Method and system for identifying transportation mode

Abstract

The invention discloses a method and system for identifying a transportation mode, and relates to the technical field of intelligent transportation, and the method comprises the steps: obtaining mobile phone signaling original data of a user in a set time and region, and carrying out the characterization processing of the original data; extracting mobile phone signaling original data of a special time node by using an OD model and carrying out independent data processing; calibrating the mobile phone signaling data by combining the mobile phone signaling data cleaned by the time threshold and the space threshold; and establishing a city double-layer model and substituting all calibrated mobile phone signaling data for training to realize identification of user travel modes in set time and regions. The invention further discloses a system for identifying the traffic mode, the system comprises a data acquisition module, a central processing module and a feature extraction module, the identification degree and accuracy of the urban double-layer model are improved, a complete traffic mode travel track path is formed by using the model, and accurate identification of the user travel mode in a set area is realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a method and system for identifying transportation travel modes. Background Art

[0002] Identifying the mode of transportation of users plays a very critical role in understanding user mobility, analyzing and predicting traffic conditions, and mining social activity patterns. The results of comprehensive urban travel mode analysis can be used for planning traffic plans to solve congestion problems, and can also be used to count the travel modes of residents in tourist areas, so as to design more efficient transportation service plans, which has high social significance. Traditional methods of urban transportation travel modes are mostly based on certain traffic models based on conditions such as residents' living conditions, the status of surrounding traffic facilities, and sampling surveys. The data collection cost is high and the data accuracy is low. With the rapid development of smart hardware and LBS technology, mobile phone signaling has the characteristics of wide data coverage, low data collection cost, and high time continuity. The conditions for identifying travel modes based on mobile phone signaling big data analysis technology and AI artificial intelligence technology are gradually met. However, due to the problems of mobile phone signaling big data itself, the accuracy of the existing method of identifying users' travel modes with the help of mobile phone signaling is low, and thus it is impossible to provide effective identification results. Therefore, a new method of identifying traffic travel modes is urgently needed to solve the current situation.

[0003] Chinese patent document CN112542045A discloses a "method and system for identifying transportation modes based on mobile phone signaling". It adopts step 1, based on the city boundary GIS data, obtains the data reported by the user's mobile communication terminal within the city as a sample, including signaling data, personal information, travel vehicles and GPS data. The signaling data is the data containing the terminal number, time and base station longitude and latitude reported by the mobile communication terminal when switching the base station sector; step 2, according to the signaling data of the mobile communication terminal, obtains the corresponding signaling trajectory point set; step 3, identifies the stop point based on the signaling trajectory point set, and determines the mid-point and end point of the trip in the signaling trajectory point set to construct the user's signaling trajectory; step 4, based on the quality The signaling trajectory of the user obtained by smoothing step 3 is replaced by the center; step 5, based on the stop point as the key point, the smoothed signaling trajectory of the user is divided into multiple OD chains with the travel end point as the end point; step 6, based on the city's proprietary scene traffic infrastructure data, the four proprietary travel modes of high-speed rail, motor vehicle, light rail and rail transit in the OD chain are identified; step 7, the OD chain of the proprietary travel mode identified in step 6 is removed from the OD chain of step 5, and the corresponding signaling trajectory points in the remaining OD chain are matched with the GIS road network data to obtain multiple corresponding sub-segment information; step 8, the road network matching results of the user every day are compared. Sort by time field, and then cut the time window according to the preset window duration and sliding step length; Step 9, match the sub-segment information in the time window cut in Step 8 with the travel mode reported by the user, and label the trajectory segment of the window with the travel mode; Step 10, obtain feature values ​​based on the user-reported data in Step 1 and each signaling trajectory point corresponding to the sub-segment in the time window cut in Step 8, the feature values ​​include the user's personal age and gender information and the maximum instantaneous speed, average speed, speed variance and movement displacement of each signaling trajectory point in the time window; Step 11, based on the feature values ​​in Step 10, The SVM binary classifier is used to train the travel mode recognition model to obtain the recognition model; Step 12, for the actual input data to be predicted, the OD chain is extracted according to the aforementioned steps 2-5, and then the travel mode is recognized: (1) First, four exclusive travel modes are identified based on the city's exclusive scene traffic infrastructure data; (2) Then, the time window obtained by cutting according to steps 7-8 and the feature value obtained based on the time window are input into the recognition model trained in step 11 to identify the travel mode. However, the accuracy of judging the user's travel mode through mobile phone signaling in this patent is insufficient to meet the user needs of the existing society. Summary of the invention

[0004] The present invention mainly solves the technical problem of insufficient accuracy in identifying the travel mode of users through mobile phone signaling, and provides a method and system for identifying transportation travel modes. After obtaining the original mobile phone signaling data, the accuracy of the data is improved by processing and calibrating the signaling data. Finally, the processed mobile phone signaling data is substituted into the urban double-layer model to realize the identification of the travel mode of users in the set time and area. The above method can solve the problem that the accuracy of the original mobile phone signaling data is unstable and prone to fluctuations, and realize the accurate identification of the travel mode of users in the required area with the help of mobile phone signaling data.

