Method and system for analyzing and predicting vehicle stopping behavior based on multitask learning

Abstract

The present invention provides a method and system for analyzing and predicting vehicle stopping behavior based on multi-task learning. [Solution] The method includes the steps of collecting vehicle GPS and OBD data, including the vehicle ID after the vehicle is anonymized, the process start time, the start longitude, the start latitude, the end time, the end longitude and the end latitude, preprocessing the vehicle GPS and OBD data to obtain vehicle stopping behavior data, including stopping points and stopping times, and performing spatio-temporal feature extraction for the preprocessed vehicle stopping behavior by a deep recurrent neural network, inputting the spatio-temporal features into a multi-task learning prediction network, and predicting the stopping points and stopping times by obtaining the correlation between the stopping point prediction task and the stopping time prediction task by the multi-task learning prediction network based on the vehicle's historical stopping behavior. The method of the present invention can provide support for application scenarios such as recommendation of points of interest, customized creation of travel, directional broadcasting of advertisements, and smart parking.

Description

[Technical Field]

[0001] The present invention mainly relates to the fields of trajectory data mining and intelligent transportation optimization, and more particularly to a method and system for analyzing and predicting vehicle stopping behavior based on multi-task learning. [Background technology]

[0002] Vehicle stopping behavior analysis means that when vehicle stopping behavior data is obtained, the relevant data is statistically analyzed to discover the rules of user travel intentions and movement attributes, and these rules are combined with policies such as optimizing urban road network traffic operation, which plays an important role in recommending points of interest, preventing traffic accidents, recommending customized routes, planning urban streets, and improving traffic efficiency.

[0003] In recent years, analyzing and mining deep features of transportation big data for modeling and prediction has gradually become a hot topic in the field of urban computing, attracting the attention of many experts and companies both in China and abroad. Previous work has mainly focused on mining the internal spatiotemporal correlations of a single prediction task, such as traffic flow, speed, and transit time, and then introducing deep learning methods to model and predict traffic data. These approaches ignore the spatiotemporal correlations and task dependencies between different prediction tasks, and only predict a single task, missing the inter-task dependencies between tasks, such as vehicle stopping locations and stopping times. Capturing these inter-task dependencies to better model traffic data would significantly improve prediction accuracy. Therefore, it is crucial to incorporate multi-task learning into vehicle stopping behavior analysis and prediction systems. Summary of the Invention [Problem to be solved by the invention]

[0004] SUMMARY OF THE INVENTION In view of the shortcomings of the prior art, the present invention aims to provide a method and system for analyzing and predicting vehicle stopping behavior based on multi-task learning. [Means for solving the problem]

[0005] To achieve the above technical objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect of an embodiment of the present invention, there is provided a method for analyzing and predicting vehicle stopping behavior based on multi-task learning, the method comprising: Step S1 of collecting vehicle GPS and OBD data including vehicle ID, process start time, start longitude, start latitude, end time, end longitude and end latitude after the vehicle has been anonymized; Preprocessing the vehicle GPS and OBD data to obtain vehicle stopping behavior data including stopping points and stopping times, and performing spatiotemporal feature extraction on the preprocessed vehicle stopping behavior using a deep recurrent neural network; Here, extracting spatiotemporal features from preprocessed vehicle stopping behavior using a deep recurrent neural network is Dividing the vehicle stopping behavior data into a plurality of hierarchical levels of vehicle stopping behavior data at time intervals; Processing vehicle stopping behavior data at a certain layer by a first-layer recurrent neural network; Processing the vehicle stopping behavior data between layers by a second-layer recurrent neural network; Step S2 includes converting the vehicle stopping behavior data between the first-layer recurrent neural network and the second-layer recurrent neural network to realize encoding of the vehicle stopping behavior data and obtain spatiotemporal features; The method includes a step S3 of inputting the spatiotemporal features into a multi-task learning prediction network, and predicting the stopping points and stopping times by obtaining correlations between the stopping point prediction task and the stopping time prediction task by the multi-task learning prediction network based on the vehicle's historical stopping behavior.

