Urban resident working duration extraction method based on long-time-sequence mobile phone signaling data

Abstract

The invention discloses an urban resident working duration extraction method based on long-time-sequence mobile phone signaling data. The method comprises the following steps: preprocessing original mobile phone signaling data; based on speed, distance and time threshold values, user track resident point identification is carried out; the residence place and the activity residence point of the core worker are identified; extracting context features, space-time attribute features, geographic semantic features, user personal information and call information of activity residence points of core workers; training a working state discrimination model based on a random forest; and based on the random forest model obtained by training, extracting the working duration of the urban residents. The invention aims to reduce the acquisition cost of the used data, and can solve the problem that the current method cannot comprehensively analyze the working state and duration of residents in a city in a large scale.

Description

Technical Field

[0001] The invention relates to a working time extraction method, in particular to a method for extracting the working time of urban residents based on long-time sequence mobile phone signaling data. Background Art

[0002] With the development of social economy, the phenomenon of overtime work of workers has become more and more serious, which has aroused social concern. Determining the work activities of urban residents, especially extracting the working hours and distribution of urban residents, can effectively discover hot spots in the city and vulnerable groups in work activities, thereby providing a unique perspective and reference for urban planning and management.

[0003] At present, the research on residents' working hours is mainly based on traditional questionnaire surveys and field interviews. Previous research on residents' work was mainly based on document surveys to explore the overwork of workers in certain industries and units. However, the sample of these studies is small and the distribution of industries is concentrated, which does not have universality for the entire industry. At the same time, it is difficult to grasp the work behavior of urban residents as a whole by mainly using methods such as questionnaires. With the development of information and communication technology, more and more new technologies and new data are being applied to urban research. Mobile phone signaling data provides good data support for urban research with its characteristics of covering many users and regions, good real-time performance and high positioning accuracy. Based on mobile phone signaling data, there are already some methods to identify the user's work and residence location. For example, the public patent CN113613174A identifies the user's stay cluster through the grid clustering method, and then extracts the user's work and residence through the time window method; the public patent CN118433649A uses the user's long-term mobile phone signaling data to extract his stable residence, and further screens and completes the identification of the city's permanent population. Although existing methods have been applied to the identification of urban mobile phone users’ workplaces and residences and the determination of permanent population, current methods still lack a focus on the accurate identification of urban residents’ work status and the effective extraction of working hours, and are unable to conduct in-depth exploration of the overall work situation of urban residents. Summary of the invention

[0004] In order to solve the shortcomings of the above-mentioned technology, the present invention provides a method for extracting the working hours of urban residents based on long-time-series mobile phone signaling data.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for extracting the working hours of urban residents based on long-time mobile phone signaling data, comprising the following extraction steps:

[0006] S1. Obtain original mobile phone signaling data of users in the city and pre-process the original mobile phone signaling data;

[0007] S2. For the pre-processed mobile phone signaling data, identify the user trajectory residence point based on speed, distance and time thresholds;

[0008] S3. Screen the core workers and identify and extract their work and residence places and their activity locations;

[0009] S4, context feature extraction of core workers’ activity residence points;

[0010] S5. Extracting the spatiotemporal attribute features of the core workers’ activity residence points;

[0011] S6, extracting geographic semantic features of the core workers’ activity locations;

[0012] S7, extracting user personal information and call information of core workers;

[0013] S8. Training of working status discrimination model based on random forest: by extracting context features, spatiotemporal attribute features, geographic semantic features and personal information of the user's stay, taking them as input variables, setting whether the user's stay status is working as a prediction variable, and performing model training of the random forest model;

[0014] S9. Based on the trained random forest model, extract the working hours of urban residents.

[0015] Preferably, in S1, data preprocessing includes deleting abnormal data and processing duplicate data.

[0016] Preferably, in S2, the user trajectory station point identification method is:

[0017] After preprocessing the user's mobile phone signaling data, the user's signaling records are sorted in chronological order;

[0018] By setting the speed threshold, we can filter out the jumping points of the user's trajectory.

[0019] By setting the stay distance threshold, the user's stay subsequence is extracted;

[0020] According to the start time and end time of the stay subsequence, the stay duration of each stay point is calculated, and the valid stays whose stay duration exceeds the set threshold are screened out;

[0021] Finally, the point with the highest occurrence frequency of different resident subsequences is calculated as the resident point of this resident;

[0022] After the residence point extraction, the user's mobile phone signaling data is converted into the user's stay point sequence. The stay point sequence is a five-tuple of base station longitude, base station latitude, start time, end time, and stay time.

