Method and system for generating dynamic economic speed interval based on vehicle physical state

Abstract

The application discloses a dynamic economic speed interval generation method and system based on a vehicle physical state, and relates to the technical field of energy-saving driving assistance. The method comprises the following steps: obtaining a current effective model version, generating branch output based on a reference parameter vector and an actual parameter vector, constructing a sensitivity matrix and determining identifiable parameters; forming and verifying model candidate updates according to directional energy consumption residuals and orthogonal complementary residuals; generating path prediction unit mileage energy consumption for candidate vehicle speeds according to six arrangement sequences of vehicle physical state parameters, combining one-sided calibration quantile values to form a one-sided calibration prediction upper bound curve, and determining a target economic speed and a dynamic economic speed interval according to a stable width cumulative index of a speed interval corresponding to a local minimum point. The application solves the problems that vehicle state changes and model drift are difficult to distinguish and economic speed intervals are unstable, and can improve the accuracy, robustness and continuity of energy consumption prediction and speed interval output.

Description

Technical Field

[0001] This invention relates to the field of energy-saving driving assistance technology, and in particular to a method and system for generating dynamic economic speed ranges based on vehicle physical conditions. Background Technology

[0002] In vehicle energy management and intelligent driving applications, it is often necessary to determine a more economical driving speed based on the vehicle's energy consumption characteristics for use in driving prompts, cruise control, or longitudinal speed planning. Existing methods for determining economical speeds are mostly based on the relationship between vehicle speed and energy consumption obtained from vehicle type approval tests, bench calibration, or specific road tests, and provide recommended speeds or economical speed ranges accordingly.

[0003] However, the load, tire condition, road conditions, and longitudinal wind field of a vehicle during actual operation will change with the usage process and driving environment, causing the vehicle's current energy consumption characteristics to differ from the pre-calibrated state. Existing methods usually use fixed calibration results or only make corrections for individual influencing factors, making it difficult to continuously and comprehensively reflect the combined impact of the vehicle's current physical state on energy consumption performance at different vehicle speeds.

[0004] Therefore, under constantly changing vehicle physical conditions, the low-energy speed positions obtained by existing methods are prone to deviating from the actual situation of the vehicle. The output economic speed or economic speed range may have inaccurate boundaries, fluctuate with the calculation cycle, or fail to be maintained stably, thus reducing its reference value for vehicle longitudinal speed planning and actual energy-saving operation. How to improve the accuracy and stability of the dynamic economic speed range generation results under changing vehicle conditions is a problem that needs to be solved in this field. Summary of the Invention

[0005] To address the problem of instability in the economic speed range under changes in vehicle condition, this invention proposes a method and system for generating dynamic economic speed ranges based on the vehicle's physical state.

[0006] The present invention achieves the above objectives through the following technical solutions:

[0007] A method for generating dynamic economic speed ranges based on vehicle physical states includes:

[0008] Obtain the current valid model version, including the road load model, the energy conversion model, and the conditional state transition models corresponding to the vehicle physical state parameters; the vehicle physical state parameters include the vehicle equivalent mass, rolling resistance coefficient, and longitudinal equivalent wind speed.

[0009] The reference branch output, actual branch output, single-parameter replacement branch output, and loop unit mileage energy consumption are generated based on the reference parameter vector and the actual parameter vector. A sensitivity matrix is ​​constructed based on the single-parameter replacement branch output and the reference branch output, and the identifiable parameters are determined by combining the state transition loop error.

[0010] The parameter candidate updates of the conditional state transition model are formed by using directional energy consumption residuals. After verification by an independent verification time window, they are retained or withdrawn, and the sensitivity matrix is ​​reconstructed based on the verified conditional state transition model. When all vehicle physical state parameters are identifiable parameters, slow-changing parameter candidate updates of the road load model and energy conversion model are formed by using orthogonal residual compensation. All verified and retained candidate updates are written into the new current effective model version.

[0011] For candidate vehicle speeds, the path prediction unit mileage energy consumption is generated according to six possible permutations of vehicle physical state parameters. The upper limit curve of the one-sided calibration prediction is generated by the sum of the maximum path prediction unit mileage energy consumption and the one-sided calibration quantile value.

[0012] Based on the cumulative stability width index of the maximum continuous speed interval corresponding to the local minimum point in the single-sided calibration prediction upper bound curve, the target economic speed and dynamic economic speed interval are determined.

[0013] As a preferred embodiment of the present invention, the conditional state transition model corresponding to any vehicle physical state parameter performs the following steps:

[0014] The system receives an input parameter vector, a target parameter value, a running context, and an input energy consumption per unit mileage. While keeping the other vehicle physical state parameters unchanged, it replaces the input parameter value of any of the vehicle physical state parameters with the target parameter value to obtain an output parameter vector.

[0015] Using the same road load model and energy conversion model, the physical predicted energy consumption per unit mileage of the input parameter vector and the output parameter vector under the operating context are calculated respectively. The physical predicted energy consumption difference is obtained by subtracting the physical predicted energy consumption per unit mileage of the input parameter vector from the physical predicted energy consumption per unit mileage of the output parameter vector.

[0016] Using the conditional state transition model, a state transition correction value is determined based on the input parameter vector, the output parameter vector, and the operating context. The sum of the input energy consumption per unit mileage, the difference between the physical predicted energy consumption and the state transition correction value is determined as the output energy consumption per unit mileage.

[0017] When exchanging the input parameter values ​​and the target parameter values, the same road load model parameters, energy conversion model parameters, and conditional state transition model parameters are used to perform the reverse state transition.

[0018] As a preferred embodiment of the present invention, both the reference parameter vector and the actual parameter vector are constructed according to a preset parameter order of vehicle equivalent mass, rolling resistance coefficient and longitudinal equivalent wind speed.

[0019] The reference branch output is generated by directly inputting the reference parameter vector into the road load model and the energy conversion model; the actual branch output is generated by sequentially converting the reference parameter vector according to the preset parameter order until the actual parameter vector is reached; any single parameter replacement branch output is generated by replacing the corresponding vehicle physical state parameter in the reference parameter vector with the actual parameter value in the actual parameter vector.

[0020] The reference branch output, the actual branch output, and the single-parameter replacement branch output all include an output vector and energy consumption per unit mileage, and use the same running context and the current effective model version;

[0021] Using the single-parameter replacement branch unit mileage energy consumption in the single-parameter replacement branch output as the input unit mileage energy consumption for the reverse state transition, the corresponding vehicle physical state parameters are reversed to the reference parameter value to obtain the loop unit mileage energy consumption; the absolute value of the difference between the loop unit mileage energy consumption and the reference branch unit mileage energy consumption is divided by the standard deviation of the unit mileage energy consumption measurement noise to obtain the state transition loop error of the corresponding vehicle physical state parameters.

[0022] As a preferred embodiment of the present invention, the wheel-end traction power sequence and the instantaneous energy input power sequence at least two sampling times within the time window are stacked into an output vector according to a preset output arrangement order;

[0023] When the absolute value of the difference between the actual parameter value and the reference parameter value of any vehicle physical state parameter is not lower than the parameter difference threshold, the difference between the output vector of the corresponding single parameter replacement branch output and the output of the reference branch is normalized according to the difference of the parameter value and the standard deviation of the measurement noise of each output channel to obtain the sensitivity column vector.

[0024] Independent components are obtained by projecting any sensitivity column vector onto the orthogonal complement of the subspace spanned by the other sensitivity column vectors. The generated sensitivity column vectors form a sensitivity matrix. Identifiable parameters are selected based on the minimum singular value and condition number of the sensitivity matrix, the norm of the independent components, and the state transition loop error.

[0025] The difference between the measured output vector and the actual branch output vector is normalized according to the preset output arrangement order and the standard deviation of the measurement noise corresponding to each output channel. The projection coefficient of the normalized difference in the direction of the independent component corresponding to the identifiable parameter is calculated. The projection coefficient is divided by the difference between the actual parameter value and the reference parameter value of the corresponding vehicle physical state parameter to obtain the dimensionless orientation coefficient. The dimensionless orientation coefficient is multiplied by the difference between the unit mileage energy consumption of the corresponding single parameter replacement branch and the unit mileage energy consumption of the reference branch to obtain the orientation energy consumption residual.

[0026] As a preferred embodiment of the present invention, the directional energy consumption residual is used to form the parameter candidate update of the corresponding conditional state transition model. After verification by an independent verification time window, the parameter candidate update is retained or withdrawn. Based on the verified conditional state transition model, the actual branch output and the single parameter replacement branch output are generated and the sensitivity matrix is ​​reconstructed.

[0027] When all vehicle physical state parameters are identifiable parameters, the difference between the measured output vector and the actual branch output vector is normalized and projected onto the orthogonal complement of the sensitivity matrix column space to obtain the orthogonal complement residual.

[0028] The normalized Jacobian matrix of the road load model is projected onto the orthogonal complement, and the rows corresponding to the wheel-end traction power sequence are extracted. When the identifiability condition is met, the same row components of the orthogonal complement residuals are used to form candidate updates of the slow-changing parameters of the road load model. The sensitivity matrix and the orthogonal complement residuals are reconstructed based on the verified road load model parameters.

[0029] The normalized Jacobian matrix of the energy conversion model is projected onto the orthogonal complement and the row corresponding to the instantaneous energy input power sequence is extracted. When the identifiability condition is met, the same row components of the orthogonal complement residual are used to form candidate updates of the slow-changing parameters of the energy conversion model.

[0030] Each candidate update is retained or withdrawn after being verified by an independent verification time window, and the next candidate update is formed only based on the verified model parameters; all retained candidate updates are written together into the new current effective model version; the rolling resistance model term of the road load model uses only the rolling resistance coefficient as the free parameter to be updated.

[0031] As a preferred embodiment of the present invention, historical samples that have not participated in the parameter update of the current effective model version are selected as calibration samples;

[0032] For each candidate vehicle speed and each calibration sample, starting from the reference parameter vector and the reference branch unit mileage energy consumption respectively, the conditional state transition model is called in sequence according to the six permutations of the vehicle physical state parameters to obtain the six path prediction unit mileage energy consumption from the reference parameter vector to the corresponding actual parameter vector.

[0033] The corresponding path difference vector is constructed by the normalized difference between the maximum path predicted energy consumption per unit mileage for each candidate vehicle speed and each calibration sample and the corresponding path predicted energy consumption per unit mileage. The corresponding path sorting sequence is constructed by the descending path number of the corresponding path predicted energy consumption per unit mileage.

[0034] For each candidate vehicle speed, the calibration samples are sorted according to the minimum number of adjacent exchanges between each calibration sample and the path sorting sequence corresponding to the candidate vehicle speed and the L2 distance between the path difference vectors, and a preset number of calibration samples are selected.

