Vehicle trajectory prediction method and device

Abstract

The invention relates to the technical field of rail transit signal control, in particular to a vehicle track prediction method and device. The vehicle determines running state information of the front vehicle based on a sensor, and processes the running state information based on a Kalman filtering algorithm to obtain a first prediction track of the front vehicle; the historical driving information of the preceding vehicle in a second preset time length is obtained, the long-term trend prediction model predicts a second prediction track of the preceding vehicle in the future according to the historical information, the target prediction track of the preceding vehicle is determined according to the first prediction track, the second prediction track, the first weight and the second weight, and the Kalman filtering algorithm can provide short-term high-confidence prediction. The long-term trend prediction model can provide long-term trend correction, the weights of the long-term trend prediction model and the long-term trend prediction model can be dynamically adjusted along with the prediction duration, the prediction results of the long-term trend prediction model and the long-term trend prediction model are fused through different weights, complementation of short-term deterministic prediction and long-term trend learning is carried out, the prediction process has duration performance and long-term prediction capability, and the accuracy of front vehicle track prediction is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rail transit signal control technology, and in particular to a vehicle trajectory prediction method and device. BACKGROUND

[0002] Virtual marshalling refers to keeping a close distance between trains through communication technology to form a dynamic train set, but not like traditional physical marshalling that is mechanically connected. In this marshalling mode, the rear train needs to obtain the accurate position of the front train to perform real-time movement authorization (MA) calculation and control the train with high precision, thereby maintaining the stability of the marshalling and ensuring the safety and efficiency of train operation.

[0003] In related technologies, a fixed safety distance is used for two-train tracking, and the rear train cannot dynamically adjust the safety distance according to the emergency braking and temporary speed limit of the front train. If the safety distance is conservatively designed, the line utilization rate will be low, and if the safety distance is aggressively designed, safety hazards will be caused. If the motion trajectory of the front train can be accurately predicted, the safety distance between the two trains can be accurately controlled according to the predicted motion trajectory.

[0004] Therefore, how to predict the motion trajectory of the front train becomes a problem to be solved. SUMMARY

[0005] Embodiments of the present application provide a vehicle trajectory prediction method and device to solve the problem that the motion trajectory of the front train cannot be accurately predicted in the prior art.

[0006] In a first aspect, the present application provides a vehicle trajectory prediction method applied to a vehicle, the method comprising:

[0007] determining the running state information of the front train based on a sensor, the running state information comprising the position, speed and acceleration of the front train; and processing the running state information based on a Kalman filtering algorithm to obtain a first predicted trajectory of the front train within a first preset time length in the future;

[0008] obtaining historical driving information of the front train within a second preset time length, inputting the historical driving information into a long-term trend prediction model trained in advance to obtain a second predicted trajectory of the front train within the first preset time length in the future; the historical driving information comprises historical running state information, historical driving environment information and historical driving route information; and the long-term trend prediction model is an LSTM;

[0009] determine the target prediction trajectory of the preceding vehicle according to the first weight corresponding to the Kalman filtering algorithm and the first prediction trajectory determined by the Kalman filtering algorithm, and the second weight corresponding to the long-term trend prediction model and the second prediction trajectory determined by the long-term trend prediction model, wherein the first preset time length is in a negative correlation relationship with the first weight, and the first preset time length is in a positive correlation relationship with the second weight.

[0010] In a second aspect, the present application provides a vehicle trajectory prediction device, the device comprising:

[0011] a first prediction module configured to determine running state information of a preceding vehicle based on a sensor, the running state information comprising a position, a speed, and an acceleration of the preceding vehicle, and process the running state information based on a Kalman filtering algorithm to obtain a first prediction trajectory of the preceding vehicle within a first preset time length in the future;

[0012] a second prediction module configured to obtain historical driving information of the preceding vehicle within a second preset time length, input the historical driving information into a long-term trend prediction model that is pre-trained to obtain a second prediction trajectory of the preceding vehicle within the first preset time length in the future, wherein the historical driving information comprises historical running state information, historical driving environment information, and historical driving route information, and the long-term trend prediction model is LSTM;

[0013] a fusion module configured to determine a target prediction trajectory of the preceding vehicle according to a first weight corresponding to the Kalman filtering algorithm and a first prediction trajectory determined by the Kalman filtering algorithm, and a second weight corresponding to the long-term trend prediction model and a second prediction trajectory determined by the long-term trend prediction model, wherein the first preset time length is in a negative correlation relationship with the first weight, and the first preset time length is in a positive correlation relationship with the second weight.

[0014] In a third aspect, an electronic device is provided, the electronic device comprising a processor configured to implement the steps of any of the vehicle trajectory prediction methods described above when executing a computer program stored in a memory.

[0015] In a fourth aspect, a computer readable storage medium is provided, the computer readable storage medium storing a computer program, the computer program being configured to implement the steps of any of the vehicle trajectory prediction methods described above when executed by a processor.

[0016] In the embodiment of the present application, the vehicle determines the running state information of the front vehicle based on the sensor, that is, determines the position, speed and acceleration of the front vehicle, and processes the running state information based on the Kalman filtering algorithm to obtain the first predicted trajectory of the front vehicle within a first preset time length in the future. Meanwhile, the historical driving information of the front vehicle within a second preset time length is obtained, and the historical driving information is input into the long-term trend prediction model which is pre-trained. Since the historical driving information includes historical running state information, historical driving environment information and historical driving route information, the long-term trend prediction model can predict the second predicted trajectory of the front vehicle within the first preset time length in the future according to these historical information. When predicting, the target predicted trajectory of the front vehicle is accurately determined according to the first weight corresponding to the Kalman filtering algorithm and the first predicted trajectory determined by the Kalman filtering algorithm, and the second weight corresponding to the long-term trend prediction model and the second predicted trajectory determined by the long-term trend prediction model. Since the Kalman filtering algorithm can provide short-term high-confidence prediction, and the long-term trend prediction model can provide long-term trend correction, the first weight decreases with the increase of the first preset time length, and the second weight increases with the increase of the first preset time length. The prediction results of the two are fused through different weights, complementary short-term deterministic prediction and long-term trend learning, so that the prediction process has both timeliness and long-term prediction ability, and the accuracy of the front vehicle trajectory prediction is improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0018] Figure 1 A flowchart of a vehicle trajectory prediction process provided by the embodiment of the present application is shown in the figure.

[0019] Figure 2 A state switching process diagram of a hierarchical communication fault-tolerant mechanism provided by the embodiment of the present application is shown in the figure.

[0020] Figure 3 A flowchart of a vehicle trajectory prediction process provided by the embodiment of the present application is shown in the figure.

[0021] Figure 4 A structure diagram of a vehicle trajectory prediction device provided by the embodiment of the present application is shown in the figure.

[0022] Figure 5 A structure diagram of an electronic device provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0023] In order to make the purposes, technical solutions and advantages of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present application.

[0024] In virtual marshalling, when the train-to-train communication quality decreases, such as signal delay, packet loss rate increases or communication interruption, the rear train will not be able to obtain the accurate position information of the front train in real time, which may cause errors in safety margin calculation, tracking interval errors and even marshalling disintegration. Therefore, in order to accurately predict the trajectory of the front train, the embodiments of the present application provide a vehicle trajectory prediction method and device, which fuses a hybrid prediction technology of physical model driving and data driving to perform front train positioning in a complex dynamic scene. The method is applied to a vehicle, the vehicle determines the running state information of the front train based on a sensor, the running state information including the position, speed and acceleration of the front train; and processes the running state information based on a Kalman filtering algorithm to obtain a first predicted trajectory of the front train within a first preset time length in the future; obtains historical driving information of the front train within a second preset time length, inputs the historical driving information into a long-term trend prediction model pre-trained to obtain a second predicted trajectory of the front train within the first preset time length in the future; the historical driving information includes historical running state information, historical driving environment information and historical driving route information; the long-term trend prediction model is LSTM; and determines a target predicted trajectory of the front train according to the first predicted trajectory determined by the first weight corresponding to the Kalman filtering algorithm and the second predicted trajectory determined by the second weight corresponding to the long-term trend prediction model, wherein the first weight is smaller and the second weight is larger as the first preset time length is larger.

