Individual travel purpose identification method and system based on mobile phone sensor positioning data

Abstract

The invention belongs to the technical field of individual travel purpose identification, and discloses an individual travel purpose identification method and system based on mobile phone sensor positioning data, and the method comprises the steps: obtaining the mobile phone sensor positioning data and individual travel log data of individual travel; obtaining an individual travel endpoint identification result according to the mobile phone sensor positioning data of individual travel; calculating an individual travel destination POI matching probability according to an individual travel endpoint identification result; according to the POI matching probability of the individual travel destination, identifying the individual travel destination through a Transform model; according to the method, data can be continuously and uninterruptedly collected, so that a large number of data samples are accumulated, the travel purpose and demand of a user can be reflected in time, an important basis is provided for making an urban traffic planning scheme and relieving urban traffic congestion, the traffic network design can be optimized, the traffic condition can be monitored in real time, and the traffic resource allocation rationality can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of individual travel purpose identification, and relates to an individual travel purpose identification method and system based on mobile phone sensor positioning data. Background Art

[0002] With the rapid economic development and continuous social progress in my country, traffic congestion is becoming increasingly serious. All travel behaviors are triggered by daily activities for specific purposes. Existing studies are mainly based on paper questionnaires, shared bicycles, bus card swiping, taxi orders and other data, and use multivariate discrete choice model probability inference methods to identify travel purposes. There are generally problems such as inaccurate identification of travel purposes and large errors in arrival and departure times.

[0003] In-depth research on the travel purpose identification method can deeply reveal the decision-making factors that drive residents' travel behavior, which is of far-reaching significance for analyzing residents' travel behavior and alleviating traffic congestion. At the same time, a deep understanding of the travel purpose of urban residents can more accurately grasp the transportation needs of urban residents, thereby providing an important basis for formulating more accurate transportation planning and transportation demand management policies. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention aims to provide a method and system for identifying individual travel purposes based on mobile phone sensor positioning data, obtain the mobile phone sensor positioning data and individual travel log data of individual travel; obtain the individual travel endpoint identification results and the individual travel destination POI matching probability; and identify the individual travel purpose.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: The present invention provides a method for identifying individual travel purposes based on mobile phone sensor positioning data, comprising: obtaining mobile phone sensor positioning data and individual travel log data of individual travel; obtaining individual travel endpoint recognition results according to the mobile phone sensor positioning data of individual travel; calculating the individual travel destination POI matching probability according to the individual travel endpoint recognition result; and identifying the individual travel purpose through a Transformer model according to the individual travel destination POI matching probability.

[0006] Furthermore, before obtaining the individual travel endpoint identification result, the method also includes: constructing a user travel attribute data set based on the mobile phone sensor positioning data of the individual travel, calculating the similarity of the positioning points, and iteratively calculating the personal support matrix and the public opinion support matrix based on the similarity.

[0007] Furthermore, the similarity of the positioning points is:

[0008] in, For individual Longitude information of the travel location, For individual Longitude information of the travel location, For individual The latitude information of the travel location, For individual The latitude information of the travel location.

[0009] Furthermore, the personal support matrix is:

[0010]

[0011] in, is the damping coefficient, is the personal support matrix (initial 0 matrix), is the personal support matrix after the iteration update, For positioning point and anchor points The similarity of is the similarity matrix Xingfeidi The value of the column, The first Xingfeidi The value of the column.

[0012] Furthermore, the public opinion support matrix is:

[0013] in, is the damping coefficient; is the public opinion support matrix at the current iteration; Include diagonal values and off-diagonal values

[0014] The diagonal value for:

[0015] in, The personal support matrix Column off-diagonal values; The off-diagonal values for:

[0016] in, is the diagonal value in the personal support matrix.

[0017] Furthermore, the mobile phone sensor positioning data of individual travel includes the order number, date, time, longitude, latitude and speed, and the individual travel log data includes the activity start time, activity end time and activity category.

[0018] Furthermore, the identification of the individual travel purpose through the Transformer model according to the POI matching probability of the individual travel destination includes: The individual travel endpoint recognition results and the individual travel destination POI matching probability are input into the Transformer model, and the travel activity information is put into the encoder for encoding. The Adam optimizer is used to update the optimization parameters and iterative training is performed. After the number of iterations reaches the preset value, the test data set is tested to obtain the classification accuracy that meets the standard and then the individual travel purpose is identified. The travel activity information includes travel time, stay duration, visit frequency, and destination type.

