CN117077928A

CN117077928A - Network appointment vehicle demand prediction method, device, equipment and storage medium

Info

Publication number: CN117077928A
Application number: CN202310891109.8A
Authority: CN
Inventors: 楼俊钢; 张心叶; 申情; 张雄涛; 赵康
Original assignee: Huzhou University
Current assignee: Huzhou University
Priority date: 2023-07-19
Filing date: 2023-07-19
Publication date: 2023-11-17

Abstract

The application provides a network about vehicle demand prediction method, a device, equipment and a storage medium, which relate to the technical field of traffic prediction, wherein the method comprises the following steps: acquiring road traffic data of network vehicle-restraining demand time, a preset time period network vehicle-restraining starting place and a destination, preprocessing the road traffic data to construct a road traffic state matrix, dividing global views based on preset time intervals, and processing the road traffic state matrix by utilizing a goblet-sea squirt algorithm to obtain a target process view state matrix; and performing feature splicing on the two state matrixes to obtain local spatial feature information, inputting a preconfigured gate-control cyclic convolutional neural network model to obtain local space-time feature information, performing dimension reduction processing to obtain regional feature information, performing dot product operation to obtain global spatial feature information, inputting the global spatial feature information and network constraint vehicle demand time into a converter model to obtain a prediction result, and improving the prediction accuracy by considering various factors such as time, local and global spatial correlation, origin-destination demand and the like.

Description

Network appointment vehicle demand prediction method, device, equipment and storage medium

Technical Field

The present application relates to the field of traffic prediction technologies, and in particular, to a method, an apparatus, a device, and a storage medium for predicting network taxi demands.

Background

With rapid urban traffic development, the demand for using network to reduce vehicles is continuously increased, so that traffic road congestion is frequently caused. Generally, inaccurate network about vehicle demand prediction can cause mismatching of path planning and current traffic conditions, so that network about vehicles are gathered in a large number of areas, and traffic pressure is overlarge, therefore, real-time accurate network about vehicle demand prediction can provide effective travel information, save travel time of travelers and relieve pressure of urban traffic systems.

In the prior art, the prior knowledge of the departure network about vehicles, such as historical network about vehicle data, network about vehicle influence factors and the like, can be obtained, so that the network about vehicle requirements of departure places can be modeled based on a time sequence method, a regression analysis method or a convolution neural network method and the like, and the network about vehicle requirements of all areas can be predicted by using the built model.

However, when the model predicts the network vehicle demand, the influence factors of the network vehicle demand are single, for example, only the convolutional neural network is used for considering the space factors, so that the prediction accuracy of the network vehicle demand is influenced, and the prediction accuracy is reduced.

Disclosure of Invention

The application provides a network taxi demand prediction method, device, equipment and storage medium, which are used for solving the problems that when the network taxi demand is predicted by the existing model, the prediction accuracy of the network taxi demand is influenced by considering that the influence factors on the network taxi demand are single, and the prediction accuracy is reduced.

In a first aspect, the present application provides a method for predicting network taxi demand, the method comprising:

acquiring network vehicle-restraining demand time and road traffic data corresponding to a network vehicle-restraining starting place and a network vehicle-restraining destination in a preset time period, preprocessing the road traffic data, and respectively constructing a road traffic state matrix corresponding to the network vehicle-restraining starting place and the network vehicle-restraining destination; the network vehicle-restraining demand time is the time when a user needs to use the network vehicle-restraining;

global view division is carried out on the road traffic state matrix based on a preset time interval to obtain a plurality of process view state matrixes, and the process view state matrixes are processed by utilizing a goblet-sea squirt algorithm to obtain a target process view state matrix;

performing feature stitching on the target process view state matrix and the road traffic state matrix to obtain local spatial feature information, and inputting the local spatial feature information into a pre-configured gating cyclic convolutional neural network model to obtain local space-time feature information;

Performing dimension reduction processing on the local space-time characteristic information to obtain regional characteristic information, calculating dot products of the regional characteristic information to obtain global space characteristic information, and inputting the network vehicle-closing demand time and the global space characteristic information into a converter model to obtain a prediction result; and the prediction result is used for predicting the number of the network taxi in the network taxi demand time.

Optionally, preprocessing the road traffic data, and respectively constructing a road traffic state matrix corresponding to the network vehicle starting place and the network vehicle destination, including:

carrying out normalization processing on the road traffic data, and carrying out grid division on the network vehicle starting place and the network vehicle destination to form a grid area;

and carrying out dimension conversion on the road traffic data subjected to normalization processing based on the grid area, and respectively constructing road traffic state matrixes corresponding to the network vehicle starting place and the network vehicle destination.

Optionally, feature stitching is performed on the view state matrix of the target process and the road traffic state matrix to obtain local spatial feature information, including:

acquiring an application scene corresponding to a network vehicle, and determining a weight value corresponding to the target process view state matrix and the road traffic state matrix in the application scene;

And carrying out weighted summation on the target process view state matrix and the road traffic state matrix based on the weight value to obtain local space characteristic information.

Optionally, the construction process of the preconfigured gated cyclic convolutional neural network model includes:

acquiring a training data set, wherein the training data set comprises a plurality of time periods, a plurality of regional information and the number of intra-regional network vehicles corresponding to each time period;

and performing iterative training on the gating cyclic convolutional neural network model by using the training data set to obtain a preconfigured gating cyclic convolutional neural network model.

