Complete vehicle state identification method and device for hybrid control of commercial vehicle and storage medium

Abstract

The invention relates to a whole vehicle state recognition method and device for hybrid control of a commercial vehicle and a storage medium, and belongs to the technical field of hybrid commercial vehicle control. The method comprises the steps that state observation parameters of the vehicle are determined based on the real-time vehicle speed and the moving average vehicle speed of the vehicle; the state observation parameters comprise a real-time driving parameter of the vehicle at the current moment, a first historical driving parameter within a first time period before the current moment, and a second historical driving parameter within a second time period before the current moment; and inputting the state observation parameters into a trained target recognition model to obtain a vehicle state output by the target recognition model. According to the whole vehicle state recognition method for hybrid control of the commercial vehicle, the vehicle state can be accurately recognized on the basis of the HMM under the condition that the hardware computing power and the storage capacity of a control system of an existing vehicle are not increased.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hybrid commercial vehicle control, and in particular to a commercial vehicle hybrid control vehicle state recognition method, device and storage medium. BACKGROUND

[0002] Compared with a traditional commercial vehicle with only an internal combustion engine as a power source, for a certain vehicle driving demand, only the gear of the gearbox can be adjusted. However, for a hybrid commercial vehicle including an electric motor and an internal combustion engine as two driving sources, the adjustment of the powertrain response selection includes: 1) the gear of the gearbox; 2) the driving mode of the powertrain (including engine alone, motor alone, engine + motor, etc.); 3) the driving distribution between the motor and the engine in the hybrid driving mode; 4) whether the engine is stopped in the motor-only driving mode; and 5) whether the motor is disconnected in the engine-only driving mode.

[0003] The hybrid powertrain has more and more flexible response selection for the vehicle. In theory, the optimal response scheme in terms of economy and power can be found for different vehicle conditions. However, when the state of the vehicle (vehicle speed and power demand) changes rapidly, the corresponding theoretical optimal scheme will also be switched rapidly. The rapid switching of the powertrain response scheme not only causes the vehicle power to be frequently and discontinuously changed, affecting the driving experience, but also causes the powertrain efficiency to be reduced, resulting in additional energy consumption and deteriorating the performance of the vehicle. Therefore, in the process of optimal control of the hybrid powertrain, the state of the vehicle needs to be quickly and accurately identified, and the control target of the hybrid control needs to be adjusted in terms of economy, power, and the continuity of the response of the physical system under different vehicle states.

[0004] The specific state of the vehicle is closely related to the running scene, road, and driver style of the vehicle. The vehicle state recognition method should be able to adaptively adjust according to the specific scene, road, and driver of the vehicle.

[0005] The existing vehicle state recognition method includes a rule-based recognition method and a deep learning model (LSTM / Transformer) based recognition method.

[0006] The recognition method based on rule definition is a deterministic method based on pre-defined logic and threshold judgment. In each moment of inspection, the recognition conditions and thresholds corresponding to different states of cloud computing are checked. Once the predetermined conditions are met, the driving state corresponding to the rule is immediately output. The recognition method based on rule definition does not require training data, can quickly build a prototype based on domain knowledge, has extremely low computational cost and complexity, and can be quickly deployed. However, the recognition method based on rule definition also has obvious disadvantages: 1) poor generalization ability. The pre-defined rules and thresholds are heavily dependent on specific vehicle models, loads, road conditions and driver styles. Any condition change (such as a larger engine) may cause the original rules to fail, and engineers need to adjust and calibrate the rules based on experience; 2) low recognition accuracy. A large number of false positives and false negatives may occur. Rules are usually based on current instantaneous or short window features for judgment, and lack of memory and use of historical state sequences. There are a lot of fuzzy and overlapping areas in real driving state transitions (such as slow following and intended parking). Simple threshold-based rules cannot fully describe this complexity.

[0007] The recognition method based on deep learning model uses deep neural networks, especially network structures designed for sequence data, to automatically learn the complex mapping relationship from input signals to driving states in an end-to-end manner. The model automatically learns the complex and high-order relationship between features and states through multiple layers of nonlinear transformation without manual definition or discretization. The network may learn basic features at the bottom layer, and the high layer may combine these features to form more abstract patterns. The recognition method based on deep learning model can describe extremely complex nonlinear relationships and long-range context dependencies, and can theoretically achieve high accuracy. After training on large-scale and diversified data, the model shows stronger robustness and generalization ability to new vehicles, drivers and environmental changes. However, the main disadvantages of the recognition method based on deep learning model for use on current vehicles include: 1) The accuracy and generalization ability of the model are directly related to the number and quality of the sample data used for model training. The cost of collecting, cleaning and labeling high-quality training data is high; 2) In addition to the demand for model training computing power, even after model compression (pruning, quantization) on the vehicle control system, the computational and storage overhead is very large, requiring high vehicle hardware computing power, involving hardware upgrades and cost increases. SUMMARY

[0008] Therefore, it is necessary to provide a commercial vehicle hybrid control whole vehicle state recognition method and device and storage medium to accurately identify the whole vehicle state at a low cost.

[0009] To achieve the above purpose, in a first aspect, the present application provides a commercial vehicle hybrid control whole vehicle state recognition method, comprising: determine a state observation parameter of the vehicle based on a real-time vehicle speed and a moving average vehicle speed of the vehicle; the state observation parameter comprises a real-time driving parameter of the vehicle at a current time, a first historical driving parameter in a first time period before the current time, and a second historical driving parameter in a second time period before the current time; the first time period is less than the second time period; input the state observation parameter into a trained target recognition model to obtain a vehicle state output by the target recognition model; the target recognition model comprises an HMM model and a GMM model, and the target recognition model is trained based on a labeled state sequence and an observation sequence corresponding to the labeled state sequence.

