CN113485102A - Method for identifying vehicle running condition based on long-term and short-term memory neural network - Google Patents
Method for identifying vehicle running condition based on long-term and short-term memory neural network Download PDFInfo
- Publication number
- CN113485102A CN113485102A CN202110693313.XA CN202110693313A CN113485102A CN 113485102 A CN113485102 A CN 113485102A CN 202110693313 A CN202110693313 A CN 202110693313A CN 113485102 A CN113485102 A CN 113485102A
- Authority
- CN
- China
- Prior art keywords
- working condition
- neural network
- condition
- long
- term memory
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 230000006403 short-term memory Effects 0.000 title claims abstract description 30
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 27
- 230000007787 long-term memory Effects 0.000 title claims abstract description 27
- 238000011217 control strategy Methods 0.000 claims abstract description 40
- 238000003062 neural network model Methods 0.000 claims abstract description 33
- 230000015654 memory Effects 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000012795 verification Methods 0.000 claims abstract description 10
- 238000012216 screening Methods 0.000 claims abstract description 7
- 239000011159 matrix material Substances 0.000 claims description 52
- 230000006870 function Effects 0.000 claims description 20
- 238000013135 deep learning Methods 0.000 claims description 12
- 238000012546 transfer Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000001413 cellular effect Effects 0.000 claims description 3
- 230000001133 acceleration Effects 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000005265 energy consumption Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 239000002283 diesel fuel Substances 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000009347 mechanical transmission Effects 0.000 description 1
- 238000013021 overheating Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
- G05B13/027—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion using neural networks only
Landscapes
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Hybrid Electric Vehicles (AREA)
Abstract
The invention belongs to the technical field of running condition recognition, and discloses a method for recognizing a vehicle running condition based on a long-term and short-term memory neural network, which is used for analyzing and screening collected running data; dividing the analyzed and screened data into a training data set and a checking data set; constructing a long-short term memory neural network model; training the long and short term memory neural network model by using a training data set to obtain a trained long and short term memory neural network model; evaluating the trained long-term and short-term memory neural network model by using the calibration data set; extracting the parameters of the long and short term memory neural network model passing the verification; establishing a driving condition classifier according to the extracted parameter data; and integrating the running condition classifier into an AMT control program, and identifying the running condition of the vehicle on line by using the running condition classifier. According to the invention, a working condition control strategy is established according to the driving working condition, so that the reliability, the safety, the economy and the comfort of the hybrid power heavy commercial vehicle are improved.
Description
Technical Field
The invention belongs to the technical field of running condition recognition, and particularly relates to a method for recognizing a running condition of a vehicle based on a long-term and short-term memory neural network.
Background
The hybrid power heavy commercial vehicle has large full-load mass (generally about 50 tons), and the large mass has large inertia, so that the acceleration and braking time is longer, and the transmission of the hybrid power heavy commercial vehicle is a device for optimizing the output of an engine and a motor according to the driving force required by the vehicle by changing the transmission ratio of a transmission system of the hybrid power heavy commercial vehicle aiming at the characteristics of the vehicle. The transmission ratios of hybrid heavy commercial vehicles generally range between 0.75 and 20, with reasonable gears and ratios being critical to the operation of the hybrid heavy commercial vehicle. Therefore, the hybrid heavy-duty commercial vehicle with an Automatic Mechanical Transmission (AMT) adopts the same gear shifting control logic under different working conditions, which can cause the problems of clutch overheating burning, excessive emission, increased energy consumption (oil consumption plus electricity consumption), excessive noise and the like, and can also endanger the safety of vehicles and personnel. For example, under congestion conditions (such as traffic jam in city centers and high-speed toll stations), frequent gear shifting can be caused by adopting a general gear shifting strategy, so that the clutch is overheated, and further, the gear shifting is failed and even the clutch is burnt; when the speed is in a low-speed working condition (such as suburbs, villages and towns), the adoption of a general gear shifting strategy can cause untimely gear shifting, and cause insufficient power or no speed; when the vehicle is in a medium-speed working condition (provincial road, national road and the like) and a high-speed working condition (expressway), the adoption of a general gear shifting strategy can cause slow gear shifting or no gear shifting of overtaking, the overtaking is powerless, the overtaking is slow, and the safety of the vehicle is influenced. In order to ensure the safety of the vehicle, increase the service life of components, reduce emission and reduce energy consumption (oil consumption plus power consumption), the mechanical automatic gearbox of the hybrid heavy commercial vehicle adopts corresponding control strategies for different working conditions.
The research on the running working condition of the hybrid power heavy commercial vehicle is almost not available, and the method aims at the identification and gear shifting strategies under different working conditions and is particularly a difficult problem in the field of AMT of the hybrid power heavy commercial vehicle.
Disclosure of Invention
The invention aims to provide a method for identifying the driving condition of a vehicle based on a long-term and short-term memory neural network, and solves the problem that an AMT control strategy of a hybrid power heavy commercial vehicle is automatically adjusted along with the driving condition.