[0007] In a second aspect of an embodiment of the present invention, there is provided a vehicle stopping behavior analysis and prediction system based on multi-task learning, the system comprising: a data collection module, a vehicle stopping behavior data acquisition module, and a multi-task learning prediction module; the data collection module collects vehicle GPS and OBD data, including a vehicle ID, a process start time, a start longitude, a start latitude, an end time, an end longitude, and an end latitude after the vehicle has been anonymized; The vehicle stopping behavior data acquisition module preprocesses the vehicle GPS and OBD data to obtain vehicle stopping behavior data including stopping points and stopping times, and performs spatiotemporal feature extraction on the preprocessed vehicle stopping behavior using a deep recurrent neural network; Here, extracting spatiotemporal features from preprocessed vehicle stopping behavior using a deep recurrent neural network is Dividing the vehicle stopping behavior data into a plurality of hierarchical levels of vehicle stopping behavior data at time intervals; Processing vehicle stopping behavior data at a certain layer by a first-layer recurrent neural network; Processing the vehicle stopping behavior data between layers by a second-layer recurrent neural network; Transforming the vehicle stopping behavior data between the first-layer recurrent neural network and the second-layer recurrent neural network to realize encoding of the vehicle stopping behavior data and obtain spatiotemporal features; The multi-task learning prediction module inputs the spatiotemporal features into a multi-task learning prediction network, and predicts the stopping points and stopping times by obtaining correlations between the stopping point prediction task and the stopping time prediction task through the multi-task learning prediction network based on the vehicle's historical stopping behavior.

[0008] In a third aspect of the present invention, there is provided an electronic device comprising a memory and a processor, the memory being coupled to the processor, the memory being used to store program data, and the processor being used to implement the above-mentioned method for analyzing and predicting vehicle stopping behavior based on multi-task learning by executing the program data.

[0009] In a fourth aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the above-described method for analyzing and predicting vehicle stopping behavior based on multi-task learning. [Effects of the Invention]

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides a method for analyzing and predicting vehicle stopping behavior based on multi-task learning, which takes into account the vehicle's historical stopping behavior and also considers the correlation between the stopping point prediction task and the stopping time prediction task; the present invention inputs spatiotemporal features into a multi-task learning prediction network, and uses the multi-task learning prediction network to obtain the correlation between the stopping point prediction task and the stopping time prediction task based on the vehicle's historical stopping behavior, thereby predicting the stopping points and stopping times, thereby analyzing the spatiotemporal dependencies of vehicle stopping behavior and improving the accuracy of stopping behavior prediction; the present invention provides support for applications such as point of interest recommendation, customized travel creation, directional advertising broadcasting, and smart parking. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a flowchart of a method for analyzing and predicting vehicle stopping behavior based on multi-task learning according to an embodiment of the present invention. [Figure 2] FIG. 2 is a flowchart of the step of extracting the stopping time according to an embodiment of the present invention. [Figure 3] FIG. 3 is a flowchart of the steps of extracting stopping points according to an embodiment of the present invention. [Figure 4] FIG. 4 is a schematic diagram of a deep recurrent neural network according to an embodiment of the present invention. [Figure 5] FIG. 5 is a schematic diagram of a vehicle stopping behavior analysis and prediction system based on multi-task learning according to an embodiment of the present invention. [Figure 6] FIG. 6 is a comparison diagram of the accuracy of the multi-task prediction network and the LSTM network according to an embodiment of the present invention. [Figure 7] FIG. 7 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0012] Illustrative embodiments will now be described in detail, examples of which are illustrated in the drawings. When the following description refers to the drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The embodiments described in the following illustrative examples do not represent all embodiments consistent with the present invention.

[0013] Hereinafter, specific embodiments of the present invention will be described in more detail with reference to the drawings. Unless inconsistent, the features of the following examples and embodiments can be combined with each other.

