CN115859123A - Vehicle running condition online identification method based on random prediction and machine learning - Google Patents
Vehicle running condition online identification method based on random prediction and machine learning Download PDFInfo
- Publication number
- CN115859123A CN115859123A CN202310183550.0A CN202310183550A CN115859123A CN 115859123 A CN115859123 A CN 115859123A CN 202310183550 A CN202310183550 A CN 202310183550A CN 115859123 A CN115859123 A CN 115859123A
- Authority
- CN
- China
- Prior art keywords
- vehicle speed
- time
- prediction
- condition
- real
- 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
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention belongs to the technical field of vehicle running condition identification, and provides a vehicle running condition online identification method based on random prediction and machine learning. The defects that the existing working condition identification method is accompanied by larger time delay and is not beneficial to real-time control of the whole vehicle control system are overcome. The short-term driving condition prediction model simultaneously reads the real-time vehicle speed output by the data preprocessing module and the real-time driving style type output by the driving style recognition model, and outputs a predicted vehicle speed sequence within a period of time after the current time point; the working condition data synthesis module splices the real-time updated historical vehicle speed sequence and the predicted vehicle speed sequence to construct combined working condition data containing the future working condition change trend; and the real-time running condition recognition model reads the characteristic parameters of the combined condition data output by the condition data synthesis module. The invention takes the combined data of the historical vehicle speed sequence and the predicted vehicle speed sequence as the identification object, thereby improving the real-time performance of the identification result while ensuring the identification precision.
Description
Technical Field
The invention belongs to the technical field of vehicle running condition identification, and particularly relates to a vehicle running condition online identification method based on random prediction and machine learning.
Background
The driving condition identification is to identify the type of the driving condition by utilizing a characteristic extraction and classification technology, and the main identification algorithm comprises a neural network, a support vector machine, fuzzy classification, cluster analysis and the like.
For a traditional vehicle, the running condition can be used for determining the emission and fuel consumption level of the vehicle and providing a basis for power matching and fuel economy optimization; for the pure electric vehicle, the calculation accuracy of the remaining driving mileage can be improved by acquiring the driving condition information in real time; for a hybrid electric vehicle, a self-adaptive energy management strategy can be established based on working condition data, the economy of the whole vehicle is improved, and the service life of parts of a power system is prolonged. Because the working condition of the vehicle in actual running is changed randomly, the online identification method of the running working condition with higher identification precision and real-time performance is extremely important.
The main difference of the existing researches is that the method for dividing the typical running working conditions and selecting the characteristic parameters is different, and the attempts to simplify the neural network structure and improve the identification precision by optimizing external input parameters all achieve better effects. However, the existing driving condition recognition method usually uses fixed-length historical operating condition data as a recognition object, and although the type of the generated driving condition can be recognized more accurately, the recognition result is often accompanied by a large time delay, which is not beneficial to the real-time control of the whole vehicle control system.
Disclosure of Invention
In order to solve at least one technical problem in the prior art, the invention provides an online vehicle running condition identification method based on random prediction and machine learning.
The invention is realized by adopting the following technical scheme: a vehicle driving condition online identification method based on random prediction and machine learning comprises the following steps:
s101, reading real-time running data of a vehicle by a data preprocessing module, wherein the real-time running data comprises vehicle speed, accelerator pedal opening and brake pedal opening;
s102: the driving style recognition model reads the real-time vehicle speed, the accelerator pedal opening, the accelerator pedal change rate, the brake pedal opening and the brake pedal change rate output by the data preprocessing module and outputs a real-time driving style type;
s103: the short-term driving condition prediction model simultaneously reads the real-time vehicle speed output by the data preprocessing module and the real-time driving style type output by the driving style recognition model, and outputs a predicted vehicle speed sequence within a period of time after the current time point;
s104: the working condition data synthesis module simultaneously reads a historical vehicle speed sequence within a period of time before the current time point output by the data preprocessing module and a predicted vehicle speed sequence within a period of time after the current time point output by the short-term running working condition prediction model, and splices the historical vehicle speed sequence updated in real time and the predicted vehicle speed sequence to construct combined working condition data containing the future working condition change trend;
s105: the real-time running condition recognition model reads characteristic parameters of the combined condition data output by the condition data synthesis module, and recognizes the real-time running condition type of the vehicle on line.
Preferably, in step S101, the real-time driving data of the vehicle is read through the entire vehicle CAN bus, the historical vehicle speed sequence is updated in real time in the form of a sliding time window, and the historical driving data exceeding the width of the time window is discarded.