[0005] The above technical problem of the present invention is mainly solved by the following technical solution: The present invention comprises S1 acquiring the original data of mobile phone signaling of users within a set time and region and performing characterization processing on the original data; S2 uses the OD model to extract the original data of mobile phone signaling at special time nodes and performs separate data processing; S3 combines the mobile phone signaling data cleaned by the time threshold and the space threshold to calibrate the mobile phone signaling data; S4 builds a two-layer model of the city and substitutes all calibrated mobile phone signaling data into training to realize the recognition of users' travel modes within the set time and region.

[0006] Preferably, when processing mobile phone signaling data at special time nodes, a corresponding daily travel OD matrix is ​​established according to different important time nodes, and then the daily travel OD matrix is ​​reduced in dimension by principal component analysis, and the OD matrices of different special time nodes are processed using dynamic fuzzy C-means clustering method to obtain the characteristic data results of the special time nodes.

[0007] Preferably, in step S3, at the time threshold level, the spatial threshold is quantitatively set based on the characteristics of the processed mobile phone signaling data, the value is taken according to the spatial interval m, and the change rate of the travel number identification at different intervals is calculated, and the average value of the travel change rate of this part of the mobile phone signaling data is combined to determine whether there is an abnormality in the value; at the spatial threshold level, a dynamic spatial threshold is adopted according to the travel chain identification result, that is, the spatial threshold is selected according to the base station density, and the mobile phone signaling data is calibrated by adjusting the threshold of the area with different travel times.

[0008] Preferably, travel characteristic indicators are set before calibrating the signaling data, and the mobile phone signaling data is calibrated according to the set travel characteristic indicators. If problems are found in the mobile phone signaling data under any travel characteristic indicator, the data error points are corrected under the travel characteristic indicator by tracing back the characteristics of the mobile phone signaling data itself and data identification.

[0009] Preferably, the urban two-layer model includes accurate fitting of paths and two-layer recognition of multi-dimensional spatiotemporal characteristics. By analyzing the differences between the spatiotemporal distribution of mobile phone signaling data and the road network structure, travel paths in the data are extracted in a fitting and matching form. Under the interaction of multi-dimensional variables such as spatiotemporal dimensions and travel rates of different modes, the differences in travel characteristics are described to complete the construction of the urban two-layer model.

[0010] Preferably, the urban two-layer model first reconstructs the travel trajectory using a linear interpolation algorithm. After the travel trajectory is reconstructed, the SG filter is used to further optimize the mutation characteristics, and the linear interpolation algorithm is used to supplement the missing data in the discrete points. After the travel path is fitted, the probability matching algorithm is used to match the fitted trajectory points with the travel path to confirm the correspondence between the travel path and the travel mode.

[0011] Preferably, the step S1 comprises: S11 obtains the original mobile phone signaling data of users within the required set time and area and performs original data processing; S12 collects the original processed mobile phone signaling raw data and performs desensitization processing; S13 matches the desensitized mobile phone signaling data with the travel trajectory map of the area; S14 extracts characteristic parameters of the user's travel based on the matching results of the mobile phone signaling data and the travel trajectory map, and completes the characterization processing of the mobile phone signaling data.

[0012] Preferably, in the process of characterization of the mobile phone signaling data, after determining the data extraction range, the feature data is cleaned, the cleaned feature data is attribute decomposed, and finally the feature data of different scales are converted to a unified scale to complete normalization processing.

[0013] Preferably, when the urban double-layer model trains mobile phone signaling data, the input travel characteristics are used as the key to the model's supervised learning algorithm. After the urban double-layer model inputs the mobile phone signaling data at the spatiotemporal attribute level, it uses the maximum value, average value, minimum value and variance for identification, and uses similarity to characterize the travel mode, thereby completing the identification of the travel mode of the mobile phone signaling data within the set time and area.

[0014] A system for identifying transportation travel modes includes: a data acquisition module, including a plurality of base station data acquisition and transmission devices, connected to a central processing module; the central processing module centrally processes mobile phone signaling data to generate processing results; a feature extraction module performs feature processing and obtains the feature result recognition module in the mobile phone signaling data, and determines the travel mode of users in different areas and time periods through the analysis results of the mobile phone signaling data.

[0015] The beneficial effects of the present invention are: combining the OD model with the time threshold and the space threshold to process and calibrate the original mobile phone signaling data, solving the problem that the existing mobile phone signaling data is prone to fluctuations and causes inaccurate feedback results. On the basis of improving the accuracy of mobile phone signaling data as a data source for identification, an urban double-layer model is established to fit and match mobile phone signaling data at travel trajectory points and travel paths, so that the travel mode can be accurately identified through mobile phone signaling data, thereby improving the rationality of builders' urban road planning and the effectiveness of users' travel mode selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flow chart of mobile phone signaling data processing.

