WO2016095708A1 - Traffic flow prediction method, and prediction model generation method and device - Google Patents

Abstract

Disclosed are a traffic flow prediction method, and a prediction model generation method and device. The prediction method comprises: for a road to be predicted, acquiring historical traffic flow data of the road to be predicted within a time period prior to the current time (301); from a correlation between a pre-existing road and a traffic flow prediction model, acquiring the traffic flow prediction model corresponding to the road to be predicted (302); and inputting the historical traffic flow data within the time period prior to the current time into the traffic flow prediction model corresponding to the road to be predicted, to obtain traffic flow data within a time period after the current time (303). Since traffic flow data has high nonlinearity and uncertainty and a neural network model has a relatively high nonlinear prediction capability, the neural network model is trained according to historical traffic flow data of a road, so that a traffic flow prediction model obtained by training can relatively accurately predict the traffic flow data of the road within the time period after the current time according to the traffic flow data of the road within the time period prior to the current time.

Description

Traffic flow prediction method, prediction model generation method and device

This application claims priority to Chinese patent application filed on Dec. 16, 2014, the Chinese Patent Office, Application No. 201410785171.X, entitled "A Traffic Flow Prediction Method, Prediction Model Generation Method and Apparatus", all of which The content is incorporated herein by reference.

Technical field

The present invention relates to the field of real-time traffic, and in particular, to a traffic flow prediction method, a prediction model generation method and apparatus.

Background technique

With the increasing popularity of intelligent transportation systems, the application of real-time traffic flow in intelligent transportation systems is becoming more and more extensive and in-depth. At present, the method of publishing real-time traffic flow data is to release real-time traffic data of the road (such as the driving speed of the road, etc.) at regular intervals (such as 5s, 10s, 30s, 1 minute or 2 minutes, etc.), which is released in the prior art. The real-time traffic data is the real-time traffic situation of the current time road, but it cannot release the traffic flow data for a certain period of time in the future. However, in real life, more and more users are more likely to know the traffic data of certain roads in advance in order to arrange the trips in advance. Therefore, the existing method of releasing real-time traffic flow data cannot meet the needs of users. .

Because traffic flow data has strong nonlinearity and uncertainty, it is difficult for people to speculate on which factors will affect the change of traffic flow in the next period. Therefore, there is no effective technical solution to accurately predict traffic flow data. .

Summary of the invention

In view of this, the embodiments of the present invention provide a traffic flow prediction method, a prediction model generation method, and a device, to provide a manner for accurately predicting traffic flow.

The embodiment of the invention provides a traffic flow prediction method, which is specifically as follows:

Obtaining historical traffic flow data of a previous time period of the current time of the road to be predicted for the road to be predicted;

Obtaining a traffic flow prediction model corresponding to the road to be predicted from a corresponding relationship between the pre-stored road and the traffic flow prediction model; wherein the traffic flow prediction model corresponding to the road is a pre-set nerve according to historical traffic flow data of the road in advance The network model is trained to obtain a traffic flow prediction model for predicting traffic flow data for a subsequent period of the current time of the road according to historical traffic flow data of a previous time period of the current time of the road;

The historical traffic flow data of the previous time period of the current time is input into the traffic flow prediction model corresponding to the road to be predicted, and the traffic flow data of the latter time period of the current time is obtained.

A traffic flow prediction device, the device comprising:

a training module, configured to pre-prepare the preset neural network model according to the historical traffic flow data of the road, and obtain the traffic of the next time period of the current time according to the historical traffic flow data of the previous time period of the current time of the road. Traffic flow prediction model for streaming data;

a storage module, configured to store a correspondence between the road and the traffic flow prediction model obtained by the training module;

a historical traffic data obtaining module, configured to acquire historical traffic flow data of a previous time period of the current time of the road to be predicted for the road to be predicted;

a traffic flow prediction model acquisition module, configured to acquire a traffic flow prediction model corresponding to the road to be predicted from a correspondence between a road and a traffic flow prediction model prestored by the storage module;

The prediction module is configured to input the historical traffic flow data of the previous time period of the current time into the traffic flow prediction model corresponding to the road to be predicted, and obtain the traffic flow data of the latter time period of the current time.

The traffic flow prediction method and apparatus provided by the embodiments of the present invention pre-prepare the preset neural network model according to the historical traffic flow data of the road, and obtain the historical traffic flow data of the previous time period of the current time of the road to predict the road. a traffic flow prediction model of traffic flow data at a later time period of the current time; when it is required to predict traffic flow data for a future time period of the current time of the road to be predicted, acquiring a traffic flow prediction model corresponding to the road to be predicted, and The historical traffic flow data of the previous time period of the current time of the road to be predicted is input into the traffic flow prediction model to obtain the traffic flow data of the future time period of the road to be predicted. Adopting the technical solution of the invention, due to traffic The flow data has strong nonlinearity and uncertainty, and the neural network model has strong nonlinear prediction ability. Therefore, the neural network model is trained according to the historical traffic flow data of the road, and the traffic flow prediction model can be trained. It is more accurate to predict the traffic flow data of the road in the latter period of the current time according to the traffic flow data of the previous time of the current time of the road.

A traffic flow prediction model generation method includes:

For each road, perform the following steps:

Step a: Obtain historical traffic flow data of consecutive P time periods of the road, wherein the durations of the P time periods are consistent, and the historical traffic flow data corresponding to each time period includes the same number of traffic data;

To traverse the consecutive P time periods, perform the following steps:

Step b: taking the current traversal time period as the previous time period of the current time, inputting the historical traffic flow data corresponding to the current traversing time period as input data into the to-be-determined neural network model, and obtaining predicted traffic flow data of the latter time period of the current time. ;

Step c: Obtain historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period, and calculate predicted traffic flow data of the latter time period of the current time and the current time The variance value of historical traffic flow data for a period of time;

Step d, determining whether the variance value is less than or equal to a preset first variance threshold, and if so, executing step e, if not, performing step f;

Step e: determining the to-be-determined neural network model as a traffic flow prediction model corresponding to the road, and establishing a correspondence between the road and the traffic flow prediction model, and saving, ending the process;

Step f: adjusting parameters of the to-be-determined neural network model according to the variance value;

Step g: Taking the next period of the current traversal period in the P periods as the current traversal period, and performing step b according to the pending neural network model after adjusting the parameters.

