A method for identifying vehicle driving conditions based on long short-term memory neural network

Abstract

The present invention belongs to the technical field of driving condition identification, and discloses a method for identifying vehicle driving conditions based on long short-term memory neural network, analyzing and screening the collected driving data; dividing the analyzed and screened data into a training data set and a verification data set; constructing a long short-term memory neural network model; using the training data set to train the long short-term memory neural network model to obtain a trained long short-term memory neural network model; using the verification data set to evaluate the trained long short-term memory neural network model; extracting the parameters of the verified long short-term memory neural network model; establishing a driving condition classifier based on the extracted parameter data; integrating the driving condition classifier into the AMT control program, and using the driving condition classifier to identify the vehicle's driving condition online. The present invention establishes a condition control strategy based on the driving condition, and improves the reliability, safety, economy and comfort of hybrid heavy-duty commercial vehicles.

Description

Technical Field

[0001] The present invention belongs to the technical field of driving condition recognition, and in particular relates to a method for recognizing vehicle driving conditions based on a long short-term memory neural network. Background Art

[0002] Since hybrid heavy-duty commercial vehicles have a large full-load mass (generally around 50 tons), large mass has large inertia, resulting in long acceleration and braking times. The hybrid heavy-duty commercial vehicle gearbox is designed to optimize the engine and motor output according to the driving force required by the vehicle by changing the transmission ratio of the hybrid heavy-duty commercial vehicle transmission system. The transmission ratio of hybrid heavy-duty commercial vehicles generally ranges from 0.75 to 20, and reasonable gears and transmission ratios are crucial to the operation of hybrid heavy-duty commercial vehicles. Therefore, hybrid heavy-duty commercial vehicles equipped with automatic mechanical transmissions (AMTs) use the same shift control logic under different working conditions, which will lead to problems such as overheating and burning of the clutch, excessive emissions, increased energy consumption (fuel consumption plus electricity consumption), and excessive noise, which may also endanger the safety of vehicles and personnel. For example, in congested conditions (such as traffic jams in city centers and highway toll stations), the use of a general shifting strategy will lead to frequent shifting, resulting in clutch overheating, which will cause shifting failure or even clutch burning; in low-speed conditions (such as suburbs, villages, etc.), the use of a general shifting strategy will lead to untimely shifting, resulting in insufficient power or vehicle speed; in medium-speed conditions (provincial roads, national roads, etc.) and high-speed conditions (highways), the use of a general shifting strategy will lead to slow or no shifting when overtaking, powerless overtaking, slow overtaking, and affecting vehicle safety. In order to ensure vehicle safety, increase component life, reduce emissions and reduce energy consumption (fuel consumption plus electricity consumption), the mechanical automatic transmission of hybrid heavy-duty commercial vehicles should adopt corresponding control strategies for different conditions.

[0003] There is almost no research on the driving conditions of hybrid heavy-duty commercial vehicles, and the identification and shifting strategies for different conditions are even more difficult problems in the field of hybrid heavy-duty commercial vehicle AMT. Summary of the invention

[0004] The purpose of the present invention is to provide a method for identifying vehicle driving conditions based on a long short-term memory neural network, so as to solve the problem of automatic adjustment of the AMT control strategy of a hybrid heavy-duty commercial vehicle according to the driving conditions.

[0005] The present invention is achieved through the following technical solutions:

[0006] A method for identifying vehicle driving conditions based on a long short-term memory neural network comprises the following steps:

[0007] S1. Collecting driving data of vehicles on different roads, and analyzing and screening the collected driving data;

[0008] S2, dividing the analyzed and screened data into two groups, one group as a training data set, and the other group as a verification data set;

[0009] S3. Construct a long short-term memory neural network model based on the screened data;

[0010] S4, using the training data set to train the long short-term memory neural network model to obtain a trained long short-term memory neural network model;

[0011] S5. Evaluate the trained LSTM neural network model using the verification data set;

[0012] S6, extracting the parameters of the verified long short-term memory neural network model;

[0013] S7, establishing a driving condition classifier according to the extracted parameter data;

[0014] S8. Integrate the driving condition classifier into the AMT control program, and use the driving condition classifier to identify the driving condition of the vehicle online.

