CN111547028A - Brake intensity fuzzy recognition method considering brake intention - Google Patents
Brake intensity fuzzy recognition method considering brake intention Download PDFInfo
- Publication number
- CN111547028A CN111547028A CN202010312924.0A CN202010312924A CN111547028A CN 111547028 A CN111547028 A CN 111547028A CN 202010312924 A CN202010312924 A CN 202010312924A CN 111547028 A CN111547028 A CN 111547028A
- Authority
- CN
- China
- Prior art keywords
- braking
- brake
- intention
- output
- braking intention
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000006073 displacement reaction Methods 0.000 claims abstract description 94
- 230000008859 change Effects 0.000 claims abstract description 24
- 238000002474 experimental method Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 45
- 238000012360 testing method Methods 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 18
- 238000012935 Averaging Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 6
- 238000007476 Maximum Likelihood Methods 0.000 claims description 4
- 238000007689 inspection Methods 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 description 20
- 238000011084 recovery Methods 0.000 description 4
- 238000012795 verification Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000013139 quantization Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60T—VEHICLE BRAKE CONTROL SYSTEMS OR PARTS THEREOF; BRAKE CONTROL SYSTEMS OR PARTS THEREOF, IN GENERAL; ARRANGEMENT OF BRAKING ELEMENTS ON VEHICLES IN GENERAL; PORTABLE DEVICES FOR PREVENTING UNWANTED MOVEMENT OF VEHICLES; VEHICLE MODIFICATIONS TO FACILITATE COOLING OF BRAKES
- B60T13/00—Transmitting braking action from initiating means to ultimate brake actuator with power assistance or drive; Brake systems incorporating such transmitting means, e.g. air-pressure brake systems
- B60T13/74—Transmitting braking action from initiating means to ultimate brake actuator with power assistance or drive; Brake systems incorporating such transmitting means, e.g. air-pressure brake systems with electrical assistance or drive
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60T—VEHICLE BRAKE CONTROL SYSTEMS OR PARTS THEREOF; BRAKE CONTROL SYSTEMS OR PARTS THEREOF, IN GENERAL; ARRANGEMENT OF BRAKING ELEMENTS ON VEHICLES IN GENERAL; PORTABLE DEVICES FOR PREVENTING UNWANTED MOVEMENT OF VEHICLES; VEHICLE MODIFICATIONS TO FACILITATE COOLING OF BRAKES
- B60T17/00—Component parts, details, or accessories of power brake systems not covered by groups B60T8/00, B60T13/00 or B60T15/00, or presenting other characteristic features
- B60T17/18—Safety devices; Monitoring
- B60T17/22—Devices for monitoring or checking brake systems; Signal devices
Landscapes
- Engineering & Computer Science (AREA)
- Transportation (AREA)
- Mechanical Engineering (AREA)
- Regulating Braking Force (AREA)
Abstract
The invention provides a brake intensity fuzzy recognition method considering brake intentions, which comprises the steps of firstly obtaining brake pedal data through experiments, classifying according to the brake intentions by taking the brake intensity as a standard, and performing parameter offline recognition on an HMM model by using a method of 'recognizing while verifying' to obtain HMM model parameters of each brake intention at each output moment; then, according to the brake pedal displacement time sequence obtained in real time, the method of 'time-sharing output, reasonable prediction and comprehensive judgment' is adopted to identify the brake intention on line, and the brake pedal displacement and the brake pedal force are further used as the input of a first-layer fuzzy controller to output a brake intention coefficient; and finally, taking the braking intention coefficient and the displacement change rate of the brake pedal as the input of a second-layer fuzzy controller, and outputting the braking strength to obtain the estimated braking strength. The invention can accurately reflect the braking intention of the driver in real time and adapt to the variable braking intention of the driver.
Description
Technical Field
The invention belongs to the technical field of automobile driving, and particularly relates to a brake intensity fuzzy identification method considering a brake intention.
Background
For electric vehicles, the adoption of a braking energy recovery strategy has become an important method for effectively increasing the driving range of the electric vehicles. The braking energy recovery efficiency of an electric vehicle is affected by various factors, wherein the braking strength is an important factor which is not negligible. Within a certain range, the greater the braking strength, the greater the braking force required, and the greater the braking energy available for recovery. Meanwhile, due to the introduction of motor braking, in order to ensure the consistency of braking feeling during braking, a brake pedal is generally decoupled from a service braking system, and the braking force required by a driver cannot be directly generated due to the change of the displacement of the brake pedal at the moment, so that the braking strength required by the driver needs to be obtained by researching the relation between the displacement of the brake pedal and the braking intention, and then the electro-hydraulic braking force is reasonably distributed according to the braking strength, so that the energy recovery efficiency is improved. Therefore, it is important to accurately estimate the braking strength.
Meanwhile, accurate estimation of the braking strength and the braking intention is also of great significance to the development of the brake-by-wire technology. The brake controller enables the brake actuator to brake according to the brake control signal sent by the electric control unit, so that the braking process is more efficient and energy-saving. However, the existing brake-by-wire system does not analyze the relationship between the braking intention and the braking strength, and the actual braking strength of the braking system does not better conform to the real intention of the driver.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a brake intensity fuzzy recognition method considering the brake intention, which accurately reflects the brake intention of a driver in real time.
The present invention achieves the above-described object by the following technical means.
A brake intensity fuzzy recognition method considering brake intentions is characterized in that brake pedal data of different brake intentions are obtained through experiments, and the brake pedal data are classified according to the brake intentions by taking the brake intensity as a standard; identifying HMM model parameters of each braking intention at each output moment in an off-line manner; identifying a braking intention on line according to a brake pedal displacement time sequence acquired in real time; taking the displacement of a brake pedal and the force of the brake pedal as the input of a first-layer fuzzy controller, and outputting a braking intention coefficient; and (5) taking the braking intention coefficient and the displacement change rate of the brake pedal as the input of the fuzzy controller of the second layer, and outputting the braking strength.
