Vehicle operating state prediction method and device

Abstract

The application provides a vehicle operation state prediction method and device, relates to the technical field of vehicle motion control and state prediction, and obtains operation state data of a current operation condition and corresponding gear signals, determines a speed direction constraint according to the current gear signals, embeds the model deduction boundary condition into a multi-step time sequence prediction process, can make the prediction result always be guided by the gear physical law in the generation process, ensures the self-consistency of the speed direction and the continuity of the trajectory, and eliminates the physical violation and action jerk caused by the after-truncation from the root.

Description

Technical Field

[0001] This invention relates to the field of vehicle motion control and state prediction technology, and in particular to a method and apparatus for predicting vehicle operating state. Background Technology

[0002] Vehicle motion control aims to achieve stable tracking of a desired path by autonomous vehicles through precise trajectory planning and low-level execution control. Accurate prediction of the vehicle's operating state (such as speed, acceleration, yaw rate, etc.) over a future time period is a prerequisite for supporting the aforementioned control decisions. In existing technologies, compared to physical mechanism models that rely on precise system parameters and complex differential equations, data-driven neural networks, with their powerful nonlinear fitting capabilities, are widely used in vehicle dynamics modeling and state prediction. Existing technologies use purely data-driven neural networks to perform long-term, multi-step predictions of the vehicle's operating state and directly use the prediction results for trajectory planning and low-level control. However, in complex conditions such as intelligent parking, involving multiple gear shifts, frequent reversing, and maneuvering, the predicted speed direction may contradict the direction of motion defined by the current gear; for example, outputting negative speed in forward gear and positive speed in reverse gear. To address this issue, existing technologies cascade a rule-based "state limiter" at the model output. This limiter corrects the predicted value post-processing based on the gear position signal, forcing speeds that violate directional constraints to zero or reverse, thus ensuring that the final output speed direction is consistent with the gear position signal in terms of numerical value.

[0003] However, this correction step is non-differentiable during the training phase, preventing the network from establishing a correspondence between gear and speed direction during training. The root cause of physical violations remains. Furthermore, sudden speed changes at the correction point lead to discontinuous predicted trajectories and abrupt changes in control commands, causing jerking and hesitation during vehicle execution. This is particularly problematic in low-speed scenarios such as parking, where smoothness cannot be guaranteed, impacting driving experience and safety. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for predicting vehicle operating status, which can convert the driving intention indicated by the gear signal into a speed and direction constraint and embed it into a multi-step time-series deduction process, so that the predicted speed is within the direction range allowed by the gear during the generation stage, thereby eliminating physical violations and motion jerks caused by post-processing cutoff from the root.

[0005] In a first aspect, the present invention provides a method for predicting vehicle operating status, wherein the method includes: acquiring operating status data of the vehicle under the current operating condition, and a gear signal corresponding to the operating status data; wherein the operating status data characterizes the vehicle operating status at multiple historical moments; determining the speed direction constraint corresponding to the operating status data based on the gear signal; and performing multi-step time-series extrapolation on the operating status data using a pre-constructed state prediction model based on the speed direction constraint to determine the predicted operating status result of the vehicle at the next moment.

[0006] In conjunction with the first aspect, the present invention provides a first implementation of the first aspect, wherein the step of determining the predicted operating state of the vehicle at the next moment by performing multi-step time-series extrapolation on the operating state data based on a pre-built state prediction model based on speed direction constraints includes: determining a reference value corresponding to the gear signal based on the operating state data; inputting the reference value into the state prediction model; and performing multi-step time-series extrapolation on the operating state data through the state prediction model to determine the predicted operating state of the vehicle at the next moment under the speed direction constraints of the gear signal.

[0007] In conjunction with the first aspect, this embodiment of the invention provides a second implementation of the first aspect, wherein the step of determining the reference value corresponding to the gear signal based on the operating status data includes: determining a normalized space based on the operating status data; obtaining the mapping value of absolute zero velocity in the normalized space; and using the mapping value as the reference value corresponding to the gear signal.

[0008] In conjunction with the first aspect, this embodiment of the invention provides a third implementation of the first aspect, wherein obtaining the mapping value of absolute zero velocity in the normalized space includes: obtaining vehicle control commands and actual vehicle driving parameters corresponding to the vehicle at multiple consecutive historical moments; mapping the vehicle control commands and actual vehicle driving parameters to a normalized interval based on the extreme values ​​of the vehicle control commands and actual vehicle driving parameters in the full time series, thereby obtaining normalized parameters corresponding to the vehicle control commands and actual vehicle driving parameters respectively; wherein the normalized parameters include the mapping value of absolute zero velocity in the normalized space.