Preferably, in step S102, the driving style recognition model is built based on the learning vector quantization neural network LVQNN, wherein the input of the input layer is real-time vehicle speed, accelerator pedal opening, accelerator pedal change rate, brake pedal opening, and brake pedal change rate; the outputs of the output layers are respectively
The three driving style types respectively correspond to a mild style, a normal style and an aggressive style.
Preferably, in step S103, the short-term driving condition prediction model is established by a markov chain stochastic prediction algorithm, the markov chain uses the vehicle speed and the acceleration as the basis for dividing the conditions, and a state transition probability matrix is established according to the transition frequency between the states; and filtering the output of the short-term running condition prediction model to obtain a final predicted vehicle speed sequence.
Preferably, in step S104, the total length of the combined condition data is equal to the sum of the length of the historical vehicle speed sequence and the length of the predicted vehicle speed sequence, and the total length of the combined condition data is a fixed value, and the length of the predicted vehicle speed sequence is variable, which is determined according to the root mean square error of the vehicle speed of the last prediction period.
Preferably, the root mean square error of the vehicle speed
The expression of (a) is:
in the formula (I), the compound is shown in the specification,
is the predicted vehicle speed value after the filtering processing>
The real vehicle speed value is obtained; />
To prepareMeasuring the length of the time domain, wherein the length is equal to the length of the predicted vehicle speed sequence;
setting the minimum value of the length of the prediction time domain as
Maximum value is->
Predicting a change step in the time domain as>
And->
，/>
Is a positive integer; />
、/>
Two preset values of the root mean square error of the vehicle speed;
root mean square error of vehicle speed in the last prediction period
If the prediction time domain length of the last prediction period is less than ^ or ^ the prediction time domain length of the last prediction period>
Then the length of the current prediction time domain is increased/compared to the last prediction period>
Otherwise, the length of the current prediction time domain is kept consistent with the length of the last prediction period;
root mean square error of vehicle speed of the last prediction cycle
If the prediction time domain length of the last prediction period is greater than ^ 4>
Then the length of the current prediction time domain is decreased ≥ compared to the last prediction period>
Otherwise, the length of the current prediction time domain is kept consistent with the length of the previous prediction period;
root mean square error of vehicle speed in the last prediction period
The length of the current prediction time domain is kept consistent with the length of the last prediction period.
Preferably, the minimum of the length of the prediction time domain
Is 10s and has a maximum value>
A change step of the prediction time domain of 30 s->
Is 5s, a preset value>
Is 3km/h, preset value>
Is 5km/h.
Preferably, in step S105, the real-time driving condition recognition model is established based on the learning vector quantization neural network LVQNN, wherein the input of the input layer is a key variable extracted from the feature parameters of the combined condition data through principal component analysis, and the key variable includes an average deceleration, a maximum acceleration, a maximum vehicle speed, an average driving vehicle speed, a low-speed driving time ratio, and an acceleration time ratio; the output of the output layer is
The three real-time running working condition types respectively correspond to a low-speed running working condition, a medium-speed running working condition and a high-speed running working condition.
Preferably, the training and testing process of the model in the online identification method of the vehicle running condition comprises the following steps:
s201: defining the type of the running working condition, and selecting typical standard cycle working conditions of various running working conditions from a standard cycle working condition library according to the amplitude of the vehicle speed and the change frequency of the vehicle speed;
s202: carrying out segment division on a typical standard cycle working condition, establishing a driving working condition data set, and dividing the driving working condition data set into a first data set and a second data set;
s203: taking the first data set as input, and training and testing the driving style recognition model;
s204: taking the first data set as input, calculating Markov state transition probability matrixes under various driving styles, and acquiring a predicted vehicle speed sequence of the tail end moment of a working condition segment;
s205: combining the working condition segments in the first data set with the corresponding predicted vehicle speed sequences to construct a combined working condition data set; taking the combined working condition data set as input, and training and testing the real-time driving working condition recognition model;
s206: and extracting the trained model parameters, implanting the trained model parameters into a corresponding model of the driving condition recognition system, and testing by adopting a second data set to obtain a complete driving condition recognition system.
Preferably, the running condition data set is composed of a large number of working condition segments with unequal lengths, which are obtained by dividing a plurality of typical standard cycle working conditions through random segments;
the working condition segments are obtained by random number sampling, namely, two-dimensional random sampling is carried out on each standard working condition in two aspects of starting time and length, and the starting time of segment extraction
And the fragment length->
Expressed as:
in the formula (I), the compound is shown in the specification,
the time length of the standard working condition; />
And &>
Maximum and minimum values for the segment length; />
And
is a random number between 0 and 1.