[0023] Preferably, in S3, the specific method for identifying the work and residence of the core worker is:

[0024] First, identify permanent residents of the city: define mobile phone users as permanent residents if they have records for at least 6 days in a week;

[0025] Second, identify the stable residence of urban permanent residents: according to the trajectory residence point records in S2, select the residence point with the longest stay time from 0:00 to 6:00 in the morning as the residence point of the day, and extract the daily residence point location with the highest frequency in a week as the stable residence of permanent residents;

[0026] Third, identification of stable work locations of permanent residents: For users with stable residences, their stable work locations are further identified. According to the trajectory residence point records in S2, the residence point with the longest stay time from 9 am to 5 pm on weekdays is defined as the work location of the day. The daily work point location with the highest frequency in a week is extracted as the user's stable work location;

[0027] Fourth, screen core workers. The screening rules include: ① The distance between the worker’s residence and workplace is greater than 1,000 meters; ② The worker’s age is between 20 and 59 years old.

[0028] Preferably, in S4, the context features of the extracted activity residence points include historical sequence state features, time context features and space context features;

[0029] The method for extracting the state feature of the historical sequence of the residence point is as follows: taking the middle moment of the residence point as the benchmark, extracting the position of the core worker's historical activity sequence at that moment, calculating the average daily frequency of appearing at the stable work place at that moment in the historical sequence, and taking this feature as the state feature of the historical sequence of the residence point;

[0030] The extracted temporal context features include: the time interval between the activity dwell point and the two previous and next dwell points;

[0031] The extracted spatial context features include: the distance between the active residence point and the two previous and next residence points.

[0032] Preferably, in S5, the specific method for extracting the spatiotemporal attribute features of the activity residence point is:

[0033] Calculate the time characteristics of the activity dwell points in S3, which include the start time, end time, dwell duration of each dwell period, and whether the day is a working day;

[0034] Calculate the spatial characteristics of the active residence point, which is the distance between the active residence point and the stable working place.

[0035] Preferably, in S6, the method for extracting geographic semantic features of the activity residence point is: extracting its geographic semantic information according to the location of the activity residence point, firstly dividing the study area into 500-meter grids, and extracting the land use type of the location of the activity residence point; secondly extracting the number of different types of POIs within the 500-meter grid range of each activity residence point, and finally calculating the POI mixed entropy within the 500-meter grid range of the activity residence point.

[0036] Preferably, in S7, the extraction of user personal information and call features includes:

[0037] (1) User personal information: Extract the personal attribute characteristics of core workers, including the gender and age of mobile phone users, by matching the mobile phone user attribute information table.

[0038] (2) Call information at the user's activity residence point: Calculate the call characteristics of the core worker during his / her stay at each activity residence point, including the number of calls, call duration, etc., and represent them as the call characteristics of the core worker.

[0039] Preferably, in S8, the specific method of model training is: utilizing the user's mobile phone signaling data with existing working tags, identifying the user's residence trajectory points and extracting their active residence points through steps S1-S3, extracting the user's residence context features, spatiotemporal attribute features, geographic semantic features and personal information through steps S4-S7, and using them as input variables, setting whether the user's residence status is working as a prediction variable, inputting them as sample data into the random forest model, training the model, and using the k-fold verification method to evaluate the model accuracy, thereby optimizing the model parameters.

[0040] Preferably, in S9, the mobile phone signaling data of urban residents is used to identify the urban residents' residence trajectory points through steps S1-S3, and the core workers and their activity residence points are identified through step S3, and then the context features, spatiotemporal attribute features, geographic semantic features and personal information of the activity residence points are extracted through steps S4-S7, and the random forest model for distinguishing the user's work status trained in S8 is used to distinguish the work status of the activity residence points of the urban core workers;

[0041] The duration of work stay is accumulated on a daily and weekly basis to calculate the daily and weekly working hours of urban residents.

[0042] The present invention discloses a method for extracting working hours of urban residents based on long-time mobile phone signaling data, which can extract the working hours of users according to their mobile phone signaling data within a period of time. This method breaks through the limitations of traditional working hours surveys that rely on questionnaires, high data mobile phone costs, and small coverage, and can effectively extract the working hours of urban residents, which helps to support managers' policy formulation and provide assistance for urban planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the overall process of the method of the present invention.