The invention is realized by the following technical scheme:
a method for identifying the running condition of a vehicle based on a long-term and short-term memory neural network comprises the following steps:
s1, collecting the driving data of the vehicle on different roads, and analyzing and screening the collected driving data;
s2, dividing the analyzed and screened data into two groups, wherein one group is used as a training data set, and the other group is used as a verification data set;
s3, constructing a long-term and short-term memory neural network model according to the screened data;
s4, training the long-short term memory neural network model by using the training data set to obtain a trained long-short term memory neural network model;
s5, evaluating the trained long-term and short-term memory neural network model by using the check data set;
s6, extracting the parameters of the verified long-term and short-term memory neural network model;
s7, establishing a driving condition classifier according to the extracted parameter data;
and S8, integrating the driving condition classifier into an AMT control program, and identifying the driving condition of the vehicle on line by using the driving condition classifier.
Further, in S8, the working condition classifications include a traffic jam working condition, a low-speed working condition, a medium-speed working condition, and a high-speed working condition, and corresponding working condition control strategies are established according to the driving working condition, and are respectively a traffic jam working condition control strategy, a low-speed working condition control strategy, a medium-speed working condition control strategy, a high-speed working condition control strategy, and a general ordinary control strategy.
Further, the driving condition classifier is used for identifying the driving condition of the vehicle on line, and the process of establishing the corresponding working condition control strategy according to the driving condition is as follows: firstly, judging whether the working condition is a high-speed working condition, if so, executing a high-speed working condition gear shifting control strategy, and otherwise, continuously judging;
secondly, judging whether the working condition is a medium-speed working condition, if so, executing a medium-speed working condition gear shifting control strategy, and otherwise, continuously judging;
judging whether the working condition is a low-speed working condition or not, if so, executing a low-speed working condition gear shifting control strategy, and otherwise, continuing to judge;
judging whether the working condition is a low-speed working condition or not, if so, executing a low-speed working condition gear shifting control strategy, and otherwise, continuing to judge;
and finally, judging whether the vehicle is in a traffic jam working condition, if so, executing a gear shifting control strategy in the traffic jam working condition, and otherwise, executing a gear shifting strategy in a general working condition.
Further, in S1, the running data includes the vehicle speed, the accelerator pedal opening, the brake pedal opening, and the steering radius.
Further, in S1, the different road surfaces include an urban road surface, a provincial road, a national road, and a highway.
Further, in S2, the data after analysis and screening is randomly divided into two groups in the ratio of 70% and 30%, one group of 70% is used as a training data set, and the other group of 30% is used as a verification data set.
Further, the long-term and short-term memory neural network model comprises an input layer, an Lstm layer, a full connection layer, a softmax layer and a classification layer;
in S7, the extracted parameters of the long-term and short-term memory neural network model include an input weight IW of the Lstm layer, a period weight RW of the Lstm layer, an offset B of the Lstm layer, a weight FW of the fully-connected layer, and an offset FB of the fully-connected layer.
Further, in S7, constructing a full link layer of the driving condition classifier using the weight FW of the full link layer and the offset FB of the full link layer; in S8, the calculation process of the driving condition classifier is:
(1) the output gtt, itt, ftt and c of the Lstm layer of the driving condition classifier are comparedt-1The vehicle speed control method is characterized in that the vehicle speed control method is input into a full connection layer, the output of the full connection layer is the probability that the current vehicle running working condition belongs to four working conditions, namely a traffic jam working condition, a low-speed working condition, a medium-speed working condition and a high-speed working condition, and the specific formula is as follows:
OUT＝FW*(gtt*itt+ftt*ct-1)+FB
the OUT represents the probability that the vehicle is in a traffic jam working condition, a low-speed working condition, a medium-speed working condition and a high-speed working condition at the current moment, and is a 1-4 matrix; FW represents the weight of the fully-connected layer, and FB represents the offset of the fully-connected layer; c. Ct-1Represents the cell state at the previous moment, gtt represents the output value of the selected memory gate, itt represents the output value of the input gate, and ftt represents the output value of the forgetting gate;
(2) calculating the type of the working condition:
[～，Kind_OUT]＝max(OUT)
wherein, kid _ OUT represents the type of the working condition, which is one of four numbers 1, 2, 3 and 4, 1 traffic jam working condition, 2 low-speed working condition, 3 medium-speed working condition and 4 high-speed working condition, max represents the max function of MATLAB, the output is a 1 x 2 matrix, the first number of the matrix is the maximum value, the second number is the position of the maximum value, the working condition is determined by the position of the maximum value, and the symbol represents the first value which ignores the output.