[0015] Furthermore, in S8, the operating condition classification includes traffic jam condition, low-speed condition, medium-speed condition and high-speed condition, and corresponding operating condition control strategies are established according to the driving conditions. The corresponding operating condition control strategies are traffic jam condition control strategy, low-speed condition control strategy, medium-speed condition control strategy, high-speed condition control strategy and general ordinary control strategy.

[0016] Furthermore, the driving condition classifier is used to identify the driving condition of the vehicle online, and the process of establishing the corresponding condition control strategy according to the driving condition is as follows: first, whether it is a high-speed condition is determined, if it is a high-speed condition, the high-speed condition shift control strategy is executed, otherwise, continue to determine;

[0017] Secondly, determine whether it is a medium-speed operating condition. If it is a medium-speed operating condition, execute the medium-speed operating condition shift control strategy, otherwise continue to determine;

[0018] Determine again whether it is a low-speed operating condition. If it is a low-speed operating condition, execute the low-speed operating condition shift control strategy, otherwise continue to determine;

[0019] Determine again whether it is a low-speed operating condition. If it is a low-speed operating condition, execute the low-speed operating condition shift control strategy, otherwise continue to determine;

[0020] Finally, determine whether it is a traffic jam condition. If it is a traffic jam condition, execute the traffic jam condition shift control strategy, otherwise execute the general condition shift strategy.

[0021] Furthermore, in S1, the driving data includes vehicle speed, accelerator pedal opening, brake pedal opening and turning radius.

[0022] Furthermore, in S1, different road surfaces include urban roads, provincial roads, national roads and expressways.

[0023] Furthermore, in S2, the analyzed and screened data are randomly divided into two groups according to the ratio of 70% and 30%, and the 70% group is used as a training data set, and the 30% group is used as a verification data set.

[0024] Furthermore, the LSTM neural network model includes an input layer, an Lstm layer, a fully connected layer, a softmax layer, and a classification layer;

[0025] In S7, the extracted parameters of the long short-term memory neural network model include the input weight IW of the Lstm layer, the periodic weight RW of the Lstm layer, the offset B of the Lstm layer, the weight FW of the fully connected layer, and the offset FB of the fully connected layer.

[0026] Further, in S7, the weight FW of the fully connected layer and the offset FB of the fully connected layer are used to construct the fully connected layer of the driving condition classifier; in S8, the calculation process of the driving condition classifier is:

[0027] (1) The outputs gtt, itt, ftt and c of the driving condition classifier Lstm layer are t-1 Input to the fully connected layer, the output of the fully connected layer is the probability that the current vehicle operating condition belongs to the four conditions of traffic jam, low speed, medium speed and high speed. The specific formula is:

[0028] OUT=FW*(gtt*itt+ftt*c t-1 )+FB

[0029] Among them, OUT represents the probability of the vehicle being in a traffic jam, low speed, medium speed and high speed at the current moment, which is a 1*4 matrix; FW represents the weight of the fully connected layer, FB represents the offset of the fully connected layer; c t-1 represents 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 forget gate;

[0030] (2) Calculation condition type:

[0031] [~, Kind_OUT] = max(OUT)

[0032] Among them, Kind_OUT represents the operating condition type, which is one of the four numbers 1, 2, 3, and 4. 1 is traffic jam condition, 2 is low speed condition, 3 is medium speed condition, and 4 is high speed condition. Max represents the max function of MATLAB, and the output is a 1*2 matrix. The first number of the matrix is ​​the value of the maximum value, and the second number is the position of the maximum value. The position of the maximum value Kind_OUT is used to determine the operating condition. The symbol ~ means ignoring the first value of the output.