A traffic flow prediction model generating device includes:

a third acquiring unit, configured to acquire historical traffic flow data of consecutive P time periods of the road, wherein the durations of the P time periods are consistent, and the historical traffic flow data corresponding to each time period includes the same number of traffic data;

a second traversal unit, configured to traverse the consecutive P time periods;

a second input unit, configured to use the current traversal period as a previous period of the current time, and input the historical traffic flow data corresponding to the current traversed period as input data into the to-be-determined neural network model, to obtain a prediction of the current time period Traffic flow data;

a fourth acquiring unit, configured to acquire historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period;

a second variance value determining unit, configured to calculate a variance value of the predicted traffic flow data of the latter time period of the current time and the historical traffic flow data of the latter time period of the current time;

a fourth determining unit, configured to determine whether the variance value is less than or equal to a preset first variance threshold; if yes, triggering a second traffic flow prediction model determining unit; if not, triggering a second parameter adjusting unit;

a second traffic flow prediction model determining unit, configured to determine the to-be-determined neural network model as a traffic flow prediction model corresponding to the road, and establish a correspondence between the road and the traffic flow prediction model, and save the relationship;

a second parameter adjustment unit, configured to adjust parameters of the to-be-determined neural network model according to the variance value;

The second triggering unit is configured to use the next time period of the current traversal period in the P time periods as the current traversal time period, and trigger the second input unit according to the to-be-determined neural network model after adjusting the parameter.

The method and device for generating a traffic flow prediction model according to an embodiment of the present invention trains a neural network model according to historical traffic flow data of the road to obtain a current traffic flow data of a previous time period of the current time of the road to predict the current road Traffic flow prediction model of traffic flow data in the latter period of time; because traffic flow data has strong nonlinearity and uncertainty, and neural network model has strong nonlinear prediction ability, therefore, according to the historical traffic of the road The flow data trains the neural network model, and the traffic flow prediction model obtained by the training can accurately predict the traffic flow data of the road in the latter period of the current time according to the traffic flow data of the previous time of the current time of the road.

DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below. Obviously, the drawings in the following description are only It is a certain embodiment of the present invention, and other drawings can be obtained from those skilled in the art without any inventive labor.

1 is a schematic diagram of a topology of a BP neural network according to an embodiment of the present invention;

2 is a schematic diagram of reverse transmission of BP algorithm error according to an embodiment of the present invention;

3 is a flowchart of a traffic flow prediction method according to Embodiment 1 of the present invention;

4 is a flowchart of a method for obtaining a traffic flow prediction model corresponding to a road according to Embodiment 1 of the present invention;

FIG. 5 is a schematic structural diagram of a traffic flow prediction apparatus according to Embodiment 3 of the present invention; FIG.

FIG. 6 is a schematic structural diagram of a training module of a traffic flow prediction apparatus according to Embodiment 3 of the present invention.

detailed description

In the technical solution of the embodiment of the present invention, the preset neural network model is trained according to the historical traffic flow data of the road in advance, and the current time of the road can be predicted according to the historical traffic flow data of the previous time period of the current time of the road. a traffic flow prediction model for traffic flow data in a later period; when it is required to predict traffic flow data for a future period of the road to be predicted, obtaining a traffic flow prediction model corresponding to the road to be predicted, and the previous one of the road to be predicted The historical traffic flow data of the time period is input into the traffic flow prediction model to obtain the traffic flow data of the future predicted time of the road to be predicted. With the technical solution of the present invention, since the traffic flow data has strong nonlinearity and uncertainty, and the neural network model has strong nonlinear prediction capability, the neural network model is trained according to the historical traffic flow data of the road. The traffic flow prediction model obtained by the training can accurately predict the traffic flow data of the road in the latter period of the current time according to the traffic flow data of the previous time of the current time of the road.

In the embodiment of the present invention, a neural network model (for example, BP (Error Back-Propagation Training) neural network model) is used to perform short-term traffic flow prediction, that is, the massive historical traffic flow data corresponding to the road is performed. Offline training and learning to obtain a traffic flow prediction model corresponding to each road. The weight matrix of the BP neural network model is used to express the road at the current time (such as t _c ) in the previous period (such as [t _{c-N+ 1} , t _c ]) The relationship between the traffic flow data and the traffic flow data of the latter period of the current time (eg [t _c+1 , t _c+L ]).

In order to facilitate the understanding of the BP neural network model by those skilled in the art, the following describes the BP neural network model in detail:

The BP neural network model includes an input layer, a hidden layer and an output layer. As shown in FIG. 1, the input layer and the hidden layer are associated by a weight matrix V, and the hidden layer and the output layer are associated by a weight matrix W, such as The input data of the input layer is multiplied by the weight matrix V, and the obtained result is the output result of the hidden layer; the output result of the hidden layer is multiplied by the weight matrix W, and the obtained result is the output data of the output layer.

Specifically, it is assumed that the input data of the input layer is X={x1, x2, x3, ..., xn}, and the weight matrix V is n rows and m columns, and the output result of the hidden layer is Y=X*V={y1, Y2, y3, ..., ym}, the weight matrix W is m rows and 1 column, then the output data of the output layer O=Y*W={o1,o2,o3,...,ol}, assuming the true corresponding to the output data The data D = {d1, d2, d3, ..., dl}, the weight matrix V and W are as follows:

The training process of the BP neural network model includes the forward propagation process of the sample and the error back propagation process:

The sample forward propagation process is specifically as follows: input data -> input layer -> hidden layer -> output layer -> output data.

The error back propagation process is specifically: error of the output data -> hidden layer -> input layer.

The above error back propagation process is shown in Figure 2.

