Individual space-time big data non-commuting activity identification method based on multi-source data fusion

Abstract

The invention relates to an individual space-time big data non-commuting activity identification method based on multi-source data fusion. The method comprises the following steps: respectively extracting first and second personal attribute features and single-day activity attribute features from individual space-time big data and resident travel survey data containing grid space information; extracting first and second built environment features based on individual space-time big data, resident travel survey data and city POI data containing grid space information; training and testing various to-be-evaluated models based on the preprocessed second personal attribute features, the single-day activity attribute features and the built environment features; selecting the to-be-evaluated model with the best performance as a final individual space-time big data non-commuting activity identification model; and based on the preprocessed first personal attribute feature, the single-day activity attribute feature and the built environment feature, obtaining an identification result by using an individual space-time big data non-commuting activity identification model. Compared with the prior art, the method has the advantages of improving the accuracy of secondary activity category inference and the like.

Description

Technical Field

[0001] The present invention relates to, in particular to, a method for identifying individual spatiotemporal big data non-commuting activities based on multi-source data fusion. Background Art

[0002] In today's rapidly developing urbanization, the growth of urban population and the diversification of activity patterns have brought new challenges to urban planning and traffic management. The daily activities of urban residents, especially non-commuting activities such as shopping and entertainment, are of great significance to understanding the functional layout of cities and the quality of life of residents. However, traditional resident travel survey methods are limited by sample size and data collection limitations, making it difficult to fully capture the complex activity patterns of urban residents.

[0003] With the rise of big data technology, individual spatiotemporal big data provides a new perspective for the analysis of urban activity patterns. These data have a wide range of spatiotemporal coverage and a representative population with a combination of multiple population attributes, which can more realistically reflect the activity dynamics of urban residents. However, how to accurately identify non-commuting activity categories from these massive and unlabeled data remains a technical challenge.

[0004] The existing technology has the following deficiencies when dealing with such problems:

[0005] 1) Focus on the identification of major activities such as commuting, while neglecting the analysis of non-commuting activities.

[0006] 2) In large-scale data processing, the accuracy and efficiency of existing methods need to be improved.

[0007] 3) Lack of comprehensive consideration of individual attributes, activity attributes and built environment characteristics. Summary of the invention

[0008] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method for identifying non-commuting activities of individuals based on multi-source data fusion based on individual spatiotemporal big data. By fusing multiple data sources, including but not limited to individual spatiotemporal trajectory data, resident travel survey data, urban points of interest (POI) data and urban spatial raster data, and using advanced data mining and machine learning algorithms, the non-commuting activities of individuals can be effectively identified and classified, thereby improving the efficiency and accuracy of identification.

[0009] The purpose of the present invention can be achieved by the following technical solutions:

[0010] The present invention provides a method for identifying non-commuting activities based on individual spatiotemporal big data and multi-source data fusion, comprising the following steps: S1, based on preset urban raster data, obtaining individual spatiotemporal big data, resident travel survey data and urban POI data containing raster space information; S2, based on the individual spatiotemporal big data, extracting a first personal attribute feature and a first single-day activity attribute feature; S3, based on the resident travel survey data, extracting a second personal attribute feature and a second single-day activity attribute feature; S4, based on the individual spatiotemporal big data and the urban POI data, extracting a first built environment feature, and based on the resident travel survey data and the urban POI data, extracting a second built environment feature; S5, merging the second personal attribute feature, the second single-day activity attribute feature and the second built environment feature. Preprocess the daily activity attribute characteristics and the second built environment characteristics; S6, based on the preprocessed second personal attribute characteristics, the second single-day activity attribute characteristics and the second built environment characteristics, train and test multiple non-commuting activity recognition models to be evaluated; S7, based on the test results, select the non-commuting activity recognition model to be evaluated with the best performance as the final individual spatiotemporal big data non-commuting activity recognition model; S8, preprocess the first personal attribute characteristics, the first single-day activity attribute characteristics and the first built environment characteristics; S9, based on the preprocessed first personal attribute characteristics, the first single-day activity attribute characteristics and the first built environment characteristics, use the individual spatiotemporal big data non-commuting activity recognition model to obtain the recognition result of non-commuting activities in the individual spatiotemporal big data.