[0014] As shown in FIG. 1, an embodiment of the present invention further provides a vehicle stopping behavior analysis and prediction method based on multi-task learning, which includes the following steps S1 to S3.

[0015] In step S1, vehicle GPS (Global Positioning System) and OBD (On-Board Diagnostics) data are collected, and the vehicle GPS and OBD data include information such as the vehicle ID after the vehicle is anonymized, the process start time, start longitude, start latitude, end time, end longitude and end latitude.

[0016] In step S2, the vehicle GPS and OBD data collected in step S1 are preprocessed to obtain vehicle stopping behavior data including stopping locations and stopping times, and spatiotemporal feature extraction is performed on the preprocessed vehicle stopping behavior using a deep recurrent neural network.

[0017] Furthermore, the step S2 specifically includes the following steps S201 and S202.

[0018] In step S201, the vehicle GPS and OBD data collected in step S1 are pre-processed to obtain vehicle stopping behavior data including stopping points and stopping times.

[0019] As shown in FIG. 2, the step of acquiring the stopping time specifically includes the following steps S2011 to S2013.

[0020] In step S2011, the vehicle GPS and OBD data are sorted according to starting time.

[0021] In step S2012, the vehicle GPS and OBD data whose end time in the previous section is not earlier than the start time in the next section is deleted, and the vehicle GPS and OBD data whose stop time is not later than the start time is deleted.

[0022] In step S2013, based on the sorted vehicle GPS and OBD data, subtract the start time of the latter item from the end time of the former item to obtain the stopping time. It is also necessary to remove the vehicle GPS and OBD data whose start and stopping points do not match, and delete the vehicle GPS and OBD data whose stopping time is less than the stopping time threshold. In this example, the stopping time threshold is set as 5 minutes.

[0023] As shown in FIG. 3, the extraction step of acquiring the stop points specifically includes the following steps S2014 and S2015.

[0024] In step S2014, the vehicle GPS data and OBD data are aligned and merged by timestamp based on the user ID, and sorted according to the user ID to obtain the vehicle trajectory data T.

[0025] In step S2015, it is determined whether the interval between any one trajectory point and its adjacent trajectory point in the vehicle trajectory data is smaller than an interval threshold, and in this example, the interval threshold is set as 50 m.

[0026] If it is smaller than the interval threshold, calculate the time difference between the two trajectory points and set the time difference as TIFF0007657965000002.tif45, and if the time difference is greater than the time difference threshold (in this example, the time difference threshold is set as 150s) and the speed of the trajectory point is 0, label the trajectory point as a stopped state at the stopping start time, and label the vehicle start until the stopping end time, and label the i-th stopping point as TIFF0007657965000003.tif859, where id is the user id, lon is the longitude of the stop, lat is the latitude of the stop, and t s is the start of stopping, and t e is the end of the stop.

[0027] Obtain the stop point set of each user, and denote the stop point set of the jth user as S j ={s1,s2,…,s n}, where j=1, ..., n.

[0028] Obtain the set of stops for all users and define S={S1,S2,...,S n} is written as

[0029] In step S202, spatiotemporal feature extraction is performed on the preprocessed vehicle stopping behavior using a deep recurrent neural network.

[0030] The vehicle user's stopping behavior data is S={s1, s2, ..., s n}, and for each stop point s i When we specify it as follows, we get a tuple consisting of latitude and longitude and time, namely s i =(id,el,et,y,m,h,min,s,d), which respectively represent the user ID, stopping longitude, stopping latitude, year, month, day, hour, minute, second and stopping time. As shown in Figure 4, the deep recurrent neural network can extract the location and time information using one embedding layer, and propagate this information through the first-layer recurrent neural network to obtain the hidden state of the stopping behavior. Then, in the second-layer recurrent neural network, its input is from the output of the first-layer recurrent neural network, and by converting the vehicle stopping behavior data between the first-layer recurrent neural network and the second-layer recurrent neural network, it can better represent the user's stopping behavior, which is specifically: This involves dividing user stopping behavior into different hierarchical trajectories with time intervals ranging from days to weeks, and transform-coding between a first-layer recurrent neural network and a second-layer recurrent neural network.