[0010] In a possible implementation, the determining of the state observation parameter of the vehicle based on the real-time vehicle speed and the moving average vehicle speed of the vehicle comprises: determining a plurality of sampling anchor points based on an intersection of the real-time vehicle speed and the moving average vehicle speed; determining a target sampling anchor point closest to a current sampling point based on a position of the current sampling point; determining a feature value vector corresponding to the current sampling point based on a feature parameter between the target sampling anchor point and the current sampling point; constructing a feature matrix based on feature value vectors corresponding to all sampling points; determining the state observation parameter based on the feature matrix.

[0011] In a possible implementation, the training parameters of the HMM model comprise a transition probability matrix of vehicle states. The training parameters of the GMM model comprise a weight vector, a mean vector, and a covariance matrix of each observation value.

[0012] In a possible implementation, the inputting of the state observation parameter into the trained target recognition model to obtain the vehicle state output by the target recognition model comprises: inputting the state observation parameter into a trained GMM model to obtain a probability distribution of the state observation parameter under different vehicle states output by the GMM model; inputting the probability distribution and the transition probability matrix into a trained HMM model to obtain a vehicle state output by the HMM model.

[0013] In a possible implementation, after the inputting of the state observation parameter into the trained target recognition model to obtain the vehicle state output by the target recognition model, the method further comprises: performing adaptive adjustment on the training parameters of the HMM model and the training parameters of the GMM model based on the vehicle state output by the target recognition model.

[0014] In a possible implementation, the real-time driving parameter comprises: a current vehicle speed; a speed difference between the current vehicle speed and a latest local vehicle speed extreme point; a time interval between a current time and the latest local vehicle speed extreme point; an average accelerator opening degree within the time interval between the current time and the latest local vehicle speed extreme point; an average brake opening degree within the time interval between the current time and the latest local vehicle speed extreme point; a latest local vehicle speed extreme point speed; a maximum accelerator opening degree from the current to the latest local vehicle speed extreme point; a maximum brake opening degree from the current to the latest local vehicle speed extreme point; The first historical driving parameter comprises: a cumulative vehicle speed increase amount within a first time period before the latest local vehicle speed extreme point speed; a cumulative vehicle speed decrease amount within the first time period before the latest local vehicle speed extreme point speed; an average accelerator opening degree within the first time period before the latest local vehicle speed extreme point speed; an average brake opening degree within the first time period before the latest local vehicle speed extreme point speed; The second historical driving parameter comprises: a cumulative vehicle speed increase amount within a second time period before the latest local vehicle speed extreme point speed; a cumulative vehicle speed decrease amount within the second time period before the latest local vehicle speed extreme point speed; an average accelerator opening degree within the second time period before the latest local vehicle speed extreme point speed; an average brake opening degree within the second time period before the latest local vehicle speed extreme point speed.

[0015] In a possible implementation, the vehicle state comprises any one of the following: a starting state, a temporary stop state, a crawling state, a smooth driving state, a continuously accelerating state during driving, or a continuously decelerating state during driving.

[0016] In a second aspect, the present application further provides a whole vehicle state recognition device for commercial vehicle hybrid control, comprising: a determination unit configured to determine a state observation parameter of a vehicle based on a real-time vehicle speed and a moving average vehicle speed of the vehicle; the state observation parameter comprises a real-time driving parameter of the vehicle at a current time, a first historical driving parameter within a first time period before the current time, and a second historical driving parameter within a second time period before the current time; the first time period is less than the second time period; The recognition unit is configured to input the state observation parameter into a trained target recognition model to obtain a vehicle state output by the target recognition model; the target recognition model comprises an HMM model and a GMM model, and the target recognition model is trained based on a labeled state sequence and an observation sequence corresponding to the labeled state sequence.

[0017] In a third aspect, the present application further provides an electronic device comprising a memory and a processor, wherein, The memory is configured to store a program. The processor, coupled to the memory, is configured to execute the program stored in the memory to implement the steps of the commercial vehicle hybrid control whole vehicle state recognition method in any of the implementation manners described above.

[0018] In a fourth aspect, the present application further provides a computer readable storage medium configured to store computer readable programs or instructions, which, when executed by a processor, can implement the steps of the commercial vehicle hybrid control whole vehicle state recognition method in any of the implementation manners described above.

[0019] The commercial vehicle hybrid control whole vehicle state recognition method, device and storage medium provided by the present application have the following beneficial effects: the state observation parameter of the vehicle is determined according to the real-time vehicle speed and the moving average vehicle speed, the state observation parameter comprises real-time observation parameters, historical information of the vehicle in a short period and historical information of the vehicle in a medium or long period, the vehicle state output by the target recognition model is obtained by inputting the state observation parameter into the trained target recognition model, and the vehicle state can be accurately recognized based on the HMM model without increasing the hardware computing power and storage capacity of the existing vehicle control system. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0021] Figure 1 An embodiment flowchart of the commercial vehicle hybrid control whole vehicle state recognition method provided by the present application is shown in the figure. Figure 2 A curve diagram of the real-time vehicle speed and the moving average vehicle speed provided by the present application is shown in the figure. Figure 3 An embodiment structure diagram of the commercial vehicle hybrid control whole vehicle state recognition device provided by the present application is shown in the figure. Figure 4An embodiment structure schematic diagram of an electronic device provided by the present application is shown in the following. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be clearly and completely described in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative work fall within the protection scope of the present application.

[0023] In the description of the embodiments of the present application, unless otherwise specified, the meaning of “a plurality of” is two or more. The association relationship of “and / or” describing the associated objects indicates that there can be three relationships, for example: A and / or B can represent the three cases of A existing alone, A and B existing together, and B existing alone.

[0024] The “first”, “second” and the like described in the embodiments of the present application are only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the technical features limited by “first”, “second” can explicitly or implicitly include at least one of the features.

[0025] In this document, the reference to “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive of other embodiments. It will be explicitly understood by a person skilled in the art that the embodiments described herein can be combined with other embodiments.

