CN108108831B - Destination prediction method and device - Google Patents

Abstract

The invention relates to the field of intelligent transportation, in particular to a destination prediction method and a destination prediction device, wherein the method comprises the following steps: presetting a destination prediction model, wherein the destination prediction model respectively records the target transition probability between every two historical position points in each historical driving path of the vehicle in each set time period, determining a time period corresponding to the current driving track by monitoring the current driving track of the vehicle, respectively calculating the prediction probability between the current driving track and each historical position point in each historical driving track by adopting a preset destination prediction model based on the time period corresponding to the current driving track, determining the historical position point of which the corresponding prediction probability meets the preset conditions as a predicted destination, therefore, time factors are considered when a destination prediction model is established or destination prediction is carried out, accuracy of destination prediction is effectively improved, and driving experience of a user is improved.

Description

Destination prediction method and device

Technical Field

The invention relates to the field of intelligent transportation, in particular to a destination prediction method and a destination prediction device.

Background

With the development of internet technology, Location Based Service (LBS) applications are emerging continuously, so as to provide services such as Location positioning, route query, and historical driving track presentation for the user driving for travel.

The driving travel route of the user is influenced by the habit of own behavior and is also restricted by external conditions, so that the driving travel destination of the user has a certain rule, and the driving travel route of the user is highly likely to reach certain specific areas, such as homes, companies, shopping centers, restaurants, movie theaters and the like.

In the prior art, a terminal can acquire a historical driving track of a user driving trip through an application program of a location service, and further, a destination of the user driving trip can be predicted based on the historical driving track of the user driving trip.

In the prior art, a method for predicting a destination where a user drives to travel by a terminal is as follows:

the terminal obtains the current driving track of the vehicle and each historical driving track, compares the current driving track with each historical driving track, and if a certain historical driving track is successfully compared with the current driving track, determines that the destination of the current vehicle is the destination of the historical driving track. Further, if the plurality of historical driving tracks are compared with the current driving track part successfully, calculating the probability of reaching each destination corresponding to the plurality of historical driving tracks according to the plurality of historical driving tracks, and determining the destination of the current vehicle as the destination with the highest probability.

However, the driving travel track of the user is greatly influenced by habits, purposes, requirements and the like of the user, and the difference between a route point and a destination is not considered when comparing the tracks, and actually, the destination in the historical driving track is more indicative of the current driving travel destination of the user than the route point, and the time factor in the historical driving track is not considered. Further, since the current driving track of the vehicle is constantly changing, when the historical driving track that is successfully compared with the current driving track cannot be found in the database of the historical driving track, the destination where the user drives the vehicle for traveling cannot be predicted.

In view of the above, a new destination prediction method is needed to overcome the above-mentioned drawbacks.

Disclosure of Invention

The embodiment of the invention provides a destination prediction method and a destination prediction device, which are used for predicting a destination of a user driving according to a current driving track and a historical driving track of a vehicle driven by the user, and solve the problem that the accuracy of destination prediction is too low when the destination is predicted according to the historical driving track in the prior art.

The embodiment of the invention provides the following specific technical scheme:

a destination prediction method, comprising:

monitoring the current driving track of a vehicle, and determining a time period corresponding to the current driving track;

acquiring a preset destination prediction model, wherein the destination prediction model respectively records the target transition probability between every two historical position points in each historical driving track of the vehicle in each set time period, and the set time period is divided based on a natural day;

respectively calculating the prediction probability between the current driving track and each historical position point in each historical driving track by adopting the destination prediction model based on the time period corresponding to the current driving track;

and determining the historical position point of which the corresponding prediction probability meets the preset condition as a prediction destination.

Optionally, before monitoring the current driving trajectory of the vehicle, the method further includes: generating the destination prediction model specifically includes:

obtaining each historical driving track of the vehicle;

after the obtained historical driving tracks of the vehicle are projected to a road network, a target transfer network is obtained, wherein the target transfer network is used for indicating the relative motion between historical position points in the historical driving tracks;

establishing a first transfer tensor based on a first dimension, a second dimension and a third dimension, wherein the first dimension represents each historical position point before the transfer occurs in the target transfer network, the second dimension represents each historical position point after the transfer occurs in the target transfer network, the third dimension represents a set of set time periods, and the first transfer tensor is used for indicating a first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period;

and processing the first transfer tensor to obtain a target transfer tensor which accords with a preset rule, wherein the target transfer tensor respectively records the target transfer probability between every two historical position points in each set time period in each historical driving track of the vehicle.

Optionally, after the obtained historical driving trajectories of the vehicle are projected to a road network, obtaining a target transfer network includes:

after each obtained historical driving track of the vehicle is projected to a road network, a first transfer network is established in the projected road network;

identifying each historical position point in each historical driving track contained in the first transfer network as a transient historical position point, and creating a mirror image absorption historical position point for each transient historical position point, wherein the transient historical position point represents a non-termination point in the historical driving track, and the absorption historical position point represents a termination point in the historical driving track;

and establishing a corresponding target transfer network based on the relative motion between each transient historical position point and each absorption historical position point in each historical driving track.

Optionally, establishing the first transfer tensor based on the first dimension, the second dimension and the third dimension includes:

dividing the target transfer network into a forward target transfer network and a recent target transfer network according to a preset time division point, wherein the forward target transfer network comprises a historical position point generated in the target transfer network before the time division point, and the recent target transfer network comprises a historical position point generated in the target transfer network after the time division point;

establishing a long-term first transfer tensor according to the long-term target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the long-term first transfer tensor represents each historical position point before transfer occurs in the long-term target transfer network, the second dimension of the long-term first transfer tensor represents each historical position point after transfer occurs in the long-term target transfer network, the third dimension of the long-term first transfer tensor represents the set time periods, and the long-term first transfer tensor is used for indicating a long-term first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period before the time division point;

establishing a recent first transfer tensor according to the recent target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the recent first transfer tensor represents each historical position point before transfer occurs in the recent target transfer network, the second dimension of the recent first transfer tensor represents each historical position point after transfer occurs in the recent target transfer network, the third dimension of the recent first transfer tensor represents the set of set time periods, and the recent first transfer tensor is used for indicating a recent first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period after the time division point;

forming the first transfer tensor using the future first transfer tensor and the near first transfer tensor.

Optionally, processing the first transfer tensor to obtain a target transfer tensor according with a preset rule includes:

respectively acquiring the number of vehicles appearing in each set time period in the target transfer network and the number of specified interest points located in each preset range of historical position points in the target transfer network to form a central tensor;

filling the first transfer tensor based on the obtained central tensor to obtain a second transfer tensor;

performing optimization training on the obtained second transfer tensor to obtain each optimization training result;

and determining the optimized training results which accord with the preset rules as the target transfer tensor based on the optimized training results.