Compared with the prior art, the invention has the beneficial effects that:
the driving style recognition and random prediction algorithm is added in the traditional driving condition recognition method, so that the combined data of the historical vehicle speed data and the prediction data of the future vehicle speed is used as a recognition object, the characteristics of the current driving condition in the recognition object are enhanced, the dependence degree of a working condition recognition system on the historical vehicle speed data is reduced, the driving condition recognition precision is ensured, and the real-time performance of a recognition result is improved. The method is applied to an Energy Management System (EMS) of a new energy automobile, so that the adaptivity of the system can be enhanced, and the dynamic property, the economical efficiency and the service life of the whole automobile can be improved.
According to the method, the standard circulating working condition is selected to establish the data set for model training and testing, so that the complex process of establishing the data set by adopting real driving working condition data is avoided, and the development efficiency of the working condition recognition system is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required in the embodiments will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic flow chart of the online driving condition identification method in the present embodiment;
FIG. 2 is a schematic diagram illustrating a rolling update process of data during data preprocessing in the present embodiment;
FIG. 3 is a schematic structural diagram of a learning vector quantization neural network in the present embodiment;
FIG. 4 is a schematic diagram of a Markov state transition probability matrix in the short-term driving condition prediction model in the embodiment;
fig. 5 is a diagram of a logic switching rule for predicting the length of the time domain in the present embodiment;
FIG. 6 is a schematic diagram of the training and testing process of the work condition recognition system in this embodiment;
FIG. 7 is a diagram illustrating exemplary standard cycle conditions for various driving conditions used to construct the condition data set in the present embodiment.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described in conjunction with the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without any creative effort, shall fall within the protection scope of the present invention.
It should be understood that the structures, ratios, sizes, and the like shown in the drawings are only used for matching the disclosure of the present disclosure, and are not used for limiting the limit of the present disclosure, which will be understood and read by those skilled in the art, and therefore, the present disclosure has no technical significance, and any structural modification, ratio relationship change or size adjustment should fall within the scope of the present disclosure without affecting the function and the achievable effect of the present disclosure, and it should be noted that the relationship terms such as first and second, and the like in the present disclosure are only used for distinguishing one entity from another entity without necessarily requiring or implying any actual relationship or order between the entities.
The present invention provides an embodiment:
as shown in FIG. 1, the method for identifying the vehicle running condition on line based on the random prediction and the machine learning comprises the following steps:
s101, reading real-time running data of a vehicle by a data preprocessing module, wherein the real-time running data comprises a vehicle speed, an accelerator pedal opening and a brake pedal opening;
s102: the driving style recognition model reads the real-time vehicle speed, the accelerator pedal opening, the accelerator pedal change rate, the brake pedal opening and the brake pedal change rate output by the data preprocessing module and outputs a real-time driving style type;
s103: the short-term driving condition prediction model simultaneously reads the real-time vehicle speed output by the data preprocessing module and the real-time driving style type output by the driving style recognition model, and outputs a predicted vehicle speed sequence within a period of time after the current time point;
s104: the working condition data synthesis module simultaneously reads a historical vehicle speed sequence within a period of time before the current time point output by the data preprocessing module and a predicted vehicle speed sequence within a period of time after the current time point output by the short-term running working condition prediction model, and splices the historical vehicle speed sequence updated in real time and the predicted vehicle speed sequence to construct combined working condition data containing the future working condition change trend;
s105: the real-time running condition recognition model reads characteristic parameters of the combined condition data output by the condition data synthesis module, and recognizes the real-time running condition type of the vehicle on line.
Specifically, real-time running data of the vehicle is read through a whole vehicle CAN bus, and a data preprocessing module stores and updates historical running data in real time; as shown in FIG. 2, the historical vehicle speed sequence is updated in real time in the form of a sliding time window, taking the vehicle speed data as an example, and assuming that the current time is
A time window width of>
In>
At the moment, historical vehicle speed data recorded in the data preprocessing process is ^>
And discarding historical driving data exceeding the width of the time window. In the present embodiment, a sequence of continuously scrolling and updating the vehicle speed in the sliding window at time intervals of 1 second is provided.
The driving style recognition model is built based on a learning vector quantization neural network LVQNN-1, as shown in FIG. 3, the learning vector quantization neural network comprises an input layer, a competition layer and an output layer, wherein the input layer and the competition layer are completely connected, and each neuron of the output layer is only connected with one group of neurons in the competition layer.