[0033] Furthermore, the forget gate formula is:

[0034] ftt = logsig(G(fInd));

[0035] Among them, ftt represents the output value of the forget gate, which is a 1*100 matrix; G(fInd) represents the 1st to 100th value of G, which is a 1*100 matrix; logsig represents the Log-sigmoid conversion function in deep learning;

[0036] The input gate formula is:

[0037] itt = logsig(G(iInd));

[0038] Among them, itt represents the output value of the input gate, which is a 1*100 matrix; G(iInd) represents the 101st to 200th value of G, which is a 1*100 matrix, and logsig represents the Log-sigmoid function in deep learning;

[0039] The formula for selecting the memory gate is:

[0040] gtt = tanh(G(zInd));

[0041] Among them, gtt represents the output value of the selected memory gate, which is a 1*100 matrix; G(zInd) represents the 201st to 300th value of G, which is a 1*100 matrix, and tanh represents the tangent function in deep learning;

[0042] The output gate formula is:

[0043] ott = logsig(G(oInd));

[0044] Among them, ott represents the output value of the output gate, which is a 1*100 matrix, G(oInd) represents the 301st to 400th values ​​of G, which is a 1*100 matrix, and logsig represents the Log-sigmoid conversion function in deep learning;

[0045] The linear gate formula is:

[0046] G=IW*x t+RW*h t-1 +B;

[0047] Among them, G represents the output value of the linear calculation part of all gates, which is a 1*400 matrix; IW represents the input weight of the Lstm layer, x t represents the eigenvalue at the current moment, RW represents the periodic weight of the Lstm layer, and h t-1 represents the hidden state of the previous moment, and B represents the offset of the Lstm layer.

[0048] Furthermore, the hidden state formula is:

[0049] h t =tanh(c t )*ott;

[0050] Among them, the operator * is a matrix operation, h t represents the hidden state at the current moment, c t Represents the cell state at the current moment;

[0051] Cell state formula:

[0052] c t =gtt*itt+ftt*c t- 1;

[0053] Among them, c t Represents the cell state at the current moment, c t-1 Represents the cell state at the previous moment.

[0054] Compared with the prior art, the present invention has the following beneficial technical effects:

[0055] The present invention discloses a method for identifying vehicle driving conditions based on long short-term memory neural network, which is specifically aimed at the dependence of hybrid heavy-duty commercial vehicles on vehicle driving conditions when they are driven by multiple power sources (generally diesel and electricity) and when the motor brakes and recovers energy. By combining with driving intention recognition, it effectively solves the control problem of recovering as much energy as possible under different working conditions and ensuring the safe operation of the vehicle, and prevents the hybrid vehicle from having insufficient braking force due to brake energy recovery at high speed, hybrid intervention exceeding the working condition requirements at low speed, and multiple braking forces being uncoordinated during braking, resulting in vehicle impact or even uncontrolled movement, and other newly discovered problems in the new energy vehicle industry. By studying the driving conditions of hybrid heavy-duty commercial vehicles, applying neural networks to identify the driving conditions, and optimizing the shifting strategy of the AMT of hybrid heavy-duty commercial vehicles based on the identified conditions, the reliability, safety, economy and comfort of hybrid heavy-duty commercial vehicles are thereby improved, providing strong support for energy conservation and emission reduction, and relieving environmental and energy pressures. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 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;

[0057] Figure 2 Schematic diagram of the layer structure of the long short-term memory neural network (LSTM);

[0058] Figure 3 Schematic diagram of the cell structure of the long short-term memory neural network (LSTM);

[0059] Figure 4 This is the processed data of a traffic jam in Nanjing; Figure 4 (a) represents the collected vehicle speed data, Figure 4 (b) represents the collected accelerator pedal opening data, Figure 4 (c) represents the collected brake pedal opening data, Figure 4 (d) represents the collected turning radius data, Figure 4 (e) represents the collected acceleration data;

[0060] Figure 5 The training accuracy curve for the Long Short-Term Memory (LSTM) neural network;

[0061] Figure 6 This is the AMT control flow chart for hybrid heavy-duty commercial vehicles. DETAILED DESCRIPTION

[0062] The present invention is further described in detail below in conjunction with specific embodiments, which are intended to explain the present invention rather than to limit it.

