A travel destination prediction method, system and storage medium based on Logit model

Abstract

The present invention discloses a method, system and storage medium for predicting travel destinations based on a Logit model, and belongs to the field of intelligent transportation technology. The method includes: S1: data acquisition stage, obtaining POI, basic OD matrix and distance matrix based on big data for each traffic community; S2: feature selection stage, constructing a lasso regression model between the travel volume of the community and the POI for feature selection; S3: selection model construction stage, constructing a multivariate Logit destination selection model based on the POI type after feature selection; S4: parameter calibration stage, using the basic OD matrix to calibrate the parameters of the multivariate Logit destination selection model; S5: travel prediction stage, using the calibrated multivariate Logit destination selection model to predict the travel volume and travel distribution after the POI scale changes. Using the lasso regression model for feature selection can effectively remove the collinearity problem between the number of various POIs, reduce the dimension of the Logit model parameter solution, and improve the model prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a travel destination prediction method, system and storage medium based on a Logit model. Background Art

[0002] As the construction of large cities in China gradually stabilizes, the overall travel distribution characteristics of cities have basically taken shape and will not undergo significant structural changes in the future. At the same time, thanks to the rapid development of big data technology, through multi-source data processing technologies such as mobile phone signaling, floating car GPS, and Internet LBS, we can obtain the temporal and spatial distribution data of urban residents' travel with high accuracy, providing a solid data foundation for transportation research.

[0003] However, with the in-depth implementation of projects such as urban renewal and urban village reconstruction, how to scientifically, efficiently and quickly predict and judge the changes in regional travel demand caused by changes in land use nature and business scale in local areas is an urgent problem to be solved in current transportation research. Traditionally, the gravity model is widely used in travel distribution prediction due to its solid theoretical foundation and mature calibration software. However, the gravity model has significant limitations. It mainly relies on traffic impedance and travel generation and attraction as variables to determine travel distribution. It is difficult to deeply explain complex destination selection behaviors, and it is also impossible to effectively reflect the dynamic impact of business scale changes on travel distribution. Although the gravity model is adjusted by introducing the K coefficient to improve the prediction accuracy, the determination of the K coefficient lacks a clear theoretical basis, and it is difficult to reveal the real mechanism behind the model adjustment.

[0004] In order to overcome the shortcomings of the traditional gravity model, the destination selection model based on the Logit model has gradually become a research hotspot. In particular, the multinomial Logit model, due to its simplicity and applicability, has shown great potential in the destination selection problem. However, existing studies or invention patents lack research on the selection of characteristic variables of the utility function when constructing the Logit model. In terms of model parameter calibration, traditional methods mainly rely on sampling data, but the sampling data often has a small number of samples and cannot fully reflect the overall characteristics of travel. In recent years, with the promotion and application of big data, a large number of studies have begun to use big data for model construction. However, the global origin-destination (OD) matrix obtained in the big data environment generally contains a large number of data points with zero values ​​or extremely low travel volume. Directly applying the traditional maximum likelihood estimation (MLE) method for model parameter calibration often leads to the model being overly inclined to OD pairs with no travel or low travel volume when predicting due to the high sparsity of the global OD matrix, while relatively ignoring those OD pairs with large travel volume but small number, thereby weakening the prediction accuracy and robustness of the model. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a travel destination prediction method, system and storage medium based on a Logit model.

[0006] The objective of the present invention is achieved through the following technical solutions: The first aspect of the present invention provides: a travel destination prediction method based on a Logit model, comprising the following steps:

[0007] S1: Data acquisition stage, obtaining POI, basic OD matrix and distance matrix of each traffic community based on big data;

[0008] S2: Feature selection stage, a lasso regression model is constructed between the travel volume in the community and POI for feature selection;

[0009] S3: In the model building phase, a multivariate Logit destination selection model is constructed based on the POI types after feature selection;

[0010] S4: parameter calibration stage, the basic OD matrix is ​​used to calibrate the parameters of the multivariate Logit destination choice model;

[0011] S5: In the travel prediction stage, the calibrated multivariate Logit destination selection model is used to predict the travel volume and distribution after the POI scale changes.