Optionally, the performing optimization training on the obtained second transfer tensor to obtain each optimization training result includes:

calculating a loss value between the first transfer tensor and the second transfer tensor;

and fitting the second transfer tensor to obtain each optimization training result based on the loss values and each regularization item, wherein each regularization item is generated by the set time period set, each historical position point in each historical driving track of the vehicle, the target transfer network, the central tensor and a preset geographic feature factor.

Optionally, calculating, by using the destination prediction model, a prediction probability between the current trajectory and any one of historical position points in each historical trajectory based on the time period corresponding to the current trajectory, including:

determining an initial position point, a current stopping position point and each passing position point between the initial position point and the current stopping position point of the current driving track;

determining target transition probabilities between the starting position point, the current stopping position point and each passing position point of the current driving track and any one historical position point in each historical driving track in a time period corresponding to the current driving track in the destination prediction model;

and determining the prediction probability between the current driving track and any one historical position point in each historical driving track based on each obtained target transition probability.

A destination prediction apparatus comprising:

the monitoring unit is used for monitoring the current driving track of the vehicle and determining the time period corresponding to the current driving track;

the system comprises an acquisition unit, a storage unit and a processing unit, wherein the acquisition unit is used for acquiring a preset destination prediction model, and the destination prediction model records the target transition probability between every two historical position points in each historical driving track of the vehicle in each set time period, wherein the set time period is divided based on a natural day;

the calculation unit is used for calculating the prediction probability between the current driving track and each historical position point in each historical driving track by adopting the destination prediction model based on the time period corresponding to the current driving track;

and the determining unit is used for determining the historical position point of which the corresponding prediction probability meets the preset condition as the prediction destination.

Optionally, the apparatus further comprises: a model building unit;

the model establishing unit is configured to generate the destination prediction model before monitoring the current driving trajectory of the vehicle, and specifically includes:

obtaining each historical driving track of the vehicle;

after the obtained historical driving tracks of the vehicle are projected to a road network, a target transfer network is obtained, wherein the target transfer network is used for indicating the relative motion between historical position points in the historical driving tracks;

establishing a first transfer tensor based on a first dimension, a second dimension and a third dimension, wherein the first dimension represents each historical position point before the transfer occurs in the target transfer network, the second dimension represents each historical position point after the transfer occurs in the target transfer network, the third dimension represents a set of set time periods, and the first transfer tensor is used for indicating a first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period;

and processing the first transfer tensor to obtain a target transfer tensor which accords with a preset rule, wherein the target transfer tensor respectively records the target transfer probability between every two historical position points in each set time period in each historical driving track of the vehicle.

Optionally, after the obtained historical driving trajectories of the vehicle are projected to a road network, when a target transfer network is obtained, the model establishing unit is configured to:

after each obtained historical driving track of the vehicle is projected to a road network, a first transfer network is established in the projected road network;

identifying each historical position point in each historical driving track contained in the first transfer network as a transient historical position point, and creating a mirror image absorption historical position point for each transient historical position point, wherein the transient historical position point represents a non-termination point in the historical driving track, and the absorption historical position point represents a termination point in the historical driving track;

and establishing a corresponding target transfer network based on the relative motion between each transient historical position point and each absorption historical position point in each historical driving track.

Optionally, when the first transfer tensor is established based on the first dimension, the second dimension, and the third dimension, the model establishing unit is configured to:

dividing the target transfer network into a forward target transfer network and a recent target transfer network according to a preset time division point, wherein the forward target transfer network comprises a historical position point generated in the target transfer network before the time division point, and the recent target transfer network comprises a historical position point generated in the target transfer network after the time division point;

establishing a long-term first transfer tensor according to the long-term target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the long-term first transfer tensor represents each historical position point before transfer occurs in the long-term target transfer network, the second dimension of the long-term first transfer tensor represents each historical position point after transfer occurs in the long-term target transfer network, the third dimension of the long-term first transfer tensor represents the set time periods, and the long-term first transfer tensor is used for indicating a long-term first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period before the time division point;

establishing a recent first transfer tensor according to the recent target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the recent first transfer tensor represents each historical position point before transfer occurs in the recent target transfer network, the second dimension of the recent first transfer tensor represents each historical position point after transfer occurs in the recent target transfer network, the third dimension of the recent first transfer tensor represents the set of set time periods, and the recent first transfer tensor is used for indicating a recent first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period after the time division point;

forming the first transfer tensor using the future first transfer tensor and the near first transfer tensor.

Optionally, when the first transfer tensor is processed to obtain a target transfer tensor conforming to a preset rule, the model establishing unit is configured to:

respectively acquiring the number of vehicles appearing in each set time period in the target transfer network and the number of specified interest points located in each preset range of historical position points in the target transfer network to form a central tensor;

filling the first transfer tensor based on the obtained central tensor to obtain a second transfer tensor;

performing optimization training on the obtained second transfer tensor to obtain each optimization training result;

and determining the optimized training results which accord with the preset rules as the target transfer tensor based on the optimized training results.

Optionally, when performing optimization training on the obtained second transfer tensor to obtain each optimization training result, the model establishing unit is configured to:

calculating a loss value between the first transfer tensor and the second transfer tensor;

and fitting the second transfer tensor to obtain each optimization training result based on the loss values and each regularization item, wherein each regularization item is generated by the set time period set, each historical position point in each historical driving track of the vehicle, the target transfer network, the central tensor and a preset geographic feature factor.

Optionally, when the prediction probability between the current driving trajectory and any one of the historical position points in each historical driving trajectory is calculated by using the destination prediction model based on the time period corresponding to the current driving trajectory, the calculating unit is configured to:

determining an initial position point, a current stopping position point and each passing position point between the initial position point and the current stopping position point of the current driving track;

determining target transition probabilities between the starting position point, the current stopping position point and each passing position point of the current driving track and any one historical position point in each historical driving track in a time period corresponding to the current driving track in the destination prediction model;

and determining the prediction probability between the current driving track and any one historical position point in each historical driving track based on each obtained target transition probability.

The invention has the following beneficial effects:

in the embodiment of the invention, a destination prediction model is established in advance based on each historical driving track of a vehicle and the difference of time factors, end points and non-end points of each historical driving track, wherein the destination prediction model records the target transition probability between every two historical position points in each historical driving track of the vehicle in each set time period, when the current driving track of the vehicle is monitored, the time period corresponding to the current driving track is determined, the preset destination prediction model is adopted based on the time periods corresponding to the current driving track and the current driving track to calculate the prediction probability between the current driving track and each historical position point in each historical driving track respectively, and the historical position point of which the corresponding prediction probability meets the preset condition is determined as the predicted destination, therefore, the difference between the non-end point and the end point in the historical behavior track can be considered, and the time factor of the historical behavior track can be fully considered, so that the established destination prediction model is more accurate, the accuracy of destination prediction is improved, and the driving experience of a user is further improved.