[0063] like Figure 1 As shown, the present invention discloses a method for identifying vehicle driving conditions based on a long short-term memory neural network, comprising the following steps:

[0064] S1. Collecting driving data of vehicles on different roads, and analyzing and screening the collected driving data;

[0065] S2, dividing the analyzed and screened data into two groups, one group as a training data set, and the other group as a verification data set;

[0066] S3. Construct a long short-term memory neural network model based on the screened data;

[0067] S4, using the training data set to train the long short-term memory neural network model to obtain a trained long short-term memory neural network model;

[0068] S5. Evaluate the trained LSTM neural network model using the verification data set;

[0069] S6, extracting the parameters of the verified long short-term memory neural network model;

[0070] S7, establishing a driving condition classifier according to the extracted parameter data;

[0071] S8. Integrate the driving condition classifier into the AMT control program of the hybrid heavy-duty commercial vehicle, use the driving condition classifier to analyze the vehicle's driving condition online, and establish a corresponding condition control strategy based on the driving condition.

[0072] The present invention mainly utilizes the long short-term memory neural network LSTM to identify the driving conditions of hybrid heavy-duty commercial vehicles and solve the problem of automatic adjustment of the AMT control strategy of hybrid heavy-duty commercial vehicles according to the driving conditions. Specifically:

[0073] (1) CANalyzer software is used to collect the original driving data of hybrid heavy-duty commercial vehicles on urban, provincial, national and expressway roads through the OBD interface of the vehicle's public CAN communication network, including vehicle speed, accelerator pedal opening, brake pedal opening and turning radius.

[0074] The vehicle speed is obtained by the FrontAxleSpeed ​​signal of the EBC2 message sent by the vehicle anti-lock braking controller ABS, the accelerator pedal opening is obtained by the AccelPedalPos1 signal of the EEC2 message sent by the engine controller, the brake pedal opening is obtained by the BrakeSwitch signal of the CCVS message sent by the engine controller, and the steering radius is calculated by the RelativeSpeedFrontAxleLeftWheel and RlativeSpeedFrontAxleRightWheel signals of the EBC2 message sent by the vehicle anti-lock braking controller ABS.

[0075] (2) Analyze and filter data. Use CANalyzer to manually group the collected data and label them. Then use the data export function of CANalyzer to convert the data format from the original .blf to the .mat format that MATLAB can recognize. Then use the data processing function of MATLAB to clean up invalid data. A data set {XData, YData} consisting of 8500 groups of data is obtained. Each group of data includes a 5*60 matrix X and a classification label Y. 5 represents the five input features of the neural network: 1 vehicle speed, 2 accelerator pedal opening, 3 brake pedal opening, 4 turning radius and 5 acceleration (acceleration is obtained by subtracting the vehicle speed in the last second from the vehicle speed in the previous second). 60 represents the 60 columns of feature values ​​of the 60-second time series. The classification label is one of the traffic jam condition, low speed condition, medium speed condition and high speed condition.

[0076] (3) The analyzed and screened data are randomly divided into two groups according to the ratio of 70% and 30%. The 70% group is used as the training data set {XTrain, YTrain}, and the 30% group is used as the verification data set {Xtest, YTest} to check the single group feature data.

[0077] (4) Based on the selected data and the neural network toolbox of MATLAB software, construct the following Figure 2 The long short-term memory neural network model shown in the figure, where x t Represents the input features, that is, the 5 feature values ​​at the current moment, t-1, t, t+1, ..., s represents the time series, where s = 60. Based on experience and continuous trial and error, the neural network model and training parameters are set: input feature dimension featureDimension = 5; number of hidden layer units numHiddenUnits = 100; number of classifications numClasses = 4; maximum iteration parameter max_epochs = 500. The generated neural network model lstm_layers includes 5 layers: input layer, Lstm layer, fully connected layer, softmax layer and classification layer.