[0009] In conjunction with the first aspect, this embodiment of the invention provides a fourth implementation of the first aspect, wherein the step of determining the speed direction constraint corresponding to the operating state data based on the gear signal includes: when the gear signal is a forward gear, the speed direction constraint is: the predicted speed is not lower than the reference value corresponding to the gear signal; when the gear signal is a reverse gear, the speed direction constraint is: the predicted speed is not higher than the reference value corresponding to the gear signal; when the gear signal is a parking gear or neutral gear, the speed direction constraint is: the predicted speed is a preset range value that floats near the reference value corresponding to the gear signal.

[0010] In conjunction with the first aspect, this embodiment of the invention provides a fifth implementation of the first aspect, wherein the state prediction model is trained by the following method: constructing a loss function, wherein the loss function includes a velocity direction constraint penalty term and a yaw rate constraint penalty term; wherein the velocity direction constraint penalty term is used to impose a directional constraint penalty on the predicted velocity output by the state prediction model corresponding to the gear signal; the yaw rate constraint penalty term is used to impose a constraint penalty on the predicted yaw rate output by the state prediction model based on the kinematic equations; and training the model parameters of the state prediction model based on the loss function to obtain the trained state prediction model.

[0011] In conjunction with the first aspect, this embodiment of the invention provides a sixth implementation of the first aspect, wherein the method further includes: determining the weight of the loss function corresponding to the yaw rate constraint penalty term based on the road conditions where the vehicle is located.

[0012] In conjunction with the first aspect, this invention provides a seventh implementation of the first aspect, wherein the kinematic equations include single-track kinematic equations, and the step of constraining and penalizing the predicted yaw rate output by the state prediction model based on the kinematic equations includes: obtaining the predicted longitudinal velocity, predicted steering wheel angle, and predicted yaw rate output by the state prediction model in the current step; determining the theoretical yaw rate corresponding to the predicted longitudinal velocity and predicted steering wheel angle based on the vehicle's wheelbase and steering ratio using the single-track kinematic equations; and constraining and penalizing the predicted yaw rate based on the deviation between the predicted yaw rate and the theoretical yaw rate.

[0013] In conjunction with the first aspect, the present invention provides an eighth implementation of the first aspect, wherein the kinematic equations further include dual-track kinematic equations; the method further includes: determining the lateral force of each wheel of the vehicle based on the tire lateral stiffness of the vehicle; and determining the theoretical yaw rate of the vehicle based on the lateral force using the dual-track kinematic equations.

[0014] Secondly, embodiments of the present invention also provide a vehicle operating state prediction device, the device comprising: a data acquisition module, configured to acquire operating state data of the vehicle under the current operating conditions, and a gear signal corresponding to the operating state data; wherein the operating state data characterizes the vehicle operating state at multiple historical moments; a data processing module, configured to determine the speed direction constraint corresponding to the operating state data based on the gear signal; and an execution module, configured to perform multi-step time-series deduction on the operating state data based on the speed direction constraint using a pre-built state prediction model, and determine the vehicle's operating state prediction result for the next moment.

[0015] The embodiments of the present invention bring the following beneficial effects: The embodiments of the present invention provide a vehicle operating state prediction method and device. By acquiring the operating state data of the current operating condition and the corresponding gear signal, the speed direction constraint is determined according to the current gear signal and embedded as the boundary condition for model deduction in a multi-step time-series prediction process. This ensures that the prediction result is always guided by the physical law of the gear during the generation process, ensuring the self-consistency of the speed direction and the continuity of the trajectory, and eliminating physical violations and motion jerks caused by post-processing cutoff from the root.

[0016] Other features and advantages of the invention will be set forth in the following description, or some features and advantages may be inferred from the description or determined without doubt, or may be learned by practicing the techniques described above.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart of a vehicle operating status prediction method provided in an embodiment of the present invention; Figure 2 A flowchart of another vehicle operating status prediction method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the joint loss calculation and training process of a state prediction model provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a normalization process provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a vehicle operating status prediction device provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0022] To facilitate understanding, a vehicle operating status prediction method provided in an embodiment of the present invention will first be described, referring to... Figure 1 The method includes the following steps: Step S102: Obtain the vehicle's operating status data under the current operating conditions, and the gear signal corresponding to the operating status data.