The error of the output data is the variance value E of the output data and the real data, and the specific calculation is as follows:

If E is less than or equal to the preset variance threshold Emin, it indicates that the output data O obtained by inputting the input data X into the BP neural network model is closer to the real data D of the output data, and is more accurate. Therefore, the BP neural network at this time is used. The model is determined to be a BP neural network model that meets the requirements;

If E is greater than the preset variance threshold Emin, it indicates that the output data O obtained by inputting the input data X into the BP neural network model is largely different from the real data D of the output data. Therefore, the BP neural network needs to be based on the variance value E. The weight matrix V and W in the model are adjusted. The specific adjustments can be as follows:

Assuming that the learning rate is η in the BP neural network model, the weight matrices V and W are adjusted as follows:

w' _j,k =w _j,k +Δw _j,k

Where 1 ≤ _j ≤ m, 1 ≤ _k ≤ l; ΔW _{j, k} is the rate of change of W _j,k ;

v' _i,j =v _i,j +Δv _i,j

Where 1 ≤ i ≤ n, 1 ≤ _j ≤ m; ΔW _{i, j} is the rate of change of W _{i, j} ;

The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

Embodiment 1

Embodiment 1 of the present invention provides a traffic flow prediction method, and a flowchart thereof is shown in FIG. 3, and the method includes the following steps:

Step 301: Obtain historical traffic flow data of a previous time period of the current time of the road to be predicted for the road to be predicted;

The previous time period of the current time refers to a time period including the current time and the current time forward. If the current time is assumed to be tc, the previous time period of the current time is [t _c-N+1 , t _c ]. In general, the traffic data of the road is released every 2 minutes. Therefore, if the previous period is half an hour, the traffic data released in the previous period is 15. Assuming that the time interval for publishing traffic data is Tinterval, the number of traffic data for a road that is released all day is (24*60)/Tinterval.

Step 302: Obtain a traffic flow prediction model corresponding to the road to be predicted from a correspondence between the pre-stored road and the traffic flow prediction model; wherein the traffic flow prediction model corresponding to the road is pre-predicted according to historical traffic flow data of the road The set neural network model is trained to obtain a traffic flow prediction model for predicting traffic flow data of the current time period of the current time according to historical traffic flow data of the previous time period of the current time of the road;

In the embodiment of the present invention, the correspondence between the pre-stored road and the traffic flow prediction model may be the correspondence between the ID/name of the road and the ID of the traffic flow prediction model. In step 302, the traffic flow prediction model corresponding to the road to be predicted is obtained, and the ID of the traffic flow prediction model corresponding to the ID/name of the road to be predicted is obtained from the foregoing correspondence according to the ID/name of the road to be predicted. And obtaining a corresponding traffic flow prediction model from the pre-stored traffic flow prediction model according to the obtained ID of the traffic flow prediction model.

In the embodiment of the present invention, the selected historical traffic flow data may be accumulated traffic data of half a month, one month, two months, one quarter or half year, and the historical traffic flow data selected for how long is selected according to actual needs. , there is no strict limit here.

Step 303: Input historical traffic flow data of a previous time period of the current time into a traffic flow prediction model corresponding to the road to be predicted, and obtain traffic flow data of a subsequent time period of the current time.

In the embodiment of the present invention, the duration of the previous period of the current time may be the same as or different from the duration of the latter period of the current time. Since the time interval for issuing the traffic data is the same, if the duration of the previous time period of the current time coincides with the duration of the latter time period of the current time, the number of historical traffic flow data included in the previous time period of the current time is the same as the current time. The number of predicted traffic flow data included in the time period is the same; if the duration of the previous time period of the current time is greater than the time length of the next time period of the current time, the number of historical traffic flow data included in the previous time period of the current time is greater than the current time interval The number of predicted traffic flow data included; if the duration of the previous time period of the current time is less than the duration of the next time period of the current time, the number of historical traffic flow data included in the previous time period of the current time is smaller than the prediction included in the latter time period of the current time The number of traffic flow data.

Specifically, in the foregoing step 302, the preset neural network model is trained according to historical traffic flow data of the road, and the current traffic time data of the previous time period of the current time of the road is predicted to predict the current time of the road. A traffic flow prediction model of traffic flow data for a period of time; can be obtained by the following method, as shown in Figure 4:

Step 401, obtaining historical traffic flow data of consecutive P time periods of the road, and traversing the consecutive P time periods, and performing step 402;

The durations of the P periods are the same, and the historical traffic flow data corresponding to each period includes the same number of traffic data, and the P is an integer greater than 1.

The adjacent periods of the above P periods may or may not overlap partially. The adjacent time periods are partially overlapped, for example, the first time period is 9:00-9:30, the second time period is 9:02-9:32, the third time period is 9:04-9:34, and the fourth time period is 9:04-9:34. -9:36, and so on. The adjacent time periods do not overlap, for example, the first time period is 9:02-9:30; the second time period is 9:32-10:00; the third time period is 10:02-10:30, and so on.

Taking the partial overlap of adjacent periods as an example, the historical traffic data streams respectively corresponding to the above consecutive P periods are used as the P group input data of the neural network:

Where: 0 ≤ t _s < 24 × 60; |t _{s+1 -} t _s | = t _interval ; ₁ ≤ _S ≤ (24 × 60) / t _interval .

Step 402: The current traversal period (assuming that the current traversal period is the ith period) is used as the previous period of the current time, and the historical traffic flow data corresponding to the current traversed period is input as input data into the pending neural network model, and the current Predicted traffic flow data for the next period of time (if the i-th period is the first period of the P period, then the pending neural network model at this time is a preset neural network model, if the i-th period is not the first period of the P period In a time period, the pending neural network model at this time is a neural network model adjusted according to the variance value of the i-1th period, and the current time is obtained. Predicting traffic flow data for the next period of time, and then performing step 403;

The duration of the previous period of the current time in this step 402 may be the same as or different from the duration of the latter period of the current time. Since the time interval for issuing the traffic data is the same, if the duration of the previous time period of the current time coincides with the duration of the latter time period of the current time, the number of historical traffic flow data included in the previous time period of the current time is the same as the current time. The time period contains the same number of predicted traffic flow data. That is, the number n of data of the input data X in the neural network model is the same as the number of data of the output data O.

Step 403: Obtain historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period, and calculate predicted traffic flow data of the subsequent time period of the current time and the current time The variance value of the historical traffic flow data for a period of time, after which step 404 is performed;

Taking the aforementioned Dinput as an example, after the historical traffic flow data of the P group of the Dinput is input to the neural network model, the obtained P group predicted traffic flow data are respectively

And the P group predicts the traffic flow data corresponding to the real historical traffic flow data respectively as Doutput as follows:

It is assumed that the current traversal period in step 402 is the pth time period (where p is less than or equal to P), and the pth time period is the previous time period of the current time, and the historical traffic flow data corresponding to the pth time period is

Predicted traffic flow data after a period of historical traffic flow data of the p-th input period neural network model obtained at the current time is O ^p, after the current time period corresponding to a real historical traffic data

details as follows:

with

among them

Step 403 calculates the variance value of the predicted traffic flow data and the corresponding historical traffic flow data for the next time period of the current time:

When the adjacent time periods of the P time periods are not overlapped, the historical traffic flow data corresponding to the next time period of the current time is obtained from the traffic flow data of the time period after the time period of the current traversal, and specifically includes:

Case 1. If the duration of the current traversal period (eg, the ith period) is greater than the duration of the next period of the current time, from the traffic flow data of the next period of the current traversal period (ie, the i+1th period) Obtaining historical traffic flow data corresponding to a later period of the current time.

