CN107529135A - User Activity type identification method based on smart machine data - Google Patents

Abstract

The invention relates to a method for discriminating user activity types based on smart device data, comprising the following steps: acquiring user smart device data; performing data cleaning to obtain the user's movement track, and obtaining the user's travel chain through itinerary identification; extracting the activity start time and activity Duration, and obtain the land use properties corresponding to the stay section according to the points of interest in the stay section; analyze the smart device data of the user's multi-day travel, and judge the location of the user's home and/or work place, so as to obtain the information of whether the user is at home or at work. Activity type: For other types of activities except home and work, the corresponding activity types are obtained by inputting the activity characteristics into the activity classifier. The invention can not only analyze the user's activity type according to the user's smart device data, but also the time threshold or space threshold involved in the analysis process is calibrated by the traditional user travel survey data, which is more usable, accurate and real-time .

Description

Discrimination method of user activity type based on smart device data

technical field

The present invention relates to user smart device data analysis, in particular to a method for discriminating user activity types based on smart device data.

Background technique

Urban residents' travel activity information is an important basis for urban planning, traffic management, and residents' activities research, and is generally obtained through residents' travel surveys. For the acquisition of information on residents' travel activities, traditional survey methods generally include home visits, telephone inquiries, and postcards. At present, the survey methods based on household interviews and paper questionnaires are mainly used in China. The disadvantages of this method are obvious, namely heavy burden of respondents, low survey accuracy, high survey cost, low sampling rate, etc., which cannot meet the needs of rapid urban development.

Smart device data has become an indispensable data source in current traffic big data research due to its wide coverage of users, no need for special data acquisition equipment, low data acquisition cost, and large data volume. Extracting user activity information based on smart device data can greatly increase the sample size of the survey, shorten the survey cycle, and reduce the cost of the survey. At the same time, the use of smart device signaling data for information acquisition is a passive information acquisition method that does not require residents to conduct questionnaires. Filling in will not increase any burden on residents, so as to provide richer basic information for the transportation industry.

At present, although time-stamped location information can be obtained from smart device data to obtain the user's daily movement trajectory, it cannot obtain the user's socioeconomic attribute information, nor can it directly obtain travel activity attribute information such as travel mode and activity type.

At present, the identification methods of activity types are mostly focused on the stages of stop point identification and job residence identification, and can only identify the two types of activities in "home" and "work". Specifically, according to the longitude and latitude of the smart device positioning data, the smart device positioning data is matched to the corresponding traffic area. Based on the smart device positioning data table after matching, the data of a user for a continuous week is extracted, and the statistics are respectively calculated between the period of residence and work The number of occurrences in each traffic zone within a time period is accurately judged, and the traffic zone with the largest number of occurrences is used as the user's residence and work place. This method of only counting the number of stays does not take into account the length of stay, and it is easy to misjudge the location of multiple short-term stays as home or work. For the judgment methods of other activity types except "home" and "work", there are few researches and the real-time performance is poor.

Contents of the invention

Based on this, it is necessary to provide a more real-time based method for judging the types of activities other than "home" and "work" in the identification of user travel activity types. Currently, there are few studies and poor real-time performance. A method for discriminating user activity types from smart device data.

A method for discriminating user activity types based on smart device data, comprising the following steps:

Obtain user smart device data;

Perform data cleaning on the smart device data to obtain the user's movement trajectory, and divide the user's stay section and travel section through itinerary identification to obtain the user's travel chain;

Extracting the activity start time and activity duration of each of the stay sections in the travel chain, and obtaining the land use properties corresponding to the stay sections according to the points of interest of the stay sections;

Analyze the smart device data of the user's travel at the preset time, and obtain the corresponding stay period, length of stay, and number of stays of each user, so as to determine the location of the user's home and/or work place, and obtain two types of activities at home or at work;

By inputting the activity start time, activity duration, and land use properties corresponding to the stay segments other than the two types of activities at home or at work into the activity classifier, the two types of activities of each user except at home or at work can be obtained outside the predefined activity types.