Optionally, the local spatial feature information includes local area information and the number of network vehicles corresponding to the local area; inputting the local spatial feature information into a preconfigured gating cyclic convolutional neural network model to obtain local space-time feature information, wherein the method comprises the following steps of:

inputting the local area information and the number of the network vehicles into a pre-configured gating cyclic convolutional neural network model to obtain local space-time characteristic information comprising prediction time; and the prediction time is a prediction time period corresponding to the local spatial feature information.

Optionally, performing dimension reduction processing on the local space-time feature information to obtain region feature information, and calculating a dot product of the region feature information to obtain global space feature information, where the method includes:

Performing dimension reduction processing on the local space-time characteristic information aiming at each region to obtain region characteristic information;

and transposing the region characteristic information, and performing dot product operation by using the transposed region characteristic information and the original region characteristic information to obtain global space characteristic information.

Optionally, the method further comprises:

acquiring a plurality of historical prediction results and the actual network vehicle number corresponding to the network vehicle demand time of the plurality of historical prediction results;

based on the plurality of historical prediction results and the actual network vehicle number, calculating prediction accuracy of the plurality of historical prediction results by using an evaluation algorithm to obtain an evaluation result; the evaluation algorithm comprises an average absolute percentage error method, an average standard error method and/or an average absolute error method;

and generating prompt information based on the evaluation result so as to prompt a user to correct the network about vehicle demand prediction method.

In a second aspect, the present application provides a network taxi demand prediction apparatus, the apparatus comprising:

the acquisition module is used for acquiring road traffic data corresponding to the network vehicle starting place and the network vehicle destination in the network vehicle demand time and the preset time period, preprocessing the road traffic data and respectively constructing a road traffic state matrix corresponding to the network vehicle starting place and the network vehicle destination; the network vehicle-restraining demand time is the time when a user needs to use the network vehicle-restraining;

The processing module is used for carrying out global view division on the road traffic state matrix based on a preset time interval to obtain a plurality of process view state matrixes, and processing the plurality of process view state matrixes by utilizing a goblet sea squirt algorithm to obtain a target process view state matrix;

the input module is used for carrying out characteristic splicing on the view state matrix of the target process and the road traffic state matrix to obtain local space characteristic information, and inputting the local space characteristic information into a pre-configured gating cyclic convolutional neural network model to obtain local space-time characteristic information;

the prediction module is used for performing dimension reduction processing on the local space-time characteristic information to obtain regional characteristic information, calculating dot products of the regional characteristic information to obtain global space characteristic information, and inputting the network vehicle demand time and the global space characteristic information into a converter model to obtain a prediction result; and the prediction result is used for predicting the number of the network taxi in the network taxi demand time.

In a third aspect, the present application also provides an electronic device, including: a processor, and a memory communicatively coupled to the processor;

The memory stores computer-executable instructions;

the processor executes computer-executable instructions stored by the memory to implement the method of any one of the first aspects.

In a fourth aspect, the present application also provides a computer-readable storage medium storing computer-executable instructions for implementing the method according to any one of the first aspects when executed by a processor.

In summary, the method, the device, the equipment and the storage medium for predicting the network vehicle demand can be used for respectively constructing a road traffic state matrix corresponding to the network vehicle starting place and the network vehicle destination by acquiring the network vehicle demand time and the road traffic data corresponding to the network vehicle starting place and the network vehicle destination in a preset time period and preprocessing the road traffic data; further, global view division is carried out on the road traffic state matrix based on a preset time interval to obtain a plurality of process view state matrixes, and the process view state matrixes are optimized by utilizing a goblet-sea squirt algorithm to obtain a target process view state matrix; further, characteristic stitching is carried out on the view state matrix of the target process and the road traffic state matrix to obtain local spatial characteristic information, and the local spatial characteristic information is processed by utilizing a pre-configured gating cyclic convolution neural network model to obtain local space-time characteristic information; further, performing dimension reduction processing on the local space-time characteristic information to obtain regional characteristic information, performing dot product operation on the regional characteristic information to obtain global space characteristic information, and inputting the network vehicle demand time and the global space characteristic information into a converter model to obtain a prediction result of the number of network vehicles corresponding to the predicted network vehicle demand time; the short-term and long-term time dependence of the data is respectively captured by taking into consideration factors such as time, local and global spatial correlation, origin-destination demand and the like and by utilizing a gate-controlled cyclic convolutional neural network model and a converter model, the short-term and long-term correlation is also considered in time, and the accuracy of the road traffic network vehicle-closing prediction is further improved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.

Fig. 1 is a schematic view of an application scenario provided in an embodiment of the present application;

fig. 2 is a schematic flow chart of a network about vehicle demand prediction method according to an embodiment of the present application;

FIG. 3 is a schematic flow chart of optimizing local spatial context by using the ascidian algorithm according to the embodiment of the application;

fig. 4 is a schematic diagram of a model structure corresponding to a network about vehicle demand prediction method according to an embodiment of the present application;

FIG. 5 is a graph comparing results of an example provided by an embodiment of the present application;

fig. 6 is a schematic flow chart of a specific network taxi demand prediction method according to an embodiment of the present application;

fig. 7 is a schematic structural diagram of a network bus demand prediction device according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Specific embodiments of the present application have been shown by way of the above drawings and will be described in more detail below. The drawings and the written description are not intended to limit the scope of the inventive concepts in any way, but rather to illustrate the inventive concepts to those skilled in the art by reference to the specific embodiments.