[0023] During vehicle operation, its longitudinal velocity, lateral velocity, yaw rate, and steering wheel angle, among other motion parameters, continuously change with driving operations and road conditions, forming high-dimensional time-series data reflecting the vehicle's real-time motion. To support trajectory planning and motion control decisions in autonomous driving systems, it is necessary to accurately predict the future evolution trend of this time series, enabling downstream modules to formulate reasonable control strategies based on the anticipated future state of the vehicle. Existing pure data-driven prediction methods typically focus only on the operational state data itself, directly inputting historical parameters such as speed and steering angle into neural networks for end-to-end time-series extrapolation. Essentially, this approach uses statistical data features to fit the evolution patterns of historical trajectories. However, the actual motion of a vehicle is not solely determined by historical inertia but is also constrained by the direction of the transmission system: in forward gear, the vehicle can only move forward, and in reverse gear, it can only move backward. This mechanical directional constraint is mandatory. Pure data-driven models cannot perceive this constraint, and in long-term time-series extrapolation, they are prone to producing predictions that contradict the gear position direction. To address this, existing technologies perform post-processing corrections by cascading state limiters at the model output. However, this correction process is non-differentiable and introduces abrupt velocity changes, failing to fundamentally resolve the issues of gradient breakage and lack of physical consistency.

[0024] Unlike the conventional methods described above, this embodiment of the invention acquires the gear position signal indicated by the vehicle at the corresponding moment in time while acquiring the operating status data. The operating status data characterizes the vehicle's operating state at multiple historical moments, including but not limited to vehicle dynamics parameters such as longitudinal velocity, yaw rate, longitudinal acceleration, lateral acceleration, and steering wheel angle over a time series. The gear position signal is the driving gear indicated by the vehicle when the aforementioned operating status data is generated under the current operating condition, including but not limited to drive, reverse, park, and neutral. This gear position signal can be the actual gear position at the current moment, or a shift command triggered by the driver or autonomous driving system during vehicle operation. The shift command characterizes the real-time intention to switch from the current gear to the target gear, enabling the gear position signal to reflect the transmission direction-limited state the vehicle will enter in the next moment, rather than being limited to the gear position at the current moment.

[0025] Operational status data reflects the actual movement trend of the vehicle at historical moments, while the gear position signal reflects the transmission direction constraint that the vehicle should follow at the corresponding moment or the next moment. Since the goal of the prediction task is to infer the future movement state of the vehicle, this embodiment of the invention, when performing time-series extrapolation on operational status data, also uses the driving intention indicated by the gear position signal as a conditional input to participate in the extrapolation process. For example, the gear position signal and historical status data jointly participate in the feature transformation of the intermediate layer, which can form a directional representation boundary corresponding to the current gear in the feature space. The predicted speed output by the model naturally falls within the allowable directional range defined by this boundary. In this regard, the constraint of the directional condition can be built into each extrapolation of the model, and the compliance of the prediction result is achieved in the generation stage, so the directional deviation will no longer accumulate and spread with the increase of the extrapolation step size.

[0026] Step S104: Based on the gear signal, determine the speed direction constraint corresponding to the running status data.

[0027] Different gear signals indicate different driving intentions and corresponding driving directions: forward gear corresponds to forward driving, reverse gear corresponds to backward driving, and park and neutral gear correspond to stationary parking. Different driving directions correspond to different speed direction constraint types. For example, when driving forward, the predicted speed should not be lower than the zero-speed reference; when driving backward, the predicted speed should not be higher than the zero-speed reference; and when stationary parking, the predicted speed should approach the zero-speed reference. In this embodiment of the invention, based on the acquired gear signal, the corresponding speed direction constraint type is determined according to the driving direction indicated by the gear signal, enabling the model to directly generate a result consistent with the gear direction within the directional range specified by the constraint type when generating the predicted speed.

[0028] Step S106: Based on the speed direction constraint, the pre-built state prediction model is used to perform multi-step time-series extrapolation on the operating state data to determine the predicted operating state of the vehicle at the next moment.

[0029] The state prediction model receives operating state data and uses the gear position signal as a conditional input. It performs multi-step time-series deductions within the directional range defined by speed direction constraints. In each deduction step, the model generates a predicted state based on the current input. The speed direction in this predicted state is governed by speed direction constraints, meaning the generated state lies within the directional range allowed by the current gear and does not exceed that range. After completing the current deduction step, the output of that step serves as the input for the next deduction step. This process is repeated until a preset number of deduction steps is reached, yielding the predicted operating state of the vehicle at the corresponding time.