[0035] The positive part of the difference between the measured energy consumption per unit mileage of the selected calibration sample at the actual vehicle speed and the maximum path prediction energy consumption per unit mileage of the calibration sample at the actual vehicle speed is determined as the one-sided calibration residual. The one-sided calibration quantile is determined based on the one-sided calibration residual and added to the maximum path prediction energy consumption per unit mileage of the candidate vehicle speed to obtain the one-sided calibration prediction upper bound curve. When the current effective model version changes, the calibration sample is reselected and the one-sided calibration quantile is recalculated.

[0036] As a preferred embodiment of the present invention, a plurality of energy consumption increment thresholds are set for the single-sided calibration prediction upper bound curve in an incremental manner.

[0037] For each local minimum point, extract the maximum continuous speed range that includes the local minimum point and whose curve value is not higher than the sum of the curve value corresponding to the local minimum point and the corresponding energy consumption increment threshold under each energy consumption increment threshold.

[0038] In multiple consecutive calculation cycles of the current effective model version, the same local minimum point is matched according to the velocity difference between local minimum points and the crossover ratio of the corresponding maximum continuous velocity interval. For each energy consumption increment threshold, the minimum value of the width of the maximum continuous velocity interval corresponding to the same local minimum point in each calculation cycle is determined as the corresponding stable width.

[0039] The product of the difference between each adjacent energy consumption increment threshold and the stable width corresponding to the smaller energy consumption increment threshold among the adjacent energy consumption increment thresholds is accumulated to obtain the stable width cumulative index. From the local minimum points where the curve value is not higher than the sum of the global minimum curve value and the preset candidate energy consumption difference threshold, the speed corresponding to the local minimum point with the largest stable width cumulative index is determined as the target economic speed.

[0040] Extract the maximum continuous speed range that includes the target economic speed and whose curve value increment relative to the curve value corresponding to the target economic speed does not exceed the energy consumption tolerance. When the maximum continuous speed range meets the preset range stability condition in multiple consecutive calculation cycles, the maximum continuous speed range is determined as the dynamic economic speed range.

[0041] As a preferred embodiment of the present invention, the vehicle physical state parameters that have been generated as non-zero sensitivity column vectors but have not been screened as identifiable parameters are determined as parameters to be verified.

[0042] For each parameter to be verified, the time window for generating the corresponding sensitivity column vector is determined as the verification time window, and the vehicle physical state parameters other than the parameters to be verified are determined as matching parameters.

[0043] Project the sensitivity column vector corresponding to the parameter to be verified onto the orthogonal complement of the subspace spanned by the sensitivity column vector corresponding to the matching parameter. After obtaining non-zero fixed independent components, select the subsequent time window where the absolute value of the difference between the actual parameter value and the reference parameter value of the parameter to be verified does not exceed the return threshold, and the difference between the matching parameter value and the running context between the time window to be verified and the time window to be verified does not exceed the matching threshold as the pairing time window.

[0044] The difference between the measured output vector and the actual branch output vector of the time window to be verified and the paired time window is normalized according to the standard deviation of the measurement noise of each output channel. The verification difference is obtained by subtracting the normalized difference of the paired time window from the normalized difference of the time window to be verified.

[0045] When the ratio of the projection norm of the verification difference along the direction of the fixed independent component to the norm of the verification difference is not lower than the attribution threshold, and the projection norm on the orthogonal complement of the direction of the fixed independent component is not higher than the verification crosstalk threshold, the time window to be verified and the paired time window are determined as candidate update samples to form the parameter candidate update of the conditional state transition model corresponding to the parameter to be verified.

[0046] As a preferred embodiment of the present invention, the vehicle physical state parameters that are not screened as identifiable parameters constitute the set of parameters to be identified.

[0047] For the set of parameters to be identified, a speed trajectory is generated that sequentially includes a first constant speed segment, an acceleration segment, a second constant speed segment, and a deceleration segment, with the same start and end speeds; the speed changes in the acceleration and deceleration segments are opposites of each other, and the running context meets preset constraints during execution;

[0048] A prediction sensitivity sub-matrix is ​​constructed based on the difference in wheel-end traction power at the same speed point between the acceleration and deceleration phases, the difference in wheel-end traction power between the first and second constant speed phases, and the difference in instantaneous energy input power.

[0049] Candidate trajectories are defined as follows: the norm of the predicted sensitivity column vector corresponding to each parameter to be identified in the predicted sensitivity submatrix is ​​not lower than the trajectory identification threshold, and the ratio of the projection norm of each predicted sensitivity column vector in the subspace spanned by the sensitivity column vectors corresponding to the identifiable parameters to the norm of the corresponding predicted sensitivity column vector does not exceed the trajectory crosstalk threshold. When there are no identifiable parameters, the ratio is zero. The candidate trajectory with the largest minimum singular value in the corresponding predicted sensitivity submatrix is ​​selected.

[0050] When there are no candidate trajectories, the velocity trajectory identification parameter is not performed. After the selected candidate trajectory is executed, the sensitivity matrix is ​​reconstructed, and parameter candidate updates are formed only for the parameters to be identified that are selected as identifiable parameters and whose corresponding state transition loop error does not exceed the loop closure threshold.

[0051] The dynamic economic speed range generation system based on vehicle physical state, applied to the dynamic economic speed range generation method based on vehicle physical state as described above, includes:

[0052] The model version acquisition module is used to obtain the currently valid model version;

[0053] The branch generation and parameter identification module is used to generate reference branch output, actual branch output, single-parameter replacement branch output and loop unit mileage energy consumption based on reference parameter vector and actual parameter vector. It constructs a sensitivity matrix based on single-parameter replacement branch output and reference branch output, and determines identifiable parameters by combining state transition loop error.

[0054] The model update module is used to form candidate parameter updates for the conditional state transition model using directional energy consumption residuals. After verification by an independent verification time window, the updates are retained or withdrawn, and the sensitivity matrix is ​​reconstructed based on the verified conditional state transition model. When all vehicle physical state parameters are identifiable parameters, slow-changing parameter updates for the road load model and energy conversion model are formed using orthogonal residuals. All verified and retained candidate updates are written into the new current effective model version.

[0055] The upper bound prediction generation module is used to generate the path prediction unit mileage energy consumption for candidate vehicle speeds according to six permutations of vehicle physical state parameters, and generate the single-sided calibration prediction upper bound curve by the sum of the maximum path prediction unit mileage energy consumption and the single-sided calibration quantile value.

[0056] The speed range determination module is used to determine the target economic speed and dynamic economic speed range based on the cumulative stability width index of the maximum continuous speed range corresponding to the local minimum point in the single-sided calibration prediction upper bound curve.

[0057] The beneficial effects of this invention are: it can effectively distinguish between the energy consumption impact caused by changes in vehicle operating status and the prediction deviation caused by changes in the model itself, and independently verify the model adjustment results, reducing the possibility of parameter error attribution, mutual compensation, and invalid updates entering the subsequent prediction process. This improves the adaptability, prediction accuracy, and long-term operational stability of the energy consumption model under changing conditions such as load, rolling resistance, and longitudinal wind field. Furthermore, it can comprehensively consider prediction differences arising from different state change processes, historical prediction deviations, and the degree to which low-energy-consumption speed regions are maintained within continuous calculation periods. This reduces the impact of single prediction results, occasional low-energy-consumption points, and local curve fluctuations on the determination of economic speed, resulting in target economic speeds and dynamic economic speed ranges with better robustness, continuity, and practical energy-saving reference value. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the modular structure of the system according to an embodiment of the present invention. Detailed Implementation

[0059] 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0060] like Figure 1 The illustration shows an embodiment of the present invention, which provides a method for generating dynamic economic speed ranges based on vehicle physical states, using a pure electric passenger vehicle as an example. For gasoline vehicles, the instantaneous energy input power can be replaced with the fuel chemical energy input power calculated based on the fuel mass flow rate and the lower heating value of the fuel; for hybrid vehicles, the electrical energy input power and the fuel chemical energy input power can be obtained separately and converted into a unified energy quantity, with the subsequent processing remaining unchanged.

[0061] Vehicle operation data is acquired via CAN, CAN FD, or in-vehicle Ethernet, and includes at least vehicle speed, longitudinal acceleration, drive wheel torque or wheel-end traction power, instantaneous energy input power, road gradient, and sampling time. It may also include powertrain operating mode, motor speed, powertrain temperature, power battery state of charge, accessory load power, air density, and signal health status. All data are resampled using a unified timestamp. Optionally, the original sampling frequency is 20Hz, the basic time window length is 2.5s, and each basic time window includes N=50 sampling moments. For parameter identification requiring longer excitation, multiple consecutive basic time windows can be combined into an analysis time window, but all outputs in the same branch calculation and sensitivity calculation use the same sampling time and the same vector dimension. N represents the total number of sampling moments in a basic time window. Throughout the text, t represents the sampling moment index within a time window.

[0062] The wheel-end traction power and instantaneous energy input power are first filtered by median filtering to remove pulse anomalies, and then low-pass filtered with a cutoff frequency no higher than one-fifth of the sampling frequency. Before the time window enters the model update or calibration sample library, timestamp continuity, signal range, consistency of drive torque and longitudinal acceleration signs, consistency of power battery voltage and current, and continuity of power system operating mode are verified. Time windows that fail the verification are not included in the model update or calibration.

[0063] Vehicle physical state parameters are arranged in a fixed order to form a parameter vector. : ;in The equivalent mass of the vehicle is expressed in kilograms. The rolling resistance coefficient is dimensionless. The longitudinal equivalent wind speed is expressed in meters per second. This implementation method specifies... Indicates headwind. Indicates tailwind, the vehicle's speed relative to the air is .

[0064] Reference parameter vector Composed of reference parameter values ​​saved in the current effective model version; actual parameter vector It consists of the parameter estimates for the current time window; the target parameter vector is the current vehicle physical state parameter vector used when predicting the path for candidate vehicle speeds.

[0065] The operating context includes vehicle and environmental variables that are not considered as the above three vehicle physical state parameters. It includes at least vehicle speed sequence, longitudinal acceleration sequence, road gradient, air density, and powertrain operating mode, and may also include powertrain temperature, state of charge, and accessory load power.

[0066] The vehicle's equivalent mass can be obtained from axle load sensors, air suspension pressure sensors, or a longitudinal dynamics estimator with boundary constraints. The rolling drag coefficient can be initially obtained from a calibration table corresponding to tire pressure, tire temperature, and road surface type, and corrected for near-constant speed segments with limited gradient variation. The longitudinal equivalent wind speed can be obtained by fusing air velocity sensors with the vehicle's ground velocity, or by a bounded search based on meteorological wind field, vehicle heading, and road load residuals.

[0067] When there is no valid estimate for any actual parameter, the most recently verified parameter value is used for energy consumption prediction, but the parameter is marked as unidentifiable and its corresponding conditional state transition model is not updated; when the active identification execution conditions are met, the parameter identification speed trajectory can be generated according to Example 9.