[0011] As an optimal technical solution, S1 specifically includes: obtaining initial individual spatiotemporal big data, initial resident travel survey data and initial city POI data; spatially connecting the spatial information of user activity points in the initial individual spatiotemporal big data with the spatial information of the city raster data to obtain individual spatiotemporal big data containing raster spatial information; spatially connecting the spatial information of the activity points of the respondents in the initial resident travel survey data with the spatial information of the city raster data to obtain resident travel survey data containing raster spatial information; spatially connecting the spatial information of the city POI point in the initial city POI data with the spatial information of the city raster data to obtain city POI data containing raster spatial information.

[0012] As an optimal technical solution, S2 specifically includes: based on individual spatiotemporal big data containing raster spatial information, directly extracting the first personal attribute feature, and simultaneously obtaining the user ID, activity start date, activity end date, activity start time and activity end time, the first personal attribute feature including the user's personal age, gender, and occupation; sorting according to the user ID, the activity start date, the activity end date, the activity start time and the activity end time to obtain the single-day activity sequence information of each user; based on the single-day activity sequence information, calculating the first single-day activity attribute feature, the first single-day activity attribute feature including the start time of this activity, the duration of this activity, whether the last activity was at home, whether the last activity was at work, the straight-line distance from home, the straight-line distance between home and workplace, whether the activity start date is a working day, and the order of this activity in the resident's day.

[0013] As an optimal technical solution, S3 specifically includes: based on individual spatiotemporal big data containing raster spatial information, directly extracting the second personal attribute feature, and simultaneously obtaining the respondent ID, activity start date, activity end date, activity start time and activity end time, the second personal attribute feature includes the respondent's personal age, gender, and occupation; sorting according to the respondent ID, the activity start date, the activity end date, the activity start time and the activity end time to obtain the single-day activity sequence information of each respondent; based on the single-day activity sequence information, calculating the second single-day activity attribute feature, the second single-day activity attribute feature includes the start time of this activity, the duration of this activity, whether the last activity was at home, whether the last activity was at work, the straight-line distance from home, the straight-line distance between home and workplace, whether the activity start date is a working day, and the order of this activity in the resident's day.

[0014] As an optimal technical solution, S4 specifically includes: based on the urban POI data containing raster spatial information, calculating the POI density of the grid where each POI type is located; based on the individual spatiotemporal big data containing raster spatial information and the resident travel survey data containing raster spatial information, directly extracting the grids where the activity place, residence and workplace are located; associating the POI density with the corresponding grids where the activity place, residence and workplace are located to obtain the first built environment feature and the second built environment feature respectively.

[0015] As an optimal technical solution, S8 specifically includes: merging the first personal attribute characteristics, the first single-day activity attribute characteristics and the first built environment characteristics; normalizing the first personal attribute characteristics, the first single-day activity attribute characteristics and the first built environment characteristics, and using the normalized data as the first feature set; obtaining the travel activity category field, and using the travel activity category field as the field to be inferred, merging it into the first feature set to obtain a prediction set.

[0016] As a preferred technical solution, S5 specifically includes: merging the second personal attribute characteristics, the second single-day activity attribute characteristics and the second built environment characteristics; normalizing the second personal attribute characteristics, the second single-day activity attribute characteristics and the second built environment characteristics, and using the normalized data as the second feature set; directly extracting non-commuting activities from the resident travel survey data as a second label set, and encoding them; merging the second feature set and the second label set as a data set for the multiple non-commuting activity recognition models to be evaluated, and dividing the data set into a training set and a test set.

[0017] As a preferred technical solution, the expression of the normalization process is:

[0018]

[0019] In the formula, represents the i-th feature after normalization, v i represents the i-th feature data, represents the maximum value of the i-th feature data set, Represents the minimum value of the i-th feature data set.

[0020] As a preferred technical solution, S6 specifically includes: inputting the training set into a plurality of pre-acquired non-commuting activity recognition models to be evaluated for training, wherein the plurality of non-commuting activity recognition models to be evaluated include an SVM model, a DT model, an RF model, a GBDT model, an ADAboost model, an XGBoost model and a Transformer-based classification model.