[0026] The present application provides a commercial vehicle hybrid control whole vehicle state recognition method, device and storage medium, which are described below.

[0027] Figure 1 An embodiment flowchart of a commercial vehicle hybrid control whole vehicle state recognition method provided by the present application is shown in the following. Figure 1 As shown in the figure, the commercial vehicle hybrid control whole vehicle state recognition method comprises: S101, determining a state observation parameter of the vehicle based on a real-time vehicle speed and a moving average vehicle speed of the vehicle; the state observation parameter comprises a real-time driving parameter of the vehicle at the current time, a first historical driving parameter in a first time period before the current time, and a second historical driving parameter in a second time period before the current time; the first time period is less than the second time period; S102, input the state observation parameter into the trained target recognition model to obtain a vehicle state output by the target recognition model; the target recognition model comprises an HMM model and a GMM model, and the target recognition model is trained based on a labeled state sequence and an observation sequence corresponding to the labeled state sequence.

[0028] In S101, a state observation parameter of the vehicle is extracted according to a real-time vehicle speed and a moving average vehicle speed of the vehicle.

[0029] The state observation parameter can comprise a speed history (vehicle speed data at different time points, including a current speed, a historical speed, and a change trend of the speed, such as acceleration and deceleration), a throttle and brake opening history (opening data of a throttle and brake pedal, and a change of the opening over time), time interval information (a time interval between a current time and a historical key point, such as a speed extreme point), a cumulative speed change (a cumulative increase or decrease of the speed in a specific time period), an average throttle and brake opening (an average value of the throttle and brake opening in a specific time period), and a speed.

[0030] The real-time driving parameter at the current time can reflect current time information of the vehicle, the first historical driving parameter is a driving parameter in a short period, and reflects historical information of the vehicle in a short period, and the second historical driving parameter is a driving parameter in a medium or long period, and reflects historical information of the vehicle in a medium or long period.

[0031] In S102, in the hybrid control vehicle state recognition method, the most probable vehicle state in the present and future time period can be recognized and estimated based on historical and present observation signal data of the vehicle.

[0032] The obtained state observation parameter is input into the trained target recognition model, and a vehicle state output by the target recognition model can be obtained.

[0033] The target recognition model is trained based on a labeled state sequence (i.e., a known actual state of the vehicle) and an observation sequence corresponding to the labeled state sequence (i.e., an observation parameter in the corresponding state) on a Hidden Markov Model (HMM) and a Gaussian Mixture Model (GMM).

[0034] The HMM model is an extended application of a Markov chain, has the basic property of "no memory" of the Markov chain, and the core idea is that the state of the system is hidden (not directly observable), but an observable signal is output, and the generation probability of the observable signal depends on the current hidden state.

[0035] By using the annotated driving state sequence and the corresponding observation value sequence, the basic parameters of the model are trained offline, including the transition probability matrix of the hidden state (driving state) of the HMM model and the GMM model parameters (including the weight vector, mean vector and covariance matrix of each component) describing the probability density of each observation value under various states.

[0036] Using the trained model, the vehicle state with the highest probability in the current and future period of time can be accurately identified according to the extracted current and historical feature parameters. The GMM model can output the probability distribution of the state observation parameters under different vehicle states, and the HMM model can output the final vehicle state as the starting state, the temporary stop state, the creeping state, the smooth driving state, the continuous acceleration state or the continuous deceleration state in driving according to the probability distribution and the transition probability matrix.

[0037] In summary, the commercial vehicle hybrid control whole vehicle state recognition method provided by the embodiment of the application determines the state observation parameters of the vehicle according to the real-time vehicle speed and the moving average vehicle speed of the vehicle, the state observation parameters including the real-time observation parameters, the historical information of the vehicle in a short period and the historical information of the vehicle in a medium or long period, and the vehicle state output by the target recognition model is obtained by inputting the state observation parameters into the trained target recognition model, so that the vehicle state can be accurately identified based on the HMM model without increasing the hardware computing power and storage capacity of the existing vehicle control system.

[0038] In some embodiments of the application, the real-time observation parameters include: the current vehicle speed; the speed difference between the current vehicle speed and the latest local vehicle speed extreme point; the time interval between the current time and the latest local vehicle speed extreme point; the average accelerator opening degree in the time interval between the current time and the latest local vehicle speed extreme point; the average brake opening degree in the time interval between the current time and the latest local vehicle speed extreme point; the latest local vehicle speed extreme point speed; the maximum accelerator opening degree from the current to the latest local vehicle speed extreme point; the maximum brake opening degree from the current to the latest local vehicle speed extreme point; The first historical driving parameter includes: the cumulative vehicle speed increase in the first time period before the latest local vehicle speed extreme point speed; the cumulative vehicle speed decrease in the first time period before the latest local vehicle speed extreme point speed; the average accelerator opening degree in the first time period before the latest local vehicle speed extreme point speed; the average brake opening degree in the first time period before the latest local vehicle speed extreme point speed; The second historical driving parameter comprises: an accumulated speed increase amount in a second time period before the speed of the most recent local speed extreme point; an accumulated speed decrease amount in the second time period before the speed of the most recent local speed extreme point; an average accelerator opening degree in the second time period before the speed of the most recent local speed extreme point; an average brake opening degree in the second time period before the speed of the most recent local speed extreme point.

[0039] In some embodiments of the present application, the vehicle state comprises any one of: a starting state, a temporary stop state, a creeping state, a smooth driving state, a continuously accelerating state during driving, or a continuously decelerating state during driving.

[0040] The hidden state (driving state) space of the HMM model is defined as S={starting state s1, temporary stop state s2, creeping state s3, smooth driving state s4, continuously accelerating state during driving s5, continuously decelerating state during driving s6}.