In one of the embodiments, the method also includes: constructing the activity classifier, including:

Based on the traditional user travel survey data, the activity start time, activity duration, and land use properties corresponding to the stay section of each user in the survey data are extracted to construct an activity classifier based on a decision tree.

In one embodiment, the data cleaning includes field missing processing, deleting abnormal IMSI number records, deleting records that cannot match the base station positioning data, deleting duplicate data, ping-pong effect processing, and signal drift processing .

In one of the embodiments, the ping-pong effect processing includes the following steps: merging the smart device data of each user into areas according to space and time, if the user signal fluctuates within a range less than the spatial threshold L1 and exceeds the time limit Threshold T1, it is considered that the user is at the same location during this period.

In one of the embodiments, the signal drift processing includes the following steps: merging the smart device data of each user into areas according to space and time, if the user leaves the space threshold L2 within the time threshold T2, and then returns to Within the space threshold L2, the users are considered to be in the same location.

In one embodiment, the itinerary identification includes the following steps: if the trajectory points of the user within the time threshold T _stay are clustered within the radius of L _stay , or the moving speed within the time threshold T _stay is lower than If the speed threshold V _stay , the corresponding segment is a stay segment, otherwise it is a travel segment.

In one of the embodiments, the following steps are also included:

Convert the travel data in the user travel survey data into activity chain data, and filter the data of users who leave home in the morning and go home in the evening;

Extract the activity duration of all types of activities, establish the activity duration distribution, and use the p% quantile of the activity duration distribution as the time threshold T _stay for trip identification, where p is any natural number from 5 to 10.

In one of the embodiments, the acquisition of the land use properties corresponding to the stay section according to the points of interest in the stay section includes the following steps:

Calculate the center coordinates of the stop position by time weighting;

According to the position and quantity of the point of interest corresponding to the center coordinates, a kernel density estimation model is established, and its formula is as follows:

K(.) represents the kernel function;

r represents the window width;

n represents the total number of points of interest;

d _{i, s} represents the distance from the center coordinates to each point of interest s;

Choose a Gaussian function as the kernel function:

Calculate the kernel density values of different interest point types at the stay position, and take the land use property corresponding to the interest point with the highest kernel density value as the land use property of the stay position.

In one of the embodiments, the analysis of the smart device data of the user's trip at the preset time obtains the corresponding stay period, length of stay, and number of stays, so as to determine the location of the user's home and/or work place, and obtain the home or Working on two activity types consists of the following steps:

extracting all of said activity start times, said activity durations, and land use properties for each user;

Filter out the data of each user's working day, and count the total number of days as N

For each type of stay position, count the total number of days N _{home where the night stay time is greater than T home ;}

If N _home is greater than the first judgment time threshold, the location is the home location. Otherwise, count the total number of days N _{work in which the stay time in the working period is greater than T work ;}

If N _work is greater than the second judging time threshold, the location is a working location.

In one of the embodiments, further comprising the following steps: further comprising the following steps:

Convert the travel data in the user travel survey data into activity chain data, and filter the smart device data of users who leave home in the morning and return home in the evening;

Extract the activities where the place of stay at night is home, establish the distribution of activity duration, and identify the time threshold T _home of the a% quantile family location of the distribution of activity duration, where a is any natural number from 5 to 10;

Extract the activities where the place of stay during the working period is the work place, establish the activity duration distribution, and use the b% quantile of the activity duration distribution as the time threshold T _work for work place identification, where b is any natural number from 5 to 10.

The method for discriminating user activity attributes based on smart device data provided by the present invention can not only analyze the user's smart device data to obtain the user's job and residence, but also make more accurate judgments and statistics on the types of activities other than at home and at work. And the judgment of other types of activities is based on the traditional user travel survey data to extract relevant activity characteristics, build an activity classifier based on a decision tree, and judge the types of activities other than home and work, and the time threshold for travel recognition T _stay , the time threshold T _home for home location identification, and the time threshold T _work for work location identification are all extracted based on the analysis of traditional user travel survey data, which play a role in calibration and are more usable, accurate and real-time sex.