Represents the connection weight between the input layer and the competition layer, and is in the training process>
Is gradually adjusted to the cluster center and is then adjusted to be greater than or equal to>
Represents the connection weight between the competition layer and the output layer, and the value is fixed to 1. And (3) calculating the distance between the competitive layer neuron and the input vector, judging that the competitive layer neuron with the minimum distance wins, wherein the output of the output neuron connected with the winning neuron is 1, and the output neurons are 0.
Wherein the input of the input layer is real-time vehicle speed, accelerator pedal opening, accelerator pedal change rate, brake pedal opening and brake pedal change rate, respectively corresponding to those in FIG. 1
(ii) a The outputs of the output layers are respectively->
The three driving style types respectively correspond to a mild type style, a normal type style and an aggressive type style. Setting input layer neuron number to 5, competition layerThe neuron number is 11, the linear output layer neuron number is 3, the learning rate is 0.01, and the error threshold is 0.01.
The short-term running condition prediction model is established by adopting a Markov chain random prediction algorithm, the Markov chain takes the speed and the acceleration as the basis for dividing the working condition states, and a state transition probability matrix is established according to the transition frequency among the states; and dividing sampling points with similar driving states into intervals by taking the vehicle speed and the acceleration as the basis for dividing the working condition states, and marking each interval as one state. The selected speed and acceleration ranges are 0 to 130km/h and-3 to 3m/s 2 The interval between the state sections is 1km/h and 0.05m/s respectively 2 。
Specifically, the model predicts the length of the time domain according to the current vehicle speed information
For the maximum time scale, counting the transition probabilities between states in the driving condition at different time scales, and constructing a multi-step state transition probability matrix at different time scales, as shown in FIG. 4, by predicting the ^ H>
At the time of the vehicle speed, a criterion of ^ is called>
The probability matrix of the time domain is used for improving the accuracy of predicting the vehicle speed in different prediction time domains, and the transition probability in each time scale is represented as:
in the formula (I), the compound is shown in the specification,
is in a state, is asserted>
Indicates that the Markov chain is->
At the moment so as to->
Is the one-step transition probability of the time scale.
State transition probability matrix
Representing on different time scales>
Then, the probability of transition between different states, each element in the matrix is represented as:
in the formula (I), the compound is shown in the specification,
indicating a driving situation when the predicted time period is->
When it is selected from the status->
Transfer into a state->
Frequency of (d);
indicating a driving situation when the predicted time period is->
In time, from state>
Transition to the sum of the arbitrary states at the next time.
According to the current driving state
And the state transition probability matrix of the corresponding time scale->
The status to which the next moment is transferred is decided by a random procedure>
. In this embodiment, the randomness of the change in vehicle speed is simulated using the Monte Carlo method, i.e., a random number between 0 and 1 is generated>
The following inequality is satisfied:
then the time
The latter predicted vehicle speed may be expressed as:
in the formula (I), the compound is shown in the specification,
is passed>
Predicted vehicle speed after time; />
Is the current vehicle speed>
Is currently status->
To a predictive state>
The vehicle speed span of (1); />
A speed value represented by a single span of vehicle speed.
Filtering the output of the short-term running condition prediction model to obtain a final predicted vehicle speed sequence, smoothing the prediction result by adopting 4-point mean filtering to eliminate high-frequency noise in the prediction result, and filtering
The predicted value of the vehicle speed at the moment is->
Expressed as:
the length of the maximum prediction time domain is set to be 30 seconds, and the interval between prediction time domains is set to be 1 second.
The total length of the combined working condition data is equal to the sum of the length of the historical vehicle speed sequence and the length of the predicted vehicle speed sequence, the total length of the combined working condition data is a fixed value, the length of the predicted vehicle speed sequence is variable, and the total length of the combined working condition data is determined according to the root mean square error of the vehicle speed of the last prediction period.