[0041] The state observation parameter of the HMM model is defined as O={current speed o1, speed difference between the current speed and the speed of the most recent local speed extreme point o2, time interval between the current time and the most recent local speed extreme point o3, average accelerator opening degree in the time interval between the current time and the most recent local speed extreme point o4, average brake opening degree in the time interval between the current time and the most recent local speed extreme point o5, speed of the most recent local speed extreme point o6, maximum accelerator opening degree from the current to the most recent local speed extreme point o7, maximum brake opening degree from the current to the most recent local speed extreme point o8, accumulated speed increase amount in a t1 time period before the speed of the most recent local speed extreme point o9, accumulated speed decrease amount in the t1 time period before the speed of the most recent local speed extreme point o10, average accelerator opening degree in the t1 time period before the speed of the most recent local speed extreme point o11, average brake opening degree in the t1 time period before the speed of the most recent local speed extreme point o12, accumulated speed increase amount in a t2 time period before the speed of the most recent local speed extreme point o13, accumulated speed decrease amount in the t2 time period before the speed of the most recent local speed extreme point o14, average accelerator opening degree in the t2 time period before the speed of the most recent local speed extreme point o15, average brake opening degree in the t2 time period before the speed of the most recent local speed extreme point o16}.

[0042] That is, the application provides a self-adaptive recognition method of vehicle state (including starting state, temporary stopping state, creeping state, smooth running state, continuously accelerating state in running, continuously decelerating state in running and the like) based on hidden Markov model, so as to realize adjustment of the optimization target of control according to different vehicle states in the optimal control process of the assembly hybrid, and realize the consideration and coordination of economy, power performance and response continuity of the physical system of the vehicle.

[0043] In some embodiments of the application, the state observation parameter of the vehicle is determined based on the real-time vehicle speed and the moving average vehicle speed, including: A plurality of sampling anchor points are determined based on the intersection of the actual vehicle speed and the moving average vehicle speed. A target sampling anchor point closest to the sampling point is determined based on the position of the current sampling point. A feature value vector corresponding to the sampling point is determined based on the feature parameter between the target sampling anchor point and the sampling point. A feature matrix is constructed based on the feature value vectors corresponding to all sampling points. The state observation parameter is determined based on the feature matrix.

[0044] The hidden Markov model outputs the recognition result of the current hidden state based on the current input observation signal, in order to make the observation signal used for training and recognition at a time point reflect the current and historical state information of the vehicle, the application further provides a signal sampling and feature extraction method for vehicle state recognition, specifically including: (1) Extracting the moving average vehicle speed of the current vehicle speed.

[0045] Figure 2 The curve diagram of the real-time vehicle speed and the moving average vehicle speed provided by the application is shown in Figure 2 , the smooth one is the moving average vehicle speed curve, and the one with large fluctuations is the actual vehicle speed curve.

[0046] As shown in Figure 2 , the iterative calculation formula of the moving average vehicle speed is shown in the following formula:

[0047] Among them, Indicates a forgetting factor, which is calibrated according to the actual situation.

[0048] (2) Sampling anchor point definition.

[0049] In a process of starting from a card to parking, the sampling anchor point of the first feature parameter is the vehicle speed starting point, as shown in Figure 2 , which is p0 point, and the other sampling anchor points are the intersection points of the moving average vehicle speed and the actual vehicle speed, as shown in Figure 2As shown in the diagram, p2, p3, p4...

[0050] (3) Extraction and recording of feature parameters.

[0051] The parameters of a total of N+1 data points, including the most recent N sampling anchor points and the current point, are recorded in the matrix MV. Each row corresponds to a record point, and different columns are the corresponding feature parameters. The parameters of the current point pn are in the first row, followed by the feature parameters corresponding to points p3, p2, and p1, respectively.

[0052] When in Figure 2 Given the location of point pn, relative to the nearest sampling anchor point p3, calculate the current actual feature vector V={ v 1, v 2, v 3, v 4, v 5, v 6, v 7, v 8, v 9, v 11}.

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] in, v 1 represents a time interval. v 2 indicates the change in vehicle speed. v 10 represents the increase in vehicle speed. v 11 represents the decrease in vehicle speed. v 3 represents the average throttle opening. v4 represents the average brake opening, v 5 represents the maximum throttle opening, v 6 represents the maximum brake opening, v 7 represents the cumulative time of stepping on the throttle pedal, v 8 represents the cumulative time of stepping on the brake pedal, v 9 represents the current vehicle speed (i.e. Figure 2 the current vehicle speed corresponding to the point Pn).

[0065] The current feature matrix MV is recorded as follows:

[0066] When reaching the p4 point, the moving average vehicle speed line intersects with the actual vehicle speed line, and the parameters corresponding to the current point pn are the parameters of the new anchor point p4. The MV matrix is moved down one row, and the last row is discarded.

[0067] (4) Extract the parameters for vehicle state recognition.

[0068] The state observation parameters of the HMM model are defined as O = {current vehicle speed o1, vehicle speed difference between the current vehicle speed and the nearest local vehicle speed extreme point o2, time interval between the current time and the nearest local vehicle speed extreme point o3, average throttle opening in the time interval between the current time and the nearest local vehicle speed extreme point o4, average brake opening in the time interval between the current time and the nearest local vehicle speed extreme point o5, nearest local vehicle speed extreme point speed o6, maximum throttle opening from the current to the nearest local vehicle speed extreme point o7, maximum brake opening from the current to the nearest local vehicle speed extreme point o8, cumulative vehicle speed increase in the t1 time period before the nearest local vehicle speed extreme point o9, cumulative vehicle speed decrease in the t1 time period before the nearest local vehicle speed extreme point o10, average throttle opening in the t1 time period before the nearest local vehicle speed extreme point o11, average brake opening in the t1 time period before the nearest local vehicle speed extreme point o12, cumulative vehicle speed increase in the t2 time period before the nearest local vehicle speed extreme point o13, cumulative vehicle speed decrease in the t2 time period before the nearest local vehicle speed extreme point o14, average throttle opening in the t2 time period before the nearest local vehicle speed extreme point o15, average brake opening in the t2 time period before the nearest local vehicle speed extreme point o16}.