[0019] Furthermore, the attractiveness of the candidate POI to each travel endpoint is calculated based on the improved gravity model as follows:

[0020] in, Indicates at time Within the scope, The spatial attraction of candidate POIs to the travel endpoints, Represents the travel endpoint, Representative Candidate POIs, is the duration of stay at the trip endpoint, represents the time weight, Indicates The comprehensive proportion of candidate POIs, is the POI area, Represents the distance between the travel endpoint and the candidate POI, is the distance squared coefficient.

[0021] The present invention also provides an individual travel purpose identification system based on mobile phone sensor positioning data, including: an acquisition module: used to obtain mobile phone sensor positioning data and individual travel log data of individual travel; a first calculation module: used to calculate the personal support matrix and the public opinion support matrix according to the mobile phone sensor positioning data of individual travel, and obtain the individual travel endpoint identification result; a second calculation module: used to calculate the individual travel destination POI matching probability according to the individual travel endpoint identification result; an identification module, used to identify the individual travel purpose through a Transformer model.

[0022] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention provides a method for identifying individual travel purposes based on mobile phone sensor positioning data. The mobile phone sensor positioning data of individual travel can be automatically collected by a mobile phone device. Compared with traditional questionnaire surveys, it can be collected continuously and uninterruptedly, thereby accumulating a large amount of data samples.

[0023] The present invention provides an individual travel purpose identification method based on mobile phone sensor positioning data. The mobile phone sensor positioning data is collected in real time and can promptly reflect the user's travel purpose and demand, making the data more timely and can be used for real-time detection and analysis.

[0024] The present invention discloses an individual travel purpose identification method based on mobile phone sensor positioning data. The mobile phone sensor positioning data can save a lot of manpower, material resources and time costs, and there is no need for tedious steps such as printing questionnaires, distributing questionnaires, collecting questionnaires and manually entering data, thereby simplifying the process.

[0025] The present invention provides a method for identifying individual travel purposes based on mobile phone sensor positioning data. The method provides an important basis for the formulation of urban traffic planning schemes and the alleviation of urban traffic congestion, and can optimize traffic network design, monitor traffic conditions in real time, and improve the rationality of traffic resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of a method for identifying individual travel purpose based on mobile phone sensor positioning data of the present invention; Figure 2 This is a structural diagram of a travel purpose identification model based on Transformer in an embodiment of the present invention; Figure 3 is the all-day travel trajectory of an individual in an embodiment of the present invention; Figure 4 is the cumulative probability distribution of the stay time at the travel endpoint in the embodiment of the present invention; Figure 5 It is a cumulative distribution diagram of the distance between the travel endpoints in an embodiment of the present invention; Figure 6 This is a diagram showing an example of a destination type recognition result in an embodiment of the present invention; Figure 7 This is the confusion matrix of the Transformer-based travel purpose recognition model result in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0028] Example 1 The present invention provides a method for identifying individual travel purposes based on mobile phone sensor positioning data, comprising obtaining mobile phone sensor positioning data and individual travel log data of individual travel; obtaining individual travel endpoint recognition results according to the mobile phone sensor positioning data of individual travel; calculating the POI matching probability of the individual travel destination according to the individual travel endpoint recognition result; and identifying the individual travel purpose through a Transformer model according to the POI matching probability of the individual travel destination.

[0029] Obtain the mobile phone sensor positioning data and individual travel log data of individual travel. Specifically, use the "Easy Travel" application to record and transmit the user's mobile phone sensor positioning data in real time. The mobile phone sensor positioning data mainly includes the order number, date, time, longitude, latitude and speed. See Table 1 for some examples of mobile phone sensor data.

[0030] Table 1 Some examples of mobile phone sensor data

[0031] While collecting the positioning data of the mobile phone sensor, the log of the entire travel process of the individual is recorded in detail. These travel records mainly involve the start time, end time and category of the activity. See Table 2 for some examples of individual travel logs. The individual travel log data is mainly used to verify the travel endpoint recognition algorithm, the calibration of the travel purpose recognition algorithm and the accuracy verification of the final recognition result, the production of the Transformer model training set and the evaluation of the model calculation results.