Detailed Description

In order to clearly describe the technical solution of the embodiments of the present application, in the embodiments of the present application, the words "first", "second", etc. are used to distinguish the same item or similar items having substantially the same function and effect. For example, the first device and the second device are merely for distinguishing between different devices, and are not limited in their order of precedence. It will be appreciated by those of skill in the art that the words "first," "second," and the like do not limit the amount and order of execution, and that the words "first," "second," and the like do not necessarily differ.

In the present application, the words "exemplary" or "such as" are used to mean serving as an example, instance, or illustration. Any embodiment or design described herein as "exemplary" or "for example" should not be construed as preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "such as" is intended to present related concepts in a concrete fashion.

In the present application, "at least one" means one or more, and "a plurality" means two or more. "and/or", describes an association relationship of an association object, and indicates that there may be three relationships, for example, a and/or B, and may indicate: a alone, a and B together, and B alone, wherein a, B may be singular or plural. The character "/" generally indicates that the context-dependent object is an "or" relationship. "at least one of" or the like means any combination of these items, including any combination of single item(s) or plural items(s). For example, at least one (one) of a, b, or c may represent: a, b, c, a-b, a-c, b-c, or a-b-c, wherein a, b, c may be single or plural.

With rapid urban traffic development, the demand for using network vehicles is increased, and thus traffic congestion often occurs. Generally, inaccurate network appointment demand prediction can cause mismatching of path planning and current traffic conditions, so that network appointment vehicles are gathered in a large number of areas, and traffic pressure is overlarge.

In one possible implementation manner, the traffic flow prediction method may include a method based on a time sequence, a method based on regression analysis, a method based on a neural network, and the like, where a certain priori knowledge, such as historical data, network vehicle influence factors, and the like, is required to construct a prediction model to predict the network vehicle demand.

In another possible implementation manner, the prediction of the network taxi demand can be performed based on a traffic flow prediction method of deep learning by considering the influence of space-time characteristics on the traffic flow prediction method.

However, the above method usually only focuses on the fact that spatially adjacent regions have similar demand patterns, neglecting that even if two regions are spatially far apart, if they have similar properties, the demand patterns may still have some spatial correlation, and the correlation needs to be considered separately from the short term and the long term in time, so that the prediction accuracy of the network traffic demand is greatly affected due to the fact that the consideration of the above factors is relatively single, so that the prediction accuracy is reduced.

It should be noted that most of the existing methods only model the driving demands of the departure place, predict the network vehicle demand of all areas, neglect the destination influence to a certain extent, and fail to perform the best selection on the origin-destination path planning, if the origin-destination path planning cannot be reasonably planned, the prediction accuracy is also reduced. In addition, the time correlation and the local and global spatial correlation are also important factors for predicting the network vehicle traffic and influencing the traffic origin-destination demand prediction performance.

In view of the above problems, the present application provides a method for predicting network about vehicle demands, in space, the network about vehicle demands are calculated from origin-destination views respectively, that is, road traffic state matrixes corresponding to origin of network about vehicles and destination of network about vehicles are used for calculation, further, a goblet sea squirt algorithm is used for optimizing and learning local space features, that is, optimizing process view state matrixes, searching for optimal target process view state matrixes, further, short-term time correlation of gate-controlled circulation units (Gate Recurrent Unit, GRU) is used for processing the target process view state matrixes, obtaining local space-time features, and combining global space context, that is, dot product operation is performed, obtaining global space features, further, the global space features are input into a converter model (transformer) for learning long-term time correlation, obtaining a prediction result, wherein, as GRU and transformer can capture short-term and long-term time dependencies in data respectively, the short-term and long-term correlations can be considered, in this way, the accuracy of network about prediction is further improved by considering factors such as time, local and global space correlations, short-term correlations, and origin-destination demands.

Embodiments of the present application are described below with reference to the accompanying drawings. Fig. 1 is a schematic diagram of an application scenario provided by an embodiment of the present application, where the method for predicting network taxi demand provided by the present application may be applied to the application scenario shown in fig. 1, and as shown in fig. 1, the application scenario includes: a data processing system 101 and a user's terminal device 102; the data processing system 101 may collect road traffic data corresponding to the network vehicle starting point a and road traffic data corresponding to the network vehicle destination B.

Specifically, the data processing system 101 may obtain the time when the user needs to use the network about vehicles, predict the number of the network about vehicles required by the corresponding time by using the network about vehicle demand prediction method based on the collected road traffic data corresponding to the network about vehicle starting point a and the road traffic data corresponding to the network about vehicle destination B, and feed back the predicted number of the network about vehicles to the terminal device 102 of the user, so that the user performs reasonable network about vehicle scheduling based on the number of the network about vehicles, and the reduction causes a large number of the network about vehicles to be aggregated due to the excessive number of the network about vehicles scheduled at the time, thereby causing a large traffic pressure condition.

The data processing system 101 may be a server of the network taxi dispatching platform or a server of the traffic management platform, which is not limited in particular, and the user terminal device 102 may be a display device corresponding to the network taxi dispatching platform or a terminal device corresponding to a manager, or a display device corresponding to the traffic management platform or a terminal device corresponding to a manager.

It is understood that the Terminal device may be various electronic devices having a display screen and supporting web browsing, and the Terminal device may also be referred to as a Terminal (Terminal), a User Equipment (UE), a Mobile Station (MS), a Mobile Terminal (MT), or the like. The terminal device may be a mobile phone, a smart television, a wearable device, a smart speaker, a smart security device, a smart gateway, a tablet computer (Pad), a computer with wireless transceiving function, a Virtual Reality (VR) terminal device, an augmented Reality (Augmented Reality, AR) terminal device, a wireless terminal in industrial control (industrial control), a wireless terminal in self-driving (self-driving), a wireless terminal in teleoperation (remote medical surgery), a wireless terminal in smart grid (smart grid), a wireless terminal in transportation security (transportation safety), a wireless terminal in smart city (smart city), a wireless terminal in smart home (smart home), etc. Such terminal devices include, but are not limited to, smartphones, tablet computers, laptop portable computers, desktop computers, and the like.