Further, the forget gate formula is:
ftt＝logsig(G(fInd))；
wherein ftt represents the output value of the forgetting gate, which is a matrix of 1 × 100; g (fnnd) represents the 1 st to 100 th values of G, which is a 1 x 100 matrix, logsig represents the Log-sigmoid transfer function in deep learning;
the input gate formula is:
itt＝logsig(G(iInd))；
wherein itt represents the output of the input gate, which is a matrix of 1 x 100; g (iind) represents the 101 st to 200 th values of G, which is a matrix of 1 x 100, logsig represents the Log-sigmoid function in deep learning;
the memory gate formula is selected as:
gtt＝tanh(G(zInd))；
wherein gtt represents the output value of the selection memory gate, and is a matrix of 1 × 100; g (zind) represents the 201 st to 300 th values of G, which is a matrix of 1 × 100, and tanh represents the tangent function in deep learning;
the output gate formula is:
ott＝logsig(G(oInd))；
wherein ott represents the output value of the output gate, which is a matrix of 1 × 100, G (oind) represents the 301 st to 400 th values of G, which is a matrix of 1 × 100, and logsig represents the Log-sigmoid transfer function in deep learning;
the linear gate formula is:
G＝IW*xt+RW*ht-1+B；
wherein, G represents the output value of all gate linear computing parts and is a matrix of 1 x 400; IW represents the input weight of Lstm layer, xtRepresenting the characteristic value at the current moment, RW representing the period weight of Lstm layer, ht-1Representing the hidden state at the previous instant and B the offset of the Lstm layer.
Further, the hidden state formula is:
ht＝tanh(ct)*ott；
wherein, the operation symbol is matrix operation, htRepresenting a hidden state at the current time, ctRepresenting the cell state at the current time;
cell state formula:
ct＝gtt*itt+ftt*c t-1；
wherein, ctRepresenting the cell state at the present moment, ct-1Representing the cellular state at the last moment.
Compared with the prior art, the invention has the following beneficial technical effects:
the invention discloses a method for identifying the running condition of a vehicle based on a long-short term memory neural network, which is specially used for solving the dependence on the running condition of the vehicle when a hybrid heavy commercial vehicle is driven by various powers (generally diesel oil and electric power) and when the braking energy of a motor is recovered. The method is combined with the driving intention recognition, so that the control problem that the energy is recovered as much as possible under different working conditions is effectively solved, the running safety of the vehicle is ensured, and the newly found new energy automobile industry problems that the braking force is insufficient due to the recovery of the braking energy when the hybrid power vehicle is at a high speed, the hybrid power is involved to exceed the working condition requirement when the hybrid power vehicle is at a low speed, and various braking forces are not coordinated during braking, so that the vehicle is impacted and even moves uncontrollably are solved. The running condition of the hybrid power heavy commercial vehicle is identified by researching the running condition of the hybrid power heavy commercial vehicle and applying the neural network, and the gear shifting strategy of the hybrid power heavy commercial vehicle AMT is optimized based on the identified working condition, so that the reliability, the safety, the economy and the comfort of the hybrid power heavy commercial vehicle are improved, and a strong support is provided for energy conservation, emission reduction, and environment and energy pressure relief.
Drawings
FIG. 1 is a flow chart of a method for identifying vehicle driving conditions based on a long-short term memory neural network according to the present invention;
FIG. 2 is a schematic diagram of the layer structure of a long short term memory neural network (LSTM);
FIG. 3 is a schematic diagram of the cell structure of a long short term memory neural network (LSTM);
FIG. 4 is data after a traffic jam condition processing in Nanjing; FIG. 4(a) represents collected vehicle speed data, FIG. 4(b) represents collected accelerator pedal opening data, FIG. 4(c) represents collected brake pedal opening data, FIG. 4(d) represents collected turning radius data, and FIG. 4(e) represents collected acceleration data;
FIG. 5 is a long short term memory neural network (LSTM) neural network training accuracy curve;
FIG. 6 is a flow chart of AMT control of the hybrid heavy-duty commercial vehicle.
Detailed Description
The present invention will now be described in further detail with reference to specific examples, which are intended to be illustrative, but not limiting, of the invention.
As shown in FIG. 1, the invention discloses a method for identifying the running condition of a vehicle based on a long-term and short-term memory neural network, which comprises the following steps:
s1, collecting the driving data of the vehicle on different roads, and analyzing and screening the collected driving data;
s2, dividing the analyzed and screened data into two groups, wherein one group is used as a training data set, and the other group is used as a verification data set;
s3, constructing a long-term and short-term memory neural network model according to the screened data;
s4, training the long-short term memory neural network model by using the training data set to obtain a trained long-short term memory neural network model;
s5, evaluating the trained long-term and short-term memory neural network model by using the check data set;
s6, extracting the parameters of the verified long-term and short-term memory neural network model;
s7, establishing a driving condition classifier according to the extracted parameter data;
and S8, integrating the driving condition classifier into an AMT control program of the hybrid heavy commercial vehicle, analyzing the driving condition of the vehicle on line by using the driving condition classifier, and establishing a corresponding working condition control strategy according to the driving condition.
