A method and system for generating a moving trajectory

Abstract

The embodiment of the present invention provides a method and device for generating a mobile trajectory. The embodiment of the present invention uses a recurrent neural network to extract the complex features of the mobile trajectory and model the complex transfer relationship in the mobile trajectory. In addition, a trajectory quality evaluation system is constructed based on reinforcement learning and adversarial generative networks and integrated with movement rules. By automatically comparing the difference between the generated trajectory and the real trajectory to guide model training and updating, a generation model that can generate a more realistic mobile trajectory can be obtained.

Description

Technical Field

[0001] The present invention relates to the field of virtual simulation technology, and in particular to a method and device for generating a movement trajectory of an individual. Background Art

[0002] The simulation of individual and group mobility in the city is of great significance for traffic control. In addition, the above mobility behavior can also be applied to the city simulator as the underlying driving model to generate relevant trajectories.

[0003] At present, the individual movement trajectory simulation model in the prior art is usually a model based on simple assumptions and artificial rules. For example, please refer to Chinese patent application CN102393928A, which is based on the traffic simulation integrated system interactively used by macro, meso and micro traffic simulation platforms. The disclosed simulation system includes: macro traffic simulation module, meso traffic simulation module, micro traffic simulation module, data acquisition and fusion module, the meso traffic simulation module is connected to the macro traffic simulation module, the micro traffic simulation module is connected to the meso traffic simulation module, and the data acquisition and fusion module is connected to the macro traffic simulation module, the meso traffic simulation module, and the micro traffic simulation module. The system realizes the interactive use of data from simulation systems at all levels and different simulation platforms on the basis of independent simulation of macro, meso and micro traffic simulation platforms, and can systematically and comprehensively simulate the traffic operation status of vehicle and passenger flows of large-scale activities, and provide a technical basis for the formulation of various traffic decision-making plans for large-scale activities.

[0004] Most of the existing technologies, including the above-mentioned solutions, describe the individual's mobility behavior through several manually specified key parameters, such as defining two special geographical locations of the individual's home and workplace, and then using the Markov model to finally output the simulation trajectory by transferring between home and workplace and randomly sampling other geographical locations.

[0005] The inventors have found that the prior art generally has the following problems when performing individual trajectory simulation:

[0006] 1) Based on a small number of parameters and a simple model, the model has poor expressiveness, the generated individual movement trajectories are simple, and it is difficult to simulate real complex movement behaviors, such as complex time-varying and high-order transfer laws.

[0007] 2) There is a lack of reasonable evaluation indicators for individual movement trajectories. The simulation results can only be evaluated through a number of statistical indicators, and then the parameters are manually adjusted according to the results, resulting in low iterative efficiency. Summary of the invention

[0008] The technical problem to be solved by the embodiments of the present invention is to provide a method and device for generating a mobile trajectory, which can simulate real assisted mobile behavior, and guide model training and updating by automatically comparing the difference between the generated trajectory and the real trajectory, so as to obtain a model that can generate a more realistic mobile trajectory.

[0009] To solve the above technical problem, according to one aspect of the present invention, a method for generating a movement trajectory is provided, comprising:

[0010] Acquire pre-collected historical trajectory data of multiple individuals moving in the target area, wherein the historical trajectory data at least includes the geographic location and time information of the individuals;

[0011] Inputting the data of the historical trajectory into a generator, predicting the next position point in the historical trajectory through the generator, and pre-training the generator; and inputting the historical trajectory and the perturbation trajectory into a discriminator, and pre-training the discriminator by training the discriminator to distinguish the perturbation trajectory, wherein the perturbation trajectory is obtained by perturbing some position points in the historical trajectory;

[0012] Repeat the following iterative process until a preset convergence condition is reached: generate a virtual trajectory using the pre-trained generator; input the virtual trajectory into the pre-trained discriminator, use the discriminator to discriminate the virtual trajectory and the sampled historical trajectory, and obtain a discrimination result; use the reinforcement learning algorithm to generate a gradient corresponding to the discrimination result and back-propagate it to the generator, and update the generator;

[0013] The generator is used to generate the simulation trajectory of the target area according to the starting point of the simulation trajectory.

[0014] According to at least one embodiment of the present invention, the generator includes an embedding layer, a bidirectional LSTM network and multiple linear layers, wherein:

[0015] The embedding layer generates a feature vector sequence for characterizing each historical trajectory according to the data of the historical trajectory, wherein the feature vector sequence includes a preset number of position point feature vectors;

[0016] The bidirectional LSTM network receives the feature vector sequence generated by the embedding layer and generates a vector representation of the next position point;

[0017] The multiple linear layers map the vector representation of the next position obtained by the bidirectional LSTM network into the geographic location of the next position point.

[0018] According to at least one embodiment of the present invention, the historical trajectory data is trajectory data within a preset time length range; and the step of generating a feature vector sequence for characterizing each historical trajectory according to the historical trajectory data comprises:

[0019] A preset number of location points are sampled from the data of the historical trajectory of each individual's movement, and a preset time interval is set between adjacent location points;

[0020] Divide the target area into a plurality of grids, and determine the grid to which each location point belongs according to the geographical location of each location point as the spatial attribute of the location point; and divide the preset time range into a plurality of time periods, and obtain the time period to which each location point belongs according to the time information of each location point as the time attribute of the location point;

[0021] According to the spatial attributes and temporal attributes of each location point, a feature vector of each location point is generated to obtain a feature vector sequence of each historical trajectory.

[0022] According to at least one embodiment of the present invention, the historical trajectory data further includes at least one of the following information: point of interest (POI) information of the area where the geographic location is located, and context information of the geographic location.

