Path prediction method, device, equipment and storage medium

Abstract

The embodiment of the present invention provides a path prediction method, device, equipment and storage medium, the method comprising: determining a first traffic path of a target traffic area; based on a preset path prediction model, using the first traffic path to obtain a predicted traffic path; based on a preset category prediction model, using the first traffic path to obtain a predicted path category; wherein the category prediction model is based on the historical traffic path of the target traffic area, combined with the at least one path category training to obtain; according to the predicted path category, adjusting the predicted traffic path to obtain a target traffic path. The embodiment of the present invention improves the accuracy of the obtained traffic path.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a path prediction method, device, equipment and storage medium. Background Art

[0002] The traffic network is a traffic network composed of various traffic sections. For the convenience of management, the traffic network can be divided into multiple traffic areas, and each traffic area corresponds to a sub-traffic network formed by multiple traffic sections. In order to obtain the driving conditions of each traffic area, the driving direction of the traffic flow in each traffic area, the traffic sections passed during the driving, and the driving speed can be monitored, wherein the traffic flow includes the traffic flow formed by the continuous driving of vehicles such as cars and trains in each traffic section. The driving speed of the traffic flow is the number of vehicles that pass through a traffic section sequence in a unit time, and the unit is vehicles / hour. The traffic path is a structured information, which can include the traffic section sequence formed by the traffic flow passing through the section and the traffic volume corresponding to each section sequence.

[0003] At present, for any traffic area, the historical traffic flow path of the traffic area can be used to make a short-term distribution prediction of the traffic flow path of the traffic area at a future moment, so as to use the obtained target traffic flow path to perform traffic optimization control on the traffic area, for example, using the traffic flow path distribution to estimate the traffic light operation cycle in the traffic area. In the prior art, when making a short-term prediction of the traffic flow path distribution of a traffic area, a path prediction model is usually used, and the current traffic flow path is input into the path prediction model to obtain multiple optional traffic flow paths, and the obtained multiple optional traffic flow paths are combined to form the target traffic flow path of the traffic area.

[0004] However, since the number of optional traffic paths corresponding to a traffic area can reach dozens to thousands, each optional traffic path has a certain error. There is error accumulation when directly combining multiple optional traffic paths to obtain the target traffic path. The deviation between the obtained target traffic path and the actual traffic path distribution is large and not accurate enough. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a path prediction method, device, equipment and storage medium to solve the technical problem that the traffic path obtained by direct superposition has a large deviation from the actual traffic path and the result error of the predicted traffic path is large.

[0006] In a first aspect, an embodiment of the present invention provides a path prediction method, comprising:

[0007] Determine the first traffic flow path of the target traffic area;

[0008] Based on a preset path prediction model, using the first traffic path to obtain a predicted traffic path;

[0009] Based on a preset category prediction model, the first traffic path is used to obtain a predicted path category; wherein the category prediction model is obtained based on the historical traffic path of the target traffic area and combined with the at least one path category training;

[0010] According to the predicted path category, the predicted traffic path is adjusted to obtain a target traffic path.

[0011] In a second aspect, an embodiment of the present invention provides a path prediction device, including:

[0012] A first determination module, used to determine a first vehicle flow path in a target traffic area;

[0013] A path prediction module, configured to obtain a predicted traffic path using the first traffic path based on a preset path prediction model;

[0014] A model prediction module, configured to obtain a predicted path category using the first traffic path based on a preset category prediction model; wherein the category prediction model is obtained based on the historical traffic path of the target traffic area and combined with the at least one path category training;

[0015] The path adjustment module is used to adjust the predicted traffic path according to the predicted path category to obtain a target traffic path.

[0016] In a third aspect, an embodiment of the present invention provides a path prediction device, comprising: a storage component and a processing component; the storage component is used to store one or more computer instructions, and the one or more computer instructions are used to be called by the processing component to execute any path prediction method.

[0017] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program; the computer program enables a computer to implement any path prediction method when executed.

[0018] An embodiment of the present invention provides a path prediction method, which determines a first traffic path in a target traffic area, obtains a predicted traffic path using the first traffic path based on a preset path prediction model, and obtains a predicted path category using the first traffic path based on a preset category prediction model, thereby achieving a double-layer prediction of future traffic paths and path categories. After determining the predicted paths corresponding to multiple optional traffic paths, it can be determined whether the predicted paths meet the predicted path categories. If so, it can be determined that the obtained traffic path is the target traffic path. If not, the traffic path can be adjusted to meet the obtained predicted path categories, so that the traffic path is more in line with the macro predicted path categories, so as to achieve effective prediction and improve the accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A flow chart of an embodiment of a path prediction method provided by an embodiment of the present invention;

[0021] Figure 2 A schematic diagram of a traffic path distribution provided by an embodiment of the present invention;

[0022] Figure 3 A flow chart of another embodiment of a path prediction method provided by an embodiment of the present invention;

[0023] Figure 4 A flow chart of an embodiment of a path training method provided by an embodiment of the present invention;

[0024] Figure 5 A flowchart of an embodiment of a category training method provided by an embodiment of the present invention;

[0025] Figure 6 A flow chart of another embodiment of a path prediction method provided by an embodiment of the present invention;

[0026] Figure 7 A schematic diagram of two traffic path distributions provided by an embodiment of the present invention;

[0027] Figure 8 A schematic diagram of the structure of an embodiment of a path prediction device provided by an embodiment of the present invention;

[0028] Fig. 9A schematic structural diagram of an embodiment of a path prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two, but does not exclude the situation of including at least one.

[0031] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0032] As used herein, the words "if" and "if" may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to identifying", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is identified" may be interpreted as "when determining" or "in response to determining" or "when identifying (stated condition or event)" or "in response to identifying (stated condition or event)", depending on the context.

[0033] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a product or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such a product or system. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the product or system including the elements.

[0034] The embodiments of the present invention are mainly used in intelligent traffic control scenarios. In addition to predicting the traffic path, the path category of the traffic can also be predicted, and the traffic path can be adjusted to conform to the predicted traffic path category, thereby combining the two prediction methods of micro traffic path and macro traffic path category to make the predicted traffic path more accurate and improve the prediction precision.

[0035] At present, when monitoring the traffic conditions of a traffic area, the driving conditions of the traffic flow or pedestrian flow in the traffic area can be monitored. The main method is to monitor the traffic flow in the traffic area and obtain the distribution of the traffic flow path in the traffic area. The traffic flow path is a kind of structured information. For a traffic area, the traffic flow path of the traffic area may include the passing sections of multiple traffic flows formed in the traffic area, and the traffic speeds corresponding to each traffic flow. The traffic speed may include the vehicle data of the road formed by the traffic flow passing through a certain sequence of traffic flow sections in a unit time, and the unit is vehicle / hour. In some monitoring scenarios, it is necessary to obtain the future traffic driving conditions of the traffic area in order to pre-manage the traffic in the traffic area to avoid traffic congestion, traffic accidents and the like. Among them, the traffic flow path of the traffic area can be used as a prediction target to obtain the future driving conditions of the traffic flow in the traffic area.

[0036] In the prior art, when predicting the traffic path of a certain traffic area, the historical traffic path of the traffic area can be obtained, and the historical traffic path can be used to predict the path to obtain the target traffic path, so as to use the target traffic path to optimize the traffic control of the traffic area. When predicting the traffic path of a traffic area, a path prediction model is usually used to input the first traffic path into the path prediction model to obtain multiple optional traffic paths, and the multiple optional traffic paths obtained are combined to form a target traffic path. However, since the multiple optional traffic paths obtained all have prediction errors, when the multiple optional traffic paths are directly combined to form the target traffic path, the error of the target traffic path obtained is large, and the prediction result is not accurate enough. In order to solve this technical problem, the inventor proposed the technical solution of the present application.

[0037] In an embodiment of the present invention, after determining the first traffic flow and the first traffic path of the target traffic area, the predicted traffic path can be obtained by using the first traffic path based on a preset path prediction model. In the technical solution of the present application, in addition to predicting the traffic path of the target traffic area, the path category is also predicted. The predicted path category is obtained by using the first traffic path based on a preset category prediction model, and the category prediction model is based on the historical traffic path of the target traffic area, combined with at least one path category training, and the at least one path category classifies the traffic path of the traffic area, thereby macroscopically classifying the traffic path, and adjusting the predicted traffic path according to the predicted path category to obtain the target traffic path, so that the predicted traffic path can be matched with the corresponding path category, with higher accuracy. By making two predictions of the microscopic traffic path and the macroscopic traffic path category, and adjusting the predicted traffic path using the path category of the traffic, the predicted traffic path is made to conform to the predicted path category, and the adjusted target traffic path does not deviate from the predicted predicted path category, with higher accuracy.

[0038] In addition, the step sequence in the following method embodiments is only an example and not a strict limitation.

[0039] Figure 1 A flow chart of a path prediction method provided by an embodiment of the present invention, the method comprising the following steps:

[0040] 101: Determine a first traffic flow path in a target traffic area.

[0041] The target traffic area is an area that needs to be monitored, and the first traffic path is a traffic path collected at the time when prediction is required. The first traffic path may include a traffic segment sequence corresponding to the traffic in the target traffic area and a traffic speed corresponding to the traffic segment sequence, wherein the traffic in the target traffic area may include multiple traffic segments.

[0042] Optionally, the first traffic flow path may include a plurality of traffic flow segment sequences corresponding to a plurality of traffic flows in the target traffic area collected at the current moment or a certain moment, and a traffic flow speed corresponding to each traffic flow segment sequence. The traffic flow speed may refer to the speed of the traffic or pedestrian flow passing through a certain continuous traffic segment sequence, in units of vehicles / hour.

[0043] As a possible implementation method, the first traffic path can be acquired in real time to perform real-time path prediction. During acquisition, the speed, road section, and direction of vehicles or pedestrians can be collected by electronic devices to obtain the traffic path distribution, and multiple traffic section sequences corresponding to the first traffic path and data such as traffic speed can be transmitted to the background server, and the background server obtains the first traffic path of the target traffic area.