[0025] Figure 1 A flowchart of a vehicle trajectory prediction process provided by the embodiments of the present application is shown in FIG. 1, which includes the following steps: Figure 1

[0026] S101: determining the running state information of the front train based on a sensor, the running state information including the position, speed and acceleration of the front train; and processing the running state information based on a Kalman filtering algorithm to obtain a first predicted trajectory of the front train within a first preset time length in the future.

[0027] The vehicle trajectory prediction method provided by the present application is applied to a vehicle.

[0028] ​In order to accurately predict the vehicle trajectory of the preceding vehicle, in the embodiments of the present application, the running state information of the preceding vehicle can be determined based on sensors. The running state information can include data such as the position, speed and acceleration of the preceding vehicle. In the embodiments of the present application, the vehicle can directly obtain the running state information based on the sensors installed by itself, or can calculate the running state information of the preceding vehicle based on the data collected by the sensors installed by itself. In the embodiments of the present application, the running state information can be understood as the real-time state of the preceding vehicle. In the embodiments of the present application, the sensors used to obtain the running state information can involve laser radar, odometer and the like. The laser radar belongs to a vehicle-mounted device and is installed at the front end of the vehicle to detect the relative position of the preceding vehicle in real time. The odometer belongs to a vehicle-mounted device and can calculate the displacement of the vehicle based on the wheel speed and electronic map data.

[0029] Since the Kalman filter algorithm has high real-time performance, relies on linear assumption and is suitable for processing linear systems, after obtaining the running state information of the preceding vehicle, the running state information can be processed based on the Kalman filter algorithm to obtain the first predicted trajectory of the preceding vehicle within a first preset time length in the future. The Kalman filter algorithm can make a "short-term" high-precision prediction, such as a 0.1-3 second trajectory prediction, based on real-time state data of the preceding vehicle such as speed and acceleration, through a linear dynamic model. Moreover, the Kalman filter algorithm can also perform real-time noise suppression, such as eliminating sensor measurement errors. Using the Kalman filter algorithm can be understood as making a deterministic prediction based on a physical kinematic model, such as a uniform acceleration model.

[0030] For ease of understanding, the process of short-term prediction of the Kalman filter algorithm is described below. When processing the running state information, the Kalman filter algorithm first determines a state equation based on the running state information:

[0031]

[0032] wherein x t represents the position of the preceding vehicle determined based on the sensors installed by itself; v t represents the speed of the preceding vehicle determined based on the sensors installed by itself; Δt represents the prediction time; a t represents the acceleration of the preceding vehicle determined based on the sensors installed by itself; x t+1 represents the predicted position of the preceding vehicle from the current time to the prediction time; v t+1 represents the predicted speed of the preceding vehicle from the current time to the prediction time; w x and w v represent process noise, which is dynamically adjusted through a covariance matrix Q. Wherein, After the state equation is determined, the Kalman filter algorithm generates a short-term prediction trajectory within T k seconds in the future based on the kinematic equation wherein, T k represents the first prediction time length; N represents a prediction period (unit: s). It should be noted that how to perform trajectory prediction based on the Kalman filtering algorithm is prior art, and the process will not be described herein.

[0033] S102: Obtain historical driving information of the preceding vehicle within a second preset time length, input the historical driving information into a long-term trend prediction model trained in advance, and obtain a second prediction trajectory of the preceding vehicle within the first preset time length in the future; the historical driving information includes historical running state information, historical driving environment information and historical driving route information; the long-term trend prediction model is LSTM.

[0034] The Kalman filtering algorithm can accurately predict short-term linear behavior, but it cannot model complex nonlinear behavior, such as emergency braking, sudden acceleration, etc. In order to further improve the accuracy of trajectory prediction, in the embodiments of the present application, a long-term trend prediction model can be used to capture long-term time dependence. The historical driving information of the preceding vehicle within a second preset time length can be obtained, which can include historical running state information, historical driving environment information and historical driving route information. For a historical time, the historical running state information can include the position, speed and acceleration of the preceding vehicle at the historical time, the historical driving environment information can include the temporary speed limit at the position of the preceding vehicle at the historical time, and the historical route information can include the slope and curvature of the track at the position of the preceding vehicle at the historical time.

[0035] For example, in the embodiments of the present application, the historical driving information of the preceding vehicle can be intercepted by a sliding window, for example, the data of the past 10 seconds is taken as the input of the long-term trend prediction model.

[0036] After obtaining the historical driving information, the historical driving information can be input into a long-term trend prediction model trained in advance to obtain a second prediction trajectory of the preceding vehicle within a first preset time length in the future. The long-term trend prediction model can be a long short-term memory (LSTM), which can predict complex nonlinear situations. LSTM can learn historical running data, such as braking curve of the preceding vehicle, variation of route slope, capture "long-term" such as 3-10 seconds of nonlinear trend such as deceleration on a curve, temporary speed limit response, and thus accurately predict the trajectory. LSTM processes long-term dependence of time series data, which can be used for prediction of complex environments such as braking performance changes in rainy and snowy weather.

[0037] It should be noted that the training data can be used to train the model when training the long-term trend prediction model. The training data can be a "previous vehicle state-environment-line" triple extracted from historical running data, the previous vehicle state being position, speed, acceleration, and the line topological feature being data such as slope and curvature.

[0038] S103: determining a target prediction trajectory of the previous vehicle according to the first weight corresponding to the Kalman filtering algorithm and the first prediction trajectory determined by the Kalman filtering algorithm, and the second weight corresponding to the long-term trend prediction model and the second prediction trajectory determined by the long-term trend prediction model, wherein the first preset time length is in a negative correlation relationship with the first weight, and the first preset time length is in a positive correlation relationship with the second weight.

[0039] Since the Kalman filtering algorithm can accurately predict the "short-term" trajectory, and the long-term trend prediction model can accurately predict the "long-term" trajectory, in the embodiments of the present application, the first weight is configured in advance for the Kalman filtering algorithm, and the second weight is configured in advance for the long-term trend prediction model. In the embodiments of the present application, different first weights and second weights can be configured in advance for different prediction time lengths. For example, the prediction time length is 5 seconds, the first weight corresponding to the Kalman filtering algorithm is 0.3, the second weight corresponding to the long-term trend prediction model is 0.7; the prediction time length is 1 second, the weight corresponding to the Kalman filtering algorithm is 0.6, and the second weight corresponding to the long-term trend prediction model is 0.4, etc. That is, the size of the first weight decreases as the first preset time length increases, and the size of the second weight increases as the first preset time length increases, that is, the larger the first preset time length, the smaller the first weight, and the larger the second weight. That is, the first preset time length is in a negative correlation relationship with the first weight, and the first preset time length is in a positive correlation relationship with the second weight. It should be noted that the first weight and the second weight can be configured by those skilled in the art as needed.