[0068] The standard deviation of the measurement noise is estimated by sensor calibration, repeated operating condition statistics, or health history data, and a positive lower limit is set to avoid zero denominator in the normalization calculation.

[0069] This method mainly includes the following steps:

[0070] S101: After the vehicle starts, obtain the current valid model version. The current valid model version mainly includes the road load model, the energy conversion model, and the conditional state transition models corresponding to the vehicle's physical state parameters. The vehicle's physical state parameters include the vehicle's equivalent mass, rolling resistance coefficient, and longitudinal equivalent wind speed.

[0071] During the initial run, the factory-calibrated model is used as the initial valid model version. In subsequent runs, only model versions that have been independently verified and consistent with the write operation are used; candidate parameters that have not yet been verified are not included in the speed range calculation.

[0072] In one specific embodiment, the road load model mainly includes the following:

[0073] Vehicle equivalent mass Vehicle weighing mass Composition of the equivalent mass of rotating parts: ;in The fixed equivalent mass of rotating components such as wheels, drive shafts, and motor rotors is referred to in the longitudinal direction of the vehicle.

[0074] When the conditional state transition changes the vehicle's equivalent mass, according to The vehicle's weighing mass is changed synchronously, so that the load change acts simultaneously on the inertia term, the slope resistance term, and the rolling resistance term, while the equivalent mass of the rotating parts only acts on the inertia term.

[0075] Wheel end demand force Represented as: ;

[0076] in, It is longitudinal acceleration; The road slope angle; air density; It is the product of the air drag coefficient and the frontal area; For the vehicle in the The longitudinal velocity relative to the ground at each sampling time; This is a fixed, calibrated mechanical resistance constant. For speed-related mechanical resistance parameters; This is the acceleration due to gravity.

[0077] The road load model does not set a constant parasitic resistance parameter that can be freely updated, nor does it set other free parameters that are equivalent to the physical effect of the rolling resistance coefficient.

[0078] Wheel end traction power for: When the unit of force is Newton and the unit of speed is meter per second, the unit of power is kilowatt. Wheel-end traction power is a signed power; a positive value indicates driving, and a negative value indicates energy recovery.

[0079] The energy conversion model mainly includes the following parts:

[0080] For pure electric vehicles, the first Instantaneous energy input power at each sampling time for:

[0081] ;

[0082] in, For the first The driving efficiency at each sampling time under the corresponding state variable conditions. For the first Energy recovery efficiency at each sampling time under the corresponding state variable conditions Indicates the first A vector of state variables affecting energy conversion efficiency at each sampling time point includes at least motor speed, torque, power system temperature, and state of charge. For the first The power of the adjacent load at each sampling time.

[0083] Drive efficiency and recovery efficiency can be established using a lookup table model based on bench tests, and bilinear or multilinear interpolation can be employed. The efficiency values ​​are limited to a preset physical range; for example, the drive efficiency is... to The recycling efficiency is to .

[0084] Energy consumption per unit distance within a time window for: ;

[0085] in, The sampling period is the time interval between two adjacent sampling times. This is a preset minimum calculation distance. This implementation method can be adopted... The unit of energy consumption per unit distance is kilowatt-hours per hundred kilometers.

[0086] The preferred time window for online updating of the conditional state transition model and for updating the slowly changing parameters of the road load model and energy conversion model is when the wheel-end traction power is positive and the power system operating mode remains unchanged; the complete speed curve prediction can still use the above-mentioned drive and recovery segmentation relationship.

[0087] For any vehicle physical state parameter The conditional state transition model receives an input parameter vector. Target parameter values runtime context and input energy consumption per unit mileage With the other two vehicle physical state parameters remaining unchanged, the parameters will be... Replace with the target parameter values ​​to obtain the output parameter vector. : ;

[0088] Using the same road load model and energy conversion model, calculate the physical predicted energy consumption per unit mileage corresponding to the input parameter vector and output parameter vector in the current operating context. , : ; Physical prediction of energy consumption difference for: ;

[0089] Conditional state transition correction value for: Its unit is the same as the energy consumption per unit mile, used to characterize the limited correction of the change in energy consumption per unit mile by the efficiency mapping nonlinearity, auxiliary power state, and parameter interactions that are not fully represented by the fundamental physical model. In the formula, This represents the correction function for the j-th conditional state transition; Let be the parameter vector of the j-th conditional state transition model.

[0090] Output energy consumption per unit mileage after state transition for: The units for all terms on both sides of the equation are kilowatt-hours per hundred kilometers. The input energy consumption per unit mileage before the state transition.

[0091] In one optional embodiment, the conditional state transition correction model is a linear basis function model: ; This is the basis function column vector for the j-th conditional state transition model, with the superscript T indicating transpose. The three vehicle physical state parameters are set as independent parameter vectors, not sharing model parameters. Input variables are normalized using the mean and standard deviation of the training samples. Parameter variation... for: ; This is the parameter value of the j-th vehicle physical state parameter in the output parameter vector; The parameter value of the j-th vehicle physical state parameter in the input parameter vector; Let be the normalized scale for the j-th vehicle physical state parameter; where the parameter scales for the vehicle equivalent mass, rolling resistance coefficient, and longitudinal equivalent wind speed can be taken as follows: , and .

[0092] basis function vectors Set as: ;

[0093] in, To standardize the average vehicle speed, To standardize the root mean square value of longitudinal acceleration, To standardize the average road slope, To standardize the temperature of the power system, To standardize the load power of accessories. , These represent the indices of the two vehicle physical state parameters other than the j-th vehicle physical state parameter. , These represent the normalized parameter values ​​of the corresponding vehicle physical state parameters that remain unchanged during this state transition, obtained by dividing the corresponding physical parameter value by the aforementioned normalization scale. Each runtime context physical quantity is subtracted from the mean of the corresponding training samples and divided by the corresponding positive standard deviation. The mean and standard deviation used are saved with the current effective model version.

[0094] thus, From time to time When performing a reverse state transition under the same conditions of all other parameters and runtime context, it is similar to... The terms of the basis functions that are proportional to each other have reversed signs, and the quadratic terms... The state transition closure error remains unchanged; therefore, it is used to measure whether the non-reciprocal component exceeds the allowable range, avoiding the closure error from being structurally zero. Since the values ​​of the other two vehicle physical state parameters depend on the order of the previously completed transitions during each transition, different arrangements can still result in different cumulative correction values.

[0095] Offline training samples are generated using controlled paired tests. Vehicle equivalent mass samples can be generated by adding known counterweights; rolling resistance coefficient samples can be generated by adjusting tire pressure and temperature, or by replacing tires calibrated for rolling resistance; longitudinal equivalent wind speed samples can be generated in a wind tunnel, a chassis dynamometer with a fan, or a test track equipped with an air velocity measurement device. Each paired test maintains the speed trajectory, road gradient, powertrain mode, temperature, state of charge, and accessory load within a preset difference range, ensuring that the paired samples primarily alter one vehicle physical state parameter.

[0096] Actual energy consumption per unit distance Obtained by integrating the terminal voltage and current of the power battery:

[0097] ;

[0098] in, This represents the terminal voltage of the power battery at the t-th sampling time, in volts. This represents the battery terminal current at the t-th sampling time, in amperes, with discharge being positive and recharge recovery being negative.

[0099] Training objective of the j-th conditional state transition model for: .

[0100] The complete test sequence is as follows and The dataset is divided into a training set and a validation set, and the time windows of the same experimental sequence are not simultaneously included in the training set and the validation set.

[0101] by Indicates the total number of offline training samples, in This represents the index of the offline training samples. The offline parameters are determined using least squares with a L2 regularization term. ;in, Indicates the first The basis function column vectors corresponding to each offline training sample This represents the non-negative regularization coefficient, which can be taken in this implementation. .

[0102] Online updates employ recursive least squares. The initial covariance matrix of the j-th conditional state transition model. for: ;in, This represents the dimension of the basis function column vector. express An identity matrix of order 1.

[0103] Forgetting factor satisfy In this embodiment, it is possible to take .by This indicates online updating of the sample index and recursive gain. Recorded as: ;in, Indicates processing the first The covariance matrix of the samples before online updates. Indicates the first The basis function column vectors corresponding to each online updated sample;

[0104] After each iteration, the covariance matrix is ​​updated according to the following formula. : ;

[0105] In the online phase, the directional energy consumption residual is used as the remaining error of the current model to form parameter increment candidates:

[0106] ;

[0107] in, Indicates according to the first The j-th candidate value of the conditional state transition model parameters is generated from the online updated samples; Indicates processing the first The conditional state transition model parameter vector before each online update sample; Indicates the first The directional energy consumption residual corresponding to the j-th vehicle physical state parameter of each online updated sample.

[0108] This update only corrects the current model parameter increments and does not reinterpret the directional energy consumption residuals as complete state transition correction values. The single change of each parameter component does not exceed both the corresponding preset absolute upper limit and relative upper limit; the relative upper limit can be taken as 5% of the larger of the absolute value of the current parameter component and the corresponding preset positive parameter scale. Both the absolute upper limit and the positive parameter scale are determined by offline training and saved with the current effective model version.

[0109] When performing a reverse state transition, the input parameter values ​​and the target parameter values ​​are swapped, and the same road load model parameters, energy conversion model parameters, and conditional state transition model parameters are used.

[0110] S102: In each calculation cycle, the reference branch output, actual branch output, single-parameter replacement branch output, and loop unit mileage energy consumption are generated based on the reference parameter vector and the actual parameter vector. A sensitivity matrix is ​​constructed based on the single-parameter replacement branch output and the reference branch output, and the identifiable parameters are determined by combining the state transition loop error.

[0111] In one embodiment, the reference parameter vector and the actual parameter vector are configured in the order of vehicle equivalent mass, rolling drag coefficient, and longitudinal equivalent wind speed. For example: ; ;

[0112] The reference branch is generated by directly inputting the reference parameter vector into the road load model and the energy conversion model: ;in, Indicates the output of the reference branch. As the reference branch output vector, Energy consumption per mileage for reference branch units.

[0113] Taking the vehicle's equivalent mass as an example, the vehicle's equivalent mass in the reference parameter vector is replaced with the actual vehicle's equivalent mass to obtain a single-parameter replacement vector. Using the energy consumption per unit mileage of the reference branch as the input energy consumption per unit mileage, the conditional state transition model corresponding to the vehicle's equivalent mass is called to obtain the energy consumption per unit mileage of the single-parameter replacement branch corresponding to the vehicle's equivalent mass. Simultaneously, the single-parameter replacement vector is input into the road load model and the energy conversion model to obtain the single-parameter replacement branch output vector corresponding to the vehicle's equivalent mass.

[0114] The single-parameter replacement branches for rolling drag coefficient and longitudinal equivalent wind speed are generated in the same manner. For any vehicle physical state parameter... The output vector of the corresponding single-parameter replacement branch is uniformly denoted as The energy consumption per unit mileage of the corresponding single parameter replacement branch is uniformly recorded as... .