Root mean square error of vehicle speed
The expression of (a) is:
in the formula (I), the compound is shown in the specification,
is the predicted vehicle speed value after the filtering processing>
The real vehicle speed value is obtained; />
The length of the prediction time domain is equal to the length of the prediction vehicle speed sequence;
setting the minimum value of the length of the prediction time domain as
Maximum value is->
Predicting a change step of ^ in the time domain>
And->
，/>
Is a positive integer; />
、/>
Two preset values of the root mean square error of the vehicle speed;
root mean square error of vehicle speed in the last prediction period
If the prediction time domain length of the last prediction period is less than ^ or ^ the prediction time domain length of the last prediction period>
Then the length of the current prediction time domain is increased ≥ from the last prediction period>
Otherwise, the length of the current prediction time domain is kept consistent with the length of the last prediction period;
root mean square error of vehicle speed in the last prediction period
If the length of the prediction time domain of the previous prediction period is greater than &>
Then the length of the current prediction time domainIs decreased compared to the last prediction period>
Otherwise, the length of the current prediction time domain is kept consistent with the length of the last prediction period;
root mean square error of vehicle speed of the last prediction cycle
The length of the current prediction time domain is kept consistent with the length of the last prediction period.
Preferably, the minimum value of the length of the time domain is predicted
Is 10s and has a maximum value>
A change step of the prediction time domain of 30 s->
Is 5s, a preset value>
Is 3km/h, preset value>
The logic switching rule for predicting the length of the time domain is shown in fig. 5 at 5km/h.
The real-time running condition recognition model is established based on a learning vector quantization neural network LVQNN-2, wherein the input of an input layer is key variables extracted from characteristic parameters of combined condition data through principal component analysis, and the key variables comprise average deceleration, maximum acceleration, maximum vehicle speed, average running vehicle speed, low-speed running time proportion and acceleration time proportion; the output of the output layer is
Three real-time running condition types respectively corresponding to a low-speed running condition, a medium-speed running condition and a high-speed running condition, setting the number of neurons in an input layer to be equal to the number of characteristic parameters, setting the number of neurons in a competition layer to be 11, and carrying out linear inputThe number of the neurons in the layer is 3, the learning rate is 0.01, and the error threshold value is 0.01.
To describe the differences between the various types of driving conditions, 16 characteristic parameters are defined, as shown in table 1.
In order to obtain the optimal characteristic parameter combination, avoid introducing repeated information due to excessive characteristic parameters and increase system calculation amount, the average deceleration, the maximum acceleration, the maximum vehicle speed, the average running vehicle speed, the low-speed running time proportion and the acceleration time proportion are determined to be used as the input of the real-time running condition identification model through Principal Component Analysis (PCA) and a large amount of comparative analysis.
As shown in fig. 6 and 7, the training and testing process of the model in the online identification method of the vehicle driving condition includes the following steps:
s201: defining the type of the running working condition, and selecting typical standard cycle working conditions of various running working conditions from a standard cycle working condition library according to the amplitude of the vehicle speed and the change frequency of the vehicle speed; the high-speed working condition has higher average running speed, and the change frequency of the speed is lower; the average running speed under the low-speed working condition is lower, and the speed change frequency is higher; the average running speed and the speed change frequency of the medium-speed working condition are both in a medium equal level;
s202: carrying out segment division on a typical standard cycle working condition, establishing a driving working condition data set, and dividing the driving working condition data set into a first data set and a second data set; the first data set and the second data set account for 70% and 30% of the driving condition data set, respectively.
S203: taking the first data set as input, and training and testing the driving style recognition model;
s204: taking the first data set as input, calculating Markov state transition probability matrixes under various driving styles, and acquiring a predicted vehicle speed sequence of the tail end moment of a working condition segment;
s205: combining the working condition segments in the first data set with the corresponding predicted vehicle speed sequences to construct a combined working condition data set; taking the combined working condition data set as input, and training and testing the real-time driving working condition recognition model;
s206: and extracting the trained model parameters, implanting the model parameters into a corresponding model of the driving condition recognition system, and testing by adopting a second data set to obtain the complete driving condition recognition system.
The running condition data set is composed of a large number of working condition segments with unequal lengths, which are obtained by dividing a plurality of typical standard cycle working conditions through random segments; the working condition segments are obtained by random number sampling, namely, two-dimensional random sampling is carried out on each standard working condition in two aspects of starting time and length, and the starting time of segment extraction
And the fragment length->
Expressed as: />
In the formula (I), the compound is shown in the specification,
the time length of the standard working condition; />
And &>
Maximum and minimum values for the segment length; />
And
is a random number between 0 and 1.
The classification index of the driving style is the average speed of each segment
Average acceleration->
Mean plus change rate->
Average deceleration->
On average rate of change of deceleration>
The calculation formula of each index is as follows:
in the formula (I), the compound is shown in the specification,
and &>
Respectively, a speed value and an acceleration value at each moment in the operating condition section>
The total number of data for the operating mode segment->
And &>
The total number of data with the acceleration value greater than 0 and less than 0 in the working condition segment respectively.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.