Description of drawings

Fig. 1 is a flow chart of a user activity type discrimination method based on smart device data provided by an embodiment of the present invention;

FIG. 2 is another flow chart of a method for discriminating user activity types based on smart device data according to an embodiment of the present invention;

Fig. 3 is a flow chart of judging the user's occupation and residence in an embodiment of the present invention;

Fig. 4 is the accumulative frequency distribution diagram of all types of activity durations of traditional user travel survey data in Shanghai in 2009 of the present invention;

Fig. 5 is the accumulative frequency distribution figure of all the activity durations at home at night in Shanghai's traditional user travel survey data in 2009 of the present invention;

Fig. 6 is the accumulative frequency distribution diagram of all work activity durations of the traditional user travel survey data in Shanghai in 2009 of the present invention;

FIG. 7 is a schematic diagram of the rules for identifying other activity types of traditional user travel survey data in Shanghai in 2009 according to the present invention.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

Fig. 1 is a flow chart of the method for discriminating user activity types based on smart device data provided by the present invention. Firstly, the operation flow of the method for discriminating user activity attributes based on smart device data provided by the present invention is generally explained. First, obtain the smart device data of all users in a city or a region from smart device network operators, smart device APPs, and other public resources. Then clean the smart device data, organize and divide the data, and then extract the activity features to determine whether the user's activity type is at home or work. If it is determined that the user's activity type is neither at home nor at work, then input the activity features into the activity classification to determine other specific types of activity. Data cleaning is carried out on smart device data. The travel/stay division time thresholds used in the data sorting and division process, and the home activity duration thresholds and workplace activity duration thresholds used in the on-the-job and residence judgment process are all determined by Shanghai’s traditional 2009 threshold. The user travel survey data analysis is obtained; in addition, the activity classifier is also constructed based on the traditional user travel survey data in Shanghai in 2009, which will be introduced in detail below. Specific smart devices can be mobile phones, Ipads, wearable devices, etc.

Referring to Fig. 2 at the same time, in one embodiment, the method for discriminating user activity attributes based on smart device data provided by the present invention includes the following steps:

S202. Obtain user smart device data.

The smart device data in the present invention can be roughly divided into two types according to the data sources. One type is smart device bill data and smart device signaling data obtained from smart device network operators (China Mobile, China Unicom, China Telecom). The main contents of the data include: anonymized user identification code, cell number of the base station, event type, and time when the event occurred. The difference between smart device bill data and smart device signaling data is mainly reflected in the event type. Smart device bill data mainly includes calling, In addition to single-data events, it also includes power-on, power-off, cell handover, location update, etc. Bill data is a subset of signaling data, and signaling data is data with a higher user location sampling rate. The longitude and latitude information of the base station can map the user's movement in the mobile network to the actual geographic system. The other type is user movement trajectory data obtained from smart device APP applications with trajectory recording functions (such as GeoLife, Bikely, SportsDo). The main content of the data includes sampling time, latitude and longitude and other information. Taking Shenzhen smart device bill data as an example, the data fields usually include user identification code, cell identification, sector identification, access time, etc., as shown in Table 1.

Table 1

S204. Perform data cleaning on the smart device data to obtain the user's movement trajectory, and divide the user's stay section and travel section through travel identification to obtain the user's travel chain.

Specifically, in one embodiment, the data cleaning of smart device data includes field missing processing, deleting abnormal IMSI (International Mobile Subscriber Identity) number records, deleting records that cannot match the base station positioning data, and deleting Duplicate data, ping-pong effect handling, and signal drift handling.

Missing field processing refers to deleting records with missing key field information in smart device data, such as the number of some base stations being 0, missing time items, etc.;

Delete the record of the abnormal IMSI number. Due to some abnormalities in the stored procedure, some IMSI numbers may be generated.

Records that could not be matched to base station data were deleted. The research of this embodiment is within the scope of Shanghai. Due to signal problems, some records may be located in the base stations of neighboring provinces. If there are base station data of neighboring provinces, the relevant data will be deleted.