The method mainly utilizes the long-term and short-term memory neural network LSTM to identify the driving condition of the hybrid power heavy commercial vehicle, and solves the problem that the AMT control strategy of the hybrid power heavy commercial vehicle is automatically adjusted along with the driving condition. The method specifically comprises the following steps:
(1) the CANalyzer software is used for acquiring original driving data of the hybrid heavy commercial vehicle on the road surfaces of urban areas, provincial roads, national roads, high speeds and the like through an OBD interface of a vehicle public CAN communication network, wherein the original driving data comprises vehicle speed, accelerator pedal opening, brake pedal opening, steering radius and the like.
The vehicle speed is obtained from a FrontAxleSpeed signal of an EBC2 message sent by an ABS (anti-lock brake system) of the automobile, the accelerator pedal opening is obtained from an AccelPedalPos1 signal of an EEC2 message sent by an engine controller, the brake pedal opening is obtained from a BrakeSwitch signal of a CCVS message sent by the engine controller, and the steering radius is calculated from a RelationsSpeedFrontAxleftWheel and a RlativeSpeedFrontAxleRightWheel signal of an EBC2 message sent by the ABS of the automobile.
(2) Analyzing screening data, initially grouping the acquired data manually by using a CANalyzer, labeling the data, converting the data format by using a data derivation function of the CANalyzer, converting the data format from original blf into a mat format which can be identified by MATLAB, cleaning invalid data by using a data processing function of the MATLAB to obtain a data set { XData, YData } consisting of 8500 groups of data, wherein each group of data comprises a matrix X of 5 × 60 and 1 classification label Y, and 5 represents 5 input characteristics of a neural network: the vehicle speed is 1, the accelerator pedal is 2, the brake pedal is 3, the turning radius is 4, the acceleration is 5 (the acceleration is obtained by subtracting the vehicle speed of the last 1 second from the vehicle speed of the first 1 second), 60 represents 60 rows of characteristic values of a 60-second time sequence, and the classification label is one of a traffic jam working condition, a low-speed working condition, a medium-speed working condition and a high-speed working condition.
(3) The data after analysis and screening are randomly divided into two groups according to the proportion of 70% and 30%, one group of 70% is used as a training data set { XTrain, YTrain }, and the other group of 30% is used as a verification data set { Xtest, YTest }, so that single group of characteristic data is checked.
(4) Based on the screened data and the neural network toolbox of MATLAB software, a long-short term memory neural network model was constructed as shown in FIG. 2, where xtThe input features, i.e. the 5 feature values at the current time, are represented, t-1, t, t +1, …, s representing a time sequence, where s is 60. Setting a neural network model and training parameters according to experience and continuous trial and error adjustment: inputting a feature dimension, featureDimension, of 5; the number of hidden layer units numHiddenUnits is 100; the classification number numClasses is 4; the maximum iteration parameter max _ epochs is 500. Generating the neural network model lstm _ layers includes 5 layers: input layer, Lstm layer, full link layer, softmax layer and classification layer.
Loading an LSTM _ layers neural Network model through an attachment program Deep Network Designer of MATLAB, checking a neural Network framework, checking multidimensional input layer parameters, checking LSTM layer parameters, checking full-connection layer parameters, checking softmax layer parameters and checking classification layer parameters.
(5) The long-short term memory neural network model is trained using a training dataset { XTrain, YTrain }.
Calling a train network function of MATLAB to train a network model, wherein the input of the function is as follows: training data set { XTrain, YTrain }, neural network model lstm _ layers, and training control parameters (max _ epochs ═ 500, etc.); the output of the function is the neural network model lstm _ net. lstm _ layers is an input model developed for training and is set manually; lstm _ net is a model of the training output, which is automatically output after training.
Looking up the training result and referring to fig. 5, if the learning effect does not meet the expected requirement, namely the expected accuracy rate does not reach the set target of 98%, adjusting the training parameters to continue learning until the learning reaches the set target required.
(6) And (3) evaluating the trained long-short term memory neural network model lstm _ net by using a verification data set { Xtest, YTest }, if the verification effect does not meet the expected requirement, namely the expected accuracy rate does not reach 98%, adjusting the training parameters to continue learning until the verification meets the requirement.
(7) Extracting the parameters of the verified long-short term memory neural network model lstm _ net:
the input weight IW of the Lstm layer is a matrix of 5 × 400, the period weight RW of the Lstm layer is a matrix of 100 × 400, the offset B of the Lstm layer is a matrix of 1 × 400, the weight FW of the fully-connected layer is a matrix of 4 × 100, and the weight FB of the fully-connected layer is a matrix of 1 × 4.
(8) Developing a working condition recognition algorithm of the hybrid power heavy commercial vehicle according to the extracted data, namely a driving working condition classifier, which specifically comprises the following steps: and constructing a full connection layer of the driving condition classifier by using the weight FW of the full connection layer and the offset FB of the full connection layer.
(9) Integrating the running condition classifier into an AMT control program, analyzing the running condition of the vehicle on line by using the running condition classifier, and establishing a corresponding working condition control strategy according to the running condition.