[0044] Figure 2 A detailed process diagram is extracted for the working hours of urban residents. DETAILED DESCRIPTION

[0045] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0046] Most of the existing methods for collecting residents' working hours rely on questionnaire survey results. However, the current methods have a small coverage, high data collection costs, and are unable to effectively extract the overall working status and working hours of urban residents. This paper proposes a method for extracting the working hours of urban residents based on long-term mobile phone signaling data, aiming to reduce the cost of obtaining the data used, and to solve the problem that the current methods cannot conduct a large-scale and comprehensive analysis of the working status and working hours of urban residents.

[0047] like Figure 1 As shown, the method for extracting working hours of urban residents based on long-time mobile phone signaling data proposed in the present invention specifically includes the following steps:

[0048] S1: mobile phone signaling data preprocessing;

[0049] The purpose of this step is to remove abnormal records and duplicate data from the original mobile phone signaling data of users in the city. The complete data after preprocessing can improve the accuracy of overtime status.

[0050] First, obtain the user's mobile phone signaling data for one week. Its main fields can be simplified into the following sequence:

[0051] Traj = {(lon 1 ,lat 1 ,t 1 ),(l 2 ,ci 2 ,t 2 ),……(lon n ,lat n ,t n )}

[0052] Among them, (lon n ,lat n ,t n ) represents a trajectory point. A trajectory point is generated when the user's mobile phone interacts with the base station; lon n and lat n Indicates the latitude and longitude of the base station location to which the user's mobile phone is connected when the interaction occurs; t n Indicates the time when the interaction between the mobile phone and the base station occurs.

[0053] Abnormal data refers to data with incomplete fields and unclear information. Abnormal data will be removed during preprocessing. Duplicate data refers to data of multiple locations where a user appears at the same time. Only one single record of the duplicate data will be retained during data preprocessing.

[0054] S2: Identify user trajectory points based on speed, distance and time thresholds;

[0055] The purpose of this step is to extract the dwelling points from the user's mobile phone signaling records. People in the city carry out complex activities every day. For employed people in the city, most of the states of participating in work activities are staying in a certain place to complete related work. Since most workers are in a state of staying when they are working. Therefore, for the identification of the dwelling points of urban residents, the key is to process their mobile phone signaling trajectory sequence, extract the dwelling subsequences of different dwelling points, and then calculate the dwelling position and dwelling duration of different dwelling points.

[0056] The main steps for mobile phone user residence point identification are as follows:

[0057] First, after S1 preprocesses the user's mobile phone signaling data, the user's signaling data is sorted in chronological order. Specifically, the mobile phone signaling data of a certain day is grouped according to the user ID from the database, and then the trajectory point records of different users for one day are obtained. Then the trajectory point records of different mobile phone users are sorted in chronological order, and finally the trajectory sequence of different mobile phone users for one day is obtained, which is a continuous sequence in time and space. The trajectory sequence can be expressed as:

[0058] S i ={p 1 ,p 2 ,……p n}

[0059] Among them, S i is the trajectory sequence of the i-th user, p n Indicates the nth trajectory point of the user.

[0060] Second, we can filter out the jumping points of the user's trajectory by setting the speed threshold. 0 =50m / s. For a user's trajectory sequence, the speed between two adjacent points in the trajectory is measured in sequence.

[0061] V i,i+1 =dist i,i+1 / Δt i,i+1

[0062] Among them, V i,i+1 represents the speed between the i-th trajectory point and the i+1-th trajectory point, dist i,i+1 represents the distance between the i-th trajectory point and the i+1-th trajectory point, Δt i,i+1 Represents the time difference between the i-th trajectory point and the i+1-th trajectory point.

[0063] If p in the user trajectory sequence i and p i+1 Speed ​​V i,i+1 Greater than V 0 , then the trajectory point p i+1 Remove from the current sequence. Continue to calculate p i and p i+2 speed between the threshold speed V 0 In comparison, if the current speed is less than V 0 , then keep p i+2 The user's trajectory sequence is traversed in sequence until all trajectory points in the sequence are checked. The output of this step is a sequence without jump points.

[0064] Third, by setting the dwell distance threshold, extract the user's dwell subsequence. Process the trajectory sequence without jump points. Set the dwell distance threshold to d 0 =200m; first, the current point p i Store in subcolumn, measure p i and p i+1 The distance between i,i+1 :If d i,i+1 If the distance is less than 200m, i+1 Add to p i In the subsequence where it is located, continue to calculate the trajectory point p i+1 and trajectory point p i+2 The distance between them is compared with the distance threshold; if d i,i+1 is greater than 200m, then the trajectory point p i The sub-sequence ends, and p i+1 Add to the new sub-column and continue to calculate the trajectory point p i+1 and trajectory point p i+2 The distance between i+1,i+2Compare with the distance threshold. Traverse all the trajectory points in the trajectory sequence until the processing is completed. The processing result of this step is the set of stay subsequences.