[0021] As a preferred technical solution, the best performance judgment indicators include accuracy and F1 score, specifically including:

[0022]

[0023] In the formula, label P is the predicted value of the label set, label T is the true value of the label set, S indicates whether the true value is correctly predicted, T represents the test set, and N Trepresents the number of samples in the test set, Accuracy represents the accuracy, P represents the precision, R represents the recall, TP represents the number of samples whose prediction results are correctly predicted in one of the labels, FP represents the number of samples whose prediction results are incorrectly predicted in one of the labels, FN represents the number of samples whose prediction results are incorrectly predicted in none of the labels, and F1score represents the f1 score.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. The present invention makes comprehensive use of multi-source data, extracts multiple features including personal attribute features, single-day activity attribute features and built environment features, combines data fusion and preprocessing based on raster spatial information, incorporates the preprocessed features into multiple mathematical models for non-commuting activity identification, and processes data more comprehensively and finely, thereby improving the accuracy and comprehensiveness of non-commuting activity identification and providing deeper insights for urban planning and traffic management;

[0026] 2. The present invention adopts advanced mathematical models and algorithms, and utilizes the pre-processed second personal attribute characteristics, the second single-day activity attribute characteristics and the second built environment characteristics to train and test a variety of non-commuting activity recognition models to be evaluated, and then selects the model with the best performance as the final non-commuting activity recognition model, which can not only ensure the optimal non-commuting activity recognition effect in different scenarios, but also improve the efficiency and intelligence level of data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A basic flow chart of a method provided in an embodiment of the present invention;

[0028] Figure 2 A diagram showing training results of various non-commuting activity recognition models to be evaluated provided in an embodiment of the present invention;

[0029] Figure 3 A heat map of the number of shopping and entertainment activities of individual spatiotemporal big data users in 2023 (working days) provided in an embodiment of the present invention;

[0030] Figure 4 This is a heat map (working day) of the frequency distribution of other activities of individual spatiotemporal big data users in 2023 provided in an embodiment of the present invention;

[0031] Figure 5 A heat map (weekend) of the frequency distribution of shopping and entertainment activities of individual spatiotemporal big data users in 2023 provided in an embodiment of the present invention;

[0032] Figure 6 This is a heat map (weekends) of the distribution of other activity frequencies of individual spatiotemporal big data users in 2023 provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0034] Example

[0035] like Figure 1 As shown, this embodiment provides a method for identifying non-commuting activities of individuals based on multi-source data fusion using spatiotemporal big data, which can more accurately identify and understand the non-commuting activities of urban residents, provide scientific decision-making support for urban development, and promote the optimization of urban transportation systems and the improvement of residents' quality of life. The specific steps include:

[0036] Step S1, obtain the city 500m×500m grid data, initial individual spatiotemporal big data with job and residence labels, initial resident travel survey data, and initial city POI data, and map the initial individual spatiotemporal big data, initial resident travel survey data, and initial city POI data to the city 500m×500m grid data through spatial information, and obtain individual spatiotemporal big data, resident travel survey data, and city POI data containing grid spatial information, and the city POI data contains city POI data of different years. Specifically:

[0037] ArcGIS software was used to spatially connect the spatial information of user activity points in the initial individual spatiotemporal big data with the spatial information of the urban 500m×500m grid data, and to spatially connect the spatial information of the activity points of the respondents in the initial resident travel survey data with the spatial information of the urban 500m×500m grid data, so as to achieve a one-to-one correspondence between the activity points of urban residents and the grids where they are located;

[0038] Using ArcGIS software, the spatial information of urban POI points in the initial urban POI data is spatially connected with the spatial information of the urban 500m×500m grid data to achieve a one-to-one correspondence between urban POI points and the grids where they are located.

[0039] Step S2, based on the individual spatiotemporal big data containing grid space information, extract the user's personal information and single-day activity sequence information, and then extract the first personal attribute feature and the first single-day activity attribute feature. Specifically:

[0040] In the individual spatiotemporal big data containing raster spatial information, the user's personal information is directly given and can be directly extracted. The first personal attribute feature extracted includes the user's personal age, gender and occupation, and the user ID, activity start date, activity end date, activity start time and activity end time are obtained at the same time;

[0041] For the single-day activity attribute features, first sort according to the obtained user ID, activity start date, activity end date, activity start time (24-hour system), and activity end time (24-hour system) to obtain the single-day activity sequence information of each user, and then calculate the single-day activity attribute features based on the sequence, including the start time of this activity, the duration of this activity, whether the last activity was at home, whether the last activity was at work, the straight-line distance from home, the straight-line distance between home and workplace, whether the activity start date is a weekday, and the order of this activity in the resident's day.