It should be noted that, the road traffic data corresponding to the network vehicle starting point a and the network vehicle destination B may be determined based on different application scenarios, and fig. 1 is merely an illustration.

The technical scheme of the application is described in detail below by specific examples. The following embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. Embodiments of the present application will be described below with reference to the accompanying drawings.

Fig. 2 is a schematic flow chart of a network about vehicle demand prediction method according to an embodiment of the present application; as shown in fig. 2, the method of the present embodiment may include:

s201, acquiring road traffic data corresponding to a network vehicle starting place and a network vehicle destination in a network vehicle demand time and a preset time period, preprocessing the road traffic data, and respectively constructing a road traffic state matrix corresponding to the network vehicle starting place and the network vehicle destination; the network vehicle-restraining demand time is the time when a user needs to use the network vehicle-restraining.

In the embodiment of the application, the preset time period may refer to a time preset in advance and used for predicting the space start-stop demand of the network vehicle, and the specific data corresponding to the preset time period is not limited by the application.

Preprocessing may refer to normalizing road traffic data by a normalization algorithm, where the road traffic data may include traffic flow data, traffic speed, traffic occupancy, and the like; the normalization algorithm comprises: the maximum-minimum normalization algorithm, standard deviation normalization algorithm, decimal scaling normalization algorithm, etc., and the embodiment of the application does not limit the specific normalization algorithm used.

In the step, the collected road traffic data of a plurality of roads are subjected to normalization preprocessing, and a road traffic data matrix is constructed according to the preprocessed road traffic data.

S202, global view division is conducted on the road traffic state matrix based on a preset time interval to obtain a plurality of process view state matrices, and the process view state matrices are processed by utilizing a goblet sea squirt algorithm to obtain a target process view state matrix.

In the embodiment of the application, because the road traffic data corresponding to the network vehicle starting place and the network vehicle destination are changed based on time, global view division is carried out on the road traffic state matrix by utilizing a preset time interval to obtain a plurality of process view state matrixes; the predetermined time interval may refer to a time interval t preset in advance for global view division, and the numerical value corresponding to t in the embodiment of the present application is not particularly limited.

In this step, the global view divides the road traffic state matrix and sets parameters, and then the target process view state matrix is optimized by using the goblet-sea squirt algorithm (Salp Swarm Algorithm, SSA).

For example, fig. 3 is a schematic flow chart of optimizing a local spatial context by using a goblet-sea squirt algorithm according to an embodiment of the present application, as shown in fig. 3, a global view dividing matrix is performed, parameters are set, specifically, a position of each goblet-sea squirt individual (net-bound vehicle) may be expressed as an N-dimensional vector, or called a position vector (x 1, … …, xn), and a search space is set to be expressed as follows by using a p×n matrix:

Where P is the spatial dimension and N is the population number (traffic volume). The position of the ith sea squirt in space is expressed asThe position of the target food (net car destination) is expressed as +.>

Further, the leader's goblet position update, i.e. during movement and foraging of the goblet chain, the leader may advance towards the food (destination), i.e. during travel of the net cart from the origin to the destination, the position update is expressed as follows:

wherein,and F _p Respectively expressed as the position and the current handling net restraint vehicle of the first sea squirt (set as the leader) in the p-th dimensionThe location, X, of the target food (net destination) _max And X _min The upper and lower boundaries c1, c2, and c3 are control parameters, respectively, and are set in advance.

And it can be seen from equation (2) that the leader position update is only related to the position of the food. Since c1 is related to the iteration number of the current population (traffic flow), and is also a convergence factor in the optimization algorithm, and plays a role in balancing global exploration and local development, and is the most important control parameter in SSA, the expression of c1 is as follows:

where K is the current iteration number and K is the maximum iteration number. The convergence factor is a decreasing function of 2-0, and the control parameters c2, c3 are both 0,1 ]C2 represents the movement length, and c3 represents the forward and reverse directions of the movement direction. The purpose of the designs c2, c3 is to enhanceThereby enhancing the global searching capability of the whole chain group.

It should be noted that, for each iteration, the value of the convergence factor is reduced by half until convergence is 0.

Further, the follower position is updated, and the goblet sea squirt iteratively searches for an optimal solution, i.e., searches for a target process view state matrix from a plurality of process view state matrices.

Specifically, the next iteration position of the ith follower is determined by the position of the ith follower and the position of the ith-1 goblet sea squirt in the current iteration. Since the followers advance in a chain-like order by the influence of each other between the front and rear individuals, the movement displacement of the followers is expressed by the following formula (4):

where a, v are defined identification letters only and do not represent acceleration and velocity, but in the optimization process, t is iterative, let t=1 in the iterative process and v0=0, the position of the follower can be expressed as the following formula (5):

when i is more than or equal to 2,and->The positions of followers before and after updating in the p-th dimension are respectively represented by X _p And (3) representing.

Further, feature stitching is performed on a target process view state matrix (process view CNN), an origin view CNN (road traffic state matrix corresponding to a network vehicle starting place) and a target view CNN (road traffic state matrix corresponding to a network vehicle destination), so as to obtain local spatial feature information, wherein the process view CNN is a state matrix of a process view convolutional neural network (Convolutional Neural Network, CNN).