[0065] Fourth, extract the location and time of the stay point. According to the start time and end time of the stay point subsequence, calculate the stay duration of each stay point, and select the valid stays whose stay duration exceeds the stay threshold; finally, calculate the point with the highest occurrence frequency of different stay subsequences as the stay point of this stay.

[0066] Specifically, for the combination of the stay subsequences processed in the third step, the stay time of the subsequences is calculated as:

[0067] Duration i =T n -T 1

[0068] Among them, Duration i is the residence time of the ith stop point, T n is the time when the last trajectory point of the stop subsequence is reached, T 1 The time when the first point of the subsequence is stopped. If the subsequence has only one point, the time is 0 and it does not constitute a stop point.

[0069] Further filter the effective stay points, keep the stay points with a stay time of more than 30 minutes, and select the point with the most occurrences of trajectory points in the stay subsequence as the area where the stay point is located. Finally, all the stay points of mobile phone users are extracted.

[0070] After the stay point extraction, the original user mobile phone signaling data is converted into the user's stay point sequence. The stay point sequence is a five-tuple of base station longitude, base station latitude, start time, end time, and stay time, which is expressed as follows:

[0071] Stay i ={(lon 1 ,lat 1 ,sdt 1 ,edt 1 ,dur 1 ),……(lon m ,lat m ,sdt m ,edt m ,dur m )}

[0072] Among them, Stay irepresents the trajectory of the i-th user, which contains m stay points. Each trajectory point is a quintuple (lon, lat, sdt, edt, dur), where (lon, lat) represents the longitude and latitude of the base station in the area where the stay point is located, (sdt, edt) represents the start time and end time, and dur represents the stay duration.

[0073] S3: Identification of the work and residence of core workers and screening of their activity sites;

[0074] The purpose of this step is to screen the core workers in the city and identify and extract their work and residence places and activity points.

[0075] The vast majority of studies have shown that in the process of work-residence identification, the location of the residence is the core area of ​​human activities, and the residence should be identified first. According to the living habits of normal urban residents, most people rest at home at night and in the early morning, while work activities are generally carried out in the morning and afternoon of weekdays. Therefore, based on this judgment, the time window method is used to identify the residence of urban residents. Work-residence separation means that the residence and the workplace are not in the same location. In order to extract working hours more effectively, the present invention only considers the extraction of working hours for workers with work-residence separation. Finally, considering factors such as retirement and schooling, it is necessary to exclude mobile phone users who are too old or too young.

[0076] like Figure 2 As shown in the figure, the steps for identifying the work and residence of urban core workers and screening their activity locations are as follows:

[0077] First, we screen the permanent residents of the city and define mobile phone users as permanent residents if they have at least 6 records in a week. Specifically, we select a week of mobile phone user stay data from the database. We group them by mobile phone user ID, check the user stay data, and screen out people who have been in the current city for at least 6 days from Monday to Sunday as permanent residents.

[0078] Second, the identification of the stable residence of permanent residents. The specific identification method is as follows: define the residence time window as 0:00 to 6:00 a.m., select the residence trajectory data of user S2, and extract the longest stay point in the time window as the daily residence point of the day. Extract the user's daily residence every week to obtain the set of daily residence points for the week. Select the daily residence with the highest frequency as the stable residence of the user.

[0079] Third, identify the stable work locations of permanent residents. For users with stable residences, further identify their stable work locations; if there is no stable residence, the user will be eliminated.

[0080] The method for identifying the stable workplace of permanent residents is as follows: define the time window of the workplace as 9 am to 5 pm on weekdays, select the residence trajectory data of user S2, and extract the longest stay point in the time window as the daily work point of the day. Then extract the user's daily work point set for a week, select the daily work point with the highest frequency as its stable workplace, and then identify the stable workplace of permanent residents in the city.

[0081] Fourth, screen core workers. After extracting the residents’ stable residence and workplace, further screen core workers. The screening rules are as follows: (1) Calculate the distance between the user’s residence and workplace. If the distance between the residence and workplace is greater than 1000m, keep it; otherwise, screen it out. (2) Considering the legal age of workers, screen out residents under 20 years old and over 59 years old, and keep users between 20 and 59 years old.

[0082] After the above steps, the urban core residents of appropriate age with separated work and residence are obtained. The objects of subsequent feature extraction are all the core workers identified in this step.