[0042] Step S3, based on the resident travel survey data containing grid space information, extract the personal information and single-day activity sequence information of the respondents, and extract the second personal attribute feature and the second single-day activity attribute feature. Specifically:

[0043] In the resident travel survey data containing raster spatial information, the personal information of the respondents is directly given and can be directly extracted. The extracted second personal attribute features include the respondent's personal age, gender and occupation, and the respondent's ID, activity start date, activity end date, activity start time and activity end time are obtained at the same time;

[0044] For the single-day activity attribute characteristics, first sort according to the obtained respondent ID, activity start date, activity end date, activity start time (24-hour system), and activity end time (24-hour system) to obtain the single-day activity sequence information of each user, and then calculate the single-day activity attribute characteristics based on the sequence, including the start time of this activity, the duration of this activity, whether the last activity was at home, whether the last activity was at work, the straight-line distance from home, the straight-line distance between home and workplace, whether the activity start date is a weekday, and the order of this activity in the resident's day.

[0045] Step S4, based on the individual spatiotemporal big data containing raster spatial information and the urban POI data containing raster spatial information in the same year, extract the first built environment feature, and based on the resident travel survey data containing raster spatial information and the urban POI data containing raster spatial information in the same year, extract the second built environment feature. Specifically:

[0046] In the city POI data containing grid spatial information, for each POI type, the POI density of the grid where it is located is calculated using the following formula:

[0047]

[0048] in, represents the POI density of type i in the kth grid, represents the number of POIs of type i in the kth grid, Represents the area of ​​the kth grid (km 2 );

[0049] In the individual spatiotemporal big data containing raster spatial information and the resident travel survey data containing raster spatial information, the grids of activity places, residence places and work places are directly obtained respectively.

[0050] The density information of various POIs in the grid is associated with the corresponding grids of activity places, residences, and workplaces to obtain the first built environment characteristics and the second built environment characteristics. The built environment characteristics include the density of regular bus stops within 500 meters and 800 meters of the center point of the grid where the activity place, home and workplace are located, and the density of POIs of shopping, life services, catering, companies, governments, commercial housing, science, education and culture, health services, landmarks, financial institutions, and tourist attractions in the grids where the activity place, home and workplace are located.

[0051] Step S5, preprocessing the second personal attribute feature, the second single-day activity attribute feature and the second built environment feature. Specifically:

[0052] The second personal attribute feature, the second single-day activity attribute feature, and the second built environment feature are combined, that is, each activity of urban residents contains the above features;

[0053] All the above features are normalized, that is, the original data is converted to the range of [0,1]. The calculation formula is:

[0054]

[0055] in, represents the i-th feature after normalization, v i represents the i-th feature data, represents the maximum value of the i-th feature data set, Represents the minimum value of the i-th feature data set;

[0056] The normalized data is used as the second feature set X model ;

[0057] Directly extract non-commuting activities (shopping and entertainment, other categories) from the resident travel survey data as the second label set Y model , and coded, the secondary activity category of shopping and entertainment is coded as 1, and the secondary activity category of other categories is coded as 0;

[0058] Merge the second feature set X model With the second label set Y model, as a dataset for various non-commuting activity recognition models to be evaluated, and divided into model training set and test set in a ratio of 8:2.

[0059] Step S6, based on the preprocessed second personal attribute feature, the second single-day activity attribute feature and the second built environment feature, obtain a data set divided into a training set and a test set, and use the data set to train and test multiple non-commuting activity recognition models to be evaluated, so as to model the non-commuting activity recognition of individual spatiotemporal big data.

[0060] Specifically: the training set obtained by preprocessing is input into a variety of non-commuting activity recognition models to be evaluated for training. The various non-commuting activity recognition models to be evaluated include SVM model, DT model, RF model, GBDT model, ADAboost model, XGBoost model and Transformer-based classification model.