Further, the offline recognition of HMM model parameters of each braking intention at each output moment specifically includes:
acquiring complete data for an HMM model parameter identification process according to experimental data, initializing model parameters when a training set in first complete data is used, calculating model parameters of each group of pedal displacement time sequences in the training set by adopting an EM (effective magnetic resonance) algorithm, and averaging the model parameters to obtain model parameters of the training set; calculating likelihood probability by using model parameters obtained by a training set for each group of pedal displacement time sequences in the test set, averaging, and if the average likelihood probability is smaller than a threshold value Pmin1, re-initializing the model parameters to calculate the model parameters until the average likelihood probability of the test set is not smaller than a threshold value Pmin 1; performing model parameter iteration by using the next complete data, and averaging the initialized model parameters of each subsequent training set by using the last training set to obtain model parameters; and the model parameters output by the last training set are used as the HMM model parameters of the braking intention at the output moment.
Further, the model parameters include an initial time state probability column vector pi, a state transition matrix A and an observation probability matrix B.
Further, the complete data acquisition process includes:
the displacement time sequence of W groups of brake pedals corresponding to a certain braking intention at a certain output time is M, M-1 parts of data are used as a training set each time, and 1 part of data are used as an inspection set to form complete data; the different training and test sets were combined into M complete data.
Further, the HMM model input in the online brake intention recognition is: brake pedal displacement time series O obtained in real time1,O2,…OnThe output is: pedal displacement time series at corresponding outputLikelihood probability P (O) of each braking intention at timen|λn(x) Where N is 1, 2, 3 … N, N denotes the output time number, OnA pedal displacement time series showing the nth output time, x ═ a, b, c, d, respectively, and represents "microstep", "light step", "middle step", and "heavy step", λn(x) HMM model parameters representing the braking intention x at the nth output time; averaging the likelihood probability of each braking intent over an identification time
Further, when a brake intention condition that can output a prediction is satisfied: that is, the output time number n is not less than the number of predictable output times Npre, and the maximum likelihood probability average value at the output timeIf the value is larger than the threshold value Pmin2, outputting the predicted braking intention x; calculating the likelihood probability average value of each braking intention after prediction output When n is equal to the maximum output time Np within one brake intensity recognition period, the maximum value of the predicted likelihood probability average value of each brake intention after output needs to be calculatedAnd comparing the corresponding braking intention with the predicted braking intention x, and outputting the final braking intention after comprehensive judgment.
Further, the comprehensive judgment comprises:
(1)the corresponding braking intention is consistent with the predicted braking intention, and the predicted braking intention is used as a final braking intention;
(2)the corresponding braking intent is not consistent with the predicted braking intent,the corresponding braking intention and the predicted braking intention are used for subsequent fuzzy recognition to output the braking strength, and the average value of the two braking strengths is taken as the final output braking strength;
(3) without outputting predicted braking intention, takeThe corresponding braking intention serves as the final braking intention.
Further, the brake pedal data are a brake pedal displacement time series, a brake pedal force and a brake pedal displacement change rate.
The invention has the beneficial effects that:
(1) when the HMM model parameters of the braking intention are identified in an off-line mode, an average likelihood probability threshold value Pmin1 is set and used for guaranteeing the accuracy of the model parameters.
(2) The displacement time sequence of W groups of brake pedals corresponding to a certain braking intention at a certain output time is M, M-1 parts of data are used as a training set every time, and 1 part of data are used as an inspection set to form complete data; combining different training sets and test sets into M complete data; the initialized model parameters of the next complete data training set all use the model parameters output by the previous training set, so that the iteration speed is increased, and the accuracy of the model parameters is improved.
(3) The invention reasonably predicts the braking intention when identifying the braking intention on line, namely the output time sequence number n is not less than the predicted output time number Npre, and the maximum likelihood probability average value under the output timeGreater than the threshold value Pmin2, the predicted braking intention x is output for identifying the braking intensity beforehand,the identification speed of the braking strength is accelerated; and calculating the likelihood probability average value of each braking intention after prediction outputAfter comprehensive judgment, outputting a final braking intention so as to adapt to the changeable braking intention of the driver, and enabling the identified braking intention to be more accurate;
(4) according to the invention, the first-layer fuzzy controller takes the displacement of a brake pedal and the force of the brake pedal as input, takes a braking intention coefficient as output, represents the change degree of the braking intention of a driver, and reduces the braking intensity range;
(5) the invention takes the real-time brake pedal displacement time sequence as the basis for identifying the braking intention, and can reflect the braking intention of the driver in real time.
Drawings
FIG. 1 is a flow chart of a brake intensity fuzzy identification method considering brake intention according to the present invention;
FIG. 2 is a flow chart of the HMM model parameter off-line recognition for a "micro-step" braking intention at the 3 rd output time according to the present invention;
FIG. 3 is a flow chart of online identification of braking intention in the 1 st braking intensity identification period according to the present invention;
FIG. 4 is a schematic diagram illustrating HMM model parameter offline recognition of each braking intention at each output moment and online recognition of braking intention in each braking intensity recognition period according to the present invention;
FIG. 5 is a graph of membership functions as used in the present invention.
Detailed Description
The invention will be further described with reference to the following figures and specific examples, but the scope of the invention is not limited thereto.