[0083] S4: Context feature extraction of core workers’ activity residence points;

[0084] The purpose of this step is to extract the spatiotemporal context features of the activity residence points, in preparation for the next step of using the random forest model to identify the user's work status.

[0085] The contextual work status features of the activity residence points can show the dynamic changes of work activities between residence points as a whole. By considering the work status association between residence points, the overall situation of work activities can be analyzed more comprehensively, and the work status of each residence point can be accurately determined, thereby improving the comprehensiveness and reliability of residence point work status identification.

[0086] The time context characteristics of the activity dwell point focus on the time interval between the activity dwell point and the two previous and next dwell points, in order to gain a deeper insight into the continuity and rhythm of work activities on the time axis. The length of the time interval can reflect the close connection between work tasks. A shorter time interval means that the work process may have a strong coherence, and the task conversion is relatively fast, reflecting the efficiency and compactness of the work; while a longer time interval may imply that there are pauses, task conversions or waiting in the work process. By analyzing these time intervals, we can better grasp the time context and rhythm characteristics of the work.

[0087] The spatial contextual characteristics of the activity residence point focus on the distance between the activity residence point and the two residence points before and after it, aiming to comprehensively characterize the spatial distribution and continuity of work activities. The distance can show whether the work activities are relatively concentrated or dispersed in space. A smaller distance indicates that the work may be carried out more intensively in a certain area, while a larger distance may mean that the work involves transfers to different areas, out-of-town work, etc., which helps to understand the spatial layout and dynamic changes of work.

[0088] Based on the above judgment, the present invention mainly divides the context feature extraction into three aspects:

[0089] (1) Historical sequence status characteristics, mainly the working status frequency of the core workers’ historical sequence. Taking the middle moment of the residence point as the benchmark, the position of the core workers’ historical activity sequence at that moment is extracted, and the average daily frequency of appearing at the stable work location at that moment in the historical sequence is calculated. This feature is used as the historical sequence status feature of the residence point.

[0090] (2) Temporal context features, such as the time interval between an activity’s dwell point and its previous dwell point, and the time interval between an activity’s dwell point and its next dwell point.

[0091] (3) Spatial context features, such as the Euclidean distance between an activity point and its previous point, and the Euclidean distance between an activity point and its next point.

[0092] This step comprehensively extracts the contextual features of the candidate residence points of the core workers, providing a strong multi-dimensional support for more accurate identification of the work status of the residence points, avoiding the isolation of the work situation of a single residence point, and thus improving the accuracy of work status identification.

[0093] S5: Extracting spatiotemporal attribute features of core workers’ activity residence points;

[0094] The purpose of this step is to extract the spatiotemporal attribute characteristics of the activity residence points, in preparation for the next step of using the random forest model to identify the user's work status.

[0095] Workers usually carry out work activities in a specific time period. For example, they are more likely to be working during the daytime on weekdays. Therefore, extracting the time characteristics of the residence record helps to better determine whether they are in a working state. The start time, end time and residence duration of each residence are the most basic time markers, which can clearly define the specific start and end range and duration of each work stage of the core workers, and provide accurate time boundaries for subsequent analysis of their working time patterns and working period distribution.

[0096] At the same time, analyzing the spatial characteristics of activity residence points helps to grasp the dynamic changes of work activities in space, the layout rules, and the relationship with the core work area, providing an important basis for studying the spatial distribution pattern of work and the impact of different spatial locations on work. Residence activities near stable workplaces are more likely to be work activities. The closer the residence activities are to a stable workplace, the greater the probability of work activities. At the same time, except for workers with a large range of activities such as takeaways, couriers, and drivers, work activities generally have a relatively stable workplace. When mobile phone users are working, their range of activities is relatively small.

[0097] Based on the above judgment, the spatiotemporal attribute features extracted by the present invention include two aspects:

[0098] (1) Calculate the time characteristics of the activity dwell points in S3, which include the start time, end time, dwell duration of each dwell period, and whether the day is a working day;

[0099] (2) Calculate the spatial characteristics of the active residence point in S3, which includes the distance between the active residence point and the stable working place.

[0100] This step comprehensively describes the core workers' active residence points from the time and space dimensions of their candidate residence points, which helps to improve the accuracy and reliability of work status identification.

[0101] S6: Extraction of geographic semantic features of core workers’ activity locations;

[0102] The purpose of this step is to extract the geographic semantic features of each user's residence record, in preparation for the next step of using the random forest model to identify the user's work status.

