188] S51, the future comprehensive influencing factor data matrix Input into the final SVM classification model for classification operation; after the classification is completed, the future comprehensive influencing factor data matrix is ​​output The classification label corresponding to each row of data in the future classification label set is obtained Represents the data matrix of future comprehensive influencing factors The classification label corresponding to the comprehensive influencing factor data at the i-th future time point in ;

[0189] S52: Based on the future classification label set Determine whether abnormal passenger flow will occur at urban rail transit stations in the future.

[0190] Embodiment 2

[0191] This embodiment discloses an abnormal passenger flow prediction system for urban rail transit based on artificial intelligence, which can implement the method of the above embodiment, including a historical data collection module, an abnormal passenger flow data extraction module, a classification module, a classification model construction module, a real-time data collection module, a classification model real-time update module, a prediction module and a final determination module;

[0192] The historical data collection module is used to collect multiple groups of different types of historical data of natural influencing factors, historical data of social influencing factors, and historical passenger flow data of different rail transit stations to obtain a historical natural influencing factor data matrix, a historical social influencing factor data matrix, and a station historical passenger flow data matrix;

[0193] The abnormal passenger flow data extraction module is used to extract the abnormal passenger flow data in the site historical passenger flow data matrix to obtain the site historical abnormal passenger flow data matrix;

[0194] The classification module is used to classify the historical abnormal passenger flow data matrix of the site to obtain a first abnormal passenger flow data classification matrix set;

[0195] The classification model building module is used to construct a final SVM classification model using the first abnormal passenger flow data classification matrix set, the historical natural influencing factor data matrix, and the historical social influencing factor data matrix;

[0196] The real-time data collection module is used to collect multiple groups of different types of real-time data of natural influencing factors, real-time data of social influencing factors, and real-time passenger flow data of different rail transit stations;

[0197] The classification model is used for real-time updating module to update the final SVM classification model in real time according to the real-time data of natural influencing factors, the real-time data of social influencing factors and the real-time passenger flow data;

[0198] The prediction module is used to predict the natural influencing factor data and social influencing factor data at future moments to obtain a future comprehensive influencing factor data matrix;

[0199] The final determination module is used to perform classification operations on the future comprehensive influencing factor data matrix using the final SVM classification model, and determine whether abnormal passenger flow phenomena will occur at urban rail transit stations at future times based on the classification results.

[0200] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0201] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. An artificial intelligence-based method for predicting abnormal passenger flow in urban rail transit, characterized in that: The following steps are involved: S1. The historical data collection module collects multiple groups of different types of historical data on natural influencing factors, historical data on social influencing factors, and historical passenger flow data at different rail transit stations; S2, the abnormal passenger flow data extraction module extracts the abnormal passenger flow data in the historical passenger flow data to obtain the site historical abnormal passenger flow data matrix; the classification module classifies the site historical abnormal passenger flow data matrix; the classification model construction module uses the classified site historical abnormal passenger flow data, natural influencing factor historical data and social influencing factor historical data to construct the final SVM classification model; S3, the real-time data collection module collects different types of real-time data of natural influencing factors, real-time data of social influencing factors, and real-time passenger flow data of different rail transit stations; The classification model real-time update module updates the final SVM classification model in real time according to the real-time data of natural influencing factors, real-time data of social influencing factors and real-time passenger flow data; S4, the prediction module predicts the natural influencing factor data and social influencing factor data at future moments to obtain a future comprehensive influencing factor data matrix; S5. The final judgment module uses the final SVM classification model to classify the future comprehensive influencing factor data matrix, and determines whether abnormal passenger flow will occur at urban rail transit stations in the future based on the classification results.

2. According to the artificial intelligence-based urban rail transit abnormal passenger flow prediction method of claim 1, it is characterized in that: The S1 comprises the following steps: S11, setting a set of natural influencing factors of passenger flow and a set of social influencing factors of passenger flow; and then setting a set of transportation track stations; S12, setting a historical data collection time point set; According to the historical data collection time point set, data on natural influencing factors, data on social influencing factors of passenger flow, and passenger flow data of transportation rail stations are collected at each time point; a historical natural influencing factor data matrix, a historical social influencing factor data matrix, and a station historical passenger flow data matrix are obtained.

3. According to claim 2, the method for predicting abnormal passenger flow of urban rail transit based on artificial intelligence is characterized in that: The S2 comprises the following steps: S21, setting the abnormal passenger flow data interval of each transportation track station in the transportation track station set to obtain an abnormal passenger flow data interval set; extracting the abnormal passenger flow data in the passenger flow data in the station historical passenger flow data matrix according to the abnormal passenger flow data interval set to obtain the station historical abnormal passenger flow data matrix; S22, classify the historical abnormal passenger flow data matrix of the site to obtain a first abnormal passenger flow data classification matrix set; set the classification label of each abnormal passenger flow data classification matrix in the first abnormal passenger flow data classification matrix set to obtain a first abnormal classification label set; use the historical passenger flow data in the historical passenger flow data matrix of the site except the first abnormal passenger flow data classification matrix set as normal classification labels; S23, horizontally merging the historical natural influencing factor data matrix and the historical social influencing factor data matrix to obtain a first historical comprehensive influencing factor data matrix; obtaining a classification label for each row of data in the first historical comprehensive influencing factor data matrix to obtain a historical comprehensive influencing factor data classification label set; S24, construct an initial SVM classification model and perform training and testing; after the training and testing are completed, the final SVM classification model is obtained.