Remove duplicate data. In addition to some real duplicate data in the implementation, there may also be duplication of records due to accuracy issues (such as time items, some records that are not originally at the same time after being accurate to seconds will also get the same result accurate to seconds).

Specifically, the ping-pong effect processing in one embodiment includes the following steps: merge the smart device data of each user into areas according to space and time, if the user signal fluctuates within a range smaller than the spatial threshold L1 and exceeds the time threshold T1, Then the user is considered to be in the same location during this time. More specifically, the space threshold L1 is a diameter range of 400-500 meters, and the time threshold T1 is 25-30 minutes.

Specifically, the signal drift processing in one embodiment includes the following steps: merge the smart device data of each user into areas according to space and time, if the user leaves the space threshold L2 within the time threshold T2, and then returns to the space Within the threshold L2, it is considered that the users are in the same location. If the data record of the user's smart device leaves the above-mentioned small space range L2 for a short period of time, and then returns quickly, the user is also considered to be in the same location. More specifically, it means that the position switching speed of the signal leaving the above-mentioned area and returning to the above-mentioned area is greater than 100km/h (upper limit of urban expressway design speed) and the time of leaving the above-mentioned area does not exceed T _clean . More specifically, the spatial threshold L2 is a diameter range of 400-500 meters, and the time threshold T2 is 25-30 minutes.

Specifically, in one embodiment, the identification of the user's movement trajectory includes the following steps: If the user's trajectory points within the time threshold T _stay are clustered within the radius of L _stay , or the user's movement within the time threshold T _stay If the speed is lower than the speed threshold V _stay , the corresponding section is a stay section, otherwise it is a travel section. Specifically, referring to FIG. 4 , in one embodiment, the traditional user travel survey data in Shanghai in 2009 is converted into activity chain data, and the data of users who leave home in the morning and return home in the evening are screened. Then extract the activity duration of all types of activities, establish the activity duration distribution, and use the p% quantile of the activity duration distribution as the time threshold T _stay for trip identification, where p is any natural number from 5 to 10. In this embodiment, the p% quantile of the activity duration distribution refers to sorting all activity times of smart device users in ascending order based on the traditional user travel survey data in Shanghai in 2009, and there are a total of n , n*p%=m, then the activity duration corresponding to the mth row in the sorting is the value of the p% quantile. If the calculated m is not an integer, for example, 12.3, then take the 12th in the sorting The mean of the time values corresponding to the 13th is taken as the value of the p% quantile. In this implementation, the 5th percentile of the activity duration distribution is taken, and 25 minutes is used as the time threshold T _stay for travel identification. In other embodiments, T _stay may be any time value corresponding to the 6th, 7th, 8th, 9th, or 10th percentile of the activity duration distribution, depending on actual application conditions. More specifically, the speed threshold V _stay is 1m/s, the L _stay is 200-500 meters, and the time threshold T _stay is 5-25 minutes. The specific numerical points of L _stay and T _stay need to be combined with the entire activity chain and specific actual conditions consider.

S206. Extract the activity start time and activity duration of each stay section in the travel chain, and obtain the land use properties corresponding to the stay section according to the points of interest in the stay section. In a specific embodiment, obtaining the land use property corresponding to the stay section includes the following steps:

Calculate the center coordinates of the stop position by time weighting. Specifically include the following steps:

First, calculate the average coordinates of two consecutive stationary points p _i and p _i+1 in the candidate stay position of the same smart device user in turn

p _i x—longitude coordinates of static point p _i ;

p _i y - the latitude coordinates of the stationary point p _i ;

Then, the ratio of the time interval Δt _{(i, i+1)} between the two static points to the duration s.Δt of the entire candidate stay position is used as the weight of the average coordinate:

Δt _{(i, i+1)} ——the time interval between two stationary points p _i , p _i+1 ;

s.Δt——the duration of the entire candidate stay position;

Finally, the center coordinates (s.x, s.y) of the candidate stop positions are calculated by weighted addition:

(s.x, s.y) is the center coordinate of the candidate stop position s.