[0115] The actual branch starts with the reference parameter vector and converts the vehicle's equivalent mass, rolling resistance coefficient, and longitudinal equivalent wind speed into actual parameter values ​​in a preset order. Each step uses the output parameter vector and output energy consumption per unit mile of the previous step as input for the next step. After all conversions are completed, the actual branch output vector is generated based on the final actual parameter vector, and the actual branch energy consumption per unit mile is obtained. ;in This indicates the actual branch output. This represents the actual branch output vector. This indicates the actual energy consumption per unit mileage of the branch.

[0116] The reference branch, the actual branch, and the three single-parameter replacement branches all use the same runtime context and the same currently valid model version. The output vector of each branch is the instantaneous power sequence generated by the road load model and the energy conversion model, and the energy consumption per unit mileage is a scalar after conditional state transition correction; the two are used for parameter identifiability analysis and path energy consumption prediction, respectively, and the state transition correction value is not repeatedly written into the instantaneous power sequence.

[0117] For the j-th vehicle physical state parameter, replace the branch unit mileage energy consumption with the corresponding single parameter. As the input unit mileage energy consumption for the reverse state transition, the j-th vehicle physical state parameter is restored from its actual value to its reference value, thus obtaining the loopback unit mileage energy consumption. State transition closure error for: ;in, This is the standard deviation of noise for energy consumption per unit mileage, and it is set to a positive number.

[0118] For example, when the reference branch energy consumption per unit mileage corresponding to the vehicle's equivalent mass is 15.20 kWh / 100 km, the loop energy consumption per unit mileage is 15.24 kWh / 100 km, and the standard deviation of the energy consumption measurement noise per unit mileage is 0.10 kWh / 100 km, the loop error of the state transition corresponding to the vehicle's equivalent mass is 0.4.

[0119] The reverse state transition does not generate a separate power output vector for loopback error calculation, thus avoiding the use of the recovered reference parameter vector to regenerate a physical output sequence that is necessarily close to the reference branch.

[0120] S103: For identifiable parameters, candidate updates for parameters of the conditional state transition model are formed using directional energy consumption residuals. After verification by an independent verification time window, these updates are retained or withdrawn. Based on the verified conditional state transition model, the actual branch output and single-parameter replacement branch output are regenerated, and the sensitivity matrix is ​​reconstructed. When all vehicle physical state parameters are identifiable parameters, candidate updates for slowly changing parameters of the road load model and energy conversion model are formed using orthogonal compensation residuals corresponding to the sensitivity matrix. All verified and retained candidate updates are written into the new current effective model version.

[0121] Specifically, the output vector within the time window It is constructed in the order of first wheel-end traction power sequence, followed by instantaneous energy input power sequence: ;in, Represents a 2N-dimensional real vector space; when At this time, the output vector is 100-dimensional. Energy consumption per unit mileage is not included in the output vector to avoid repeatedly using the instantaneous energy input power sequence and its integral in the sensitivity matrix.

[0122] The standard deviation of the output channel measurement noise forms a diagonal matrix. : ;in, This represents the standard deviation of the measurement noise corresponding to the t-th wheel-end traction power output channel. This represents the standard deviation of the measurement noise corresponding to the t-th instantaneous energy input power output channel. One feasible value is: the standard deviation of the wheel-end traction power measurement noise is... The standard deviation of the instantaneous energy input power measurement noise is .

[0123] The physical parameter difference between the actual parameter value and the reference parameter value of the j-th vehicle physical state parameter Recorded as: ;in, This represents the physical parameter value of the j-th vehicle physical state parameter in the actual parameter vector; This represents the physical parameter value of the j-th vehicle physical state parameter in the reference parameter vector.

[0124] When the absolute value of the physical parameter difference is not lower than the corresponding physical parameter difference threshold, a sensitivity column vector is generated based on the difference between the output vector of the corresponding single-parameter replacement branch and the reference branch output vector, the physical parameter difference, and the standard deviation of the measurement noise of each output channel. :

[0125] ;

[0126] When using this basic implementation method, the minimum singular value threshold, condition number threshold, and independent component threshold of the sensitivity matrix (described later) are all calibrated under the fixed metrology system and saved with the current valid model version, without being directly used across different metrology systems. The parameter incremental projection coefficients described later have the same dimensions as the corresponding vehicle physical state parameters, and the parameter incremental projection coefficients are divided by the corresponding... The dimensionless orientation coefficient is obtained. The relationships for the remaining independent components, sensitivity matrix, identifiable parameter selection, and orientation energy consumption residual calculation are the same as those in the preferred numerical implementation method described below.

[0127] As a preferred numerical implementation method, to make the sensitivity columns corresponding to the three vehicle physical state parameters comparable, a fixed-scale normalization is used within the sensitivity calculation: ; ; ;in, , , The coordinates represent the dimensionless parameters of the vehicle's equivalent mass, rolling resistance coefficient, and longitudinal equivalent wind speed, respectively.

[0128] In the following preferred numerical implementations, the parameter difference refers to the difference between the actual parameter value and the reference parameter value in the normalized parameter coordinates. The dimensionless parameter difference corresponding to the j-th vehicle physical state parameter is uniformly denoted as:

[0129] ;

[0130] in, This represents the dimensionless parameter difference between the actual parameter value and the reference parameter value of the j-th vehicle physical state parameter. This represents the dimensionless parameter coordinates corresponding to the actual parameter values; This represents the dimensionless parameter coordinates corresponding to the reference parameter value.

[0131] The threshold values ​​for the parameter differences between the vehicle's equivalent mass, rolling resistance coefficient, and longitudinal equivalent wind speed can be taken as follows: , and The dimensionless parameter difference threshold of the j-th vehicle physical state parameter is denoted as . Only when At that time, a corresponding sensitivity column vector is generated. ,For example: .

[0132] Since the output difference has been normalized to the standard deviation of the measurement noise, and the parameter difference uses dimensionless coordinates, the sensitivity column vector is also dimensionless. When the parameter difference is below a threshold, the sensitivity column vector for that parameter is not generated, and that parameter is determined to be an unidentifiable parameter for the current time window.

[0133] parameters The matrix is ​​composed of the previously generated sensitivity column vectors. Independent components for: ,in, Representation matrix Moore-Penrose generalized inverse, This represents a 2N-order identity matrix. When no other generated sensitivity column vectors exist, the following is defined: ;

[0134] For each non-empty combination of the generated sensitivity column vectors, form a sensitivity submatrix, and denote the parameter index set as... The corresponding sensitivity submatrix is A qualified combination simultaneously satisfies:

[0135] ;

[0136] ;

[0137] ;

[0138] ;

[0139] In the formula, Represents the smallest singular value of a matrix. This indicates the preset minimum singular value threshold; The condition number represents the matrix. This indicates the threshold number of preset conditions; Indicates the preset independent component threshold; This represents the preset state transition lap time error threshold. A set of implementable thresholds is as follows: .

[0140] Choose the combination with the most parameters from the qualified combinations; if the number of parameters is the same, choose the combination with the largest minimum singular value. The parameters in the selected combination are identifiable parameters.

[0141] The difference between the measured output vector and the actual branch output vector is normalized according to the same output arrangement and measurement noise standard deviation. The measured output vector is composed of the measured wheel-end traction power sequence and the measured instantaneous energy input power sequence within the same time window, arranged in the aforementioned order. ; This represents the normalized output residual corresponding to the current time window. This represents the measured output vector corresponding to the current time window.

[0142] For identifiable parameters Calculate the parameter increment projection coefficients of the normalized output residuals in the directions of independent components. : In the preferred numerical implementation, the dimensionless orientation coefficient for: ;

[0143] The difference between the energy consumption per unit mileage of the single-parameter replacement branch and the energy consumption per unit mileage of the reference branch. for: ;

[0144] Directional energy consumption residual for: ; Dimensionless The unit is kilowatt-hours per hundred kilometers, therefore the unit of directional energy consumption residual is also kilowatt-hours per hundred kilometers.

[0145] When the norm of independent components is below the independence threshold, the projection coefficients are not calculated; when... When the parameter difference is below the threshold, the dimensionless orientation coefficient is not calculated.

[0146] For each identifiable parameter, candidate parameters for the corresponding conditional state transition model are updated according to a preset order of vehicle equivalent mass, rolling resistance coefficient, and longitudinal equivalent wind speed.

[0147] Each candidate update is written to the candidate parameter area and does not directly modify the current valid model version. Independent validation time windows must not participate in the formation of this candidate update and should use the same dynamic system operating mode as the candidate update time window, and pass data health checks.

[0148] Candidate parameters and parent model parameters are applied to independent validation time windows, and the mean of the absolute prediction error per unit mileage and the maximum value of the state transition loopback error are calculated separately. A candidate parameter update is retained only if the mean of the absolute prediction error corresponding to the candidate parameter is at least lower than a preset validation improvement threshold relative to the parent model parameter, and the maximum value of the loopback error is not higher than the sum of the corresponding value of the parent model parameter and the preset loopback tolerance; otherwise, it is revoked. The validation improvement threshold and loopback tolerance are both preset non-negative values ​​and are saved with each model version.

[0149] The candidate update of the next conditional state transition model is formed only based on the model state after the previous candidate update has been verified.

[0150] After the three conditional state transition models are processed, the actual branch output and single-parameter replacement branch output are regenerated based on the validated conditional state transition models, and the sensitivity matrix is ​​reconstructed.

[0151] Since the sensitivity matrix is ​​composed of the output vectors generated by the road load model and the energy conversion model, and the state transition correction value of the conditional state transition model only applies to the energy consumption per unit mileage and is not repeatedly written into the output vector, the reconstructed sensitivity matrix is ​​allowed to have the same value as the sensitivity matrix before the conditional state transition model update, provided that the parameters of the road load model and the energy conversion model remain unchanged. In this case, the actual branch output and the single-parameter replacement branch output are still regenerated, and the sensitivity matrix is ​​reconstructed; the reconstruction step is not replaced by "matrix values ​​remaining unchanged".

[0152] When the parameters of the road load model or energy conversion model undergo validated changes, the branch outputs are regenerated and the sensitivity matrix is ​​reconstructed based on the changed model parameters. When all three vehicle physical state parameters are determined to be identifiable parameters, the sensitivity matrix... for: ;

[0153] Orthogonal complement projection matrix of the sensitivity matrix column space for: ; Represents the sensitivity matrix Moore-Penrose generalized inverse;

[0154] Orthogonal residual compensation for: The residual represents the portion of the normalized output error that cannot be explained by the three vehicle physical state parameters.

[0155] Define the wheel-end traction power row selection matrix and the instantaneous energy input power row selection matrix:

[0156] ;

[0157] ;

[0158] in, This represents the row selection matrix for extracting the corresponding rows of the wheel-end traction power sequence from the 2N-dimensional output vector; This represents the row selection matrix for extracting the rows corresponding to the instantaneous energy input power sequence from the 2N-dimensional output vector; Represents an N-order identity matrix; This represents an N-order zero matrix.