[0023] According to at least one embodiment of the present invention, the preset convergence condition is: a success rate of the discriminator in distinguishing the virtual trajectory is less than a preset threshold.

[0024] According to at least one embodiment of the present invention, the step of using the discriminator to discriminate the virtual trajectory and the sampled historical trajectory to obtain a discrimination result includes:

[0025] Using the discriminator to distinguish the virtual trajectory from the sampled historical trajectory, and obtaining a first judgment result of the authenticity of the virtual trajectory;

[0026] Calculating a sum of distances between all adjacent position points on the virtual track, and generating a second judgment result of the authenticity of the virtual track according to the sum, wherein the authenticity is negatively correlated with the sum;

[0027] Calculating the difference in geographical location between relevant time points on the virtual trajectory, and generating a third judgment result of the authenticity of the virtual trajectory according to the difference, wherein the authenticity is negatively correlated with the difference; the relevant time points include the same periodic time points and different time points set according to the periodicity of the individual movement law;

[0028] A determination result of the authenticity of the virtual trajectory is generated according to the first determination result, the second determination result and the third determination result.

[0029] According to at least one embodiment of the present invention, the starting point of the simulation trajectory is generated in the following manner:

[0030] According to the historical trajectory data of the target area, the population distribution characteristics in the preset starting period are counted;

[0031] Determine the distribution probability of the starting point of the simulation trajectory in the target area according to the population distribution characteristics, wherein the distribution probability of the starting point in a grid is positively correlated with the population density in the grid;

[0032] The starting point of the simulation trajectory is generated according to the distribution probability of the starting point in the target area.

[0033] According to at least one embodiment of the present invention, the step of generating a virtual trajectory using the pre-trained generator includes:

[0034] After generating some position points of a virtual trajectory using the pre-trained generator, the remaining position points of the virtual trajectory are generated by random sampling to obtain the virtual trajectory.

[0035] According to another aspect of the present invention, a system for generating a movement trajectory is also provided, comprising:

[0036] A data acquisition module, used to acquire pre-collected historical trajectory data of multiple individuals moving in a target area, wherein the historical trajectory data at least includes the geographic location and time information of the individuals;

[0037] A pre-training module, for inputting the data of the historical trajectory into a generator, predicting the next position point in the historical trajectory through the generator, and pre-training the generator; and inputting the historical trajectory and the perturbation trajectory into a discriminator, and pre-training the discriminator by training the discriminator to distinguish the perturbation trajectory, wherein the perturbation trajectory is obtained by perturbing some position points in the historical trajectory;

[0038] The formal training module is used to repeatedly perform the following iterative processing until a preset convergence condition is reached: using the pre-trained generator to generate a virtual trajectory; inputting the virtual trajectory into the pre-trained discriminator, using the discriminator to discriminate the virtual trajectory and the sampled historical trajectory to obtain a discrimination result; using a reinforcement learning algorithm to generate a gradient corresponding to the discrimination result and back-propagating it to the generator, and updating the generator;

[0039] The trajectory simulation module is used to generate the simulation trajectory of the target area using the generator according to the starting point of the simulation trajectory.

[0040] According to at least one embodiment of the present invention, the generator includes an embedding layer, a bidirectional LSTM network and multiple linear layers, wherein:

[0041] The embedding layer generates a feature vector sequence for characterizing each historical trajectory according to the data of the historical trajectory, wherein the feature vector sequence includes a preset number of position point feature vectors;

[0042] The bidirectional LSTM network receives the feature vector sequence generated by the embedding layer and generates a vector representation of the next position point;

[0043] The multiple linear layers map the vector representation of the next position obtained by the bidirectional LSTM network into the geographic location of the next position point.

[0044] According to at least one embodiment of the present invention, the formal training module is further used to:

[0045] Using the discriminator to distinguish the virtual trajectory from the sampled historical trajectory, and obtaining a first judgment result of the authenticity of the virtual trajectory;

[0046] Calculating a sum of distances between all adjacent position points on the virtual track, and generating a second judgment result of the authenticity of the virtual track according to the sum, wherein the authenticity is negatively correlated with the sum;

[0047] Calculating the difference in geographical location between relevant time points on the virtual trajectory, and generating a third judgment result of the authenticity of the virtual trajectory according to the difference, wherein the authenticity is negatively correlated with the difference; the relevant time points include the same periodic time points and different time points set according to the periodicity of the individual movement law;

[0048] A determination result of the authenticity of the virtual trajectory is generated according to the first determination result, the second determination result and the third determination result.

[0049] According to at least one embodiment of the present invention, the trajectory simulation module is further used to generate the starting point of the simulation trajectory in the following manner:

[0050] According to the historical trajectory data of the target area, the population distribution characteristics in the preset starting period are counted;

[0051] Determine the distribution probability of the starting point of the simulation trajectory in the target area according to the population distribution characteristics, wherein the distribution probability of the starting point in a grid is positively correlated with the population density in the grid;

[0052] The starting point of the simulation trajectory is generated according to the distribution probability of the starting point in the target area.

[0053] According to at least one embodiment of the present invention, the formal training module is further used to generate some position points of the virtual trajectory by using the generator obtained by pre-training, and then generate the remaining position points of the virtual trajectory by random sampling to obtain the virtual trajectory.

[0054] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for generating a movement trajectory as described above are implemented.