[0012] Preferably, the S1: data acquisition stage further includes the following steps:

[0013] S11: Define the scope of model application;

[0014] S12: Divide the model application scope into multiple traffic zones according to the model granularity requirements;

[0015] S13: Acquire POI data with longitude and latitude coordinates within the target range and OD data based on LBS, where the data attributes of the POI data at least include POI major category, POI medium category, longitude, and latitude fields; the data attributes of the OD data based on LBS at least include starting point coordinates, end point coordinates, travel volume, and travel distance fields;

[0016] S14: Associating the POI data with longitude and latitude coordinates and the OD data based on LBS with the traffic districts, and performing spatial correlation analysis, classifying and summing the POI data by traffic districts, and obtaining the number X of each type of POI in each traffic district; summing the travel volume of the LBS data by the starting traffic district and the starting and ending traffic districts, respectively, to obtain the travel volume P of each traffic district and the travel volume matrix Y; averaging the travel distances of the LBS data by the starting and ending traffic districts, and obtaining the distance matrix D between each traffic district.

[0017] Preferably, the S2: feature selection stage further includes the following steps:

[0018] S21: Construct a linear model of the travel volume of the traffic zone and the number of POIs in the traffic zone:

[0019] ;

[0020] In the formula, p i is the model prediction value of the travel volume in the ith traffic zone; x ik is the number of POIs of the kth category contained in the i-th traffic zone; , ε are the parameters to be fitted;

[0021] S22: Construct a lasso model of fitting parameters:

[0022] ;

[0023] In the formula, represents the objective function of parameter optimization; is the standardized form of the residual sum of squares, which is used to measure the degree of fit to the data; is the actual value of the travel volume in the ith traffic zone; is the L1 regularization term, where is the regularization parameter used to control the strength of regularization; n is the number of samples; k is the number of POI types;

[0024] S23: Use cross-validation to determine the best , using the coordinate descent algorithm to find the optimal ;

[0025] S24: Select POI types that are not equal to zero are used as input variables SX in the multivariate logit destination choice model.

[0026] Preferably, the step S3: selecting a model building stage further comprises the following steps:

[0027] S31: Construct the utility function of the destination traffic zone: In the formula, Select the utility value of the jth traffic zone for the i-th traffic zone; for observable determined utility; is a random item;

[0028] S32: Assume that the utility function consists of the interaction benefits and interaction costs between the i-th traffic zone and the j-th traffic zone, then: In the formula, is the interaction benefit between the ith traffic area and the jth traffic area; is the interaction cost between the ith traffic area and the jth traffic area; is the intercept term;

[0029] S33: According to the relationship that the gravitational force between two objects is proportional to their mass, and the application of the Weber-Fechner law in determining the benefit function, It is expressed as:

[0030] ;

[0031] In the formula, is the interaction correlation coefficient between m-type POI and n-type POI; is the number of POIs of the mth type contained in the i-th traffic zone; is the number of POIs of the nth type contained in the jth traffic zone; S is the number of POI types used to construct the profit function;

[0032] S34: Yes To make corrections:

[0033] ;

[0034] S35: Based on the approximate logarithmic relationship between travel cost and travel distance, It is expressed as: In the formula, is the travel cost conversion factor; is the comprehensive travel distance between the ith traffic area and the jth traffic area;

[0035] S36: Assume that the random terms in S31 obey the standard Gumbel distribution that is independent and identically distributed. Then the discrete choice model of the i-th traffic zone choosing the j-th traffic zone for travel is expressed as:

[0036] ;

[0037] S37: According to S32-S36, The final expression of is:

[0038] .

[0039] Preferably, the said S4: parameter calibration stage further comprises the following steps:

[0040] S41: Calculate the OD travel ratio matrix based on the basic OD matrix ;

[0041] S42: Introduce the weighted cross entropy function as the objective function of parameter estimation:

[0042] ;

[0043] In the formula, is the weight coefficient of the i-th traffic zone and the j-th traffic zone in the parameter calibration process; is the proportion of all trips departing from the ith transportation zone in the basic OD matrix that select the jth transportation zone;

[0044] S43: Determine the weight coefficient of each travel ratio data in the weighted cross entropy function;

[0045] S44: Use the BFGS algorithm to solve the model parameters and obtain the optimized ;

[0046] S45: McFadden's coefficient of determination is used to evaluate the fitting accuracy of the multivariate Logit destination choice model, and a calibrated multivariate Logit destination choice model is obtained by achieving the preset fitting accuracy.