[0044] When dividing the traffic area, it can be divided as needed. Figure 2 is an example image of a traffic area, where any traffic area can be used as a target traffic area. Figure 2 The area shown in 201 can be used as a target traffic area. Figure 2 The traffic path distribution of the target traffic area 201 at a certain moment may include a road segment sequence A, a road segment sequence B, a road segment sequence C, and a road segment sequence D, wherein the traffic speed corresponding to the road segment sequence A is 300 vehicles / hour, the traffic speed corresponding to the road segment sequence B is 168 vehicles / hour, the traffic speed corresponding to the road segment sequence C is 300 vehicles / hour, and the traffic speed corresponding to the road segment sequence D is 355 vehicles / hour. Each road segment sequence may also include the driving direction of the traffic, for example Figure 2 The road segment sequence C and the road segment sequence D in the figure are actually composed of the same road segments, but the traffic directions are opposite, forming two road segment sequences. In practical applications, the sequence arrangement direction of the road segment sequence is the driving direction of the traffic flow, and the traffic flow travels in the arrangement direction of the road segment sequence to generate the corresponding traffic flow speed.

[0045] 102: Based on a preset path prediction model, using the first traffic path, obtain a predicted traffic path.

[0046] The path prediction model can be obtained based on the historical traffic path training of the target traffic area. In order to reduce the processing complexity, the training process can be separated from the path prediction method of the embodiment of the present application. When predicting the traffic path, the pre-trained path prediction model can be directly loaded to obtain the predicted traffic path using the first traffic path prediction.

[0047] As an optional method, the path prediction model can use a linear regression model. The algorithm complexity of the linear regression model is low and can be predicted quickly. At the same time, since the linear regression model can set multiple independent variables, it can process multiple road section sequences and obtain accurate processing effects. Then, when the corresponding model parameters are obtained, the path prediction model can be used to obtain accurate prediction results.

[0048] The method of obtaining a predicted traffic path using the first traffic path based on a preset path prediction model may include: inputting the first traffic path into the preset path prediction model to calculate and obtain a plurality of optional traffic paths.

[0049] The multiple optional traffic paths are predicted traffic path distribution results, and each optional traffic path includes multiple path sequences and the traffic speed corresponding to each path sequence. The unit of the traffic speed is vehicle / hour.

[0050] Combining the multiple optional traffic paths to form a predicted traffic path may specifically refer to combining the multiple optional traffic paths in a certain arrangement order to form a predicted traffic path.

[0051] 103: Based on a preset category prediction model, using the first traffic path, obtain a predicted path category.

[0052] The category prediction model is obtained based on the historical traffic paths of the target traffic area and combined with at least one path category training.

[0053] A route category is actually a collection of traffic routes with the same route distribution characteristics.

[0054] The category prediction model can be based on the historical traffic paths of the target traffic area and can be obtained by pre-training in combination with at least one path category, so that the first traffic path is input into a preset category prediction model to predict the predicted path category.

[0055] In order to ensure the accuracy of the prediction, when the path prediction model and the category prediction model perform short-term prediction, the prediction time interval is equal. Compared with the collection time of the first traffic flow path, the path prediction model and the category prediction model both predict the traffic flow path and path category at the target prediction time. The path prediction model predicts the distribution of traffic flow paths at the target prediction time, and the category prediction model also predicts the distribution of path categories at the target prediction time. The time interval between the target prediction time and the collection time of the first traffic flow path, that is, the current moment or a certain moment, can be set according to actual needs, and will not be repeated here. For example, assuming that the collection time of the first traffic flow path is 10:00 am, the path prediction model predicts the short-term prediction time period for the future is 10 minutes, the target prediction time corresponding to the predicted traffic flow path is 10:10, and the target prediction time corresponding to the predicted traffic flow category is 10:10, that is, the path prediction model predicts the traffic flow path at 10:10, and the category prediction model predicts the traffic flow category to which the traffic flow path at 10:10 belongs.

[0056] 104: According to the predicted path category, adjust the predicted traffic path to obtain a target traffic path.

[0057] The adjusting the predicted traffic path to obtain the target traffic path according to the predicted path category may specifically be adjusting the predicted traffic path so that the predicted traffic path conforms to the predicted path category to obtain the target traffic path.

[0058] Optionally, adjusting the predicted traffic path to obtain a target traffic path according to the predicted path category includes:

[0059] Determining whether the predicted traffic flow path meets the predicted path category;

[0060] If yes, determining the predicted traffic flow path as the target traffic flow path;

[0061] If not, the predicted traffic path is adjusted to conform to the predicted path category to obtain a target traffic path.

[0062] Optionally, adjusting the predicted traffic flow path to obtain a target traffic flow path according to the predicted path category may include: adjusting a plurality of traffic segment sequences of the predicted traffic flow path and a traffic flow corresponding to each traffic segment sequence according to the predicted path category to obtain a plurality of target segment sequences and a target traffic flow corresponding to each target segment sequence;

[0063] The target traffic flow paths are determined to be the target road segment sequences and the target traffic flow corresponding to each road segment sequence.

[0064] In actual applications, adjusting the multiple traffic segment sequences of the predicted traffic path and the traffic flow corresponding to each traffic segment sequence according to the predicted path category to obtain multiple target traffic segment sequences and the target traffic flow corresponding to each target segment sequence may include: adjusting the multiple traffic segment sequences of the predicted traffic path and the traffic flow corresponding to each traffic segment sequence so that the multiple traffic segment sequences and the traffic flow corresponding to each traffic segment sequence belong to the predicted path category to obtain multiple target segment sequences and the target traffic flow corresponding to each target segment sequence.

[0065] In an embodiment of the present invention, after determining the first traffic path of the target traffic area, the first traffic path can be used to obtain a predicted traffic path based on a preset path prediction model. Afterwards, the predicted path category is obtained by using the first traffic path based on a preset category prediction model, and the category prediction model is obtained based on the historical traffic path of the target traffic area, combined with at least one path category training, and at least one path category classifies the traffic path of the traffic area, thereby macroscopically classifying the traffic path, and adjusting the predicted traffic path according to the predicted path category to obtain the target traffic path, so that the predicted traffic path can be matched with the corresponding path category, with higher accuracy. By combining the two prediction methods of microscopic traffic path and macroscopic traffic path category, and adjusting the predicted traffic path using the path category of the traffic, so that the predicted traffic path meets the predicted path category, the adjusted target traffic path does not deviate from the predicted predicted path category, and the accuracy is higher.

[0066] In some embodiments, the path category includes a set of multiple traffic paths having the same path distribution characteristics.

[0067] Through clustering, the original traffic paths that have no distribution rules and are relatively complex can be classified, and the characteristics of different traffic path distributions can be reflected, so that different traffic path distributions can be arranged, and then the traffic paths can be adjusted using path categories to improve the accuracy of the predicted traffic paths.

[0068] like Figure 3 FIG. 1 is a flow chart of another embodiment of a path prediction method provided by an embodiment of the present invention. The method may include the following steps:

[0069] 301: Determine a first traffic flow path in a target traffic area.

[0070] 302: Determine first impact data of at least one path impact factor.

[0071] The collection time of the first influencing data is the same as the collection time of the first vehicle flow path. For example, when the collection time of the first vehicle flow path is the current moment, the collection time of the first influencing data is also the current moment.

[0072] 303: Based on a preset path prediction model, a predicted traffic path is obtained using the first traffic path and the first influencing data.

[0073] The path prediction model is obtained by training based on the historical traffic paths of the target traffic area and the historical impact data of the at least one path impact factor.

[0074] 304: Based on a preset category prediction model, using the first traffic path and the first impact data, obtain a predicted path category.

[0075] The category prediction model is obtained based on the historical traffic path of the target traffic area and the historical impact data of the at least one path impact factor, combined with the at least one path category training.

[0076] 305: According to the predicted path category, adjust the predicted traffic path to obtain a target traffic path.

[0077] To ensure the accuracy of the prediction, the path prediction model and the category prediction model have the same prediction time interval when making short-term predictions. Compared with the collection time of the first traffic path, the path prediction model and the category prediction model both predict the traffic path and path category at the target prediction time. The path prediction model predicts the traffic path distribution at the target prediction time, and the category prediction model also predicts the path category distribution at the target prediction time.

[0078] In the embodiment of the present invention, by adding the path influence factor to the path prediction process, various influence factors that affect the path prediction are taken into consideration. By increasing the prediction reference amount, prediction can be made from multiple angles to improve the accuracy of the prediction.

[0079] In some embodiments, Figure 4 As shown, the path prediction model can be trained and obtained in the following way:

[0080] 401: Determine at least one path training sample and a path true value corresponding to each path training sample; wherein each path training sample is composed of a historical traffic flow path and historical influence data of at least one path influence factor corresponding to the historical traffic flow path.

[0081] The path truth value of each path training sample is the historical traffic path corresponding to the time of the collection time interval prediction time interval corresponding to the historical traffic path of the path training sample. That is, the path truth value is the historical traffic path corresponding to the time of the historical traffic path detection prediction time interval.

[0082] 402: Build a path prediction model.

[0083] 403: Based on the at least one path training sample and the path true value corresponding to each path training sample, obtain model parameters of the path prediction model through training.

[0084] The path prediction model may be a linear regression model. The model parameters of the constructed path prediction model are unknown. The model parameters of the path prediction model may be obtained through at least one path training sample and the path truth value training corresponding to each path training sample.

[0085] By adding at least one path influencing factor to the traffic path prediction process, factors that may affect the path distribution can be obtained, the impact of the path distribution can be considered from multiple aspects, and the accuracy of the prediction result of the path distribution can be improved.

[0086] The path true value corresponding to each path training sample may include a historical traffic flow path that has a preset time interval with the path training sample.

[0087] When performing path prediction, the input of the path prediction model is the first traffic path, and the output is the obtained predicted traffic path. During actual prediction, there is a certain time difference between the obtained predicted traffic path and the input first traffic path. The time difference is the preset time interval. For example, in practical applications, if it is necessary to predict the distribution of traffic paths in the next half an hour, the preset time interval is half an hour. The path truth value corresponding to each path training sample is the historical traffic path separated from the historical traffic path of the training sample by a preset time interval. For example, when the time interval is 10 minutes, the path truth value corresponding to the historical traffic path collected at 10:00 a.m. is the historical traffic path collected at 10:10 a.m.

[0088] As a possible implementation manner, the training to obtain the model parameters of the path prediction model based on the at least one path training sample and the path true value corresponding to each path training sample may include:

[0089] Randomly generating path reference parameters of the path prediction model;

[0090] Inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, respectively, to predict and obtain a first prediction result corresponding to each path training sample;

[0091] Determine a path prediction error based on a first prediction result corresponding to each path training sample and a path true value;

[0092] If the path prediction error satisfies a first error condition, determining the path reference parameter as a model parameter of the path prediction model;

[0093] If the path prediction error does not meet the first error condition, based on the path prediction error, the path reference parameter is updated, and the process returns to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, and the step of predicting and obtaining the first prediction result corresponding to each path training sample continues.