And inputting the input characteristic matrix of the full connection layer into the full connection layer, and calculating to obtain the output of the full connection layer, namely the probability that the current vehicle running working condition belongs to four working conditions, namely a traffic jam working condition, a low-speed working condition, a medium-speed working condition and a high-speed working condition.
the inputs to the LSTM layer (see fig. 3) at time t are: at the present momentCharacteristic value xtMatrix 1 x 5, cell status c at the previous momentt-1Matrix of 1 x 100, and hidden state h at the previous momentt-1Matrix of 1 x 100.
Linear gate formula:
G＝IW*xt+RW*ht-1+B
wherein, G represents the output value of all gate linear computing parts and is a matrix of 1 x 400; IW represents the input weight of Lstm layer, xtRepresenting the characteristic value at the current moment, RW representing the period weight of Lstm layer, ht-1Representing the hidden state at the previous instant and B the offset of the Lstm layer.
Forget gate formula:
ftt＝logsig(G(fInd))
wherein ftt represents the output value of the forgetting gate, which is a matrix of 1 × 100; g (fnnd) represents the 1 st to 100 th values of G, a matrix of 1 × 100, logsig represents the Log-sigmoid transfer function in deep learning.
Input gate formula:
itt＝logsig(G(iInd))
wherein itt represents the output of the input gate, which is a matrix of 1 x 100; g (iind) represents the 101 th to 200 th values of G, a matrix of 1 x 100, logsig represents the Log-sigmoid function in deep learning.
Selecting a memory gate formula:
gtt＝tanh(G(zInd))
wherein gtt represents the output value of the selection memory gate, and is a matrix of 1 × 100; g (zind) represents the 201 st to 300 th values of G, which is a matrix of 1 × 100, and tanh represents the tangent function in deep learning. g output gate formula:
ott＝logsig(G(oInd))
wherein ott represents the output value of the output gate, which is 1 × 100 matrix, G (oind) represents the 301 st to 400 th values of G, which is 1 × 100 matrix, and logsig represents the Log-sigmoid transfer function in deep learning.
Updating the cell state formula:
ct＝gtt*itt+ftt*c t-1
wherein the operation symbols are matrix operation, ctRepresenting the cell state at the present moment, ct-1Representing the cellular state at the last moment.
Updating the hidden state formula:
ht＝tanh(ct)*ott
wherein, the operation symbol is matrix operation, htRepresenting the hidden state at the current time.
At time t, the output calculation formula of the full connection layer is as follows:
OUT＝FW*ct+FB
the operation symbols x and + are matrix operation, and OUT represents the probability that the vehicle is in four working conditions, namely a traffic jam working condition, a low-speed working condition, a medium-speed working condition and a high-speed working condition at the current moment, and is a 1 x 4 matrix.
Calculating a working condition formula:
[～，Kind_OUT]＝max(OUT)
wherein, kid _ OUT represents the type of the operating condition, which is one of four numbers 1, 2, 3, 4, 1 traffic jam, 2 low speed, 3 medium speed and 4 high speed, max represents the max function of MATLAB, the input of which can be a matrix, the output of which can be a 1 x 2 matrix, the first number of the matrix is the maximum value, the second number is the position of the maximum value, the classifier does not care about the value of the maximum value, only needs the position of the maximum value, kid _ OUT, to determine the operating condition, and the symbol represents the first value to ignore the output.
As shown in fig. 4, it can be seen from fig. 4(a) that the vehicle speed of the feature 1 is low, from fig. 4(b) and 4(c) that the brake of the feature 3 is immediately performed after the accelerator pedal of the feature 2 is hardened, from fig. 4(d) that the vehicle turns at a low speed at the traffic light of the feature 4, and from fig. 4(e) that the acceleration of the feature 5 repeatedly fluctuates in a wide range. Is a set of input data labeled traffic congestion conditions, and the OUT identified by a verified classifier is [0.9982, 0.4473, 0.0258, 0.2040], which is seen to have the highest probability of being in a traffic congestion condition, with a probability of 0.9982.
(9) Integrating the running condition classifier into an AMT control program of the hybrid power heavy commercial vehicle, and according to the four obtained classifications: the method comprises the following steps of respectively developing a traffic jam working condition control strategy, a low-speed working condition control strategy, a medium-speed working condition control strategy, a high-speed working condition control strategy and a general common control strategy under the traffic jam working condition, the low-speed working condition, the medium-speed working condition and the high-speed working condition.
As shown in fig. 6, the process of analyzing the driving condition of the vehicle on line by using the driving condition classifier includes:
firstly, judging whether the working condition is a high-speed working condition, if so, executing a high-speed working condition gear shifting control strategy, and otherwise, continuously judging;
secondly, judging whether the working condition is a medium-speed working condition, if so, executing a medium-speed working condition gear shifting control strategy, and otherwise, continuously judging;
judging whether the working condition is a low-speed working condition or not, if so, executing a low-speed working condition gear shifting control strategy, and otherwise, continuing to judge;
judging whether the working condition is a low-speed working condition or not, if so, executing a low-speed working condition gear shifting control strategy, and otherwise, continuing to judge;
and finally, judging whether the vehicle is in a traffic jam working condition, if so, executing a gear shifting control strategy in the traffic jam working condition, and otherwise, executing a gear shifting strategy in a general working condition.