Drawings

FIG. 1 is a flow chart of a method for modeling a destination prediction in an embodiment of the invention;

FIG. 2 is a diagram illustrating a first transfer tensor in an embodiment of the present invention;

FIG. 3 is a flow chart of a method of predicting a destination in an embodiment of the present invention;

fig. 4 is a schematic structural diagram of an apparatus for predicting a destination according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In order to improve the accuracy of destination prediction, in the embodiment of the invention, a method for predicting a destination is designed, the method comprises the steps of presetting a destination prediction model, wherein a target transition probability between every two historical position points in each historical driving track of a vehicle in each set time period is recorded in the destination prediction model, determining a time period corresponding to the current driving track by monitoring the current driving track of the vehicle, adopting the preset destination prediction model based on the time period corresponding to the current driving track to respectively calculate the prediction probability between the current driving track and each historical position point in each historical driving track, and determining the historical position point of which the corresponding prediction probability meets preset conditions as a predicted destination.

In the embodiment of the invention, a terminal is utilized to monitor a vehicle driven by a user, the driving tracks of the vehicle are recorded at any time, a plurality of historical driving tracks can be recorded by the terminal after a period of time, and a destination prediction model can be established by the terminal based on the historical driving tracks, so that the terminal can predict the destination ahead in time according to the current driving track of the vehicle in the subsequent driving process of the vehicle, wherein the terminal can be a vehicle-mounted terminal, an intelligent mobile phone, an intelligent bracelet, a tablet computer and the like.

Specifically, referring to fig. 1, a specific process of the terminal establishing the destination prediction model is as follows:

step 100: the terminal acquires historical driving tracks of a vehicle driven by a user and projects each acquired historical driving track to a road network.

Specifically, each historical driving track of the vehicle driven by the user is obtained by the terminal through a Global Positioning System (GPS), and because the GPS provides coordinate data points including longitude, latitude and other information, each historical driving track obtained by the terminal is composed of a series of coordinate data points, and a specific driving route of the vehicle cannot be accurately and intuitively reflected.

Further, the terminal projects each obtained historical driving track into the road network according to the corresponding coordinate in the coordinate data points contained in each historical driving track, so as to obtain the specific driving route of each historical driving track of the vehicle in the road network.

In the embodiment of the present invention, a map matching algorithm (e.g., a semi-deterministic method, a probabilistic statistical algorithm, a pattern recognition algorithm, etc.) is preferably used for projection, but in a specific implementation, the present invention is not limited thereto.

Step 101: and the terminal establishes a first transfer network in the projected road network, wherein the first transfer network is used for indicating the relative motion between each historical position point in each historical driving track before the conversion.

Specifically, the terminal determines any one road intersection (such as a crossroad and a T-junction) passing through in the specific driving route of the vehicle as a historical position point according to the obtained specific driving route of each historical driving track of the vehicle in the road network, determines a continuous track connecting any two historical position points as a transfer side, and the transfer sides between all the historical position points and any two historical position points form a first transfer network which is used for indicating the relative movement between the historical position points in each historical driving track before conversion.

In the embodiment of the present invention, a preferred expression manner of the first transition network is a directed graph, which is denoted by G ' (V ', E '), where V ' represents a group of history location points, E ' represents a set of transition edges connecting any two history location points, and all the transition edges are vectors and have directions.

For example, suppose V' contains 5 historical location points, respectively V₁、v₂、v₃、v₄、v₅If v is₁To v₃There is a continuous track e between, and v₃To v₄There is a continuous trace f between them, then E' contains 2 transition edges, E and f, respectively.

Step 102: and the terminal converts the established first transfer network into a target transfer network, wherein the target transfer network is used for indicating the relative motion among the historical position points in the historical driving tracks after conversion.

Specifically, since the attributes of the historical location points are not distinguished from a group of historical location points included in the first transfer network established by the terminal, that is, whether the historical location points are termination points or non-termination points is not distinguished, and actually, when destination prediction is performed, the termination points are more referential than the non-termination points, the attributes of the historical location points need to be distinguished to ensure that the difference between the termination points and the non-termination points is fully utilized when the destination prediction is performed subsequently, so that the prediction result is more accurate.

In the embodiment of the invention, the preferable mode is that a special type of absorption state in a mirror image absorption Markov chain model is adopted, a distinction is made between a set of historical location points contained in the first transit network described above, since in the special type of sink state of the mirror-absorbing markov chain model, the nodes that are not lossable are called sink nodes, the non-sink nodes are called transient nodes, whereas, in reality, the end point belongs to a point that is not losable, and therefore, the history position point expressed as the end point is marked as the absorption history position point, the history position point expressed as the non-end point is marked as the transient history position point, and accordingly, the set of transition edges of the first transition network also follows the transformation of the historical location points, including all transition edges connecting the transient historical location points and all transition edges connecting the transient historical location points to the sink historical location points.

However, the unidirectional distinction of each historical position point of the first transfer network into the absorption historical position point or the transient historical position point cannot avoid the situation that one transfer node exists in a certain historical driving track, not only is a non-termination point of the certain historical driving track, but also is a termination point of another historical driving track, that is, each historical position point may be a termination point or a non-termination point, therefore, in the embodiment of the present invention, a preferred implementation manner is to create a corresponding mirror image node for each historical position point in the first transfer network by using a mirror image concept in a mirror image absorption markov chain model, wherein the mirror image node is used for distinguishing the absorption state and the transient state of the historical position point, that is, any historical position point which is not originally used as an attribute distinction, the method comprises the steps of distinguishing a transient historical position point and an absorption historical position point, establishing a corresponding target transfer network based on relative motion between each transient historical position point and each absorption historical position point in each historical driving track, wherein correspondingly, a transfer edge set in the established target transfer network comprises all transfer edges connecting the transient historical position points and all transfer edges connecting the transient historical position points to the absorption historical position points.

For example, assuming that the first transfer network is represented as a directed graph G ' (V ', E '), where V is represented as n historical location points and E is represented as m transfer edges, if the target transfer network is represented as a directed graph G (V, E), then, according to the conversion relationship, V will have n transient historical location points and n absorption historical location points, and if it is continuously assumed that there is only one historical driving track in the target transfer network, and the historical driving track is represented by V₀To v_n-1N transient history location points, then, v₀To v_n-1For a total of n transient history location points, create respective corresponding mirror nodes as absorption history location points, e.g., assuming transient history location point v_iFor the ith transient history location point of the n transient history location points, and in the target transfer network, for v_iCreating a mirror node v_2i-1By analogy, v is₀To v_n-1The mirror nodes of n transient historical position points are respectively v_nTo v_2n-1。

At this point, the terminal converts the established first transition network into a target transition network.

Step 103: the terminal establishes a first transfer tensor according to a target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first transfer tensor is used for indicating a first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period, and the set time period is divided based on a natural day.