A brake intensity fuzzy identification method considering a brake intention specifically comprises the following steps:
the method comprises the following steps: in the real vehicle experiment, the whole vehicle controller acquires W groups of brake pedal data and driving state data of different braking intentions by a sensor, wherein the brake pedal data is classified according to the braking intentions by taking the braking intensity as a standard; the brakeThe pedal data is: the system comprises a brake pedal displacement time sequence, a brake pedal force and a brake pedal displacement change rate, wherein the brake pedal displacement time sequence is determined by data transmitted to a vehicle control unit by a brake pedal displacement sensor, the brake pedal force is acquired by a force sensor, and the brake pedal displacement change rate is obtained by brake pedal displacement difference; the driving state is braking deceleration and is obtained by a deceleration sensor; relationship between braking intensity and braking deceleration:wherein Z is the braking strength, g is the acceleration of gravity,is the braking deceleration; the braking intentions include microsteps, light steps, medium steps, and heavy steps.
And preprocessing (such as Kalman filtering) the brake pedal displacement time sequence, eliminating abnormal values and reducing the influence of measurement errors on subsequent model parameter offline identification.
Step two: HMM model (hidden Markov model) parameters of the braking intention are identified off-line.
Determining an identification time length T according to the time length of the brake pedal displacement time sequence obtained in the first step, and determining an identification time interval T according to the sampling frequency of the brake pedal displacement sensor; the number of output moments of a certain braking intention is determined to be N according to the recognition time length T and the recognition time interval T, 4 groups of HMM model parameters exist at each output moment, and 4N groups of HMM model parameters are required to be recognized. And respectively carrying out off-line recognition on HMM model parameters of the four braking intentions at each output moment, wherein the HMM model parameters comprise an initial moment state probability column vector pi, a state transition matrix A and an observation probability matrix B.
Specifically, the HMM model parameter offline identification process of a certain braking intention at a certain output time is as follows:
and M parts of the W groups of brake pedal displacement time sequences corresponding to the brake intention at the output time, wherein each part of data comprises W groups of data, and W is equal to M.w. The method of 'identification and verification' is adopted, M-1 parts of data are used as a training set and 1 part of data are used as a checking set each time to form a complete data for a model parameter iteration process, different training sets and checking sets are combined to form M complete data, and each group of pedal displacement time sequences in the checking set in each complete data are not repeated.
When model parameter calculation is carried out, the input of the EM algorithm is a pedal displacement time sequence in a training set, and the output is HMM model parameters of a braking intention; when model parameter verification is carried out, the HMM model of the braking intention is input into a pedal displacement time sequence in the test set, and the output is a likelihood probability average value. When a training set in the first complete data is used, initializing model parameters, calculating the model parameters of each group of pedal displacement time sequences in the training set by adopting an EM (expectation-maximization) algorithm, and averaging w x (M-1) obtained model parameters to obtain the model parameters obtained by the training set; calculating likelihood probability for each group of pedal displacement time sequences in the test set by using model parameters obtained by the training set, and averaging w likelihood probabilities; if the average likelihood probability is smaller than the threshold Pmin1, model parameters need to be reinitialized for model parameter calculation until the average likelihood probability of the test set is not smaller than the threshold Pmin1, and a complete model parameter iteration process is finished. And continuing to use the next complete data to carry out model parameter iteration, wherein the initialization model parameters of each subsequent training set all use the model parameters output by the previous training set, so that the iteration speed is accelerated. Meanwhile, the average likelihood probability under each test set is not less than the threshold value Pmin1, otherwise, the model parameters need to be recalculated from the first complete data. When all the complete data are used, the complete model parameter identification process is finished. And taking the model parameter output by the last training set as the HMM model parameter of the braking intention at the output moment, wherein the performance of the HMM model parameter is represented by the average likelihood probability under the corresponding test set, and the higher the average likelihood probability value is, the more accurate the model parameter is relatively. The same method can obtain the HMM model parameters of other braking intentions at the output moment and the HMM model parameters of each braking intention at the output moment.
Step three: and identifying the braking intention on line.
The recognition period Tp of the brake intensity is determined by the sampling frequency of the deceleration sensor, and the number of output times Np in one brake intensity recognition period satisfies t.Np ═ Tp. The HMM model input when the braking intention is identified on line is as follows: brake pedal displacement time series O obtained in real time1,O2,…OnN is 1, 2, 3 … N, N indicates an output time number, OnA pedal displacement time series representing the nth output time, the output being: likelihood probability P (O) of each braking intention of pedal displacement time series at corresponding output timen|λn(x) Where x ═ a, b, c, d, respectively, represent "microstep", "light step", "medium step", and "heavy step", λ ═ a, b, c, d, respectivelyn(x) HMM model parameters representing the braking intention x at the nth output time; simultaneously, the likelihood probability of each braking intention in the recognition time is averaged to obtainWhen a brake intention condition that can output a prediction is satisfied: that is, the output time number is not less than the number Npre of predictable output times, and the maximum likelihood probability average value at the output timeAbove the threshold Pmin2, the predicted braking intent x may be output; while continuously calculating the likelihood probability average value of each braking intention after prediction outputWhen n is equal to the maximum output time Np within one braking intensity recognition period Tp, the maximum value of the predicted likelihood probability average value of each braking intention after output needs to be predictedComparing the corresponding braking intention with the predicted braking intention x, outputting the final braking intention after comprehensive judgment, wherein the comprehensive judgment comprises three conditions:
(1) braking device with output predictionWithout significant change in braking intent after map and prediction of output, i.e.The corresponding braking intention is consistent with the predicted braking intention, and the predicted braking intention is used as the braking intention finally used for estimating the braking strength.
(2) With output of predicted braking intent but with a significant change in predicted braking intent, i.e. with output of predictionAnd if the corresponding braking intention is not consistent with the predicted braking intention, both the two braking intentions are used for subsequent fuzzy recognition and the braking strength is output, and the average value of the two braking strengths is taken as the final output braking strength.