[0094] During the initial calculation, since the model parameters of the path prediction model are unknown, the model parameters of the path distribution model need to be initialized. Usually, the model parameters of the path distribution model can be randomly generated to start the training of the path prediction model.

[0095] Inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, respectively, and predicting to obtain a first prediction result corresponding to each path training sample may include:

[0096] The path prediction error satisfying the first error condition may include the path prediction error satisfying a convergence condition. For example, the path prediction error satisfying the convergence condition may be that the path prediction error is less than a preset path error threshold.

[0097] The step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter to predict and obtain the first prediction result corresponding to each path training sample may specifically be inputting each path training sample as input data into the path prediction model corresponding to the path reference parameter to predict and obtain the first prediction result corresponding to each path training sample. The first prediction result is the predicted traffic path, which may include multiple predicted traffic sections and the traffic speed corresponding to each predicted traffic section.

[0098] In some possible designs, if the path prediction error does not satisfy the first error condition, based on the path prediction error, updating the path reference parameter, and returning to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, respectively, and predicting to obtain the first prediction result corresponding to each path training sample may include:

[0099] If the path prediction error does not satisfy the first error condition, calculating a partial derivative of the path prediction error with respect to a model parameter of the path prediction model to obtain a path adjustment amount;

[0100] Determining a path parameter adjustment amount according to the path adjustment step length and the path adjustment amount;

[0101] The path reference parameter is adjusted using the path parameter adjustment amount, and the process returns to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, and predicting and obtaining the first prediction result corresponding to each path training sample.

[0102] When updating the path reference parameter based on the path prediction error, an optimization algorithm such as the least square method and the gradient descent algorithm can be used for reference optimization to update the path reference parameter. In the embodiment of the present application, a gradient descent algorithm is used, that is, by calculating the partial derivative of the path prediction error relative to the model parameter of the path prediction model, the partial derivative is used as the path adjustment amount, and the product of the path adjustment amount and the adjustment step length is calculated to obtain the path parameter adjustment amount, and the difference between the path reference parameter and the path parameter adjustment is calculated to obtain the adjusted path reference parameter.

[0103] By optimizing the path reference parameters, accurate reference parameters can be obtained, the path reference parameters can be adjusted, and accurate path prediction results can be obtained.

[0104] In some further possible designs, determining the path prediction error based on the first prediction result corresponding to each path training sample and the path true value includes:

[0105] Calculating the path distance between the first prediction result corresponding to each path training sample and the path true value to obtain the path error of each path training sample;

[0106] A path prediction error is determined according to the path errors respectively corresponding to at least one path training sample.

[0107] Calculating the path distance between the first prediction result corresponding to each path sample and the path true value to obtain the path error of each path training sample may include:

[0108] According to the multiple predicted road segment sequences corresponding to the first prediction result of each path training sample and the vehicle flow speed corresponding to each road segment prediction sequence, the number of vehicles passing through the first prediction result per unit time is calculated;

[0109] Calculate the number of vehicles passing the path true value per unit time according to a plurality of real road segment sequences corresponding to the path true value and the vehicle flow speed corresponding to each real road segment sequence;

[0110] Determine a plurality of change path pairs having changes between the plurality of predicted road segment sequences and the plurality of real road segment sequences;

[0111] Determine the number of vehicles that pass between the first prediction result and the path true value per unit time according to the vehicle flow speeds corresponding to the multiple change path pairs respectively;

[0112] The path error is calculated by using the number of vehicle changes and combining the vehicle data that passes the first prediction result and the vehicle data that passes the path true value within a unit time.

[0113] Among them, determining multiple changed path pairs in which changes exist between multiple predicted road segment sequences and multiple real path sequences may refer to: road segment sequences in which the traffic segment sequences are the same as those in the multiple predicted road segment sequences but have different traffic speeds, and paths in which the traffic segment sequences are completely different between the two, and the corresponding traffic speeds.

[0114] Determining the changed number of vehicles passing between the first prediction result and the path true value in unit time according to the traffic speeds corresponding to the multiple changed path pairs may include: calculating a first difference in the number of vehicles generated in unit time by a sequence of sections with the same traffic section sequence and different corresponding traffic speeds, and a second difference in the number of vehicles generated in unit time by a path with completely different traffic section sequences, calculating the sum of the first difference in the number of vehicles and the second difference in the number of vehicles, and obtaining the changed number of vehicles.

[0115] The path distance between the first prediction result corresponding to each path training sample and the path true value can be calculated by the following formula:

[0116]

[0117] Where Δf i,j The number of unit vehicle changes generated by the path change between the first prediction result and the path truth value, f i is the number of vehicles passing the path in unit time, f j is the number of vehicles that pass the first prediction result per unit time. i is the iteration order value of the reference traffic flow path, j is the iteration order value of the predicted traffic flow path, and when the number of paths is one, i and j can usually be 1 or ignored.

[0118] In some embodiments, Figure 5 As shown, the category prediction model in the embodiment of the present invention is obtained by training in the following way:

[0119] 501: Determine at least one category training sample and a category truth value corresponding to each category training sample; wherein each category training sample is composed of a historical traffic flow path and historical influence data of at least one path influence factor corresponding to the historical traffic flow path; and the category truth value corresponding to each category training sample is a path category in the at least one path category that matches a historical traffic flow path that is spaced at a preset time interval from the corresponding historical traffic flow path.

[0120] The category true value of each category training sample can be determined in the following manner: determine a historical traffic flow path corresponding to a category training sample, and a target historical traffic flow path in at least one historical traffic flow path that is separated from the historical traffic flow path by a preset time interval, and determine that the path category matching the target historical traffic flow path is the category true value of the category training sample.

[0121] 502: Build a category prediction model.

[0122] 503: Based on the at least one category training sample and the category truth value corresponding to each category training sample, obtain model parameters of the category prediction model through training.

[0123] By dividing each traffic flow path into different path categories, paths without distribution patterns can be classified, so that different traffic flow paths are defined as different path categories, and then the path categories are applied to the path prediction scenario. The predicted traffic flow path is obtained by predicting the category to which the distribution of the path belongs and using the predicted path category constraints, so that the adjusted traffic flow path can meet the corresponding path category, without generating large prediction errors, thereby improving the accuracy of the prediction.

[0124] As a possible implementation manner, the training to obtain the model parameters of the category prediction model based on the at least one category training sample and the category true value corresponding to each category training sample may include:

[0125] Randomly generating category reference parameters of the category prediction model;

[0126] Inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter respectively, and predicting to obtain a second prediction result corresponding to each category training sample;

[0127] Determine a category prediction error based on the second prediction result corresponding to each category training sample and the category true value;

[0128] If the category prediction error satisfies a second error condition, determining the category reference parameter as a model parameter of the category prediction model;

[0129] If the category prediction error does not meet the second error condition, the category reference parameter is adjusted based on the category prediction error, and the process returns to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and the step of predicting the second prediction result corresponding to each category training sample continues.

[0130] In some possible designs, if the category prediction error does not satisfy the second error condition, adjusting the category reference parameter based on the category prediction error, and returning to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, respectively, and predicting to obtain the second prediction result corresponding to each category training sample may include:

[0131] If the category prediction error does not satisfy the second error condition, calculating the partial derivative of the category prediction error with respect to the model parameters of the category prediction model to obtain a category adjustment amount;

[0132] Determining a category parameter adjustment amount according to the category adjustment step length and the category adjustment amount;

[0133] The category reference parameter is adjusted using the category parameter adjustment amount, and the step of returning to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and predicting and obtaining the second prediction result corresponding to each category training sample continues.

[0134] In some further possible designs, determining the category prediction error based on the second prediction result corresponding to each category training sample and the category true value may include:

[0135] Calculating the category distance between the second prediction result corresponding to each category training sample and the category true value to obtain the category error of each category training sample;

[0136] The category prediction error is determined based on the category error corresponding to each category training sample.

[0137] Further, optionally, the at least one path influencing factor may include: a surrounding traffic influencing factor, a time influencing factor, a weather influencing factor and / or a date influencing factor.

[0138] The historical impact data of the at least one path impact factor includes: surrounding traffic paths, path collection time, path collection weather and / or path collection date.

[0139] Path influencing factors refer to factors that can affect the distribution of traffic paths.

[0140] When collecting the first traffic path of the target traffic area, the first influencing data for determining at least one path influencing factor can be collected at the same time, that is, the collection time of the first traffic path is the same as the collection time of the first influencing data of at least one path influencing factor. The surrounding traffic influencing factor may include a surrounding traffic path, which may include a traffic path of at least one traffic area around the target traffic area, and the traffic path may include multiple road section distribution sequences and the traffic speed corresponding to each road section distribution sequence.

[0141] Time influencing factors may include the path collection time. At different time points in a day, the traffic conditions will change. For example, during the morning and evening rush hours, there are more vehicles and the traffic volume is larger. At ordinary time points, there are fewer vehicles and the traffic volume is smaller.

[0142] The weather influencing factor may include the path collection weather, which may specifically refer to the weather conditions when the traffic path is collected. As a possible implementation, the weather influencing factor may be identified by a weather character, for example, 1 represents a sunny day, 0 represents a cloudy day, etc., which may be specifically set according to usage requirements.

[0143] The date influencing factor may include the path collection date. The distribution of traffic paths on different dates is also different. The path collection date is the collection date of each traffic path. In practical applications, when the path collection date is used as data, the path collection date can be used in the form of a date string.

[0144] like Figure 6 FIG. 1 is a flow chart of another embodiment of a path prediction method provided by an embodiment of the present invention. The method may include the following steps:

[0145] 601: Determine a first traffic flow path in a target traffic area.

[0146] 602: Based on a preset path prediction model, using the first traffic path, obtain a predicted traffic path.

[0147] 603: Based on a preset category prediction model, using the first traffic path, obtain a predicted path category.

[0148] The category prediction model is obtained based on the historical traffic paths of the target traffic area and combined with at least one path category training.

[0149] 604: Determine a plurality of reference traffic paths corresponding to the predicted path category.

[0150] 605: Using the multiple reference traffic paths, adjust the predicted traffic path to obtain a target traffic path.

[0151] The predicted path category may be composed of a plurality of traffic paths having the same path distribution characteristics, and the plurality of traffic paths may be determined as a plurality of reference traffic paths.

[0152] In an embodiment of the present invention, when adjusting the predicted traffic flow path according to the predicted path category, the predicted traffic flow path can be adjusted using multiple reference traffic flow paths corresponding to the predicted path category, thereby improving the accuracy of the prediction and achieving accurate adjustment.