[0017] Figure 2 It is a flow chart of the operation of a city double-layer model of the present invention. DETAILED DESCRIPTION

[0018] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0019] Embodiment: A kind of embodiment of this embodiment is as follows Figure 1 and Figure 2 As shown, it includes obtaining the original data of mobile phone signaling of users within the set time and region and performing characterization processing on the original data. When processing, the original data of mobile phone signaling of users within the required set time and region is obtained and the original data is processed; the original processed mobile phone signaling data is collected and desensitized; the desensitized mobile phone signaling data is matched with the travel trajectory map of the area; the characteristic parameters of the user's travel are extracted according to the matching results of the mobile phone signaling data and the travel trajectory map, and the characterization processing of the mobile phone signaling data is completed. The processing flow of the original data is as follows: The user identification code in the original data is used as the unique ID of the user data and processed for easy use; Extract valid information from the original data, including travel start and end time, user identification code, base station location area code and sector number. Clear data with special characters in the user identification code and incomplete time and space information; Convert the recorded time data information into the form of year, month, day, hour and minute, and arrange them in ascending order according to chronological order; 4. The processed data contains the base station location area code and latitude and longitude information, and the corresponding base station latitude and longitude data have been merged. The two data sources are associated and the null value data is deleted to obtain more complete data.

[0020] After the original data is processed, the mobile phone signaling data is cleaned. First, the data is retrieved to remove the duplicate parts. Next, the missing information is processed. The so-called missing information refers to the situation where the data cannot be fully presented, aggregated or combined due to incomplete or lost information. This article uses the mean filling method for processing. If the missing information attribute is data type, the average value of other non-missing data is used as filling; for non-data type missing information, the value with the most occurrences of the attribute is selected for filling; and for outliers where the deviation between the measured value and the mean value exceeds three times the standard deviation, it should also be treated as missing data and processed.

[0021] Two phenomena will occur during the collection of mobile phone signaling data: data ping-pong effect and data drift. The ping-pong effect refers to the situation that when the signal strength of two base stations in a specific area changes dramatically, the location information of the mobile phone will be constantly switched between the two base stations. This may result in a large number of records of mobile phone signaling data without the user actually changing location. Data drift is actually a protection mechanism and optimization strategy. For example, when the base station load in the user's area is too high, the mobile phone signal will be connected to a relatively distant but lower-loaded base station, resulting in anomalies, which will bring greater instability to the signaling data. For the ping-pong effect, combined with the characteristics of mobile phone signaling, the travel chain of the signaling data user is extracted, and a circle is drawn with the current signaling data location as the center. When the next signaling data jumps outside the circle within a specific time and returns to the original location, the user is considered to be staying. Common ping-pong effect data forms are 1-2-1 and 1-2-3-1. The processing flow is as follows: 1. Extract the time and base station code fields of the mobile phone user and use them as the user's spatiotemporal information; 2. Process 1-2-1 type ping-pong data. Take the user's nth data, if the base station code of the n+1th data is different from that of the nth data, but the base station code of the n+2th data is the same as that of the nth data, and the time difference between the n+1th data and the nth data is less than the ping-pong effect data switching time threshold, then it is considered to be 1-2-1 ping-pong data, and the n+1th data is deleted; 3. Process 1-2-3-1 type ping-pong data. Take the user's nth data, if the base station codes of the n+1th data and the nth data, the n+1th data and the nth data, and the n+2th data and the nth data are all different, and the time difference between the n+2th data and the nth data is less than the ping-pong effect data switching time threshold, then it is considered to be 1-2-3-1 ping-pong data, and the n+1th and n+2th data are deleted; 4. Set n = n + 1 until all the user's data is traversed; 5. Return to the first step and identify the next user until all users are identified.

[0022] For data drift, speed detection and screening are first used. Data drift often occurs in an instant, which causes the travel speed of residents during this period to increase sharply. The distance of drift is calculated through the geographic location information obtained from the mobile phone signaling data. To complete the cleaning of mobile phone signaling data, it is also necessary to desensitize the mobile phone signaling data. In the process of extracting and using the mobile phone signaling data, some of the user's private information will appear, such as user name, location coordinates, mobile phone contact information and other specific information. This embodiment uses special characters to replace sensitive content, thereby destroying the readability of private data, and does not retain the original meaning and format, thereby achieving the effect of privacy protection.

[0023] After the basic processing steps of mobile phone signaling data are completed, it is necessary to match the residents' travel trajectory with the map of the set area. When matching, the travel chain is first obtained to obtain the user's residence and travel location data. Combined with the basic characteristics of the user's access to the network and communication-related preference attributes, as well as the fusion of multi-source data, it provides support for the acquisition of the travel trajectory chain. The process of obtaining the user's travel trajectory chain through the big data platform is as follows: 1. Confirm the user's location status. The user's status of the day can be divided into stay and move. During the user's movement, the base station collects the user's travel trajectory point data every 15 seconds to identify the user's real-time status. Generally, a user who stays in the same place for more than 30 minutes is considered to be staying. The user's daily stay status is recorded in the platform's table. Each record represents a stay behavior, which includes information such as the stay location grid, start and end time, and number. The stay_poi table is set to summarize the user's monthly stay behavior, including the number of all the user's stay locations in that month, the location grid, the length of stay and the frequency of stay in each time period, and other information.