[0159] Slowly changing parameter vector of road load model Set as: ;in, This is a slowly varying correction amount corresponding to the product of the drag coefficient and the frontal area; the allowable range can be set to... to ; This is the slow-change correction amount corresponding to the speed-related mechanical resistance parameter, and the allowable range can be set to... to .

[0160] Set the parameter scaling matrix: , This represents the scaling matrix of slowly varying parameters in the road load model; The positive parameter scale representing the correction amount corresponding to the product of air drag coefficient and windward area can be taken as 0.15 m² in this embodiment; The positive parameter scale corresponding to the correction amount of speed-related mechanical resistance parameters can be taken as 2 N·s / m in this embodiment.

[0161] Fixed mechanical resistance constant Not updated online. The rolling resistance term uses only the rolling resistance coefficient from the vehicle's physical state parameters. In the road load model, no other free parameters equivalent to the rolling resistance coefficient are set.

[0162] Jacobian matrix of road load model output vector with respect to slowly varying parameters Recorded as ;

[0163] The local derivative of the correction factor for the product of the air resistance coefficient and the frontal area at the t-th sampling time is:

[0164] ;

[0165] The local derivative of the correction amount of the wheel-end traction power to the speed-related mechanical resistance parameter at the t-th sampling time is:

[0166] .

[0167] The Jacobian line corresponding to the instantaneous energy input power is obtained by using the chain rule based on the local derivative of the wheel-end traction power in the energy conversion model.

[0168] After normalization and orthogonal projection, extract the rows corresponding to the wheel-end traction power sequence: ;

[0169] Corresponding orthogonal complemented residual row components for: ;

[0170] This represents the row Jacobian matrix of the wheel-end traction power of the road load model extracted after output noise normalization, orthogonal complement projection, and parameter scaling.

[0171] when and At that time, candidate updates for slowly changing parameters of the road load model are generated: ; This represents the dimensionless, slowly changing parameter increment of the road load model. This represents the dimensionless nonnegative regularization coefficient corresponding to the slow-changing parameter update of the road load model. In this implementation, it can be taken as... ; This represents the second-order identity matrix. It corresponds to the increment of physical parameters. for: .

[0172] Parameter scaling matrix It is also used to unify the columns of the Jacobian matrix and the parameter increments to dimensionless coordinates, ensuring that the minimum singular value, condition number, and regularization term do not change with the choice of physical units. Candidate values ​​must not exceed the upper and lower limits of the physical parameters, and the amount of change in a single instance must not exceed the preset maximum amount of change.

[0173] During the independent validation time window, the orthogonal complement projection matrix formed before the candidate update is used as the fixed projection reference. The road load model candidate update is retained only if the norm of the orthogonal complement residual decreases after the candidate update and the norm of the column space projection residual of the sensitivity matrix does not increase; otherwise, it is revoked.

[0174] After the candidate update of the road load model is verified, the output of each branch, the sensitivity matrix, the orthogonal complementary projection matrix and the orthogonal complementary residual are regenerated based on the road load model state after retention or cancellation.

[0175] Slow-changing parameters of energy conversion model Set as: ;in, , and These are efficiency correction factors for the low-power, medium-power, and high-power regions, with an allowable range of 0.90 to 1.10 for each. For accessory power bias that is not individually metered, the permissible range is -0.5 to 0.5 kW.

[0176] Setting the scaling matrix for slowly changing parameters of the energy conversion model : ; , , The positive parameter scales representing the efficiency correction coefficients in the low-power, medium-power, and high-power regions are respectively taken as 0.10 in this embodiment; The positive parameter scale corresponding to the power bias of the accessory can be taken as 0.5kW in this embodiment.

[0177] The low-power, medium-power, and high-power ranges can be set to 0% to 20%, 20% to 60%, and above 60% of the rated wheel end power, respectively.

[0178] In driving mode, first determine the correction efficiency after limiting: Then calculate the corrected instantaneous energy input power. ; This represents the corrected driving efficiency after amplitude limiting at the t-th sampling time. This represents the driving efficiency corresponding to the basic efficiency table at the t-th sampling time; b∈{L,M,H} represents the low-power, medium-power, or high-power region to which the current sampling time belongs; This represents the efficiency correction factor for the corresponding power region; Indicates the preset lower limit of efficiency; This indicates a preset efficiency upper limit, and satisfies... .

[0179] For sampling times that are not triggered by efficiency limiting and are located in power region b: ,and .

[0180] This leads to the normalized Jacobian matrix of the energy conversion model. And calculate: , ; This represents the Jacobian matrix of the actual branch output vector with respect to the slowly varying parameter vector of the energy conversion model. This represents the instantaneous energy input power row Jacobian matrix of the energy conversion model extracted after output noise normalization, orthogonal complement projection, and parameter scaling. This represents the row component corresponding to the instantaneous energy input power sequence in the orthogonal complementary residual.

[0181] when and At that time, candidate updates for slowly changing parameters of the energy conversion model are generated: ; This represents the dimensionless, slowly changing parameter increment of the energy conversion model; This represents the dimensionless nonnegative regularization coefficient corresponding to the slow-changing parameter update of the energy conversion model. In this implementation, it can be taken as... ; Represents a 4th-order identity matrix; corresponding to the physical parameter increment. for: .

[0182] Parameter scaling matrix normalization effect and The energy conversion model candidate updates use the same physical boundaries, single change limits, and independent verification rules as the road load model.

[0183] Each candidate update is based on the parent model state at the time it was formed, and the next candidate update is formed only based on the model state after the previous candidate update has been validated. After all candidate updates are completed, the validated and retained parameters in each model are written into the new model version as a consistent transaction. The model version includes at least the model version number, parent version number, road load model parameters, energy conversion model parameters, three conditional state transition model parameters, reference parameter vector, input normalization parameters, measurement noise standard deviation, training time window identifier, validation time window identifier, validation index, generation time, and integrity check code.

[0184] Only after a candidate version is fully written and passes verification will the current valid model version be switched at the boundary of the computation cycle. If no candidate update is retained, the original valid model version will continue to be used.

[0185] S104: For candidate vehicle speeds, generate path prediction energy consumption per unit mileage according to six possible arrangements of vehicle physical state parameters, and generate a single-sided calibration prediction upper bound curve by summing the maximum path prediction energy consumption per unit mileage with the single-sided calibration quantile value.

[0186] In one embodiment, calibration samples are selected from historical samples that have never participated in parameter updates for the current valid model version. Training samples and calibration samples are kept mutually exclusive through sample identifiers. Each calibration sample includes at least the measured energy consumption per unit mileage, actual vehicle speed, actual parameter vector, operating context, data quality flags, and model version identifier.

[0187] There are six possible arrangements of the three vehicle physical state parameters:

[0188] Vehicle equivalent mass — rolling resistance coefficient — longitudinal equivalent wind speed;

[0189] Vehicle equivalent mass — longitudinal equivalent wind speed — rolling resistance coefficient;

[0190] Rolling drag coefficient — vehicle equivalent mass — longitudinal equivalent wind speed;

[0191] Rolling resistance coefficient — longitudinal equivalent wind speed — vehicle equivalent mass;

[0192] Longitudinal equivalent wind speed — vehicle equivalent mass — rolling resistance coefficient;

[0193] Longitudinal equivalent wind speed — rolling resistance coefficient — vehicle equivalent mass.

[0194] For candidate vehicle speeds, the vehicle speed sequence in the running context vector is set as the predicted vehicle speed sequence corresponding to the candidate vehicle speed, and the remaining components adopt the current road, environment and power system state; when the candidate vehicle speed represents a constant speed cruise speed, each element of the predicted vehicle speed sequence is taken as the candidate vehicle speed and the longitudinal acceleration sequence is set to zero; when the candidate vehicle speed corresponds to the planned trajectory, the vehicle speed sequence and the longitudinal acceleration sequence are calculated from the planned trajectory.

[0195] Starting with the reference parameter vector and the energy consumption per unit mileage corresponding to the candidate vehicle speed, the vector is transformed to the current target parameter vector in six different orders, resulting in: ; Indicates the candidate vehicle speed; to These represent the order of the six vehicle physical state parameters in the candidate vehicle speed. The corresponding path prediction unit mileage energy consumption is shown below. The following text will use... Indicate the path number, and then... The predicted energy consumption per unit mileage for each path is uniformly denoted as: ,in .

[0196] The energy consumption per unit mile predicted for the six paths of the calibration sample must be calculated within the actual vehicle speed, actual parameter vector, and operating context of the calibration sample itself.

[0197] Maximum path prediction energy consumption per unit mile for: ;

[0198] Extreme difference in energy consumption per unit mileage in route prediction for: ;

[0199] Set preset positive energy consumption value : ;

[0200] No. Normalized difference of the path for: ;

[0201] Path difference vectors are formed by arranging them according to fixed path numbers. : .

[0202] The paths are numbered in descending order of predicted energy consumption per unit mileage to form a path sorting sequence; when the predicted energy consumption per unit mileage is the same, the path with the smaller number comes first.

[0203] For each candidate vehicle speed, it is first sorted in ascending order according to the minimum number of adjacent swaps between the calibration sample path sorting sequence and the candidate vehicle speed path sorting sequence; if the number of swaps is the same, it is then sorted in ascending order according to the L2 distance between the path difference vectors.

[0204] This implementation selects the first 30 calibration samples. If there are fewer than 30 available calibration samples, all available samples are selected; if there are fewer than 10, the previous valid one-sided calibration quantile value is used; for the first run and when there are insufficient samples, the factory offline calibration quantile value is used.

[0205] No. One-sided calibration residuals of each selected calibration sample for: ;in, Indicates the first The measured energy consumption per unit mileage of each selected calibration sample under its actual vehicle speed, actual parameter vector and operating context; Indicates the first The maximum path prediction energy consumption per unit mileage for each selected calibration sample under the same actual vehicle speed, actual parameter vector, and operating context.

[0206] Let the number of selected samples be... 1-sided quantile level is ,in To determine the acceptable unilateral coverage rate, this implementation method takes... Percentage number for: ;

[0207] Arrange the one-sided calibration residuals in ascending order, and then... The residual is used as the one-sided calibration quantile. .

[0208] One-sided calibration prediction upper bound curve value at candidate vehicle speed for: .

[0209] This curve is a conservative prediction curve formed under the selected calibration sample and preset quantile level, and does not represent an absolute upper limit that will not be exceeded under all unknown operating conditions.

[0210] When the current effective model version changes, the original path features and calibration residuals become invalid. The new model version is used to recalculate the predicted energy consumption per unit mileage, path difference vector, and path ranking sequence for the six paths of the historical calibration samples, and the calibration samples are reselected and the one-sided calibration quantile values ​​are determined.

[0211] S105: Determine the target economic speed and dynamic economic speed range based on the cumulative stability width index of the maximum continuous speed range corresponding to the local minimum point in the single-sided calibration prediction upper bound curve.