[0032] Table 2 Some examples of individual travel logs

[0033] A user travel attribute dataset is constructed based on the mobile phone sensor positioning data of individual travel, and the similarity of the positioning points is calculated. The personal support matrix and the public opinion support matrix are iteratively calculated based on the similarity, and the individual travel endpoint identification results are obtained based on the mobile phone sensor positioning data of individual travel.

[0034] Specifically, based on the longitude and latitude information in the mobile phone sensor positioning data of individual travel, a user travel attribute dataset is constructed. ,in Represents the user Longitude and latitude travel location feature information .

[0035] The similarity between two-way positioning points is determined based on the negative Euclidean distance between the two-dimensional travel attributes of individuals, where a longitude and latitude data is a point on the map with spatial attributes, and two-way positioning points refer to two longitude and latitude data of the same individual.

[0036] Similarity between two positioning points Determined by negative Euclidean distance,

[0037] in, For individual Longitude information of the travel location, For individual Longitude information of the travel location, For individual The latitude information of the travel location, For individual The latitude information of the travel location.

[0038] The similarity matrix is ​​constructed based on the similarity of the two positioning points. When constructing the similarity matrix, if the value calculated directly using the similarity formula is 0, in order to avoid this situation, you can usually set The minimum value among is used to replace the calculated value of 0.

[0039] In order to make the rows of the similarity matrix comparable, the similarity matrix needs to be normalized to obtain a standardized personal support matrix.

[0040] The normalization process is:

[0041] That is, each element in each row is subtracted from the other elements in the row, where is the similarity of the positioning points, is the similarity matrix Xingfeidi The value of the column, is the public opinion support matrix, a zero matrix with the same structure as the similarity matrix in the initial stage.

[0042] In order to avoid the non-convergence of the results caused by excessive update of the personal support matrix, the damping coefficient is introduced In this embodiment, the value is 0.5 to control the update amplitude of the matrix.

[0043]

[0044] in, is the damping coefficient, It is a zero matrix in the initial stage.

[0045] The diagonal values ​​and off-diagonal values ​​of the public opinion support matrix are calculated using different methods. The diagonal values ​​are calculated as follows:

[0046] The off-diagonal values ​​are calculated as follows:

[0047] Personal support matrix The sum of the positive numbers in each column of the matrix The value of the diagonal; the personal support matrix The positive numbers in each column are accumulated, and the corresponding personal support matrix of the current row The non-positive number obtained by adding the diagonal values ​​in is used as the matrix Off-diagonal values.

[0048] In order to avoid the update amplitude being too large and causing the result to not converge, the damping coefficient is introduced λ In this embodiment, the value is 0.5 to control the update amplitude of the matrix.

[0049] in, is the damping coefficient; is the public opinion support matrix at the current iteration.

[0050] Determine the decision matrix through the personal support matrix and the public opinion support matrix , through the decision matrix To decide whether to end the algorithm. Decision matrix The calculation of is as follows:

[0051] Decision Matrix The column corresponding to the value greater than 0 on the diagonal is the current cluster center, and the cluster center is the individual travel endpoint identification result.

[0052] The decision matrix generated last time As the next similarity matrix, loop to

[0053] Then continue to update the matrix The above iterative process is repeated until the cluster center does not change for multiple consecutive times, which is 15 times in this embodiment, or the maximum number of iterations is reached, which is 200 times in this embodiment. The update is stopped and a series of high-quality user travel endpoints are generated, and the corresponding travel endpoint cluster labels are calibrated for all user data points.

[0054] Search for the index of data points with the same data label, and find all data points related to each cluster center. For each cluster center and all data points with the same label, arrange them in timestamp order, calculate the time interval from the first data point to the last data point, and if the interval is greater than the threshold, which is 10 minutes in this embodiment, judge that the individual is in a stationary state at this time, otherwise it is in a moving state.

[0055] The matching probability of the POI of the individual travel destination is calculated based on the individual travel endpoint identification results. This includes determining the individual's activity area near the travel endpoint based on the individual travel endpoint identification results, and extracting all candidate POIs in the area. The attractiveness of all candidate POIs to each travel endpoint is calculated using the improved gravity model. The matching probability of each candidate POI is calculated by combining the improved gravity model and the posterior probability of the Bayesian criterion, and the POI with the highest matching probability is selected as the user's final destination.