[0103] The land use type of a residence point can reveal the functional attributes of the environment in which the residence point is located. Different land use types are usually closely related to specific work types and patterns. For example, commercial areas may focus on business negotiations, sales and other work, while industrial areas are more involved in production and manufacturing. Understanding land use types helps to correspond work to specific geographical functional areas. The number of different types of POIs reflects the resource richness and functional diversity of the surrounding area. A larger number of POIs means that more commercial services, social activities and other resources are available. Therefore, the type and number of POIs near the residence point are of great significance for judging the type of residence activities.

[0104] In this step, the geographic semantic feature extraction method of the activity residence point is as follows: the geographic semantic information is extracted according to the location of the activity residence point. First, the study area is divided into 500-meter grids, and the land use type of the location of the activity residence point is extracted. Secondly, the number of different types of POIs within the 500-meter grid range of each activity residence point is extracted. Finally, the POI mixed entropy within the 500-meter grid range of the activity residence point is calculated.

[0105] The geographic semantic features of the activity residence point include: the land use type of the residence point, the number of POIs of different types within 500 meters of the residence point, and the mixed entropy of POIs within 500 meters of the residence point.

[0106] Based on the spatial connection between the location of the residence point and the urban land use distribution, the land use type of the residence point can be extracted. Urban land use types can be divided into residential areas, commercial areas, industrial areas, transportation areas and public service areas.

[0107] According to the POI classification specification, the number of different types of POIs within 500 meters of the station point is determined. The first-level classification of POIs includes: catering services, accommodation services, resident services, wholesale and retail, automobile services, finance and insurance, education and culture, health care, sports, parks and attractions, public facilities, commercial facilities, companies and enterprises, transportation, scientific research institutes, agriculture, forestry, animal husbandry and fishery, etc. Different types of POIs can represent different urban functional areas.

[0108] The calculation of the mixed entropy of POIs within 500 meters of the station point includes: counting the number of different types of POIs within 500 meters of the station point, and then calculating their mixed entropy.

[0109]

[0110] Among them, E is the POI mixing entropy, R i is the proportion of the i-th type POI in the area.

[0111] This step extracts the geographical semantic features of the activity residence point in a relatively detailed manner, aiming to deeply explore the geographical environment connotation of the place where the work activity occurs and its potential impact on the work. It helps to understand the development of the work and the environmental constraints or promotion factors from the perspective of geographical background, and further enriches the basis for identifying the work status of the residence point.

[0112] S7: Extraction of user personal information and call information of core workers;

[0113] The purpose of this step is to extract user personal information and call information in preparation for the next step of using the random forest model to determine the user's work status.

[0114] The gender and age characteristics of core workers are basic personal attribute information. Workers of different genders and ages may have differences in work methods, work preferences, work capabilities, etc. Phone calls are a common way of communication at work. A large number of phone calls and a long call duration may mean that the workload at this stage of residence is heavy, requiring more communication and coordination, or involving important business negotiations. The call information such as the number of calls and call duration of each core worker's stay can reflect the communication status of the worker at each stage of residence.

[0115] Based on the above judgment, the present invention extracts the user personal information and call information of the core workers in the following ways:

[0116] (1) User personal information: Extract the personal attribute characteristics of core workers, including the gender and age of mobile phone users, by matching the mobile phone user attribute information table.

[0117] (2) Call information at the user's activity residence point: Calculate the call characteristics of the core worker during his / her stay at each activity residence point, including the number of calls, call duration, etc., and represent them as the call characteristics of the core worker.

[0118] S8: Training of working status discrimination model based on random forest;

[0119] The purpose of this step is to train a random forest model using mobile phone signaling data with labels indicating that the user is in a working state, so as to identify the user's working state based on the contextual features, spatiotemporal attribute features, geographic semantic features, and personal information of the residence.

[0120] For the mobile phone signaling data with the user in working state label, the user's residence trajectory points are identified and their active residence points are extracted through steps S1-S3, and the user's residence context features, spatiotemporal attribute features, geographic semantic features and personal information are extracted through steps S4-S7 to construct a training data set containing M training sample points:

[0121] T={(x i ,y i )|i=1,2,…,M}

[0122] where x i =(x i1 ,x i2 ,…x in ) is the input feature of the user’s i-th residence point, x in The context features, spatiotemporal attribute features, geographic semantic features, and personal information of the activity residence point are represented by y i Indicates whether the station is in working state: (1) working; (2) not working.