According to the position and quantity of the interest points corresponding to the center coordinates, the kernel density estimation model is established, and the formula is as follows:

K(.) represents the kernel function;

r represents the window width;

n represents the total number of points of interest;

d _{i, s} represents the distance from the center coordinates to each point of interest s;

Choose a Gaussian function as the kernel function:

Calculate the kernel density values of different types of interest points at the stay position, and take the land use property corresponding to the interest point with the highest kernel density value as the land use property of the stay position.

S208, analyze the smart device data of the user's travel at the preset time, obtain the corresponding stay period, stay duration, and number of stays of each user, so as to judge the location of the user's home and/or work place, and obtain two activities at home or at work Types of. Specifically, the preset time in this implementation is multiple days, and in other embodiments, the preset time can also be a period of time, half a month, a month, a few months or a year, or several years, depending on the actual situation. Certainly.

Referring to Figure 3, in one embodiment, the specific steps are as follows:

Extract all the activity start time, activity duration, and land use properties of each user; filter out the data of each user's working day, and count the total number of days as N;

For each type of stay position, count the total number of days N _{home where the night stay time is greater than T home ;}

If N _home is greater than the first judgment time threshold, the location is the home location. Otherwise, count the total number of days N _{work in which the stay time in the working period is greater than T work ;}

If N _work is greater than the first judging time threshold, the location is a work place; otherwise, the location is a destination for other activities. Specifically, nighttime refers to 20:00-7:00 the next day; working hours refers to 9:00-17:00. In this implementation, both the first judgment time threshold and the second time threshold are 60% of the total number of days N, and other percentages are determined according to actual conditions in other embodiments.

More specifically, the time threshold T _homme for family location identification in this embodiment is based on the traditional user travel survey data in Shanghai in 2009 as a sample. The smart device data of users who leave and go home at night; extract the activities that stay at home from 20:00 at night to 7:00 the next day, establish the activity duration distribution, and identify the a% quantile family location of the activity duration distribution The time threshold T _home , a is any natural number from 5 to 10. In this embodiment, the a% quantile of the activity duration distribution refers to sorting all the activity times of smart device users in ascending order based on the traditional user travel survey data in Shanghai in 2009, and there are a total of n , n*a%=m, then the activity duration corresponding to the mth row in the sorting is the value of the a% quantile. If the calculated m is not an integer, for example, 12.3, take the 12th in the sorting The average of the time values corresponding to the 13th is taken as the value of the a% quantile. Specifically, referring to FIG. 5 , the 5th percentile of the activity duration distribution is taken as 540 minutes, which is used as the time threshold T _home for family location identification, that is, the time spent at home is 9 hours. In other embodiments, T _home may be any time value corresponding to the 6th, 7th, 8th, 9th, or 10th percentile of the activity duration distribution, depending on actual application conditions.

The time threshold T _work for workplace identification takes the traditional user travel survey data in Shanghai in 2009 as a sample, converts the travel data in the user travel survey data into activity chain data, and screens the smart device data of users who leave home in the morning and return home in the evening ;Extract the activities that stay at the work place during the working period 9:00-17:00, establish the activity duration distribution, and use the b% quantile of the activity duration distribution as the time threshold T _work for the work place identification, b is 5 to Any one of 10 natural numbers. In this embodiment, the b% quantile of the activity duration distribution refers to sorting all activity times of smart device users in ascending order based on the traditional user travel survey data in Shanghai in 2009, and there are a total of n , n*b%=m, then the activity duration corresponding to the mth row in the sorting is the value of the b% quantile. If the calculated m is not an integer, such as 12.3, take the 12th in the sorting The average of the time values corresponding to the 13th is taken as the value of the b% quantile. Specifically, in this embodiment, referring to FIG. 6, taking the traditional user travel survey data in Shanghai in 2009 as a sample, the 5th percentile of the activity duration distribution is 165 minutes, that is, the time spent in a certain place is More than 2 hours, close to 3 hours. Generally, the normal working time is 7-8 hours, and 165 minutes is taken as the time threshold because some jobs may not be in one place for a long time, such as company managers, or teachers, etc., they work in one place The time is 2-3 hours. At the same time, in addition to the reference time threshold of 165 minutes, we will also consider the number of stays and whether the location of the stay is consistent to rule out occasional shopping or eating out. In other embodiments, T _work may be any time value corresponding to the 6th, 7th, 8th, 9th, or 10th percentile of the activity duration distribution, depending on actual application conditions.