Claims (10)
1. A method for identifying the running condition of a vehicle based on a long-term and short-term memory neural network is characterized by comprising the following steps:
s1, collecting the driving data of the vehicle on different roads, and analyzing and screening the collected driving data;
s2, dividing the analyzed and screened data into two groups, wherein one group is used as a training data set, and the other group is used as a verification data set;
s3, constructing a long-term and short-term memory neural network model according to the screened data;
s4, training the long-short term memory neural network model by using the training data set to obtain a trained long-short term memory neural network model;
s5, evaluating the trained long-term and short-term memory neural network model by using the check data set;
s6, extracting the parameters of the verified long-term and short-term memory neural network model;
s7, establishing a driving condition classifier according to the extracted parameter data;
and S8, integrating the driving condition classifier into an AMT control program, and identifying the driving condition of the vehicle on line by using the driving condition classifier.
2. The method according to claim 1, wherein in S8, the working condition classifications include a traffic jam working condition, a low speed working condition, a medium speed working condition and a high speed working condition, and corresponding working condition control strategies are established according to the driving working conditions, and are respectively a traffic jam working condition control strategy, a low speed working condition control strategy, a medium speed working condition control strategy, a high speed working condition control strategy and a general common control strategy.
3. The method for identifying the vehicle running condition based on the long-short term memory neural network as claimed in claim 2, wherein the running condition classifier is used for identifying the running condition of the vehicle on line, and the process of establishing the corresponding working condition control strategy according to the running condition is as follows:
firstly, judging whether the working condition is a high-speed working condition, if so, executing a high-speed working condition gear shifting control strategy, and otherwise, continuously judging;
secondly, judging whether the working condition is a medium-speed working condition, if so, executing a medium-speed working condition gear shifting control strategy, and otherwise, continuously judging;
judging whether the working condition is a low-speed working condition or not, if so, executing a low-speed working condition gear shifting control strategy, and otherwise, continuing to judge;
judging whether the working condition is a low-speed working condition or not, if so, executing a low-speed working condition gear shifting control strategy, and otherwise, continuing to judge;
and finally, judging whether the vehicle is in a traffic jam working condition, if so, executing a gear shifting control strategy in the traffic jam working condition, and otherwise, executing a gear shifting strategy in a general working condition.
4. The method for identifying the driving conditions of the vehicle based on the long-short term memory neural network as claimed in claim 1, wherein in S1, the driving data comprises vehicle speed, accelerator pedal opening, brake pedal opening and steering radius.
5. The method for identifying the driving conditions of the vehicle based on the long and short term memory neural network as claimed in claim 1, wherein in S1, the different road surfaces comprise urban road surfaces, provincial road surfaces, national road surfaces and expressway surfaces.
6. The method for identifying the driving conditions of the vehicle based on the long-short term memory neural network as claimed in claim 1, wherein in S2, the analyzed and screened data are randomly divided into two groups according to the proportion of 70% and 30%, one group of 70% is used as the training data set, and the other group of 30% is used as the verification data set.
7. The method for identifying the driving condition of the vehicle based on the long-short term memory neural network is characterized in that the long-short term memory neural network model comprises an input layer, an Lstm layer, a full connection layer, a softmax layer and a classification layer;
in S7, the extracted parameters of the long-term and short-term memory neural network model include an input weight IW of the Lstm layer, a period weight RW of the Lstm layer, an offset B of the Lstm layer, a weight FW of the fully-connected layer, and an offset FB of the fully-connected layer.
8. The method for identifying the driving condition of the vehicle based on the long-short term memory neural network as claimed in claim 7, wherein in S7, the full link layer of the driving condition classifier is constructed by using the weight FW of the full link layer and the offset FB of the full link layer; in S8, the calculation process of the driving condition classifier is:
(1) the output gtt, itt, ftt and c of the Lstm layer of the driving condition classifier are comparedt-1When the current vehicle running working condition is input into the full connection layer, the output of the full connection layer, namely the current vehicle running working condition belongs to the traffic jam working condition and the low-speed working conditionThe probability of four working conditions of medium-speed working condition and high-speed working condition is as follows:
OUT＝FW*(gtt*itt+ftt*ct-1)+FB
the OUT represents the probability that the vehicle is in a traffic jam working condition, a low-speed working condition, a medium-speed working condition and a high-speed working condition at the current moment, and is a 1-4 matrix; FW represents the weight of the fully-connected layer, and FB represents the offset of the fully-connected layer; c. Ct-1Represents the cell state at the previous moment, gtt represents the output value of the selected memory gate, itt represents the output value of the input gate, and ftt represents the output value of the forgetting gate;
(2) calculating the type of the working condition:
[～，Kind_OUT]＝max(OUT)
wherein, kid _ OUT represents the type of the working condition, which is one of four numbers 1, 2, 3 and 4, 1 traffic jam working condition, 2 low-speed working condition, 3 medium-speed working condition and 4 high-speed working condition, max represents the max function of MATLAB, the output is a 1 x 2 matrix, the first number of the matrix is the maximum value, the second number is the position of the maximum value, the working condition is determined by the position of the maximum value, and the symbol represents the first value which ignores the output.