[0031] The process of processing vehicle stopping behavior data at a certain layer by the first-layer recurrent neural network is specifically shown in the following equation 1:

number

[0032] Specifically, the process of processing vehicle stopping behavior data between layers using a second-layer recurrent neural network is as follows: If the next user movement record to be processed is large, the state of the lower layer needs to be input into the higher layer recurrent neural network to represent the user movement record together. The specific details of the formula are shown in the following Equation 2:

number

[0033] The stopping behaviors of different layers are coded by recurrent neural networks of different layers, and the movement features of the stopping behaviors of the first layer are extracted and processed via the second layer recurrent neural network, so that the stopping behavior features of the user can be retained to the maximum extent.

[0034] The user's stopping behavior is divided into different layers, namely the first layer and the second layer. The user's stopping behavior record is processed holistically by encoding different stopping behaviors, thereby simplifying the user's complex stopping behavior mode. el,et are the longitude and latitude coordinates of the user's stopping behavior point, and the embedding of the first layer of the i-th layer is shown in Equation 1 below:

number

[0035] The second layer embedding inputs the first layer state representing the user's parking behavior within a given time into a high-level recurrent neural network and integrates it. The user's parking behavior is shown in Equation 2 below.

number

[0036] Now, after understanding how a recurrent neural network with a hierarchical structure processes user parking behavior, it is intuitively understood that user parking behavior is a segmented record. The first-layer recurrent neural network processes one segment of parking behavior record, but there is generally a large time interval between different parking behavior records. The second-layer recurrent neural network processes the situations between different parking behaviors. By converting user parking behavior with recurrent neural networks between two different layers, longer user parking behavior records can be processed. This hierarchical structure method can better preserve information about user parking behavior at different layers, thereby better modeling user parking behavior.

[0037] In step S3, the spatiotemporal features are input into a multi-task learning prediction network, and the multi-task learning prediction network obtains the correlation between the stopping point prediction task and the stopping time prediction task based on the vehicle's historical stopping behavior, thereby predicting the stopping point and stopping time.

[0038] The spatiotemporal features are input into a multi-task learning prediction network, and a stop location task model and a stop time task model are respectively constructed by a long short-term memory network based on the vehicle's historical stopping behavior, and the stop location task model and the stop time task model are simultaneously learned by the multi-task learning prediction network, and the correlation between the stop location prediction task and the stop time prediction task is mined to predict and obtain the stop locations and stop times within a future time period; In this example, both the stop point task model and the stop time task model are realized by long short-term memory networks. The difference between the stop point task model and the stop time task model is that predicting stop time is a regression problem, while predicting stop points is a classification problem.

[0039] When predicting the stopping time task, the present invention sorts the stopping behavior data in time to obtain the sequence data of the vehicle stopping behavior. Then, the next stopping time and the location category are used as labels, so as to train and obtain the information of the next stopping time and the location category. In addition, in the task learning process by the long short-term memory network, the parameter change process is shown in the following formula 3:

number

[0040] In the parameter learning process, the stopping time task model not only uses the input x of the previous layer, but also uses the input h of the previous time point in this layer together with the inputs of various gates to form a new vector. In this way, the stopping time task model can be guaranteed to remember the situation of the previous vehicle stopping behavior and predict the next stopping behavior based on this.

number

[0041] The stopping time task model uses the relu function to activate the stopping time prediction task. This is because stopping time prediction is a regression problem, and the relu activation function generally has a better representation. The relu activation function is a maximum function. It outputs values ​​greater than 0, meaning that the neuron is activated when it is greater than 0. It suppresses outputs of values ​​less than 0, meaning that it outputs 0 when the input is negative. This allows the stopping time task model to discard some values ​​and become sparse, resulting in a better prediction effect.