(3) Without outputting the predicted braking intention, takeThe corresponding braking intention serves as the braking intention which is ultimately used for estimating the braking intensity.
Step four: the brake intention factor is identified in a fuzzy manner. Under the same braking intention, the expression degree of the braking intention is different, and the corresponding braking strength value is different. The first layer fuzzy controller takes the brake pedal displacement and the brake pedal force as input and takes the brake intention coefficient as output. The braking intention coefficient is in the range of [0, 1], and a value closer to 1 means that the braking intention is more strongly expressed and the braking strength is relatively larger.
Step five: and (5) fuzzy recognition of braking strength. The intensity of braking is related to the rate of change of brake pedal displacement, in addition to the intent of braking. Under the same braking intention coefficient, the larger the pedal displacement change rate is, the more urgent the braking is, and the required braking strength is larger. The second level fuzzy controller thus takes as input the brake intent factor and the rate of change of brake pedal displacement and as output the brake strength.
Examples
Firstly, acquiring and preprocessing experimental data:
(1) the experimenter: five male and female testers of 10 different ages are selected to complete the experiment.
(2) Experimental tools: electric automobile, displacement sensor, force sensor and deceleration sensor.
(3) The experimental contents are as follows: each experimenter drives the automobile to complete the following four braking processes with braking intentions: micro-stepping, light-stepping, middle-stepping and heavy-stepping to obtain the displacement time sequence of the brake pedal, the displacement change rate of the brake pedal, the force of the brake pedal and the braking deceleration under four braking intentions, and classifying according to the braking intentions by taking the braking intensity as a standard (as shown in table 1); and eliminating abnormal values of the brake pedal displacement time series. Each experimenter repeatedly completes 20 experiments on the braking process of each braking intention, and finally 200 groups of braking intention experimental data exist at each output moment.
TABLE 1 relationship between braking intention and braking intensity
Intention of braking | Micro-treading | Light stepping | Middle step | Heavy tread |
Strength of |
0≤Z<0.1 | 0.1≤Z<0.4 | 0.4≤Z<0.7 | 0.7≤ |
Secondly, identifying HMM model parameters of braking intention in an off-line manner
Determining that the recognition time length T is 0.1s, the recognition time interval T is 0.0005s, and the recognition period Tp of the braking intensity is 0.001s, wherein the number of output moments N is 200, the number of output moments Np in one braking intensity recognition period is 20, and each output moment has 4 groups of HMM model parameters to be recognized, so that 800 HMM model parameters are required to be recognized; let the average likelihood probability threshold value Pmin1 be 0.7 for each test set. At each output time, 200 groups of brake pedal displacement time sequence data are collected for each braking intention, and are divided into 10 parts, and 20 groups of data are provided for each part. And taking 9 parts of data as a training set and 1 part of data as a test set so as to obtain complete data, wherein 10 complete data are obtained, and data in the test set in the 10 complete data are not repeated.
Specifically, the HMM model parameter recognition of the "micro-stepping" braking intention at the 3 rd output time is taken as an example for specific description: when calculating model parameters, inputting pedal displacement time sequence O3=(O3-1,O3-1,…O3-200) Outputting HMM model identification parameters; wherein O is3-s=(o1,o2,…ou),s=1,2,…200,ou∈(v1,v2,…vk),ouIs an element of the pedal displacement time series, vkIs a possible value of the pedal displacement time series element. The specific process is as follows with reference to the attached figure 2:
firstly, a training set in first complete data is used for model parameter calculation, and a model parameter lambda is initialized3-1-1 (0)Inputting a first set of pedal displacement time series O in a first complete training set of data3-1-1Using EM algorithm to parameter aij、bj(k) And piiPerforming a calculation (a)ij、bj(k) And piiRespectively as elements in pi, a and B) to obtain a model parameter lambda output by the first calculation3-1-1 (1)=(A(1),B(1),π(1)) Then calculating the pedal displacement time sequenceColumn O3-1-1Likelihood probability P (O) under model parameters of first calculation output3-1-1|λ3-1-1 (1)) If it is greater than the pedal displacement time series O3-1-1At an initialization parameter lambda3-1-1 (0)Likelihood probability of lower P (O)3-1-1|λ3-1-1 (0)) Then is at O3-1-1And λ3-1-1 (1)Continuing to calculate model parameters and updating the model parameters to lambda by using an EM algorithm3-1-1 (2)=(A(2),B(2),π(2)) Then calculating the pedal displacement time series O3-1-1Likelihood probability P (O) under model parameters output by second calculation3-1-1|λ3-1-1 (2)) If it is greater than the pedal displacement time series O3-1-1Model parameter lambda output in last calculation3-1-1 (1)Likelihood probability of lower P (O)3-1-1|λ3-1-1 (1)) Then is at O3-1-1And λ3-1-1 (2)Continuing to calculate model parameters and updating the model parameters to lambda by using an EM algorithm3-1-1 (3)=(A(3),B(3),π(3)) (ii) a Calculating model parameters until the m-th time and updating the model parameters to be lambda3-1-1 (m)=(A(m),B(m),π(m)) Time sequence of pedal displacement O3-1-1Likelihood probability P (O) under the model parameters3-1-1|λ3-1-1 (m)) Not greater than observation sequence O3-1-1Model parameter lambda output in last calculation3-1-1 (m-1)Likelihood probability of lower P (O)3-1-1|λ3-1-1 (m-1)) At this time, the first group of data in the first complete data training set is used up, and the model parameter lambda is obtained3-1-1=(A(m-1),B(m-1),π(m-1)). Followed by a pedal displacement time series O3-1-2Repeating the calculation process of the model parameters of the previous group of data to obtain the model parameters lambda as the pedal displacement time sequence3-1-2(ii) a When all the 180 pedal displacement time sequences are used, the average value of each model parameter can be calculatedWherein "r" represents the serial number of the pedal displacement time sequence in the training set, r is 1, 2, … 180, and a represents the intention of 'micro-stepping' brake.