[0047] Preferably, the S43 further comprises the following steps:

[0048] S431: Calculate the average value of the number of trips departing from the i-th traffic zone:

[0049] ;

[0050] In the formula, p i is the travel volume of the ith traffic zone; N i is the number of optional destination transportation zones for the ith transportation zone;

[0051] S432: As the group distance, the OD data from the i-th traffic zone is divided into There are K non-overlapping intervals in total;

[0052] S433: Count the number of OD point pairs in each interval n k :

[0053] ;

[0054] In the formula, is an indicator function, indicating that if Falling in interval I k ,but is equal to 1, otherwise it is 0;

[0055] S434: Determine the weight value of the OD data by judging the number of data and the number of intervals in the interval where each OD data is located: , when calculating Falling in interval I k .

[0056] The second aspect of the present invention provides: a travel destination prediction system based on a Logit model, which is used to implement any of the above-mentioned travel destination prediction methods based on a Logit model, comprising:

[0057] The data acquisition module is used to obtain the POI, basic OD matrix and distance matrix of each traffic community based on big data;

[0058] The feature selection module is used to construct a lasso regression model between the travel volume in the community and POI for feature selection;

[0059] The selection model building module is used to build a multivariate Logit destination selection model based on the POI type after feature selection;

[0060] Parameter calibration module, used to calibrate the parameters of the multivariate Logit destination choice model using the basic OD matrix;

[0061] The travel prediction module is used to predict the travel volume and distribution after the POI scale changes using the calibrated multivariate Logit destination selection model.

[0062] The third aspect of the present invention provides: a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned travel destination prediction methods based on the Logit model is implemented.

[0063] The beneficial effects of the present invention are:

[0064] 1) Using the lasso regression model for feature selection can effectively remove the collinearity problem between the numbers of various POIs, reduce the dimension of the Logit model parameter solution, and improve the model prediction accuracy.

[0065] 2) The weighted cross-entropy function is used as the objective function of parameter estimation. By paying more attention to OD pairs with high travel volume but small number, the overfitting tendency of a large number of zero-value or low-value data is reduced, thereby optimizing the parameter calibration process of the model to ensure that the prediction results can reflect the general travel mode and accurately capture important abnormal or extreme travel behaviors; compared with the parameter estimation method based on traditional sampling survey samples, it effectively reduces the impact of the sparsity problem in the big data OD matrix on the model accuracy and improves the model prediction accuracy.

[0066] 3) By integrating multi-source big data resources, we deeply explore the intrinsic connection between urban POI and residents' travel behavior, enrich the theoretical system based on the Logit destination choice model, and provide a highly operational technical method for the application of POI in urban traffic travel distribution prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of the travel destination prediction method based on the Logit model. DETAILED DESCRIPTION

[0068] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0069] See also Figure 1 The first aspect of the present invention provides: a travel destination prediction method based on a Logit model, comprising the following steps:

[0070] S1: Data acquisition stage, obtaining POI, basic OD matrix and distance matrix of each traffic community based on big data;

[0071] S2: Feature selection stage, a lasso regression model is constructed between the travel volume in the community and POI for feature selection;

[0072] S3: In the model building phase, a multivariate Logit destination selection model is constructed based on the POI types after feature selection;

[0073] S4: parameter calibration stage, the basic OD matrix is ​​used to calibrate the parameters of the multivariate Logit destination choice model;

[0074] S5: In the travel prediction stage, the calibrated multivariate Logit destination selection model is used to predict the travel volume and distribution after the POI scale changes.

[0075] In some embodiments, the S1: data acquisition phase further includes the following steps:

[0076] S11: Define the scope of model application;

[0077] S12: Divide the model application scope into multiple traffic zones according to the model granularity requirements;

[0078] S13: Acquire POI data with longitude and latitude coordinates within the target range and OD data based on LBS, where the data attributes of the POI data at least include POI major category, POI medium category, longitude, and latitude fields; the data attributes of the OD data based on LBS at least include starting point coordinates, end point coordinates, travel volume, and travel distance fields;

[0079] S14: Associating the POI data with longitude and latitude coordinates and the OD data based on LBS with the traffic districts, and performing spatial correlation analysis, classifying and summing the POI data by traffic districts, and obtaining the number X of each type of POI in each traffic district; summing the travel volume of the LBS data by the starting traffic district and the starting and ending traffic districts, respectively, to obtain the travel volume P of each traffic district and the travel volume matrix Y; averaging the travel distances of the LBS data by the starting and ending traffic districts, and obtaining the distance matrix D between each traffic district.

[0080] In this embodiment, the LBS-based OD data can be acquired from a map service provider through purchase.