4. According to claim 3, the method for predicting abnormal passenger flow of urban rail transit based on artificial intelligence is characterized in that: The S22 comprises the following steps: S221, randomly selecting a number of rows of data from the historical abnormal passenger flow data matrix of the site as an initial classification center set; S222, calculating the Euclidean distance of each row of data in the historical abnormal passenger flow data matrix of the station except the initial classification center set to each classification center in the initial classification center set; selecting the classification center with the smallest Euclidean distance to the corresponding row of data in the historical abnormal passenger flow data matrix of the station as the classification of the row of data; after the calculation is completed, the initial abnormal passenger flow data classification matrix set is obtained; S223, set a discreteness threshold; calculate the sum of the Euclidean distances from each row of data in each classification matrix in the initial abnormal passenger flow data classification matrix set to the corresponding classification center to obtain the sum of initial Euclidean distances; when the sum of initial Euclidean distances is less than or equal to the discreteness threshold, use the initial abnormal passenger flow data classification matrix set as the first abnormal passenger flow data classification matrix set; otherwise, repeat S221 and S222 until the sum of initial Euclidean distances is less than or equal to the discreteness threshold to obtain the first abnormal passenger flow data classification matrix set.

5. According to claim 4, the method for predicting abnormal passenger flow of urban rail transit based on artificial intelligence is characterized in that: The S3 comprises the following steps: S31, collecting real-time data on natural influencing factors of passenger flow, data on social influencing factors of passenger flow, and passenger flow data of sites, and obtaining a real-time data set of natural influencing factors, a real-time data set of social influencing factors, and a real-time passenger flow data matrix of sites; S32. Classify the real-time site passenger flow data matrix using the method of S22 to obtain the classification label of the real-time site passenger flow data matrix, which is recorded as the real-time classification label; input the set of natural influencing factors of passenger flow and the set of social influencing factors of passenger flow into the final SVM classification model to obtain a second classification label; when the second classification label is different from the real-time classification label, update the final SVM classification model; otherwise, there is no need to update the final SVM classification model.

6. The method for predicting abnormal passenger flow of urban rail transit based on artificial intelligence according to claim 5 is characterized in that: The S4 comprises the following steps: S41, setting a future time point set; S42, predicting the natural influencing factor data and the social influencing factor data of each future time point in the future time point set to obtain a future comprehensive influencing factor data matrix.

7. The method for predicting abnormal passenger flow of urban rail transit based on artificial intelligence according to claim 6 is characterized in that: The S42 comprises the following steps: S421, dividing the second historical comprehensive influencing factor data matrix to obtain a comprehensive influencing factor training data matrix, a comprehensive influencing factor test data matrix, and a comprehensive influencing factor to-be-predicted data matrix; S422, constructing a first initial LSTM neural network model and setting a first training error threshold; inputting the comprehensive influencing factor training data matrix into the first initial LSTM neural network model for training; during the training process, when the training error is less than the first training error threshold, stopping the training to obtain a trained first LSTM neural network model; S423: Setting the first test accuracy threshold The comprehensive influencing factor test data matrix is ​​input into the trained first LSTM neural network model for testing to obtain the first test accuracy. when When the first LSTM neural network model is trained, the first LSTM neural network model is used as the first final LSTM neural network model; otherwise, the first LSTM neural network model is optimized; after the optimization is completed, the first final LSTM neural network model is obtained; S424, inputting the to-be-predicted data matrix of the comprehensive influencing factors into the first final LSTM neural network model for prediction operation to obtain the future comprehensive influencing factor data matrix.

8. The method for predicting abnormal passenger flow of urban rail transit based on artificial intelligence according to claim 7 is characterized in that: In S423, the trained first LSTM neural network model is optimized; after the optimization is completed, obtaining the first final LSTM neural network model includes the following steps: S4231, constructing a sparrow population; setting the maximum number of iterations of the sparrow population to l1′ and the current number of iterations to l2′; S4232, setting the initial position of each sparrow in the sparrow population to obtain an initial position matrix; S4233: Based on the first test accuracy threshold And the first test accuracy Set the fitness function of the sparrow population; S4234, start the iteration operation; in each round of iteration, calculate the fitness value of each sparrow according to the fitness function of the sparrow population and update the position of each sparrow; S4235. When l2′≥l1′, stop iteration and obtain an optimal position set; substitute the optimal position set into the trained first LSTM neural network model to obtain an optimized first LSTM neural network model; S4236: Input the comprehensive influencing factor test data matrix in S423 into the optimized first LSTM neural network model for testing to obtain the second test accuracy. when , the optimized first LSTM neural network model is used as the first final LSTM neural network model; otherwise, the iterative operation of S4234 is continued until So far, the first final LSTM neural network model is obtained.

9. The method for predicting abnormal passenger flow of urban rail transit based on artificial intelligence according to claim 8 is characterized in that: The S5 comprises the following steps: S51, inputting the future comprehensive influencing factor data matrix into the final SVM classification model for classification operation; after the classification is completed, outputting the classification label corresponding to each row of data in the future comprehensive influencing factor data matrix to obtain a future classification label set; S52. Determine whether abnormal passenger flow will occur at the urban rail transit station in the future according to the future classification label set.

10. A system for implementing the artificial intelligence-based urban rail transit abnormal passenger flow prediction method as described in any one of claims 1 to 9.

Images