[0153] In some possible applications, the number of the multiple reference traffic paths in the predicted path category may be relatively large. If the multiple reference traffic paths in the traffic path category are directly used to adjust the predicted traffic path in the above embodiment, the amount of calculation is relatively large. An appropriate reference traffic path can be selected from the multiple reference traffic paths to reduce the amount of calculation when adjusting the predicted traffic path. The use of the multiple reference traffic paths to adjust the predicted traffic path to obtain the target traffic path includes:

[0154] Selecting an adjusted traffic flow path having a first correlation relationship with the predicted traffic flow path from the plurality of reference traffic flow paths;

[0155] The predicted traffic flow path is adjusted using the adjusted traffic flow path to obtain a target traffic flow path.

[0156] The first association relationship may mean that the selected reference traffic path and the predicted traffic path have a high similarity, and the path distance between the two is small.

[0157] As a possible implementation manner, the selecting a preset number of adjusted traffic paths from the multiple reference traffic paths includes: selecting a preset number of adjusted traffic paths that have a high similarity to the predicted traffic path from the multiple reference traffic paths.

[0158] In some possible designs, the path similarity between the reference traffic path and the predicted traffic path can be expressed by calculating the path distance between the reference traffic path and the predicted traffic path, wherein the greater the path distance, the smaller the similarity, and the smaller the path distance, the higher the similarity. Therefore, further, optionally, the selecting a preset number of adjusted traffic paths from the multiple reference traffic paths may include:

[0159] respectively calculating the path distances between the multiple reference traffic paths and the predicted traffic path to obtain multiple reference path distances;

[0160] According to the multiple reference path distances, a reference traffic path whose path distance from the predicted traffic path is less than a first preset threshold is selected from the multiple reference traffic paths to obtain an adjusted traffic path.

[0161] The path distance between any reference traffic path and the predicted traffic path can be calculated using the following formula:

[0162]

[0163] Where Δf i,j The number of unit vehicle changes caused by the path change between the reference traffic path and the predicted traffic path, f i is the number of vehicles passing through the reference traffic path per unit time, f j is the number of vehicles passing through the predicted traffic flow path per unit time. i is the iteration order value of the reference traffic flow path, j is the iteration order value of the predicted traffic flow path, and when the number of predicted traffic flow paths is one, j can usually be 1 or ignored.

[0164] Optionally, determining, according to the multiple reference path distances, a preset number of reference traffic paths closest to the predicted traffic path as the adjusted traffic path comprises:

[0165] Sorting the multiple reference path distances in ascending order;

[0166] A preset number of reference path distances respectively corresponding to the reference traffic flow paths are selected from the sorted multiple reference path distances as the adjusted traffic flow paths.

[0167] Respectively calculating the path distances between the multiple reference traffic paths and the predicted traffic path, and obtaining multiple reference path distances can specifically refer to calculating the path distance between each reference traffic path in the multiple reference traffic paths and the predicted traffic path, and obtaining the path distance between each reference traffic path and the predicted traffic path.

[0168] For any reference traffic flow path, when calculating the path distance between the reference traffic flow path and the predicted traffic flow path, a section difference calculation can be performed based on multiple reference section sequences corresponding to the reference traffic flow path and the reference traffic speed corresponding to each reference section sequence, and multiple predicted section sequences corresponding to the multiple predicted traffic flow paths and the predicted traffic speed corresponding to each predicted section sequence, so as to determine the weighted change in the changed section and the changed traffic speed, which is the path distance between the above two paths.

[0169] In some embodiments, adjusting the predicted traffic path to obtain the target traffic path according to the predicted path category may specifically include: adjusting the predicted traffic path to match the predicted path category, and obtaining the adjusted predicted traffic path as the target traffic path.

[0170] In order to adjust the predicted traffic flow path so that the adjusted traffic flow path can match the predicted path category, when actually adjusting the predicted traffic flow path, over-adjustment is prone to occur, causing the final target traffic flow path to deviate greatly from the predicted traffic flow path, resulting in a large adjustment error. Furthermore, when adjusting the predicted traffic flow path, certain constraints can be used to constrain the adjustment process to avoid over-adjustment and thereby improve the adjustment accuracy.

[0171] Further, optionally, using the preset number of adjusted traffic paths to adjust the predicted traffic path to obtain the target traffic path may include:

[0172] The predicted traffic flow path is adjusted using the adjusted traffic flow path so that the predicted traffic flow path satisfies the adjustment constraint conditions, thereby obtaining a target traffic flow path.

[0173] Optionally, the step of adjusting the predicted traffic path to obtain the target traffic path using the adjusted traffic path under the path adjustment constraint condition is as follows: under the path adjustment constraint condition, adjusting the predicted traffic path to reduce the sum of the path distances with the adjusted traffic path, and obtaining the target traffic path after the adjustment. During the adjustment, the path distance between the predicted traffic path and the adjusted traffic path is continuously reduced until a distance threshold is met.

[0174] The adjustment constraint condition may be that the path distance between the adjusted traffic path and the central traffic path is less than the path distance between the reference traffic path and the central traffic path. The central traffic path is obtained by performing center clustering based on multiple reference traffic paths of the target traffic path category.

[0175] As an embodiment, the adjusting the predicted traffic flow path by using the adjusted traffic flow path so that the predicted traffic flow path satisfies the adjustment constraint condition and obtaining the target traffic flow path may include:

[0176] Determining a central traffic path of the predicted path category according to the multiple reference traffic paths;

[0177] Calculating the multiple reference traffic paths and the central traffic path respectively to obtain multiple reference distances;

[0178] Determine the largest path distance among the multiple reference path distances as the constraint distance;

[0179] When the predicted traffic flow path is adjusted using the adjusted traffic flow path, the path distance between the adjusted predicted traffic flow path and the central traffic flow path is less than the constraint distance, and a target traffic flow path is obtained.

[0180] In the embodiment of the present invention, the adjustment process of the predicted traffic flow path can be constrained by adjusting the constraint conditions to avoid excessive adjustment, thereby ensuring the accuracy of the adjustment of the predicted traffic flow path and improving the adjustment effect.

[0181] The path distance between any reference traffic flow path and the central traffic flow path can be calculated using the following formula:

[0182]

[0183] Where Δf i,j The number of unit vehicle changes caused by the path change between the reference traffic path and the central traffic path, f i is the number of vehicles passing through the reference traffic path per unit time, f j is the number of vehicles passing through the central traffic path per unit time.

[0184] As a possible implementation manner, determining the central traffic path of the predicted path category according to the multiple reference traffic paths includes:

[0185] Based on the central clustering algorithm, the central traffic path of the predicted path category is determined using the multiple reference traffic paths.

[0186] The center clustering algorithm described in the embodiment of the present invention can adopt a clustering algorithm such as a K-center point clustering algorithm, which can improve the accuracy of clustering, obtain an accurate center traffic path, and further improve the constraint effect, thereby ensuring the accuracy of the target traffic path.

[0187] Further, optionally, the determining the central traffic path of the predicted path category by using the multiple reference traffic paths based on the central clustering algorithm may include:

[0188] Randomly generate central traffic paths;

[0189] Calculating the cluster path distance between the central traffic flow path and each reference traffic flow path to obtain multiple cluster path distances;

[0190] If the sum of the squares of the distances of the plurality of cluster paths meets the optimization objective, determining the central traffic flow path;

[0191] If the sum of the squares of the multiple cluster path distances does not meet the optimization objective, the central traffic path is updated based on the cluster path distance, and the process returns to the step of calculating the cluster path distance between the central traffic path and each reference traffic path, and the step of obtaining multiple cluster path distances continues.

[0192] The candidate traffic path is the traffic path obtained after the predicted traffic path is adjusted. If the candidate traffic path satisfies the path constraint condition, determining the candidate traffic path as the target traffic path includes: if the path distance between the candidate traffic path and the central traffic path is less than the constraint distance, determining the candidate traffic path as the target traffic path.

[0193] When predicting the central traffic flow path through a clustering algorithm, an optimization target is usually required. When the optimization target is achieved, the central traffic flow path can be determined. In practical applications, the optimization target is of high importance and plays an important role in the prediction results. In order to obtain an accurate optimization target, the optimization target used in the embodiment of the present invention can ensure that the sum of the squares of the path distances between the central traffic flow path and other traffic flow paths is minimized, and the obtained central traffic flow path can accurately represent the center point of the predicted path category, achieve accurate clustering, and improve the accuracy of the prediction results.

[0194] If the sum of the squares of the distances of the plurality of cluster paths meets the optimization target, determining the central traffic path may include: if the sum of the squares of the distances of the plurality of cluster paths reaches a minimum, determining the central traffic path,

[0195] The optimization objective is expressed by the formula:

[0196] in, is the traffic volume on the traffic path rt corresponding to the central traffic path, is the traffic flow of the reference traffic path rt among the k reference vehicle paths; i is the number of iterations.

[0197] The method for determining the predicted path category adopted in the embodiment of the present invention is to first cluster the traffic path categories using the historical traffic flow paths of the target traffic area, and then use the first traffic path to perform category prediction to achieve the prediction of the traffic path category. When predicting, it is necessary to use the historical traffic paths and the path category corresponding to each historical traffic path. The path category corresponding to each historical traffic path needs to be obtained through prior training. In actual applications, the path category may include one or more, and the one or more path categories can be obtained through prior training.

[0198] As an embodiment, the at least one path category is determined by:

[0199] The historical traffic paths of the target traffic area are clustered to obtain at least one path category.

[0200] Each path category includes a set of multiple traffic paths with the same path distribution characteristics.

[0201] Optionally, before clustering, the method may further include:

[0202] Determine the historical traffic paths of the target traffic area.

[0203] When clustering the historical traffic paths of the target traffic area, the traffic paths with high similarity can be classified into the same path category, and the similarity of the traffic paths can be expressed according to the path distance between the traffic paths. In practical applications, the traffic paths with high path similarity can be classified into the same graph. When specifically dividing, each traffic path can be abstracted as a node, and the association relationship between the traffic paths can be constructed through the nodes.

[0204] As a possible implementation manner, clustering the historical traffic paths of the target traffic area to obtain at least one path category includes:

[0205] generating a plurality of nodes according to a plurality of historical traffic paths of the target traffic area;

[0206] Based on the node distance between any two nodes, a node relationship graph is constructed;

[0207] Dividing the node relationship graph into at least one sub-node relationship graph;

[0208] The traffic path corresponding to the node included in each sub-node relationship graph is determined as a path category, and at least one path category is obtained.

[0209] Further, optionally, dividing the node relationship graph into at least one sub-node relationship graph includes:

[0210] Nodes with the same position features in the node relationship graph are divided into the same sub-node relationship graph to obtain at least one sub-node relationship graph.