Then, each group of pedal displacement time sequence in the first complete data test set is used for model parameter verification, and likelihood probability P (O) is calculated3-1-h|λ3-1(a) (h denotes the number of pedal displacement time series in the test set, h 181, 182, … 200), and averagingAnd ending the complete model parameter iteration process. Then, the likelihood probability average value of the test set is judged if P (O)3-1|λ3-1(a) Is not more than 0.7), the initialization parameter is needed to use the first complete data training set to complete a complete model parameter iteration process again until the likelihood probability average value is more than 0.7, and the model parameter lambda under the first training set is obtained3-1(a) And the first perfect data use ends. Followed by a training set O using second complete data3-2-rAnd test set O3-2-hCompleting model parameter iteration process, and using model parameter lambda outputted by first complete data training set3-1(a) The initial model parameters are used as the initial model parameters when each group of data in the second complete data training set is subjected to model parameter calculation, and finally the model parameters lambda under the second complete data are obtained3-2(a) In that respect And the initialization parameters when each group of data in each complete data training set is subjected to model parameter calculation all use the model parameters output by the previous complete data training set until all the complete data are used. Model parameter lambda output by taking the last complete data training set3-10(a) The HMM model parameters of the micro-stepping brake intention at the 3 rd output time have the performance represented by the average likelihood probability P (O) under the last complete data test set3-10|λ3-10(a) The greater the average likelihood probability, the more accurate the model parameters are relatively. In this way, HMM model parameters λ of other braking intentions at the 3 rd output time can be obtained3(b)、λ3(c)、λ3(d) And the average likelihood probability lambda of each braking intention at other output timesn(a)、λn(b)、λn(c)、λn(d),n=1,2,…200。
The iterative formula of the EM algorithm for calculating the model parameters is as follows:
wherein m represents the mth iteration and t represents a time hidden state;
the right end of the above equations is in accordance with a given pedal displacement time series O3-q-r(q is the number of complete data) and m-1 model parameters of iterative output3-q-r (m-1)=(A(m-1),B(m-1),π(m-1)) The calculation is carried out, and the specific calculation formula is as follows:
representing the probability that the 'time t is in a hidden state i' under the condition of given model parameters and pedal displacement time sequences, wherein v represents the number of possible values of the hidden state;
representing the probability that the 'time t is in a hidden state i and the time t +1 is in a hidden state j' under the given model parameters and pedal displacement time sequence, aij、bj(ot+1) Is λ3_q_r (m-1)αt(i)、βt(i) α calculated by respectively using a forward algorithm and a backward algorithmt(i) Indicating that "at time t is in state i" and "the output sequence before that time is (o)1,o2,……ot) "the probability that both are true at the same time is calculated by the following recursion formula:
α initialization1(i)=π(i)bi(o1);
βt(i) indicating that the time t has been in state i, after which the output sequence is (o)t+1,ot+2,……ou) "is calculated by the following recursion formula:
β initializationu(i)=1;
third, brake intention on-line recognition
The online identification of the braking intention in the first braking intensity identification period will be specifically described below with reference to fig. 3 and 4. One braking intensity recognition cycle includes 20 output times, and the time number at which the output of the braking intention can be predicted is 12, and the time number at which the output of the braking intention is actually predicted is Npre.
Firstly, inputting a displacement time sequence O of the brake pedal1And calculating the likelihood probability of each braking intention at the first output time. When the output time number n is 1, not more than 11 and less than 20, the brake pedal displacement time sequence O is input2And calculating the likelihood probability of each braking intention at the output moment 2, and meanwhile, averaging the likelihood probability of each braking intention in the recognition time. When the output time n is 2 and is not more than 11 and less than 20, the pedal displacement time sequence O is continuously input3Calculating until the number of output time n is greater than 11 to judge whether the predicted braking intention is output, if the maximum average likelihood probability is not greater than 0.8, and if the number of output time n is not less than 20, determining the maximum average likelihood probabilityThe corresponding braking intention Y is comprehensively judged and then the braking intention is output; when the output time n is less than 20, the brake pedal displacement time sequence at the next time is continuously input, and the corresponding calculation is carried outOutputting likelihood probability of each braking intention at the moment, averaging likelihood probability of each braking intention in the recognition time until the maximum average likelihood probability value is greater than 0.8, outputting predicted braking intention, namely braking intention corresponding to the maximum average likelihood probability, and recording the maximum average likelihood probabilityThe output time number at this time is Npre for the corresponding braking intention Y. If the output time number n is equal to 20, outputting the braking intention after comprehensive judgment; if the number n of the output time is not equal to 20, continuing to input the brake pedal displacement time sequence of the next time, calculating the likelihood probability of each brake intention at the corresponding output time, and simultaneously calculating the average likelihood probability value of each brake intention from the output time with the serial number Npre to the current timeIf the output time number n is less than 20, continuing to input the brake pedal displacement time sequence of the next time and calculating the average likelihood probability valueRecording the maximum average likelihood probability value until the output time number n is not less than 20And (4) outputting the braking intention after comprehensive judgment, wherein the output time n of the corresponding braking intention X is equal to 20. The following 3 conditions are comprehensively judged: if the braking intention X is inconsistent with the predicted braking intention, the braking intention of the driver is obviously changed in the braking process, the braking intentions before and after the change are all output to a subsequent fuzzy controller, corresponding braking intensities are respectively calculated, and an average value is taken as the final estimated braking intensity; if the braking intention X is consistent with the predicted braking intention, the output predicted braking intention is kept and used for estimating the braking strength. If there is no predicted braking intention output, then the maximum average likelihood probability over the entire recognition time period is takenThe corresponding braking intention Y is output and used to estimate the braking intensity.