[0056] Specifically, the trip is divided into several segments based on the individual travel endpoint identification results. With the travel endpoint as the center, the user's destination activity area is created with the "space-time neighborhood" as the radius. The parameters of the "space-time neighborhood" are determined by the relevant statistical results of the actual mobile phone sensor positioning data. The POIs contained in the activity area are the destinations that the travelers may go to during this activity. These POIs are called candidate POIs. Their relevant information is collected through the Amap API port, mainly including id, name, location, distance, type and other attributes, where id is the number of all POIs within the radius of the endpoint, name represents the POI name, location represents the center position of the POI, distance is the distance between the POI within the area and the travel endpoint, and type represents the type of these POIs. Table 3 shows an example of a candidate POI list for a travel endpoint.

[0057] Table 3 Candidate POI search results

[0058] The attraction of each candidate POI is calculated based on the improved gravity model. According to the calculation principle of the gravity model, the travel endpoint is regarded as an object. , candidate POIs are considered as objects , the set of candidate POIs is Considering the impact of the built environment of the destination, this model uses the comprehensive proportion of POIs at the travel endpoints and POI area The spatial attractiveness calculation is performed as follows:

[0059] in, Indicates at time Within the scope, The spatial attraction of candidate POIs to the travel endpoints, Represents the travel endpoint, Representative Candidate POIs, The time of the trip endpoint, represents the time weight, Indicates The comprehensive proportion of candidate POIs, The POI area is defined as 1 for the POI represented by a point on the Amap, and the area defined by a surface attribute is calculated based on its actual area. Represents the distance between the travel endpoint and the candidate POI, is the distance squared coefficient.

[0060] The comprehensive proportion of POIs at travel endpoints The calculation method is as follows:

[0061] in, Represents the activity area of ​​the travel endpoint Number of POIs of the type, Representatives in the study area The number of POIs of the type, It is the number of POI types within the travel endpoint range.

[0062] Considering that people's travel time selection is closely related to their travel purpose, this embodiment adjusts and optimizes the time weight according to the dynamic change table of the probability of visiting POIs at different time periods throughout the day, and recommends The values ​​are shown in Table 4.

[0063] Table 4 POI probability dynamic change table

[0064] in addition, is the POI area. According to the definition of POI and POI area on Amap, the POI area represented by a point is defined as 1, and the POI area defined by the surface attribute is calculated according to its true value. At the same time, existing studies have shown that the distance square coefficient When it is set to 1.5, the fit is the highest, so 1.5 is selected as the distance power coefficient.

[0065] Calculate the matching probability of individual travel destination POI. Specifically, based on the spatial attractiveness calculated by the gravity model, the matching probability of each candidate POI is calculated using the posterior probability. When calculating the spatial attractiveness of POI, an improved gravity model is introduced to consider distance, time, comprehensive proportion, and area, and finally obtain The matching probability of the POI with the greatest spatial attractiveness at the travel endpoint is also the greatest. , candidate POI set , travelers visit destinations The probability of is calculated as follows:

[0066] in, For in time Within the scope, Candidate POI pairs for travel endpoints The spatial attraction For travelers in time ,Place , POI is The matching probability.

[0067] The POI with the highest matching probability in the activity area is selected as the real destination at the travel endpoint, and the algorithm is applied to all travel segments of the traveler, giving the endpoint of each travel segment a POI type label. This label can be used as a label for the subsequent Transformer model training data set, and the trained model can be used to identify individual travel purposes.

[0068] According to the POI matching probability of individual travel destinations, the individual travel purpose is identified through the Transformer model. Specifically, the travel purpose identification model structure of the Transformer model is first determined, such as Figure 2 As shown. Then, the travel activity information data and travel destination labels are input into the embedding layer as four-dimensional and one-dimensional vectors to obtain embedded data and class labels for classification. User travel activity information data includes travel time, stay duration, visit frequency, and destination type. User travel activity information data is Gaussian embedded as shown below:

[0069] in, Input data for the model after adding timing information; is the standard deviation. Gaussian embedding has a higher stable rank upper bound, which can be obtained by control.

[0070] The user trajectory data is sorted according to the id number in the order of travel time to help the model learn the location information. The location information is the time series vector. The trajectory data and the time series vector are vector-added to obtain the final embedding result.

[0071] There are N encoders with the same structure in the Transformer-based travel purpose recognition model. The multi-head attention layer is the core of the encoder, and each "head" uses an independent linear projection matrix to transform the input embedding data and generate query vectors respectively. , key vector Sum value vector .