Specifically, since the driving travel of the user may be related to the behavior habit of the user, there may be a certain regularity in the travel at different time periods, for example, for a user of a working family, the early peak period of the working day is usually from the family to the company, and the late peak period is usually from the company to the house, so in order to make the destination prediction model more accurate, it is necessary to pre-process the obtained second transfer network in combination with the occurrence time of each historical behavior trajectory.

Further, a first transition tensor is established based on a first dimension, a second dimension and a third dimension, wherein the first dimension represents each historical position point before transition occurs in the target transition network, the second dimension represents each historical position point after transition occurs in the target transition network, the third dimension represents a set of set time periods, and the first transition tensor is used for indicating a first transition probability between every two historical position points in each historical driving track of the vehicle in each set time period.

In a preferred embodiment of the present invention, the set of time periods is composed of a plurality of time periods divided by a natural day.

For example, still taking the example in step 100 and step 101, referring to fig. 2, the first transfer tensor is denoted as a ', and a' ═ V₁，V₂T'), wherein V₁N transient history location points and n absorption history location points, V, before the transition in the target transition network₂The target transfer network represents n transient history location points and n absorption history location points after the transfer, and if one natural day is divided into h set time periods, that is, the set time period set T 'includes h set time periods T, and if one entry a' (i, j, k) 'of the first transfer tensor a' is assumed to be e, the target transfer network represents the history location point v_iAnd v_jAt a set time period t_k(e.g., 10:00-10:30) with a transition probability of e.

Furthermore, as the historical driving tracks closer to the current occurrence time are closer to the current driving habits of the user, in order to make the destination prediction model more accurate, the historical driving tracks are divided into a long-term historical driving track and a short-term historical driving track according to a preset time division point, that is, the target transfer network is divided into a long-term target transfer network and a short-term target transfer network, and correspondingly, the first transfer tensor can also be divided into a long-term first transfer tensor and a short-term first transfer tensor.

Specifically, the construction processes of the future first transfer tensor and the near first transfer tensor are respectively as follows:

and creating a long-term first transfer tensor according to the long-term object transfer network, wherein the first dimension of the long-term first transfer tensor represents each historical position point before the transfer occurs in the long-term object transfer network, the second dimension of the long-term first transfer tensor represents each historical position point after the transfer occurs in the long-term object transfer network, the third dimension of the long-term first transfer tensor represents the set time period, and the long-term first transfer tensor is used for indicating the long-term first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period before the time division point.

And establishing a recent first transfer tensor according to the recent target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the recent first transfer tensor represents each historical position point before the transfer occurs in the recent target transfer network, the second dimension of the recent first transfer tensor represents each historical position point after the transfer occurs in the recent target transfer network, the third dimension of the recent first transfer tensor represents the set of set time periods, and the recent first transfer tensor is used for indicating the recent first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period after the time division point.

And composing the first transfer tensor by using the long-term first transfer tensor and the short-term first transfer tensor.

In the embodiment of the present invention, the preferable time division point is 20%, that is, the historical driving trajectories are sorted according to the time sequence of the occurrence of the historical driving trajectories, the historical driving trajectory close to the current time is before, the later historical driving trajectory is after, the target transfer network composed of the historical driving trajectories of the previous 20% is represented as a recent target transfer network, and the target transfer network composed of the historical driving trajectories of the remaining 80% is represented as a long-term target transfer network.

For example, assume that each of the historical driving trajectories of the vehicle is recordedR, and dividing each historical driving track R into long-term historical driving tracks R_mAnd recent historical driving track R_cThen, the obtained first transfer tensor A 'may be divided into a forward-phase first transfer tensor A'_mAnd a recent first transition tensor A'_cIn the above example, if the set time period set T' includes h set time periods when the forward first transfer tensor and the near first transfer tensor are divided, the set time period set T includes 2h set time periods T after the forward first transfer tensor and the near first transfer tensor are divided.

To this end, the terminal completes the construction of the first transfer tensor a', wherein the constructed first transfer tensor can be divided into the future first transfer tensor and the near-term first transfer tensor.

Step 104: the terminal acquires the central tensor.

Specifically, since the effective transition probability obtained from the first transition tensor is very small during actual calculation, and there is a data sparsity problem, in order to solve the data sparsity problem, the number of vehicles appearing in each set time period in the target transition network and the number of specified points of interest located in a preset range of each historical position point in the target transition network are respectively obtained to form a central tensor.

Further, the number of vehicles appearing in each set time period in the target transfer network represents the correlation between different set time periods captured under coarse-grained traffic conditions, is a statistical value and can be directly obtained from the outside, the number of specified points of interest located in a preset range of any one historical location point in the target transfer network represents the number of specified points of interest appearing near any one historical location point, wherein the specified points of interest may be a shopping mall, a pharmacy, a hospital, a school, and the like, the vicinity of the specified points of interest may also be specifically set to be located within 100 meters of the historical location point, and the adjustment is performed according to actual needs, which is not limited herein.

Step 105: the terminal fills the first transfer tensor based on the obtained central tensor to obtain a second transfer tensor.

Specifically, after the terminal obtains the central tensor, the terminal respectively uses the number of vehicles appearing in each set time period in the target transfer network and the number of specified interest points located in each preset range of historical position points in the target transfer network to form the central tensor, and fills the first transfer tensor to obtain a second transfer tensor.

Assuming that the central tensor is represented as S, the number of vehicles appearing in each set time period in the target transfer network can be represented by a matrix X and can be used for dense filling, the number of specified interest points in each preset range of historical position points in the target transfer network can be represented by a matrix Y, and through actual checking, the error of each transfer probability contained in the first transfer tensor can be reduced by multiplying the factors decomposed by the central tensor S.

In the embodiment of the present invention, preferably, the second transfer tensor can be obtained by using the following formula:

A”＝S×V₁V₁×V₂V₂×TT

wherein A' represents a second transfer tensor, V₁N transient history location points and n absorption history location points, V, before the transition in the target transition network₂N transient history position points and n absorption history position points after the transition in the target transition network are represented, T represents a set of set time periods of the first transition tensor, and S represents the central tensor.

At this point, the terminal completes the filling of the first transfer tensor, and obtains a second transfer tensor a ″.

Step 106: and the terminal carries out optimization training on the obtained second transfer tensor to obtain each optimization training result, screens out the optimization training results which accord with a preset rule as a target transfer tensor, and constructs a destination prediction model based on the target transfer tensor.

Specifically, a loss value between the first transfer tensor and the second transfer tensor is calculated, the second transfer tensor is fitted based on the loss value and each regularization term generated by the set time period set, each historical position point in each historical driving track of the vehicle, the target transfer network, the central tensor, and a preset geographic feature factor to obtain each optimized training result, and the optimized training result conforming to a preset rule is determined as a target transfer tensor based on each optimized training result.