[0055] Compared with the prior art, the method and device for generating a movement trajectory provided by the embodiment of the present invention have at least the following beneficial effects:

[0056] 1) Using recurrent neural networks to extract complex features of mobile trajectories, combined with carefully constructed multiple mobile behavior features, complex transfer relationships in mobile trajectories can be modeled;

[0057] 2) A trajectory quality evaluation system is constructed based on reinforcement learning and generative adversarial networks and integrated with movement rules. By automatically comparing the differences between the generated trajectory and the real trajectory to guide model training and updating, a generative model that can generate more realistic movement trajectories can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0059] Figure 1 A system block diagram of a system for generating a movement trajectory provided by an embodiment of the present invention;

[0060] Figure 2 A schematic diagram of the structure of a generator provided by an embodiment of the present invention;

[0061] Figure 3 A schematic diagram of a flow chart of a method for generating a moving trajectory provided by an embodiment of the present invention;

[0062] Figure 4 A schematic diagram of the structure of a system for generating a movement trajectory provided by an embodiment of the present invention;

[0063] Figure 5 Another schematic diagram of the structure of the system for generating a movement trajectory provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail in conjunction with the accompanying drawings and specific embodiments. In the following description, specific details such as specific configurations and components are provided only to help fully understand the embodiments of the present invention. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. In addition, for clarity and brevity, the description of known functions and structures is omitted.

[0065] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present invention. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0066] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0067] As described in the background technology, the mobile trajectory generation methods of the prior art usually have problems such as difficulty in simulating real and complex mobile behaviors or difficulty in reasonably evaluating the simulation results. In order to solve at least one of the above problems, an embodiment of the present invention provides a mobile trajectory generation method, which uses a recurrent neural network to extract the complex features of the mobile trajectory, and combines a variety of carefully constructed mobile behavior features to model the complex transfer relationship in the mobile trajectory. In addition, the embodiment of the present invention also constructs a trajectory quality evaluation system based on reinforcement learning and adversarial generative networks and integrates movement rules, and guides model training and updating by automatically comparing the differences between the generated trajectory and the real trajectory, so as to obtain a generation model that can generate a more realistic mobile trajectory.

[0068] Figure 1 The figure is a schematic diagram of a system block diagram of a system for generating a movement trajectory according to an embodiment of the present invention. Figure 1 The mobile trajectory generation system 10 shown in the figure adopts the network architecture of Generative Adversarial Networks (GAN), which includes two main network models, namely a generator and a discriminator. Figure 2An example of a specific structure of a generator of an embodiment of the present invention is given, and the generator includes an embedding layer, an LSTM layer, and multiple linear layers. The embedding layer generates a feature vector sequence for characterizing each historical trajectory based on the data of the historical trajectory, and the feature vector sequence includes a preset number of location point feature vectors. The LSTM layer is a bidirectional LSTM network, which is used to receive the feature vector sequence generated by the embedding layer and generate a vector representation of the next location point. The multiple linear layers map the vector representation of the next location obtained by the bidirectional LSTM network into the geographical location of the next location point.

[0069] Here, the embodiment of the present invention introduces a bidirectional LSTM network to extract time series features from unknown feature vectors. The starting point is that the geographical location of an individual's next time point (or next hop) is not only related to the individual's previous time point, but also to the subsequent time point of the next time point. For example, a student's decision on where to eat after class is related to both the location of his class and the location of his class after dinner. Therefore, the location of the next hop is not simply determined by the current location. The use of a bidirectional LSTM network can more effectively capture the above features; the linear layer reduces the feature dimension through multi-layer perception, and generates the probability distribution of the next hop in the full space based on this, usually taking the location point with the highest probability as the location of the next hop.

[0070] like Figure 3 As shown, a method for generating a movement trajectory provided by an embodiment of the present invention includes:

[0071] Step 31 : obtaining pre-collected historical trajectory data of multiple individuals moving in the target area, wherein the historical trajectory data at least includes the geographic location and time information of the individuals.

[0072] Here, the target area may be a city or other area with a certain spatial range that needs to be simulated. The individuals may be pedestrians and various transportation vehicles (such as bicycles, cars, and motorcycles). The embodiment of the present invention obtains data on the historical trajectories of multiple individuals in the target area collected in advance, and the data may generally include the geographic location of the individual (such as latitude and longitude information) and the time information when the geographic location is collected.

[0073] In addition, in order to better reflect the individual's movement patterns, the data of the historical trajectory may also include at least one of the following information: information about points of interest (POI) in the area where the geographic location is located, and context information of the geographic location. The context information of the geographic location is usually information related to the source of the trajectory data. For example, taking the historical trajectory obtained from a social network as an example, the context of the geographic location may be a textual description of the geographic location; taking the historical trajectory obtained from an operator of a mobile communication network as an example, the context of the geographic location may be relevant information of the base station that collected the geographic location, such as the location of the base station, coverage, and number of sectors.

[0074] Step 32: input the data of the historical trajectory into a generator, predict the next position point in the historical trajectory through the generator, and pre-train the generator; and input the historical trajectory and the perturbation trajectory into a discriminator, train the discriminator to distinguish the perturbation trajectory, and pre-train the discriminator, wherein the perturbation trajectory is obtained by perturbing some position points in the historical trajectory.

[0075] In step 32, the embodiment of the present invention pre-trains the generator and the discriminator. Before pre-training, the data of the historical trajectory needs to be pre-processed, that is, the individual's geographical location and related information (such as POI and context) are extracted from each trajectory at a preset time interval, so as to obtain the geographical location and related information of multiple time points (also referred to as time points in this article), and the geographical location of each time point is used as a location point of the trajectory (sometimes also referred to as trajectory point in this article), so that multiple location points can be obtained, and the attributes of each location point include time, geographical location, and may also include POI and context. Then, the attributes of each location point are converted into a high-dimensional feature vector (also referred to as location point feature vector in this article), so that each trajectory can be characterized by a feature vector sequence of multiple location points of the trajectory. Each location point feature vector in the feature vector sequence is sorted in chronological order. For ease of processing, the embodiment of the present invention can generate a preset number of location point feature vectors. For example, assuming that the sampled trajectory is the movement trajectory of an individual on a natural day, sampling can be performed every hour to generate a feature vector sequence of 24 location points.