[0042] When predicting stop points, this is a classification problem, and the classification labels are point categories obtained by clustering vehicle trajectory points in vehicle trajectory data using a clustering algorithm, so the stop point prediction model uses a softmax activation function. The formula for the softmax activation function is shown below, and is commonly used for multi-classification problems. Softmax maps each label value to the (0,1) interval, and the result of adding these label values ​​is 1. The label value represents the probability, and the label with the highest probability is the prediction result.

number

[0043] When predicting stop times, the loss function used by the stop time prediction model is the mean squared error (MSE), which is the result of calculating the sum of the squares of the differences between the true value and the predicted value and averaging them, and the formula is as follows:

number

[0044] When predicting stop locations, the loss function used by the stop location prediction model is cross entropy. Generally, cross entropy is used to explain the discrepancy between probability values. When two probability distributions are close, the cross entropy value is generally relatively small. The calculation method is shown below.

number

[0045] In the present invention, the loss function of the multi-task deep learning model is the result of adding formulas (6) and (7). By simultaneously optimizing these two loss functions, the multi-task deep learning model can simultaneously learn the stopping point prediction task and the stopping time prediction task to predict stopping behavior within a future time period.

[0046] As shown in FIG. 5 , the present invention further provides a vehicle parking behavior analysis and prediction system based on multi-task learning, the system includes: a data collection terminal, a parking behavior modeling terminal, a multi-task prediction terminal and a user terminal; The data collection terminal includes a vehicle GPS and OBD data collection module and a data transmission module. The vehicle GPS (Global Positioning System) and OBD (On-Board Diagnostics) data collection module mainly collects vehicle trajectory data combining GPS and OBD and transmits all collected trajectory data to the data transmission module, which is a data pre-processing module that transmits the collected vehicle trajectory data to the parking behavior modeling terminal. The data collected by the vehicle GPS and OBD data collection module includes information such as the anonymized vehicle ID, process start time, start longitude, start latitude, end time, end longitude, and end latitude. The vehicle GPS and OBD data are collectively referred to as vehicle trajectory data and are transmitted to the data pre-processing module.

[0047] The vehicle stopping behavior modeling terminal includes a data pre-processing module and a vehicle stopping behavior modeling module, the data pre-processing module first receives vehicle trajectory data transmitted by the data transmission module, and then extracts vehicle stopping behavior data from the vehicle trajectory data, the vehicle stopping behavior data specifically including stopping points and stopping times, the vehicle stopping behavior module performs spatio-temporal feature extraction on the pre-processed vehicle stopping behavior using a deep recurrent neural network method to lay the foundation for the multi-tasking prediction terminal. The data pre-processing module accepts the vehicle trajectory data and performs data pre-processing to obtain vehicle stopping behavior data, the vehicle stopping behavior data specifically including the vehicle ID, stopping start time, stopping longitude, stopping latitude and stopping end time after the vehicle is anonymized.

[0048] The multi-task prediction terminal includes a stop point classification module and a stop time regression module, and the terminal captures correlations between multiple tasks based on multi-task learning to predict tasks, and predicts stop behavior within a future time period based on the correlations between the vehicle's historical stop behavior and the multiple tasks.

[0049] The user terminal includes a prediction result receiving module and a user matching module. The prediction result receiving module receives the vehicle stopping behavior prediction result output from the multitasking prediction terminal and transmits it to a user matching module. The user matching module matches the acquired vehicle stopping behavior prediction result with an individual user and outputs the matching result to the individual user terminal.

[0050] The raw data is shown in Table 1 below, and the predicted results are shown in Table 2 below.

[0051] Table 1 Raw data table TIFF0007657965000016.tif19154

[0052] The raw data is as above. The data indicates that the vehicle is stopping at a point with longitude and latitude (108.3646, 22.82886), the point category of the point is 12, the time is June 2017, Monday, 8:00 a.m., and the vehicle has been stopped at the point for 5 minutes.