In the embodiment of the present invention, preferably, the following algorithm (gradient descent algorithm) may be adopted for the optimization training to obtain the target transfer tensor:

wherein, omega (S, V)₁,V₂The specific calculation of T, G, F) is as follows:

where A 'and A' are the first and second transfer tensors, respectively, and L (A ', A') is a loss function of A 'and A', a loss value between A 'and A' can be calculated, S represents the central tensor, and V represents the central tensor₁N transient history location points and n absorption history location points, V, before the transition in the target transition network₂N transient history location points and n absorption history location points after the transfer in the target transfer network, T a set of time periods of the first transfer network, G a number of grids, F a predetermined geographic feature factor (e.g., a dimension of a geographic feature), λ₄Is a meta-parameter to control the contribution of the different parts, omega (S, V) above₁,V₂T, G, F) are regularization terms that prevent overfitting.

In the optimization training process, a plurality of optimization training results are obtained, and in the embodiment of the present invention, optimization training is performed according to a gradient descent algorithm, so that according to the gradient descent algorithm, a smallest one of the optimization training results is selected as a target transfer tensor a, where the target transfer tensor a records a target transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period, and each two historical position points include a transient historical position point and an absorption historical position point, and a calculation process of the specific transfer tensor a is as follows:

specifically, the obtained target transfer tensor is divided based on the time dimension to obtain each set time period unit transfer matrix, wherein each element P in each set time period unit transfer matrix_k(i, j) denotes at t_kAt a set time period, from a historical position point v_iTo historical location point v_jTransition probability (v) of_iAnd v_jEither transient history location points or absorption history location points).

In the embodiment of the present invention, it is assumed that one unit of the set time period (set time period t) exists_k) Transition matrix P_kAll the absorption history position points are represented as Z, all the transient history position points are represented as W, and preferably, the time period t can be set in the form recorded in Table 1_kEach transfer matrix P of_kAnd (3) carrying out recombination:

TABLE 1

Wherein, I is an n-dimensional unit matrix, 0 is an n-dimensional zero matrix, Q is an n-dimensional matrix representing the transition probability between transient nodes, and S is an n-dimensional matrix representing the transition probability from the transient node to the absorbing node.

In the embodiment of the present invention, preferably, a unit of the set time period (assuming that the set time period is t) can be obtained by the following formula_k) Transition matrix P_k：

Specifically, the following are shown:

when v is_iIs v₀To v_n-1One of the historical location points, v_jIs v₀To v_n-1Of (i.e., v) is determined_iAnd v_jAll transient historical location points), v_iTo v_jHas a transition probability of P_Qk(ii) one of the elements of (i, j);

when v is_iIs v₀To v_n-1One of the historical location points, v_jIs v_nTo v_2n-1Of (i.e., v) is determined_iFor transient historical location points, v_jTo absorb historical location points), v_iTo v_jHas a transition probability of P_Sk(ii) one of the elements of (i, j);

when v is_iIs v_nTo v_2n-1One of the historical location points, v_jIs v₀To v_n-1Of (i.e., v) is determined_iTo absorb the historical location points, v_jA transient historical location point), v_iTo v_jThe transition probability between is one of the elements of the 0 matrix (the absorption history location point cannot be transferred to the transient history location point);

when v is_iIs v_nTo v_2n-1One of the historical location points, v_jIs v_nTo v_2n-1Of (i.e., v) is determined_iAnd v_jAll absorption historical location points), v_iTo v_jThe transition probability between is one of the elements of the identity matrix.

Specifically, in the embodiment of the present invention, preferably, P can be obtained by using the following formula_Qk(i, j) setting a time period t_kTransition probability between every two transient history position points:

specifically, in the embodiment of the present invention, preferably, P can be obtained by using the following formula_Sk(i, j) setting a time period t_kTransition probability between each transient history position point to the absorption history position point:

wherein, in the two calculation formulas, | R_m(i, k) | denotes that the time period t is set_kAll passes v_iThe number of long-term historical driving trajectories; | R_c(i, k) | denotes that the time period t is set_kAll passes v_iThe number of recent historical driving trajectories; | R_m(i, j, k) | represents that the time period t is set_kAll passes v_iTo v_jAnd v is_iTo v_jThe number of long-term historical driving tracks without a historical position point exists; | R_c(i, j, k) | represents that the time period t is set_kAll passes v_iTo v_jAnd v is_iTo v_jThe number of recent historical driving tracks of a historical position point does not exist;

represents | R_mIn all long-term historical driving tracks represented by (i, j, k) |, the termination point is v_jThe number of long-term historical driving trajectories;

represents | R_cIn all recent historical driving paths represented by (i, j, k) |, all termination points are v_jThe number of recent historical trajectories.

The above calculation process is to set the time period t_kFor example, for the set of set time periods T', any one set time period can be calculated according to the set time period T_kObtaining each of the vehicles in each of the set time periodsAnd (4) target transition probability between every two historical position points in the historical driving track.

In this embodiment, a target transition tensor is obtained through the above steps, and a destination prediction model is built based on the target transition tensor, where the target transition tensor records a target transition probability between every two historical position points (which may be transient historical position points or absorption historical position points) in each historical driving track of the vehicle in each set time period, that is, the destination prediction model records a target transition probability between every two historical position points in each historical driving track of the vehicle in each set time period.

In the running process of the vehicle, the terminal predicts the destination ahead in time by monitoring the current driving track of the vehicle according to the established destination prediction model.

Specifically, referring to fig. 3, the specific process of the terminal predicting the destination ahead is as follows:

step 300: the terminal monitors the current driving track of the vehicle and determines the time period corresponding to the current driving track.

Specifically, the terminal monitors a current driving track of the vehicle and determines a time period corresponding to the current driving track in the running process of the vehicle, wherein the current driving track comprises a group of current position points, and determines each time period of the group of current position points.

For example, assume that the current trajectory of the vehicle monitored by the terminal is R_pDetermining R_pContains a set of current location points of { v_s，v_i，…，v_i+j，v_cIf v is_sAnd v_cRespectively the current driving track R_pAnd a current stopping position point, and v_i、v_i+jAnd v_iV between_i+jAll current position points of (2) are respectively the current driving track R_pDetermines the set of current location points { v }_s，v_i，…，v_i+j，v_cThe respective periods of occurrence correspond to { t } respectively_s，t_i，…，t_i+j，t_c}。

Step 301: the terminal obtains a preset destination prediction model.

Specifically, the terminal determines, according to a time period corresponding to the acquired current driving track, a target transition probability between the starting position point, the current stopping position point and each passing position point of the current driving track and any one of the historical position points in the historical driving tracks in the time period corresponding to the current driving track in the destination prediction model.