[0072]

[0073]

[0074]

[0075] The final embedding result , is the weight vector, which is randomly generated in the initial stage. , , , is the dimension of the attention layer.

[0076] Each set of Q, K, and V vectors is used to implement the self-attention mechanism of the Transformer model, helping the model capture the dependencies between different positions in the sequence data. The similarity between the query vector Q and all key vectors K is calculated to generate an attention score, which is used to weight the value vector V. The attention score calculation formula is as follows:

[0077] Among them, softmax transforms each row of the obtained matrix so that the sum of the elements in each row is 1. For the matrix Dimension.

[0078] The attention score calculation process occurs in parallel, and each "head" is completed independently to capture different characteristics of the data; the outputs of all "heads" are combined through splicing, and transformed through another linear projection layer to produce the final output, which is a comprehensive score based on the data's own label. The mathematical expression of this layer is as follows:

[0079]

[0080] in, is the weight matrix.

[0081] In order to ensure the stability of data distribution characteristics and prevent overfitting, it is necessary to perform residual connection between the output of the multi-head attention layer and the original input data, and perform layer normalization, as shown below:

[0082]

[0083] in, is the attention output obtained by residual connection, is the scaling factor, the initial value is 1, is the offset factor, the initial value is 0, is a constant term to avoid the denominator being 0.

[0084] The obtained attention output is processed by the loss function, and the loss function is calculated using the cross entropy loss function:

[0085] in, is the true distribution, is the predicted distribution.

[0086] The Adam optimizer is used to optimize the weight matrix by combining the advantages of the momentum method. By estimating the first-order moment (mean) and second-order moment (uncentered variance) of the gradient, the historical gradient information is comprehensively utilized and the learning rate of each parameter is adaptively adjusted. The weight matrix is ​​optimized using the Adam optimizer as follows:

[0087] in, for The weight of the moment, for The weight of the moment, is the learning rate, which is 0.001 in this embodiment, is the learning rate, which is 10 in this embodiment -8 . is the number of iterations, for The first moment estimate of the gradient at that moment, yes The first moment estimate of the gradient at that moment, is the learning rate of the first-order moment estimate, which is 0.9 in this embodiment. For partial guidance, for The second moment estimate of the gradient at that moment, for The second moment estimate of the gradient at that moment, is the learning rate of the second-order moment estimation, which is 0.999 in this embodiment. is the bias-corrected first-order moment estimate, is the bias-corrected second moment estimate.

[0088] Update the weight matrix for iterative training. When the loss function value tends to be stable, or the number of iterations reaches the preset value (200 times in this embodiment), stop model training, input the test data set for testing to obtain the classification accuracy, and then identify the individual travel purpose. Since the individual travel activity trajectory data set is small, there are 8 types of travel purpose classification, the number of attention layers is 8, and the number of encoders is 1.

[0089] Example 2 A method for identifying individual travel purposes based on mobile phone sensor positioning data. First, the mobile phone sensor positioning data and individual travel log data of individual travel are obtained. 60 volunteers are selected to collect 150 hours of mobile phone sensor data and travel activity log data, totaling more than 270,000 valid data records. The scope covers eight common daily travel activities of residents, including going to work, going to school, going home, seeking medical treatment, dining, shopping, entertainment, and picking up students. The target groups for this data collection include six major travel groups, including staff, teachers, students, medical personnel, self-employed individuals, and civil servants, covering the daily travel patterns of urban residents.

[0090] Then, the individual travel endpoint identification results are obtained based on the mobile phone sensor positioning data of the individual travel. Taking the travel data of a user as an example, the travel endpoint identification results are shown, such as Figure 3 As shown in the figure, the collected mobile phone sensor positioning data shows the actual travel path of the individual. In order to accurately represent the actual travel endpoints and movement directions of the experimental route, the numbers ①, ② and straight arrows are used for annotation, where the actual travel endpoints and activity information are determined based on the individual travel log.

[0091] The affinity propagation clustering algorithm was applied to the measured data of 60 volunteers. The statistical results of the travel endpoint recognition errors of all measured data are shown in Table 5. There are 1112 real travel endpoints in total, and 1043 were accurately recognized, reaching an overall recognition accuracy of 93.8%. In particular, in regular travel activities such as going to work, going to school, going home, dining, shopping, and seeking medical treatment, the recognition accuracy of travel endpoints generally exceeds 90%.