[0069] Based on the feature matrix MV, the observation parameters at the current time can be calculated as follows: a. The observation parameters extracted directly from the MV matrix include o1-o8.

[0070] o1 = MV(1,9); o2 = MV(1,2); o3 = MV(1,1); o4 = MV(1,3); o5 = MV(1,4); o6 = MV(2,4); o7 = MV(2,5); o8 = MV(2,6).

[0071] b、based on the MV to calculate from the current time before the first historical driving parameters (including o9~o12), and from the current time before the second historical driving parameters (including o13~o16) t2 time.

[0072] Cumulative time axis in the MV matrix: Time_axes = [0,cumsum(MV(:,1))]; t1 time corresponding to the position: Pos_t1= interp1(Time_axes,1:numel(Time_axes),t1); Pos1_t1 = floor(Pos_t1); Pos2_t1 = Pos_t1-floor(Pos_t1); o9=sum(MV(1: Pos1_t1,10) + Pos2_t1× MV(Pos1_t1+1,10); o10=sum(MV(1: Pos1_t1,10) + Pos2_t1× MV(Pos1_t1+1,10); o11=[sum(MV(1: Pos1_t1,5) × MV(1: Pos1_t1,1) + Pos2_t1× MV(Pos1_t1+1,5) × MV(1: Pos1_t1,1)] / t1; o12=[sum(MV(1: Pos1_t1,6) × MV(1: Pos1_t1,1) + Pos2_t1× MV(Pos1_t1+1,6) × MV(1: Pos1_t1,1)] / t1。

[0073] t2 time corresponding to the position: Pos_t2= interp1(Time_axes,1:numel(Time_axes),t2); Pos1_t2 = floor(Pos_t2); Pos2_t2 = Pos_t1-floor(Pos_t2); o13 = sum(MV(1: Pos1_t2,10) + Pos2_t1× MV(Pos1_t2+1,10); o14 = sum(MV(1: Pos1_t2,10) + Pos2_t1× MV(Pos1_t2+1,10); o15 = [sum(MV(1: Pos1_t2,5) × MV(1: Pos1_t2,1) + Pos2_t2× MV(Pos1_t2+1,5) × MV(1: Pos1_t2,1)] / t2; o16 = [sum(MV(1: Pos1_t2,6) ×MV(1: Pos1_t2,1) + Pos2_t2×MV(Pos1_t2+1,6) × MV(1: Pos1_t2,1)] / t2.

[0074] The application provides a feature parameter sampling, extraction and storage method based on the intersection of real-time vehicle speed and moving average vehicle speed, solves the problem of recording historical feature information related to vehicle state recognition using small memory, and extracts feature observation values used in hidden Markov model recognition based on the information recorded by the method, so that the HMM recognition process uses more current and historical feature information, and the model is adaptively adjusted online based on the storage data structure of the feature information.

[0075] The vehicle state recognition method for commercial vehicle hybrid control provided by the application includes vehicle current time information, short-term vehicle historical information (first historical driving parameter) and medium-long-term vehicle historical information (second historical driving parameter), so as to more accurately recognize the vehicle state.

[0076] In some embodiments of the application, the training parameters of the HMM model include a transition probability matrix of the vehicle state. The training parameters of the GMM model include a weight vector, a mean vector and a covariance matrix of each observation value.

[0077] The basic parameter training process of the hidden Markov model is as follows: (1) Processing and labeling of offline data for training.

[0078] According to the signal sampling and feature extraction method of the vehicle state recognition, the corresponding feature parameters and observation parameters of each offline data sampling point are extracted, and the vehicle state of the data before and after each sampling point is determined.

[0079] And the data is divided into different data sets according to the determined vehicle state, as follows: Data1 = {MV | State = 1}; Data2 = {MV | State = 2}; … Data6 = {MV | State = 6}.

[0080] (2) Training based on transition probability matrix A.

[0081] Transition probability matrix A is a 6x6 matrix, where each element A(i, j) represents the probability of switching from state i to state j in the whole.

[0082] a. In offline data, count the number of times from state i to state j, and record it in AN(i, j). Count all transitions in offline data and get the complete AN matrix. b. Normalize AN matrix to get transition probability matrix A.

[0083] rowSums = sum(AN, 2) A = AN / repmat(rowSums, 1, 6) AN matrix is a 6x6 matrix (because there are 6 vehicle states: starting, stopping, crawling, etc.). The meaning of AN(i, j): in all historical data, how many times does it switch from state i to state j. 2 means summing in the row direction, that is, for an MxN matrix (here 6x6), sum(AN, 2) will sum all elements of each row (i.e. the first column to the Nth column of each row).

[0084] The output result rowSums is an Mx1 column vector (i.e. 6x1), rowSums is equal to AN(i,1) + AN(i,2) +... + AN(i,6), that is, the total number of times from state i to all states.

[0085] repmat(A, M, N) copies matrix A as a "block" to generate an MxN "block array".

[0086] rowSums: is the matrix to be copied (here it is a 6x1 column vector).

[0087] 1 means copying 1 time in the vertical direction (row direction), which means that the new matrix after copying has the same number of rows as the original rowSums. 6 means copying 6 times in the horizontal direction (column direction). A new 6x6 matrix is generated, and each column of the new matrix is exactly the same as the original rowSums column vector.

[0088] Suppose rowSums = [100; 200; 50; 300; 150; 250] (a 6x1 vector), then the result of repmat(rowSums, 1, 6) is as follows: [100, 100, 100, 100, 100, 100; 200, 200, 200, 200, 200, 200; 50, 50, 50, 50, 50, 50; 300, 300, 300, 300, 300, 300; 150, 150, 150, 150, 150, 150; 250, 250, 250, 250, 250, 250] (3) Training of the probability distribution model (GMM) parameters of the observation data under various vehicle states.