For example, assume that a historical position point v is associated with each of the historical driving paths_kIf the current driving track is still R in step 300_pThen, in the above destination prediction model, R is determined_pContaining a set of current location points v_s，v_i，…，v_i+j，v_cWithin the corresponding time periods, the historical position points v are respectively corresponding to_kTarget transition probabilities between.

Step 302: and the terminal respectively calculates the prediction probability between the current driving track and each historical position point in each historical driving track by adopting the acquired destination prediction model based on the time period corresponding to the current driving track.

Specifically, after the terminal determines the current driving track and the time period corresponding to the current driving track, the terminal determines the prediction probability between the current driving track and each historical position point in each historical driving track by adopting each target transition probability obtained in the destination prediction model.

In the embodiment of the present invention, a bayesian conditional probability calculation method is preferably used to calculate the prediction probability between the current trajectory and each historical position point in each historical trajectory.

For example, the vehicle still uses one historical position point v in each historical driving track_kFor example, in the Bayesian conditional probability calculation, the current driving is givenTrack, calculating the historical position point v under the current driving track_kIs the predicted probability P (v) of the destination_k|R_p) In the embodiment of the present invention, preferably, the historical position point v under the current driving track can be obtained by using the following formula_kIs the predicted probability P (v) of the destination_k|R_p)：

P(v_k|R_p)

＝P(v_k|v_s,v_i,...,v_i+j,v_c)

＝P(v_k|v_s)·P(v_k，v_s|v_i)·...·P(v_k，v_s，v_i,...,v_i+_j1|v_i+_j)P(v_k,v_s,v_i,...,v_i+_j|v_c)

＝B_s(s,k)·[B_i(i,k)·P_Qs(s,i)]·...·[B_i+j(i+j,k)·P_Q(i+j-1)(i+j-1,i+j)]·[B_c(c,k)·P_Q(i+j)(i+j,c)]

Wherein, B_s(s, k) represents a time period t set_sFrom transient historical location points v_sTo the absorption historical position point v_kAbsorption probability of (B)_i(i, k) represents a time period t set_iFrom transient historical location points v_iTo the absorption historical position point v_kAbsorption probability of P_Qs(s, i) represents a time period t set_sFrom transient historical location points v_sTo the transient historical location point v_iTarget transition probability of, B_i+j(i + j, k) represents a time period t set_i+jFrom transient historical location points v_i+jTo the absorption historical position point v_kAbsorption probability of P_Q(i+j-1)(i + j-1, i + j) represents the time period t set_i+j-1From transient historical location points v_i+j-1To the transient historical location point v_i+jTarget transition probability of, B_c(c, k) represents a time period t set_cFrom transient historical location points v_cTo the absorption historical position point v_kAbsorption probability of P_Q(i+j)(i + j, c) represents the time period t set_i+jFrom transient historical location points v_i+jTo the transient historical location point v_cTarget transition probability of (2).

Further, each absorption probability is obtained based on the target transition probability between the transient history location points and the absorption history location points_kRespective absorption probability of (2):

B_k＝H_kS_k＝(I-Q_k)^-1S_k

wherein Q is_kFor a set time period t in the transfer tensor A for the target_kProbability of target transfer between any two transient history location points, S_kFor a set time period t in the transfer tensor A for the target_kAnd target transition probability between any transient historical position point and any absorption historical position point.

B_kTo set a time period t_kB is the respective absorption probability, i.e. the absorption probability matrix, for each set time period, i.e. the absorption tensor, for the set time period t_kAn absorption probability B of_k(i, j) indicates that the time period t is set_kFrom transient historical location points v_iTo the absorption historical position point v_jThe probability of absorption of (c).

In the above calculation process, each of the transition probabilities of the target involved may be obtained in the destination prediction model in advance through step 302, and through the above calculation, it can be known that the historical position point v is located under the current driving track_kIs the predicted probability P (v) of the destination_k|R_p) Specific values of (a).

By analogy, under the current driving track, the prediction probability that other historical position points included in each historical driving track of the vehicle are the destination can also be obtained by adopting the calculation method.

And the terminal obtains the prediction probability between the current driving track and each historical position point in each historical driving track.

Step 303: and the terminal determines the historical position point of which the corresponding prediction probability meets the preset condition as a prediction destination.

Specifically, the terminal screens the obtained prediction probabilities, selects the corresponding historical position points of which the prediction probabilities meet the preset conditions and determines the historical position points as the prediction destinations.

Therefore, the terminal can predict the destination ahead in time by monitoring the current driving track of the vehicle according to the established destination prediction model.

Referring to fig. 4, in the embodiment of the present invention, the destination predicting device at least includes a monitoring unit 41, an obtaining unit 42, a calculating unit 43 and a determining unit 44, wherein,

the monitoring unit 41 is configured to monitor a current driving track of a vehicle, and determine a time period corresponding to the current driving track;

an obtaining unit 42, configured to obtain a preset destination prediction model, where a target transition probability between every two historical position points in each historical driving track of the vehicle in each set time period is recorded in the destination prediction model, where the set time period is divided based on one natural day;

a calculating unit 43, configured to calculate, based on a time period corresponding to the current driving trajectory, prediction probabilities between the current driving trajectory and each historical position point in each historical driving trajectory by using the destination prediction model;

and the determining unit 44 is used for determining the historical position point of which the corresponding prediction probability meets the preset condition as the prediction destination.

Optionally, the apparatus further comprises: a model building unit 45;

the model establishing unit 45 is configured to, before monitoring the current driving trajectory of the vehicle, generate the destination prediction model, which specifically includes:

obtaining each historical driving track of the vehicle;

after the obtained historical driving tracks of the vehicle are projected to a road network, a target transfer network is obtained, wherein the target transfer network is used for indicating the relative motion between historical position points in the historical driving tracks;

establishing a first transfer tensor based on a first dimension, a second dimension and a third dimension, wherein the first dimension represents each historical position point before the transfer occurs in the target transfer network, the second dimension represents each historical position point after the transfer occurs in the target transfer network, the third dimension represents a set of set time periods, and the first transfer tensor is used for indicating a first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period;

and processing the first transfer tensor to obtain a target transfer tensor which accords with a preset rule, wherein the target transfer tensor respectively records the target transfer probability between every two historical position points in each set time period in each historical driving track of the vehicle.

Optionally, after the obtained historical driving trajectories of the vehicle are projected to a road network, when a target transfer network is obtained, the model establishing unit 45 is configured to:

after each obtained historical driving track of the vehicle is projected to a road network, a first transfer network is established in the projected road network;

identifying each historical position point in each historical driving track contained in the first transfer network as a transient historical position point, and creating a mirror image absorption historical position point for each transient historical position point, wherein the transient historical position point represents a non-termination point in the historical driving track, and the absorption historical position point represents a termination point in the historical driving track;

and establishing a corresponding target transfer network based on the relative motion between each transient historical position point and each absorption historical position point in each historical driving track.