[0053] Table 2 Prediction result table TIFF0007657965000017.tif19154

[0054] As can be seen from the data in Table 2, there is a certain error in the prediction results, but it is within an acceptable range. The final term indicates the stop time, the raw data is 5, and the predicted result is 5.58. Looking at the third term in the data, sub_class, this term indicates the location number, and the predicted location classification number is 12, which represents that this location belongs to a point near location 12, and the predicted result is also location 12, which is almost accurate.

[0055] The correct results of the multi-task learning prediction network obtained after training the multi-task learning prediction network 100 times are shown in Figure 6. It can be seen that the accuracy of the multi-task learning prediction network remains unchanged after a short period of training. This is because the amount of data for the multi-task learning prediction network is too small. Since the multi-task learning prediction network can only model each vehicle user, it can only use data from one personal car. Next, as can be seen from Figure 6, the accuracy of the multi-task deep learning model quickly reaches over 90%, which is because the multi-task deep learning neural network is deeper and can fit the data better.

[0056] In summary, the method of the present invention takes into account the vehicle's historical stopping behavior, and also considers the correlation between the stopping point prediction task and the stopping time prediction task. The present invention inputs spatiotemporal features into a multi-task learning prediction network, and uses the multi-task learning prediction network to obtain the correlation between the stopping point prediction task and the stopping time prediction task based on the vehicle's historical stopping behavior, thereby predicting the stopping points and stopping times. This allows the spatiotemporal dependency of vehicle stopping behavior to be analyzed, and improves the accuracy of stopping behavior prediction. The present invention provides support for applications such as point of interest recommendation, customized travel creation, directional advertising broadcasting, and smart parking.

[0057] Corresponding to the above-mentioned embodiment of the method for analyzing and predicting vehicle stopping behavior based on multi-task learning, the present invention further provides an embodiment of an apparatus for analyzing and predicting vehicle stopping behavior based on multi-task learning.

[0058] Referring to FIG. 7, the device for analyzing and predicting vehicle stopping behavior based on multi-task learning according to an embodiment of the present invention includes one or more processors, and is for realizing the method for analyzing and predicting vehicle stopping behavior based on multi-task learning according to the above embodiment.

[0059] An embodiment of the device for analyzing and predicting vehicle stopping behavior based on multi-task learning according to the present invention may be applied to any device having data processing capabilities, which may be a computer or other device. The device embodiment may be implemented using software, hardware, or a combination of software and hardware. For example, in the case of software implementation, the device is formed by a processor in the device having data processing capabilities reading corresponding computer program instructions from a non-volatile memory into an internal memory and executing them. From a hardware perspective, FIG. 7 is a hardware structural diagram of the device having data processing capabilities in which the device for analyzing and predicting vehicle stopping behavior based on multi-task learning according to the present invention is located. In addition to the processor, internal memory, network interface, and non-volatile memory shown in FIG. 7, the device having data processing capabilities in which the device in the embodiment is located may generally include other hardware depending on the actual functions of the device having data processing capabilities, and detailed descriptions thereof will be omitted.

[0060] The process of realizing the functions and actions of each unit in the above device specifically refers to the process of realizing the corresponding step in the above method, and detailed description is omitted here.

[0061] The device embodiments basically correspond to the method embodiments, so please refer to the relevant parts of the description of the method embodiments. The device embodiments described above are merely schematic, and the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e., they may be located in one place or may be arranged in multiple network units. Depending on actual needs, some or all of the modules may be selected to achieve the objectives of the proposed invention. Those skilled in the art can understand and implement them without any creative effort.

[0062] An embodiment of the present invention further provides a computer-readable storage medium on which a program is stored, which, when executed by a processor, realizes the method for analyzing and predicting vehicle stopping behavior based on multi-task learning in the above embodiment.