[0024] 2. Record the travel behavior between two stay points. The travel behavior between two stay points is regarded as a travel chain, and the travel behavior between two important road nodes (nodes that determine the direction) is regarded as a travel segment. A travel chain consists of multiple travel segments, and a travel segment consists of several track points. The travel segments are recorded in the move_month table, including the number of each user's trip, the starting and ending location grid and longitude and latitude, starting and ending time, distance and other information.

[0025] 3. Associate user attribute information. Associate the user basic attribute table with the base station and road network location table through fields to achieve feature statistics in different areas and time periods.

[0026] 4. Trip segment sorting: Sort the extracted user stay and trip segment information in chronological order to form a travel trajectory chain.

[0027] When performing map matching, the method of associating the user base station location sequence collected by the mobile communication network with the road network on the electronic map is the process of converting the sampling sequence under the base station coordinates into the road network coordinate sequence. Map matching is widely used in the fields of global positioning system trajectory positioning, mobile device data positioning, etc., and has been used as a key means to solve the deviation of user path positioning in the field of traffic information. Due to the excessive amount of travel trajectory data during matching, a candidate matching road range is set with the base station positioning coordinates of the travel segment as the center and 1km as the radius. Roads are selected within this range and used as the user's original trajectory data. The replaced travel segments are sorted to determine the user's travel trajectory. Taking some roads in the central urban area as an example, this paper adopts the line-to-line matching method in the geometric analysis method to determine the road with the highest similarity corresponding to the travel segment in the road network, so as to achieve road matching of user trajectory points. The travel trajectory matching process is as follows: Basic data preparation. Use ArcGIS software to associate the route node table and obtain the path longitude and latitude information. For each base station location information, search for candidate road sections within 1 km around the point to form a correspondence table between base stations and surrounding roads.

[0028] Arrange the road sequence table to obtain the road candidate set. Calculate the similarity between the travel segment and each road in the road candidate set. This paper uses the longest common subsequence (LCSS) algorithm to calculate the similarity between the trajectory and the road. Let A and B be a trajectory of length n and a road of length m, respectively.

[0029] Spatial fitting is performed between the user's travel segment and the road with the highest similarity. The matching result of the user's travel trajectory is obtained. The matched travel segment is reflected in the geographic information software.

[0030] Finally, the travel characteristics are extracted and analyzed. The travel characteristics include the analysis of the spatial and temporal distribution characteristics of residents' travel. Mastering the time distribution characteristics of residents' travel is conducive to the early deployment of traffic planning schemes, which can effectively alleviate traffic congestion, traffic flow distribution and other problems, and is of great significance to the identification of residents' travel modes. Users in the set area will generate many travel trajectories in their daily production and life. According to the trajectories, the travel chains of mobile phone users can be obtained. By analyzing a large number of complex travel chains, it can usually be found that they have certain regularities, and then the travel spatial distribution characteristics can be obtained. The flow of users between signal base station cells is used as the embodiment of spatial distribution characteristics. If the number of users per unit time in the set area increases, it means that residents' travel during this period tends to flow into the area, and vice versa. In order to more clearly grasp the mobility of residents' travel, the main urban area cells are rasterized based on the latitude and longitude of the signal base station. The k clustering algorithm is used to cluster the base station cells according to the latitude and longitude coordinates, and they are numbered one by one. Based on the clustered areas, the Moran index is calculated to understand the correlation of residents' travel in the urban space. The clustering algorithm can process large data sets with clear effects, has low sensitivity to outliers and noise, and can achieve base station cell clustering at a high level. The specific implementation of the clustering algorithm in this embodiment is as follows: 1. First, randomly select K centroids (the centroids must be the values ​​of some sample points).

[0031] 2. Calculate the distance from each point to the centroid.

[0032] 3. Divide the class of the point into the centroid closest to it to form K clusters.

[0033] 4. Based on the divided clusters, recalculate the centroid in each cluster (calculate the sum of the distances from all samples in the cluster to one of the points, and select the sample with the smallest sum of distances as the new centroid).

[0034] 5. Repeat the iteration 2-4 times until the classification errors are within the allowable range.