Fourthly, fuzzy recognition of braking intention coefficient
The following takes the micro-stepping braking intention as an example to specifically describe the fuzzy recognition of the braking intention coefficient:
(1) inputs to the first fuzzy controller are determined as brake pedal displacement and brake pedal force, and outputs are brake intent coefficients. The ranges of brake pedal displacement and brake pedal force are determined by the results of on-line identification of braking intent, here exemplified by "microstep" braking intent. The discourse domain of the brake pedal displacement is [0, 40] and the unit is mm; the argument of the brake pedal force is [0, 100], in N; the brake intent coefficient argument field is [0, 1 ].
(2) Determining a fuzzy control knowledge base
Firstly, mapping each input and output quantity to a standardized discourse field [0, 1], and determining a quantization level and a fuzzy subset. And (3) quantizing the brake pedal displacement, the brake pedal force and the brake intention coefficient on the standardized domain into 11 grades, wherein the fuzzy subset is { S, MS, M, MB, B }, wherein S: small; MS: is small; m: performing the following steps; MB: is large; b: is large. Secondly, determining a membership function, wherein the triangular membership function has large change, so that the control sensitivity and the resolution are high; the trapezoidal membership function is stable and stable in change. Therefore, the advantages of the triangular membership function and the trapezoidal membership function are combined, and the control effect is improved. Each input and output quantity has the same membership function, as shown in fig. 5:
the discretized table of membership functions is shown in table 1 below:
TABLE 1 discretization of membership functions
According to the above table:
μS(x)=μS(y)=μS(z)=[1,1,0.5,0,0,0,0,0,0,0,0]
μMS(x)=μMS(y)=μMS(z)=[0,0,0.5,1,0.5,0,0,0,0,0,0]
μM(x)=μM(y)=μM(z)=[0,0,0,0,0.5,1,0.5,0,0,0,0]
μMB(x)=μMB(y)=μMB(z)=[0,0,0,0,0,0,0.5,1,0.5,0,0]
μB(x)=μB(y)=μB(z)=[0,0,0,0,0,0,0,0,0.5,1,1]
x, y and z respectively represent discrete value points of the brake pedal displacement, the brake pedal force and the brake intention coefficient in the normalization theory domain, and are respectivelyc(k) And representing the membership function corresponding to the fuzzy value c.
(3) Determining a fuzzy control rule base
The greater the brake pedal displacement, the greater the brake pedal force, and the correspondingly greater the brake intent factor. The fuzzy control rules are shown in Table 2, And adopt double-front multi-rule statements, i.e. "IfX Is AAnd Y Is B, Then Z Is C", such as "IfX Is SAnd Y Is S, Then Z Is S" or "IfX Is S And Y Is MS, Then Z Is S".
TABLE 2 fuzzy control rules
(4) Computing a fuzzy relation matrix
According to the definition of the Mamdani fuzzy relation:
RXYZ=((μX(x))T×μY(y))T1×μZ(z);x,y,z=0,0.1,0.2,......0.9,1;X,Y,Z=S,MS,M,MB,B
the multi-rule fuzzy relation matrix is a union of each rule fuzzy matrix:
RXYZ=RXYZ1∪RXYZ2∪RXYZ3.......∪RXYZn。
wherein:
the cartesian product "x" represents a small operation, and the union "U" represents a large operation.
②((μx(x))T×μY(y))T1Indicates that will (mu)x(x))T×μYThe result of the calculation of (y) is straightened out row vector by row and then transposed into column vector.
For example, the first fuzzy rule is taken as an example, the fuzzy relation matrix is RXYZ1The calculation formula is as follows:
(μ1)121×1=((μS(x))T×μS(y))T1=((1,1,0.5,0,0,0,0,0,0,0,0)T×(1,1,0.5,0,0,0,0,0,0,0,0))T1
(RXYZ1)121×11=(μ1)121×1×(μS(z))1×11
then calculate R one by oneXYZ2,RXYZ3......RXYZ25Then, the union set is solved to obtain the overall fuzzy relation matrix R of multiple rules and multiple antecedentsXYZ。
(5) Fuzzy decision making
Fuzzy output U ═ E · RXYZWherein "·" means "max-min synthesis", E ═ μ (x)T×μ(y))T2And μ (x) and μ (y) represent fuzzy sets of actually inputted brake pedal force and brake pedal displacement, respectively (((μ (x))Tμ(y))T2Represents that (mu (x))TThe × μ (y) calculation straightens the row vectors by row.
(6) Defuzzification
And (3) defuzzifying the fuzzy output quantity of the braking intention coefficient by adopting a gravity center method to obtain an accurate value of the braking intention coefficient positioned at [0, 1], and taking the accurate value as the input of the fuzzy controller of the next layer. The specific calculation formula is as follows:
record output U ═ mu (z)1),μ(z2),......μ(z10),μ(z11)),ziTo normalize discrete values of discourse field, braking intention coefficient
Fifthly, fuzzy identification of braking strength
The following takes the micro-stepping brake intention as an example to specifically describe the fuzzy recognition of the brake strength:
(1) the input of the second fuzzy controller is determined as the displacement change rate of the brake pedal and the braking intention coefficient, and the output is the braking strength. The range of the brake pedal displacement change rate and the brake intensity is determined by the online identification result of the brake intention, and the 'micro-stepping' brake intention is taken as an example. The discourse domain of the displacement change rate of the brake pedal is [0, 200] and the unit is mm/s; the argument of the braking intention coefficient is [0, 1 ]; the brake strength discourse field is [0, 0.1 ].