[0211] Based on the node distance between any two nodes, determine the K similar nodes closest to each node;

[0212] Each of the multiple nodes is taken as a vertex, and K edges are extended, and K similar nodes are respectively another vertex of the K edges, so as to obtain the node relationship graph.

[0213] The node distance between any two nodes may refer to the path distance between the traffic paths corresponding to the two nodes.

[0214] The path distance between any two historical traffic paths can be calculated using the following formula:

[0215]

[0216] Where Δf i,j The number of unit vehicle changes caused by the path change between two traffic paths, f i is the number of vehicles passing through one of the historical traffic paths per unit time, f j is the number of vehicles passing through another historical traffic path in unit time. Wherein, i and j are the values ​​of the arrangement order of the two traffic paths in any iteration process.

[0217] Figure 7 The traffic path of the traffic area 701 at the first moment may include a road segment sequence A, a road segment sequence B, a road segment sequence C, and a road segment sequence D, wherein the traffic speed corresponding to the road segment sequence A is 300 vehicles / hour, the traffic speed corresponding to the road segment sequence B is 168 vehicles / hour, the traffic speed corresponding to the road segment sequence C is 300 vehicles / hour, and the traffic speed corresponding to the road segment sequence D is 355 vehicles / hour.

[0218] The traffic path of the traffic area 701 at a second moment separated by a preset interval from the first moment can be as follows: Figure 7 As shown in 702, the traffic paths at this moment are: section sequence A, section sequence B, section sequence C and section sequence E, wherein the traffic speed corresponding to section sequence A is 340 vehicles / hour, the traffic speed corresponding to section sequence B is 168 vehicles / hour, the traffic speed corresponding to section sequence C is 270 vehicles / hour, and the traffic speed corresponding to section sequence E is 227 vehicles / hour.

[0219] At this time, the link sequence A includes 5 links, the link sequence B includes 5 links, the link sequence C includes 3 links, the link sequence D includes 3 links, and the link sequence E includes 3 links.

[0220] Among them, the road sections with changes in traffic speed are A, D, and E. The traffic speed changes are: A increases by 40 cars / hour, D decreases by 156 cars / hour, and E increases by 227 cars / hour.

[0221] Then the change in unit vehicle volume caused by the change in the traffic path between the first moment and the second moment is:

[0222] Δf i,j =40*5+156*3+227*3=1349;

[0223] f i =300*5+270*3+156*3+168*5=3618

[0224] f j =340*5+270*3+227*3+168*5=4031

[0225]

[0226] d is the path distance between the traffic path corresponding to the first moment and the traffic path corresponding to the second moment.

[0227] Further, optionally, dividing the nodes having the same position features in the node relationship graph into the same sub-node relationship graph to obtain at least one sub-node relationship graph includes:

[0228] The nodes in the node relationship graph belonging to the same area are divided into the same sub-node relationship graph to obtain at least one sub-node relationship graph.

[0229] The nodes belonging to the same area in the node relationship graph are divided into the same sub-node relationship graph, which may specifically include: dividing the nodes with very close distances in the node relationship graph into the same sub-node relationship graph.

[0230] The node relationship graph can express the association relationship between each traffic path, associate them according to the path distance between the traffic paths, and divide them according to the region. In practical applications, the node relationship graph can be segmented based on a graph segmentation algorithm, a hierarchical clustering algorithm, a modularity optimization algorithm, etc., wherein the graph segmentation algorithm can be such as the Kernighan-Lin algorithm, the spectral bisection method, the hierarchical clustering algorithm can be such as the GN algorithm, the Newman fast algorithm, and the modularity optimization algorithm can be such as the greedy algorithm, the simulated annealing algorithm, the Memetic algorithm, the PSO algorithm, and the evolutionary multi-objective optimization algorithm. The input of each algorithm is the constructed node relationship graph, and the output is at least one child node relationship graph. The division process is the same as the operation process of the algorithm, which will not be repeated here.

[0231] As a possible implementation manner, dividing the node relationship graph into at least one sub-node relationship graph includes:

[0232] Nodes with the same position features in the node relationship graph are divided into a sub-node relationship graph to obtain at least one sub-node relationship graph.

[0233] The nodes with the same position features in the node relationship graph may specifically refer to nodes in the same node relationship graph having a higher similarity to each other and a lower similarity to nodes in other child node relationship graphs.

[0234] As a possible implementation manner, dividing the nodes having the same position features in the node relationship graph into the same sub-node relationship graph to obtain at least one sub-node relationship graph includes:

[0235] Randomly generate at least one central node;

[0236] Calculating the node distance between each node and the at least one central node respectively;

[0237] Determine the central node with the shortest node distance to each node as the category label of the node;

[0238] If the sum of the node distances corresponding to all nodes meets the optimal condition, determine that the node corresponding to each central node is a child node relationship graph, and obtain at least one child node relationship graph;

[0239] If the sum of the node distances corresponding to all nodes does not meet the optimal condition, for each node corresponding to the central node, update the at least one central node, and return to the step of respectively calculating the node distance between each node and the at least one central node to continue execution.

[0240] Dividing nodes according to node distance can improve the accuracy of node relationship division and obtain accurate path category classification results.

[0241] As a possible implementation manner, constructing a node relationship graph based on the node distance between any two nodes may include:

[0242] Based on the node distance between any two nodes, determine the K similar nodes closest to each node;

[0243] Each node and the K similar nodes closest to it are connected to form the node relationship graph; wherein the length of each edge in the node relationship graph corresponds to the node distance between the two nodes of the edge.

[0244] like Figure 8 FIG. 1 is a schematic diagram of a structure of an embodiment of a path prediction device provided by an embodiment of the present invention. The device may include the following modules:

[0245] The first determination module 801 is used to determine a first vehicle flow path in a target traffic area;

[0246] A path prediction module 802 is configured to obtain a predicted traffic path using the first traffic path based on a preset path prediction model;

[0247] Model prediction module 803: used to obtain a predicted path category using the first traffic path based on a preset category prediction model.

[0248] Wherein, the category prediction model is obtained based on the historical traffic paths of the target traffic area and combined with the at least one path category training;

[0249] The path adjustment module 804 is used to adjust the predicted traffic path according to the predicted path category to obtain a target traffic path.

[0250] In some embodiments, the path category is a collection of multiple traffic paths having the same path distribution characteristics.

[0251] Through clustering, the original traffic paths that have no distribution rules and are relatively complex can be classified, and the characteristics of different traffic path distributions can be reflected, so that different traffic path distributions can be arranged, and then the traffic paths can be adjusted using path categories to improve the accuracy of the predicted traffic paths.

[0252] As an embodiment, the device may further include:

[0253] A data determination module, used to determine first impact data of at least one path impact factor;

[0254] The path prediction module comprises:

[0255] A path prediction unit, configured to obtain a predicted traffic path based on a preset path prediction model using the first traffic path and the first influence data; wherein the path prediction model is obtained by training based on the historical traffic path of the target traffic area and the historical influence data of the at least one path influence factor;

[0256] The model prediction module includes:

[0257] A model prediction unit is used to obtain a predicted path category based on a preset category prediction model using the first traffic path and the first influence data; wherein the category prediction model is based on the historical traffic path of the target traffic area and the historical influence data of the at least one path influence factor, combined with the at least one path category training.

[0258] In some embodiments, the apparatus further comprises:

[0259] A path sample module, used to determine at least one path training sample and a path truth value corresponding to each path training sample; wherein each path training sample is composed of a historical traffic path and historical influence data of at least one path influence factor corresponding to the historical traffic path;

[0260] The first building module is used to build a path prediction model;

[0261] The first training module is used to train and obtain model parameters of the path prediction model based on the at least one path training sample and the path true value corresponding to each path training sample.

[0262] As a possible implementation manner, the first training module includes:

[0263] A first generating unit, used for randomly generating a path reference parameter of the path prediction model;

[0264] A first prediction unit, configured to input the at least one path training sample into a path prediction model corresponding to the path reference parameter, to obtain a first prediction result corresponding to each path training sample;

[0265] A first comparison unit, configured to determine a path prediction error based on a first prediction result corresponding to each path training sample and a path true value;

[0266] a first judgment unit, configured to determine that the path reference parameter is a model parameter of the path prediction model if the path prediction error satisfies a first error condition;

[0267] The second judgment unit is used to update the path reference parameter based on the path prediction error if the path prediction error does not meet the first error condition, and return to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, and continuing to execute the step of obtaining the first prediction result corresponding to each path training sample.

[0268] In some possible designs, the second determination unit includes:

[0269] A first judgment subunit is configured to calculate a partial derivative of the path prediction error with respect to a model parameter of the path prediction model to obtain a path adjustment amount if the path prediction error does not satisfy a first error condition;

[0270] A first determining subunit, configured to determine a path parameter adjustment amount according to the path adjustment step length and the path adjustment amount;

[0271] The first adjustment subunit is used to adjust the path reference parameter using the path parameter adjustment amount, and return to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, and obtaining the first prediction result corresponding to each path training sample.

[0272] In some further possible designs, the first comparing unit may include:

[0273] A first calculation subunit is used to calculate the path distance between the first prediction result corresponding to each path training sample and the path true value, and obtain the path error of each path training sample;

[0274] The second determining subunit is used to determine the path prediction error according to the path error respectively corresponding to at least one path training sample.

[0275] In some embodiments, the apparatus may further include:

[0276] A category sample module is used to determine at least one category training sample and a category truth value corresponding to each category training sample; wherein each category training sample is composed of a historical traffic path and historical influence data of at least one path influence factor corresponding to the historical traffic path; and the category truth value corresponding to each category training sample is a path category that matches a historical traffic path in the at least one path category that is separated from the historical traffic path corresponding to the historical traffic path by a preset time interval;

[0277] The second building module is used to build a category prediction model;

[0278] The second training module is used to train and obtain model parameters of the category prediction model based on the at least one category training sample and the category true value corresponding to each category training sample.

[0279] As a possible implementation manner, the second training module may include:

[0280] A second generating unit, used for randomly generating category reference parameters of the category prediction model;

[0281] A second prediction unit, configured to input the at least one category training sample into a category prediction model corresponding to the category reference parameter, to obtain a second prediction result corresponding to each category training sample;

[0282] A second comparison unit, used to determine a category prediction error based on a second prediction result corresponding to each category training sample and a category true value;

[0283] a third judgment unit, configured to determine that the class reference parameter is a model parameter of the class prediction model if the class prediction error satisfies a second error condition;

[0284] The fourth judgment unit is used to adjust the category reference parameter based on the category prediction error if the category prediction error does not meet the second error condition, and return to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and continuing to execute the step of obtaining the second prediction result corresponding to each category training sample.