[0212] Specifically, a candidate speed grid is established at the intersection of the safe permissible speed range and the legal speed range. For example, under high-speed cruise conditions, speeds from 60 to 120 km / h are discretized in steps of 0.5 km / h. When the intersection is empty, no target economic speed or dynamic economic speed range is generated, and a range unavailable flag is output.

[0213] To avoid confusion with the sampling time index t, This represents the grid index of the candidate velocity, and the corresponding candidate velocity is denoted as . .

[0214] If candidate speed If the upper bound curve value of the unilateral calibration prediction at a given velocity is not higher than the curve value at the adjacent velocity, then that velocity is identified as a local minimum. If multiple consecutive velocity points have the same minimum curve value, then the midpoint of the platform velocity range is taken as the velocity corresponding to the local minimum. The endpoint of the velocity search range is identified as a local minimum only if its curve value is not higher than the only adjacent velocity point.

[0215] Set to increment sequentially Individual energy consumption increment threshold: , Indicates the first One energy consumption increment threshold; one feasible value is:

[0216] .

[0217] For the There are local minimum points, and their corresponding velocities are denoted as . In the first Under a threshold of incremental energy consumption, from satisfying Extracting from the candidate velocity set containing Maximum continuous speed range , represented as: The width of the interval corresponding to the current calculation period. for: ; This represents the low-speed boundary of the maximum continuous speed range; This represents the high-speed boundary of the maximum continuous speed range;

[0218] consecutive within the same currently valid model version Within each calculation cycle, the same local minimum point is matched based on the velocity difference between local minimum points and the intersection-combination ratio of the maximum continuous velocity interval. This implementation method takes... .

[0219] by This indicates the calculation of periodic indexes and interval intersection-union ratios. for: ; and This represents the two maximum continuous velocity intervals participating in cross-cycle matching; Indicates speed range The number of candidate velocity grid points within the range. Since each maximum continuous velocity interval contains at least the corresponding local minimum, the denominator is not zero.

[0220] When the speed difference between local minimum points does not exceed 2 km / h, and the intersection-to-exchange ratio of the intervals under the preset matching energy consumption increment threshold is not less than 0.60, they are determined to be the same local minimum point.

[0221] The first In the calculation cycle, the first The local minimum point is at the . The interval width corresponding to each energy consumption increment threshold is denoted as . For each energy consumption increment threshold, the minimum interval width of the same local minimum point within a continuous calculation period is determined as the stable width. : Stable width cumulative index for: Among them, the stable width corresponding to the smaller threshold among the adjacent energy consumption increment thresholds is adopted.

[0222] Determine the global minimum curve value : , will satisfy The local minimum point is determined as a candidate local minimum point; This indicates the preset candidate energy consumption difference threshold, which can be taken in this implementation: .

[0223] The speed corresponding to the candidate local minimum point with the largest cumulative stability width index is determined as the target economic speed. When the cumulative stability width indices are the same, the curve value with the lower value is selected; when the curve values ​​are still the same, the vehicle speed with the lower value is selected.

[0224] Set energy consumption tolerance From satisfying Extract the target economic speed from the speed set. The maximum continuous speed range.

[0225] The preset interval stability condition can be set as follows: within three consecutive calculation cycles, the intersection-union ratio of the current interval and the corresponding interval of the previous calculation cycle is not less than 0.80, and the change of the low-speed boundary and the high-speed boundary does not exceed 1km / h.

[0226] Once the stability condition is met, the maximum continuous speed range is defined as the dynamic economic speed range. If the stability condition is not met, the most recent valid dynamic economic speed range is maintained and a pending confirmation flag is output, or a range unavailable flag is output.

[0227] When the current effective model version changes, the local minimum point matching state under the previous model version is cleared, and the stable period is re-accumulated.

[0228] Vehicle physical state parameters that have generated non-zero sensitivity column vectors but have not been screened as identifiable parameters are identified as parameters to be verified. The time window for generating these sensitivity column vectors is the verification time window, and vehicle physical state parameters other than those to be verified are the matching parameters.

[0229] Set the parameters to be reviewed The corresponding sensitivity column vector is The sensitivity column vectors generated in the matching parameters are used to form a matrix. Fixed independent components for: ;

[0230] When there is no generated sensitivity column vector corresponding to the matching parameters, the following is specified: ;

[0231] Only when satisfied Perform pairing verification at the time. To preset the positive fixed independent component threshold, this embodiment can take 0.10. Otherwise, the fixed independent component is not used for pairing verification, and the parameters to be verified are retained in the set of parameters to be identified.

[0232] When selecting a subsequent time window as the paired time window, the actual parameter value of the parameter to be verified should return to near the reference value:

[0233] ;

[0234] in, This represents the actual parameter value of the parameter to be reviewed in the paired time window; Indicates the reference parameter value of the parameter to be reviewed; This represents a preset positive return threshold with the same unit as the physical state parameter of the j-th vehicle.

[0235] Simultaneously, the differences between the matching parameters and each component of the runtime context relative to the time window to be reviewed are divided by the corresponding matching scale to form a normalized matching difference vector. Each matching scale is calculated using the standard deviation of its corresponding component in the health history samples; when this standard deviation is lower than the corresponding preset positive lower limit, the preset positive lower limit is used, and each matching scale is saved along with the current effective model version. At that time, it is determined that the difference between the matching parameters and the runtime context does not exceed the matching threshold. This implementation method 1.5 is acceptable.

[0236] Normalized output residuals of the time window to be reviewed for: ; This represents a positive definite diagonal matrix composed of the standard deviations of the measurement noise of each output channel within the time window to be reviewed; This represents the measured output vector corresponding to the time window to be reviewed; This represents the actual branch output vector corresponding to the time window to be reviewed.

[0237] Normalized output residual of paired time window for: ; This represents the measured output vector corresponding to the paired time window; This represents the actual branch output vector corresponding to the paired time window;

[0238] Review the difference for: ;

[0239] when At that time, the projection scale is not calculated; This implementation method represents a preset positive verification residual threshold. 0.20 is acceptable.

[0240] Verify the projection components of the difference in the direction of the fixed independent components. for: ;

[0241] The orthogonal components with fixed independent component directions are: ;

[0242] when and At that time, the time window to be reviewed and the paired time window are determined as candidate update samples.

[0243] Candidate update samples are used to form candidate parameter updates for the conditional state transition model corresponding to the parameters to be reviewed. The parameter incremental projection coefficients can be calculated using fixed independent components, and the decision on whether to retain candidate updates is made according to the independent verification and model version writing rules in Example 5. Passing the paired review does not mean that the candidate parameters can be directly written into the current valid model version.

[0244] Vehicle physical state parameters that were not selected as identifiable parameters constitute the set of parameters to be identified. A parameter identification speed trajectory is generated for the set of parameters to be identified, sequentially including a first constant speed segment, an acceleration segment, a second constant speed segment, and a deceleration segment.

[0245] The parameter identification speed trajectory has the same starting speed and ending speed, and the speed changes in the acceleration and deceleration phases are opposites.

[0246] One candidate trajectory is as follows: the first constant speed section is 60km / h for 4 seconds; the acceleration section accelerates to 70km / h at 0.4 m / s²; the second constant speed section is 70km / h for 4 seconds; and the deceleration section decelerates to 60km / h at -0.4 m / s².

[0247] The runtime context during execution must meet the following constraints: the change in road slope does not exceed... The longitudinal equivalent wind speed variation shall not exceed 0.5 m / s; the power system temperature variation shall not exceed 2℃; the accessory load power variation shall not exceed 0.3 kW; the power system operating mode shall remain unchanged; and the longitudinal acceleration, jerk, legal speed, and vehicle safety limits shall all meet the constraints of the vehicle controller.

[0248] The parameter identification speed trajectory optimization is performed only on closed test roads, chassis dynamometers, approved test roads, or authorized parameter calibration modes with equivalent safety guarantees, and is supervised by the vehicle safety controller. When the vehicle safety controller cancels the trajectory, no parameter update sample is generated.

[0249] Select the same vehicle speed sampling points during acceleration and deceleration phases to calculate the difference in wheel-end traction power. : ; Indicates the acceleration phase at vehicle speed The corresponding wheel-end traction power; Indicates the deceleration phase at the same vehicle speed The corresponding wheel-end traction power; This indicates the vehicle speed at the same speed point during the acceleration and deceleration phases.

[0250] Calculate the difference in wheel-end traction power between the first constant speed range and the second constant speed range. : ; This indicates the wheel-end traction power corresponding to the second constant speed range; This indicates the wheel-end traction power corresponding to the first constant speed range;

[0251] The corresponding instantaneous energy input power difference is calculated. The trajectory feature output vector is then composed of the wheel-end traction power difference at each point of equal speed, the wheel-end traction power difference between the two constant speed segments, and the instantaneous energy input power difference, arranged in a fixed order. .

[0252] For each parameter to be identified, the predicted sensitivity column vector is calculated using central difference: ; This represents the column vector of predicted sensitivity for the j-th parameter to be identified under the current candidate parameter identification velocity trajectory; This represents a positive definite diagonal matrix composed of the standard deviations of the prediction noise of each component output by the trajectory features; This indicates that the j-th vehicle physical state parameter is increased by a disturbance based on its current physical parameter value. The resulting trajectory feature output vector; This indicates that the disturbance amount for the j-th vehicle physical state parameter is reduced based on the current physical parameter value. The resulting trajectory feature output vector; This represents the positive physical quantity disturbance value corresponding to the j-th vehicle physical state parameter;

[0253] The parameter disturbances can be taken as follows: vehicle equivalent mass 20kg; rolling resistance coefficient 0.0002; longitudinal equivalent wind speed 0.2m / s.

[0254] Each prediction noise standard deviation is estimated from the prediction residuals output by the trajectory characteristics of repeated parameter identification trajectory experiments or independent validation samples, and a preset positive lower limit is set, which is saved with the current effective model version. The resulting prediction sensitivity column vector corresponds to the dimensionless parameter change.

[0255] The predicted sensitivity column vectors of each parameter to be identified form a predicted sensitivity submatrix; for parameters that have been screened as identifiable, a reference predicted sensitivity column vector is generated using the same trajectory feature output, the standard deviation of the predicted noise, and the parameter normalization scale.

[0256] In the parameter identification velocity trajectory evaluation process, the sensitivity column vector corresponding to the identifiable parameter is specifically the reference predicted sensitivity column vector regenerated according to the velocity trajectory identified by the current candidate parameter, under the same trajectory feature output, prediction noise standard deviation, and parameter normalization scale; the identity of the identifiable parameter is still determined by the screening results of S102 mentioned above. Therefore, the predicted sensitivity column vector corresponding to the parameter to be identified and the reference predicted sensitivity column vector corresponding to the identifiable parameter are in the same trajectory feature space.