Optionally, when the first transfer tensor is established based on the first dimension, the second dimension and the third dimension, the model establishing unit 45 is configured to:

dividing the target transfer network into a forward target transfer network and a recent target transfer network according to a preset time division point, wherein the forward target transfer network comprises a historical position point generated in the target transfer network before the time division point, and the recent target transfer network comprises a historical position point generated in the target transfer network after the time division point;

establishing a long-term first transfer tensor according to the long-term target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the long-term first transfer tensor represents each historical position point before transfer occurs in the long-term target transfer network, the second dimension of the long-term first transfer tensor represents each historical position point after transfer occurs in the long-term target transfer network, the third dimension of the long-term first transfer tensor represents the set time periods, and the long-term first transfer tensor is used for indicating a long-term first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period before the time division point;

establishing a recent first transfer tensor according to the recent target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the recent first transfer tensor represents each historical position point before transfer occurs in the recent target transfer network, the second dimension of the recent first transfer tensor represents each historical position point after transfer occurs in the recent target transfer network, the third dimension of the recent first transfer tensor represents the set of set time periods, and the recent first transfer tensor is used for indicating a recent first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period after the time division point;

forming the first transfer tensor using the future first transfer tensor and the near first transfer tensor.

Optionally, when the first transfer tensor is processed to obtain a target transfer tensor conforming to a preset rule, the model establishing unit 45 is configured to:

respectively acquiring the number of vehicles appearing in each set time period in the target transfer network and the number of specified interest points located in each preset range of historical position points in the target transfer network to form a central tensor;

filling the first transfer tensor based on the obtained central tensor to obtain a second transfer tensor;

performing optimization training on the obtained second transfer tensor to obtain each optimization training result;

and determining the optimized training results which accord with the preset rules as the target transfer tensor based on the optimized training results.

Optionally, when performing optimization training on the obtained second transfer tensor to obtain each optimization training result, the model establishing unit 45 is configured to:

calculating a loss value between the first transfer tensor and the second transfer tensor;

and fitting the second transfer tensor to obtain each optimization training result based on the loss values and each regularization item, wherein each regularization item is generated by the set time period set, each historical position point in each historical driving track of the vehicle, the target transfer network, the central tensor and a preset geographic feature factor.

Optionally, when the prediction probability between the current driving trajectory and any one of the historical position points in each historical driving trajectory is calculated by using the destination prediction model based on the time period corresponding to the current driving trajectory, the calculating unit 43 is configured to:

determining an initial position point, a current stopping position point and each passing position point between the initial position point and the current stopping position point of the current driving track;

determining target transition probabilities between the starting position point, the current stopping position point and each passing position point of the current driving track and any one historical position point in each historical driving track in a time period corresponding to the current driving track in the destination prediction model;

and determining the prediction probability between the current driving track and any one historical position point in each historical driving track based on each obtained target transition probability.

In the embodiment of the invention, a destination prediction model is established in advance based on each historical driving track of a vehicle and the difference of time factors, end points and non-end points of each historical driving track, wherein the destination prediction model records the target transition probability between every two historical position points in each historical driving track of the vehicle in each set time period, when the current driving track of the vehicle is monitored, the time period corresponding to the current driving track is determined, the preset destination prediction model is adopted based on the time periods corresponding to the current driving track and the current driving track to calculate the prediction probability between the current driving track and each historical position point in each historical driving track respectively, and the historical position point of which the corresponding prediction probability meets the preset condition is determined as the predicted destination, therefore, the difference between the non-end point and the end point in the historical behavior track can be considered, and the time factor of the historical behavior track can be fully considered, so that the established destination prediction model is more accurate, the accuracy of destination prediction is improved, and the driving experience of a user is further improved.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.

It will be apparent to those skilled in the art that various modifications and variations can be made in the embodiments of the present invention without departing from the spirit or scope of the embodiments of the invention. Thus, if such modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to encompass such modifications and variations.

Claims (12)

1. A destination prediction method, comprising:

generating a destination prediction model, specifically comprising:

after each obtained historical driving track of the vehicle is projected to a road network, a first transfer network is established in the projected road network, and the first transfer network is used for indicating the relative motion among historical position points in each historical driving track before being converted;

identifying each historical position point in each historical driving track contained in the first transfer network as a transient historical position point, and creating a mirror image absorption historical position point for each transient historical position point, wherein the transient historical position point represents a non-termination point in the historical driving track, and the absorption historical position point represents a termination point in the historical driving track;

establishing a corresponding target transfer network based on the relative motion between each transient historical position point and each absorption historical position point in each historical driving track, wherein the target transfer network is used for indicating the relative motion between each historical position point in each historical driving track after conversion;

monitoring the current driving track of a vehicle, and determining a time period corresponding to the current driving track;

acquiring the destination prediction model, wherein the destination prediction model respectively records the target transition probability between every two historical position points in each historical driving track of the vehicle in each set time period, wherein the set time period is divided based on a natural day;

respectively calculating the prediction probability between the current driving track and each historical position point in each historical driving track by adopting the destination prediction model based on the time period corresponding to the current driving track;

and determining the historical position point of which the corresponding prediction probability meets the preset condition as a prediction destination.

2. The method of claim 1, wherein generating the destination predictive model further comprises, after establishing the corresponding target transition network:

establishing a first transfer tensor based on a first dimension, a second dimension and a third dimension, wherein the first dimension represents each historical position point before the transfer occurs in the target transfer network, the second dimension represents each historical position point after the transfer occurs in the target transfer network, the third dimension represents a set of set time periods, and the first transfer tensor is used for indicating a first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period;

and processing the first transfer tensor to obtain a target transfer tensor which accords with a preset rule, wherein the target transfer tensor respectively records the target transfer probability between every two historical position points in each set time period in each historical driving track of the vehicle.

3. The method of claim 2, wherein establishing the first transfer tensor based on the first dimension, the second dimension, and the third dimension comprises:

dividing the target transfer network into a forward target transfer network and a recent target transfer network according to a preset time division point, wherein the forward target transfer network comprises a historical position point generated in the target transfer network before the time division point, and the recent target transfer network comprises a historical position point generated in the target transfer network after the time division point;

establishing a long-term first transfer tensor according to the long-term target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the long-term first transfer tensor represents each historical position point before transfer occurs in the long-term target transfer network, the second dimension of the long-term first transfer tensor represents each historical position point after transfer occurs in the long-term target transfer network, the third dimension of the long-term first transfer tensor represents the set time periods, and the long-term first transfer tensor is used for indicating a long-term first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period before the time division point;

establishing a recent first transfer tensor according to the recent target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the recent first transfer tensor represents each historical position point before transfer occurs in the recent target transfer network, the second dimension of the recent first transfer tensor represents each historical position point after transfer occurs in the recent target transfer network, the third dimension of the recent first transfer tensor represents the set of set time periods, and the recent first transfer tensor is used for indicating a recent first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period after the time division point;

forming the first transfer tensor using the future first transfer tensor and the near first transfer tensor.