S203, performing feature stitching on the view state matrix of the target process and the road traffic state matrix to obtain local spatial feature information, and inputting the local spatial feature information into a pre-configured gate control cyclic convolution neural network model to obtain local space-time feature information.

In the embodiment of the application, the gating cyclic convolutional neural network model can be used for adjusting the structure of the network on the basis of the convolutional cyclic neural network, adding the model of the gating cyclic unit for controlling the transmission of information in the convolutional neural network, wherein the gating cyclic unit can be used for controlling how much information in a memory unit needs to be reserved and how much information needs to be discarded, and how much new state data information needs to be saved in the memory unit, so that the gating cyclic convolutional neural network model can learn the dependency relationship with relatively long span without the problems of gradient disappearance and gradient explosion.

Illustratively, local spatial feature information can be obtained by referring to the embodiment shown in fig. 3, further, the local spatial feature information is input into a preconfigured gated cyclic convolutional neural network model, and the local spatial feature information is obtained by prediction based on the short-term time correlation of the GRU learning.

S204, performing dimension reduction processing on the local space-time characteristic information to obtain regional characteristic information, calculating dot products of the regional characteristic information to obtain global space characteristic information, and inputting the network vehicle demand time and the global space characteristic information into a converter model to obtain a prediction result; and the prediction result is used for predicting the number of the network taxi in the network taxi demand time.

In the embodiment of the application, the converter model can be referred to as a converter model, a plurality of converter layers are arranged in the converter model, and the global space characteristic information and the network contract vehicle demand time are input into the converter layers, so that the predicted long-term time correlation can be enhanced, and a prediction result is obtained.

Specifically, in the global space feature information F ^g For node n, a query matrix, a key matrix, and a value matrix are obtained using the following formula:

wherein,all are learnable parameters, and self-focusing operation is performed on the time dimension, so that the dependency relationship among all time slices in the node n is as follows:

it will be appreciated that the time-self-care operation may discover dynamic time patterns of different nodes in traffic data, modeling long-term time dependencies between all time slices.

Further, the spatiotemporal characteristics of the output of the temporal self-attention module can be obtained:

and further can be used for the space-time characteristic F ^gt And inputting the network appointment vehicle demand time into a linear regression model, wherein the formula of the linear regression model is as follows:

X′ _t+1 ＝tanh(τ(F ^gt )) (9)

wherein X 'is' _t+1 And (3) representing the network vehicle demand time, wherein τ is a linear regression parameter realized by a convolution layer, and the hyperbolic tangent can ensure that the output is between-1 and 1 so as to obtain a prediction result meeting the network vehicle demand time.

It should be noted that, in the training process of the model, the network vehicle demand time in the formula (9) can be predicted, and the accuracy of the model is judged by comparing with the actual data.

Therefore, the application provides a network vehicle demand prediction method, which can be used for respectively constructing a road traffic state matrix corresponding to a network vehicle starting place and a network vehicle destination by acquiring the network vehicle demand time and road traffic data corresponding to the network vehicle starting place and the network vehicle destination in a preset time period and preprocessing the road traffic data; further, global view division is carried out on the road traffic state matrix based on a preset time interval to obtain a plurality of process view state matrixes, and the process view state matrixes are optimized by utilizing a goblet-sea squirt algorithm to obtain a target process view state matrix; further, characteristic stitching is carried out on the view state matrix of the target process and the road traffic state matrix to obtain local spatial characteristic information, and the local spatial characteristic information is processed by utilizing a pre-configured gating cyclic convolution neural network model to obtain local space-time characteristic information; further, performing dimension reduction processing on the local space-time characteristic information to obtain regional characteristic information, performing dot product operation on the regional characteristic information to obtain global space characteristic information, and inputting the network vehicle demand time and the global space characteristic information into a converter model to obtain a prediction result of the number of network vehicles corresponding to the predicted network vehicle demand time; the short-term and long-term time dependence of the data is captured by taking into consideration factors such as time, local and global spatial correlation, origin-destination demand and the like and by utilizing a gate-controlled cyclic convolutional neural network model and a converter model, the short-term and long-term correlation is considered in time, and the traffic prediction precision of the network traffic origin-destination demand is further improved.

An exemplary implementation of the network taxi demand prediction method has a corresponding model structure, and fig. 4 is a schematic diagram of a model structure corresponding to the network taxi demand prediction method according to an embodiment of the present application, as shown in fig. 4, based on the embodiment described in fig. 2, an X is shown _t ，X _t-n+2 ，X _t-n+1 Optimizing local space context (Local Spatial Context-Salp Swarm Algorithm, LSC-SSA) by using the Zun sea squirt algorithm to obtain multiple process view state matrixes, and inputting the multiple process view state matrixes into a GRU model (gated cyclic convolutional neural network model) to obtain short-term predicted local space-time characteristics F ^l Further, F ^l Input into a convolution layer to generate an embedded feature F _s ，F _s The original three-dimensional matrix is reduced in dimension Cheng Erwei matrix, and thenAnd its transposed matrix->Performing dot product operation to obtain a similarity matrix S, further adding all the region features and the calculated similarity weight, and performing dot product operation to calculate global space features F of each region ^g Further F can be made ^g Network contract vehicle demand time X _t ' ₊₁ Input into a converter model (transducer) to obtain a space-time characteristic F meeting the demand time of a network vehicle ^gt 。

In this step, the collected road traffic data is normalized by using the formula (10), and the formula (10) is as follows:

wherein X is _i X is the original road traffic data _min X is the minimum value in the original road traffic data _max And X is the road traffic data after preprocessing, and is the maximum value in the original road traffic data.