[0030] In summary, this embodiment of the invention acquires the corresponding gear signal while obtaining the operating status data, and determines the speed direction constraint type based on the driving direction indicated by the gear signal. The determined constraint type can be applied to the speed direction in the form of a direction interval during the prediction process. When generating the predicted speed at each moment, the model can directly restrict it to the direction interval allowed by the current gear. Thus, the predicted speed already satisfies the gear direction constraint at each step of generation, eliminating the need for any correction steps after model output, and avoiding physical violations and motion stuttering caused by post-processing truncation. The gear signal can change at different times, and the constraint type switches instantly to the direction constraint of the corresponding gear with the real-time input of the gear signal. The same model can maintain a prediction result consistent with the current gear direction under different operating conditions such as forward, reverse, and parking. The predicted speed at each step is restricted to the direction interval allowed by the gear, and the direction deviation is governed by the constraint in the current step after it occurs, preventing it from proceeding to the next step. This effectively suppresses the accumulation and diffusion of direction deviation in long-term iterations, ensuring the continuity and physical consistency of the predicted trajectory.

[0031] Furthermore, based on the above embodiments, this invention also provides another method for predicting vehicle operating status, referring to... Figure 2 The method includes the following steps: Step S202: Obtain the vehicle's operating status data under the current operating conditions, and the gear signal corresponding to the operating status data.

[0032] Step S204: Based on the gear signal, determine the speed direction constraint corresponding to the running status data.

[0033] Step S206: Based on the operating status data, determine the reference value corresponding to the gear signal.

[0034] Based on the above embodiments, this invention introduces a reference value as a reference for determining the speed direction. This reference value can be predetermined from the speed data during vehicle operation and is used to distinguish the direction of speed in the prediction calculation. After receiving the operating status data and gear signal, the model can determine the corresponding reference value based on the currently input gear signal, and determine the directional constraints that the predicted speed should satisfy based on the reference value. Therefore, the predicted speed is restricted to the directional range defined by the reference value during generation, rather than being judged whether it exceeds the limit after generation. During the deduction process, each deduction step is based on the operating status data at the current moment, and the predicted speed is generated under the directional constraints corresponding to the gear signal. The deduction result satisfies the continuity of speed change and always stays within the directional boundary allowed by the gear, without exceeding the boundary. The operating status data carries the actual trend of vehicle movement, and the reference value carries the directional limitation indicated by the gear signal. Both participate in the generation process of each deduction step, and the predicted speed naturally follows the continuity of vehicle movement and the physical constraints of gear direction in its form. In one embodiment, the zero speed during vehicle operation can be used as the reference value. The predicted speed is on different sides of the reference value, corresponding to different driving directions. For example, the predicted speed should not be lower than the reference value when in forward gear, should not be higher than the reference value when in reverse gear, and should be close to the reference value when in park and neutral gear.

[0035] In one implementation, a normalization space can be determined based on operational status data. This normalization space maps physical data with different dimensions to a unified numerical range, enabling the model to perform calculations on the same scale. The mapping value of absolute zero velocity in the normalization space is obtained and used as a reference value for the gear signal. This reference value serves as a reference for determining the velocity direction and remains fixed throughout the simulation process, regardless of gear position or time. During actual vehicle operation, vehicle control commands and actual vehicle driving parameters change in real time with driving operations and road conditions, forming high-dimensional data reflecting the vehicle's dynamic response. Since different physical quantities have different dimensions and numerical ranges, in one implementation, vehicle control commands and actual vehicle driving parameters at multiple consecutive historical moments can be pre-collected. Based on the extreme values ​​of these data in the full time series, the vehicle control commands and actual vehicle driving parameters are mapped to a normalization interval to obtain corresponding normalized parameters. These normalized parameters include the mapping value of absolute zero velocity in the normalization space, which can be used as a reference value for determining the velocity direction in subsequent prediction processes. Using the above method, during the prediction phase, once the current operating status data is obtained, the corresponding position of the current operating status data in the normalization space can be determined according to the pre-determined normalization parameters, thereby determining its corresponding baseline value.

[0036] In this invention, multiple consecutive historical moments are used to characterize the observed motion state of the vehicle at different points in time, while the full time series is used to characterize the entire range of values ​​reached by each physical quantity throughout the entire observation period. Normalization based on the extreme values ​​of the full time series maps different physical quantities to a unified numerical range. Due to differences in the structure and performance of different vehicles, the distribution of extreme values ​​in their full time series differs, resulting in different normalization parameters. In predicting the operating state, this embodiment of the invention uses the specific mapping position of the absolute zero velocity in the corresponding vehicle's normalization space as a benchmark value. Different gears use this benchmark value as a boundary to determine their respective allowable speed direction ranges. The predicted speed is constrained by the current gear direction during the generation stage, eliminating the need for additional correction steps after model output. The benchmark value has different values ​​for different vehicles. By determining the benchmark value corresponding to the operating state data in the normalization space, the driving intention indicated by the gear signal can be associated with the vehicle's actual physical direction boundary, ensuring that the predicted speed direction output by the model in different gears always aligns with the vehicle's driving pattern.