9. The method for identifying the driving condition of the vehicle based on the long-short term memory neural network as claimed in claim 8, wherein the forgetting gate formula is as follows:
ftt＝logsig(G(fInd))；
wherein ftt represents the output value of the forgetting gate, which is a matrix of 1 × 100; g (fnnd) represents the 1 st to 100 th values of G, which is a 1 x 100 matrix, logsig represents the Log-sigmoid transfer function in deep learning;
the input gate formula is:
itt＝logsig(G(iInd))；
wherein itt represents the output of the input gate, which is a matrix of 1 x 100; g (iind) represents the 101 st to 200 th values of G, which is a matrix of 1 x 100, logsig represents the Log-sigmoid function in deep learning;
the memory gate formula is selected as:
gtt＝tanh(G(zInd))；
wherein gtt represents the output value of the selection memory gate, and is a matrix of 1 × 100; g (zind) represents the 201 st to 300 th values of G, which is a matrix of 1 × 100, and tanh represents the tangent function in deep learning;
the output gate formula is:
ott＝logsig(G(oInd))；
wherein ott represents the output value of the output gate, which is a matrix of 1 × 100, G (oind) represents the 301 st to 400 th values of G, which is a matrix of 1 × 100, and logsig represents the Log-sigmoid transfer function in deep learning;
the linear gate formula is:
G＝IW*xt+RW*ht-1+B；
wherein, G represents the output value of all gate linear computing parts and is a matrix of 1 x 400; IW represents the input weight of Lstm layer, xtRepresenting the characteristic value at the current moment, RW representing the period weight of Lstm layer, ht-1Representing the hidden state at the previous instant and B the offset of the Lstm layer.
10. The method for identifying the driving condition of the vehicle based on the long-term and short-term memory neural network as claimed in claim 9, wherein the hidden state formula is as follows:
ht＝tanh(ct)*ott；
wherein, the operation symbol is matrix operation, htRepresenting a hidden state at the current time, ctRepresenting the cell state at the current time;
cell state formula:
ct＝gtt*itt+ftt*ct-1；
wherein, ctRepresenting the cell state at the present moment, ct-1Representing the cellular state at the last moment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110693313.XA CN113485102A (en) | 2021-06-22 | 2021-06-22 | Method for identifying vehicle running condition based on long-term and short-term memory neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110693313.XA CN113485102A (en) | 2021-06-22 | 2021-06-22 | Method for identifying vehicle running condition based on long-term and short-term memory neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113485102A true CN113485102A (en) | 2021-10-08 |
Family
ID=77935855
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110693313.XA Pending CN113485102A (en) | 2021-06-22 | 2021-06-22 | Method for identifying vehicle running condition based on long-term and short-term memory neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113485102A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115782595A (en) * | 2022-12-06 | 2023-03-14 | 东南大学 | Electric bus instantaneous energy consumption estimation method based on energy recovery state |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108197648A (en) * | 2017-12-28 | 2018-06-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models |
| CN108595632A (en) * | 2018-04-24 | 2018-09-28 | 福州大学 | A kind of hybrid neural networks file classification method of fusion abstract and body feature |
| CN110515456A (en) * | 2019-08-14 | 2019-11-29 | 东南大学 | EEG signals emotion method of discrimination and device based on attention mechanism |
| CN111079348A (en) * | 2019-12-27 | 2020-04-28 | 北京旋极信息技术股份有限公司 | Method and device for detecting slowly-varying signal |
| US20200242382A1 (en) * | 2019-01-25 | 2020-07-30 | Fujitsu Limited | Deep learning model used for driving behavior recognition and training apparatus and method thereof |
| CN111832225A (en) * | 2020-07-07 | 2020-10-27 | 重庆邮电大学 | Method for constructing driving condition of automobile |
| CN111967566A (en) * | 2019-05-20 | 2020-11-20 | 天津科技大学 | Edge computing offloading decision making based on long-short term memory neural network in Internet of vehicles environment |
| CN112061136A (en) * | 2020-08-26 | 2020-12-11 | 西北工业大学 | Automatic analysis method for driver driving behavior based on long-term and short-term memory network |
| CN112265546A (en) * | 2020-10-26 | 2021-01-26 | 吉林大学 | Networked automobile speed prediction method based on time-space sequence information |
| CN112435503A (en) * | 2020-10-30 | 2021-03-02 | 江苏大学 | Intelligent automobile active collision avoidance method for identifying intention of high-risk pedestrians |
-
2021
- 2021-06-22 CN CN202110693313.XA patent/CN113485102A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108197648A (en) * | 2017-12-28 | 2018-06-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models |