[0040] After the first prediction trajectory and the second prediction trajectory are determined, the target prediction trajectory of the previous vehicle can be determined according to the first prediction trajectory, the second prediction trajectory, the first weight, and the second weight. For example, assuming that the position corresponding to T+1 time in the first prediction trajectory is X1, and the position corresponding to T+1 time in the second prediction trajectory is X2, then the position corresponding to T+1 time in the determined target prediction trajectory is X1*first weight+X2*second weight. For example, the target prediction trajectory can be determined based on the following formula:

[0041]

[0042] wherein, represents the target prediction trajectory at t+τ time; w KF represents the first weight; represents a first predicted trajectory at t+τ time; w LSTM (τ) represents a second weight; represents a second predicted trajectory at t+τ time.

[0043] In the embodiment of the present application, the vehicle determines the running state information of the preceding vehicle based on the sensor, that is, determines the position, speed and acceleration of the preceding vehicle, and processes the running state information based on the Kalman filtering algorithm to obtain the first predicted trajectory of the preceding vehicle within a first preset time length in the future. At the same time, the historical driving information of the preceding vehicle within a second preset time length is obtained, and the historical driving information is input into the long-term trend prediction model which is trained in advance. Since the historical driving information includes historical running state information, historical driving environment information and historical driving route information, the long-term trend prediction model can predict the second predicted trajectory of the preceding vehicle within the first preset time length in the future according to these historical information. When predicting, the target predicted trajectory of the preceding vehicle is accurately determined according to the first weight corresponding to the Kalman filtering algorithm and the first predicted trajectory determined by the Kalman filtering algorithm, and the second weight corresponding to the long-term trend prediction model and the second predicted trajectory determined by the long-term trend prediction model. Since the Kalman filtering algorithm can provide short-term high-confidence prediction, and the long-term trend prediction model can provide long-term trend correction, the first weight decreases with the increase of the first preset time length, and the second weight increases with the increase of the first preset time length. The prediction results of the two are fused through different weights, which is complementary to the short-term deterministic prediction and long-term trend learning, so that the prediction process has both timeliness and long-term prediction ability, and the accuracy of the preceding vehicle trajectory prediction is improved.

[0044] In order to further improve the accuracy of the preceding vehicle trajectory prediction, on the basis of the above embodiment, in the embodiment of the present application, the long-term trend prediction model comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises a first LSTM and a second LSTM.

[0045] The historical driving information is input into the long-term trend prediction model to obtain the second predicted trajectory of the preceding vehicle within the first preset time length in the future, comprising:

[0046] The historical driving information is input into the input layer;

[0047] The input layer inputs the received historical driving information to the first LSTM of the hidden layer, the first LSTM extracts features from the historical driving information to obtain a first feature matrix and inputs it to the second LSTM, and the second LSTM adjusts and extracts the first feature matrix to obtain a second feature matrix and inputs it to the output layer.

[0048] The output layer performs full connection processing on the second feature matrix to obtain a trajectory offset of the preceding vehicle in a future first preset time length; and determines the second predicted trajectory according to the trajectory offset and a current position of the vehicle.

[0049] In the embodiments of the present application, the long-term trend prediction model can include an input layer, a hidden layer and an output layer, the hidden layer can include two layers of LSTM units, respectively a first LSTM and a second LSTM, and each layer of LSTM units can include 64 neurons.

[0050] When predicting the second predicted trajectory based on the long-term trend prediction model, historical driving information can be input to the input layer. For example, 10 seconds of historical driving information is input to the input layer.

[0051] After receiving the historical driving information, the input layer can input the historical driving information to the first LSTM of the hidden layer, the first LSTM performs feature extraction on the historical driving information to obtain a first feature matrix and input to the second LSTM, and the second LSTM performs adjustment extraction on the first feature matrix to obtain a second feature matrix and input to the output layer.

[0052] After receiving the second feature matrix, the output layer can perform full connection processing on the second feature matrix to obtain a trajectory offset of the preceding vehicle in a future first prediction time length. That is, a trajectory offset of T L seconds in the future

[0053] After obtaining the trajectory offset, output correction can be performed, and the second predicted trajectory can be determined according to the trajectory offset and the current position of the vehicle. That is, the trajectory offset is added to the current position of the vehicle to obtain the second predicted trajectory. That is, the offset is superimposed on the baseline prediction value to obtain the long-term prediction trajectory. It should be noted that the output correction process can be completed by the output layer, or by the electronic device, that is, the output correction process can be completed inside the model or outside the model.

[0054] In order to further improve the accuracy of the preceding vehicle trajectory prediction, on the basis of the above embodiments, in the embodiments of the present application, the determination process of the first weight and the second weight includes:

[0055] According to a preset decay factor and the first preset time length, the first weight and the second weight are determined, the sum of the first weight and the second weight is 1, the first weight exponentially decays with the first preset time length, and the second weight exponentially grows with the first preset time length.

[0056] In order to further improve the accuracy of the trajectory prediction of the preceding vehicle, in the embodiments of the present application, the weight distribution of the Kalman filter and the long-term trend prediction model can be adjusted in a dynamic manner. In the embodiments of the present application, the dynamic weight distribution strategy can be that the weight coefficient dynamically changes with the prediction length τ, the first weight corresponding to the Kalman filter algorithm exponentially decays with the prediction length, and the second weight corresponding to the long-term trend prediction model exponentially grows with the prediction length.

[0057] In the embodiments of the present application, a preset decay factor can be preconfigured, which can be used to control the model switching rate, and the preset decay factor can be 0.5-1.0. Of course, those skilled in the art can also configure it according to the needs.

[0058] In the embodiments of the present application, the first weight and the second weight can be determined according to the preset decay factor and the first preset time length. Specifically, the first weight corresponding to the Kalman filter algorithm can be determined based on the following formula:

[0059] w KF (τ)=e -λτ

[0060] wherein w KF (τ) represents the first weight; λ represents the preset decay factor; and τ represents the first preset time length.

[0061] In the embodiments of the present application, the sum of the first weight and the second weight is 1, and after the first weight is determined, the second weight can be determined according to the first weight. Specifically, the second weight can be determined based on the following formula:

[0062] w LSTM (τ)=1-e -λτ

[0063] wherein w LSTM (τ) represents the second weight; λ represents the preset decay factor; and τ represents the first preset time length.

[0064] In order to further improve the accuracy of the trajectory prediction of the preceding vehicle, on the basis of the above embodiments, in the embodiments of the present application, the determination process of the first weight and the second weight comprises:

[0065] determining the quotient value of the prediction error standard deviation and the maximum allowable error threshold value; and determining a first value according to an error feedback factor weight and the quotient value, wherein the error feedback factor weight is used to correct the prediction error, and the error feedback factor weight is determined according to the quotient value and a preset balance value;

[0066] According to the acceleration mutation factor weight and the acceleration of the front vehicle, a second value is determined, the acceleration mutation factor weight is used to reduce the first weight corresponding to the Kalman filtering algorithm in the sudden acceleration / braking event, and the acceleration mutation factor weight is configured in advance according to historical data, and the acceleration mutation factor weight corresponding to a straight road section is smaller than the acceleration mutation factor weight corresponding to a curve / slope road section;

[0067] According to the dynamic attenuation factor, the first preset time length and the time attenuation factor weight, a third value is determined, the dynamic attenuation factor is used to control the attenuation rate of the weight with the prediction length, and the time attenuation factor weight is used to control the attenuation rate of the Kalman filtering algorithm with the prediction length;

[0068] The sum of the first value, the second value and the third value is determined as the first weight;

[0069] The second weight is determined according to the first weight, and the sum of the first weight and the second weight is 1.