Further, a road traffic state matrix is constructed according to the preprocessed road traffic data, namely dimension conversion is performed, and the network vehicle departure place requirement at the time interval t is expressed as a three-dimensional OD (Origin-Destination) matrix X _i ∈R ^N ^×H×W Where H and W are the height and width, respectively, of the urban grid map, and the Destination demand represents the transposed matrix of the Origin, represented as a three-dimensional DO (Destination-Origin) matrix

For example, a taxi trip record is adopted to construct a network taxi demand prediction data set, a departure place and a destination of most network taxi is divided into 15×5 grid areas based on the most network taxi, each grid area represents a road traffic state matrix with the size of about 0.75km×0.75km, the sampling time interval t is 10min, and the road traffic state matrix corresponding to the departure place and the destination is further constructed.

Therefore, the embodiment of the application normalizes the road traffic data, so that the calculation can be more convenient and rapid, and the obtained result is more reasonable by acquiring the data corresponding to the network vehicle starting place and the network vehicle destination, thereby improving the prediction precision of the model.

In the embodiment of the application, the application scene may refer to a space and time environment where the network vehicle is located, and under different application scenes, the weight values corresponding to the target process view state matrix and the road traffic state matrix are different, for example, the weight values corresponding to the target process view state matrix and the road traffic state matrix are different in the time of the region a and the time of the region B, and the weight values of the application scene and the matrix correspondingly set by each application scene are not particularly limited and are determined according to actual conditions.

It can be understood that the weight value corresponding to the application scene can be set in advance, and can be directly mobilized based on the application scene when in use, or can be manually input and modified, and the embodiment of the application is not particularly limited.

In this step, the designed origin view CNN may contain 2 convolution layers, where each convolution layer has 16 filters with a kernel size of 3×3, and one ReLU layer immediately follows the filters for improving performance and expression capability, so that the gradient vanishing problem can be effectively alleviated, and further, the stride of all the convolution layers may be set to 1 so that it maintains the same resolution in space.

It should be noted that, the target view CNN and the origin view CNN have the same network structure, so that the convolution of 32 filters can be used to fuse three features, and the specific formula is as follows:

wherein the method comprises the steps ofIs feature stitching, i.e. weighting and summing, F _i ^o ，F _i ^d ，F _i ^P The output features, ω, of the three views respectively _o ，ω _d ，ω _P Parameters of an origin view CNN, a target view CNN and a process view CNN respectively, ω is a parameter of a fusion convolution layer, F _i ^l Is the generated local spatial feature information.

Therefore, the embodiment of the application can generate reasonable local spatial characteristic information based on characteristic splicing, thereby improving the rationality and diversity of the construction data.

In the step, modeling the network vehicle demand in a short-term time by using a gated cyclic convolutional neural network model, specifically, inputting a training data set into the gated cyclic convolutional neural network model for iterative k times of training, wherein a given input X at the moment t _k And hidden layer state H at time t-1 _k-1 Acquiring two reset gates R _k Updating door Z _k State, candidate hidden layer stateFinal hidden state H _k The formulas are shown below, respectively:

R _k ＝σ(X _k W _xr +H _k-1 W _hr +b _r ) (15)

Z _k ＝σ(X _k W _xz +H _k-1 W _hz +b _z ) (16)

wherein sigma is a sigmoid function, which can transform data into a numerical value in the range of 0-1, and serve as a gating signal, which represents multiplication by element, R _k And Z is _k The range is 0-1, and the closer the value is to 0, the more information is forgotten.

When using a gated cyclic convolutional neural network model, the features can be obtained by applying equation (14) Sequentially introducing into ConvGRU to obtain local space-time characteristic F ^l The formula is as follows:

F ^l ＝Conv(H _k ,ω _l ) (19)

therefore, the embodiment of the application can control the transmission of information in the neural network by constructing the gated circular convolution neural network model, and improves the feature extraction capability of the model.

In the step, local spatial feature information is processed by utilizing a gating cyclic convolutional neural network model which is built in advance to obtain local space-time feature information comprising time and spatial features, wherein the gating cyclic neural network model can learn a dependency relationship with relatively long span without the problems of gradient disappearance and gradient explosion, and the local spatial feature information not only comprises local area information, but also comprises the number of net-about cars corresponding to the local area, and the local area information is used for indicating the local area position where the net-about cars are located, so that the data has spaciousness.

Therefore, the embodiment of the application processes the data through the gating cyclic convolutional neural network model, and can combine the space and time characteristics, thereby improving the comprehensiveness of the acquired data and further improving the prediction precision.

In this step, local spatiotemporal characteristic information F ^l Input to C _s Generating embedded features F in convolutional layers of (1) _s The formula is as follows:

F _s ＝Conv(F ^l ,ω _s ) (20)

wherein F is _s The original three-dimensional matrix may be reduced in dimension Cheng Erwei matrix, n=h×w, further,and its transposed matrix->The dot product operation is performed to obtain a similarity matrix S, and the formula is as follows:

further, the global spatial feature information F of each region is calculated by adding all the region features and the calculated similarity weight and performing dot product calculation ^g The corresponding formula is shown below:

therefore, the embodiment of the application can calculate the global space characteristics by utilizing dot product operation, and improves the calculation rate.