[0285] In some possible designs, the fourth determination unit may include:

[0286] A second judgment subunit is used to calculate the partial derivative of the class prediction error with respect to the model parameter of the class prediction model to obtain a class adjustment amount if the class prediction error does not meet the second error condition;

[0287] A third determining subunit, configured to determine a category parameter adjustment amount according to the category adjustment step length and the category adjustment amount;

[0288] The second adjustment subunit is used to adjust the category reference parameter using the category parameter adjustment amount, and return to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and continuing to execute the step of obtaining the second prediction result corresponding to each category training sample.

[0289] In some further possible designs, the second comparing unit includes:

[0290] A second calculation subunit is used to calculate the category distance between the second prediction result corresponding to each category training sample and the category true value, so as to obtain the category error of each category training sample;

[0291] The fourth determination subunit is used to determine the category prediction error according to the category error corresponding to each category training sample.

[0292] Further, optionally, the at least one path influencing factor includes: a surrounding traffic influencing factor, a time influencing factor, a weather influencing factor and / or a date influencing factor.

[0293] The historical impact data of the at least one path impact factor includes: surrounding traffic paths, path collection time, path collection weather and / or path collection date.

[0294] As an embodiment, the path adjustment module includes:

[0295] A reference determination unit, used to determine a plurality of reference traffic flow paths corresponding to the predicted path category;

[0296] The path adjustment unit is used to adjust the predicted traffic path by using the multiple reference traffic paths to obtain a target traffic path.

[0297] In some possible applications, the path adjustment unit includes:

[0298] A path selection subunit, configured to select, from the plurality of reference traffic paths, an adjusted traffic path having a first association relationship with the predicted traffic path;

[0299] The path adjustment subunit is used to use the adjusted traffic path to adjust the predicted traffic path to obtain a target traffic path.

[0300] As a possible implementation manner, the path selection subunit may include:

[0301] The distance calculation module calculates the path distances between the multiple reference traffic paths and the predicted traffic path respectively to obtain multiple reference path distances;

[0302] The reference adjustment module is used to select, from the multiple reference traffic paths, a reference traffic path whose path distance from the predicted traffic path is less than a first preset threshold value, to obtain an adjusted traffic path according to the multiple reference path distances.

[0303] In some embodiments, the path adjustment module includes:

[0304] The constraint adjustment unit is used to adjust the predicted traffic flow path using the adjusted traffic flow path so that the predicted traffic flow path satisfies the adjustment constraint conditions and obtains the target traffic flow path.

[0305] As an embodiment, the constraint adjustment unit may include:

[0306] a center determination subunit, configured to determine a center traffic flow path of the predicted path category according to the plurality of reference traffic flow paths;

[0307] A distance determination subunit, used for respectively calculating the multiple reference traffic paths and the central traffic path to obtain multiple reference distances;

[0308] A constraint determination subunit, configured to determine the largest path distance among the plurality of reference path distances as a constraint distance;

[0309] The constraint adjustment subunit is used to adjust the predicted traffic flow path using the adjusted traffic flow path, so that the path distance between the adjusted predicted traffic flow path and the central traffic flow path is less than the constraint distance, thereby obtaining a target traffic flow path.

[0310] As a possible implementation manner, the center determination subunit includes:

[0311] The algorithm clustering module is used to determine the central traffic path of the predicted path category based on the central clustering algorithm and using the multiple reference traffic paths.

[0312] Further, optionally, the algorithm clustering module includes:

[0313] The third generation unit is used to randomly generate a central traffic flow path;

[0314] A first calculation unit is used to calculate the cluster path distance between the central traffic flow path and each reference traffic flow path to obtain a plurality of cluster path distances;

[0315] A fifth judgment subunit, configured to determine the central traffic flow path if the sum of the squares of the distances of the plurality of cluster paths meets the optimization target;

[0316] The sixth judgment subunit is used to update the central traffic path based on the cluster path distance if the sum of the squares of the multiple cluster path distances does not meet the optimization target, return to the step of calculating the cluster path distance between the central traffic path and each reference traffic path, and continue to execute the step of obtaining multiple cluster path distances.

[0317] As an embodiment, the device may further include:

[0318] The clustering module is used to cluster the historical traffic paths of the target traffic area to obtain at least one path category; wherein each path category includes a set of multiple traffic paths with the same path distribution characteristics.

[0319] As a possible implementation, the clustering module includes:

[0320] A node generation unit, used for generating a plurality of nodes according to a plurality of historical vehicle flow paths in the target traffic area;

[0321] A relationship building unit, used for building a node relationship graph based on the node distance between any two nodes;

[0322] A relationship division unit, used for dividing the node relationship graph into at least one sub-node relationship graph;

[0323] The category determination unit is used to determine that the traffic path corresponding to the node included in each sub-node relationship graph is a path category, and obtain at least one path category.

[0324] Further, optionally, the dividing the node into units may include:

[0325] The first division subunit divides the nodes with the same position features in the node relationship graph into the same sub-node relationship graph to obtain at least one sub-node relationship graph.

[0326] Further, optionally, the first dividing subunit may include:

[0327] The node division module is used to divide the nodes belonging to the same area in the node relationship graph into the same sub-node relationship graph to obtain at least one sub-node relationship graph.

[0328] As a possible implementation manner, the node division module may include:

[0329] A fourth generating unit, used to randomly generate at least one central node;

[0330] A distance calculation unit, used to respectively calculate the node distance between each node and the at least one central node;

[0331] A label determination unit, used to determine the central node with the shortest node distance to each node as the category label of the node;

[0332] A fifth judgment unit, configured to determine that the node corresponding to each central node is a sub-node relationship graph if the sum of the node distances corresponding to all nodes meets the optimal condition, and obtain at least one sub-node relationship graph;

[0333] The sixth judgment unit is used to update the at least one central node for each node corresponding to the central node if the sum of the node distances corresponding to all nodes does not meet the optimal condition, and return to the step of respectively calculating the node distance between each node and at least one central node to continue execution.

[0334] As a possible implementation manner, the relationship building unit may include:

[0335] A node determination subunit, used for determining K similar nodes closest to each node based on the node distance between any two nodes;

[0336] The relationship graph subunit is used to connect each node and the K similar nodes closest to it to form the node relationship graph; wherein the length of each edge in the node relationship graph corresponds to the node distance between the two nodes of the edge.

[0337] Figure 8 The path prediction device can perform Figure 1 , Figure 3 to Figure 6 The implementation principle and technical effect of the path prediction method described in the illustrated embodiment will not be described in detail. The specific manner in which the various modules, units, and subunits of the path prediction device in the above embodiment perform operations has been described in detail in the embodiment of the method, and will not be described in detail here.

[0338] Figure 8 The device shown can be implemented as a path prediction device, such as Fig. 9 FIG. 1 is a schematic diagram of a structure of an embodiment of a path prediction device provided by an embodiment of the present invention, and the device may include:

[0339] A storage component 901 and a processing component 902 ; the storage component 901 is used to store one or more computer instructions, and the one or more computer instructions are used to be called by the processing component 902 .

[0340] The processing component is used to: determine a first vehicle flow path in a target traffic area;

[0341] Based on a preset path prediction model, using the first traffic path to obtain a predicted traffic path;

[0342] Based on a preset category prediction model, the first traffic path is used to obtain a predicted path category; wherein the category prediction model is obtained based on the historical traffic path of the target traffic area and combined with the at least one path category training;

[0343] According to the predicted path category, the predicted traffic path is adjusted to obtain a target traffic path.

[0344] In some embodiments, the path category may be a collection of multiple traffic paths having the same path distribution characteristics.

[0345] As an embodiment, the processing component may also be used for:

[0346] Determining first impact data of at least one path impact factor;

[0347] The processing component may obtain the predicted traffic flow path by using the first traffic flow path based on the preset path prediction model, specifically:

[0348] Based on a preset path prediction model, a predicted traffic path is obtained using the first traffic path and the first influence data; wherein the path prediction model is obtained by training based on the historical traffic path of the target traffic area and the historical influence data of the at least one path influence factor;

[0349] The processing component may obtain the predicted path category using the first vehicle flow path based on the preset category prediction model:

[0350] Based on a preset category prediction model, the predicted path category is obtained using the first traffic path and the first influence data; wherein the category prediction model is obtained based on the historical traffic path of the target traffic area and the historical influence data of the at least one path influence factor, combined with the at least one path category training.

[0351] In some embodiments, the processing component may train and obtain a path prediction model in the following manner:

[0352] Determine at least one path training sample and a path true value corresponding to each path training sample; wherein each path training sample is composed of a historical traffic flow path and historical influence data of at least one path influence factor corresponding to the historical traffic flow path;

[0353] Constructing a path prediction model;

[0354] Based on the at least one path training sample and the path truth value corresponding to each path training sample, model parameters of the path prediction model are obtained through training.

[0355] As a possible implementation manner, the processing component may train and obtain the model parameters of the path prediction model based on the at least one path training sample and the path true value corresponding to each path training sample, specifically:

[0356] Randomly generating path reference parameters of the path prediction model;

[0357] Inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter to obtain a first prediction result corresponding to each path training sample;

[0358] Determine a path prediction error based on a first prediction result corresponding to each path training sample and a path true value;

[0359] If the path prediction error satisfies a first error condition, determining the path reference parameter as a model parameter of the path prediction model;

[0360] If the path prediction error does not meet the first error condition, based on the path prediction error, the path reference parameter is updated, and the process returns to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, and obtaining the first prediction result corresponding to each path training sample.

[0361] In some possible designs, if the path prediction error does not satisfy the first error condition, the processing component updates the path reference parameter based on the path prediction error, and returns to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, respectively, to obtain the first prediction result corresponding to each path training sample, and further performs the following steps:

[0362] If the path prediction error does not satisfy the first error condition, calculating a partial derivative of the path prediction error with respect to a model parameter of the path prediction model to obtain a path adjustment amount;

[0363] Determining a path parameter adjustment amount according to the path adjustment step length and the path adjustment amount;

[0364] The path reference parameter is adjusted using the path parameter adjustment amount, and the process returns to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, and obtaining the first prediction result corresponding to each path training sample.

[0365] In some other possible designs, the processing component determines the path prediction error based on the first prediction result corresponding to each path training sample and the path true value by:

[0366] Calculating the path distance between the first prediction result corresponding to each path training sample and the path true value to obtain the path error of each path training sample;

[0367] A path prediction error is determined according to the path errors respectively corresponding to at least one path training sample.