[0081] In some embodiments, the S2: feature selection stage further includes the following steps:

[0082] S21: Construct a linear model of the travel volume of the traffic zone and the number of POIs in the traffic zone:

[0083] ;

[0084] In the formula, p i is the model prediction value of the travel volume in the ith traffic zone; x ik is the number of POIs of the kth category contained in the i-th traffic zone; , ε are the parameters to be fitted;

[0085] S22: Construct a lasso model of fitting parameters:

[0086] ;

[0087] In the formula, represents the objective function of parameter optimization; is the standardized form of the residual sum of squares, which is used to measure the degree of fit to the data; is the actual value of the travel volume in the ith traffic zone; is the L1 regularization term, where is the regularization parameter used to control the strength of regularization; n is the number of samples; k is the number of POI types;

[0088] S23: Use cross-validation to determine the best , using the coordinate descent algorithm to find the optimal ;

[0089] S24: Select POI types that are not equal to zero are used as input variables SX in the multivariate logit destination choice model.

[0090] In this embodiment, step S2 can effectively handle the collinearity problem between POIs of different types, realize the key POI type characteristics that affect residents' travel, and reduce the complexity of the multivariate Logit destination selection model and the dimension of model parameters.

[0091] In some embodiments, the S3: selecting a model building stage further includes the following steps:

[0092] S31: Construct the utility function of the destination traffic zone: In the formula, Select the utility value of the jth traffic zone for the i-th traffic zone; for observable determined utility; is a random item;

[0093] S32: Assume that the utility function consists of the interaction benefits and interaction costs between the i-th traffic zone and the j-th traffic zone, then: In the formula, is the interaction benefit between the ith traffic area and the jth traffic area; is the interaction cost between the ith traffic area and the jth traffic area; is the intercept term;

[0094] S33: According to the relationship that the gravitational force between two objects is proportional to their mass, and the application of the Weber-Fechner law in determining the benefit function, It is expressed as:

[0095] ;

[0096] In the formula, is the interaction correlation coefficient between m-type POI and n-type POI; is the number of POIs of the mth type contained in the i-th traffic zone; is the number of POIs of the nth type contained in the jth traffic zone; S is the number of POI types used to construct the profit function;

[0097] S34: Yes To make corrections:

[0098] ;

[0099] S35: Based on the approximate logarithmic relationship between travel cost and travel distance, It is expressed as: In the formula, is the travel cost conversion factor; is the comprehensive travel distance between the ith traffic area and the jth traffic area;

[0100] S36: Assume that the random terms in S31 obey the standard Gumbel distribution that is independent and identically distributed. Then the discrete choice model of the i-th traffic zone choosing the j-th traffic zone for travel is expressed as:

[0101] ;

[0102] S37: According to S32-S36, The final expression of is:

[0103] .

[0104] In this embodiment, step S3 integrates the advantages of the gravity model and the Logit model, and incorporates the strength of association between different POI types at the starting and ending points and the impact of travel distance on travel into the model variable structure, effectively improving the model's sensitivity to the mutual influence between different types of POIs. In step S33, the interaction benefit between the i-th traffic area and the j-th traffic area is closely related to the POI type and number of the starting and ending traffic areas, but lacks a specific functional relationship with the number of POIs in the traffic area. Therefore, the relationship between the gravity and mass of two objects in the gravity model is referenced, as well as the application definition of the Weber-Fechner law in determining the benefit function. In step S34, since not all traffic areas contain all types of POI data, This may cause data overflow. To avoid model data overflow, Make corrections.

[0105] In some embodiments, the S4: parameter calibration phase further includes the following steps:

[0106] S41: Calculate the OD travel ratio matrix based on the basic OD matrix ;

[0107] S42: Introduce the weighted cross entropy function as the objective function of parameter estimation:

[0108] ;

[0109] In the formula, is the weight coefficient of the i-th traffic zone and the j-th traffic zone in the parameter calibration process; is the proportion of all trips departing from the ith transportation zone in the basic OD matrix that select the jth transportation zone;

[0110] S43: Determine the weight coefficient of each travel ratio data in the weighted cross entropy function;

[0111] S44: Use the BFGS algorithm to solve the model parameters and obtain the optimized ;

[0112] S45: McFadden's coefficient of determination is used to evaluate the fitting accuracy of the multivariate Logit destination choice model, and a calibrated multivariate Logit destination choice model is obtained by achieving the preset fitting accuracy.