(2) Determining a fuzzy control knowledge base
And mapping the brake pedal displacement change rate and the brake intensity to a standardized domain [0, 1], and determining a quantization level and a fuzzy subset. And (3) measuring the displacement change rate of the brake pedal, the brake intention coefficient and the brake strength into 11 grades, wherein the fuzzy subset is { S, MS, M, MB, B }. The trigonometric membership function and the trapezoidal membership function are also selected. Specifically, as shown in fig. 5, the discretization table of membership functions is shown in table 1:
(3) determining a fuzzy control rule base
The greater the braking intent factor, the greater the rate of change of pedal displacement, and the correspondingly greater the required braking intensity. The fuzzy control rules are shown in table 3:
TABLE 3 fuzzy control rules
(4) The process of the fuzzy relation matrix obtaining, fuzzy decision making and defuzzification is similar to the process of the braking intention coefficient fuzzy reasoning and calculation, and is not described here.
The present invention is not limited to the above-described embodiments, and any obvious improvements, substitutions or modifications can be made by those skilled in the art without departing from the spirit of the present invention.
Claims (8)
1. A fuzzy recognition method for brake strength considering brake intention is characterized in that brake pedal data of different brake intentions are obtained through experiments, and the brake pedal data are classified according to the brake intention by taking the brake strength as a standard; identifying HMM model parameters of each braking intention at each output moment in an off-line manner; identifying a braking intention on line according to a brake pedal displacement time sequence acquired in real time; taking the displacement of a brake pedal and the force of the brake pedal as the input of a first-layer fuzzy controller, and outputting a braking intention coefficient; and (5) taking the braking intention coefficient and the displacement change rate of the brake pedal as the input of the fuzzy controller of the second layer, and outputting the braking strength.
2. The fuzzy recognition method for braking intensity considering braking intention according to claim 1, wherein the offline recognition of HMM model parameters of each braking intention at each output moment specifically comprises:
acquiring complete data for an HMM model parameter identification process according to experimental data, initializing model parameters when a training set in first complete data is used, calculating model parameters of each group of pedal displacement time sequences in the training set by adopting an EM (effective magnetic resonance) algorithm, and averaging the model parameters to obtain model parameters of the training set; calculating likelihood probability by using model parameters obtained by a training set for each group of pedal displacement time sequences in the test set, averaging, and if the average likelihood probability is smaller than a threshold value Pmin1, re-initializing the model parameters to calculate the model parameters until the average likelihood probability of the test set is not smaller than a threshold value Pmin 1; performing model parameter iteration by using the next complete data, wherein the initialized model parameters of each subsequent training set all use the model parameters output by the previous training set; and the model parameters output by the last training set are used as the HMM model parameters of the braking intention at the output moment.
3. The fuzzy recognition method of braking intensity considering braking intention according to claim 2, wherein the model parameters include an initial time state probability column vector pi, a state transition matrix a and an observation probability matrix B.
4. The fuzzy recognition method for braking strength considering braking intention according to claim 2, wherein the complete data is obtained by:
the displacement time sequence of W groups of brake pedals corresponding to a certain braking intention at a certain output time is M, M-1 parts of data are used as a training set each time, and 1 part of data are used as an inspection set to form complete data; the different training and test sets were combined into M complete data.
5. The braking intensity fuzzy recognition method considering the braking intention according to claim 1, wherein the HMM model input at the time of the online recognition of the braking intention is: brake pedal displacement time series O obtained in real time1,O2,…OnThe output is: likelihood probability P (O) of each braking intention of pedal displacement time series at corresponding output timen|λn(x) Where N is 1, 2, 3 … N, N denotes the output time number, OnA pedal displacement time series showing the nth output time, x ═ a, b, c, d, respectively, and represents "microstep", "light step", "middle step", and "heavy step", λn(x) HMM model parameters representing the braking intention x at the nth output time; averaging the likelihood probabilities of individual braking intents for each braking intention over an identification time
6. The fuzzy recognition method for brake intensity considering brake intention according to claim 5, wherein the predicted brake intention is outputted when satisfiedGraph conditions: that is, the output time number n is not less than the number of predictable output times Npre, and the maximum likelihood probability average value at the output timeIf the value is larger than the threshold value Pmin2, outputting the predicted braking intention x; calculating the likelihood probability average value of each braking intention after prediction output When n is equal to the maximum output time Np within one brake intensity recognition period, the maximum value of the predicted likelihood probability average value of each brake intention after output needs to be calculatedAnd comparing the corresponding braking intention with the predicted braking intention x, and outputting the final braking intention after comprehensive judgment.