Optionally, the method further comprises:

In the embodiment of the application, the historical prediction result can refer to a result predicted by using the network vehicle demand prediction method in a period of time, and whether the accuracy of the network vehicle demand prediction method is accurate or not is judged by comparing the result with the actual network vehicle number in the period of time.

In this step, an absolute percentage error method (Mean Absolute Percentage Error, MAPE), a mean standard error method (Root Mean Square Error, RMSE) and a mean absolute error method (Mean Square Error, MSE) may be selected as the evaluation methods of the road traffic origin-destination demand prediction accuracy, and the corresponding calculation formulas are respectively as follows:

Wherein z is the number of test samples, X' _t X is the prediction result of time interval t _t Is the actual value in reality of the time interval t, namely the actual net car number.

Taking the statistics of MAPE and RMSE results as an example, as shown in Table 1:

TABLE 1

Method	MAPE	RMSE
			Historical average model (History Average Model, HA)	35.52％	1.88％
Space-time residual error network model (ST-ResNet)	28.43％	1.36％
			Network appointment vehicle demand prediction method	26.57％	1.31％

As can be seen from Table 1, the network vehicle demand prediction method provided by the application HAs better prediction accuracy than other methods, such as HA and ST-ResNet.

FIG. 5 is an exemplary comparison of results provided by embodiments of the present application; as shown in FIG. 5, the network contract vehicle demand prediction method provided by the application is used for predicting the number of network contract vehicles, the actual situation is closer to the prediction result, and the prediction accuracy is more accurate.

Furthermore, the application can also generate prompt information based on the evaluation result, wherein the prompt information can prompt the prediction precision of the user model, and can prompt the user to correct the network vehicle demand prediction method by using other methods; other methods may be other types of algorithms or models added on the basis of the present application, or parameters may be adjusted, which are not particularly limited in the embodiment of the present application, and other more optimized algorithms may be further studied later.

It should be noted that, the prompt message generated in the embodiment of the present application may be sent to the terminal device of the user in the form of a short message, or may be displayed on the display device corresponding to the data processing system in the form of a display frame.

Therefore, the method and the device can evaluate the prediction precision, ensure the prediction effect of the network taxi demand prediction method, remind the user to correct the method based on the evaluation result, and improve the flexibility.

In combination with the foregoing embodiments, fig. 6 is a schematic flow chart of a specific network vehicle demand prediction method according to an embodiment of the present application, as shown in fig. 6, where the network vehicle demand prediction method includes the following flows:

step A: and B, collecting road traffic data of the network vehicle starting place and the destination for a period of time, preprocessing the data, constructing a data set of a road traffic state matrix of the network vehicle starting place and the destination, and executing the step B.

And (B) step (B): and C, dividing a road traffic state matrix by the global view, setting parameters, optimizing a local space context (LSC-SSA) by utilizing a goblet sea squirt algorithm, obtaining a process view CNN state matrix, performing feature stitching with the road traffic state matrix to obtain local space features, and executing the step C.

Step C: and D, capturing short-term time correlation by utilizing the GRU, namely, carrying out convolution and repeatedly iterating the GRU to obtain local space-time characteristics of short-term prediction, and executing the step D.

Step D: the global spatial context (Global Spatial Context, GSC) calculates the similarity between the global features, i.e. the local spatial context dot products all the regional features, thereby generating global spatial features for each region, and step E is performed.

Step E: and capturing long-term correlation by combining a transducer layer, namely inputting global space features into the transducer layer, and enhancing the predicted long-term time correlation to obtain a prediction result.

In the foregoing embodiment, the network taxi demand prediction method provided by the embodiment of the present application is described, and in order to implement each function in the method provided by the embodiment of the present application, the electronic device as the execution body may include a hardware structure and/or a software module, and each function may be implemented in the form of a hardware structure, a software module, or a hardware structure plus a software module. Some of the functions described above are performed in a hardware configuration, a software module, or a combination of hardware and software modules, depending on the specific application of the solution and design constraints.

For example, fig. 7 is a schematic structural diagram of a network about vehicle demand prediction device according to an embodiment of the present application, where, as shown in fig. 7, the device includes: an acquisition module 701, a processing module 702, an input module 703 and a prediction module 704; the acquiring module 701 is configured to acquire road traffic data corresponding to an on-line vehicle starting place and an on-line vehicle destination in a preset time period and perform preprocessing on the road traffic data, and respectively construct a road traffic state matrix corresponding to the on-line vehicle starting place and the on-line vehicle destination; the network vehicle-restraining demand time is the time when a user needs to use the network vehicle-restraining;

the processing module 702 is configured to perform global view division on the road traffic state matrix based on a predetermined time interval to obtain a plurality of process view state matrices, and process the plurality of process view state matrices by using a goblet-sea squirt algorithm to obtain a target process view state matrix;

the input module 703 is configured to perform feature stitching on the target process view state matrix and the road traffic state matrix to obtain local spatial feature information, and input the local spatial feature information into a preconfigured gated cyclic convolutional neural network model to obtain local space-time feature information;

The prediction module 704 is configured to perform dimension reduction processing on the local space-time feature information to obtain area feature information, calculate a dot product of the area feature information to obtain global space feature information, and input the network vehicle demand time and the global space feature information into a converter model to obtain a prediction result; and the prediction result is used for predicting the number of the network taxi in the network taxi demand time.