[0113] In this embodiment, in step S4, in order to improve the adaptability and prediction accuracy of the model to sparse OD matrix data, a weighted cross-entropy function is introduced as the objective function of parameter estimation. The weighted cross-entropy function can more evenly consider the contribution of various data points to model fitting by assigning different weights to OD observation data based on LBS big data (especially OD pairs with low travel volume and high travel volume). Specifically, this method mainly increases the attention to OD pairs with high travel volume but small number, reduces the tendency of overfitting to a large number of zero-value or low-value data, and then optimizes the parameter calibration process of the model to ensure that the prediction results can not only reflect the general travel mode, but also accurately capture important abnormal or extreme travel behaviors. In step S42, considering that the global origin-destination (OD) matrix constructed in a big data environment generally contains a large number of data points with zero values ​​or extremely low travel volume, the traditional maximum likelihood estimation (MLE) method is directly applied to calibrate the model parameters. Due to the high sparsity of the basic OD matrix, the model is often overly inclined to OD pairs with no travel or low travel volume when predicting, and relatively ignores those OD pairs with large travel volume but low data frequency, thereby weakening the prediction accuracy and robustness of the model. In order to effectively alleviate this problem and improve the adaptability and prediction accuracy of the model to sparse OD matrix data, the weighted cross-entropy function is introduced as the objective function of parameter estimation.

[0114] In some embodiments, the S43 further includes the following steps:

[0115] S431: Calculate the average value of the number of trips departing from the i-th traffic zone:

[0116] ;

[0117] In the formula, p iis the travel volume of the ith traffic zone; N i is the number of optional destination transportation zones for the ith transportation zone;

[0118] S432: As the group distance, the OD data from the i-th traffic zone is divided into There are K non-overlapping intervals in total;

[0119] S433: Count the number of OD point pairs in each interval n k :

[0120] ;

[0121] In the formula, is an indicator function, indicating that if Falling in interval I k ,but is equal to 1, otherwise it is 0;

[0122] S434: Determine the weight value of the OD data by judging the number of data and the number of intervals in the interval where each OD data is located: , when calculating Falling in interval I k .

[0123] The second aspect of the present invention provides: a travel destination prediction system based on a Logit model, which is used to implement any of the above-mentioned travel destination prediction methods based on a Logit model, comprising:

[0124] The data acquisition module is used to obtain the POI, basic OD matrix and distance matrix of each traffic community based on big data;

[0125] The feature selection module is used to construct a lasso regression model between the travel volume in the community and POI for feature selection;

[0126] The selection model building module is used to build a multivariate Logit destination selection model based on the POI type after feature selection;

[0127] Parameter calibration module, used to calibrate the parameters of the multivariate Logit destination choice model using the basic OD matrix;

[0128] The travel prediction module is used to predict the travel volume and distribution after the POI scale changes using the calibrated multivariate Logit destination selection model.

[0129] The third aspect of the present invention provides: a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned travel destination prediction methods based on the Logit model is implemented.