[0076] Here, the data of the historical trajectory is the trajectory data within a preset time range, that is, the trajectory data records the trajectory of an individual within a preset time range, and the preset length can be 24 hours a day, or multiple days, etc. Specifically, the embodiment of the present invention can sample a preset number of location points from the historical trajectory data of each individual's movement, and the interval between adjacent location points is preset; then, the target area is divided into multiple grids, and the grid to which each location point belongs is determined according to the geographical location of each location point, as the spatial attribute of the location point; and, the preset time range is divided into multiple time periods, and the time period to which each location point belongs is obtained according to the time information of each location point, as the time attribute of the location point; then, according to the spatial attribute and time attribute of each location point, a feature vector of each location point is generated to obtain a feature vector sequence of each historical trajectory. Of course, when considering more attributes, such as POI and context information, the embodiment of the present invention can convert the above attributes into corresponding feature vectors according to the word embedding algorithm, and form the feature vector of the location point together with the geographical location and time attribute.

[0077] In order to enable the network model to converge faster in the subsequent formal training process, the embodiment of the present invention pre-trains the generator and the discriminator in step 32. Through pre-training, the generator can learn some sequence transfer rules, so that the discriminator can distinguish between the real trajectory and the perturbed trajectory. Here, the perturbed trajectory can be obtained by perturbing some position points in the real historical trajectory, for example, changing the geographical location of one or some position points in the historical trajectory, and / or changing the time of some position points in the historical trajectory, etc.

[0078] Step 33, repeatedly perform the following iterative processing until a preset convergence condition is reached: generate a virtual trajectory using the pre-trained generator; input the virtual trajectory into the pre-trained discriminator, use the discriminator to discriminate the virtual trajectory and the sampled historical trajectory, and obtain a discrimination result; use a reinforcement learning algorithm to generate a gradient corresponding to the discrimination result and back-propagate it to the generator, and update the generator.

[0079] After the pre-training in step 32, the embodiment of the present invention performs formal training on the above-mentioned network model through step 33. During the formal training process, training is performed based on the principle of generative adversarial networks. Among them, the generator generates a virtual trajectory and provides it to the discriminator. The discriminator discriminates the virtual trajectory and the real historical trajectory, obtains a judgment result of whether the virtual trajectory is a real trajectory, and uses a reinforcement learning algorithm to generate a gradient corresponding to the judgment result and feeds it back to the generator, so that the generator updates the model parameters until a preset convergence condition is reached. For example, the discriminator cannot distinguish between the real historical trajectory and the virtual trajectory generated by the generator. In practical applications, the convergence condition can be set to: the success rate of the discriminator in distinguishing the virtual trajectory is less than a preset threshold.

[0080] In addition, when generating a virtual trajectory, the generator may first determine the starting point of the virtual trajectory, and then generate the location point of the next jump based on the starting point, and combine it with the generated location points to continue generating subsequent location points until a virtual trajectory including a preset number of location points is generated. The starting point of the virtual trajectory can be based on the population distribution characteristics of the target area, for example, the population density of the population in the target area in a preset starting time period (for example, most individual movement trajectories start in a certain time period in the early morning or the morning), and the starting point of the virtual trajectory is generated based on the population density, wherein the greater the population density of a certain location point, the greater the probability that the location point is selected as the starting point.

[0081] Step 34, generating a simulation trajectory of the target area using the generator according to the starting point of the simulation trajectory.

[0082] Through steps 31 to 33, the embodiment of the present invention can train a generator, and the generator can be used for the subsequent generation of simulation trajectories. When generating a simulation trajectory, the embodiment of the present invention can count the population distribution characteristics in a preset starting period based on the data of the historical trajectory of the target area, such as the population density in each grid; according to the population distribution characteristics, determine the distribution probability of the starting point of the simulation trajectory in the target area, wherein the distribution probability of the starting point in a grid is positively correlated with the population density in the grid; then, based on the distribution probability of the starting point in the target area, generate the starting point of the simulation trajectory. Based on the above starting point, the generator is used to generate the location point of the next hop hop by hop to obtain one or more complete simulation trajectories.

[0083] Through the above steps, the embodiment of the present invention realizes the generation of simulation trajectories. In the above process, the embodiment of the present invention uses a recurrent neural network to extract the complex features of the mobile trajectory and model the complex transfer relationship in the mobile trajectory. In addition, a trajectory quality evaluation system is constructed based on reinforcement learning and adversarial generative networks and integrated with movement rules. By automatically comparing the difference between the generated trajectory and the real trajectory to guide model training and updating, a generation model that can generate a more realistic mobile trajectory can be obtained.

[0084] According to at least one embodiment of the present invention, in the above step 33, when the discriminator discriminates the authenticity of the trajectory, the embodiment of the present invention also introduces more laws of individual mobility, such as distance laws and periodic laws. For example, individual movement is affected by the traffic conditions that can be provided by the transportation vehicles and the target area, and its movement distance is usually limited, which can be reflected by the distance law of individual movement.

[0085] For another example, the movement of an individual usually has a certain periodicity. For example, an individual is usually in the company during working hours on different weekdays, and is usually at home in the morning or evening on weekdays, that is, an individual is usually in the same geographical location at the same time point on different natural days, or the distance between geographical locations is relatively small, and at certain time points on the same natural day, the individual is usually in the same geographical location, or the distance between geographical locations is relatively small. Therefore, the embodiment of the present invention can pre-set relevant time points, including the same periodic time points and different time points set according to the periodicity of the individual movement law, and reflect the periodic law of individual movement through the geographical location of the relevant time points.