[0035] The spatiotemporal characteristics of mobile phone signaling data include user travel distance characteristics. Travel distance refers to the distance of residents' travel. Combined with the travel trajectory extracted from the signaling data, the travel distance of residents can be calculated, so as to conduct statistical analysis; user travel speed. Travel speed includes instantaneous speed and average speed. Since instantaneous speed fluctuates greatly and is affected by many factors, the average speed is selected as the speed characteristic indicator of this section in this embodiment. At the same time, the average speeds of different travel modes are bound to be different; user travel destination. Combined with the starting and ending point locations, travel time, residence time and other parameters in the travel chain data, the user travel chain data is divided according to the travel purpose as the classification principle. The specific classification is as follows: the traveler's residence point in the city of arrival is obtained, and the residents' point of interest data is extracted from the data platform in combination with the mobile phone signaling data. Extract the location vector data of public facilities such as schools, shopping malls, and entertainment venues from the geographic spatial data, and match the traveler's stay point with the corresponding location vector data. If the traveler's stay point appears within the scope of shopping malls, entertainment venues, etc. and the stay time exceeds the time threshold, the purpose of travel is determined to be work, otherwise it is life; if the traveler's stay point appears within the scope of the school and the stay time exceeds the time threshold, the purpose of travel is determined, and the stay point information is combined with the POI data. If the travel end point appears within the scope of the user's home location and the stay time exceeds the time threshold, the purpose of travel is determined to be going home, otherwise it is life; if the traveler's stay point appears within the scope of the workplace and the stay time exceeds the time threshold, the purpose of travel is determined to be school, otherwise it is life, and the travel purposes of the remaining travel chains are judged as others; extract and analyze the social attributes of travelers. Combined with the results of the extraction of residents' travel characteristics in the previous section, the social attributes of travelers such as gender, age, occupation, education level, and personal average monthly income are extracted and analyzed.

[0036] After acquiring the characteristic data of the mobile phone signaling data, the characteristic data is preprocessed. The preprocessing of the characteristic data includes determining the data extraction range and cleaning the data. The specific steps are as follows: 1. Determine the data extraction scope and ensure data association matching. Combined with the analysis of travel time characteristics and travel space characteristics, the time range of mobile phone signaling data used to extract feature data is set in summer, and the spatial range is set in areas 3, 7, and 9 in the central urban area. When building and applying the model, the road network data and mobile phone signaling data are associated and matched.

[0037] 2. Data cleaning. Information confusion is inevitable in the process of resident travel surveys. Therefore, during the data processing process, the feature data needs to be processed to a certain extent, which can not only improve the authenticity and effectiveness of the model, but also reduce the complexity of the data and improve work efficiency.

[0038] After the feature data is preprocessed, it is complete and usable data, but it cannot be brought into the model for training. It is necessary to decompose the attribute of the feature data, convert the feature value in text form when input into digital form, decompose the feature value of the category attribute, and use numbering processing on the basis of classification to improve the information utilization rate of the input data. The remaining features in the feature data are numbered and classified respectively. After that, the feature data is normalized. In order to convert the pure numerical data of different scales to the same scale, it is necessary to perform dimensionless processing so that the features of different attributes are comparable. Common dimensionless methods include extreme value, standardization, mean value, normalization and standard deviation methods. In this embodiment, the numerical values ​​of the feature value intervals of each feature data set are quite different, and the feature data is processed by max-min normalization.

[0039] Since users in the set area and time are affected by different forms of festivals, politics, and activities at special time nodes, the travel mode of users at special time nodes is quite different from the normal mode. Therefore, it is impossible to uniformly process the mobile phone signaling data of special time nodes and the normal mobile phone signaling data. It is necessary to consider and analyze them separately based on their specific conditions. Therefore, the characterization processing of mobile phone signaling data also needs to use the OD model to extract the original data of mobile phone signaling at special time nodes and perform separate data processing. Based on the mobile phone signaling data, the travel of special time nodes is extracted and gridded. The user's mobile phone signaling data record is matched to the grid, and the coordinates of the center point of the grid are used as the user's position at this time. Then, the binning process is performed, and the data is divided into several intervals of equal width according to the time dimension. Then, a representative value is selected in each box to represent the data in the box. This processing method can overcome the data noise caused by the "ping-pong switching" switching back and forth in a short time. Through binning and grid processing, not only can the problems of ping-pong switching be solved, but also the calculation efficiency of large-scale mobile phone signaling data can be improved. After pre-processing the OD of residents' travel, the grid OD is aggregated. Based on the above, the set area is divided into several small areas, and the dynamic fuzzy C-means clustering method is used for further processing. By establishing the uncertainty description of sample attributes, the comprehensive effect of multiple indicators can be considered to cluster the time series data. The specific implementation steps are as follows: 1. Preprocess the original time series matrix to determine the length of the time series and the number of variables m. 2. Calculate the weight of each point in the class relative to the cluster center. Assuming that it needs to be clustered into k classes, randomly generate k cluster center coordinates in the time dimension and set them as the kth cluster center coordinates; 3. Calculate the degree of membership of each point to each cluster center; 4. Determine the DFCM objective function. Under the premise of determining the number of clusters, a smaller objective function means greater similarity within the class, greater difference between classes, and better clustering effect; 5. Calculate the ordered clustering validity function to describe the clustering validity and indicate the minimum distance between two adjacent cluster centers; 6. Using the ordered clustering validity function as the optimization function, a multivariate nonlinear programming problem with cluster centers and cluster numbers as variables is established. The accelerated genetic algorithm is used to solve the nonlinear optimization function of the optimal cluster centers and cluster numbers, and the comprehensive membership matrix is ​​recorded.