[0070] Although the scheme for determining the first weight and the second weight in the above embodiment can improve the accuracy of positioning the front vehicle in the virtual marshalling, the model switching rate only changes with the prediction time, and the adaptability to curves, slopes, environments and the like still needs to be improved, so in the embodiment of the present application, a way of determining the first weight and the second weight is also provided.

[0071] In the embodiment of the present application, the quotient of the prediction error standard deviation and the maximum allowable error threshold value can be determined. The prediction error standard deviation is the standard deviation of the recent prediction error. In the embodiment of the present application, the prediction error can be determined according to the target prediction trajectory and the actual running information of the front vehicle at the corresponding time, and the prediction error standard deviation can be determined according to the prediction error in the recent set time length. The maximum allowable error threshold value can be set according to the line design specification, for example, defined according to the safety standard, such as SIL4 level requiring error ≤ 3m, therefore, the maximum allowable error threshold value can be set to 3m.

[0072] After the quotient of the prediction error and the maximum allowable error threshold value is determined, the first value can be determined according to the error feedback factor weight and the quotient. For example, the product of the error feedback factor weight and the quotient can be determined as the first value. The error feedback factor is used to dynamically correct the model weight to cope with the prediction error, that is, the error feedback factor is used to correct the prediction error. In the embodiment of the present application, the error feedback factor weight can be determined based on the following formula:

[0073]

[0074] Wherein, γ represents the error feedback factor weight; σerror denotes the standard deviation of the prediction error; σ max denotes the maximum allowable error threshold; 0.1, 0.2 can be understood as a preset balance value.

[0075] In the embodiment of the present application, an acceleration mutation factor weight is configured, which is used to suppress the dependence of the Kalman filtering algorithm on the sudden acceleration / braking event. That is, the acceleration mutation factor weight is used to reduce the first weight corresponding to the Kalman filtering algorithm in the sudden acceleration / braking event. In the embodiment of the present application, the acceleration mutation factor weight can be set according to historical data statistics, for example, 0.2 is set for a straight road section, and 0.4 is set for a curve / slope road section. That is, the acceleration mutation factor weight corresponding to the straight road section is less than the acceleration mutation factor weight corresponding to the curve / slope road section. In the embodiment of the present application, the setting of the acceleration mutation factor weight is not limited to the above examples, and those skilled in the art can also set it according to needs. In the embodiment of the present application, the second value can be determined according to the acceleration mutation factor weight and the acceleration of the preceding vehicle. For example, the second value can be determined according to the following formula:

[0076]

[0077] wherein β denotes the acceleration mutation factor weight, a front denotes the acceleration of the preceding vehicle.

[0078] In the embodiment of the present application, a dynamic attenuation factor and a time attenuation factor weight are also configured, the dynamic attenuation factor is used to control the attenuation rate of the weight with the prediction length. The time attenuation factor weight is used to control the attenuation rate of the Kalman filtering algorithm with the prediction length, and the time attenuation factor weight can be set according to the prediction needs of "short term" and "long term", for example, the time attenuation factor weight can be set as 0.6 by default. In the embodiment of the present application, the third value can be determined according to the dynamic attenuation factor, the first preset time length and the time attenuation factor weight. For example, the third value can be determined based on the following formula:

[0079] The third value = a e -λτ

[0080] wherein a denotes the time attenuation factor weight; λ denotes the dynamic attenuation factor; τ denotes the first preset time length.

[0081] After the first value, the second value and the third value are determined, the sum of the first value, the second value and the third value can be determined as the first weight. Specifically, the first weight can be represented by the following formula:

[0082]

[0083] wherein w KF represents a first weight; a represents a time decay factor weight; l represents a dynamic decay factor; t represents a first preset time length; b represents an acceleration mutation factor weight, a front represents an acceleration of the preceding vehicle; g represents an error feedback factor; s error represents a prediction error standard deviation; s max represents a maximum allowable error threshold.

[0084] In the embodiments of the present application, the sum of the first weight and the second weight is 1, after the first weight is determined, the second weight can be determined according to the first weight. Specifically, the second weight can be determined based on the following formula:

[0085] w LSTM =1-w KF

[0086] wherein w KF represents a first weight; w LSTM represents a second weight.

[0087] The Kalman filter algorithm can provide short-term high-confidence prediction, and the LSTM can provide long-term trend correction. Through dynamic weight fusion, the short-term deterministic prediction and the long-term trend learning are complementary, which can improve the accuracy of the preceding vehicle trajectory prediction in the virtual marshalling scene.

[0088] In order to further improve the accuracy of the preceding vehicle trajectory prediction, in the embodiments of the present application, the determination process of the dynamic decay factor comprises:

[0089] determining an environment complexity score according to the track curvature radius, the slope change rate of the current location and the weather influence coefficient corresponding to the current weather;

[0090] determining the dynamic decay factor according to the basic decay factor, the environment sensitivity coefficient and the environment complexity score.

[0091] In order to further improve the accuracy of the preceding vehicle trajectory prediction, in the embodiments of the present application, the dynamic decay factor can be determined in combination with the current environment. In the embodiments of the present application, the environment complexity score can be determined according to the track curvature radius, the slope change rate of the current location and the weather influence coefficient corresponding to the current weather. Specifically, the environment complexity score can be determined based on the following formula:

[0092]

[0093] wherein EnvScore represents the environment complexity score; R curve represents the track curvature radius (m), normalized to denotes the slope change rate (rad / s), normalized to [0, 1]; W weather denotes the weather influence coefficient, which is obtained by a sensor on the trackside, when the weather condition obtained by the sensor is fine, the W weather is 0, and when the weather condition obtained by the sensor is not fine, such as rainstorm or heavy snow, the W weather is 1.

[0094] After the environmental complexity score is determined, a dynamic attenuation factor can be determined according to the basic attenuation factor, the environmental sensitivity coefficient, and the determined environmental complexity score. Specifically, the dynamic attenuation factor can be determined based on the following formula:

[0095] λ = λ base · (1 + k · EnvScore)

[0096] wherein λ denotes the dynamic attenuation factor; λ base denotes the basic attenuation factor, and the default value is 0.01; k denotes the environmental sensitivity coefficient, and the default value is 0.5. In the embodiments of the present application, the value range of the environmental sensitivity coefficient can be 0.3-0.7.

[0097] In order to further improve the accuracy of the trajectory prediction of the preceding vehicle, on the basis of the above-mentioned embodiments, in the embodiments of the present application, the method further comprises:

[0098] determining a prediction error according to the target prediction trajectory and the actual running information of the preceding vehicle at the corresponding time;

[0099] if the prediction error is greater than an error threshold value for N consecutive times, increasing the second weight according to a preset value to obtain an increased weight;

[0100] selecting the minimum value between the increased weight and a preset weight value as a latest weight, updating the second weight using the latest weight, and adjusting the first weight according to the updated second weight.

[0101] In order to further improve the accuracy of the trajectory prediction of the preceding vehicle, in the embodiments of the present application, an error feedback mechanism is configured. In the embodiments of the present application, a prediction error can be determined according to the target prediction trajectory and the actual running information of the preceding vehicle at the corresponding time. That is, the prediction error is determined according to the predicted data and the actual data of the train actually running to the corresponding time. For example, if the predicted preceding vehicle runs to point A at time A, but actually runs to point B at time A, it can be considered that there is a prediction error, and the prediction error can be determined according to the distance between point A and point B. If the predicted preceding vehicle runs to point A at time A, and actually also runs to point A at time A, it means that there is no prediction error.