[0078] Load the lstm_layers neural network model through MATLAB's accessory program Deep Network Designer, and view the neural network architecture, multi-dimensional input layer parameters, LSTM layer parameters, fully connected layer parameters, softmax layer parameters, and classification layer parameters.

[0079] (5) Use the training dataset {XTrain, YTrain} to train the long short-term memory neural network model.

[0080] Call MATLAB's trainNetwork function to train the network model. The function inputs are: training data set {XTrain, YTrain}, neural network model lstm_layers and training control parameters (max_epochs = 500, etc.); the function outputs the neural network model lstm_net. lstm_layers is the input model developed for training and needs to be set manually; lstm_net is the training output model, which is automatically output after training.

[0081] View the training results at Figure 5 If the learning effect does not meet the expected requirements, that is, the expected accuracy rate does not reach the set target of 98%, adjust the training parameters and continue learning until the learning reaches the required set target.

[0082] (6) Use the verification data set {Xtest, YTest} to evaluate the trained long short-term memory neural network model lstm_net. If the verification effect does not meet the expected requirements, that is, the expected accuracy rate does not reach 98%, adjust the training parameters and continue learning until the verification meets the requirements.

[0083] (7) Extract the parameters of the verified long short-term memory neural network model lstm_net:

[0084] The input weight IW of the Lstm layer is a 5*400 matrix, the periodic weight RW of the Lstm layer is a 100*400 matrix, the offset B of the Lstm layer is a 1*400 matrix, the weight FW of the fully connected layer is a 4*100 matrix, and the weight FB of the fully connected layer is a 1*4 matrix.

[0085] (8) Based on the extracted data, a hybrid heavy-duty commercial vehicle operating condition recognition algorithm, namely, a driving condition classifier, is developed. Specifically, the fully connected layer of the driving condition classifier is constructed using the weight FW of the fully connected layer and the offset FB of the fully connected layer.

[0086] (9) The driving condition classifier is integrated into the AMT control program, and the driving condition classifier is used to analyze the vehicle's driving condition online, and the corresponding condition control strategy is established according to the driving condition.

[0087] The input feature matrix of the fully connected layer is input into the fully connected layer, and the output of the fully connected layer is calculated, that is, the probability that the current vehicle operating condition belongs to the four conditions of traffic jam, low speed, medium speed and high speed.

[0088] LSTM layer at time t (see Figure 3 ) is: the eigenvalue x at the current moment t , is a 1*5 matrix, the cell state c at the last moment t-1 , a 1*100 matrix, and the hidden state h of the previous moment t-1 , which is a 1*100 matrix.

[0089] Linear gate formula:

[0090] G=IW*x t +RW*h t-1 +B

[0091] Among them, G represents the output value of the linear calculation part of all gates, which is a 1*400 matrix; IW represents the input weight of the Lstm layer, x t represents the eigenvalue at the current moment, RW represents the periodic weight of the Lstm layer, and h t-1 represents the hidden state of the previous moment, and B represents the offset of the Lstm layer.

[0092] Forget gate formula:

[0093] ftt=logsig(G(fInd))

[0094] Among them, ftt represents the output value of the forget gate, which is a 1*100 matrix; G(fInd) represents the 1st to 100th value of G, which is a 1*100 matrix, and logsig represents the Log-sigmoid conversion function in deep learning.

[0095] Input gate formula:

[0096] itt=logsig(G(iInd))

[0097] Among them, itt represents the output value of the input gate, which is a 1*100 matrix; G(iInd) represents the 101st to 200th values ​​of G, which is a 1*100 matrix, and logsig represents the Log-sigmoid function in deep learning.