[0089]

[0090] The number K of corresponding Gaussian components in the GMM model of the observation data under each vehicle state needs to be determined through iterative calibration.

[0091] The parameters of the GMM model include:

[0092] The training of the model starts with initial values .

[0093] The posterior probability of each observation value under the current parameters is calculated, and the posterior probability of each sample belonging to the kth Gaussian component is calculated:

[0094] (4) Update the GMM parameters according to the posterior probability of each sample:

[0095]

[0096]

[0097] (5) Iteratively execute the above steps until convergence.

[0098] For each state data set, perform the above process to complete the offline training of all GMM parameters.

[0099] The commercial vehicle hybrid control whole vehicle state recognition method provided by the application can better apply long-term historical data experience information based on offline data training of model basic parameters.

[0100] In some embodiments of the application, the state observation parameter is input into the trained target recognition model to obtain the vehicle state output by the target recognition model, including: The state observation parameter is input into the trained GMM model to obtain the probability distribution of the state observation parameter under different vehicle states output by the GMM model. The probability distribution and the transition probability matrix are input into the trained HMM model to obtain the vehicle state output by the HMM model.

[0101] The online vehicle state recognition process based on the hidden Markov model includes: according to the transition probability matrix A and the probability distribution of the observation parameter under each state , and the current observation vector O, when the vehicle starts to start, the possible probability distribution of each state at the corresponding time is sequentially calculated, and the state with the highest probability is taken as the current recognized vehicle state.

[0102] (1) The initial probability distribution, because the vehicle always starts from the start, so the initial probability distribution is:

[0103] (2) The current observation value , the probability of appearing under different vehicle states:

[0104]

[0105] (3) Calculate the probability of each state as the hidden state corresponding to the current time t:

[0106]

[0107] That is, the probability of the current time hidden state j = the probability distribution of each state at the last time × the probability of each state switching to j × the probability of the current feature observation value belonging to j.

[0108] The recognition state at the current time is the state with the highest probability:

[0109] The application uses a hidden Markov model of continuous observation values for vehicle state recognition of optimal control of a hybrid assembly, can use more offline identified experience knowledge through offline data training, all parameters of the hidden Markov model are limited and have explicit meanings, and the training, explanation and adjustment of the model are relatively easy.

[0110] The application can be applied to a hybrid assembly control system based on vehicle states to perform hybrid assembly mode, engine start-stop, gearbox gear, assembly drive distribution.

[0111] In some embodiments of the application, after the state observation parameter is input into the trained target recognition model, the vehicle state output by the target recognition model is obtained, and the method further comprises: Based on the vehicle state output by the target recognition model, the training parameters of the HMM model and the training parameters of the GMM model are adaptively adjusted.

[0112] In the sampling method of the application, the historical records are mainly feature anchors near local extreme points of the vehicle, when the data of a period of time before and after the corresponding anchor point are known states, the state to which the anchor point belongs can be intuitively judged based on rules (the judgment rules are the same as the rules required for clutch calibration).

[0113] If the vehicle state judged after the event is inconsistent with the state judged in real time online on the corresponding anchor point, the feature transition matrix A in the HMM model is activated using the vehicle state judged after the event on the corresponding anchor point and the features recorded in the MV. The corresponding vehicle state observation values of all anchors are used as new sample points to update the probability distribution GMM model parameters of the observation values.

[0114] Suppose that the actual state on an anchor point is k→i, the state confirmed after the event is k→j, and the corresponding observation parameter is vector Ok.

[0115] (1) Update of the transition matrix A: A(k,j) = A(k,j) + △P A(k,j) = A(k,i) – △P After adjustment, it still satisfies: sum(A(k,:)) = 1, wherein △P is a calibration value.

[0116] (2) Update of the observation probability density model GMM j parameters of state j.

[0117] (a) Based on the original parameters odd, calculate the posterior probability that the new observation value Ok belongs to the kth Gaussian component:

[0118] (b) Update the relevant parameters of the GMM based on the posterior probability of the new observation Ok:

[0119]

[0120]

[0121] Because the parameters of a Hidden Markov Model are limited and each has a clear meaning, this invention develops an online adaptive adjustment method for the relational parameters in the model, enabling the recognition model to quickly understand different road and driver conditions.

[0122] This invention proposes a vehicle state adaptive recognition method based on a Hidden Markov Model (HMM) for deploying vehicle state recognition systems on existing commercial vehicles for optimal control of hybrid powertrains. The method involved in this invention comprises three parts: (1) A method for signal sampling and feature extraction for vehicle state recognition, providing the recognition model with process information including different time horizons.

[0123] (2) A state recognition method based on a hidden Markov model, including a Gaussian mixture model that describes the probability density of the distribution of each feature parameter under different vehicle states.

[0124] The hidden state (driving state) space of the HMM model is defined as S, and the state observation parameters of the HMM model are defined as O.

[0125] (3) Based on the information extracted and saved in the first part, the historical sampling points of the vehicle state can be clearly determined by using the data before and after the relevant points, which are used to update the hidden state (driving state) transition matrix A in the HMM model, and the corresponding weights, mean and covariance matrix in the GMM model are updated by using the feature parameters on the corresponding sampling points.

[0126] The method of the present invention includes the following three steps in use: S1, Training of the model's basic parameters.

[0127] Using labeled driving state sequences and corresponding observation sequences, the basic parameters of the offline training model are obtained, including the hidden state (driving state) transition matrix A (6×6) of the Hidden Markov Model, and the GMM model parameters describing the probability density of each observation under various states (including the weight vector, mean vector and covariance matrix of each component).