[0037] Step S208: Input the baseline value into the state prediction model, and use the state prediction model to perform multi-step time-series deduction on the operating state data to determine the operating state prediction result of the vehicle at the next moment under the speed direction constraint of the gear signal.

[0038] When the model performs multi-step time-series simulations, the baseline value serves as a numerical reference for determining the velocity direction in the normalized space. This baseline value remains effective in each simulation step. When generating the predicted velocity for the current step, the model uses this baseline value as a boundary reference for the velocity direction, ensuring that the predicted velocity direction falls within the allowable direction range of the current gear. After completing the current simulation step, the predicted state output of that step serves as the input for the next simulation step. This process is repeated until a preset number of simulation steps is reached. At each moment in the resulting predicted state sequence, the velocity direction satisfies the constraints of the gear signal.

[0039] Furthermore, embodiments of the present invention also provide a training method for the state prediction model, enabling the model to generate predicted speeds under gear signal constraints. In one implementation, the model can be constructed through the following steps: 1) Construct the loss function.

[0040] To incorporate the driving intention indicated by the gear position signal into the model parameters during the training phase, this embodiment of the invention considers a conventional prediction error term, a velocity direction constraint penalty term, and a yaw rate constraint penalty term when constructing the loss function. These are used as the basic fitting term and the two types of physical constraint terms, respectively, to form a joint loss function. The three types of loss terms participate in gradient backpropagation in a differentiable manner, internalizing the gear position direction constraint and yaw rate constraint into the model parameters. During the inference phase, the model can directly generate a predicted speed that satisfies the direction constraint based on the input gear position signal.

[0041] The conventional prediction error term is used to measure the deviation between the predicted state output by the model and the actual state sequence, ensuring the model's basic fitting ability to the vehicle's motion state.

[0042] The velocity direction constraint penalty term is mapped to the value X in the normalized space at absolute zero velocity. zero Based on this, conditional constraints are applied to the predicted speed direction output by the model, incorporating the current gear information, so that the model establishes a correspondence between the gear signal and the speed direction during the learning process. In one implementation, let the mapping value of the true zero speed in the normalized space be... The prediction normalization speed for the current frame is .when (As with regular driving):

[0043] That is, if When ReLU takes effect, a penalty is imposed, preventing the vehicle from reversing.

[0044] when (e.g., reversing into a parking space):

[0045] That is, if When ReLU takes effect, a penalty is applied, preventing the vehicle from sliding forward.

[0046] when (e.g., parked and stationary):

[0047] Through the aforementioned conditional penalty branch, the network automatically corrects the direction error of speed prediction based on the gear intention of the real trajectory in the dataset during backpropagation training, so that the speed direction output by the model under different gears is consistent with the gear logic.

[0048] The yaw rate constraint penalty term imposes a constraint on the predicted yaw rate output by the model based on the kinematic equations. In one implementation, taking the single-track kinematic yaw rate penalty (loss_yaw) as an example, the theoretical yaw rate can be determined based on the vehicle's wheelbase and steering ratio, using the single-track kinematic equations, according to the predicted longitudinal velocity and predicted steering wheel angle output by the model in the current step. The deviation between this theoretical yaw rate and the predicted yaw rate output by the model is used as the constraint penalty.

[0049] The penalty item corresponding to this implementation method can be represented by the following steps: Predicted speed in the computational graph and steering wheel angle Gradient-based denormalization restores its true physical dimensions. and ).

[0050] Further utilize vehicle wheelbase and steering ratio Estimate the theoretical yaw rate:

[0051] It should be noted that under reversing conditions, the vehicle kinematic characteristics in the above formula exhibit polarity reversal. This model utilizes real-world signed values. Direct substitution makes The symbols are automatically matched with forward / reverse driving scenarios.