In addition, it is understandable that the activity characteristics of the activity type of going to school are similar to that of work, so the judgment of activities such as going to school has been incorporated into the judgment of work and residence.

S210, by inputting the activity start time, activity duration, and land use properties corresponding to the stay segments other than the two types of activities at home or at work into the activity classifier, to obtain the two types of activities for each user except at home or at work. Activity types other than predefined ones. Predefined activity types refer to activity types defined in advance according to statistical requirements. For example, the predefined activity types may include going to work, going to school, shopping, culture and entertainment, business, picking up people, going home, etc., but are not limited thereto.

The steps to build an activity classifier are as follows: Based on the traditional user travel survey data in Shanghai in 2009, extract the activity start time, activity duration, and land use properties corresponding to the stay sections of the above survey data to construct an activity based on a decision tree Classifier. Specifically, the nature of land use first corresponds to the category of stay points summarized from the traditional user travel survey data in Shanghai in 2009, and then combined with the activity start time and activity duration input into the activity classifier, other specific activity types can be obtained. The judgment rules of the activity classifier after training are shown in Figure 7, which will be introduced in detail below. User travel surveys are held regularly in various cities, and the collected information is relatively complete, covering the input characteristic data required for model calibration in the present invention. Although the sampling rate is not high (generally 1% of the total urban population), the number of samples can fully meet the needs of model training and calibration. In the embodiment of the present invention, an activity classifier model is constructed by using the survey data of user trips in Shanghai in 2009. In other embodiments, user travel survey data of other cities or years may also be used to construct an activity classifier model.

Specifically, the corresponding relationship between land use properties and stay points is shown in Table 2:

Table 2

Through the statistics of Shanghai traditional user travel survey data in 2009, the analysis shows that the types of activities include: ①going to work, ②going to school, ③shopping, ④cultural entertainment, ⑤business, ⑥pick-up, ⑦going home, and ⑧other life.

Now in conjunction with Figure 7, refer to Table 2 and the above eight types of activity types to elaborate on the judgment rules of the activity classifier. Specifically, GIS2 in Fig. 7 is the type of stay point corresponding to the nature of the land use, dur is the duration of the activity, and startTime is the start time of the activity. First, input the stay point category GIS2, if GIS2=3, the stay point category is shopping mall, store, then the activity type is ③ shopping, and the judgment is completed. If GIS2 is not equal to 3, continue to judge whether it is another type of stay, such as Figure 7, continue to judge, if GIS2=8, the type of stay point is entertainment place, tourist place, exhibition hall, sports venue, then the activity type is ④ culture entertainment. If GIS2 is not equal to 8, then continue to judge other categories, if GIS2=7 or 12, further judge whether the activity duration dur is less than 38 minutes, if then further judge whether the start time startTime of this activity is less than 432 minutes (by zero, that is, at night 12:00 is the starting point), if the activity type is ④ culture and entertainment, if not, further judge whether the startTime is less than 1040 minutes, if the activity type is ⑥ pick-up, if not, the activity type is ⑧ other life. If the start time startTime is not less than 432 minutes, then the stay point category GIS2 is an external transportation hub such as a railway station, a long-distance bus station, a wharf, an airport, etc. At the same time, it is further judged whether the start time of the activity is less than 488 minutes, and if so, the activity type is ⑥ pick up and drop off people, otherwise ⑧ other life.