[0092] Table 5 Trip endpoint recognition results

[0093] The POI matching probability of individual travel destinations is calculated based on the individual travel endpoint identification results. The relevant statistics of the mobile phone sensor data of 60 volunteers are obtained. The data show that for endpoints of various travel purposes, when the stay time is 600 seconds, the cumulative probability distribution reaches 95%. The statistical results are as follows: Figure 4 As shown, Figure 5 As shown in the figure, by analyzing the proximity distance between each positioning point and the 600th second point, it is found that there is an obvious inflection point in the curve at 62 meters. The curve is stable before this point and rises sharply after this point. Based on this, 62 meters is selected as the expansion radius of the travel endpoint.

[0094] Take a user's travel activity as an example. According to the user's travel log for the day, the day's activities are arranged as sending children to school - going to work - getting off work - picking up children from school - going home. By obtaining the individual travel endpoint recognition results based on the mobile phone sensor positioning data of the individual travel, 5 travel segments are obtained, such as Figure 6 As shown in the figure, it is a travel segment of activity 1, and the travel time of this activity is 7:36am-7:47am. In the figure, point A is the endpoint of this activity. It can be seen from the figure that there are several POIs within the clustering threshold radius, including Jianfeng Primary School, Runhe Department Store, Guo Jianzhang Roujiamo Shop, Printing Photography, Typing Color Printing, etc. The individual travel endpoint identification results are obtained based on the mobile phone sensor positioning data of the individual travel, and the matching probability of all POIs in the activity range is calculated. The calculation results show that Jianfeng Primary School is the most attractive POI, and it is used as the destination of this activity. By querying the travel log, this destination identification result is correct.

[0095] The Transformer model is used to identify individual travel purposes, using the Adam optimizer and cross entropy loss function, and the sample data is used for model training and verification in a ratio of 7:3. The statistics of the travel purpose recognition errors of all measured data are shown in Table 6. It can be seen that the recognition accuracy of different travel purposes is above 85%, especially for medical treatment, entertainment activities, and catering, the recognition accuracy is 100%. The main reason for the accurate recognition of medical treatment is that the labels corresponding to its destination have a large degree of distinction, and the length of stay is much different from the doctor's working hours; the accurate recognition of entertainment activities is mainly due to its unique destination label. The algorithm has a good distinction between activities with large differences. In addition, from the "Recognition Error" row data in Table 6, it can be found that the number of recognition errors for work activities is the largest, resulting in the lowest recognition. The reason is that there are differences in identity activities and generalization of activity scope in work activities. For example, students and teachers have the same work and rest schedule, and there will be recognition errors for going to work and going to school; travelers with other identities are within the scope of the workplace, and the travel endpoint / destination type cannot be accurately identified, resulting in travel purpose recognition errors. Figure 7The confusion matrix diagram of the recognition results of all measured data trip purposes. It can be seen from the figure that based on the sample adoption strategy of stratified sampling, going home is the most frequent travel activity, and the frequency of medical treatment and entertainment activities is relatively low. The activity of going to work has the largest number of errors among all travel activities, and its error results tend to be inclined to going to school. The recognition errors of going home activities tend to be inclined to entertainment and shopping. The similarity of travel time and the similarity of the length of stay between entertainment and shopping activities and short-term going home activities are the main reasons for the errors.

[0096] Table 6 Statistical results of the Transformer-based travel purpose recognition model

[0097] Example 3 The present invention provides an individual travel purpose identification system based on mobile phone sensor positioning data. The individual travel purpose identification system based on mobile phone sensor positioning data can be used to implement the above-mentioned individual travel purpose identification method based on mobile phone sensor positioning data. Specifically, the individual travel purpose identification system based on mobile phone sensor positioning data includes an acquisition module, a first calculation module, a second calculation module, and an identification module.

[0098] Among them, the acquisition module is used to obtain the mobile phone sensor positioning data and individual travel log data of individual travel; the first calculation module is used to calculate the personal support matrix and the public opinion support matrix based on the mobile phone sensor positioning data of individual travel, and obtain the individual travel endpoint identification result; the second calculation module is used to calculate the POI matching probability of the individual travel destination based on the individual travel endpoint identification result; the identification module is used to identify the individual travel purpose through the Transformer model.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying individual travel purpose based on mobile phone sensor positioning data, characterized in that: The following steps are involved: Obtaining mobile phone sensor positioning data and individual travel log data of individual travel; Obtain individual travel endpoint identification results based on the mobile phone sensor positioning data of individual travel; Calculate the POI matching probability of individual travel destination based on the individual travel endpoint identification results; According to the POI matching probability of individual travel destination, the individual travel purpose is identified through the Transformer model.