[0052] The estimated values ​​were further normalized to Finally, the calculation is as follows:

[0053] Correspondingly, the total loss function of the state prediction model can be expressed as:

[0054] In another implementation, the aforementioned kinematic equations can be replaced with dual-track kinematic equations that include tire lateral stiffness estimation to adapt to complex scenarios such as four-wheel independent drive or steering. These dual-track kinematic equations determine the lateral force of each wheel based on tire lateral stiffness and estimate the theoretical yaw rate based on the lateral force of each wheel. Specifically, on low-traction surfaces, the weight of the yaw rate constraint penalty term can be adaptively reduced, allowing the model to rely more on data-driven learning of nonlinear characteristics such as slippage and sideslip, avoiding the rigid constraint errors introduced by purely physical equations.

[0055] 2) Train the model parameters based on the above loss function until the loss function converges, and obtain the trained state prediction model.

[0056] Figure 3 This diagram illustrates the joint loss calculation and training process of the state prediction model provided in an embodiment of the present invention. Figure 3As shown, the state prediction model includes a state network, a numerical integrator, a conditional physical constraint module, and an output network. The model receives normalized control commands, the previous state, and coded gear information as inputs. The state network predicts the time derivative of the current input state quantity, which is then accumulated by a numerical integrator (such as an Euler integrator) to obtain the predicted state for the next time step. This process is repeated, and the model cyclically generates a multi-step predicted state sequence within a preset time window. The generated multi-step predicted state sequence is input to two branches: one branch outputs to the output network, which predicts the longitudinal and lateral accelerations; the other branch inputs to the conditional physical constraint module, which calculates the velocity direction constraint penalty (loss_vx) and yaw rate constraint penalty (loss_yaw). The conventional state prediction error (L1 Loss) and the physical penalty term output by the conditional physical constraint module together constitute the joint state loss. The error between the predicted acceleration value output by the output network and the actual acceleration constitutes the output prediction error (L1 Loss).

[0057] During the backpropagation phase, the joint state loss is backpropagated to the state network, and the output prediction error is backpropagated to the output network. Since the output of the state network participates in both the calculation of the physical penalty term and the calculation of the regular prediction error, the computational graph of the state network can be retained (retain_graph=True) in a single batch to support the complete backpropagation of gradients to the state network via both computational paths. The gradients of both the state network and the output network are used for joint parameter updates. The two types of networks perform collaborative optimization within the same training framework, thereby ensuring that the predicted state sequence output by the state network simultaneously satisfies the requirements of physical constraint fitting accuracy and kinematic consistency. In one implementation, the physical loss is also periodically evaluated in a gradient-free mode using a validation set during training to monitor the model's generalization performance and prevent overfitting.

[0058] Furthermore, Figure 4 A schematic diagram of the normalization process corresponding to an embodiment of the present invention is shown. This diagram illustrates the normalization process of the original data, as well as the inverse kinematics derivation to generate a physical penalty term. Figure 4As shown, raw multimodal data (including velocity, steering wheel angle, etc.) can be input into the dynamic extremum extraction module, which outputs the extrema (including minimum and maximum values) of each physical quantity in the full time series. Further, the adaptive normalization module can utilize these extrema to map the raw data to a preset normalization interval. During the training phase, the neural network prediction module can perform state prediction within the normalized space and output the normalized predicted longitudinal velocity and predicted steering wheel angle. Further, the gradient-based denormalization module can receive the predicted values ​​and saved scaling parameters to restore the normalized predicted longitudinal velocity and predicted steering wheel angle to their physical dimensions. This restored data can be input into the kinematic estimation module (illustrated using single-track kinematic equations in the figure), which estimates the theoretical yaw rate based on the vehicle's wheelbase and steering ratio. The renormalization module receives the scaling parameters and maps the theoretical yaw rate back to the normalized space. The absolute deviation calculation module can compare the normalized theoretical yaw rate with the network-predicted yaw rate to generate a physical penalty term.

[0059] Furthermore, in the actual in-vehicle inference phase, the predicted state output by the model already possesses physical compliance under the directional constraints internalized during the training phase. However, neural networks exhibit probabilistic biases in generalizing to extreme conditions. For example, training data often comes from dry road surfaces. When a vehicle enters a low-traction road surface covered by snow, the model trained based on conventional road data may still predict higher longitudinal acceleration or yaw rate, exceeding the physically reasonable range. Similarly, when the model performs long-term inferences, small deviations from previous steps are amplified step by step as the prediction step size increases, causing the predicted state to gradually deviate from the true physical boundary during iteration. In these rare cases, the model may output predicted values ​​exceeding the physical boundary. To address this, this embodiment of the invention sets a numerical safety boundary after each step of numerical integration, applying corresponding polarity constraints to the predicted speed based on the current gear signal. For example, in forward gear, the predicted speed below the reference value is corrected to the reference value, and in reverse gear, the predicted speed above the reference value is corrected to the reference value. This clamping mechanism can be used in conjunction with directional constraints internalized in the model parameters during the training phase. For example, physical constraints during training ensure directional compliance at the mechanism level, enabling the model to fundamentally understand and follow gear shift logic. Numerical clamping during inference ensures error elimination at the numerical level, guaranteeing that the calculation results do not deviate from the physical range due to numerical noise in rare cases. This numerical safety boundary is not triggered when the prediction results conform to the gear shift logic; it only intervenes in the event of accidental numerical errors. As an engineering-level safeguard beyond the physical mechanism, it can jointly guarantee the reliability and stability of the prediction results from different dimensions.