If GIS2 is not 7 or 12, then further judge that GIS2 is 2 or 5, if the duration is greater than or equal to 7 minutes, then further judge whether the start time startTime is greater than or equal to 598 minutes, if not, then the activity type is ⑥ pick-up person; If yes, further judge whether the stay point category GIS2 of the activity is 2; if yes, the activity type is ⑤ business; if not, it is ⑧ other life. If the start time startTime is not greater than or equal to 598 minutes, then further judge whether the start time startTime is less than 438 minutes, if so, then the activity type is ⑧ other life, if the startTime is not less than 438 minutes, then further judge whether the duration of the activity is greater than It is equal to 72 minutes. If not, the activity type is ⑧ other life. If it is further judged whether the type of stay is administrative or business office space, if it is, the activity type is ⑤ business. If not, it is ⑧ other life.

If the stay category is not equal to 2 or 5, further judge whether the activity duration of the activity is less than 22 minutes, if not, then the activity type is ⑧ other life, if so, further judge whether the stay category is 4 or 6 or 9 , if so, the activity type is ⑧ other life, if not, further judge whether the duration of the activity is less than 9.5 minutes, if yes, then the activity type is ⑥ pick-up person, if not, further judge whether the activity start time startTime is less than 500 Minutes, if yes, the activity type is ⑥ pick-up person, otherwise it is ⑧ other life.

According to this judgment rule, inputting the start time and duration of the activity, and the type of stay point corresponding to the nature of the land use can determine other types of activities except work, or at home, or going to school.

The user activity type discrimination method based on smart device data provided by the present invention can analyze the user's activity type in different time periods of a day according to the user's smart device data, thereby providing important references for urban planning, traffic management and user activity research. Moreover, the time threshold or space threshold involved in the analysis process is calibrated by traditional user travel survey data, which is more usable, accurate and real-time.

The technical features of the above-mentioned embodiments can be combined arbitrarily. To make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, should be considered as within the scope of this specification.

The above-mentioned embodiments only express several implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but should not be construed as limiting the patent scope of the invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be based on the appended claims.

Claims (10)