[0102] In the embodiments of the present application, the deviation between the prediction error and the error threshold value can be calculated in real time. If the prediction error is greater than the error threshold value for N consecutive times, it can be considered that the weight setting may be unreasonable, and the second weight can be increased by a preset value to obtain an increased weight. The error threshold value can be 2σ error , that is, 2 times the standard deviation of the prediction error. N can be any integer, such as 3, 4, 7, etc. When the second group of weights is determined to be increased, the preset value can be added to the original second weight.

[0103] After obtaining the increased weight, the minimum value between the increased weight and a preset weight can be selected as the latest weight. After the latest weight is determined, the second weight can be updated using the latest weight. That is, the latest weight determined is used as the second weight.

[0104] Since the second weight is updated, the first weight also needs to be adjusted.

[0105] To further improve the accuracy of the front vehicle trajectory prediction, on the basis of the above embodiments, in the embodiments of the present application, the method further comprises:

[0106] If the prediction error is lower than the error threshold value for K consecutive times, the updated second weight is decreased each time when the target prediction trajectory is determined subsequently, until the value of the second weight returns to the second weight before updating.

[0107] In the embodiments of the present application, a recovery condition is also provided. If the prediction error is lower than the error threshold value for K consecutive times, the updated second weight is decreased each time when the target prediction trajectory is determined subsequently, until the value of the second weight returns to the second weight before updating.

[0108] Exemplarily, the trigger condition of the error feedback mechanism: the prediction error exceeds 2σ error for 3 consecutive periods; weight correction: w LSTM = min(w LSTM + 0.2, 0.9), w LSTM represents the updated second weight, w LSTM represents the second weight before updating; 0.2 represents the preset value; and 0.9 represents the preset weight value. Recovery condition: the error is lower than σ error for 5 consecutive periods, and the weight is reduced by 0.05 each period until the second weight before updating.

[0109] To further improve the accuracy of the front vehicle trajectory prediction, on the basis of the above embodiments, in the embodiments of the present application, before determining the state information of the front vehicle based on the sensor, the method further comprises:

[0110] If the position information sent by the front vehicle is not received within the third preset time length, and at least one sensor is in an abnormal running state, a marshalling decoupling early warning is triggered, wherein the sensor is a sensor for determining the state information of the front vehicle.

[0111] The front and rear vehicle following control based on vehicle-to-vehicle communication in the related art virtual marshalling has no hierarchical fault-tolerant mechanism, and emergency braking or decoupling is triggered when communication is interrupted. In the embodiment of the present application, in order to ensure the normal operation of the vehicle, before determining the state information of the front vehicle based on the sensor, it can be determined whether the position information sent by the front vehicle is received within the third preset time length, and whether all the sensors installed by itself are in a normal running state. The third preset time length can be configured by a person skilled in the art according to safety indicators.

[0112] If it is determined that the position information sent by the front vehicle is not received within the third preset time length, and at least one sensor is in an abnormal running state, the subsequent step of determining the state information of the front vehicle based on the sensor installed by itself is continued. Since the vehicle is installed with more sensors, and the functions of each sensor are not the same, in the embodiment of the present application, when determining whether the vehicle trajectory prediction can be performed, it can be determined whether the sensors for determining the state information of the front vehicle are all in a normal running state. Which sensors are used to determine the state information of the front vehicle can be pre-set and identified.

[0113] If the position information sent by the front vehicle is not received within the third preset time length, and the sensors installed by itself are not all in a normal running state, a marshalling decoupling early warning is triggered.

[0114] The process of the state switching of the hierarchical communication fault-tolerant mechanism will be described below in combination with a specific embodiment, Figure 2 A process diagram of the state switching of the hierarchical communication fault-tolerant mechanism provided in the embodiment of the present application is shown in FIG. 1. Figure 2 The communication delay time with the front vehicle is obtained in real time.

[0115] If the delay time is less than or equal to the third preset time length 100 ms, it means that the communication is normal, and the position of the front vehicle can be set as the position in the report of the front vehicle. That is, in the vehicle distance control process, the position of the front vehicle is determined according to the position sent by the front vehicle.

[0116] If the delay time is not less than or equal to the third preset time length 100 ms, it means that the communication is abnormal, and it can be further determined whether all the sensors installed by itself are in a normal running state. If yes, the position of the front vehicle can be calculated using a hybrid prediction model, that is, the step of determining the state information of the front vehicle based on the sensor installed by itself is continued.

[0117] If the delay time is not less than or equal to the third preset time length 100 ms, and at least one sensor is in an abnormal operating state, it is indicated that the vehicle can not be able to obtain the state information of the preceding vehicle based on the sensors installed by itself, the position of the preceding vehicle can be set to be invalid, and a marshalling decoupling warning can be triggered.

[0118] If a single model is used for positioning the preceding vehicle, there are defects, pure physical models (such as Kalman filter algorithm) are difficult to capture complex nonlinear behavior, and pure data-driven models (such as LSTM) have high response delay to sudden changes in the preceding vehicle. In the embodiments of the present application, in the case that the vehicle-to-vehicle communication condition is poor, such as unstable or interrupted, high-precision short-term deterministic trajectory prediction of the preceding vehicle is realized by Kalman filtering, long-term nonlinear trend under complex environment (including line slope, curvature, speed limit, etc.) is captured by combining LSTM, and dynamic weight distribution strategy is used to realize smooth transition of the two types of models, which improves the real-time and accuracy of the preceding vehicle positioning in complex dynamic scenarios, and solves the problems of poor model adaptability and static safety margin in traditional preceding vehicle positioning methods.

[0119] In the embodiments of the present application, through the dynamic fusion prediction model of Kalman filter algorithm and LSTM neural network, the real-time and accuracy of the preceding vehicle positioning are significantly improved while ensuring safety, compared with single model method, the short-term prediction accuracy is improved, the long-term prediction error is reduced, and the line utilization rate is improved, which solves the core contradiction of excessive safety distance redundancy and insufficient adaptability in dynamic scenarios in traditional technology, and is especially suitable for efficient and safe control in high-density and high-dynamic scenarios such as virtual marshalling.

[0120] The vehicle trajectory prediction process will be described below in conjunction with a specific embodiment, Figure 3 A vehicle trajectory prediction process provided in the embodiments of the present application is shown in Figure 3 As shown, first, the sensor data, such as the operating state information of the preceding vehicle, is obtained. Then, the environment data is obtained, such as the line topology data (slope θ, curvature κ) from the electronic map, the environment data (track friction coefficient μ, temporary speed limit) from the electronic map, and the weather conditions of the driving section, etc.

[0121] After obtaining the above data, the obtained data is made as a historical data set for subsequent trajectory prediction.

[0122] In order to facilitate subsequent processing, the collected data can be normalized. For example, the slope, curvature, etc. are mapped to the range of 0-1, etc.

[0123] In the trajectory prediction, the Kalman filtering algorithm can be used to process the latest acquired running state information to obtain a first predicted trajectory. Then, the LSTM neural network is used to process the historical driving information of the preceding vehicle within a second preset time length to obtain a second predicted trajectory.

[0124] When the first predicted trajectory and the second predicted trajectory are acquired, the dynamic weight distribution module can determine a first weight corresponding to the Kalman filtering algorithm and a second weight corresponding to the LSTM based on the current running state.

[0125] Finally, the target predicted trajectory of the preceding vehicle is determined according to the first predicted trajectory, the second predicted trajectory, the first weight and the second weight.