[0040] Through the above clustering method, the user's mobile phone signaling travel data at special time nodes is statistically processed to obtain the recognition results of the user's travel mode at different special time nodes. The judgment of the user's travel mode at different special time nodes and normal conditions is uniformly collected, and the mobile phone signaling data after cleaning and processing with the time threshold and space threshold is combined to calibrate the mobile phone signaling data. Data cleaning includes travel chain identification, identifying the user's stay through the time threshold and space threshold, and identifying a trip between two stays in different places. The smaller the space threshold and time threshold, the more trips are identified. At the time threshold level, this embodiment starts from the characteristics of the mobile phone signaling data itself, takes the space threshold respectively, and sets the value interval. In each case, the change rate of the identification of the number of trips at different nodes in each case is calculated. At the spatial threshold level, the travel chain identification of this embodiment adopts a dynamic spatial threshold, that is, the selection of the spatial threshold is related to the base station density. The K fan method is selected as the dynamic threshold algorithm. Under this dynamic threshold algorithm, the spatial threshold is low in places with high base station density, and vice versa. The threshold is high. Thereby solving the recognition error caused by the uneven base stations in different regions of the fixed spatial threshold, and obtaining a more accurate recognition result.

[0041] The comparison of travel characteristic indicators includes: travel intensity. The travel intensity characteristics are analyzed by two indicators: the proportion of travel occurrence and the number of trips per capita. Travel time. The travel time characteristics are compared using two indicators: travel time consumption and travel time. Travel distance and travel OD. In terms of travel space characteristics, this embodiment compares the average travel distance and the distribution difference of the starting and ending points of the trip. The straight-line distance between the starting and ending points of the trip is used for calculation, and the difference in OD distribution is further compared. It can be found that the OD distribution of the two types of data at the street scale and above has a high degree of consistency. Travel purpose. The residents' travel survey data divides the residents' travel purposes into 8 categories: going to work, going to school, culture, sports, entertainment and leisure, business, picking up people, going home, medical treatment, and daily life, while mobile phone signaling data cannot obtain such information. Therefore, this article mainly divides the travel purpose into two categories: commuting travel and non-commuting travel, and compares and analyzes it with the travel survey data.

[0042] When making comparisons, it is necessary to judge the factors that affect the quality of mobile phone signaling data. In the data collection process, some low-quality data exist in the mobile phone signaling data, which makes it difficult to accurately describe the user's spatiotemporal behavior, and thus leads to deviations in travel characteristics. Its impact can be analyzed from three dimensions: the integrity of the recording period, the integrity of the travel chain, and the time interval of location recording. In terms of the integrity of the recording period, the location records of a large number of low-quality signaling data are not uniform and continuous in time, making it difficult to fully describe the user's spatiotemporal behavior, resulting in deviations in travel characteristics. Among them, the original signaling data recording period is incomplete, which has a significant impact on the identification of the number of trips. In terms of the integrity of the travel chain, the conventional method of cutting mobile phone signaling data will destroy the integrity of the travel chain. In order to improve the calculation efficiency of MSD, mobile phone signaling data is usually cut, but this data processing method will lead to under-identification of travel. When the number of location records at the residence is small at night, the cutting of the data makes it difficult to clearly identify the situation of staying at home. In addition, the long time interval of location recording is the main reason for the error in travel time. In the process of processing and identifying mobile phone signaling data, the selection of time thresholds for travel chain identification will cause two forms of under-identification of travel, thus causing deviations in the travel volume measurement results. In terms of the deviation caused by the selection of base station density and spatial thresholds, the spatial accuracy of mobile phone signaling data depends on the base station density distribution. When the user is in the overlapping area of ​​the base station service range, even if the actual location has not changed, it may be recorded by several neighboring base stations, also known as the "base station hopping" phenomenon. This phenomenon is due to the fact that the recording time interval characteristics are no different from normal data, and cannot be removed by a similar method of cleaning "ping-pong data", resulting in deviations in travel volume identification. In response to the above problems, it is necessary to use a spatial threshold as a "spatial tolerance" to prevent the "base station hopping" phenomenon of neighboring base stations from being mistakenly identified as travel.

[0043] However, due to the uneven distribution of base stations at the urban scale, the central urban area base stations are densely distributed, and the suburban base stations are sparsely distributed. Therefore, the spatial threshold of each base station will dynamically adjust the threshold size according to the density and distance of the surrounding base stations, that is, the "dynamic spatial threshold" algorithm is used to solve the problems at the above levels. The conventional dynamic spatial threshold has the following problems: 1. The base station itself has a certain signal coverage range. If the user's location moves within the service range of the same base station, it cannot be identified as a trip. This part of the deviation is related to the base station density of the mobile phone signaling data, and it is difficult to directly verify and correct it. 2. The spatial threshold can largely prevent the "base station jumping" phenomenon between adjacent base stations during the identification process and misidentify it as a trip, but it also causes short-distance trips between adjacent base stations to be ignored. The deviation of short-distance travel caused by the spatial threshold setting can be verified based on HTS. Comparing the difference in the number of per capita trips between MSD and HTS short-distance trips, the mobile phone signaling data calibrated at the time threshold and spatial threshold levels are obtained.