[0086] Based on the above analysis, in the above step 33, when the virtual trajectory and the sampled historical trajectory are judged and a judgment result of whether the virtual trajectory is real is obtained, the embodiment of the present invention may include:

[0087] 1) using the discriminator to distinguish the virtual trajectory from the sampled historical trajectory to obtain a first judgment result of the authenticity of the virtual trajectory;

[0088] 2) calculating the sum of the distances between all adjacent position points on the virtual trajectory, and generating a second judgment result of the authenticity of the virtual trajectory according to the sum, wherein the authenticity is negatively correlated with the sum, which can reflect whether the virtual trajectory conforms to the distance law of individual movement;

[0089] 3) calculating the difference in geographical location between relevant time points on the virtual trajectory, and generating a third judgment result of the authenticity of the virtual trajectory according to the difference, wherein the authenticity is negatively correlated with the difference, which can reflect whether the virtual trajectory conforms to the periodic law of individual movement; the relevant time points include the same periodic time points and different time points set according to the periodicity of the individual movement law;

[0090] 4) Generate a result of determining the authenticity of the virtual track based on the first, second and third determination results. For example, the first, second and third determination results may be weighted and summed to obtain a final determination result for determining whether the virtual track is authentic, where authenticity refers to whether the virtual track is similar to the real track in terms of feature morphology.

[0091] In addition, when the discriminator makes a judgment, it may make a judgment on a plurality of virtual trajectories, and generate a final judgment result according to the judgment results of these virtual trajectories.

[0092] Through the above processing, the embodiment of the present invention introduces more features that can reflect the laws of real moving trajectories into the discriminator, which can improve the discriminant performance of virtual trajectories, and further guide the update of the generator to obtain a generator with better performance.

[0093] In the process of the generator generating a preset number of position points of a virtual sequence, a large deviation from the real trajectory may occur starting from a certain position point. On this basis, more position points of the trajectory subsequently generated will also be far away from the real trajectory, resulting in a large difference in the morphology of the virtual trajectory and the real trajectory. According to at least one embodiment of the present invention, in order to improve the efficiency of model training, in the above step 33, the embodiment of the present invention can use the generator obtained by pre-training to generate some position points of the virtual trajectory, and then generate the remaining position points of the virtual trajectory by random sampling to obtain the virtual trajectory. Here, each position point can be represented by the feature vector corresponding to the position point, and each trajectory can be represented by a feature vector sequence composed of the feature vectors of all the position points of the trajectory. In addition, the number of the partial position points can be preset. For example, assuming that the number of position points of each trajectory is 24, it can be set to 5, 8 or 12, etc.

[0094] That is to say, in adversarial training, the generator samples according to the next hop access probability and obtains the specific position of the next hop, and combines the position with the historical jump position into a partial sequence. Since the discriminator cannot accept partial sequences as input, the partial sequence needs to be completed using the currently trained generator, that is, the partial sequence is input into the generator and each hop is sampled in the same way until it is fully completed; since the sampling process is random, a single sampling process cannot give a fair reward value. The embodiment of the present invention can use Monte Carlo Tree Search (MTS) to find the reward value. The MC algorithm searches in the space of all the completion data, inputs the search results into the discriminator for discrimination and obtains the average probability of its output as the reward value, and updates the generative model responsible for completing the sequence and the generative model responsible for generating jumps according to the reward gradient corresponding to the strategy; the updating of the main generative model (i.e., the model responsible for jumping) is to back-propagate the network by treating rewards as losses and update the weights according to the gradient, and then delay the update of the completion model at a certain rate after the updated network, because the task of the completion model is different from that of the jump model, and the jump model cannot be blindly followed, otherwise it would be meaningless to design such a completion mechanism; then the generator is used to generate a batch of virtual trajectory data and mixed with the real trajectory data into the discriminator to train the recognition ability of the discriminator.

[0095] Based on the above method for generating a moving trajectory, an embodiment of the present invention further provides a device for implementing the above method.

[0096] Please refer to Figure 4 , a moving trajectory generation system 40 provided by an embodiment of the present invention includes:

[0097] A data acquisition module 41 is used to acquire pre-collected historical trajectory data of multiple individuals moving in a target area, wherein the historical trajectory data at least includes geographic location and time information of the individuals;

[0098] A pre-training module 42 is used to input the data of the historical trajectory into a generator, predict the next position point in the historical trajectory through the generator, and pre-train the generator; and input the historical trajectory and the perturbation trajectory into a discriminator, and pre-train the discriminator by training the discriminator to distinguish the perturbation trajectory, wherein the perturbation trajectory is obtained by perturbing some position points in the historical trajectory;

[0099] The formal training module 43 is used to repeatedly perform the following iterative processing until a preset convergence condition is reached: using the pre-trained generator to generate a virtual trajectory; inputting the virtual trajectory into the pre-trained discriminator, using the discriminator to discriminate the virtual trajectory and the sampled historical trajectory to obtain a discrimination result; using a reinforcement learning algorithm to generate a gradient corresponding to the discrimination result and back-propagating it to the generator, and updating the generator;

[0100] The trajectory simulation module 44 is used to generate the simulation trajectory of the target area using the generator according to the starting point of the simulation trajectory.

[0101] Through the above modules, the embodiment of the present invention can simulate real assisted movement behavior, and guide model training and updating by automatically comparing the difference between the generated trajectory and the real trajectory, so as to obtain a model that can generate a more realistic movement trajectory.