[0126] In the embodiments of the present application, the hybrid architecture of physical model driving + data driving is used to combine real-time performance and long-term prediction capability, and to solve the key problem of calculating the position of the preceding vehicle in the virtual marshalling. The innovation is not only in the algorithm combination, but also in the design of the dynamic weight distribution strategy and the hierarchical communication fault-tolerant mechanism. Through the hierarchical communication fault-tolerant mechanism and the fusion algorithm, the positioning accuracy of the following vehicle to the preceding vehicle can be maintained under unreliable communication conditions, and the safety margin is avoided to be excessively compressed or false alarm braking, so as to ensure the stability and driving safety of the virtual marshalling, and significantly improve the system robustness in complex scenarios.

[0127] Based on the same inventive concept, the embodiments of the present application provide a vehicle trajectory prediction device, Figure 4 For the structure schematic diagram of the vehicle trajectory prediction device provided by the embodiments of the present application, please refer to Figure 4 The device comprises:

[0128] The first prediction module 401 is configured to determine the running state information of the preceding vehicle based on the sensor, wherein the running state information comprises the position, speed and acceleration of the preceding vehicle; and process the running state information based on the Kalman filtering algorithm to obtain a first predicted trajectory of the preceding vehicle within a first preset time length in the future;

[0129] The second prediction module 402 is configured to acquire the historical driving information of the preceding vehicle within a second preset time length, input the historical driving information into a pre-trained long-term trend prediction model, and obtain a second predicted trajectory of the preceding vehicle within the first preset time length in the future; the historical driving information comprises historical running state information, historical driving environment information and historical driving route information; and the long-term trend prediction model is an LSTM.

[0130] The fusion module 403 is configured to determine a target prediction trajectory of the preceding vehicle according to a first weight corresponding to the Kalman filtering algorithm and a first prediction trajectory determined by the Kalman filtering algorithm, and a second weight corresponding to the long-term trend prediction model and a second prediction trajectory determined by the long-term trend prediction model, wherein the first preset time length is in a negative correlation relationship with the first weight, and the first preset time length is in a positive correlation relationship with the second weight.

[0131] In a possible implementation, the historical running state information includes a position, a speed, and an acceleration; the historical driving environment information includes a temporary speed limit at the position; and the historical route information includes a slope and a curvature of a track at the position.

[0132] In a possible implementation, the long-term trend prediction model includes an input layer, a hidden layer, and an output layer, and the hidden layer includes a first LSTM and a second LSTM.

[0133] The second prediction module 402 is configured to input the historical driving information into the input layer, and the input layer inputs the received historical driving information into the first LSTM of the hidden layer, the first LSTM extracts features from the historical driving information to obtain a first feature matrix and input the first feature matrix into the second LSTM, the second LSTM adjusts and extracts the first feature matrix to obtain a second feature matrix and input the second feature matrix into the output layer, the output layer performs full connection processing on the second feature matrix to obtain a track offset of the preceding vehicle within the first preset time length in the future, and the second prediction trajectory is determined according to the track offset and a current position of the vehicle.

[0134] In a possible implementation, the device further includes:

[0135] The determination module 404 is configured to determine the first weight and the second weight according to a preset attenuation factor and the first preset time length, a sum of the first weight and the second weight is 1, the first weight exponentially attenuates with the first preset time length, and the second weight exponentially grows with the first preset time length.

[0136] In a possible implementation, the determining module 404 is further configured to determine a quotient value of the prediction error standard deviation and the maximum allowed error threshold value; determine a first value according to an error feedback factor weight and the quotient value, the error feedback factor weight being used to correct the prediction error, the error feedback factor weight being determined according to the quotient value and a preset balance value; determine a second value according to an acceleration mutation factor weight and the acceleration of the preceding vehicle, the acceleration mutation factor weight being used to reduce the first weight of the Kalman filtering algorithm in a sudden acceleration / braking event, the acceleration mutation factor weight being preconfigured according to historical data, and the acceleration mutation factor weight of a straight road section being smaller than that of a curve / slope road section; determine a third value according to a dynamic attenuation factor, the first preset time length, and a time attenuation factor weight, the dynamic attenuation factor being used to control the attenuation rate of the weight with the prediction time length, and the time attenuation factor weight being used to control the attenuation rate of the Kalman filtering algorithm with the prediction time length; determine a sum of the first value, the second value, and the third value as the first weight; determine the second weight according to the first weight, and a sum of the first weight and the second weight being 1.

[0137] In a possible implementation, the determining module 404 is further configured to determine an environment complexity score according to a track curvature radius, a slope change rate, and a weather influence coefficient corresponding to a current weather of a current location; and determine the dynamic attenuation factor according to a basic attenuation factor, an environment sensitivity coefficient, and the environment complexity score.

[0138] In a possible implementation, the determining module 404 is further configured to determine a prediction error according to the target prediction trajectory and actual running information of the preceding vehicle at a corresponding time; if the prediction error is greater than an error threshold value for consecutive N times, increase the second weight according to a preset value to obtain an increased weight; select a minimum value of the increased weight and a preset weight value as a latest weight, update the second weight using the latest weight, and adjust the first weight according to the updated second weight.

[0139] In a possible implementation, the determining module 404 is further configured to, if the prediction error is lower than the error threshold value for consecutive K times, perform decreasing processing on the updated second weight each time the target prediction trajectory is determined subsequently, until the value of the second weight returns to that of the second weight before the update.

[0140] In a possible implementation, the determining module 404 is further configured to, if no position information sent by the preceding vehicle is received within a third preset time length, and at least one sensor is in an abnormal running state, trigger a marshalling and uncoupling early warning, wherein the sensor is a sensor used to determine running state information of the preceding vehicle.

[0141] Based on the same inventive concept, the embodiment of the present application provides an electronic device which can implement the steps of the vehicle trajectory prediction method discussed above, Figure 5 An electronic device structure schematic diagram provided by the embodiment of the present application is shown in the figure, which includes a processor 501, a communication interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communication interface 502 and the memory 503 complete mutual communication through the communication bus 504; Figure 5

[0142] The memory 503 stores a computer program, and when the program is executed by the processor 501, the processor 501 executes the following steps:

[0143] Determine the running state information of the preceding vehicle based on the sensor, wherein the running state information includes the position, speed and acceleration of the preceding vehicle; and process the running state information based on the Kalman filtering algorithm to obtain the first predicted trajectory of the preceding vehicle within a first preset time length in the future;

[0144] Obtain the historical driving information of the preceding vehicle within a second preset time length, input the historical driving information into a long-term trend prediction model which is pre-trained, and obtain the second predicted trajectory of the preceding vehicle within the first preset time length in the future; the historical driving information includes historical running state information, historical driving environment information and historical driving route information; the long-term trend prediction model is LSTM;

[0145] Determine the target predicted trajectory of the preceding vehicle according to the first predicted trajectory determined by the first weight corresponding to the Kalman filtering algorithm and the second predicted trajectory determined by the second weight corresponding to the long-term trend prediction model, wherein the first preset time length and the first weight are in a negative correlation relationship, and the first preset time length and the second weight are in a positive correlation relationship.

[0146] In a possible implementation, the historical running state information includes position, speed and acceleration; the historical driving environment information includes temporary speed limit at the location; and the historical route information includes the slope and curvature of the track at the location.

[0147] In a possible implementation, the long-term trend prediction model includes an input layer, a hidden layer and an output layer, and the hidden layer includes a first LSTM and a second LSTM.