[0061] Step S7: Based on the test results, the performance of various mathematical models is output, and the non-commuting activity recognition model with the best performance to be evaluated is selected as the final individual spatiotemporal big data non-commuting activity recognition model. Specifically:

[0062] The overall accuracy of the indicators is used to describe the performance of multiple mathematical models, and the model with the best indicators is extracted as the subsequent individual spatiotemporal big data non-commuting activity identification model;

[0063] The indicators are: Accuracy and F1score, including,

[0064]

[0065] Among them, label P The second label set Y model The predicted value of label T The second label set Y model The true value, S indicates whether the true value is correctly predicted, T indicates the test set, N T represents the number of test set samples, and Accuracy represents the accuracy;

[0066]

[0067] Among them, P is the precision rate, R is the recall rate, TP is the number of samples whose prediction results are correctly predicted in a certain class of labels, FP is the number of samples whose prediction results are incorrectly predicted in a certain class of labels, FN is the number of samples whose prediction results are incorrectly predicted for non-certain class of labels, and F1score represents the F1 score.

[0068] Step S8, preprocessing the first personal attribute feature, the first single-day activity attribute feature and the first built environment feature. Specifically:

[0069] The first personal attribute feature, the first single-day activity attribute feature, and the first built environment feature are merged, that is, each non-commuting activity of each person in the individual spatiotemporal big data contains the above features;

[0070] All the above features are normalized, and the processing method is consistent with the method in step S5;

[0071] The normalized data is used as the first feature set X Pre ;

[0072] Get the travel activity category field, and use the travel activity category field as the field to be inferred, and merge it into the first feature set X Pre In the above example, a data prediction set is formed.

[0073] In step S9, a data prediction set is formed based on the preprocessed first personal attribute feature, the first single-day activity attribute feature and the first built environment feature. The prediction set data is input into the individual spatiotemporal big data non-commuting activity recognition model obtained in step S7 to obtain the recognition results of non-commuting activities in the individual spatiotemporal big data, including shopping and entertainment categories and other categories.

[0074] The individual spatiotemporal big data non-commuting activity identification method based on multi-source data fusion provided in this embodiment can more comprehensively capture personal attribute characteristics, single-day activity attribute characteristics and built environment characteristics, while combining more advanced mathematical models to improve the accuracy of secondary activity category inference, providing support for urban planning and traffic management.

[0075] Next, the effectiveness of the method proposed in this embodiment is verified:

[0076] The research data of this embodiment is 500m×500m raster data of a certain city, individual spatiotemporal big data with job and residence labels in 2023 of the city (data provider platform: Mengxiang and Ke data platform), 2019 resident travel survey data of the city, and 2019 and 2023 urban POI data of the city, among which the 2019 urban POI data and the 2019 resident travel survey data are used together for model training, and the 2023 urban POI data and the individual spatiotemporal big data with job and residence labels in 2023 are used together for non-commuting activity identification.

[0077] Firstly, the residents’ travel survey data, urban POI data, and individual spatiotemporal big data are mapped onto a 500m×500m urban raster layer to unify the spatial information.

[0078] Based on the resident travel survey data and urban POI data, the relevant influencing features required for the modeling of individual spatiotemporal big data non-commuting activity identification are extracted. Based on individual spatiotemporal big data and urban POI data, the relevant influencing features for individual spatiotemporal big data non-commuting activity identification are extracted. It should be noted that the types of features included in modeling and inference are consistent, including personal attribute features, single-day activity attribute features and built environment features. The detailed features are shown in Table 1.

[0079] Table 1. Feature set of non-commuting activity recognition model for individual spatiotemporal big data

[0080]

[0081]

[0082]

[0083] After the acquired feature values ​​are normalized, the feature data obtained from the resident survey data and the city POI data are used for model construction and need to be divided into a training set and a test set in a ratio of 8:2. The feature data obtained from the individual spatiotemporal big data and the city POI data are used as prediction data, which is the prediction set.

[0084] The training set is input into a variety of machine learning models, including SVM, DT, RF, GBDT, ADAboost, XGBoost, and Transformer-based classification models, in order to establish an inference model for the secondary activity categories of individual spatiotemporal big data. After the model training is completed, the performance of each model is evaluated by Accuracy and F1score to select the best performing model.