[0368] In some embodiments, the processing component may train and obtain a category prediction model in the following manner:

[0369] Determine at least one category training sample and a category truth value corresponding to each category training sample; wherein each category training sample is composed of a historical traffic path and historical influence data of at least one path influence factor corresponding to the historical traffic path; the category truth value corresponding to each category training sample is a path category that matches a historical traffic path in the at least one path category that is separated from the historical traffic path corresponding to the historical traffic path by a preset time interval;

[0370] Build a category prediction model;

[0371] Based on the at least one category training sample and the category truth value corresponding to each category training sample, model parameters of the category prediction model are obtained through training.

[0372] As a possible implementation manner, the processing component may train and obtain the model parameters of the category prediction model based on the at least one category training sample and the category truth value corresponding to each category training sample, specifically:

[0373] Randomly generating category reference parameters of the category prediction model;

[0374] Inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter respectively to obtain a second prediction result corresponding to each category training sample;

[0375] Determine a category prediction error based on the second prediction result corresponding to each category training sample and the category true value;

[0376] If the category prediction error satisfies a second error condition, determining the category reference parameter as a model parameter of the category prediction model;

[0377] If the category prediction error does not meet the second error condition, the category reference parameter is adjusted based on the category prediction error, and the process returns to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and obtaining the second prediction result corresponding to each category training sample.

[0378] In some possible designs, if the category prediction error does not satisfy the second error condition, the processing component adjusts the category reference parameter based on the category prediction error, and returns to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and obtaining the second prediction result corresponding to each category training sample, and further performs the following steps:

[0379] If the category prediction error does not satisfy the second error condition, calculating the partial derivative of the category prediction error with respect to the model parameters of the category prediction model to obtain a category adjustment amount;

[0380] Determining a category parameter adjustment amount according to the category adjustment step length and the category adjustment amount;

[0381] The category reference parameter is adjusted using the category parameter adjustment amount, and the step of returning to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and obtaining the second prediction result corresponding to each category training sample continues.

[0382] In some further possible designs, the processing component determines the category prediction error based on the second prediction result corresponding to each category training sample and the category true value, which may be:

[0383] Calculating the category distance between the second prediction result corresponding to each category training sample and the category true value to obtain the category error of each category training sample;

[0384] The category prediction error is determined based on the category error corresponding to each category training sample.

[0385] Further, optionally, the at least one path influencing factor may include: a surrounding traffic influencing factor, a time influencing factor, a weather influencing factor and / or a date influencing factor.

[0386] The historical impact data of the at least one path impact factor may include: surrounding traffic paths, path collection time, path collection weather and / or path collection date.

[0387] As an embodiment, adjusting the predicted traffic path to obtain the target traffic path according to the predicted path category may include:

[0388] Determining a plurality of reference traffic paths corresponding to the predicted path category;

[0389] The predicted traffic path is adjusted using the multiple reference traffic paths to obtain a target traffic path.

[0390] In some possible applications, the processing component uses the multiple reference traffic paths to adjust the predicted traffic path to obtain the target traffic path, which may specifically be:

[0391] Selecting, from the plurality of reference traffic paths, an adjusted traffic path having a first correlation relationship with the predicted traffic path;

[0392] The predicted traffic flow path is adjusted using the adjusted traffic flow path to obtain a target traffic flow path.

[0393] As a possible implementation manner, the processing component may select, from the multiple reference traffic flow paths, an adjusted traffic flow path having a first correlation relationship with the predicted traffic flow path by:

[0394] respectively calculating the path distances between the multiple reference traffic paths and the predicted traffic path to obtain multiple reference path distances;

[0395] According to the multiple reference path distances, a reference traffic path whose path distance from the predicted traffic path is less than a first preset threshold is selected from the multiple reference traffic paths to obtain an adjusted traffic path.

[0396] In some embodiments, the processing component uses the adjusted traffic flow path to adjust the predicted traffic flow path to obtain the target traffic flow path, which may specifically be:

[0397] The predicted traffic flow path is adjusted using the adjusted traffic flow path so that the predicted traffic flow path satisfies the adjustment constraint conditions, thereby obtaining a target traffic flow path.

[0398] As an embodiment, the method of adjusting the predicted traffic flow path by using the adjusted traffic flow path so that the predicted traffic flow path satisfies the adjustment constraint condition and obtaining the target traffic flow path may specifically be:

[0399] Determining a central traffic path of the predicted path category according to the multiple reference traffic paths;

[0400] Calculating the multiple reference traffic paths and the central traffic path respectively to obtain multiple reference distances;

[0401] Determine the largest path distance among the multiple reference path distances as the constraint distance;

[0402] When the predicted traffic flow path is adjusted using the adjusted traffic flow path, the path distance between the adjusted predicted traffic flow path and the central traffic flow path is less than the constraint distance, and a target traffic flow path is obtained.

[0403] As a possible implementation manner, the processing component may determine the central traffic path of the predicted path category according to the multiple reference traffic paths:

[0404] Based on the central clustering algorithm, the central traffic path of the predicted path category is determined using the multiple reference traffic paths.

[0405] Further, optionally, the processing component determines the central traffic path of the predicted path category based on the central clustering algorithm using the multiple reference traffic paths, which may specifically be:

[0406] Randomly generate central traffic paths;

[0407] Calculating the cluster path distance between the central traffic flow path and each reference traffic flow path to obtain multiple cluster path distances;

[0408] If the sum of the squares of the distances of the plurality of cluster paths meets the optimization objective, determining the central traffic flow path;

[0409] If the sum of the squares of the multiple cluster path distances does not meet the optimization objective, the central traffic path is updated based on the cluster path distance, and the process returns to the step of calculating the cluster path distance between the central traffic path and each reference traffic path, and the step of obtaining multiple cluster path distances continues.

[0410] As an embodiment, the processing component may determine at least one path category in the following manner:

[0411] The historical traffic paths of the target traffic area are clustered to obtain at least one path category; wherein each path category includes a set of multiple traffic paths with the same path distribution characteristics.

[0412] As a possible implementation manner, the processing component clusters the historical traffic paths of the target traffic area, and obtaining at least one path category may specifically be:

[0413] generating a plurality of nodes according to a plurality of historical traffic paths of the target traffic area;

[0414] Based on the node distance between any two nodes, a node relationship graph is constructed;

[0415] Dividing the node relationship graph into at least one sub-node relationship graph;

[0416] The traffic path corresponding to the node included in each sub-node relationship graph is determined as a path category, and at least one path category is obtained.

[0417] Further, optionally, the processing component divides the node relationship graph into at least one sub-node relationship graph by:

[0418] Nodes with the same position features in the node relationship graph are divided into the same sub-node relationship graph to obtain at least one sub-node relationship graph.

[0419] Further, optionally, the processing component divides the nodes having the same position features in the node relationship graph into the same sub-node relationship graph to obtain at least one sub-node relationship graph, which may specifically be:

[0420] The nodes in the node relationship graph belonging to the same area are divided into the same sub-node relationship graph to obtain at least one sub-node relationship graph.

[0421] As a possible implementation manner, the processing component divides the nodes having the same position features in the node relationship graph into the same sub-node relationship graph to obtain at least one sub-node relationship graph, which may specifically be:

[0422] Randomly generate at least one central node;

[0423] Calculating the node distance between each node and the at least one central node respectively;

[0424] Determine the central node with the shortest node distance to each node as the category label of the node;

[0425] If the sum of the node distances corresponding to all nodes meets the optimal condition, determine that the node corresponding to each central node is a child node relationship graph, and obtain at least one child node relationship graph;

[0426] If the sum of the node distances corresponding to all nodes does not meet the optimal condition, for each node corresponding to the central node, update the at least one central node, and return to the step of respectively calculating the node distance between each node and the at least one central node to continue execution.

[0427] As a possible implementation, the processing component constructs a node relationship graph based on the node distance between any two nodes by:

[0428] Based on the node distance between any two nodes, determine the K similar nodes closest to each node;

[0429] Each node and the K similar nodes closest to it are connected to form the node relationship graph; wherein the length of each edge in the node relationship graph corresponds to the node distance between the two nodes of the edge.

[0430] Fig. 9 The path prediction device can perform Figure 1 , Figure 3 to Figure 6 The implementation principle and technical effect of the path prediction method described in the illustrated embodiment will not be described in detail. The specific manner in which the processing component of the path prediction device in the above embodiment performs the operation has been described in detail in the embodiment of the method, and will not be described in detail here.

[0431] In addition, the present application also provides a computer-readable storage medium, which stores a computer program; the computer program enables a computer to implement the path prediction method described in any one of the embodiments of the present application when executed.

[0432] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0433] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by adding a necessary general hardware platform, and of course can also be implemented by combining hardware and software. Based on such an understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a computer product, and the present invention can be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0434] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0435] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0436] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0437] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0438] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0439] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0440] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A path prediction method, characterized in that: include: Determine a first traffic flow path of a target traffic area, wherein the first traffic flow path includes a traffic flow segment sequence corresponding to the traffic flow in the target traffic area and a traffic volume corresponding to the traffic flow segment sequence; Based on a preset path prediction model, using the first traffic path to obtain a predicted traffic path; Based on a preset category prediction model, the first traffic path is used to obtain a predicted path category; wherein the category prediction model is obtained based on the historical traffic path of the target traffic area and combined with at least one path category training; The predicted traffic path is adjusted so that the predicted traffic path conforms to the predicted path category to obtain a target traffic path.

2. The method according to claim 1, characterized in that The path category includes a set of multiple traffic paths with the same path distribution characteristics.

3. The method according to claim 1, characterized in that Also includes: Determining first impact data of at least one path impact factor; The step of obtaining a predicted traffic path using the first traffic path based on a preset path prediction model includes: Based on a preset path prediction model, a predicted traffic path is obtained using the first traffic path and the first influence data; wherein the path prediction model is obtained by training based on the historical traffic path of the target traffic area and the historical influence data of the at least one path influence factor; The step of obtaining a predicted path category by using the first traffic path based on a preset category prediction model includes: Based on a preset category prediction model, the predicted path category is obtained using the first traffic path and the first influence data; wherein the category prediction model is obtained based on the historical traffic path of the target traffic area and the historical influence data of the at least one path influence factor, combined with the at least one path category training.

4. The method according to claim 3, characterized in that The path prediction model is trained in the following way: Determine at least one path training sample and a path true value corresponding to each path training sample; wherein each path training sample is composed of a historical traffic flow path and historical influence data of at least one path influence factor corresponding to the historical traffic flow path; Constructing a path prediction model; Based on the at least one path training sample and the path true value corresponding to each path training sample, model parameters of the path prediction model are obtained through training.