| CN108595632A (en) * | 2018-04-24 | 2018-09-28 | 福州大学 | A kind of hybrid neural networks file classification method of fusion abstract and body feature |
| US20200242382A1 (en) * | 2019-01-25 | 2020-07-30 | Fujitsu Limited | Deep learning model used for driving behavior recognition and training apparatus and method thereof |
| CN111967566A (en) * | 2019-05-20 | 2020-11-20 | 天津科技大学 | Edge computing offloading decision making based on long-short term memory neural network in Internet of vehicles environment |
| CN110515456A (en) * | 2019-08-14 | 2019-11-29 | 东南大学 | EEG signals emotion method of discrimination and device based on attention mechanism |
| CN111079348A (en) * | 2019-12-27 | 2020-04-28 | 北京旋极信息技术股份有限公司 | Method and device for detecting slowly-varying signal |
| CN111832225A (en) * | 2020-07-07 | 2020-10-27 | 重庆邮电大学 | Method for constructing driving condition of automobile |
| CN112061136A (en) * | 2020-08-26 | 2020-12-11 | 西北工业大学 | Automatic analysis method for driver driving behavior based on long-term and short-term memory network |
| CN112265546A (en) * | 2020-10-26 | 2021-01-26 | 吉林大学 | Networked automobile speed prediction method based on time-space sequence information |
| CN112435503A (en) * | 2020-10-30 | 2021-03-02 | 江苏大学 | Intelligent automobile active collision avoidance method for identifying intention of high-risk pedestrians |
Non-Patent Citations (2)
| Title |
|---|
| 孙倩 等: "基于长短期记忆神经网络方法的车辆跟驰模型", 吉林大学学报(工学版), vol. 20, no. 04, 31 August 2020 (2020-08-31), pages 1380 - 1386 * |
| 张凤娇 等: "基于深度学习的极限工况下车辆的状态估计", 重庆理工大学学报( 自然科学), vol. 32, no. 10, 31 October 2018 (2018-10-31), pages 64 - 70 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115782595A (en) * | 2022-12-06 | 2023-03-14 | 东南大学 | Electric bus instantaneous energy consumption estimation method based on energy recovery state |
| CN115782595B (en) * | 2022-12-06 | 2024-04-02 | 东南大学 | Electric bus instantaneous energy consumption estimation method based on energy recovery state |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111238825B (en) | Intelligent driving automatic emergency braking performance testing method for combined test pavement | |
| Murphey et al. | Driver's style classification using jerk analysis | |
| CN107291972B (en) | The Intelligent Vehicle Driving System efficiency evaluation method excavated based on multi-source data | |
| Wu et al. | Predictive shift strategy of dual-clutch transmission for driving safety on the curve road combined with an electronic map | |
| CN104890669B (en) | A kind of hybrid power automobile power assembly control method | |
| CN102506160B (en) | Ramp based on longitudinal dynamics and vehicle load identification method | |
| CN111775925B (en) | Working mode decision method and device for power split hybrid electric vehicle | |
| CN108333921B (en) | Automobile gear shifting rule optimization method based on dynamic programming algorithm | |
| Luo et al. | Multi-objective decoupling algorithm for active distance control of intelligent hybrid electric vehicle | |
| Ganji et al. | A study on look-ahead control and energy management strategies in hybrid electric vehicles | |
| CN110262491B (en) | Method and system for predicting vehicle braking intention based on hybrid learning mode | |
| CN107421752A (en) | A kind of intelligent automobile test scene accelerates reconstructing method | |
| CN111038488B (en) | Energy optimization control method and device for hybrid electric vehicle | |
| CN111459995A (en) | Multi-mode vehicle speed prediction method based on driving data | |
| CN115476861B (en) | Safety evaluation system of intelligent network-connected automobile | |
| CN108068814A (en) | The method and the vehicle of application method travelled using vehicle traveling information control vehicle | |
| CN116186884A (en) | Intelligent network-connected automobile function safety and expected function safety linkage verification method | |
| CN113485102A (en) | Method for identifying vehicle running condition based on long-term and short-term memory neural network | |
| CN108711285A (en) | A kind of mixed traffic analogy method based on road cross | |
| CN115186594A (en) | Energy-saving speed optimization method under influence of man-vehicle-road coupling | |
| Salunkhe et al. | Energy optimization for CAN bus and media controls in electric vehicles using deep learning algorithms | |
| CN113657432A (en) | Commercial vehicle driving behavior risk level identification method based on Internet of vehicles data | |
| Liu et al. | Energy management strategy of hybrid electric vehicles based on driving condition prediction | |
| CN113781770B (en) | Simulation double-clutch type confluence area vehicle-road cooperative control system | |
| Beşkardeş et al. | Data-driven-based fuzzy control system design for a hybrid electric vehicle |
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 |