[0128] Since state recognition is not required when the vehicle is stationary, the vehicle state at the start of the recognition system is the starting state. That is, the initial probability distribution of the HMM model is [1, eps, eps, eps, eps, eps]. No training is required because data overflow will occur during probability calculation. Instead, logarithmic calculation is used, and a minimum value is used to replace the 0 probability.

[0129] S2, Online use of the model.

[0130] Using the Hidden Markov Model (HMM), the vehicle state with the highest probability in the current and future periods is identified based on the extracted current and historical feature parameters.

[0131] S3. Adaptive adjustment of model parameters.

[0132] For historical sampling points where the vehicle state can be clearly determined based on the previous and subsequent states, if the results differ from the HMM recognition results, the transition matrix A and related parameters of the GMM model are adaptively adjusted based on the state and feature parameters of the sampling points.

[0133] The present invention provides a vehicle state recognition method for hybrid commercial vehicle control. This method is based on a Hidden Markov Chain model with smaller parameter scale and clearly defined parameters to identify the vehicle state of the hybrid commercial vehicle control system. It also proposes a method for sampling, extracting, and storing relevant identification parameters and features. Without increasing the computing power and storage capacity of existing vehicle control systems, the following technical effects are achieved: (1) The basic parameters of the model are trained based on offline data, which makes better use of the experience information of long-term historical data.

[0134] (2) The feature parameters used for identification include the current time of the vehicle, the short-term vehicle historical signal, and the medium- and long-term vehicle historical information, so as to more accurately identify the vehicle status.

[0135] (3) Based on the clear scale and meaning of the model parameters of the present invention, the historical information stored in the structure realizes the adaptive update of the model, so as to adjust the model to adapt to and generalize different road scenarios and driver driving characteristics.

[0136] To better implement the vehicle state recognition method for commercial vehicle hybrid control in this embodiment of the invention, based on the existing method, the corresponding implementation is as follows: Figure 3 As shown, this embodiment of the invention also provides a vehicle status identification device for hybrid control of commercial vehicles. The vehicle status identification device 300 for hybrid control of commercial vehicles includes: The determining unit 301 is configured to determine a state observation parameter of the vehicle based on a real-time vehicle speed and a moving average vehicle speed of the vehicle; the state observation parameter comprises a real-time driving parameter of the vehicle at a current time, a first historical driving parameter in a first time period before the current time, and a second historical driving parameter in a second time period before the current time; the first time period is less than the second time period. The identifying unit 302 is configured to input the state observation parameter into a trained target identification model to obtain a vehicle state output by the target identification model; the target identification model comprises an HMM model and a GMM model, and the target identification model is trained based on an annotated state sequence and an observation sequence corresponding to the annotated state sequence.

[0137] The commercial vehicle hybrid control whole vehicle state identification device 300 provided by the above embodiments can implement the technical solutions described in the commercial vehicle hybrid control whole vehicle state identification method embodiments, and the principles of the implementation of the above modules or units can be referred to the corresponding content in the commercial vehicle hybrid control whole vehicle state identification method embodiments, which will not be described here.

[0138] As shown in Figure 4 The present application also correspondingly provides an electronic device 400. The electronic device 400 comprises a processor 401, a memory 402 and a display 403. Figure 4 Only part of the components of the electronic device 400 are shown, but it should be understood that all the shown components are not required to be implemented, and more or less components can be alternatively implemented.

[0139] The processor 401 can be a central processing unit (CPU), a microprocessor or other data processing chip in some embodiments, used to run the program code or process data stored in the memory 402, such as the commercial vehicle hybrid control whole vehicle state identification method in the present application.

[0140] In some embodiments, the processor 401 can be a single server or a group of servers. The group of servers can be centralized or distributed. In some embodiments, the processor 401 can be local or remote. In some embodiments, the processor 401 can be implemented in a cloud platform. In some embodiments, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multiple cloud, etc., or any combination of the above.

[0141] The memory 402 can be an internal storage unit of the electronic device 400, such as a hard disk or a memory of the electronic device 400 in some embodiments. The memory 402 can also be an external storage device of the electronic device 400, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device 400 in other embodiments.

[0142] Further, the memory 402 can include both an internal storage unit and an external storage device of the electronic device 400. The memory 402 is used to store application software installed on the electronic device 400 and various types of data.

[0143] The display 403 can be an LED display, a liquid crystal display, a touch liquid crystal display, an Organic Light-Emitting Diode (OLED) touch, etc. in some embodiments. The display 403 is used to display information of the electronic device 400 and to display a visualized user interface. The components 401-403 of the electronic device 400 communicate with each other through a system bus.

[0144] In an embodiment, when the processor 401 executes the whole vehicle state recognition program of commercial hybrid vehicle control in the memory 402, the following steps can be implemented: Determine a state observation parameter of the vehicle based on a real-time vehicle speed and a moving average vehicle speed of the vehicle; the state observation parameter includes a real-time driving parameter of the vehicle at a current time, a first historical driving parameter in a first time period before the current time, and a second historical driving parameter in a second time period before the current time; the first time period is less than the second time period; Input the state observation parameter into a trained target recognition model to obtain a vehicle state output by the target recognition model; the target recognition model includes an HMM model and a GMM model, and the target recognition model is trained based on a labeled state sequence and an observation sequence corresponding to the labeled state sequence.

[0145] It should be understood that, when the processor 401 executes the whole vehicle state recognition program of commercial hybrid vehicle control in the memory 402, in addition to the above functions, it can also implement other functions, which can be referred to the description of the corresponding method embodiments.

[0146] Further, the embodiments of the present application do not make specific limitation on the type of the electronic device 400 mentioned above, and the electronic device 400 can be a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, or the like. Exemplary embodiments of the portable electronic device include, but are not limited to, a portable electronic device running an IOS, an android, a microsoft, or other operating system. The portable electronic device described above can also be other portable electronic devices, such as a laptop computer having a touch-sensitive surface (e.g., a touch panel), and the like. It should also be understood that in some other embodiments of the present application, the electronic device 400 can also not be a portable electronic device, but a desktop computer having a touch-sensitive surface (e.g., a touch panel).