Claims (10)
1. A vehicle running condition online identification method based on random prediction and machine learning is characterized by comprising the following steps:
s101, reading real-time running data of a vehicle by a data preprocessing module, wherein the real-time running data comprises a vehicle speed, an accelerator pedal opening and a brake pedal opening;
s102: the driving style recognition model reads the real-time vehicle speed, the accelerator pedal opening, the accelerator pedal change rate, the brake pedal opening and the brake pedal change rate output by the data preprocessing module and outputs a real-time driving style type;
s103: the short-term driving condition prediction model simultaneously reads the real-time vehicle speed output by the data preprocessing module and the real-time driving style type output by the driving style recognition model, and outputs a predicted vehicle speed sequence within a period of time after the current time point;
s104: the working condition data synthesis module simultaneously reads a historical vehicle speed sequence within a period of time before the current time point output by the data preprocessing module and a predicted vehicle speed sequence within a period of time after the current time point output by the short-term running working condition prediction model, and splices the historical vehicle speed sequence updated in real time and the predicted vehicle speed sequence to construct combined working condition data containing the future working condition change trend;
s105: the real-time running condition recognition model reads characteristic parameters of the combined condition data output by the condition data synthesis module, and recognizes the real-time running condition type of the vehicle on line.
2. The vehicle driving condition online identification method based on random prediction and machine learning according to claim 1, characterized in that: in the step S101, the real-time driving data of the vehicle is read through the entire vehicle CAN bus, the historical vehicle speed sequence is updated in real time in the form of a sliding time window, and the historical driving data exceeding the width of the time window is discarded.
3. The vehicle driving condition online identification method based on random prediction and machine learning according to claim 1, characterized in that: the steps areIn step S102, a driving style recognition model is established based on a learning vector quantization neural network LVQNN, where the input of an input layer is a real-time vehicle speed, an accelerator pedal opening, an accelerator pedal change rate, a brake pedal opening, and a brake pedal change rate; the outputs of the output layers are respectively
The three driving style types respectively correspond to a mild type style, a normal type style and an aggressive type style.
4. The vehicle driving condition online identification method based on random prediction and machine learning according to claim 1, characterized in that: in the step S103, a short-term running condition prediction model is established by adopting a Markov chain stochastic prediction algorithm, the Markov chain takes the speed and the acceleration as the basis for dividing the working condition states, and a state transition probability matrix is established according to the transition frequency among the states; and filtering the output of the short-term running condition prediction model to obtain a final predicted vehicle speed sequence.
5. The vehicle driving condition online identification method based on random prediction and machine learning according to claim 1, characterized in that: in step S104, the total length of the combined operating condition data is equal to the sum of the length of the historical vehicle speed sequence and the length of the predicted vehicle speed sequence, the total length of the combined operating condition data is a fixed value, the length of the predicted vehicle speed sequence is variable, and the total length of the combined operating condition data is determined according to the root mean square error of the vehicle speed in the previous prediction period.
6. The vehicle driving condition online identification method based on random prediction and machine learning according to claim 5, characterized in that: root mean square error of the vehicle speed
The expression of (a) is:
in the formula (I), the compound is shown in the specification,
is the predicted vehicle speed value after the filtering processing>
The real vehicle speed value is obtained; />
The length of the prediction time domain is equal to the length of the prediction vehicle speed sequence; />
Setting the minimum value of the length of the prediction time domain as
Maximum value is->
Predicting a change step of ^ in the time domain>
And is and
，/>
is a positive integer; />
、/>
Two preset values of the root mean square error of the vehicle speed;
root mean square error of vehicle speed in the last prediction period
If the prediction time domain length of the last prediction period is less than
Then the length of the current prediction time domain is increased ≥ from the last prediction period>
Otherwise, the length of the current prediction time domain is kept consistent with the length of the last prediction period;
root mean square error of vehicle speed in the last prediction period
If the prediction time domain length of the last prediction period is larger than
Then the length of the current prediction time domain is decreased compared to the last prediction period>
Otherwise, the length of the current prediction time domain is kept consistent with the length of the previous prediction period;
root mean square error of vehicle speed in the last prediction period
The length of the current prediction time domain is kept consistent with the length of the last prediction period.
7. The vehicle driving condition online identification method based on random prediction and machine learning according to claim 6, characterized in that: minimum value of length of the prediction time domain
Is 10s and has a maximum value>
A change step of the prediction time domain of 30 s->
Is 5s, a preset value>
Is 3km/h, preset value>
Is 5km/h.