[0148] The inputting of the historical driving information into the long-term trend prediction model which is pre-trained to obtain the second predicted trajectory of the preceding vehicle within the first preset time length in the future includes:

[0149] ​inputting the historical driving information to the input layer;

[0150] The input layer inputs the received historical driving information to a first LSTM of the hidden layer, the first LSTM performs feature extraction on the historical driving information to obtain a first feature matrix and inputs the first feature matrix to a second LSTM, the second LSTM performs adjustment extraction on the first feature matrix to obtain a second feature matrix and inputs the second feature matrix to the output layer;

[0151] The output layer performs full connection processing on the second feature matrix to obtain a trajectory offset of the preceding vehicle within the first preset time length in the future; and determines the second predicted trajectory according to the trajectory offset and a current position of the vehicle.

[0152] In a possible implementation, the determination of the first weight and the second weight comprises:

[0153] The first weight and the second weight are determined according to a preset decay factor and the first preset time length, a sum of the first weight and the second weight is 1, the first weight exponentially decays with the first preset time length, and the second weight exponentially grows with the first preset time length.

[0154] In a possible implementation, the determination of the first weight and the second weight comprises:

[0155] A quotient of a prediction error standard deviation and a maximum allowed error threshold is determined, a first value is determined according to an error feedback factor weight and the quotient, the error feedback factor weight is used to correct a prediction error, and the error feedback factor weight is determined according to the quotient and a preset balance value;

[0156] A second value is determined according to an acceleration mutation factor weight and an acceleration of the preceding vehicle, the acceleration mutation factor weight is used to reduce a first weight corresponding to the Kalman filtering algorithm in a sudden acceleration / braking event, the acceleration mutation factor weight is configured in advance according to historical data, and the acceleration mutation factor weight corresponding to a straight road section is smaller than the acceleration mutation factor weight corresponding to a curved road / slope road section;

[0157] A third value is determined according to a dynamic decay factor, the first preset time length and a time decay factor weight, the dynamic decay factor is used to control a decay rate of a weight with a prediction length, and the time decay factor weight is used to control a decay rate of the Kalman filtering algorithm with the prediction length;

[0158] A sum of the first value, the second value and the third value is determined as the first weight;

[0159] The second weight is determined according to the first weight, and a sum of the first weight and the second weight is 1.

[0160] In a possible implementation, the process of determining the dynamic attenuation factor comprises:

[0161] According to a track curvature radius, a slope change rate of a current position, and a weather influence coefficient corresponding to a current weather, an environmental complexity score is determined.

[0162] According to a basic attenuation factor, an environmental sensitivity coefficient, and the environmental complexity score, the dynamic attenuation factor is determined.

[0163] In a possible implementation, the method further comprises:

[0164] According to the target prediction trajectory and actual running information of the preceding vehicle at a corresponding time, a prediction error is determined.

[0165] If the prediction error is greater than an error threshold value for consecutive N times, the second weight is increased according to a preset value to obtain an increased weight.

[0166] The minimum value between the increased weight and a preset weight value is selected as a latest weight, the second weight is updated using the latest weight, and the first weight is adjusted according to the updated second weight.

[0167] In a possible implementation, the method further comprises:

[0168] If the prediction error is lower than the error threshold value for consecutive K times, the updated second weight is processed in a decreasing manner each time the target prediction trajectory is determined subsequently, until the value of the second weight returns to the second weight before the update.

[0169] In a possible implementation, before the state information of the preceding vehicle is determined based on the sensor, the method further comprises:

[0170] If no position information sent by the preceding vehicle is received within a third preset time length, and at least one sensor is in an abnormal running state, a marshalling and uncoupling early warning is triggered, wherein the sensor is a sensor used to determine the running state information of the preceding vehicle.

[0171] Since the principle of solving the problem of the electronic device is similar to that of the vehicle trajectory prediction method, implementation of the electronic device can refer to the embodiments of the method, and repeated parts will not be described herein.

[0172] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface 502 is used for communication between the above electronic device and other devices. The memory can include a Random Access Memory (RAM) and can also include a Non-Volatile Memory (NVM), such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the aforementioned processor.

[0173] The processor mentioned above can be a general-purpose processor, including a central processing unit, a network processor (NP), etc.; can also be a Digital Signal Processing (DSP), an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc.

[0174] Based on the same inventive concept, the embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores a computer program executable by a processor. When the program runs on the processor, the processor executes the following steps:

[0175] Based on the sensor, the running state information of the preceding vehicle is determined, and the running state information includes the position, speed and acceleration of the preceding vehicle; and based on a Kalman filtering algorithm, the running state information is processed to obtain a first predicted trajectory of the preceding vehicle within a first preset time length in the future;

[0176] The historical driving information of the preceding vehicle within a second preset time length is obtained, the historical driving information is input into a long-term trend prediction model which is pre-trained, and a second predicted trajectory of the preceding vehicle within the first preset time length in the future is obtained; the historical driving information includes historical running state information, historical driving environment information and historical driving route information; and the long-term trend prediction model is an LSTM;

[0177] According to the first weight corresponding to the Kalman filtering algorithm and the first predicted trajectory determined by the Kalman filtering algorithm, and the second weight corresponding to the long-term trend prediction model and the second predicted trajectory determined by the long-term trend prediction model, a target predicted trajectory of the preceding vehicle is determined, wherein the first preset time length is in a negative correlation relationship with the first weight, and the first preset time length is in a positive correlation relationship with the second weight.

[0178] In a possible implementation, the historical running state information includes a position, a speed, and an acceleration; the historical driving environment information includes a temporary speed limit at the position; and the historical route information includes a slope and a curvature of a track at the position.

[0179] In a possible implementation, the long-term trend prediction model includes an input layer, a hidden layer, and an output layer, and the hidden layer includes a first LSTM and a second LSTM.

[0180] The inputting of the historical driving information into the long-term trend prediction model that is pre-trained includes:

[0181] The historical driving information is input into the input layer.

[0182] The input layer inputs the received historical driving information into the first LSTM of the hidden layer, the first LSTM extracts features from the historical driving information to obtain a first feature matrix and input the first feature matrix into the second LSTM, and the second LSTM adjusts and extracts the first feature matrix to obtain a second feature matrix and input the second feature matrix into the output layer.

[0183] The output layer performs full connection processing on the second feature matrix to obtain a trajectory offset of the preceding vehicle in the future first preset time length, and determines the second predicted trajectory according to the trajectory offset and a current position of the vehicle.

[0184] In a possible implementation, the determination of the first weight and the second weight includes:

[0185] The first weight and the second weight are determined according to a preset attenuation factor and the first preset time length, a sum of the first weight and the second weight is 1, the first weight exponentially attenuates with the first preset time length, and the second weight exponentially grows with the first preset time length.

[0186] In a possible implementation, the determination of the first weight and the second weight includes:

[0187] determining a first value according to a quotient of the prediction error standard deviation and the maximum allowed error threshold, and an error feedback factor weight used to correct the prediction error, the error feedback factor weight being determined according to the quotient and a preset balance value;

[0188] determining a second value according to the acceleration mutation factor weight and the acceleration of the preceding vehicle, the acceleration mutation factor weight being used to reduce the first weight of the Kalman filtering algorithm in a sudden acceleration / braking event, the acceleration mutation factor weight being configured in advance according to historical data, and the acceleration mutation factor weight corresponding to a straight road section being smaller than the acceleration mutation factor weight corresponding to a curved road / slope road section;

[0189] determining a third value according to a dynamic attenuation factor, the first preset time length and a time attenuation factor weight, the dynamic attenuation factor being used to control the attenuation rate of the weight with the prediction time length, and the time attenuation factor weight being used to control the attenuation rate of the Kalman filtering algorithm with the prediction time length;

[0190] determining the first weight as a sum of the first value, the second value and the third value;

[0191] determining the second weight according to the first weight, and a sum of the first weight and the second weight being 1.