[0130] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not deviate from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A travel destination prediction method based on the Logit model, characterized by: The following steps are involved: S1: Data acquisition stage, obtaining POI, basic OD matrix and distance matrix of each traffic community based on big data; S2: Feature selection stage, a lasso regression model is constructed between the travel volume in the community and POI for feature selection; S3: In the model building phase, a multivariate Logit destination selection model is constructed based on the POI types after feature selection; S4: parameter calibration stage, the basic OD matrix is ​​used to calibrate the parameters of the multivariate Logit destination choice model; S5: In the travel prediction stage, the calibrated multivariate Logit destination selection model is used to predict the travel volume and travel distribution after the POI scale changes; The S2: feature selection stage further includes the following steps: S21: Construct a linear model of the travel volume of the traffic zone and the number of POIs in the traffic zone: ； In the formula, p i is the model prediction value of the travel volume in the ith traffic zone; x ik is the number of POIs of the kth category contained in the i-th traffic zone; , ε are the parameters to be fitted; S22: Construct a lasso model of fitting parameters: ； In the formula, represents the objective function of parameter optimization; is the standardized form of the residual sum of squares, which is used to measure the degree of fit to the data; is the actual value of the travel volume in the ith traffic zone; is the L1 regularization term, where is the regularization parameter used to control the strength of regularization; n is the number of samples; k is the number of POI types; S23: Use cross-validation to determine the best , using the coordinate descent algorithm to solve the optimal ; S24: Select POI types that are not equal to zero are used as input variables SX in the multivariate logit destination choice model; The S3: model building phase further includes the following steps: S31: Construct the utility function of the destination traffic zone: In the formula, Select the utility value of the jth traffic zone for the i-th traffic zone; for observable determined utility; is a random item; S32: Assume that the utility function consists of the interaction benefits and interaction costs between the i-th traffic zone and the j-th traffic zone, then: In the formula, is the interaction benefit between the ith traffic area and the jth traffic area; is the interaction cost between the ith traffic area and the jth traffic area; is the intercept term; S33: According to the relationship that the gravitational force between two objects is proportional to their mass, and the application of the Weber-Fechner law in determining the benefit function, It is expressed as: ； In the formula, is the interaction correlation coefficient between m-type POI and n-type POI; is the number of POIs of the mth type contained in the i-th traffic zone; is the number of POIs of the nth type contained in the jth traffic zone; S is the number of POI types used to construct the profit function; S34: Yes To make corrections: ； S35: Based on the approximate logarithmic relationship between travel cost and travel distance, It is expressed as: In the formula, is the travel cost conversion factor; is the comprehensive travel distance between the ith traffic area and the jth traffic area; S36: Assume that the random terms in S31 obey the standard Gumbel distribution that is independent and identically distributed. Then the discrete choice model of the i-th traffic zone choosing the j-th traffic zone for travel is expressed as: ； S37: According to S32-S36, The final expression of is: ； The S4: parameter calibration phase also includes the following steps: S41: Calculate the OD travel ratio matrix based on the basic OD matrix ; S42: Introduce the weighted cross entropy function as the objective function of parameter estimation: ； In the formula, is the weight coefficient of the i-th traffic zone and the j-th traffic zone in the parameter calibration process; is the proportion of all trips departing from the ith transportation zone in the basic OD matrix that select the jth transportation zone; S43: Determine the weight coefficient of each travel ratio data in the weighted cross entropy function; S44: Use the BFGS algorithm to solve the model parameters and obtain the optimized ; S45: McFadden's coefficient of determination is used to evaluate the fitting accuracy of the multivariate Logit destination choice model, and a calibrated multivariate Logit destination choice model is obtained by achieving the preset fitting accuracy.

2. The method for predicting travel destinations based on the Logit model according to claim 1, characterized in that: The S1: data acquisition phase further includes the following steps: S11: Define the scope of model application; S12: Divide the model application scope into multiple traffic zones according to the model granularity requirements; S13: Acquire POI data with longitude and latitude coordinates within the target range and OD data based on LBS, where the data attributes of the POI data at least include POI major category, POI medium category, longitude, and latitude fields; the data attributes of the OD data based on LBS at least include starting point coordinates, end point coordinates, travel volume, and travel distance fields; S14: Associating the POI data with longitude and latitude coordinates and the OD data based on LBS with the traffic districts, and performing spatial correlation analysis, classifying and summing the POI data by traffic districts, and obtaining the number X of each type of POI in each traffic district; summing the travel volume of the LBS data by the starting traffic district and the starting and ending traffic districts, respectively, to obtain the travel volume P of each traffic district and the travel volume matrix Y; averaging the travel distances of the LBS data by the starting and ending traffic districts, and obtaining the distance matrix D between each traffic district.

3. The method for predicting travel destinations based on the Logit model according to claim 1, characterized in that: The S43 further comprises the following steps: S431: Calculate the average value of the number of trips departing from the i-th traffic zone: ； In the formula, p i is the travel volume of the ith traffic zone; N i is the number of optional destination transportation zones for the ith transportation zone; S432: As the group distance, the OD data from the i-th traffic zone is divided into There are K non-overlapping intervals in total; S433: Count the number of OD point pairs in each interval n k : ； In the formula, is an indicator function, indicating that if Falling in interval I k ,but is equal to 1, otherwise it is 0; S434: Determine the weight value of the OD data by judging the number of data and the number of intervals in the interval where each OD data is located: , when calculating Falling in interval I k .

4. A travel destination prediction system based on the Logit model, characterized by: The method for predicting a travel destination based on a Logit model according to any one of claims 1 to 3 comprises: The data acquisition module is used to obtain the POI, basic OD matrix and distance matrix of each traffic community based on big data; The feature selection module is used to construct a lasso regression model between the travel volume in the community and POI for feature selection; The selection model building module is used to build a multivariate Logit destination selection model based on the POI type after feature selection; Parameter calibration module, used to calibrate the parameters of the multivariate Logit destination choice model using the basic OD matrix; The travel prediction module is used to use the calibrated multivariate Logit destination selection model to predict the travel volume and distribution after the POI scale changes.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by the processor, the travel destination prediction method based on the Logit model as described in any one of claims 1 to 3 is implemented.

Images