For example, taking traffic data every 2 minutes as an example, the first time period is 9:02-9:30, the corresponding historical traffic flow data is {P1, P1,..., P15}; the second time period is 9:32. -10:00, the corresponding historical traffic flow data is {P16, P17, ..., P30}; the third time period is 10:02-10:30, and the corresponding historical traffic flow data is {P31, P32, ..., P45} ,....; then the first time period is the previous time period of the current time, and the current time period is 9:32-9:50, then the current time period is obtained from the corresponding historical traffic flow data in the second time period. The corresponding historical traffic data stream is {P16, P17, ..., P25}.

Case 2: If the duration of the current traversal period (assuming the ith period) is equal to the duration of the next period of the current time, the traffic flow data of the next period of the current traversal period (ie, the (i+1)th period) is taken as The historical traffic flow data corresponding to the next time period of the current time.

For example, taking traffic data every 2 minutes as an example, the first time period is 9:02-9:30, the corresponding historical traffic flow data is {P1, P1,..., P15}; the second time period is 9:32. -10:00, the corresponding historical traffic flow data is {P16, P17, ..., P30}; the third time period is 10:02-10:30, and the corresponding historical traffic flow data is {P31, P32, ..., P45} ,....; then the first time period is the previous time period of the current time, and the current time period is 9:32-10:00, then the historical traffic flow data corresponding to the second time period is used as the corresponding time period of the current time. The historical traffic data stream is {P16, P17,..., P30}.

Case 3: if the duration of the current traversal period (assuming the ith period) is less than the duration of the next period of the current time, at least two consecutive periods from the current traversal period (eg, the i+1th period, the i-th Obtaining historical traffic corresponding to the latter period of the current time in the traffic flow data of +2 period, etc.) Stream data.

For example, taking traffic data every 2 minutes as an example, the first time period is 9:02-9:30, the corresponding historical traffic flow data is {P1, P1,..., P15}; the second time period is 9:32. -10:00, the corresponding historical traffic flow data is {P16, P17, ..., P30}; the third time period is 10:02-10:30, and the corresponding historical traffic flow data is {P31, P32, ..., P45} ,....; then the first time period is the previous time period of the current time, and the current time period is 9:32-10:12, then the current time flow data obtained from the second time period and the third time period is obtained. The historical traffic data flow corresponding to the time period after the time is {P16, P17, ..., P36}.

Step 404, determining whether the variance value calculated in step 403 is less than or equal to the preset first variance threshold; if yes, executing step 405, and if not, executing step 406;

Step 405: Determine the to-be-determined neural network model as a traffic flow prediction model corresponding to the road, and establish a correspondence between the road and the traffic flow prediction model, and save, and end the process;

Step 406, adjusting the parameters of the pending neural network model according to the variance value, and then performing step 407;

In this step 406, specifically, the adjustment of the weight matrix V and W of the neural network model according to the variance value can be referred to the foregoing content, and details are not described herein again.

Step 407: The next time period of the current traversal period in the P time periods is used as the current traversal time period, and step 402 is performed according to the to-be-determined neural network model after adjusting the parameter.

Preferably, for the historical traffic flow data traversing to the last time period, if the calculated variance value is still greater than or equal to the preset first variance threshold, between step 406 and step 407, the method further includes:

Step 406a, it is determined whether the current traversal time period is the last time period of the P time periods, and if so, step 406b is performed, and if not, step 407 is performed;

Step 406b, calculating the sum of the variance values corresponding to the P time periods, and then performing step 406c;

Step 406c, determining whether the sum value is less than or equal to a preset second variance threshold, and if yes, performing step 405, wherein the second variance threshold is greater than the first variance threshold; if not, according to the step The pending neural network model after adjusting the parameters obtained in step 406 re-traverses the P time periods.

In the technical solution of the embodiment of the present invention, the preset neural network model is trained according to the historical traffic flow data of the road in advance, and the current time of the road can be predicted according to the historical traffic flow data of the previous time period of the current time of the road. a traffic flow prediction model for traffic flow data in a later period; when it is required to predict traffic flow data for a future period of the road to be predicted, obtaining a traffic flow prediction model corresponding to the road to be predicted, and the previous one of the road to be predicted The historical traffic flow data of the time period is input into the traffic flow prediction model to obtain the traffic flow data of the future predicted time of the road to be predicted. With the technical solution of the present invention, since the traffic flow data has strong nonlinearity and uncertainty, and the neural network model has strong nonlinear prediction capability, the neural network model is trained according to the historical traffic flow data of the road. The traffic flow prediction model obtained by the training can accurately predict the traffic flow data of the road in the latter period of the current time according to the traffic flow data of the previous time of the current time of the road.

Embodiment 2

To further enable those skilled in the art to understand the present solution, a specific example will be described below.

Assume that the road to be predicted is R0, the historical traffic flow data is published every 2 minutes, and the selected historical traffic flow data is 150 historical traffic data from 9:02 to 14:00. The preset initial BP neural network model The parameters are as follows:

The input layer allows the number of input data to be N=15, the number of output data in the middle layer (hidden layer) is M=15, the number of output data in the output layer is L=15, the first variance threshold E _min = 0.02, and training learning Rate η = 0.1;

The weight matrix V and W of the BP neural network model are as follows:

Since the input layer of the neural network model allows the input traffic data to be N=15, 15 historical traffic data corresponding to each half hour can be used as a set of input data, and overlap for 28 minutes according to adjacent time periods (of course, the overlap duration can also be smaller than In 28 minutes), 5 hours can be divided into 136 time periods, such as the first time period is 9:02-9:30, the second time period is 9:04-9:32, and the third time period is 9:06-9. :34....; At this time, the historical traffic flow data set corresponding to each time period is obtained.