[0085] The performance evaluation results are shown in Figure 2 It can be seen that the XGBoost model outperforms other models in various evaluation indicators, including SVM, DT, RF, GBDT, ADAboost and Transformer-based classification models. Specifically, the overall recognition accuracy of the XGBoost model reached 86%, the F1 score for shopping and entertainment secondary activity category inference was 0.89, and the F1 score for other secondary activity category inference was 0.83. These results show that the XGBoost model can more accurately identify shopping and entertainment and other types of secondary activity categories. Therefore, this embodiment finally selects the XGBoost model as the model with the best performance for further individual spatiotemporal big data secondary activity category inference.

[0086] The personal attribute characteristics, daily activity attribute characteristics and built environment characteristics obtained from individual spatiotemporal big data containing raster spatial information and urban POI data containing raster spatial information were preprocessed and normalized and then imported into the optimal performance model to identify non-commuting activities in individual spatiotemporal big data. Then, a heat map of non-commuting activities in individual spatiotemporal big data was drawn based on the identified non-commuting activity information (see Figure 3 to Figure 6 ). From the heat map results, it can be observed that, whether on weekdays or weekends, shopping and entertainment activities are mainly concentrated in the central areas of Shanghai's main urban area and the other five new cities, while other types of activities show a wider distribution in comparison. This inference result is consistent with real observations, confirming that the proposed non-commuting activity recognition technology can effectively serve the non-commuting activity recognition in individual spatiotemporal big data.

[0087] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for identifying non-commuting activities based on individual spatiotemporal big data based on multi-source data fusion, characterized in that: The following steps are involved: S1, based on the preset urban grid data, obtain individual spatiotemporal big data containing grid space information, resident travel survey data and urban POI data; S2, extracting the first personal attribute feature and the first single-day activity attribute feature based on the individual spatiotemporal big data; S3, extracting a second personal attribute feature and a second single-day activity attribute feature based on the resident travel survey data; S4, extracting a first built environment feature based on the individual spatiotemporal big data and the city POI data, and extracting a second built environment feature based on the resident travel survey data and the city POI data; S5, preprocessing the second personal attribute feature, the second single-day activity attribute feature, and the second built environment feature; S6, based on the pre-processed second personal attribute features, the second single-day activity attribute features and the second built environment features, training and testing a variety of non-commuting activity recognition models to be evaluated; S7, based on the test results, the non-commuting activity recognition model with the best performance is selected as the final individual spatiotemporal big data non-commuting activity recognition model; S8, preprocessing the first personal attribute feature, the first single-day activity attribute feature, and the first built environment feature; S9, based on the preprocessed first personal attribute feature, the first single-day activity attribute feature and the first built environment feature, use the individual spatiotemporal big data non-commuting activity recognition model to obtain recognition results of non-commuting activities in the individual spatiotemporal big data.

2. The method for identifying non-commuting activities of individuals based on multi-source data fusion using spatiotemporal big data according to claim 1 is characterized in that: The S1 specifically includes: Obtain initial individual spatiotemporal big data, initial resident travel survey data and initial city POI data; Performing spatial connection between the spatial information of user activity points in the initial individual spatiotemporal big data and the spatial information of the city grid data to obtain individual spatiotemporal big data containing grid spatial information; Performing spatial connection between the spatial information of the activity points of the respondents in the initial resident travel survey data and the spatial information of the urban grid data to obtain the resident travel survey data containing the grid spatial information; The city POI point spatial information in the initial city POI data is spatially connected with the city grid data spatial information to obtain the city POI data containing grid spatial information.

3. The method for identifying non-commuting activities based on individual spatiotemporal big data based on multi-source data fusion according to claim 1 is characterized in that: The S2 specifically includes: Based on the individual spatiotemporal big data containing raster spatial information, directly extract the first personal attribute feature, and simultaneously obtain the user ID, activity start date, activity end date, activity start time, and activity end time, wherein the first personal attribute feature includes the user's personal age, gender, and occupation; Sort by the user ID, the activity start date, the activity end date, the activity start time, and the activity end time to obtain the single-day activity sequence information of each user; Based on the single-day activity sequence information, the first single-day activity attribute feature is calculated, and the first single-day activity attribute feature includes the start time of this activity, the duration of this activity, whether the last activity was at home, whether the last activity was at work, the straight-line distance from home, the straight-line distance between home and workplace, whether the activity start date is a weekday, and the order of this activity in the resident's day.