[0123] Therefore, the present invention uses context features, spatiotemporal attribute features, geographic semantic features and personal information as input variables, and sets whether the user's residence status is working as a prediction variable; that is, the constructed training data set is input into the random forest model for pattern training, and the formula of the random forest is:

[0124]

[0125] Wherein, RF(x) is the final discrimination result, Fi(x) is the discrimination result of the i-th decision tree, ntree and mtry are model parameters, which respectively represent the number of decision trees in the random forest and the number of randomly selected features for each decision tree. The model can be continuously iterated and updated with the accumulation of training data sets, thereby improving the discrimination accuracy. The present invention specifically adopts the k-fold verification method to evaluate the model accuracy and then optimize the model parameters.

[0126] S9: Extraction of working hours of urban residents;

[0127] Based on the random forest model trained in the previous step, it is possible to determine whether the user's residence point is in working condition.

[0128] like Figure 2 As shown in the figure, given a week of mobile phone signaling data of any user, firstly, the mobile phone signaling data of urban residents is used to identify the user's residence trajectory points and extract their active residence points through steps S1-S3, and the contextual features, spatiotemporal attribute features, geographic semantic features and personal information of the user's residence are extracted through steps S4-S7. The random forest model trained in S8 for distinguishing the user's work status is further used to distinguish the work status of the active residence points of urban core workers. Finally, the duration of work residence is accumulated by day and week respectively to calculate the daily working hours and weekly working hours of urban residents.

[0129] In summary, the method for extracting working hours of urban residents based on long-term mobile phone signaling data proposed in the present invention takes mobile phone signaling data as the data basis, completes the user stay subsequence identification based on speed, distance and time threshold methods, and realizes the extraction of user stay points; before extracting various features of the stay record, the city's core workers are first identified, and the identification rule is that the user must have a stable place of work and residence, with the characteristics of separation of work and residence, and the user's age is between 20 and 59 years old; on this basis, the time characteristics, spatial characteristics, spatiotemporal context characteristics and personal information of the user's stay record are used to judge the user's work status, and combined with a random forest or neural network model, the extraction of working hours of urban residents is realized.

[0130] Compared with the prior art, the method of the present invention has the following technical advantages:

[0131] (1) High accuracy: Mobile phone signaling data can provide real-time location information of users, with an accuracy of tens to hundreds of meters. This helps to accurately identify whether users are at work and can effectively identify the working status of mobile phone users in the city, and then extract their working hours.

[0132] (2) Good real-time performance. This method can understand the user's work behavior pattern by analyzing the long-term mobile phone signaling data. As long as the user's mobile phone signaling data over a period of time is obtained, the working status of urban residents can be analyzed in real time, which has good real-time performance;

[0133] (3) Wide coverage and low cost. Compared with the questionnaire survey method, this method is lower in cost and has a wider coverage, and does not require large-scale questionnaire surveys or field observations.

[0134] The above implementation modes are not limitations of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the scope of the technical solution of the present invention also belong to the protection scope of the present invention.

Claims

1. A method for extracting working hours of urban residents based on long-time mobile phone signaling data, characterized in that: The extraction steps include: S1. Obtain original mobile phone signaling data of users in the city and pre-process the original mobile phone signaling data; S2. For the pre-processed mobile phone signaling data, identify the user trajectory residence point based on speed, distance and time thresholds; S3. Screen the core workers and identify and extract their work and residence places and their activity locations; S4, context feature extraction of core workers’ activity residence points; S5. Extracting the spatiotemporal attribute features of the core workers’ activity residence points; S6, extracting geographic semantic features of the core workers’ activity locations; S7, extracting user personal information and call information of core workers; S8. Training of working status discrimination model based on random forest: by extracting context features, spatiotemporal attribute features, geographic semantic features and personal information of the user's stay, taking them as input variables, setting whether the user's stay status is working as a prediction variable, and performing model training of the random forest model; S9. Based on the trained random forest model, extract the working hours of urban residents.

2. The method for extracting working hours of urban residents based on long-time mobile phone signaling data according to claim 1 is characterized in that: In S1, data preprocessing includes deleting abnormal data and processing duplicate data.

3. The method for extracting working hours of urban residents based on long-time mobile phone signaling data according to claim 1 is characterized in that: In S2, the user trajectory station point identification method is: After preprocessing the user's mobile phone signaling data, the user's signaling records are sorted in chronological order; By setting the speed threshold, we can filter out the jumping points of the user's trajectory. By setting the stay distance threshold, the user's stay subsequence is extracted; According to the start time and end time of the stay subsequence, the stay duration of each stay point is calculated, and the valid stays whose stay duration exceeds the set threshold are screened out; Finally, the point with the highest occurrence frequency of different resident subsequences is calculated as the resident point of this resident; After the residence point extraction, the user's mobile phone signaling data is converted into the user's stay point sequence. The stay point sequence is a five-tuple of base station longitude, base station latitude, start time, end time, and stay time.