7. The fuzzy recognition method of braking intensity considering braking intention according to claim 6, wherein said comprehensive judgment comprises:
(1)the corresponding braking intention is consistent with the predicted braking intention, and the predicted braking intention is used as a final braking intention;
(2)the corresponding braking intent is not consistent with the predicted braking intent,the corresponding braking intention and the predicted braking intention are used for subsequent fuzzy recognition to output the braking strengthTaking the average value of the two brake intensities as the final output brake intensity;
8. The braking intensity fuzzy recognition method considering braking intention according to claim 1, characterized in that the brake pedal data are a brake pedal displacement time series, a brake pedal force and a brake pedal displacement change rate.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010312924.0A CN111547028B (en) | 2020-04-20 | 2020-04-20 | Brake intensity fuzzy recognition method considering brake intention |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010312924.0A CN111547028B (en) | 2020-04-20 | 2020-04-20 | Brake intensity fuzzy recognition method considering brake intention |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111547028A true CN111547028A (en) | 2020-08-18 |
CN111547028B CN111547028B (en) | 2021-07-20 |
Family
ID=72007544
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010312924.0A Active CN111547028B (en) | 2020-04-20 | 2020-04-20 | Brake intensity fuzzy recognition method considering brake intention |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111547028B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113147711A (en) * | 2021-04-06 | 2021-07-23 | 南京航空航天大学 | Nonlinear braking force compensation method of giant magnetostrictive brake-by-wire system |
CN113177359A (en) * | 2021-04-30 | 2021-07-27 | 上海电机学院 | Dummy model-based body tissue state prediction method |
CN113759804A (en) * | 2021-09-24 | 2021-12-07 | 湖南大学 | Method and system for vehicle remote control |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103318181A (en) * | 2013-06-19 | 2013-09-25 | 电子科技大学 | Driver intention recognition method |
CN106960189A (en) * | 2017-03-20 | 2017-07-18 | 重庆邮电大学 | A kind of driving intention decision method based on hidden Markov model |
US20180286220A1 (en) * | 2017-04-03 | 2018-10-04 | Here Global B.V. | Vehicle traffic state determination |
CN108773376A (en) * | 2018-05-07 | 2018-11-09 | 南京航空航天大学 | A kind of the automobile objective layered Collaborative Control and optimization method of fusion driving intention |
CN110626339A (en) * | 2019-07-10 | 2019-12-31 | 聊城大学 | Emergency obstacle avoidance automatic driving assistance torque calculation method based on driver intention |
-
2020
- 2020-04-20 CN CN202010312924.0A patent/CN111547028B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103318181A (en) * | 2013-06-19 | 2013-09-25 | 电子科技大学 | Driver intention recognition method |
CN106960189A (en) * | 2017-03-20 | 2017-07-18 | 重庆邮电大学 | A kind of driving intention decision method based on hidden Markov model |
US20180286220A1 (en) * | 2017-04-03 | 2018-10-04 | Here Global B.V. | Vehicle traffic state determination |
CN108773376A (en) * | 2018-05-07 | 2018-11-09 | 南京航空航天大学 | A kind of the automobile objective layered Collaborative Control and optimization method of fusion driving intention |
CN110626339A (en) * | 2019-07-10 | 2019-12-31 | 聊城大学 | Emergency obstacle avoidance automatic driving assistance torque calculation method based on driver intention |
Non-Patent Citations (1)
Title |
---|
赵轩、马建、汪贵平: "基于制动驾驶意图辨识的纯电动客车复合制动控制策略", 《交通运输工程学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113147711A (en) * | 2021-04-06 | 2021-07-23 | 南京航空航天大学 | Nonlinear braking force compensation method of giant magnetostrictive brake-by-wire system |
CN113147711B (en) * | 2021-04-06 | 2022-04-05 | 南京航空航天大学 | Nonlinear braking force compensation method of giant magnetostrictive brake-by-wire system |
CN113177359A (en) * | 2021-04-30 | 2021-07-27 | 上海电机学院 | Dummy model-based body tissue state prediction method |
CN113759804A (en) * | 2021-09-24 | 2021-12-07 | 湖南大学 | Method and system for vehicle remote control |
CN113759804B (en) * | 2021-09-24 | 2023-06-16 | 湖南大学 | Method and system for vehicle remote control |
Also Published As
Publication number | Publication date |
---|---|
CN111547028B (en) | 2021-07-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111547028B (en) | Brake intensity fuzzy recognition method considering brake intention | |
CN106971194B (en) | Driving intention recognition method based on improved HMM and SVM double-layer algorithm | |
CN107958269B (en) | Driving risk degree prediction method based on hidden Markov model | |
CN108280484B (en) | Driver acceleration characteristic online classification and identification method | |
CN111832225A (en) | Method for constructing driving condition of automobile | |
CN110910531B (en) | Rapid pavement friction coefficient detection method based on vehicle-mounted OBD information | |
CN109255094B (en) | Commercial truck quality estimation method based on SVR-Adaboost improved algorithm | |
CN112507915B (en) | Bolt connection structure loosening state identification method based on vibration response information | |
CN113188807B (en) | Automatic abs result judging algorithm | |
CN111414932A (en) | Classification identification and fault detection method for multi-scale signals of aircraft | |
US11535214B2 (en) | System of modeling anti-lock brake system controller of vehicle | |
CN111191824B (en) | Power battery capacity attenuation prediction method and system | |
CN115742855A (en) | Electric automobile remaining mileage prediction method and device, electric automobile and medium | |
CN110490275A (en) | A kind of driving behavior prediction technique based on transfer learning | |
CN109167546B (en) | Asynchronous motor parameter online identification method based on data generation model | |
Singh et al. | Deep learning in the automotive industry: Recent advances and application examples | |
CN111275004A (en) | Bearing fault diagnosis method based on LMD and impulse neural network | |
CN112985830A (en) | Abs result automatic judging algorithm | |
CN112763967A (en) | BiGRU-based intelligent electric meter metering module fault prediction and diagnosis method | |
CN115859077A (en) | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions | |
CN111027618A (en) | Automobile dynamic property and economic property expectation quantification method | |
CN114547963A (en) | Tire modeling method and medium based on data driving | |
CN113326962B (en) | Subway train ATO speed curve prediction method based on BP neural network | |
CN113799788A (en) | Method, computer program and device for processing a signal | |
Boada et al. | Neural-empirical tyre model based on recursive lazy learning under combined longitudinal and lateral slip conditions |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20240423 Address after: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee after: Dragon totem Technology (Hefei) Co.,Ltd. Country or region after: China Address before: Zhenjiang City, Jiangsu Province, 212013 Jingkou District Road No. 301 Patentee before: JIANGSU University Country or region before: China |