8. The vehicle driving condition online identification method based on random prediction and machine learning according to claim 1, characterized in that: in step S105, the real-time driving condition recognition model is built based on the learning vector quantization neural network LVQNN, wherein the input of the input layer is a key variable extracted from the characteristic parameters of the combined condition data through principal component analysis, and the key variable includes an average deceleration, a maximum acceleration, a maximum vehicle speed, an average driving vehicle speed, a low-speed driving time ratio, and an acceleration time ratio; the output of the output layer is
The three real-time running working condition types respectively correspond to a low-speed running working condition, a medium-speed running working condition and a high-speed running working condition.
9. The vehicle driving condition online identification method based on random prediction and machine learning according to claim 1, characterized in that: the training and testing process of the model in the vehicle running condition online identification method comprises the following steps:
s201: defining the type of the running working condition, and selecting typical standard cycle working conditions of various running working conditions from a standard cycle working condition library according to the amplitude of the vehicle speed and the change frequency of the vehicle speed;
s202: carrying out segment division on a typical standard cycle working condition, establishing a driving working condition data set, and dividing the driving working condition data set into a first data set and a second data set;
s203: taking the first data set as input, and training and testing the driving style recognition model;
s204: taking the first data set as input, calculating Markov state transition probability matrixes under various driving styles, and acquiring a predicted vehicle speed sequence of the tail end moment of a working condition segment;
s205: combining the working condition segments in the first data set with the corresponding predicted vehicle speed sequences to construct a combined working condition data set; training and testing a real-time driving condition recognition model by taking the combined working condition data set as input;
s206: and extracting the trained model parameters, implanting the trained model parameters into a corresponding model of the driving condition recognition system, and testing by adopting a second data set to obtain a complete driving condition recognition system.
10. The on-line vehicle driving condition identification method based on random prediction and machine learning of claim 9, characterized in that: the running condition data set is composed of a large number of working condition segments with unequal lengths, which are obtained by dividing a plurality of typical standard cycle working conditions through random segments;
the working condition segments are obtained by random number sampling, namely, two-dimensional random sampling is carried out on each standard working condition in two aspects of starting time and length, and the starting time of segment extraction
And the fragment length->
Expressed as: />
In the formula (I), the compound is shown in the specification,
the time length of the standard working condition; />
And &>
Is a segment longMaximum and minimum values of degrees; />
And &>
Is a random number between 0 and 1. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310183550.0A CN115859123B (en) | 2023-03-01 | 2023-03-01 | Vehicle driving condition online identification method based on random prediction and machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310183550.0A CN115859123B (en) | 2023-03-01 | 2023-03-01 | Vehicle driving condition online identification method based on random prediction and machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115859123A true CN115859123A (en) | 2023-03-28 |
| CN115859123B CN115859123B (en) | 2023-05-02 |
Family
ID=85659492
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310183550.0A Active CN115859123B (en) | 2023-03-01 | 2023-03-01 | Vehicle driving condition online identification method based on random prediction and machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115859123B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100023183A1 (en) * | 2008-07-24 | 2010-01-28 | Gm Global Technology Operations, Inc. | Adaptive vehicle control system with integrated maneuver-based driving style recognition |
| EP2610836A1 (en) * | 2011-12-30 | 2013-07-03 | Seat, S.A. | Device and method for the on-line prediction of the driving cycle in an automotive vehicle |
| CN107527113A (en) * | 2017-08-01 | 2017-12-29 | 北京理工大学 | A kind of operating mode Forecasting Methodology of hybrid car travel operating mode |
| CN110254417A (en) * | 2019-06-27 | 2019-09-20 | 清华大学苏州汽车研究院（吴江） | Method for controlling hybrid power vehicle based on actual condition and the double identifications of driving style |