Optionally, the acquiring module 701 is specifically configured to:

Optionally, the input module 703 includes a splicing unit and an input unit, where the splicing unit is configured to:

Optionally, the local spatial feature information includes local area information and the number of network vehicles corresponding to the local area; the input unit is used for:

Optionally, the prediction module 704 is specifically configured to:

Optionally, the apparatus further includes an evaluation module, configured to:

The specific implementation principle and effect of the network taxi demand prediction device provided by the embodiment of the present application can be referred to the relevant description and effect corresponding to the above embodiment, and will not be repeated here.

The embodiment of the application also provides a schematic structural diagram of an electronic device, and fig. 8 is a schematic structural diagram of an electronic device provided by the embodiment of the application, as shown in fig. 8, the electronic device may include: a processor 801 and a memory 802 communicatively coupled to the processor; the memory 802 stores a computer program; the processor 801 executes the computer program stored in the memory 802, so that the processor 801 performs the method described in any of the above embodiments.

Wherein the memory 802 and the processor 801 may be connected by a bus 803.

Embodiments of the present application also provide a computer-readable storage medium storing computer program-executable instructions that, when executed by a processor, are configured to implement a method as described in any of the foregoing embodiments of the present application.

The embodiment of the application also provides a chip for running instructions, and the chip is used for executing the method in any of the previous embodiments executed by the electronic equipment in any of the previous embodiments.

Embodiments of the present application also provide a computer program product comprising a computer program which, when executed by a processor, performs a method as in any of the preceding embodiments of the present application, as in any of the preceding embodiments performed by an electronic device.

In the several embodiments provided by the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of modules is merely a logical function division, and there may be additional divisions of actual implementation, e.g., multiple modules or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or modules, which may be in electrical, mechanical, or other forms.

The modules illustrated as separate components may or may not be physically separate, and components shown as modules may or may not be physical units, may be located in one place, or may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to implement the solution of this embodiment.

In addition, each functional module in the embodiments of the present application may be integrated in one processing unit, or each module may exist alone physically, or two or more modules may be integrated in one unit. The units formed by the modules can be realized in a form of hardware or a form of hardware and software functional units.

The integrated modules, which are implemented in the form of software functional modules, may be stored in a computer readable storage medium. The software functional modules described above are stored in a storage medium and include instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or processor to perform some of the steps of the methods described in the various embodiments of the application.

It should be appreciated that the processor may be a central processing unit (Central Processing Unit, CPU for short), other general purpose processors, digital signal processor (Digital Signal Processor, DSP for short), application specific integrated circuit (Application Specific Integrated Circuit, ASIC for short), etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the present application may be embodied directly in a hardware processor for execution, or in a combination of hardware and software modules in a processor for execution.

The Memory may include a high-speed random access Memory (Random Access Memory, abbreviated as RAM), and may further include a Non-volatile Memory (NVM), such as at least one magnetic disk Memory, and may also be a U-disk, a removable hard disk, a read-only Memory, a magnetic disk, or an optical disk.

The bus may be an industry standard architecture (Industry Standard Architecture, ISA) bus, an external device interconnect (Peripheral Component Interconnect, PCI) bus, or an extended industry standard architecture (Extended Industry Standard Architecture, EISA) bus, among others. The buses may be divided into address buses, data buses, control buses, etc. For ease of illustration, the buses in the drawings of the present application are not limited to only one bus or to one type of bus.

The storage medium may be implemented by any type of volatile or non-volatile Memory device or combination thereof, such as Static Random-Access Memory (SRAM), electrically erasable programmable Read-Only Memory (Electrically Erasable Programmable Read Only Memory, EEPROM), erasable programmable Read-Only Memory (Erasable Programmable Read-Only Memory, EPROM), programmable Read-Only Memory (Programmable Read-Only Memory, PROM), read-Only Memory (ROM), magnetic Memory, flash Memory, magnetic disk, or optical disk. A storage media may be any available media that can be accessed by a general purpose or special purpose computer.

An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an application specific integrated circuit (Application Specific Integrated Circuits, ASIC for short). It is also possible that the processor and the storage medium reside as discrete components in an electronic device or a master device.

The foregoing is merely a specific implementation of the embodiment of the present application, but the protection scope of the embodiment of the present application is not limited to this, and any changes or substitutions within the technical scope disclosed in the embodiment of the present application should be covered in the protection scope of the embodiment of the present application. Therefore, the protection scope of the embodiments of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for predicting demand of a network taxi, the method comprising:

2. The method of claim 1, wherein preprocessing the road traffic data to construct road traffic state matrices corresponding to the network about vehicle origin and the network about vehicle destination respectively comprises:

3. The method of claim 1, wherein feature stitching the target process view state matrix and the road traffic state matrix to obtain local spatial feature information comprises:

4. The method of claim 1, wherein the constructing of the preconfigured gated circular convolutional neural network model comprises:

5. The method of claim 4, wherein the local spatial feature information includes local area information and a net contract number corresponding to the local area; inputting the local spatial feature information into a preconfigured gating cyclic convolutional neural network model to obtain local space-time feature information, wherein the method comprises the following steps of:

6. The method of claim 1, wherein performing dimension reduction processing on the local space-time feature information to obtain region feature information, calculating a dot product of the region feature information to obtain global space feature information, comprising:

7. The method according to any one of claims 1-6, further comprising:

8. A network appointment vehicle demand prediction device, the device comprising:

9. An electronic device, comprising: a processor, and a memory communicatively coupled to the processor;

The memory stores computer-executable instructions;

the processor executes computer-executable instructions stored in the memory to implement the method of any one of claims 1-7.

10. A computer readable storage medium storing computer executable instructions which when executed by a processor are adapted to carry out the method of any one of claims 1 to 7.