Since the number of data outputted by the output layer of the neural network model is 15, the first time period is 9:02-9:30, the second time period is 9:04-9:32, and the third time period is 9:06. -9:34.... As the input historical traffic flow data of the neural network model, predict the predicted traffic with time period 9:32-10:00, 10:02-10:30, 10:32-11:00.... The flow data predicts the historical traffic flow data corresponding to the traffic flow data (ie, the actual traffic flow data) as follows:

After the historical traffic flow data and the parameters of the preset BP neural network are prepared, the steps 401 to 407 in the first embodiment are started to obtain the BP neural network model corresponding to the road R ₀ , that is, the historical data flow set. The process of offline training and learning is as follows:

Step 401: Acquire a historical traffic flow data set of consecutive 136 time periods of the road R ₀ , and traverse from the first time period;

Step 402: The first time period (that is, the current traversal time period) is used as the previous time period of the current time, and the historical traffic flow data corresponding to the first time period is

And inputting historical traffic flow data corresponding to the first time period into a pending neural network model (that is, the preset BP neural network model described above), and obtaining predicted traffic flow data for a later period of the current time;

In the step 402, the predicted traffic flow data of the next time period of the current time is obtained, which can be implemented by the following two steps:

The first step is to multiply the historical traffic flow data corresponding to the first time period and the weight matrix V of the neural network model to obtain a hidden layer output result:

Y ¹ =X ¹ ×V;

Step 2: Multiply the result of the hidden layer output by the weight matrix W to obtain the result of the output layer:

O ¹ = Y ¹ × W;

Step 403: Acquire historical traffic flow data corresponding to a time period after the current time

For d ¹ and O ¹ , the variance values of the two are calculated as:

Step 404: Determine whether the variance value E ¹ is greater than or equal to the first variance threshold 0.02, and if so, determine the preset neural network model as the traffic flow prediction model of the road R0, and if not, proceed to step 406;

Step 406: Adjust the pending neural network parameter according to the variance value E ¹ (because it is the first time period, therefore, here specifically adjust the preset neural network parameters);

Specifically, the above X ¹ , d ¹ and O ¹ are adjusted to the weight matrix of the neural network model as follows:

w' _j,k =w _j,k +Δw _j,k →w _j,k ;

v' _i,j =v _i,j +Δv _i,j →v _i,j ;

The adjusted weight matrices V and W are:

Step 406a: determining that the first time period of the current traversal is not the last time period of 136 time periods, step 407 is performed;

Step 407: Jump to step 402 by using the next period 9:04-9:32 of the 9:02-9:30 period of the 136 periods as the current traversal period and the pending neural network model after adjusting the parameters.

Through the foregoing process, the weight matrix V and W of the neural network model corresponding to the road R0 are finally obtained as follows:

For example, the current time is t _c =10:30:00 (ie 10:30 am), and the historical traffic flow data of the previous time period of the current time is as follows:

That is, as shown in Table 1 below.

Numbering	Time point		input value
1	10:02:00	54

2	10:04:00	55
3	10:06:00	55
4	10:08:00	59
5	10:10:00	60
6	10:12:00	62
7	10:14:00	58
8	10:16:00	53
9	10:18:00	52
10	10:20:00	54
11	10:22:00	57
12	10:24:00	57
13	10:26:00	57
14	10:28:00	57
15	10:30:00	59

Table 1)

Data set

As the input data of the neural network model, the resulting output data is as follows:

That is, as shown in Table (2):

Numbering	Time point		output value
1	10:32:00	54
2	10:34:00	53
3	10:36:00	51
4	10:38:00	50
5	10:40:00	49
6	10:42:00	48

7	10:44:00	48
8	10:46:00	47
9	10:48:00	47
10	10:50:00	48
11	10:52:00	49
12	10:54:00	51
13	10:56:00	51
14	10:58:00	52
15	11:00:00	53

Table 2)

Embodiment 3

Based on the foregoing same concept of a traffic flow prediction method, a third embodiment of the present invention provides a traffic flow prediction device, which is shown in FIG. 5 and includes a training module 51, a storage module 52, and a historical traffic data acquisition module 53. The traffic flow prediction model acquisition module 54 and the prediction module 55, wherein:

The training module 51 is configured to train the preset neural network model according to the historical traffic flow data of the road in advance, and obtain the historical traffic flow data of the previous time period of the current time of the road to predict the time period of the current time of the road. Traffic flow prediction model for traffic flow data;

a storage module 52, configured to store a correspondence between the road and the traffic flow prediction model obtained by the training module 51;

The historical traffic data obtaining module 53 is configured to acquire historical traffic flow data of a previous time period of the current time of the road to be predicted for the road to be predicted;

The traffic flow prediction model acquisition module 54 is configured to obtain, from the correspondence between the road and the traffic flow prediction model prestored by the storage module 52, the traffic flow prediction model corresponding to the road to be predicted;

The prediction module 55 is configured to input the historical traffic flow data of the previous time period of the current time into the traffic flow prediction model corresponding to the road to be predicted, and obtain the traffic of the next time period of the current time. Stream data.

Preferably, the training module 51, as shown in FIG. 6, specifically includes:

a first acquisition unit 5101, configured to acquire historical traffic flow data of consecutive P time periods of the road, wherein the durations of the P time periods are consistent, and the historical traffic flow data corresponding to each time period includes the same number of traffic data;

a first traversal unit 5102, configured to traverse the consecutive P time periods;

The first input unit 5103 is configured to use the current traversal period as the previous time period of the current time, and input the historical traffic flow data corresponding to the current traversed time period as input data into the to-be-determined neural network model, to obtain the next time period of the current time. Forecast traffic flow data;

a second obtaining unit 5104, configured to acquire historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period;

a first variance value determining unit 5105, configured to calculate a variance value of the predicted traffic flow data of the latter time period of the current time and the historical traffic flow data of the latter time period of the current time;

The first determining unit 5106 is configured to determine whether the variance value is less than or equal to a preset first variance threshold; if yes, triggering the first traffic flow prediction model determining unit 5107; if not, triggering the first parameter adjusting unit 5108;

The first traffic flow prediction model determining unit 5107 is configured to determine the to-be-determined neural network model as a traffic flow prediction model corresponding to the road, and establish a correspondence between the road and the traffic flow prediction model, and save the relationship;

The first parameter adjustment unit 5108 is configured to adjust parameters of the to-be-determined neural network model according to the variance value.

The first triggering unit 5109 is configured to use the next time period of the current traversal period in the P time periods as the current traversal time period, and trigger the first input unit 5103 according to the to-be-determined neural network model after adjusting the parameter.