[0098] Select the memory gate formula:

[0099] gtt=tanh(G(zInd))

[0100] Among them, gtt represents the output value of the selected memory gate, which is a 1*100 matrix; G(zInd) represents the 201st to 300th value of G, which is a 1*100 matrix, and tanh represents the tangent function in deep learning. g output gate formula:

[0101] ott=logsig(G(oInd))

[0102] Among them, ott represents the output value of the output gate, which is a 1*100 matrix, G(oInd) represents the 301st to 400th values ​​of G, which is a 1*100 matrix, and logsig represents the Log-sigmoid conversion function in deep learning.

[0103] Update cell status formula:

[0104] c t =gtt*itt+ftt*c t- 1

[0105] Among them, the operator symbols * and + are matrix operations, c t Represents the cell state at the current moment, c t-1 Represents the cell state at the previous moment.

[0106] Update hidden state formula:

[0107] h t =tanh(c t)*ott

[0108] Among them, the operator * is a matrix operation, h t Represents the hidden state at the current moment.

[0109] At time t, the output calculation formula of the fully connected layer is:

[0110] OUT=FW*c t +FB

[0111] Among them, the operator symbols * and + are matrix operations, and OUT represents the probability that the vehicle is in a traffic jam, a low-speed condition, a medium-speed condition, and a high-speed condition at the current moment, which is a 1*4 matrix.

[0112] Calculation formula:

[0113] [~, Kind_OUT] = max(OUT)

[0114] Among them, Kind_OUT represents the operating condition type, which is one of the four numbers 1, 2, 3, and 4. 1 is traffic jam condition, 2 is low speed condition, 3 is medium speed condition, and 4 is high speed condition. Max represents the max function of MATLAB, whose input can be a matrix and output can be a 1*2 matrix. The first number of the matrix is ​​the value of the maximum value, and the second number is the position of the maximum value. This classifier does not care about the value of the maximum value. Only the position of the maximum value Kind_OUT is needed to determine the operating condition. The symbol ~ means ignoring the first value of the output.

[0115] like Figure 4 As shown, from Figure 4 (a) It can be seen that feature 1 has a low vehicle speed. Figure 4 (b) and 4(c) show that after feature 2, the accelerator pedal is turned off, and feature 3 is immediately braked to a stop. Figure 4 (d) It can be seen that feature 4 is a low-speed turn at the traffic light. Figure 4 (e) It can be seen that the acceleration of feature 5 fluctuates repeatedly over a large range. This is a set of input data marked as traffic jam conditions. OUT = [0.9982, 0.4473, 0.0258, 0.2040] is identified by a verified classifier. It can be seen that the probability of it belonging to the traffic jam condition is the largest, with a probability of 0.9982.

[0116] (9) The driving condition classifier is integrated into the AMT control program of hybrid heavy-duty commercial vehicles. According to the four classifications obtained: traffic jam condition, low-speed condition, medium-speed condition and high-speed condition, the traffic jam condition control strategy, low-speed condition control strategy, medium-speed condition control strategy, high-speed condition control strategy and general ordinary control strategy are developed respectively.

[0117] like Figure 6As shown in FIG. 1 , the process of using the driving condition classifier to analyze the driving condition of the vehicle online is as follows:

[0118] First, determine whether it is a high-speed operating condition. If it is a high-speed operating condition, execute the high-speed operating condition shift control strategy, otherwise continue to determine;

[0119] Secondly, determine whether it is a medium-speed operating condition. If it is a medium-speed operating condition, execute the medium-speed operating condition shift control strategy, otherwise continue to determine;

[0120] Determine again whether it is a low-speed operating condition. If it is a low-speed operating condition, execute the low-speed operating condition shift control strategy, otherwise continue to determine;

[0121] Determine again whether it is a low-speed operating condition. If it is a low-speed operating condition, execute the low-speed operating condition shift control strategy, otherwise continue to determine;

[0122] Finally, determine whether it is a traffic jam condition. If it is a traffic jam condition, execute the traffic jam condition shift control strategy, otherwise execute the general condition shift strategy.