| CN113297685A (en) * | 2021-07-27 | 2021-08-24 | 中汽研（天津）汽车工程研究院有限公司 | Vehicle operation condition mode identification method |
| CN115718886A (en) * | 2022-11-21 | 2023-02-28 | 长安大学 | Construction method and device for pure electric vehicle running condition |
-
2023
- 2023-03-01 CN CN202310183550.0A patent/CN115859123B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100023183A1 (en) * | 2008-07-24 | 2010-01-28 | Gm Global Technology Operations, Inc. | Adaptive vehicle control system with integrated maneuver-based driving style recognition |
| EP2610836A1 (en) * | 2011-12-30 | 2013-07-03 | Seat, S.A. | Device and method for the on-line prediction of the driving cycle in an automotive vehicle |
| CN107527113A (en) * | 2017-08-01 | 2017-12-29 | 北京理工大学 | A kind of operating mode Forecasting Methodology of hybrid car travel operating mode |
| CN110254417A (en) * | 2019-06-27 | 2019-09-20 | 清华大学苏州汽车研究院（吴江） | Method for controlling hybrid power vehicle based on actual condition and the double identifications of driving style |
| CN113297685A (en) * | 2021-07-27 | 2021-08-24 | 中汽研（天津）汽车工程研究院有限公司 | Vehicle operation condition mode identification method |
| CN115718886A (en) * | 2022-11-21 | 2023-02-28 | 长安大学 | Construction method and device for pure electric vehicle running condition |
Non-Patent Citations (6)
| Title |
|---|
| R LIESSNER 等: "Derivation of Real-World Driving Cycles Corresponding to Traffic Situation and Driving Style on the Basis of Markov Models and Cluster Analyses", 《DERIVATION OF REAL-WORLD DRIVING CYCLES CORRESPONDING TO TRAFFIC SITUATION AND DRIVING STYLE ON THE BASIS OF MARKOV MODELS AND CLUSTER ANALYSES》 * |
| RUI WANG 等: "Review of driving conditions prediction and driving style recognition based control algorithms for hybrid electric vehicles", 《REVIEW OF DRIVING CONDITIONS PREDICTION AND DRIVING STYLE RECOGNITION BASED CONTROL ALGORITHMS FOR HYBRID ELECTRIC VEHICLES》 * |
| YONGGANG LIU 等: "Prediction of vehicle driving conditions with incorporation of stochastic forecasting and machine learning and a case study in energy management of plug-in hybrid electric vehicles", 《MECHANICAL SYSTEMS AND SIGNAL PROCESSING》 * |
| 刘炳文 等: "工况特征参数对客车燃油经济性的影响", 《科学技术与工程》 * |
| 刘荣昌 等: "驾驶风格和行驶工况对纯电动汽车续驶里程的影响研究", 汽车 《2020中国工程学会年会论文集》 * |
| 黄春: "电动汽车城市行驶工况构建与驾驶风格识别研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115859123B (en) | 2023-05-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110775065B (en) | Hybrid electric vehicle battery life prediction method based on working condition recognition | |
| CN110245390B (en) | Automobile engine oil consumption prediction method based on RS-BP neural network | |
| CN113222385B (en) | Method for constructing and evaluating driving condition of electric automobile | |
| CN112182962A (en) | Hybrid electric vehicle running speed prediction method | |
| CN112327168A (en) | XGboost-based electric vehicle battery consumption prediction method | |
| CN106875681B (en) | A kind of traffic congestion tendency prediction technique and system | |
| CN111198501B (en) | Method for determining fuel equivalent factor by RBF neural network | |
| CN115166534A (en) | Method and system for predicting remaining service life of lithium ion battery | |
| CN110059938B (en) | Power distribution network planning method based on association rule driving | |
| CN115051929A (en) | Network fault prediction method and device based on self-supervision target perception neural network | |
| CN107301478A (en) | A kind of cable run short-term load forecasting method | |
| CN114358367A (en) | Non-invasive load decomposition method and device | |
| CN117056765A (en) | Vehicle speed multi-time scale prediction method, system, equipment, medium and terminal | |
| CN115859123B (en) | Vehicle driving condition online identification method based on random prediction and machine learning | |
| CN110097757B (en) | Intersection group critical path identification method based on depth-first search | |
| Yang et al. | Optimization of fuzzy controller based on genetic algorithm | |
| CN113537620B (en) | Vehicle speed prediction method based on Markov model optimization and working condition recognition | |
| CN112489422B (en) | Traffic flow prediction method based on algorithm model analysis | |
| CN113788007A (en) | Layered real-time energy management method and system | |
| CN114169226A (en) | Short-term power load prediction method, computer device, and storage medium | |
| CN117901724B (en) | Control method, system and equipment for thermal management system of pure electric vehicle | |
| CN111563543A (en) | Method and device for cleaning wind speed-generated power data of wind turbine generator | |
| CN113569948B (en) | Electric automobile working condition classification and evaluation method and system | |
| Xiao et al. | A time Series Prediction method based on self-adaptive RBF neural network | |
| CN110807290B (en) | Road gradient-containing automobile operation condition design method based on self-adaptive Markov chain evolution |
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 |