[0044] After calibration, the urban double-layer model is established and all calibrated mobile phone signaling data are substituted for training to realize the recognition of user travel modes within the set time and region. At the level of travel path recognition, SG filtering can effectively smooth the fluctuation of signaling data relative to the actual travel path, and the linear interpolation algorithm can make up for the spatial and temporal loss of signaling data. At the level of travel mode recognition, key factors such as recognition path similarity, travel time similarity, acceleration, wavelet velocity, etc. are explored, and the K-nearest neighbor algorithm is used to identify buses and cars. The results show that the method proposed in this paper can effectively subdivide bus and car travel in the dense urban road network environment, and the recognition accuracy reaches 88.29% and 82.28% respectively. From the perspectives of different travel distances, travel time periods, congestion status, road grades, road types and recognition path similarity, the recognition effect is better than that of random forest algorithms. The research supports the precise mining of travel characteristics based on signaling data, and provides an important foundation for road planning and construction, bus network planning, etc. In the process of establishing the urban double-layer model, the path-mode motorization recognition process needs to be completed first. Considering the differences between the spatiotemporal distribution of signaling data and the road network structure, the motorized travel path is extracted through the "fitting-matching" step, and then the differences in travel characteristics between buses and cars are characterized through multi-dimensional variables such as time, space, speed, and speed fluctuations, and a travel mode identification model is constructed. According to the spatial distribution and triggering time interval of the signaling data, optimization is performed from the two levels of path fitting and matching, respectively, to achieve effective identification of travel paths in a dense road network environment and complete travel path identification based on signaling data. Then, the SG filter is used to smooth the fluctuations of the signaling data relative to the travel road to achieve travel path fitting. Among them, the SG filter can smooth the high-frequency part of the data without changing the overall change trend, and has the characteristics of fast and stable. Mobile phone signaling data has irregular triggering time. When the time interval is too large, direct matching will cause large-scale path loss. In this regard, after smoothing, this embodiment first uses a linear interpolation algorithm to reconstruct the travel trajectory, and then uses the SG filter again to further optimize the mutation characteristics and reconstruct the complete travel trajectory. Finally, the travel path matching and optimization are completed. On the basis of the distance-based matching algorithm, a probability matching algorithm based on quantity is proposed to ensure that each fitting trajectory point can match at least one road. It is assumed that the matching road segment set μ of each signaling point p contains m nearest road segments r. Then, for any sub-signaling sequence, match this sub-sequence to the same road segment. The fluctuation of signaling data relative to the travel road is greatly reduced by SG filtering, but due to the complexity of communication technology and network, some signaling data still has a large offset, resulting in incorrect path matching results. The specific steps for optimizing the offset are as follows: 1. Starting from any matching road segment, determine whether the road numbers are the same, and if so, continue to determine. 2. When the road numbers are different, determine whether the matching roads are the same, and if they are the same, remove them; if they are different, delete the duplicate values ​​in the road and ensure that the order remains unchanged.3. If all matching data have been identified, end the step; otherwise, return to step 1.

[0045] After completing the travel path identification, it is necessary to identify the travel mode by integrating multi-source features again. Affected by many factors such as the external environment, the travel characteristics between different motorized modes are slightly different, and there is even a certain degree of interweaving, which is difficult to distinguish directly. Use the K-nearest neighbor algorithm to predict the travel mode. This method analyzes the K data classifications closest to the target point to determine the category to which the target point belongs. It has the advantages of being insensitive to outliers and fast training speed. End the identification work step and input features for supervised learning of the model. The two-layer model input in this embodiment considers the spatial and temporal attributes of the travel space path, time, speed, acceleration, wavelet transform speed, etc., and further uses the maximum value, average value, minimum value, and variance to characterize. Among them, the travel time is mainly characterized by similarity, which represents the degree of difference between the signaling recognition time and the estimated time of the same OD point in the network map; the wavelet transform can highlight the local changes in the time or spatial frequency of the target data, and finally obtains the input features including: recognition path similarity, travel time similarity, travel speed average, maximum value and variance, travel acceleration average, maximum value and variance, wavelet transform speed average, maximum value and variance. The above-mentioned multiple parameter features enable accurate identification of the user's travel mode within a set area and time.

[0046] Therefore, in this embodiment, the current traditional travel mode identification methods and models are relatively backward and cannot identify the travel modes of different residents in the urban transportation system, and mobile phone signaling data inevitably produce deviations during the collection process, resulting in the existing dilemma of being unable to effectively use mobile phone signaling data to identify user travel modes. This embodiment ensures the key to the authenticity of trajectory data by performing map matching on mobile phone signaling travel chain data, which can increase the credibility of subsequent feature processing. Combined with map matching of computer programs, it is possible to effectively integrate models and calculation methods to solve the problem of trajectory data positioning. Combined with the travel characteristics of residents extracted from multi-source data, it can further enhance the applicability of machine learning models for travel mode identification, which is conducive to solving the problem of single mobile phone signaling data source in travel mode identification. At the same time, innovations have been made in the research models and methods, expanding the application of mobile phone signaling data and improving the accuracy of identifying and judging user travel modes with the help of mobile phone signaling.