[0102] According to at least one embodiment of the present invention, the generator includes an embedding layer, a bidirectional LSTM network and multiple linear layers, wherein:

[0103] The embedding layer generates a feature vector sequence for characterizing each historical trajectory according to the data of the historical trajectory, wherein the feature vector sequence includes a preset number of position point feature vectors;

[0104] The bidirectional LSTM network receives the feature vector sequence generated by the embedding layer and generates a vector representation of the next position point;

[0105] The multiple linear layers map the vector representation of the next position obtained by the bidirectional LSTM network into the geographic location of the next position point.

[0106] According to at least one embodiment of the present invention, the data of the historical trajectory is trajectory data within a preset time length range. Specifically, the embedding layer samples a preset number of position points from the data of the historical trajectory of each individual's movement, and the interval between adjacent position points is preset for a preset time length; the target area is divided into multiple grids, and the grid to which each position point belongs is determined according to the geographical location of each position point, as the spatial attribute of the position point; and the preset time range is divided into multiple time periods, and the time period to which each position point belongs is obtained according to the time information of each position point, as the time attribute of the position point; the feature vector of each position point is generated according to the spatial attribute and time attribute of each position point, and the feature vector sequence of each historical trajectory is obtained.

[0107] According to at least one embodiment of the present invention, the historical trajectory data further includes at least one of the following information: point of interest (POI) information of the area where the geographic location is located, and context information of the geographic location.

[0108] According to at least one embodiment of the present invention, the preset convergence condition is: a success rate of the discriminator in distinguishing the virtual trajectory is less than a preset threshold.

[0109] According to at least one embodiment of the present invention, the formal training module 43 is further used to:

[0110] Using the discriminator to distinguish the virtual trajectory from the sampled historical trajectory, and obtaining a first judgment result of the authenticity of the virtual trajectory;

[0111] Calculating a sum of distances between all adjacent position points on the virtual track, and generating a second judgment result of the authenticity of the virtual track according to the sum, wherein the authenticity is negatively correlated with the sum;

[0112] Calculating the difference in geographical location between relevant time points on the virtual trajectory, and generating a third judgment result of the authenticity of the virtual trajectory according to the difference, wherein the authenticity is negatively correlated with the difference; the relevant time points include the same periodic time points and different time points set according to the periodicity of the individual movement law;

[0113] A determination result of the authenticity of the virtual trajectory is generated according to the first determination result, the second determination result and the third determination result.

[0114] According to at least one embodiment of the present invention, the trajectory simulation module is further used to generate the starting point of the simulation trajectory in the following manner:

[0115] According to the historical trajectory data of the target area, the population distribution characteristics in the preset starting period are counted;

[0116] Determine the distribution probability of the starting point of the simulation trajectory in the target area according to the population distribution characteristics, wherein the distribution probability of the starting point in a grid is positively correlated with the population density in the grid;

[0117] The starting point of the simulation trajectory is generated according to the distribution probability of the starting point in the target area.

[0118] According to at least one embodiment of the present invention, the formal training module is further used to generate some position points of the virtual trajectory by using the generator obtained by pre-training, and then generate the remaining position points of the virtual trajectory by random sampling to obtain the virtual trajectory.

[0119] like Figure 5As shown, the embodiment of the present invention further provides another mobile trajectory generation system 50, which specifically includes a processor 51, a memory 52, a bus system 53, a receiver 54 and a transmitter 55. The processor 51, the memory 52, the receiver 54 and the transmitter 55 are connected via the bus system 53, the memory 52 is used to store instructions, and the processor 51 is used to execute the instructions stored in the memory 52 to control the receiver 54 to receive signals and control the transmitter 55 to send signals;

[0120] The processor 51 is used to read the program in the memory and execute the following process:

[0121] Acquire pre-collected historical trajectory data of multiple individuals moving in the target area, wherein the historical trajectory data at least includes the geographic location and time information of the individuals;

[0122] Inputting the data of the historical trajectory into a generator, predicting the next position point in the historical trajectory through the generator, and pre-training the generator; and inputting the historical trajectory and the perturbation trajectory into a discriminator, and pre-training the discriminator by training the discriminator to distinguish the perturbation trajectory, wherein the perturbation trajectory is obtained by perturbing some position points in the historical trajectory;

[0123] Repeat the following iterative process until a preset convergence condition is reached: generate a virtual trajectory using the pre-trained generator; input the virtual trajectory into the pre-trained discriminator, use the discriminator to discriminate the virtual trajectory and the sampled historical trajectory, and obtain a discrimination result; use the reinforcement learning algorithm to generate a gradient corresponding to the discrimination result and back-propagate it to the generator, and update the generator;

[0124] The generator is used to generate the simulation trajectory of the target area according to the starting point of the simulation trajectory.

[0125] It should be understood that in the embodiment of the present invention, the processor 51 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0126] The memory 52 may include a read-only memory and a random access memory, and provides instructions and data to the processor 51. A portion of the memory 52 may also include a non-volatile random access memory. For example, the memory 52 may also store information on the device type.

[0127] The bus system 53 may include, in addition to the data bus, a power bus, a control bus, a status signal bus, etc. However, for the sake of clarity, various buses are labeled as the bus system 53 in the figure.

[0128] In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 51 or the instruction in the form of software. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as the execution of the hardware processor, or the execution is completed by the combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 52, and the processor 51 reads the information in the memory 52 and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0129] According to at least one embodiment of the present invention, the generator includes an embedding layer, a bidirectional LSTM network and multiple linear layers, wherein:

[0130] The embedding layer generates a feature vector sequence for characterizing each historical trajectory according to the data of the historical trajectory, wherein the feature vector sequence includes a preset number of position point feature vectors;

[0131] The bidirectional LSTM network receives the feature vector sequence generated by the embedding layer and generates a vector representation of the next position point;

[0132] The multiple linear layers map the vector representation of the next position obtained by the bidirectional LSTM network into the geographic location of the next position point.