Preferably, the training module 51 further includes:

The second determining unit 5110 is configured to determine whether the current traversal period is the last period of the P periods, and if yes, trigger the first variance sum value determining unit 5111; if not, trigger the first trigger Unit 5109;

a first variance sum value determining unit 5111, configured to calculate a sum value of the variance values corresponding to the P time periods;

The third determining unit 5112 is configured to determine whether the sum value is less than or equal to a preset second variance threshold, where the second variance threshold is greater than the first variance threshold; if yes, triggering the first traffic flow prediction model to determine The unit 5107, if not, triggers the first traversal unit 5102 according to the pending neural network model after the parameter adjustment unit 5108 adjusts the parameter.

Preferably, if the time periods of the P time periods are adjacent and do not overlap, the second acquiring unit 5104 is specifically configured to:

If the duration of the current traversal period is greater than the duration of the next period of the current time, the historical traffic flow data corresponding to the later period of the current time is obtained from the traffic flow data of the next period of the current traversal period;

If the duration of the current traversal period is equal to the duration of the next period of the current time, the traffic flow data of the next period of the current traversal period is used as historical traffic flow data corresponding to the next period of the current time;

If the duration of the current traversal period is less than the duration of the next period of the current time, the historical traffic flow corresponding to the latter period of the current time is obtained from the traffic flow data of the at least two consecutive periods after the current traversal period data.

Embodiment 4

The fourth embodiment of the present invention further provides a method for generating a traffic flow prediction model. For each road, the method includes:

Step a: Obtain historical traffic flow data of consecutive P time periods of the road, wherein the durations of the P time periods are consistent, and the historical traffic flow data corresponding to each time period includes the same number of traffic data;

To traverse the consecutive P time periods, perform the following steps:

Step b: taking the current traversal time period as the previous time period of the current time, inputting the historical traffic flow data corresponding to the current traversing time period as input data into the to-be-determined neural network model, and obtaining predicted traffic flow data of the latter time period of the current time. ;

Step c: Obtain historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period, and calculate predicted traffic flow data of the latter time period of the current time and the current time The variance value of historical traffic flow data for a period of time;

In the step c, the historical traffic flow data corresponding to the next time period of the current time is obtained from the traffic flow data of the time period after the current traversing time period, which may specifically include:

If the duration of the current traversal period is greater than the duration of the next period of the current time, the historical traffic flow data corresponding to the later period of the current time is obtained from the traffic flow data of the next period of the current traversal period;

If the duration of the current traversal period is equal to the duration of the next period of the current time, the traffic flow data of the next period of the current traversal period is used as historical traffic flow data corresponding to the next period of the current time;

If the duration of the current traversal period is less than the duration of the next period of the current time, the historical traffic flow corresponding to the latter period of the current time is obtained from the traffic flow data of the at least two consecutive periods after the current traversal period data.

Step d, determining whether the variance value is less than or equal to the preset first variance threshold, and if so, executing step e, if not, executing step f;

Step e: determining the to-be-determined neural network model as a traffic flow prediction model corresponding to the road, and establishing a correspondence between the road and the traffic flow prediction model, and saving, ending the process;

Step f: adjusting parameters of the to-be-determined neural network model according to the variance value;

Step g: Taking the next period of the current traversal period in the P periods as the current traversal period, and performing step b according to the pending neural network model after adjusting the parameters.

Preferably, between the step f and the step g, the steps f1 - f3 are further included:

Step f1, determining whether the current traversal time period is the last time period of the P time periods, and if so, executing step f2, and if not, executing step g;

Step f2, calculating a sum of variance values corresponding to P time periods;

Step f3, determining whether the sum value is less than or equal to a preset second variance threshold, wherein the second variance threshold is greater than the first variance threshold; if yes, executing step e; if not, according to the step The pending neural network model after f obtained the adjusted parameters re-traverses the P time periods.

For the specific implementation of each step in the foregoing method, refer to the steps shown in FIG. 4 above, and details are not described herein again.

Embodiment 5

Based on the same concept of the traffic flow prediction model generation method provided in the foregoing fourth embodiment, the fifth embodiment of the present invention provides a traffic flow prediction model generation device, where the device includes:

a third acquiring unit, configured to acquire historical traffic flow data of consecutive P time periods of the road, wherein the durations of the P time periods are consistent, and the historical traffic flow data corresponding to each time period includes the same number of traffic data;

a second traversal unit, configured to traverse the consecutive P time periods;

a second input unit, configured to use the current traversal period as a previous period of the current time, and input the historical traffic flow data corresponding to the current traversed period as input data into the to-be-determined neural network model, to obtain a prediction of the current time period Traffic flow data;

a fourth acquiring unit, configured to acquire historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period;

And a fourth acquiring unit, configured to: obtain, after the current time, the traffic flow data of the next time period of the current traversal time period, if the duration of the current traversal time period is greater than the time length of the current time period The historical traffic flow data corresponding to the time period; if the duration of the current traversal period is equal to the duration of the next time period of the current time, the traffic flow data of the next time period of the current traversal time period is used as the corresponding time period of the current time Historical traffic flow data; if the duration of the current traversal period is less than the duration of the next time period of the current time, the traffic flow data of at least two consecutive time periods after the current traversal time period is acquired and the time period after the current time is acquired Corresponding historical traffic flow data.

a second variance value determining unit, configured to calculate a variance value of the predicted traffic flow data of the latter time period of the current time and the historical traffic flow data of the latter time period of the current time;

a fourth determining unit, configured to determine whether the variance value is less than or equal to a preset first variance threshold; If yes, triggering the second traffic flow prediction model determining unit; if not, triggering the second parameter adjusting unit;

a second traffic flow prediction model determining unit, configured to determine the to-be-determined neural network model as a traffic flow prediction model corresponding to the road, and establish a correspondence between the road and the traffic flow prediction model, and save the relationship;

a second parameter adjustment unit, configured to adjust parameters of the to-be-determined neural network model according to the variance value;

The second triggering unit is configured to use the next time period of the current traversal period in the P time periods as the current traversal time period, and trigger the second input unit according to the to-be-determined neural network model after adjusting the parameter.

Preferably, in the embodiment of the present invention, the foregoing apparatus further includes:

a fifth determining unit, configured to determine whether the current traversing period is the last period of the P periods, and if yes, triggering the second variance sum value determining unit; if not, triggering the second trigger unit;

a second variance sum value determining unit, configured to calculate a sum value of the variance values corresponding to the P time periods;

a sixth determining unit, configured to determine whether the sum value is less than or equal to a preset second variance threshold, the second variance threshold is greater than the first variance threshold; if yes, triggering a second traffic flow prediction model determining unit If not, the second traversal unit is triggered according to the pending neural network model after the parameter is adjusted by the second parameter adjustment unit.