4. The method for extracting working hours of urban residents based on long-time mobile phone signaling data according to claim 1 is characterized in that: In S3, the specific method for identifying the work and residence of core workers is: First, identify permanent residents of the city: define mobile phone users as permanent residents if they have records for at least 6 days in a week; Second, identify the stable residence of urban permanent residents: according to the trajectory residence point records in S2, select the residence point with the longest stay time from 0:00 to 6:00 in the morning as the residence point of the day, and extract the daily residence point location with the highest frequency in a week as the stable residence of permanent residents; Third, identification of stable work locations of permanent residents: For users with stable residences, their stable work locations are further identified. According to the trajectory residence point records in S2, the residence point with the longest stay time from 9 am to 5 pm on weekdays is defined as the work location of the day. The daily work point location with the highest frequency in a week is extracted as the user's stable work location; Fourth, screen core workers. The screening rules include: ① The distance between the worker’s residence and workplace is greater than 1,000 meters; ② The worker’s age is between 20 and 59 years old.

5. The method for extracting working hours of urban residents based on long-time mobile phone signaling data according to claim 1 is characterized in that: In said S4, the context features of the extracted activity residence points include historical sequence state features, time context features and space context features; The method for extracting the state feature of the historical sequence of the residence point is as follows: taking the middle moment of the residence point as the benchmark, extracting the position of the core worker's historical activity sequence at that moment, calculating the average daily frequency of appearing at the stable work place at that moment in the historical sequence, and taking this feature as the state feature of the historical sequence of the residence point; The extracted temporal context features include: the time interval between the activity dwell point and the two previous and next dwell points; The extracted spatial context features include: the distance between the active residence point and the two previous and next residence points.

6. The method for extracting working hours of urban residents based on long-time mobile phone signaling data according to claim 1 is characterized in that: In S5, the specific method for extracting the spatiotemporal attribute features of the activity residence point is: Calculate the time characteristics of the activity dwell points in S3, which include the start time, end time, dwell duration of each dwell period, and whether the day is a working day; Calculate the spatial characteristics of the active residence point, which is the distance between the active residence point and the stable working place.

7. The method for extracting working hours of urban residents based on long-time mobile phone signaling data according to claim 1 is characterized in that: In S6, the method for extracting geographic semantic features of the activity residence point is: extracting its geographic semantic information according to the location of the activity residence point, firstly dividing the study area into 500-meter grids, and extracting the land use type of the location of the activity residence point; secondly extracting the number of different types of POIs within the 500-meter grid range of each activity residence point, and finally calculating the POI mixed entropy within the 500-meter grid range of the activity residence point.

8. The method for extracting working hours of urban residents based on long-time mobile phone signaling data according to claim 1 is characterized in that: In S7, the extraction of user personal information and call features includes: (1) User personal information: Extract the personal attribute characteristics of core workers, including the gender and age of mobile phone users, by matching the mobile phone user attribute information table. (2) Call information at the user's activity residence point: Calculate the call characteristics of the core worker during his / her stay at each activity residence point, including the number of calls, call duration, etc., and represent them as the call characteristics of the core worker.

9. The method for extracting working hours of urban residents based on long-time mobile phone signaling data according to claim 1 is characterized in that: In S8, the specific method of model training is: using the user's mobile phone signaling data with existing working tags, identifying the user's residence trajectory points and extracting their active residence points through steps S1-S3, extracting the user's residence context features, spatiotemporal attribute features, geographic semantic features and personal information through steps S4-S7, and using them as input variables, setting whether the user's residence status is working as a prediction variable, and inputting it as sample data into the random forest model to train the model, and using the k-fold verification method to evaluate the model accuracy, thereby optimizing the model parameters.

10. The method for extracting working hours of urban residents based on long-time sequence mobile phone signaling data according to claim 1 is characterized in that: In said S9, the mobile phone signaling data of urban residents is used to identify the urban residents' residence trajectory points through steps S1-S3, and the core workers and their activity residence points are identified through step S3, and then the context features, spatiotemporal attribute features, geographic semantic features and personal information of the activity residence points are extracted through steps S4-S7, and the random forest model for distinguishing the user's work status trained in S8 is used to distinguish the work status of the activity residence points of the urban core workers; The duration of work stay is accumulated on a daily and weekly basis to calculate the daily and weekly working hours of urban residents.

Images