[0133] According to at least one embodiment of the present invention, the historical trajectory data is trajectory data within a preset time length range; when the program is executed by the processor 51, the following steps may also be implemented:

[0134] A preset number of location points are sampled from the data of the historical trajectory of each individual's movement, and a preset time interval is set between adjacent location points;

[0135] Divide the target area into a plurality of grids, and determine the grid to which each location point belongs according to the geographical location of each location point as the spatial attribute of the location point; and divide the preset time range into a plurality of time periods, and obtain the time period to which each location point belongs according to the time information of each location point as the time attribute of the location point;

[0136] According to the spatial attributes and temporal attributes of each location point, a feature vector of each location point is generated to obtain a feature vector sequence of each historical trajectory.

[0137] According to at least one embodiment of the present invention, the historical trajectory data further includes at least one of the following information: point of interest (POI) information of the area where the geographic location is located, and context information of the geographic location.

[0138] According to at least one embodiment of the present invention, the preset convergence condition is: a success rate of the discriminator in distinguishing the virtual trajectory is less than a preset threshold.

[0139] According to at least one embodiment of the present invention, when the program is executed by the processor 51, the following steps may also be implemented:

[0140] Using the discriminator to distinguish the virtual trajectory from the sampled historical trajectory, and obtaining a first judgment result of the authenticity of the virtual trajectory;

[0141] Calculating a sum of distances between all adjacent position points on the virtual track, and generating a second judgment result of the authenticity of the virtual track according to the sum, wherein the authenticity is negatively correlated with the sum;

[0142] Calculating the difference in geographical location between relevant time points on the virtual trajectory, and generating a third judgment result of the authenticity of the virtual trajectory according to the difference, wherein the authenticity is negatively correlated with the difference; the relevant time points include the same periodic time points and different time points set according to the periodicity of the individual movement law;

[0143] A determination result of the authenticity of the virtual trajectory is generated according to the first determination result, the second determination result and the third determination result.

[0144] According to at least one embodiment of the present invention, when the program is executed by the processor 51, the following steps may also be implemented:

[0145] The starting point of the simulation trajectory is generated as follows:

[0146] According to the historical trajectory data of the target area, the population distribution characteristics in the preset starting period are counted;

[0147] Determine the distribution probability of the starting point of the simulation trajectory in the target area according to the population distribution characteristics, wherein the distribution probability of the starting point in a grid is positively correlated with the population density in the grid;

[0148] The starting point of the simulation trajectory is generated according to the distribution probability of the starting point in the target area.

[0149] According to at least one embodiment of the present invention, when the program is executed by the processor 51, the following steps may also be implemented:

[0150] After generating some position points of a virtual trajectory using the pre-trained generator, the remaining position points of the virtual trajectory are generated by random sampling to obtain the virtual trajectory.

[0151] In some embodiments of the present invention, a computer-readable storage medium is further provided, on which a program is stored, and when the program is executed by a processor, the following steps can be implemented:

[0152] Acquire pre-collected historical trajectory data of multiple individuals moving in the target area, wherein the historical trajectory data at least includes the geographic location and time information of the individuals;

[0153] Inputting the data of the historical trajectory into a generator, predicting the next position point in the historical trajectory through the generator, and pre-training the generator; and inputting the historical trajectory and the perturbation trajectory into a discriminator, and pre-training the discriminator by training the discriminator to distinguish the perturbation trajectory, wherein the perturbation trajectory is obtained by perturbing some position points in the historical trajectory;

[0154] Repeat the following iterative process until a preset convergence condition is reached: generate a virtual trajectory using the pre-trained generator; input the virtual trajectory into the pre-trained discriminator, use the discriminator to discriminate the virtual trajectory and the sampled historical trajectory, and obtain a discrimination result; use the reinforcement learning algorithm to generate a gradient corresponding to the discrimination result and back-propagate it to the generator, and update the generator;

[0155] The generator is used to generate the simulation trajectory of the target area according to the starting point of the simulation trajectory.

[0156] When the program is executed by the processor, it can achieve Figure 3 All implementations of the method for generating the moving trajectory shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0157] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0158] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0159] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0160] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0161] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0162] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0163] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for generating a moving trajectory, characterized in that: include: Acquire pre-collected historical trajectory data of multiple individuals moving in the target area, wherein the historical trajectory data at least includes the geographic location and time information of the individuals; Inputting the data of the historical trajectory into a generator, predicting the next position point in the historical trajectory through the generator, and pre-training the generator; and, inputting the historical trajectory and the perturbation trajectory into a discriminator, and pre-training the discriminator by training the discriminator to distinguish the perturbation trajectory, wherein the perturbation trajectory is obtained by perturbing some position points in the historical trajectory; Repeat the following iterative process until a preset convergence condition is reached: generate a virtual trajectory using the pre-trained generator; input the virtual trajectory into the pre-trained discriminator, use the discriminator to discriminate the virtual trajectory and the sampled historical trajectory, and obtain a discrimination result; use the reinforcement learning algorithm to generate a gradient corresponding to the discrimination result and back-propagate it to the generator, and update the generator; According to the starting point of the simulation trajectory, the simulation trajectory of the target area is generated by the generator, The step of using the discriminator to discriminate the virtual trajectory and the sampled historical trajectory to obtain a discrimination result includes: Using the discriminator to distinguish the virtual trajectory from the sampled historical trajectory, and obtaining a first judgment result of the authenticity of the virtual trajectory; Calculating a sum of distances between all adjacent position points on the virtual track, and generating a second judgment result of the authenticity of the virtual track according to the sum, wherein the authenticity is negatively correlated with the sum; Calculating the difference in geographical location between relevant time points on the virtual trajectory, and generating a third judgment result of the authenticity of the virtual trajectory according to the difference, wherein the authenticity is negatively correlated with the difference; the relevant time points include the same periodic time points and different time points set according to the periodicity of the individual movement law; A determination result of the authenticity of the virtual trajectory is generated according to the first determination result, the second determination result and the third determination result.