[0063] The computer-readable storage medium may be an internal storage unit of any device having data processing capabilities described in any one of the above embodiments, such as a hard disk or internal memory. The computer-readable storage medium may also be any device having data processing capabilities, such as a plug-in hard disk, Smart Media Card (SMC), SD card, or flash card installed in the device. Furthermore, the computer-readable storage medium may include not only an internal storage unit of any device having data processing capabilities, but also an external storage device. The computer-readable storage medium is intended to store the computer program and other programs and data required by the device having data processing capabilities, and may also be intended to temporarily store data that has already been output or is to be output.

[0064] The above embodiments are merely for the purpose of illustrating the design concepts and features of the present invention, and are intended to enable those skilled in the art to understand and practice the contents of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design concepts disclosed in the present invention are included within the scope of protection of the present invention.

Claims

1. A method for analyzing and predicting vehicle stopping behavior based on multitask learning, comprising the following steps S1 to S3: In step S1, vehicle GPS and OBD data are collected, including a vehicle ID, a process start time, a start longitude, a start latitude, an end time, an end longitude and an end latitude after the vehicle is anonymized; In step S2, the vehicle GPS and OBD data are pre-processed to obtain vehicle stopping behavior data including stopping points and stopping times, and a deep recurrent neural network is used to perform spatio-temporal feature extraction on the pre-processed vehicle stopping behavior; Here, the extraction of spatiotemporal features from preprocessed vehicle stopping behavior using a deep recurrent neural network is Dividing the vehicle stopping behavior data into a plurality of hierarchical levels of vehicle stopping behavior data by time intervals; Processing a layer of vehicle stopping behavior data by a first layer recurrent neural network; Processing the inter-layer vehicle stopping behavior data by a second layer recurrent neural network; Transforming the vehicle stopping behavior data between the first layer recurrent neural network and the second layer recurrent neural network to realize encoding of the vehicle stopping behavior data and obtain spatio-temporal features; In step S3, the spatiotemporal features are input to a multi-task learning prediction network, and the multi-task learning prediction network obtains correlations between the stopping point prediction task and the stopping time prediction task based on the vehicle's historical stopping behavior, thereby predicting the stopping point and the stopping time. A method for analyzing and predicting vehicle stopping behavior based on multitask learning, characterized by:

2. Pre-processing the vehicle GPS and OBD data to obtain the stop time includes: Sorting the vehicle GPS and OBD data according to starting time; deleting vehicle GPS and OBD data whose end time in the preceding paragraph is not earlier than the start time in the following paragraph, deleting vehicle GPS and OBD data whose stop time in the preceding paragraph is not later than the start time, and deleting vehicle GPS and OBD data whose start and stop points do not match; and subtracting a start time of the latter item from an end time of the former item of the vehicle GPS and OBD data to obtain a stop time. The method for analyzing and predicting vehicle stopping behavior based on multi-task learning as claimed in claim 1.

3. Pre-processing vehicle GPS and OBD data to obtain stopping points includes: Aligning and merging the vehicle GPS data and OBD data with time stamps based on the user id and sorting according to the user id to obtain vehicle trajectory data; determining whether a distance between any one trajectory point and its adjacent trajectory point in the vehicle trajectory data is less than a distance threshold; If the time difference is smaller than the interval threshold, calculate the time difference between the two trajectory points; if the time difference is greater than the time difference threshold and the speed of the trajectory point is zero, label the trajectory point as a stopped state at the start of stopping, label the start of the vehicle until the end of stopping, and record the stopping point; Obtaining a set of stops for each user; obtaining a set of stops for all users. The method for analyzing and predicting vehicle stopping behavior based on multi-task learning as claimed in claim 1.

4. Processing a certain layer of vehicle stopping behavior data by a first layer recurrent neural network includes: Including, Where: indicates splicing vectors, el is the longitude of the user's vehicle stop, et is the dimension of the user's vehicle stop, and h ｉ－１ represents the execution state of the first-layer recurrent neural network in the (i-1)th iteration, Processing vehicle stopping behavior data between layers by a second-layer recurrent neural network: Including, Here, eu denotes vector information that uniquely identifies a user id, h denotes hidden state information acquired iteratively by the first-layer recurrent neural network, and h ｊ－１ shows the execution state of the second-layer recurrent neural network that iterates j-1 times. The method for analyzing and predicting vehicle stopping behavior based on multi-task learning as claimed in claim 1.