1. A user activity type discrimination method based on smart device data, is characterized in that, comprises the following steps: Obtain user smart device data; Perform data cleaning on the smart device data to obtain the user's movement trajectory, and divide the user's stay section and travel section through itinerary identification to obtain the user's travel chain; Extracting the activity start time and activity duration of each of the stay sections in the travel chain, and obtaining the land use properties corresponding to the stay sections according to the points of interest of the stay sections; Analyze the smart device data of the user's travel at the preset time, and obtain the corresponding stay period, length of stay, and number of stays of each user, so as to determine the location of the user's home and/or work place, and obtain two types of activities at home or at work; By inputting the activity start time, activity duration, and land use properties corresponding to the stay segments other than the two types of activities at home or at work into the activity classifier, the two types of activities of each user except at home or at work can be obtained outside the predefined activity types. 2. The method for discriminating user activity types based on smart device data according to claim 1, wherein the method further comprises: building the activity classifier, comprising: Based on the traditional user travel survey data, the activity start time, activity duration, and land use properties corresponding to the stay section of each user in the survey data are extracted to construct an activity classifier based on a decision tree. 3. The method for discriminating user activity types based on smart device data according to claim 1, wherein the data cleaning includes field missing processing, deleting abnormal IMSI number records, and deleting records that cannot be matched with the base station positioning data. Matching records, deduplication, ping-pong effect handling, and signal drift handling. 4. The method for discriminating user activity types based on smart device data according to claim 3, wherein the ping-pong effect processing comprises the steps of: combining the smart device data of each user with regions according to space and time , if the user signal fluctuates within a range smaller than the spatial threshold L1 and exceeds the time threshold T1, it is considered that the user is at the same location during this period. 5. The method for discriminating user activity types based on smart device data according to claim 3, wherein the signal drift processing comprises the following steps: combining the smart device data of each user into areas according to space and time , if the user leaves the space threshold L2 within the time threshold T2 and then returns to the space threshold L2, it is considered that the user is at the same location. 6. The method for discriminating user activity types based on smart device data according to claim 1, wherein the itinerary identification comprises the following steps: if the user's trajectory points within the time threshold T _stay are clustered within the radius of L _stay range, or the moving speed within the time threshold T _stay is lower than the speed threshold V _stay , then the corresponding section is a stay section, otherwise it is a travel section. 7. The user activity type discrimination method based on smart device data according to claim 6, further comprising the steps of: Convert the travel data in the user travel survey data into activity chain data, and filter the data of users who leave home in the morning and go home in the evening; Extract the activity duration of all types of activities, establish the activity duration distribution, and use the p% quantile of the activity duration distribution as the time threshold T _stay for trip identification, where p is any natural number from 5 to 10. 8. The method for discriminating user activity types based on smart device data according to claim 1, wherein said obtaining the land use properties corresponding to said stay section according to the points of interest of said stay section comprises the following steps: Calculate the center coordinates of the stop position by time weighting; According to the position and quantity of the point of interest corresponding to the center coordinates, a kernel density estimation model is established, and its formula is as follows: <mrow><mi>f</mi><mrow><mo>(</mo><mi>s</mi><mo>)</mo></mrow><mo>=</mo><munderover><mo>&amp;Sigma;</mo><mrow><mi>i</mi><mo>=</mo><mn>1</mn></mrow><mi>n</mi></munderover><mfrac><mn>1</mn><mrow><msup><mi>nr</mi><mn>2</mn></msup></mrow></mfrac><mi>K</mi><mrow><mo>(</mo><mfrac><msub><mi>d</mi><mrow><mi>i</mi><mo>,</mo><mi>s</mi></mrow></msub><mi>r</mi></mfrac><mo>)</mo></mrow></mrow> K(.) represents the kernel function; r represents the window width; n represents the total number of points of interest; d _{i, s} represents the distance from the center coordinates to each point of interest; Choose a Gaussian function as the kernel function: <mrow><mi>K</mi><mrow><mo>(</mo><mfrac><msub><mi>d</mi><mrow><mi>i</mi><mo>,</mo><mi>s</mi></mrow></msub><mi>r</mi></mfrac><mo>)</mo></mrow><mo>=</mo><mfrac><mn>1</mn><msqrt><mrow><mn>2</mn><mi>&amp;pi;</mi></mrow></msqrt></mfrac><mi>exp</mi><mrow><mo>(</mo><mo>-</mo><mfrac><mrow><msup><msub><mi>d</mi><mrow><mi>i</mi><mo>,</mo><mi>s</mi></mrow></msub><mn>2</mn></msup></mrow><mrow><mn>2</mn><msup><mi>r</mi><mn>2</mn></msup></mrow></mfrac><mo>)</mo></mrow></mrow> Calculate the kernel density values of different interest point types at the stay position, and take the land use property corresponding to the interest point with the highest kernel density value as the land use property of the stay position. 9. The method for discriminating user activity types based on smart device data according to claim 1, wherein the smart device data of the user's preset travel time is analyzed to obtain the corresponding period of stay, duration of stay, and number of stays, Based on this, the location of the user's home and/or work place is determined to obtain two types of activities at home or at work, including the following steps: extracting all of said activity start times, said activity durations, and land use properties for each user; Filter out the data of each user's working day, and count the total number of days as N For each type of stay position, count the total number of days N _{home where the night stay time is greater than T home ;} If N _home is greater than the first judgment time threshold, then this position is the home location, otherwise, count the total number of days N _{work where the stay time during working hours is greater than T work ;} If N _work is greater than the second judging time threshold, the location is a working location. 10. The user activity type discrimination method based on smart device data according to claim 9, further comprising the steps of: Convert the travel data in the user travel survey data into activity chain data, and filter the smart device data of users who leave home in the morning and return home in the evening; Extract the activities where the place of stay at night is home, establish the distribution of activity duration, and identify the time threshold T _home of the a% quantile family location of the distribution of activity duration, where a is any natural number from 5 to 10; Extract the activities where the place of stay during the working period is the work place, establish the activity duration distribution, and use the b% quantile of the activity duration distribution as the time threshold T _work for work place identification, where b is any natural number from 5 to 10.