5. The method according to claim 4, characterized in that The training to obtain the model parameters of the path prediction model based on the at least one path training sample and the path true value corresponding to each path training sample includes: Randomly generating path reference parameters of the path prediction model; Inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, respectively, to predict and obtain a first prediction result corresponding to each path training sample; Determine a path prediction error based on a first prediction result corresponding to each path training sample and a path true value; If the path prediction error satisfies a first error condition, determining the path reference parameter as a model parameter of the path prediction model; If the path prediction error does not meet the first error condition, based on the path prediction error, the path reference parameter is updated, and the process returns to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, and the step of predicting and obtaining the first prediction result corresponding to each path training sample continues.

6. The method according to claim 5, characterized in that If the path prediction error does not satisfy the first error condition, based on the path prediction error, updating the path reference parameter, and returning to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, respectively, and predicting to obtain the first prediction result corresponding to each path training sample, further comprising: If the path prediction error does not satisfy the first error condition, calculating a partial derivative of the path prediction error with respect to a model parameter of the path prediction model to obtain a path adjustment amount; Determining a path parameter adjustment amount according to the path adjustment step length and the path adjustment amount; The path reference parameter is adjusted using the path parameter adjustment amount, and the process returns to the step of inputting the at least one path training sample into the path prediction model corresponding to the path reference parameter, and predicting and obtaining the first prediction result corresponding to each path training sample.

7. The method according to claim 5, characterized in that The determining of the path prediction error based on the first prediction result corresponding to each path training sample and the path true value comprises: Calculating the path distance between the first prediction result corresponding to each path training sample and the path true value to obtain the path error of each path training sample; A path prediction error is determined according to the path errors respectively corresponding to at least one path training sample.

8. The method according to claim 3, characterized in that The category prediction model is trained in the following way: Determine at least one category training sample and a category truth value corresponding to each category training sample; wherein each category training sample is composed of a historical traffic path and historical influence data of at least one path influence factor corresponding to the historical traffic path; the category truth value corresponding to each category training sample is a path category in the at least one path category that matches a historical traffic path with a preset time interval between the historical traffic path corresponding to the historical traffic path; Build a category prediction model; Based on the at least one category training sample and the category truth value corresponding to each category training sample, model parameters of the category prediction model are obtained through training.

9. The method according to claim 8, characterized in that The step of training and obtaining the model parameters of the category prediction model based on the at least one category training sample and the category true value corresponding to each category training sample includes: Randomly generating category reference parameters of the category prediction model; Inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter respectively to obtain a second prediction result corresponding to each category training sample; Determine a category prediction error based on the second prediction result corresponding to each category training sample and the category true value; If the category prediction error satisfies a second error condition, determining the category reference parameter as a model parameter of the category prediction model; If the category prediction error does not meet the second error condition, the category reference parameter is adjusted based on the category prediction error, and the process returns to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and the step of predicting the second prediction result corresponding to each category training sample continues.

10. The method according to claim 9, characterized in that If the category prediction error does not meet the second error condition, adjusting the category reference parameter based on the category prediction error, and returning to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and predicting to obtain the second prediction result corresponding to each category training sample, further comprises: If the category prediction error does not satisfy the second error condition, calculating the partial derivative of the category prediction error with respect to the model parameters of the category prediction model to obtain a category adjustment amount; Determining a category parameter adjustment amount according to the category adjustment step length and the category adjustment amount; The category reference parameter is adjusted using the category parameter adjustment amount, and the step of returning to the step of inputting the at least one category training sample into the category prediction model corresponding to the category reference parameter, and predicting and obtaining the second prediction result corresponding to each category training sample continues.

11. The method according to claim 10, characterized in that The determining of the category prediction error based on the second prediction result corresponding to each category training sample and the category true value comprises: Calculating the category distance between the second prediction result corresponding to each category training sample and the category true value to obtain the category error of each category training sample; The category prediction error is determined based on the category error corresponding to each category training sample.

12. The method according to any one of claims 3 to 11, characterized in that: The at least one path influencing factor includes: a surrounding traffic influencing factor, a time influencing factor, a weather influencing factor and / or a date influencing factor; The historical impact data of the at least one path impact factor includes: surrounding traffic paths, path collection time, path collection weather and / or path collection date.

13. The method according to claim 1, characterized in that According to the predicted path category, adjusting the predicted traffic flow path to obtain a target traffic flow path includes: Determining a plurality of reference traffic paths corresponding to the predicted path category; The predicted traffic path is adjusted using the multiple reference traffic paths to obtain a target traffic path.

14. The method according to claim 13, characterized in that The step of adjusting the predicted traffic path to obtain the target traffic path by using the multiple reference traffic paths comprises: Selecting, from the plurality of reference traffic paths, an adjusted traffic path having a first correlation relationship with the predicted traffic path; The predicted traffic flow path is adjusted using the adjusted traffic flow path to obtain a target traffic flow path.

15. The method according to claim 14, characterized in that The selecting, from the plurality of reference traffic flow paths, an adjusted traffic flow path having a first correlation relationship with the predicted traffic flow path comprises: respectively calculating the path distances between the multiple reference traffic paths and the predicted traffic path to obtain multiple reference path distances; According to the multiple reference path distances, a reference traffic path whose path distance from the predicted traffic path is less than a first preset threshold is selected from the multiple reference traffic paths to obtain an adjusted traffic path.

16. The method according to claim 14, characterized in that The adjusting the predicted traffic flow path to obtain the target traffic flow path comprises: The predicted traffic flow path is adjusted using the adjusted traffic flow path so that the predicted traffic flow path satisfies the adjustment constraint condition, thereby obtaining the target traffic flow path.

17. The method according to claim 16, characterized in that The step of adjusting the predicted traffic flow path by using the adjusted traffic flow path so that the predicted traffic flow path satisfies the adjustment constraint condition and obtaining the target traffic flow path comprises: Determining a central traffic path of the predicted path category according to the multiple reference traffic paths; Calculating the multiple reference traffic paths and the central traffic path respectively to obtain multiple reference distances; Determine the largest path distance among the multiple reference path distances as the constraint distance; When the predicted traffic flow path is adjusted using the adjusted traffic flow path, the path distance between the adjusted predicted traffic flow path and the central traffic flow path is less than the constraint distance, and a target traffic flow path is obtained.

18. The method according to claim 17, characterized in that Determining the central traffic path of the predicted path category according to the multiple reference traffic paths includes: Based on the central clustering algorithm, the central traffic path of the predicted path category is determined using the multiple reference traffic paths.

19. The method according to claim 18, characterized in that The method of determining the central traffic path of the predicted path category by using the multiple reference traffic paths based on the central clustering algorithm includes: Randomly generate central traffic paths; Calculating the cluster path distance between the central traffic flow path and each reference traffic flow path to obtain multiple cluster path distances; If the sum of the squares of the distances of the plurality of cluster paths meets the optimization objective, determining the central traffic flow path; If the sum of the squares of the multiple cluster path distances does not meet the optimization objective, the central traffic path is updated based on the cluster path distance, and the process returns to the step of calculating the cluster path distance between the central traffic path and each reference traffic path, and the step of obtaining multiple cluster path distances continues.

20. The method according to claim 1, characterized in that The at least one path category is determined by: The historical traffic paths of the target traffic area are clustered to obtain at least one path category; wherein each path category includes a set of multiple traffic paths with the same path distribution characteristics.

21. The method according to claim 20, characterized in that The clustering of the historical traffic paths of the target traffic area to obtain at least one path category comprises: generating a plurality of nodes according to a plurality of historical traffic paths of the target traffic area; Based on the node distance between any two nodes, a node relationship graph is constructed; Dividing the node relationship graph into at least one sub-node relationship graph; The traffic path corresponding to the node included in each sub-node relationship graph is determined as a path category, and at least one path category is obtained.

22. The method according to claim 21, characterized in that The dividing the node relationship graph into at least one sub-node relationship graph comprises: Nodes with the same position features in the node relationship graph are divided into the same sub-node relationship graph to obtain at least one sub-node relationship graph.

23. The method according to claim 22, characterized in that The dividing the nodes having the same position features in the node relationship graph into the same sub-node relationship graph to obtain at least one sub-node relationship graph comprises: The nodes in the node relationship graph belonging to the same area are divided into the same sub-node relationship graph to obtain at least one sub-node relationship graph.

24. The method according to claim 22, characterized in that The dividing the nodes having the same position features in the node relationship graph into the same sub-node relationship graph to obtain at least one sub-node relationship graph comprises: Randomly generate at least one central node; Calculating the node distance between each node and the at least one central node respectively; Determine the central node with the shortest node distance to each node as the category label of the node; If the sum of the node distances corresponding to all nodes meets the optimal condition, determine that the node corresponding to each central node is a child node relationship graph, and obtain at least one child node relationship graph; If the sum of the node distances corresponding to all nodes does not meet the optimal condition, for each node corresponding to the central node, update the at least one central node, and return to the step of respectively calculating the node distance between each node and the at least one central node to continue execution.

25. The method according to claim 21, characterized in that The step of constructing a node relationship graph based on the node distance between any two nodes includes: Based on the node distance between any two nodes, determine the K similar nodes closest to each node; Each node and the K similar nodes closest to it are connected to form the node relationship graph; wherein the length of each edge in the node relationship graph corresponds to the node distance between the two nodes of the edge.

26. A path prediction device, characterized in that: include: A first determination module is used to determine a first traffic path of a target traffic area, wherein the first traffic path includes a traffic segment sequence corresponding to the traffic in the target traffic area and a traffic volume corresponding to the traffic segment sequence; A path prediction module, configured to obtain a predicted traffic path using the first traffic path based on a preset path prediction model; A model prediction module, used to obtain a predicted path category using the first traffic path based on a preset category prediction model; wherein the category prediction model is obtained based on the historical traffic path of the target traffic area and combined with at least one path category training; The path adjustment module is used to adjust the predicted traffic path so that the predicted traffic path conforms to the predicted path category to obtain a target traffic path.

27. A path prediction device, characterized in that: It comprises: a storage component and a processing component; the storage component is used to store one or more computer instructions, and the one or more computer instructions are used to be called by the processing component to execute the path prediction method described in any one of claims 1 to 25.

28. A storage medium, characterized in that: The storage medium is a computer-readable storage medium storing a computer program; The computer program enables a computer to implement the path prediction method according to any one of claims 1 to 25 when executed.

Images