For the specific implementation of each unit in the foregoing device, reference may be made to FIG. 6 , and details are not described herein again.

The method and device for generating a traffic flow prediction module according to an embodiment of the present invention trains a neural network model according to historical traffic flow data of the road, so as to obtain a current traffic flow data of a previous time period of the current time of the road to predict the current road. Traffic flow prediction model of traffic flow data in the latter period of time; because traffic flow data has strong nonlinearity and uncertainty, and neural network model has strong nonlinear prediction ability, therefore, according to the historical traffic of the road The flow data trains the neural network model, and the traffic flow prediction model obtained by the training can accurately predict the traffic flow data of the road in the latter period of the current time according to the traffic flow data of the previous time of the current time of the road.

Those skilled in the art will appreciate that embodiments of the present invention can 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 a combination of software and hardware. Moreover, the invention can 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, etc.) including computer usable program code.

The present invention has been described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (system), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or FIG. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine for the execution of instructions for execution by a processor of a computer or other programmable data processing device. Means for implementing the functions specified in one or more of the flow or in a block or blocks of the flow chart.

The computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device. The apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.

These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device. The instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.

While the preferred embodiment of the invention has been described, it will be understood that Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and the modifications and

It is apparent that those skilled in the art can make various modifications and variations to the invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention are claimed in the present invention The invention is also intended to cover such modifications and variations within the scope of the invention.

Claims (12)

1. A traffic flow prediction method, the method comprising:

   Obtaining historical traffic flow data of a previous time period of the current time of the road to be predicted for the road to be predicted;

   Obtaining a traffic flow prediction model corresponding to the road to be predicted from a corresponding relationship between the pre-stored road and the traffic flow prediction model; wherein the traffic flow prediction model corresponding to the road is a pre-set nerve according to historical traffic flow data of the road in advance The network model is trained to obtain a traffic flow prediction model for predicting traffic flow data for a subsequent period of the current time of the road according to historical traffic flow data of a previous time period of the current time of the road;

   The historical traffic flow data of the previous time period of the current time is input into the traffic flow prediction model corresponding to the road to be predicted, and the traffic flow data of the latter time period of the current time is obtained.
2. The method according to claim 1, wherein the preset neural network model is trained in advance based on historical traffic flow data of the road, and the road is predicted based on historical traffic flow data of a previous time period of the current time of the road. The traffic flow prediction model of the traffic flow data in the latter period of the current time includes:

   Step a: Obtain historical traffic flow data of consecutive P time periods of the road, wherein the durations of the P time periods are consistent, and the historical traffic flow data corresponding to each time period includes the same number of traffic data;

   To traverse the consecutive P time periods, perform the following steps:

   Step b: taking the current traversal time period as the previous time period of the current time, inputting the historical traffic flow data corresponding to the current traversing time period as input data into the to-be-determined neural network model, and obtaining predicted traffic flow data of the latter time period of the current time. ;

   Step c: Obtain historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period, and calculate predicted traffic flow data of the latter time period of the current time and the current time The variance value of historical traffic flow data for a period of time;

   Step d, determining whether the variance value is less than or equal to the preset first variance threshold, and if so, executing step e, if not, executing step f;

   Step e: determining the to-be-determined neural network model as a traffic flow prediction model corresponding to the road, and establishing a correspondence between the road and the traffic flow prediction model, and saving, ending the process;

   Step f: adjusting parameters of the to-be-determined neural network model according to the variance value;

   Step g: Taking the next period of the current traversal period in the P periods as the current traversal period, and performing step b according to the pending neural network model after adjusting the parameters.
3. The method of claim 2, further comprising: between the step f and the step g,

   Step f1, determining whether the current traversal time period is the last time period of the P time periods, and if so, executing step f2, and if not, executing step g;

   Step f2, calculating a sum of variance values corresponding to P time periods;

   Step f3, determining whether the sum value is less than or equal to a preset second variance threshold, wherein the second variance threshold is greater than the first variance threshold; if yes, performing step e; if not, obtaining according to step f The pending neural network model after adjusting the parameters re-traverses the P time periods.
4. The method according to claim 2, wherein if each of the P time periods is adjacent and does not overlap, then the current time is obtained from the traffic flow data of the time period after the current traversed time period in step c The historical traffic flow data corresponding to a period of time includes:

   If the duration of the current traversal period is greater than the duration of the next period of the current time, the historical traffic flow data corresponding to the later period of the current time is obtained from the traffic flow data of the next period of the current traversal period;

   If the duration of the current traversal period is equal to the duration of the next period of the current time, the traffic flow data of the next period of the current traversal period is used as historical traffic flow data corresponding to the next period of the current time;

   If the duration of the current traversal period is less than the duration of the next period of the current time, the historical traffic flow corresponding to the latter period of the current time is obtained from the traffic flow data of the at least two consecutive periods after the current traversal period data.
5. A method for generating a traffic flow prediction model, comprising:

   For each road, perform the following steps:

   Step a: Obtain historical traffic flow data of consecutive P time periods of the road, wherein the durations of the P time periods are consistent, and the historical traffic flow data corresponding to each time period includes the same number of traffic data;

   To traverse the consecutive P time periods, perform the following steps:

   Step b: taking the current traversal time period as the previous time period of the current time, inputting the historical traffic flow data corresponding to the current traversing time period as input data into the to-be-determined neural network model, and obtaining predicted traffic flow data of the latter time period of the current time. ;

   Step c: Obtain historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period, and calculate predicted traffic flow data of the latter time period of the current time and the current time The variance value of historical traffic flow data for a period of time;

   Step d, determining whether the variance value is less than or equal to the preset first variance threshold, and if so, executing step e, if not, executing step f;

   Step e: determining the to-be-determined neural network model as a traffic flow prediction model corresponding to the road, and establishing a correspondence between the road and the traffic flow prediction model, and saving, ending the process;

   Step f: adjusting parameters of the to-be-determined neural network model according to the variance value;

   Step g: Taking the next period of the current traversal period in the P periods as the current traversal period, and performing step b according to the pending neural network model after adjusting the parameters.
6. The method of claim 5, further comprising: between step f and step g,