[0147] Correspondingly, the embodiments of the present application also provide a computer readable storage medium for storing computer readable programs or instructions, which, when executed by a processor, can implement the steps or functions in the whole vehicle state identification method for commercial hybrid control provided by the above-mentioned method embodiments.

[0148] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiments can be completed by a computer program instructing relevant hardware (such as a processor, a controller, etc.) to complete, and the computer program can be stored in a computer readable storage medium. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory, or a random access memory, etc.

[0149] The above describes in detail the whole vehicle state identification method for commercial hybrid control, the device and the storage medium provided by the present application. The principle and implementation mode of the present application are described by applying specific examples in this paper. The above-mentioned embodiments are only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed; in view of the above, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A commercial vehicle hybrid control whole vehicle state recognition method, characterized in that, The method comprises the following steps: determining a state observation parameter of the vehicle based on a real-time vehicle speed and a moving average vehicle speed of the vehicle; the state observation parameter comprises a real-time driving parameter of the vehicle at a current time, a first historical driving parameter within a first time period before the current time, and a second historical driving parameter within a second time period before the current time; the first time period is less than the second time period; inputting the state observation parameter into a trained target recognition model to obtain a vehicle state output by the target recognition model; the target recognition model comprises an HMM model and a GMM model, and the target recognition model is trained based on a labeled state sequence and an observation sequence corresponding to the labeled state sequence.

2. The method of claim 1, wherein the method further comprises: The method for determining the state observation parameter of the vehicle based on the real-time vehicle speed and the moving average vehicle speed of the vehicle comprises the following steps: determining a plurality of sampling anchor points based on the intersection of the actual vehicle speed and the moving average vehicle speed; determining a target sampling anchor point closest to a current sampling point based on the position of the sampling point; determining a feature value vector corresponding to the sampling point based on a feature parameter between the target sampling anchor point and the sampling point; constructing a feature matrix based on the feature value vectors corresponding to all sampling points; determining the state observation parameter based on the feature matrix.

3. The method of claim 1, wherein the method further comprises: The training parameters of the HMM model comprise a transition probability matrix of the vehicle state. The training parameters of the GMM model comprise a weight vector, a mean vector, and a covariance matrix of each observation value.

4. The method of claim 3, wherein, The method for inputting the state observation parameter into the trained target recognition model to obtain the vehicle state output by the target recognition model comprises the following steps: inputting the state observation parameter into the trained GMM model to obtain a probability distribution of the state observation parameter under different vehicle states output by the GMM model; inputting the probability distribution and the transition probability matrix into the trained HMM model to obtain the vehicle state output by the HMM model.

5. The commercial vehicle hybrid control overall vehicle state recognition method according to claim 4, characterized in that, After the step of inputting the state observation parameter into the trained target recognition model to obtain the vehicle state output by the target recognition model, the method further comprises the following step: adaptively adjusting the training parameters of the HMM model and the training parameters of the GMM model based on the vehicle state output by the target recognition model.

6. The method of claim 1, wherein: The real-time driving parameter comprises: a current vehicle speed; a vehicle speed difference between the current vehicle speed and a nearest local vehicle speed extreme point; a time interval between the current time and the nearest local vehicle speed extreme point; an average accelerator opening degree within the time interval between the current time and the nearest local vehicle speed extreme point; an average brake opening degree within the time interval between the current time and the nearest local vehicle speed extreme point; a nearest local vehicle speed extreme point speed; a maximum accelerator opening degree from the current to the nearest local vehicle speed extreme point; a maximum brake opening degree from the current to the nearest local vehicle speed extreme point. The first historical driving parameter comprises: a cumulative vehicle speed increase within the first time period before the nearest local vehicle speed extreme point speed; a cumulative vehicle speed decrease within the first time period before the nearest local vehicle speed extreme point speed; an average accelerator opening degree within the first time period before the nearest local vehicle speed extreme point speed; an average brake opening degree within the first time period before the nearest local vehicle speed extreme point speed. The second historical driving parameter comprises: an accumulated speed increase amount in a second time period before a speed of a most recent local speed extreme point; an accumulated speed decrease amount in the second time period before the speed of the most recent local speed extreme point; an average accelerator opening degree in the second time period before the speed of the most recent local speed extreme point; an average brake opening degree in the second time period before the speed of the most recent local speed extreme point.

7. The method of claim 1, wherein: The vehicle state comprises any one of: a starting state, a temporary stop state, a crawling state, a smooth driving state, a continuously accelerating state during driving, or a continuously decelerating state during driving.

8. A commercial vehicle hybrid control whole vehicle state recognition device, characterized by, The method comprises: determining a state observation parameter of the vehicle based on a real-time speed of the vehicle and a moving average speed of the vehicle; the state observation parameter comprises a real-time driving parameter of the vehicle at a current time, a first historical driving parameter in a first time period before the current time, and a second historical driving parameter in a second time period before the current time; the first time period is less than the second time period; inputting the state observation parameter into a trained target recognition model to obtain a vehicle state output by the target recognition model; the target recognition model comprises an HMM model and a GMM model, and the target recognition model is trained based on a labeled state sequence and an observation sequence corresponding to the labeled state sequence.

9. An electronic device, comprising: comprise a memory and a processor, wherein the memory is configured to store a program; the processor is coupled to the memory and is configured to execute the program stored in the memory to implement the steps of the commercial vehicle hybrid control whole vehicle state recognition method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer readable program or instruction is stored, and the program or instruction is executed by a processor to implement the steps of the commercial vehicle hybrid control whole vehicle state recognition method according to any one of claims 1 to 7.

Images