Claims

1. A method for identifying a mode of transportation, characterized in that: include S1 obtains the original mobile phone signaling data of users in a set time and region and performs characterization processing on the original data; S2 uses the OD model to extract the original data of mobile phone signaling at special time nodes and performs separate data processing; S3 combines the mobile phone signaling data cleaned by the time threshold and the space threshold to calibrate the mobile phone signaling data; S4 builds a two-layer model of the city and substitutes all calibrated mobile phone signaling data into training to realize the recognition of users' travel modes within the set time and region.

2. A method for identifying a transportation mode according to claim 1, characterized in that: When processing mobile phone signaling data at special time nodes, a corresponding daily travel OD matrix is ​​established according to different important time nodes, and then the daily travel OD matrix is ​​reduced in dimension through principal component analysis. The dynamic fuzzy C-means clustering method is used to process the OD matrices of different special time nodes to obtain the characteristic data results of the special time nodes.

3. A method for identifying a transportation mode according to claim 1, characterized in that: In step S3, at the time threshold level, the spatial threshold is quantitatively set based on the characteristics of the processed mobile phone signaling data, and the value is taken according to the spatial interval m and the change rate of the travel number identification at different intervals is calculated. Combined with the average travel change rate of this part of the mobile phone signaling data, it is judged whether there is an abnormality in the value; at the spatial threshold level, a dynamic spatial threshold is adopted according to the travel chain identification result, that is, the spatial threshold is selected according to the base station density, and the mobile phone signaling data is calibrated by adjusting the threshold of the area with different travel times.

4. A method for identifying a transportation mode according to claim 3, characterized in that: Before calibrating the signaling data, a travel characteristic index is set, and the mobile phone signaling data is calibrated according to the set travel characteristic index. If a problem is found in the mobile phone signaling data under any travel characteristic index, the data error point is corrected under the travel characteristic index by tracing back the characteristics of the mobile phone signaling data itself and identifying the data.

5. The method for identifying a transportation mode according to claim 1, characterized in that: The urban two-layer model includes accurate fitting of paths and two-layer recognition of multi-dimensional spatiotemporal characteristics. By analyzing the differences between the spatiotemporal distribution of mobile phone signaling data and the road network structure, the travel paths in the data are extracted in a fitting and matching form. Under the interaction of multi-dimensional variables such as spatiotemporal dimensions and travel rates of different modes, the differences in travel characteristics are described to complete the construction of the urban two-layer model.

6. A method for identifying a transportation mode according to claim 5, characterized in that: The urban two-layer model first reconstructs the travel trajectory using a linear interpolation algorithm. After the travel trajectory is reconstructed, the SG filter is used to further optimize the mutation characteristics, and the linear interpolation algorithm is used to supplement the missing data in the discrete points. After the travel path is fitted, the probability matching algorithm is used to match the fitted trajectory points with the travel path to confirm the correspondence between the travel path and the travel mode.

7. The method for identifying a transportation mode according to claim 1, characterized in that: The step S1 includes: S11 obtaining the original mobile phone signaling data of users within the required set time and area and performing original data processing; S12 collects the original processed mobile phone signaling raw data and performs desensitization processing; S13 matches the desensitized mobile phone signaling data with the travel trajectory map of the area; S14 extracts characteristic parameters of the user's travel based on the matching results of the mobile phone signaling data and the travel trajectory map, and completes the characterization processing of the mobile phone signaling data.

8. A method for identifying a transportation mode according to any one of claims 1 or 7, characterized in that: In the process of characterization of the mobile phone signaling data, after determining the data extraction range, the characteristic data is cleaned, the cleaned characteristic data is attribute decomposed, and finally the characteristic data of different scales are converted to a unified scale to complete normalization processing.

9. The method for identifying a transportation mode according to claim 6, characterized in that: When the urban double-layer model trains mobile phone signaling data, the input travel characteristics are used as the key to the model's supervised learning algorithm. After the urban double-layer model inputs the mobile phone signaling data at the spatiotemporal attribute level, it uses the maximum value, average value, minimum value and variance for identification, and uses similarity to characterize the travel mode, thereby completing the identification of the travel mode of the mobile phone signaling data within the set time and area.

10. A system for identifying a mode of transportation, adapted to the method for identifying a mode of transportation as claimed in any one of claims 1 to 9, characterized in that: include: The data acquisition module includes a plurality of base station data acquisition and transmission devices connected to the central processing module; The central processing module performs centralized processing on the mobile phone signaling data to generate processing results; The feature extraction module performs feature processing and obtains the feature results in the mobile phone signaling data; The identification module determines the travel mode of users in different areas and time periods through the analysis results of mobile phone signaling data.

Images