[0060] In summary, this invention, by introducing kinematic direction constraints based on gear position conditions, enables the same neural network to seamlessly adapt to complex scenarios such as driving and parking. Unlike existing technologies that only design unidirectional speed penalties for forward gears, this invention determines the corresponding speed direction constraint type based on the gear signal. Furthermore, this speed direction constraint is embedded in the loss function as a differentiable physical information constraint term, maintaining complete gradient backpropagation during the training phase. This ensures that the model weights inherently possess physical continuity and regularity during parameter updates, avoiding the gradient breakage problem caused by hard truncation at the output end in existing technologies and improving the smoothness of model convergence. Building upon this, this invention also introduces a dynamic extremum normalization mechanism, enabling real-time inverse scaling between physical units and network activation values ​​in closed-loop physical loss calculations. This allows the network to adaptively determine normalized parameters and zero-speed reference values ​​on different vehicle platforms (different wheelbases, steering ratios), demonstrating plug-and-play generalization capabilities.

[0061] Based on the above embodiments, this invention also provides a vehicle operating status prediction device, referring to... Figure 5 The device includes: a data acquisition module 10, used to acquire the vehicle's operating status data under the current operating conditions, and the gear signal corresponding to the operating status data; wherein, the operating status data represents the vehicle's operating status at multiple historical moments; a data processing module 20, used to determine the speed direction constraint corresponding to the operating status data based on the gear signal; and an execution module 30, used to perform multi-step time-series extrapolation on the operating status data based on the speed direction constraint using a pre-built state prediction model, and determine the predicted operating status result of the vehicle at the next moment.

[0062] The vehicle operating status prediction device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0063] This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described... Figures 1 to 2 The steps of any of the methods shown. Embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the above-described steps. Figures 1 to 2 The steps of any of the methods shown. Embodiments of the present invention also provide a structural schematic diagram of an electronic device, such as... Figure 6The diagram shows the structure of the electronic device, which includes a processor 101 and a memory 100. The memory 100 stores computer-executable instructions that can be executed by the processor 101. The processor 101 executes the computer-executable instructions to implement the above-mentioned... Figures 1 to 2 Any of the methods shown.

[0064] exist Figure 6 In the illustrated embodiment, the electronic device further includes a bus 102 and a communication interface 103, wherein the processor 101, the communication interface 103, and the memory 100 are connected via the bus 102. The memory 100 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk drive. Communication between this system network element and at least one other network element is achieved through at least one communication interface 103 (which can be wired or wireless), using the Internet, wide area network, local area network, metropolitan area network, etc. Bus 102 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, or an AMBA (Advanced Microcontroller Bus Architecture) bus. AMBA defines three types of buses: APB (Advanced Peripheral Bus), AHB (Advanced High-performance Bus), and AXI (Advanced eXtensible Interface). Bus 102 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6The diagram uses only a single double-headed arrow, but this does not imply a single bus or a single type of bus. Processor 101 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 101 or by instructions in software form. Processor 101 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor 101 reads information from the memory and, in conjunction with its hardware, completes the aforementioned tasks. Figures 1 to 2 Any of the methods shown.