2. The method for identifying individual travel purpose based on mobile phone sensor positioning data according to claim 1, characterized in that: The step of obtaining the individual travel endpoint identification result also includes: A user travel attribute dataset is constructed based on the mobile phone sensor positioning data of individual travel, and the similarity of the positioning points is calculated. The personal support matrix and the public opinion support matrix are iteratively calculated based on the similarity.

3. The method for identifying individual travel purpose based on mobile phone sensor positioning data according to claim 2, characterized in that: The similarity of the positioning points is: in, For individual Longitude information of the travel location, For individual Longitude information of the travel location, For individual The latitude information of the travel location, For individual The latitude information of the travel location.

4. The method for identifying individual travel purpose based on mobile phone sensor positioning data according to claim 2, characterized in that: The personal support matrix is: in, is the damping coefficient, is the personal support matrix (initialized as 0 matrix), is the personal support matrix after the iteration update, For positioning point and anchor points The similarity of is the similarity matrix Xingfeidi The value of the column, The first Xingfeidi The value of the column.

5. The method for identifying individual travel purpose based on mobile phone sensor positioning data according to claim 2, characterized in that: The public opinion support matrix is: in, is the damping coefficient; is the public opinion support matrix at the current iteration; Include diagonal values and off-diagonal values The diagonal value for: in, The personal support matrix Column off-diagonal values; The off-diagonal values for: in, is the diagonal value in the personal support matrix.

6. The method for identifying individual travel purpose based on mobile phone sensor positioning data according to claim 1, characterized in that: The mobile phone sensor positioning data of the individual travel includes the order number, date, time, longitude, latitude and speed, and the individual travel log data includes the activity start time, activity end time and activity category.

7. The method for identifying individual travel purpose based on mobile phone sensor positioning data according to claim 1, characterized in that: The calculating the individual travel destination POI matching probability according to the individual travel endpoint identification result includes: The activity area of ​​the travel endpoint is determined according to the individual travel endpoint recognition results, and the candidate POIs in the activity area are extracted. The attractiveness of the candidate POIs to each travel endpoint is calculated based on the improved gravity model. The matching probability of each candidate POI is calculated by combining the improved gravity model and the posterior probability of the Bayesian criterion, and the POI with the highest probability is selected as the user's final destination.

8. The method for identifying individual travel purpose based on mobile phone sensor positioning data according to claim 1, characterized in that: The method of identifying the individual travel purpose through the Transformer model according to the POI matching probability of the individual travel destination includes: The individual travel endpoint recognition results and the individual travel destination POI matching probability are input into the Transformer model, and the travel activity information is put into the encoder for encoding. The Adam optimizer is used to update the optimization parameters and iterative training is performed. After the number of iterations reaches the preset value, the test data set is tested to obtain the classification accuracy that meets the standard and then the individual travel purpose is identified. The travel activity information includes travel time, stay duration, visit frequency, and destination type.

9. The method for identifying individual travel purpose based on mobile phone sensor positioning data according to claim 7, characterized in that: The attractiveness of the candidate POI to each travel endpoint is calculated based on the improved gravity model as follows: in, Indicates at time Within the scope, The spatial attraction of candidate POIs to the travel endpoints, Represents the travel endpoint, Representative Candidate POIs, is the duration of stay at the trip endpoint, represents the time weight, Indicates The comprehensive proportion of candidate POIs, is the POI area, Represents the distance between the travel endpoint and the candidate POI, is the distance squared coefficient.

10. A system for identifying individual travel purpose based on mobile phone sensor positioning data, comprising: Acquisition module: used to obtain mobile phone sensor positioning data and individual travel log data of individual travel; The first calculation module is used to calculate the personal support matrix and the public opinion support matrix based on the mobile phone sensor positioning data of the individual travel, and obtain the individual travel endpoint identification result; The second calculation module is used to calculate the POI matching probability of the individual travel destination according to the individual travel endpoint identification result; The recognition module is used to identify individual travel purposes through the Transformer model.

Images