   Step f1, determining whether the current traversal time period is the last time period of the P time periods, and if so, executing step f2, and if not, executing step g;

   Step f2, calculating a sum of variance values corresponding to P time periods;

   Step f3, determining whether the sum value is less than or equal to a preset second variance threshold, wherein the second variance threshold is greater than the first variance threshold; if yes, performing step e; if not, obtaining according to step f The pending neural network model after adjusting the parameters re-traverses the P time periods.
7. A traffic flow prediction device, characterized in that the device comprises:

   a training module, configured to pre-prepare the preset neural network model according to historical traffic flow data of the road, and obtain the historical traffic flow data of the previous time period of the current time of the road to predict the article a traffic flow prediction model of traffic flow data for a later period of the current time of the road;

   a storage module, configured to store a correspondence between the road and a traffic flow prediction model obtained by the pre-training module;

   a historical traffic data obtaining module, configured to acquire historical traffic flow data of a previous time period of the current time of the road to be predicted for the road to be predicted;

   a traffic flow prediction model acquisition module, configured to acquire a traffic flow prediction model corresponding to the road to be predicted from a correspondence between a road and a traffic flow prediction model prestored by the storage module;

   The prediction module is configured to input the historical traffic flow data of the previous time period of the current time into the traffic flow prediction model corresponding to the road to be predicted, and obtain the traffic flow data of the latter time period of the current time.
8. The traffic flow prediction apparatus according to claim 7, wherein the training module specifically includes:

   a first acquisition unit, configured to acquire historical traffic flow data of consecutive P time periods of the road, wherein the durations of the P time periods are consistent, and the historical traffic flow data corresponding to each time period includes the same number of traffic data;

   a first traversal unit, configured to traverse the consecutive P time periods;

   a first input unit, configured to use the current traversal period as a previous time period of the current time, and input historical traffic flow data corresponding to the current traversed time period as input data into the to-be-determined neural network model, to obtain a prediction of the current time period Traffic flow data;

   a second acquiring unit, configured to acquire historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period;

   a first variance value determining unit, configured to calculate a variance value of the predicted traffic flow data of the latter time period of the current time and the historical traffic flow data of the latter time period of the current time;

   a first determining unit, configured to determine whether the variance value is less than or equal to a preset first variance threshold; if yes, triggering a first traffic flow prediction model determining unit; if not, triggering a first parameter adjusting unit;

   a first traffic flow prediction model determining unit, configured to determine the pending neural network model as a traffic flow prediction model corresponding to the road, and establishing a correspondence between the road and the traffic flow prediction model, and saving;

   a first parameter adjustment unit, configured to adjust parameters of the to-be-determined neural network model according to the variance value;

   The first triggering unit is configured to use the next time period of the current traversal period in the P time periods as the current traversal time period, and trigger the first input unit according to the to-be-determined neural network model after adjusting the parameter.
9. The traffic flow prediction apparatus according to claim 8, wherein the training module further comprises:

   a second determining unit, configured to determine whether the current traversal period is the last period of the P periods, and if yes, triggering the first variance sum value determining unit; if not, triggering the first trigger unit;

   a first variance sum value determining unit, configured to calculate a sum value of the variance values corresponding to the P time periods;

   a third determining unit, configured to determine whether the sum value is less than or equal to a preset second variance threshold, the second variance threshold is greater than the first variance threshold; if yes, triggering the first traffic flow prediction model determining unit If not, the first traversal unit is triggered according to the pending neural network model after the parameter is adjusted by the first parameter adjustment unit.
10. The traffic flow prediction apparatus according to claim 8, wherein the second acquisition unit is specifically configured to: if the time periods of the P time periods are adjacent and do not overlap:

    If the duration of the current traversal period is greater than the duration of the next period of the current time, the historical traffic flow data corresponding to the later period of the current time is obtained from the traffic flow data of the next period of the current traversal period;

    If the duration of the current traversal period is equal to the duration of the next period of the current time, the traffic flow data of the next period of the current traversal period is used as historical traffic flow data corresponding to the next period of the current time;

    If the duration of the current traversal period is less than the duration of the next period of the current time, the historical traffic flow corresponding to the latter period of the current time is obtained from the traffic flow data of the at least two consecutive periods after the current traversal period data.
11. A traffic flow prediction model generating device, comprising:

    a third acquiring unit, configured to acquire historical traffic flow data of consecutive P time periods of the road, wherein the durations of the P time periods are consistent, and the historical traffic flow data corresponding to each time period includes the same number of traffic data;

    a second traversal unit, configured to traverse the consecutive P time periods;

    a second input unit, configured to use the current traversal period as a previous period of the current time, and input the historical traffic flow data corresponding to the current traversed period as input data into the to-be-determined neural network model, to obtain a prediction of the current time period Traffic flow data;

    a fourth acquiring unit, configured to acquire historical traffic flow data corresponding to a subsequent time period of the current time from the traffic flow data of the time period after the current traversed time period;

    a second variance value determining unit, configured to calculate a variance value of the predicted traffic flow data of the latter time period of the current time and the historical traffic flow data of the latter time period of the current time;

    a fourth determining unit, configured to determine whether the variance value is less than or equal to a preset first variance threshold; if yes, triggering a second traffic flow prediction model determining unit; if not, triggering a second parameter adjusting unit;

    a second traffic flow prediction model determining unit, configured to determine the to-be-determined neural network model as a traffic flow prediction model corresponding to the road, and establish a correspondence between the road and the traffic flow prediction model, and save the relationship;

    a second parameter adjustment unit, configured to adjust parameters of the to-be-determined neural network model according to the variance value;

    The second triggering unit is configured to use the next time period of the current traversal period in the P time periods as the current traversal time period, and trigger the second input unit according to the to-be-determined neural network model after adjusting the parameter.
12. The device of claim 11 further comprising:

    a fifth determining unit, configured to determine whether the current traversing period is the last period of the P periods, and if yes, triggering the second variance sum value determining unit; if not, triggering the second trigger unit;

    a second variance sum value determining unit, configured to calculate a sum value of the variance values corresponding to the P time periods;

    a sixth determining unit, configured to determine whether the sum value is less than or equal to a preset second variance threshold, the second variance threshold is greater than the first variance threshold; if yes, triggering a second traffic flow prediction model determining unit If not, the second traversal unit is triggered according to the pending neural network model after the parameter is adjusted by the second parameter adjustment unit.

Images