4. The method for identifying non-commuting activities of individuals based on multi-source data fusion using spatiotemporal big data according to claim 1 is characterized in that: The S3 specifically includes: Based on the individual spatiotemporal big data containing raster spatial information, the second personal attribute feature is directly extracted, and the respondent ID, activity start date, activity end date, activity start time and activity end time are obtained at the same time. The second personal attribute feature includes the respondent's personal age, gender, and occupation; Sort by the respondent ID, the activity start date, the activity end date, the activity start time, and the activity end time to obtain the single-day activity sequence information of each respondent; Based on the single-day activity sequence information, a second single-day activity attribute feature is calculated, and the second single-day activity attribute feature includes the start time of this activity, the duration of this activity, whether the last activity was at home, whether the last activity was at work, the straight-line distance from home, the straight-line distance between home and workplace, whether the activity start date is a weekday, and the order of this activity in the resident's day.

5. The method for identifying non-commuting activities based on individual spatiotemporal big data based on multi-source data fusion according to claim 1 is characterized in that: The S4 specifically includes: Based on the city POI data containing grid spatial information, the POI density of each POI type grid is calculated; Based on individual spatiotemporal big data containing raster spatial information and resident travel survey data containing raster spatial information, the grids of activity places, residence places and work places are directly extracted respectively; The POI density is associated with the grids where the corresponding activity sites, residences and workplaces are located, and a first built environment feature and a second built environment feature are obtained respectively.

6. The method for identifying non-commuting activities based on individual spatiotemporal big data based on multi-source data fusion according to claim 1 is characterized in that: The S8 specifically includes: Combining the first personal attribute feature, the first single-day activity attribute feature, and the first built environment feature; Normalizing the first personal attribute feature, the first single-day activity attribute feature, and the first built environment feature, and using the normalized data as a first feature set; A travel activity category field is obtained, and the travel activity category field is used as a field to be inferred and merged into the first feature set to obtain a prediction set.

7. The method for identifying non-commuting activities based on individual spatiotemporal big data based on multi-source data fusion according to claim 1 is characterized in that: The S5 specifically includes: Combining the second personal attribute feature, the second single-day activity attribute feature and the second built environment feature; Normalizing the second personal attribute feature, the second single-day activity attribute feature, and the second built environment feature, and using the normalized data as a second feature set; Directly extracting non-commuting activities from the resident travel survey data as a second label set and encoding them; The second feature set and the second label set are combined as a data set of the multiple non-commuting activity recognition models to be evaluated, and the data set is divided into a training set and a test set.

8. The method for identifying non-commuting activities of individuals based on multi-source data fusion using spatiotemporal big data according to any one of claims 6-7, characterized in that: The expression of the normalization process is: In the formula, represents the i-th feature after normalization, v i represents the i-th feature data, represents the maximum value of the i-th feature data set, Represents the minimum value of the i-th feature data set.

9. The method for identifying non-commuting activities based on individual spatiotemporal big data based on multi-source data fusion according to claim 7 is characterized in that: The S6 specifically includes: The training set is input into a plurality of pre-acquired non-commuting activity recognition models to be evaluated for training, wherein the plurality of non-commuting activity recognition models to be evaluated include an SVM model, a DT model, an RF model, a GBDT model, an ADAboost model, an XGBoost model, and a Transformer-based classification model.

10. The method for identifying non-commuting activities based on individual spatiotemporal big data based on multi-source data fusion according to claim 7, characterized in that: The best performance judgment indicators include accuracy and F1 score, specifically including: In the formula, label P is the predicted value of the label set, label T is the true value of the label set, S indicates whether the true value is correctly predicted, T represents the test set, and N T represents the number of samples in the test set, Accuracy represents the accuracy, P represents the precision, R represents the recall, TP represents the number of samples whose prediction results are correctly predicted in one of the labels, FP represents the number of samples whose prediction results are incorrectly predicted in one of the labels, FN represents the number of samples whose prediction results are incorrectly predicted in none of the labels, and F1score represents the F1 score.

Images