[0257] Candidate trajectories should satisfy , This implementation method represents a preset trajectory recognition threshold. A value of 0.20 can be taken, as this condition applies to all parameters to be identified.

[0258] When identifiable parameters exist, their sensitivity column vectors are combined into a matrix. The projection ratio of the column vector of predicted sensitivity of the identifiable parameter into the column space of identifiable parameter sensitivity. for: .

[0259] Candidate trajectories should also satisfy: This holds true for all parameters to be identified. The preset trajectory crosstalk threshold in this implementation method... 0.30 is acceptable.

[0260] When no identifiable parameters are available, the projection scale is set to zero.

[0261] From the candidate trajectories that meet the safety constraints, runtime context constraints, trajectory duration constraints, prediction sensitivity norm constraints, and projection scale constraints, select the trajectory with the largest minimum singular value of the prediction sensitivity submatrix.

[0262] If the intersection of the velocity ranges is empty or there are no candidate trajectories that meet all the conditions, the parameter identification of velocity trajectories will not be performed, and the current effective model version will continue to be used.

[0263] After executing the selected trajectory, the sensitivity matrix is ​​reconstructed based on the measured wheel-end traction power sequence and the instantaneous energy input power sequence. Only for the parameters to be identified that have been re-selected as identifiable parameters and whose corresponding state transition closure error does not exceed the closure threshold, parameter candidate updates for their conditional state transition models are generated.

[0264] The predicted sensitivity submatrix is ​​only used for trajectory selection. The actual model update still uses the measured output after trajectory execution and the independent verification time window.

[0265] like Figure 2 As shown, another embodiment of the present invention provides a dynamic economic speed range generation system based on vehicle physical state, which can be deployed on a vehicle central computing platform, vehicle controller or intelligent driving domain controller, including a model version acquisition module, a branch generation and parameter identification module, a model update module, a prediction upper bound generation module and a speed range determination module.

[0266] The model version acquisition module connects to non-volatile memory to read the currently valid model version. The non-volatile memory contains a current version area and a candidate version area. The current version pointer is not changed until a candidate version is fully written and passes verification.

[0267] The branch generation and parameter identification module includes a parameter vector generation unit, a reference branch generation unit, an actual branch generation unit, a single parameter replacement branch generation unit, a lap-loop error calculation unit, a sensitivity matrix calculation unit, and an identifiable parameter filtering unit.

[0268] The parameter vector generation unit generates reference parameter vectors, actual parameter vectors, and the runtime context. Each branch generation unit calls the road load model, energy conversion model, and conditional state transition model, outputting the wheel-end traction power sequence, instantaneous energy input power sequence, and energy consumption per unit mileage. The conditional state transition model is used to correct the energy consumption per unit mileage, without repeatedly writing correction values ​​into the instantaneous power sequence. The loopback error calculation unit performs a reverse state transition to obtain the loopback energy consumption per unit mileage. The sensitivity matrix calculation unit constructs a sensitivity matrix based on the single-parameter replacement branch output and the reference branch output. The identifiable parameter selection unit determines identifiable parameters based on the minimum singular value, condition number, independent component norm, and state transition loopback error.

[0269] The model update module includes a directional energy consumption residual calculation unit, a conditional state transition model update unit, an orthogonal residual calculation unit, a road load model update unit, an energy conversion model update unit, an independent verification unit, and a model version management unit.

[0270] The directional energy consumption residual calculation unit generates directional energy consumption residuals based on the independent components corresponding to the identifiable parameters. The conditional state transition model update unit generates candidate updates for the conditional state transition model parameters and regenerates the actual branch output and single-parameter replacement branch output based on the independently verified conditional state transition model; the sensitivity matrix calculation unit reconstructs the sensitivity matrix based on the regenerated actual branch output and single-parameter replacement branch output.

[0271] If the parameters of the road load model and the energy conversion model remain unchanged, since the state transition correction value of the conditional state transition model only applies to the energy consumption per unit mileage and is not repeatedly written into the instantaneous power sequence, the reconstructed sensitivity matrix is ​​allowed to have the same value as the sensitivity matrix before the conditional state transition model update, but the sensitivity matrix reconstruction process is still performed.

[0272] When all three vehicle physical state parameters are identifiable, the orthogonal residual calculation unit generates orthogonal residuals; the road load model update unit and the energy conversion model update unit generate corresponding slow-changing parameter candidate updates in the order described in the above embodiments. The independent verification unit verifies each candidate update item by item, and the model version management unit writes all verified and retained candidate updates as a consistent transaction into the new currently valid model version.

[0273] The upper bound prediction module includes a candidate velocity generation unit, a six-path calculation unit, a path difference calculation unit, a calibration sample retrieval unit, and a one-sided calibration quantile calculation unit.

[0274] The candidate speed generation unit generates discrete candidate speeds based on the safe speed range and the legal speed range; the six-path calculation unit generates the predicted energy consumption per unit mileage according to the six possible permutations of the three vehicle physical state parameters; the path difference calculation unit forms the path difference vector and the path sorting sequence; the calibration sample retrieval unit selects calibration samples according to the minimum number of adjacent exchanges and the L2 distance; and the one-sided calibration quantile calculation unit generates the one-sided calibration prediction upper bound curve.

[0275] The speed range determination module includes a local minimum point detection unit, a maximum continuous speed range extraction unit, a cross-cycle matching unit, a stable width cumulative index calculation unit, and a dynamic economic speed range output unit.

[0276] The local minimum point detection unit detects local minimum points in the single-sided calibration prediction upper bound curve; the maximum continuous speed range extraction unit extracts the maximum continuous speed range containing local minimum points under multiple energy consumption increment thresholds; the cross-cycle matching unit matches the same local minimum point based on the speed difference of local minimum points and the interval intersection ratio; the stability width cumulative index calculation unit determines the target economic speed; and the dynamic economic speed range output unit outputs the dynamic economic speed range after satisfying the preset stability conditions.

[0277] The system outputs the target economic speed, the low-speed and high-speed boundaries of the dynamic economic speed range, the current valid model version number, the range validity flag, and a timestamp to the vehicle longitudinal planning module. Vehicle safety constraints, collision avoidance constraints, road adhesion constraints, driver instructions, and legal speed limits have higher priority than the dynamic economic speed range.

[0278] Model updates are performed using a background thread, while the upper bound curve for one-sided calibration prediction and the dynamic economic velocity range are calculated using a foreground thread. At the beginning of each computation cycle, the foreground thread reads a complete version of the currently valid model and maintains that version throughout the cycle. Background candidate updates do not interrupt foreground output; the new model version only takes effect at the boundary of the next computation cycle.

[0279] When data involves vehicle usage records or driving behavior data that can be linked to natural persons, data collection, storage, and processing are conducted with the user's consent or under other legally valid processing conditions, and are subject to access control, transmission encryption, de-identification, and retention period control. Model input does not include facial features, biometrics, gender, age, ethnicity, or other personal attributes unrelated to vehicle energy consumption.

[0280] In summary, this invention can reduce the erroneous attribution between changes in vehicle physical state and changes in the model itself, suppress the impact of invalid model updates and local prediction fluctuations on the determination of economic speed, thereby improving the accuracy of energy consumption prediction per unit mileage and the robustness, continuity and adaptability of dynamic economic speed range output.

[0281] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating dynamic economic speed ranges based on vehicle physical states, characterized in that, include: Obtain the current valid model version, including the road load model, the energy conversion model, and the conditional state transition models corresponding to the vehicle physical state parameters; the vehicle physical state parameters include the vehicle equivalent mass, rolling resistance coefficient, and longitudinal equivalent wind speed. The reference branch output, actual branch output, single-parameter replacement branch output, and loop unit mileage energy consumption are generated based on the reference parameter vector and the actual parameter vector. A sensitivity matrix is ​​constructed based on the single-parameter replacement branch output and the reference branch output, and the identifiable parameters are determined by combining the state transition loop error. The parameter candidate updates of the conditional state transition model are formed by using directional energy consumption residuals. After verification by an independent verification time window, they are retained or withdrawn, and the sensitivity matrix is ​​reconstructed based on the verified conditional state transition model. When all vehicle physical state parameters are identifiable parameters, slow-changing parameter candidate updates of the road load model and energy conversion model are formed by using orthogonal residual compensation. All verified and retained candidate updates are written into the new current effective model version. For candidate vehicle speeds, the path prediction unit mileage energy consumption is generated according to six permutations of vehicle physical state parameters. The upper limit curve of the one-sided calibration prediction is generated by the sum of the maximum path prediction unit mileage energy consumption and the one-sided calibration quantile value. Based on the cumulative stability width index of the maximum continuous speed interval corresponding to the local minimum point in the single-sided calibration prediction upper bound curve, the target economic speed and dynamic economic speed interval are determined.

2. The method for generating dynamic economic speed ranges based on vehicle physical states according to claim 1, characterized in that, The conditional state transition model corresponding to any vehicle physical state parameter performs the following steps: The system receives an input parameter vector, a target parameter value, a running context, and an input energy consumption per unit mileage. While keeping the other vehicle physical state parameters unchanged, it replaces the input parameter value of any of the vehicle physical state parameters with the target parameter value to obtain an output parameter vector. Using the same road load model and energy conversion model, the physical predicted energy consumption per unit mileage of the input parameter vector and the output parameter vector under the operating context are calculated respectively. The physical predicted energy consumption difference is obtained by subtracting the physical predicted energy consumption per unit mileage of the input parameter vector from the physical predicted energy consumption per unit mileage of the output parameter vector. Using the conditional state transition model, a state transition correction value is determined based on the input parameter vector, the output parameter vector, and the operating context. The sum of the input energy consumption per unit mileage, the difference between the physical predicted energy consumption and the state transition correction value is determined as the output energy consumption per unit mileage. When exchanging the input parameter values ​​and the target parameter values, the same road load model parameters, energy conversion model parameters, and conditional state transition model parameters are used to perform the reverse state transition.

3. The method for generating dynamic economic speed ranges based on vehicle physical states according to claim 2, characterized in that, Both the reference parameter vector and the actual parameter vector are constructed according to the preset parameter order of vehicle equivalent mass, rolling resistance coefficient and longitudinal equivalent wind speed. The reference branch output is generated by directly inputting the reference parameter vector into the road load model and the energy conversion model; the actual branch output is generated by sequentially converting the reference parameter vector according to the preset parameter order until the actual parameter vector is reached. The output of any single-parameter replacement branch is generated by replacing the corresponding vehicle physical state parameter in the reference parameter vector with the actual parameter value in the actual parameter vector. The reference branch output, the actual branch output, and the single-parameter replacement branch output all include an output vector and energy consumption per unit mileage, and use the same running context and the current effective model version; Using the single-parameter replacement branch unit mileage energy consumption in the single-parameter replacement branch output as the input unit mileage energy consumption for the reverse state transition, the corresponding vehicle physical state parameters are reversed to the reference parameter value to obtain the loop unit mileage energy consumption. The absolute value of the difference between the loop unit mileage energy consumption and the reference branch unit mileage energy consumption is divided by the standard deviation of the unit mileage energy consumption measurement noise to obtain the state transition loop error of the corresponding vehicle physical state parameters.