5. Inputting the spatiotemporal features into a multi-task learning prediction network, and obtaining a correlation between a stopping point prediction task and a stopping time prediction task by the multi-task learning prediction network based on the vehicle's historical stopping behavior, thereby predicting the stopping point and the stopping time; Inputting the spatio-temporal features into a multi-task learning prediction network, respectively constructing a stop point task model and a stop time task model by a long short-term memory network according to the vehicle's historical stopping behavior, simultaneously learning the stop point task model and the stop time task model by the multi-task learning prediction network, mining the correlation between the stop point prediction task and the stop time prediction task, and predicting and obtaining the stop point and stop time in a future time period; Here, the process of multi-task learning includes: selecting a relu function as an activation function, using a mean squared error function as a loss function for predicting the stop time, and using a cross entropy function as a loss function for predicting the stop point; setting the loss function of the multi-task deep learning network as the sum of the mean squared error function and the cross entropy function; simultaneously optimizing the mean squared error function and the cross entropy function; and learning and predicting the stop points and stop times in a future time period to obtain them. The method for analyzing and predicting vehicle stopping behavior based on multi-task learning as claimed in claim 1.

6. A vehicle stopping behavior analysis and prediction system based on multi-task learning, comprising: a data collection module, a vehicle stopping behavior data acquisition module, and a multi-task learning prediction module; the data collection module collects vehicle GPS and OBD data including vehicle ID, process start time, start longitude, start latitude, end time, end longitude, and end latitude after the vehicle has been anonymized; The vehicle stopping behavior data acquisition module pre-processes vehicle GPS and OBD data to obtain vehicle stopping behavior data including stopping points and stopping times, and performs spatio-temporal feature extraction on the pre-processed vehicle stopping behavior through a deep recurrent neural network; Here, the extraction of spatiotemporal features from preprocessed vehicle stopping behavior using a deep recurrent neural network is Dividing the vehicle stopping behavior data into a plurality of hierarchical levels of vehicle stopping behavior data by time intervals; Processing a layer of vehicle stopping behavior data by a first layer recurrent neural network; Processing the inter-layer vehicle stopping behavior data by a second layer recurrent neural network; Transforming the vehicle stopping behavior data between the first layer recurrent neural network and the second layer recurrent neural network to realize encoding of the vehicle stopping behavior data and obtain spatio-temporal features; The multi-task learning prediction module inputs the spatio-temporal features into a multi-task learning prediction network, and predicts the stopping point and the stopping time by obtaining a correlation between the stopping point prediction task and the stopping time prediction task through the multi-task learning prediction network based on the vehicle's historical stopping behavior. A vehicle stopping behavior analysis and prediction system based on multitask learning.

7. The vehicle stopping behavior analysis and prediction system based on multi-task learning further includes a user terminal, a prediction result receiving module, and a user matching module; The user terminal includes a prediction result receiving module and a user matching module; The prediction result receiving module receives a vehicle stopping behavior prediction result output from the multi-task learning prediction module and transmits it to a user matching module; The user matching module matches the parking behavior prediction result with a user and feeds back the matching result to the user. The vehicle stopping behavior analysis and prediction system based on multi-task learning according to claim 6.

8. An electronic device comprising a memory and a processor, the memory is coupled to the processor; Here, the memory is used to store program data, and the processor is used to execute the program data to implement the method for analyzing and predicting vehicle stopping behavior based on multitask learning according to any one of claims 1 to 5.

1. An electronic device comprising:

9. A computer-readable storage medium on which a computer program is stored, When the computer program is executed by a processor, the method for analyzing and predicting vehicle stopping behavior based on multitask learning according to any one of claims 1 to 5 is implemented. A computer-readable storage medium comprising:

Images