[0065] The computer program product of the vehicle operating status prediction method and device provided in this embodiment of the invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the preceding method embodiments, which will not be repeated here. In addition, in the description of the embodiments of the invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "connection" should be interpreted broadly. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0066] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting vehicle operating status, characterized in that, The method includes: The vehicle's operating status data under the current operating conditions is acquired, as well as the gear signal corresponding to the operating status data; wherein, the operating status data represents the vehicle's operating status at multiple historical moments. Based on the gear signal, determine the speed direction constraint corresponding to the operating status data; Based on the speed direction constraint, a pre-built state prediction model is used to perform multi-step time-series deduction on the operating state data to determine the predicted operating state of the vehicle at the next moment. The step of determining the predicted operating state of the vehicle at the next moment by performing multi-step time-series extrapolation on the operating state data based on the velocity direction constraint using a pre-built state prediction model includes: Based on the operating status data, determine the reference value corresponding to the gear signal; The baseline value is input into the state prediction model, and the state prediction model performs multi-step time-series deduction on the operating state data to determine the operating state prediction result of the vehicle at the next moment under the speed direction constraint of the gear signal. The step of determining the reference value corresponding to the gear signal based on the operating status data includes: The normalization space is determined based on the aforementioned operational status data; Obtain the mapping value of absolute zero velocity in the normalized space; The mapped value is used as the reference value corresponding to the gear position signal.

2. The method according to claim 1, characterized in that, Obtaining the mapping value of absolute zero velocity in the normalized space includes: Obtain the vehicle control commands and actual vehicle driving parameters corresponding to the vehicle at multiple consecutive historical moments; Based on the extreme values ​​of the vehicle control commands and actual vehicle driving parameters in the full time series, the vehicle control commands and actual vehicle driving parameters are mapped to a normalized interval to obtain normalized parameters corresponding to the vehicle control commands and actual vehicle driving parameters respectively; wherein, the normalized parameters include the mapping value of absolute zero velocity in the normalized space.

3. The method according to claim 1, characterized in that, The step of determining the speed direction constraint corresponding to the operating status data based on the gear signal includes: When the gear signal is a forward gear, the speed direction constraint is: the predicted speed is not lower than the reference value corresponding to the gear signal; When the gear signal is reverse gear, the speed direction constraint is: the predicted speed is not higher than the reference value corresponding to the gear signal; When the gear position signal is parking or neutral, the speed direction constraint is: the predicted speed is a preset range value that floats around the reference value corresponding to the gear position signal.

4. The method according to claim 1, characterized in that, The state prediction model is trained using the following method: A loss function is constructed, comprising a velocity direction constraint penalty term and a yaw rate constraint penalty term; wherein the velocity direction constraint penalty term is used to impose a direction constraint penalty on the predicted velocity output by the state prediction model corresponding to the gear signal; and the yaw rate constraint penalty term is used to impose a constraint penalty on the predicted yaw rate output by the state prediction model based on the kinematic equations. The model parameters of the state prediction model are trained based on the loss function to obtain the trained state prediction model.

5. The method according to claim 4, characterized in that, The method further includes: Based on the road conditions where the vehicle is located, the weight of the loss function corresponding to the yaw rate constraint penalty term is determined.

6. The method according to claim 4, characterized in that, The kinematic equations include single-track kinematic equations. The step of constraining and penalizing the predicted yaw rate output by the state prediction model based on the kinematic equations includes: Obtain the predicted longitudinal velocity, predicted steering wheel angle, and predicted yaw rate output by the current step of the state prediction model; Based on the vehicle's wheelbase and steering ratio, the predicted longitudinal velocity and the theoretical yaw rate corresponding to the predicted steering wheel angle are determined using the single-track kinematic equations. The predicted yaw rate is constrained and penalized based on the deviation between the predicted yaw rate and the theoretical yaw rate.

7. The method according to claim 6, characterized in that, The kinematic equations also include dual-track kinematic equations; the method further includes: Based on the tire lateral stiffness of the vehicle, the lateral force of each wheel of the vehicle is determined; The theoretical yaw rate of the vehicle is determined based on the lateral force using the dual-track kinematic equations.

8. A vehicle operating status prediction device, characterized in that, The device includes: The data acquisition module is used to acquire the vehicle's operating status data under the current operating conditions, and the gear signal corresponding to the operating status data; wherein, the operating status data represents the vehicle's operating status at multiple historical moments. The data processing module is used to determine the speed direction constraint corresponding to the operating status data based on the gear signal. The execution module is used to perform multi-step time-series deduction on the operating state data based on the speed direction constraint and using a pre-built state prediction model to determine the predicted operating state of the vehicle at the next moment. The execution module is further configured to: determine a reference value corresponding to the gear signal based on the operating status data; input the reference value into the state prediction model; perform multi-step time-series deduction on the operating status data through the state prediction model; and determine the operating status prediction result of the vehicle at the next moment under the speed direction constraint of the gear signal. The step of determining the reference value corresponding to the gear signal based on the operating status data includes: determining a normalization space based on the operating status data; obtaining the mapping value of absolute zero velocity in the normalization space; and using the mapping value as the reference value corresponding to the gear signal.

Images