2. The method according to claim 1, characterized in that The generator includes an embedding layer, a bidirectional LSTM network and multiple linear layers, wherein: The embedding layer generates a feature vector sequence for characterizing each historical trajectory according to the data of the historical trajectory, wherein the feature vector sequence includes a preset number of position point feature vectors; The bidirectional LSTM network receives the feature vector sequence generated by the embedding layer and generates a vector representation of the next position point; The multiple linear layers map the vector representation of the next position obtained by the bidirectional LSTM network into the geographic location of the next position point.

3. The method according to claim 2, characterized in that The historical trajectory data is trajectory data within a preset time range; The step of generating a feature vector sequence for characterizing each historical trajectory according to the historical trajectory data comprises: A preset number of location points are sampled from the data of the historical trajectory of each individual's movement, and a preset time interval is set between adjacent location points; Divide the target area into a plurality of grids, and determine the grid to which each location point belongs according to the geographical location of each location point as the spatial attribute of the location point; and divide the preset time length range into a plurality of time periods, and obtain the time period to which each location point belongs according to the time information of each location point as the time attribute of the location point; According to the spatial attributes and temporal attributes of each location point, a feature vector of each location point is generated to obtain a feature vector sequence of each historical trajectory.

4. The method according to claim 1, characterized in that The historical trajectory data also includes at least one of the following information: point of interest (POI) information of the area where the geographic location is located, and context information of the geographic location.

5. The method according to claim 1, characterized in that The preset convergence condition is: the success rate of the discriminator in distinguishing the virtual trajectory is less than a preset threshold.

6. The method according to claim 1, characterized in that The starting point of the simulation trajectory is generated in the following way: According to the historical trajectory data of the target area, the population distribution characteristics in the preset starting period are counted; Determine the distribution probability of the starting point of the simulation trajectory in the target area according to the population distribution characteristics, wherein the distribution probability of the starting point in a grid is positively correlated with the population density in the grid; The starting point of the simulation trajectory is generated according to the distribution probability of the starting point in the target area.

7. The method according to any one of claims 1 to 6, characterized in that: The step of generating a virtual trajectory using the pre-trained generator includes: After generating some position points of a virtual trajectory using the pre-trained generator, the remaining position points of the virtual trajectory are generated by random sampling to obtain the virtual trajectory.

8. A system for generating a moving trajectory, characterized in that: include: A data acquisition module, used to acquire pre-collected historical trajectory data of multiple individuals moving in a target area, wherein the historical trajectory data at least includes the geographic location and time information of the individuals; A pre-training module, used for inputting the data of the historical trajectory into a generator, predicting the next position point in the historical trajectory through the generator, and pre-training the generator; and, inputting the historical trajectory and the perturbation trajectory into a discriminator, and pre-training the discriminator by training the discriminator to distinguish the perturbation trajectory, wherein the perturbation trajectory is obtained by perturbing some position points in the historical trajectory; The formal training module is used to repeatedly perform the following iterative processing until a preset convergence condition is reached: using the pre-trained generator to generate a virtual trajectory; inputting the virtual trajectory into the pre-trained discriminator, using the discriminator to discriminate the virtual trajectory and the sampled historical trajectory to obtain a discrimination result; using a reinforcement learning algorithm to generate a gradient corresponding to the discrimination result and back-propagating it to the generator, and updating the generator; A trajectory simulation module is used to generate a simulation trajectory of the target area using the generator according to the starting point of the simulation trajectory. Wherein, the formal training module is also used for: Using the discriminator to distinguish the virtual trajectory from the sampled historical trajectory, and obtaining a first judgment result of the authenticity of the virtual trajectory; Calculating a sum of distances between all adjacent position points on the virtual track, and generating a second judgment result of the authenticity of the virtual track according to the sum, wherein the authenticity is negatively correlated with the sum; Calculating the difference in geographical location between relevant time points on the virtual trajectory, and generating a third judgment result of the authenticity of the virtual trajectory according to the difference, wherein the authenticity is negatively correlated with the difference; the relevant time points include the same periodic time points and different time points set according to the periodicity of the individual movement law; A determination result of the authenticity of the virtual trajectory is generated according to the first determination result, the second determination result and the third determination result.

9. The system according to claim 8, characterized in that The generator includes an embedding layer, a bidirectional LSTM network and multiple linear layers, wherein: The embedding layer generates a feature vector sequence for characterizing each historical trajectory according to the data of the historical trajectory, wherein the feature vector sequence includes a preset number of position point feature vectors; The bidirectional LSTM network receives the feature vector sequence generated by the embedding layer and generates a vector representation of the next position point; The multiple linear layers map the vector representation of the next position obtained by the bidirectional LSTM network into the geographic location of the next position point.

10. The system according to claim 8, characterized in that The trajectory simulation module is further used to generate the starting point of the simulation trajectory in the following manner: According to the historical trajectory data of the target area, the population distribution characteristics in the preset starting period are counted; Determine the distribution probability of the starting point of the simulation trajectory in the target area according to the population distribution characteristics, wherein the distribution probability of the starting point in a grid is positively correlated with the population density in the grid; The starting point of the simulation trajectory is generated according to the distribution probability of the starting point in the target area.

11. The system according to any one of claims 8 to 10, characterized in that The formal training module is further used to generate some position points of the virtual trajectory by using the generator obtained by pre-training, and then generate the remaining position points of the virtual trajectory by random sampling to obtain the virtual trajectory.

Images