4. The method of claim 2, wherein processing the first transfer tensor to obtain a target transfer tensor that satisfies a predetermined rule comprises:

respectively acquiring the number of vehicles appearing in each set time period in the target transfer network and the number of specified interest points located in each preset range of historical position points in the target transfer network to form a central tensor;

filling the first transfer tensor based on the obtained central tensor to obtain a second transfer tensor;

performing optimization training on the obtained second transfer tensor to obtain each optimization training result;

and determining the optimized training results which accord with the preset rules as the target transfer tensor based on the optimized training results.

5. The method of claim 4, wherein performing optimization training on the obtained second transfer tensor to obtain respective optimization training results comprises:

calculating a loss value between the first transfer tensor and the second transfer tensor;

and fitting the second transfer tensor to obtain each optimization training result based on the loss values and each regularization item, wherein each regularization item is generated by the set time period set, each historical position point in each historical driving track of the vehicle, the target transfer network, the central tensor and a preset geographic feature factor.

6. The method according to any one of claims 1 to 5, wherein calculating the prediction probability between the current driving trajectory and any one of the historical position points in each historical driving trajectory by using the destination prediction model based on the time period corresponding to the current driving trajectory comprises:

determining an initial position point, a current stopping position point and each passing position point between the initial position point and the current stopping position point of the current driving track;

determining target transition probabilities between the starting position point, the current stopping position point and each passing position point of the current driving track and any one historical position point in each historical driving track in a time period corresponding to the current driving track in the destination prediction model;

and determining the prediction probability between the current driving track and any one historical position point in each historical driving track based on each obtained target transition probability.

7. A destination prediction apparatus, comprising:

the model establishing unit is used for generating a destination prediction model, and specifically comprises:

after each obtained historical driving track of the vehicle is projected to a road network, a first transfer network is established in the projected road network, and the first transfer network is used for indicating the relative motion among historical position points in each historical driving track before being converted;

identifying each historical position point in each historical driving track contained in the first transfer network as a transient historical position point, and creating a mirror image absorption historical position point for each transient historical position point, wherein the transient historical position point represents a non-termination point in the historical driving track, and the absorption historical position point represents a termination point in the historical driving track;

establishing a corresponding target transfer network based on the relative motion between each transient historical position point and each absorption historical position point in each historical driving track, wherein the target transfer network is used for indicating the relative motion between each historical position point in each historical driving track after conversion;

the monitoring unit is used for monitoring the current driving track of the vehicle and determining the time period corresponding to the current driving track;

the obtaining unit is used for obtaining the destination prediction model, and the destination prediction model records the target transition probability between every two historical position points in each historical driving track of the vehicle in each set time period, wherein the set time period is divided based on one natural day;

the calculation unit is used for calculating the prediction probability between the current driving track and each historical position point in each historical driving track by adopting the destination prediction model based on the time period corresponding to the current driving track;

and the determining unit is used for determining the historical position point of which the corresponding prediction probability meets the preset condition as the prediction destination.

8. The apparatus of claim 7, wherein after establishing the respective target transition network, the model establishing unit is further configured to:

establishing a first transfer tensor based on a first dimension, a second dimension and a third dimension, wherein the first dimension represents each historical position point before the transfer occurs in the target transfer network, the second dimension represents each historical position point after the transfer occurs in the target transfer network, the third dimension represents a set of set time periods, and the first transfer tensor is used for indicating a first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period;

and processing the first transfer tensor to obtain a target transfer tensor which accords with a preset rule, wherein the target transfer tensor respectively records the target transfer probability between every two historical position points in each set time period in each historical driving track of the vehicle.

9. The apparatus of claim 8, wherein in establishing the first transfer tensor based on the first dimension, the second dimension, and the third dimension, the model establishing unit is to:

dividing the target transfer network into a forward target transfer network and a recent target transfer network according to a preset time division point, wherein the forward target transfer network comprises a historical position point generated in the target transfer network before the time division point, and the recent target transfer network comprises a historical position point generated in the target transfer network after the time division point;

establishing a long-term first transfer tensor according to the long-term target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the long-term first transfer tensor represents each historical position point before transfer occurs in the long-term target transfer network, the second dimension of the long-term first transfer tensor represents each historical position point after transfer occurs in the long-term target transfer network, the third dimension of the long-term first transfer tensor represents the set time periods, and the long-term first transfer tensor is used for indicating a long-term first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period before the time division point;

establishing a recent first transfer tensor according to the recent target transfer network and based on a first dimension, a second dimension and a third dimension, wherein the first dimension of the recent first transfer tensor represents each historical position point before transfer occurs in the recent target transfer network, the second dimension of the recent first transfer tensor represents each historical position point after transfer occurs in the recent target transfer network, the third dimension of the recent first transfer tensor represents the set of set time periods, and the recent first transfer tensor is used for indicating a recent first transfer probability between every two historical position points in each historical driving track of the vehicle in each set time period after the time division point;

forming the first transfer tensor using the future first transfer tensor and the near first transfer tensor.

10. The apparatus of claim 8, wherein when the first transfer tensor is processed to obtain the target transfer tensor according to a preset rule, the model building unit is configured to:

respectively acquiring the number of vehicles appearing in each set time period in the target transfer network and the number of specified interest points located in each preset range of historical position points in the target transfer network to form a central tensor;

filling the first transfer tensor based on the obtained central tensor to obtain a second transfer tensor;

performing optimization training on the obtained second transfer tensor to obtain each optimization training result;

and determining the optimized training results which accord with the preset rules as the target transfer tensor based on the optimized training results.

11. The apparatus of claim 10, wherein the obtained second transfer tensor is optimally trained, and when obtaining each optimized training result, the model building unit is configured to:

calculating a loss value between the first transfer tensor and the second transfer tensor;

and fitting the second transfer tensor to obtain each optimization training result based on the loss values and each regularization item, wherein each regularization item is generated by the set time period set, each historical position point in each historical driving track of the vehicle, the target transfer network, the central tensor and a preset geographic feature factor.

12. The apparatus according to any one of claims 7 to 11, wherein when the prediction probability between the current trajectory and any one of the historical position points in the respective historical trajectories is calculated by using the destination prediction model based on a time period corresponding to the current trajectory, the calculation unit is configured to:

determining an initial position point, a current stopping position point and each passing position point between the initial position point and the current stopping position point of the current driving track;

determining target transition probabilities between the starting position point, the current stopping position point and each passing position point of the current driving track and any one historical position point in each historical driving track in a time period corresponding to the current driving track in the destination prediction model;

and determining the prediction probability between the current driving track and any one historical position point in each historical driving track based on each obtained target transition probability.

Images