[0192] In a possible implementation, the determination process of the dynamic attenuation factor includes:

[0193] determining an environment complexity score according to a track curvature radius, a slope change rate of a current position and a weather influence coefficient corresponding to a current weather;

[0194] determining the dynamic attenuation factor according to a basic attenuation factor, an environment sensitivity coefficient and the environment complexity score.

[0195] In a possible implementation, the method further includes:

[0196] determining a prediction error according to the target prediction trajectory and actual running information of the preceding vehicle at a corresponding time;

[0197] if the prediction error is greater than an error threshold for N consecutive times, increasing the second weight according to a preset value to obtain an increased weight;

[0198] selecting a minimum value between the increased weight and a preset weight value as a latest weight, updating the second weight using the latest weight, and adjusting the first weight according to the updated second weight.

[0199] In a possible implementation, the method further comprises:

[0200] If the prediction error is lower than the error threshold for K consecutive times, then the updated second weight is decreased in each subsequent determination of the target prediction trajectory until the value of the second weight returns to the second weight before the update.

[0201] In a possible implementation, before determining the state information of the preceding vehicle based on the sensor, the method further comprises:

[0202] If the position information sent by the preceding vehicle is not received within the third preset time length, and at least one sensor is in an abnormal operating state, a marshalling and uncoupling early warning is triggered, wherein the sensor is a sensor for determining the operating state information of the preceding vehicle.

[0203] Since the above computer-readable storage medium solves the problem by the same principle as the vehicle trajectory prediction method, the implementation of the above computer-readable storage medium can be referred to the implementation of the method, and the repeated parts will not be described here.

[0204] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take 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 code.

[0205] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as a combination of flows and / or blocks in the flowcharts and / or block diagrams, 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 apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0206] These computer program instructions can also be stored in a computer-readable memory that can guide the computer or other programmable data processing apparatus to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1the function specified in the one or more blocks.

[0207] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of user operations steps are performed on the computer or other programmable data processing devices to generate a computer-implemented process, so that the instructions executed on the computer or other programmable devices provide a process for implementing the flowchart Figure 1 the flowchart or flowcharts and / or a block Figure 1 the steps of the function specified in the one or more blocks.

[0208] Obviously, various modifications and changes can be made to the present application by those skilled in the art without departing from the spirit and scope of the application. Accordingly, it is intended that all such modifications and changes be included within the scope of the application as long as the modified and changed application fall within the scope of the claims and their equivalents.

Claims

1. A vehicle trajectory prediction method, characterized in that: Applied to a vehicle, the method comprises: determining, based on sensors, operating status information of a leading vehicle, the operating status information including a position, speed, and acceleration of the leading vehicle; and processing the operating status information based on a Kalman filter algorithm to obtain a first predicted trajectory of the leading vehicle within a first preset time period in the future; Obtaining historical driving information of the preceding vehicle within a second preset time period, inputting the historical driving information into a pre-trained long-term trend prediction model to obtain a second predicted trajectory of the preceding vehicle within the first preset time period in the future; the historical driving information includes historical operating state information, historical driving environment information, and historical driving route information; the long-term trend prediction model is an LSTM; The target predicted trajectory of the leading vehicle is determined based on the first weight corresponding to the Kalman filter algorithm and the first predicted trajectory determined thereby, and the second weight corresponding to the long-term trend prediction model and the second predicted trajectory determined thereby, wherein the first preset time length is negatively correlated with the first weight, and the first preset time length is positively correlated with the second weight.

2. The method according to claim 1, characterized in that The historical operating status information includes location, speed, and acceleration; the historical driving environment information includes the temporary speed limit at the location; and the historical route information includes the slope and curvature of the track at the location.

3. The method according to claim 1, characterized in that The long-term trend prediction model includes an input layer, a hidden layer and an output layer, wherein the hidden layer includes a first LSTM and a second LSTM; Inputting the historical driving information into a pre-trained long-term trend prediction model to obtain a second predicted trajectory of the preceding vehicle within the first preset time period in the future includes: Inputting the historical driving information into the input layer; The input layer inputs the received historical driving information into the first LSTM of the hidden layer. The first LSTM extracts features from the historical driving information to obtain a first feature matrix and inputs the first feature matrix into the second LSTM. The second LSTM adjusts and extracts the first feature matrix to obtain a second feature matrix and inputs the second feature matrix into the output layer. The output layer performs full connection processing on the second feature matrix to obtain a trajectory offset of the leading vehicle within the first preset time length in the future; and determines the second predicted trajectory based on the trajectory offset and the current position of the vehicle.

4. The method according to claim 1, wherein The process of determining the first weight and the second weight includes: The first weight and the second weight are determined according to a preset attenuation factor and the first preset time length, the sum of the first weight and the second weight is 1, the first weight exponentially decays with the first preset time length, and the second weight exponentially increases with the first preset time length.

5. The method according to claim 1, wherein The process of determining the first weight and the second weight includes: Determining a quotient of a prediction error standard deviation and a maximum allowable error threshold; and determining a first value based on an error feedback factor weight and the quotient, wherein the error feedback factor weight is used to correct the prediction error, and the error feedback factor weight is determined based on the quotient and a preset balance value; determining a second value based on an acceleration mutation factor weight and the acceleration of the preceding vehicle, wherein the acceleration mutation factor weight is used to reduce the first weight corresponding to the Kalman filter algorithm during a sudden acceleration / braking event, and the acceleration mutation factor weight is pre-configured based on historical data, with the acceleration mutation factor weight corresponding to a straight section being smaller than the acceleration mutation factor weight corresponding to a curved / sloped section; Determining a third value based on a dynamic attenuation factor, the first preset time length, and a time attenuation factor weight, wherein the dynamic attenuation factor is used to control the attenuation rate of the weight with the predicted time length, and the time attenuation factor weight is used to control the attenuation rate of the Kalman filter algorithm with the predicted time length; determining a sum of the first value, the second value, and the third value as the first weight; The second weight is determined according to the first weight, and the sum of the first weight and the second weight is 1.

6. The method according to claim 5, characterized in that The process of determining the dynamic attenuation factor includes: Determine the environmental complexity score based on the track curvature radius, slope change rate, and weather impact coefficient corresponding to the current weather at the current location; The dynamic attenuation factor is determined according to the basic attenuation factor, the environmental sensitivity coefficient and the environmental complexity score.

7. The method according to any one of claims 1 to 6, characterized in that The method further comprises: Determining a prediction error based on the target predicted trajectory and actual operating information of the preceding vehicle at a corresponding time; If the prediction error is greater than the error threshold for N consecutive times, the second weight is increased according to a preset value to obtain an increased weight; The minimum value between the increased weight and the preset weight value is selected as the latest weight, and the second weight is updated using the latest weight, and the first weight is adjusted according to the updated second weight.

8. The method according to claim 7, characterized in that The method further comprises: If the prediction error is lower than the error threshold for K consecutive times, the updated second weight is decreased each time the target prediction trajectory is determined subsequently until the value of the second weight is restored to the second weight before the update.

9. The method according to claim 1, characterized in that Before determining the status information of the preceding vehicle based on the sensor, the method further includes: If the position information sent by the preceding vehicle is not received within the third preset time length and at least one sensor is in an abnormal operating state, a train separation warning is triggered, wherein the sensor is a sensor used to determine the operating status information of the preceding vehicle.

10. An electronic device, characterized in that: The electronic device includes a processor, and the processor is used to implement the steps of the vehicle trajectory prediction method according to any one of claims 1 to 9 when executing a computer program stored in a memory.

Images