4. The method for generating dynamic economic speed ranges based on vehicle physical states according to claim 3, characterized in that, The wheel-end traction power sequence and instantaneous energy input power sequence at least two sampling times within the time window are stacked into an output vector according to a preset output arrangement order; When the absolute value of the difference between the actual parameter value and the reference parameter value of any vehicle physical state parameter is not lower than the parameter difference threshold, the difference between the output vector of the corresponding single parameter replacement branch output and the output of the reference branch is normalized according to the difference of the parameter value and the standard deviation of the measurement noise of each output channel to obtain the sensitivity column vector. Independent components are obtained by projecting any sensitivity column vector onto the orthogonal complement of the subspace spanned by the other sensitivity column vectors. The sensitivity matrix is ​​formed by the generated sensitivity column vectors. Identifiable parameters are selected based on the minimum singular value and condition number of the sensitivity matrix, the norm of the independent components, and the state transition loop error. The difference between the measured output vector and the actual branch output vector is normalized according to the preset output arrangement order and the standard deviation of the measurement noise corresponding to each output channel. The projection coefficient of the normalized difference in the direction of the independent component corresponding to the identifiable parameter is calculated. The projection coefficient is divided by the difference between the actual parameter value and the reference parameter value of the corresponding vehicle physical state parameter to obtain the dimensionless orientation coefficient. The dimensionless orientation coefficient is multiplied by the difference between the unit mileage energy consumption of the corresponding single parameter replacement branch and the unit mileage energy consumption of the reference branch to obtain the orientation energy consumption residual.

5. The method for generating dynamic economic speed ranges based on vehicle physical states according to claim 4, characterized in that, The parameter candidate update of the corresponding conditional state transition model is formed by using the directional energy consumption residual. After verification by an independent verification time window, the parameter candidate update is retained or withdrawn. Based on the verified conditional state transition model, the actual branch output and the single parameter replacement branch output are generated and the sensitivity matrix is ​​reconstructed. When all vehicle physical state parameters are identifiable parameters, the difference between the measured output vector and the actual branch output vector is normalized and projected onto the orthogonal complement of the sensitivity matrix column space to obtain the orthogonal complement residual. The normalized Jacobian matrix of the road load model is projected onto the orthogonal complement, and the rows corresponding to the wheel-end traction power sequence are extracted. When the identifiability condition is met, the same row components of the orthogonal complement residuals are used to form candidate updates of the slow-changing parameters of the road load model. The sensitivity matrix and the orthogonal complement residuals are reconstructed based on the verified road load model parameters. The normalized Jacobian matrix of the energy conversion model is projected onto the orthogonal complement and the row corresponding to the instantaneous energy input power sequence is extracted. When the identifiability condition is met, the same row components of the orthogonal complement residual are used to form candidate updates of the slow-changing parameters of the energy conversion model. Each candidate update is retained or withdrawn after being verified by an independent verification time window, and the next candidate update is formed only based on the verified model parameters; all retained candidate updates are written together into the new current effective model version; the rolling resistance model term of the road load model uses only the rolling resistance coefficient as the free parameter to be updated.

6. The method for generating dynamic economic speed ranges based on vehicle physical states according to claim 5, characterized in that, Historical samples that did not participate in the parameter update of the current effective model version were selected as calibration samples; For each candidate vehicle speed and each calibration sample, starting from the reference parameter vector and the reference branch unit mileage energy consumption respectively, the conditional state transition model is called in sequence according to the six permutations of the vehicle physical state parameters to obtain the six path prediction unit mileage energy consumption from the reference parameter vector to the corresponding actual parameter vector. The normalized difference between the maximum path predicted energy consumption per unit mileage for each candidate vehicle speed and each calibration sample and the corresponding path predicted energy consumption per unit mileage is used to construct the corresponding path difference vector, and the corresponding path sorting sequence is constructed based on the descending path number of the corresponding path predicted energy consumption per unit mileage. For each candidate vehicle speed, the calibration samples are sorted according to the minimum number of adjacent exchanges between each calibration sample and the path sorting sequence corresponding to the candidate vehicle speed and the L2 distance between the path difference vectors, and a preset number of calibration samples are selected. The positive part of the difference between the measured energy consumption per unit mileage of the selected calibration sample at the actual vehicle speed and the maximum path prediction energy consumption per unit mileage of the calibration sample at the actual vehicle speed is determined as the one-sided calibration residual. The one-sided calibration quantile is determined based on the one-sided calibration residual and added to the maximum path prediction energy consumption per unit mileage of the candidate vehicle speed to obtain the one-sided calibration prediction upper bound curve. When the current effective model version changes, the calibration sample is reselected and the one-sided calibration quantile is recalculated.

7. The method for generating dynamic economic speed ranges based on vehicle physical states according to claim 6, characterized in that, Set multiple energy consumption increment thresholds that increase sequentially for the upper bound curve of the unilateral calibration prediction; For each local minimum point, extract the maximum continuous speed range that includes the local minimum point and whose curve value is not higher than the sum of the curve value corresponding to the local minimum point and the corresponding energy consumption increment threshold under each energy consumption increment threshold. In multiple consecutive calculation cycles of the current effective model version, the same local minimum point is matched according to the velocity difference between local minimum points and the crossover ratio of the corresponding maximum continuous velocity interval. For each energy consumption increment threshold, the minimum value of the width of the maximum continuous velocity interval corresponding to the same local minimum point in each calculation cycle is determined as the corresponding stable width. The product of the difference between each adjacent energy consumption increment threshold and the stable width corresponding to the smaller energy consumption increment threshold among the adjacent energy consumption increment thresholds is accumulated to obtain the stable width cumulative index. From the local minimum points where the curve value is not higher than the sum of the global minimum curve value and the preset candidate energy consumption difference threshold, the speed corresponding to the local minimum point with the largest stable width cumulative index is determined as the target economic speed. Extract the maximum continuous speed range that includes the target economic speed and whose curve value increment relative to the curve value corresponding to the target economic speed does not exceed the energy consumption tolerance. When the maximum continuous speed range meets the preset range stability condition in multiple consecutive calculation cycles, the maximum continuous speed range is determined as the dynamic economic speed range.

8. The method for generating dynamic economic speed ranges based on vehicle physical states according to claim 7, characterized in that, Vehicle physical state parameters that have generated non-zero sensitivity column vectors but have not been screened as identifiable parameters are identified as parameters to be reviewed. For each parameter to be verified, the time window for generating the corresponding sensitivity column vector is determined as the verification time window, and the vehicle physical state parameters other than the parameters to be verified are determined as matching parameters. Project the sensitivity column vector corresponding to the parameter to be verified onto the orthogonal complement of the subspace spanned by the sensitivity column vector corresponding to the matching parameter. After obtaining non-zero fixed independent components, select the subsequent time window where the absolute value of the difference between the actual parameter value and the reference parameter value of the parameter to be verified does not exceed the return threshold, and the difference between the matching parameter value and the running context between the time window to be verified and the time window to be verified does not exceed the matching threshold as the pairing time window. The difference between the measured output vector and the actual branch output vector of the time window to be verified and the paired time window is normalized according to the standard deviation of the measurement noise of each output channel. The verification difference is obtained by subtracting the normalized difference of the paired time window from the normalized difference of the time window to be verified. When the ratio of the projection norm of the verification difference along the direction of the fixed independent component to the norm of the verification difference is not lower than the attribution threshold, and the projection norm on the orthogonal complement of the direction of the fixed independent component is not higher than the verification crosstalk threshold, the time window to be verified and the paired time window are determined as candidate update samples to form the parameter candidate update of the conditional state transition model corresponding to the parameter to be verified.

9. The method for generating dynamic economic speed ranges based on vehicle physical states according to claim 8, characterized in that, The vehicle physical state parameters that were not filtered as identifiable parameters constitute the set of parameters to be identified. For the set of parameters to be identified, a speed trajectory is generated that sequentially includes a first constant speed segment, an acceleration segment, a second constant speed segment, and a deceleration segment, with the same start and end speeds; the speed changes in the acceleration and deceleration segments are opposites of each other, and the running context meets preset constraints during execution; A prediction sensitivity sub-matrix is ​​constructed based on the difference in wheel-end traction power at the same speed point between the acceleration and deceleration phases, the difference in wheel-end traction power between the first and second constant speed phases, and the difference in instantaneous energy input power. Candidate trajectories are defined as follows: the norm of the predicted sensitivity column vector corresponding to each parameter to be identified in the predicted sensitivity submatrix is ​​not lower than the trajectory identification threshold, and the ratio of the projection norm of each predicted sensitivity column vector in the subspace spanned by the sensitivity column vectors corresponding to the identifiable parameters to the norm of the corresponding predicted sensitivity column vector does not exceed the trajectory crosstalk threshold. When there are no identifiable parameters, the ratio is zero. The candidate trajectory with the largest minimum singular value in the corresponding predicted sensitivity submatrix is ​​selected. When there are no candidate trajectories, the velocity trajectory identification parameter is not performed. After the selected candidate trajectory is executed, the sensitivity matrix is ​​reconstructed, and parameter candidate updates are formed only for the parameters to be identified that are selected as identifiable parameters and whose corresponding state transition loop error does not exceed the loop closure threshold.

10. A dynamic economic speed range generation system based on vehicle physical state, applied to the dynamic economic speed range generation method based on vehicle physical state as described in claim 1, characterized in that, include: The model version acquisition module is used to obtain the currently valid model version; The branch generation and parameter identification module is used to generate reference branch output, actual branch output, single-parameter replacement branch output and loop unit mileage energy consumption based on reference parameter vector and actual parameter vector. It constructs a sensitivity matrix based on single-parameter replacement branch output and reference branch output, and determines identifiable parameters by combining state transition loop error. The model update module is used to form candidate parameter updates for the conditional state transition model using directional energy consumption residuals. After verification by an independent verification time window, the updates are retained or withdrawn, and the sensitivity matrix is ​​reconstructed based on the verified conditional state transition model. When all vehicle physical state parameters are identifiable parameters, slow-changing parameter updates for the road load model and energy conversion model are formed using orthogonal residuals. All verified and retained candidate updates are written into the new current effective model version. The upper bound prediction generation module is used to generate the path prediction unit mileage energy consumption for candidate vehicle speeds according to six permutations of vehicle physical state parameters, and generate the single-sided calibration prediction upper bound curve by the sum of the maximum path prediction unit mileage energy consumption and the single-sided calibration quantile value. The speed range determination module is used to determine the target economic speed and dynamic economic speed range based on the cumulative stability width index of the maximum continuous speed range corresponding to the local minimum point in the single-sided calibration prediction upper bound curve.

Images