Claims

1. A method for identifying vehicle driving conditions based on long short-term memory neural network, characterized in that: The following steps are involved: S1. Collecting driving data of the vehicle on different roads and analyzing and screening the collected driving data; the driving data includes vehicle speed, accelerator pedal opening, brake pedal opening and turning radius; S2, dividing the analyzed and screened data into two groups, one group as a training data set, and the other group as a verification data set; S3. Construct a long short-term memory neural network model based on the screened data; S4, using the training data set to train the long short-term memory neural network model to obtain a trained long short-term memory neural network model; S5. Evaluate the trained LSTM neural network model using the verification data set; S6, extracting the parameters of the verified long short-term memory neural network model; S7, establishing a driving condition classifier according to the extracted parameter data; S8, integrating the driving condition classifier into the AMT control program, and using the driving condition classifier to identify the driving condition of the vehicle online; The long short-term memory neural network model includes input layer, Lstm layer, fully connected layer, softmax layer and classification layer; In S7, the extracted parameters of the long short-term memory neural network model include the input weight IW of the Lstm layer, the period weight RW of the Lstm layer, the offset B of the Lstm layer, the weight FW of the fully connected layer, and the offset FB of the fully connected layer; The fully connected layer weight FW and the offset FB of the fully connected layer are used to construct the fully connected layer of the driving condition classifier; in S8, the calculation process of the driving condition classifier is: (1) The outputs gtt, itt, ftt and c of the driving condition classifier Lstm layer are t-1 Input to the fully connected layer, the output of the fully connected layer is the probability that the current vehicle operating condition belongs to the four conditions of traffic jam, low speed, medium speed and high speed. The specific formula is: OUT=FW*(gtt*itt+ftt*c t-1 )+FB Among them, OUT represents the probability of the vehicle being in a traffic jam, low speed, medium speed and high speed at the current moment, which is a 1*4 matrix; FW represents the weight of the fully connected layer, FB represents the offset of the fully connected layer; c t-1 represents 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 forget gate; (2) Calculation condition type: [~, Kind_OUT] = max(OUT) Among them, Kind_OUT represents the working condition type, which is one of the four numbers 1, 2, 3, and 4. 1 is the traffic jam condition, 2 is the low-speed condition, 3 is the medium-speed condition, and 4 is the high-speed condition. Max represents the max function of MATLAB, and the output is a 1*2 matrix. The first number of the matrix is ​​the value of the maximum value, and the second number is the position of the maximum value. The position of the maximum value Kind_OUT determines the working condition. The symbol ~ means ignoring the first value of the output; The forget gate formula is: ftt = logsig(G(fInd)); Among them, ftt represents the output value of the forget gate, which is a 1*100 matrix; G(fInd) represents the 1st to 100th value of G, which is a 1*100 matrix; logsig represents the Log-sigmoid conversion function in deep learning; The input gate formula is: itt = logsig(G(iInd)); Among them, itt represents the output value of the input gate, which is a 1*100 matrix; G(iInd) represents the 101st to 200th value of G, which is a 1*100 matrix, and logsig represents the Log-sigmoid function in deep learning; The formula for selecting the memory gate is: gtt = tanh(G(zInd)); Among them, gtt represents the output value of the selected memory gate, which is a 1*100 matrix; G(zInd) represents the 201st to 300th value of G, which is a 1*100 matrix, and tanh represents the tangent function in deep learning; The output gate formula is: ott = logsig(G(oInd)); Among them, ott represents the output value of the output gate, which is a 1*100 matrix, G(oInd) represents the 301st to 400th values ​​of G, which is a 1*100 matrix, and logsig represents the Log-sigmoid conversion function in deep learning; The linear gate formula is: G=IW*x t +RW*h t-1 +B; Among them, G represents the output value of the linear calculation part of all gates, which is a 1*400 matrix; IW represents the input weight of the Lstm layer, x t represents the eigenvalue at the current moment, RW represents the periodic weight of the Lstm layer, and h t-1 represents the hidden state of the previous moment, and B represents the offset of the Lstm layer.

2. The method for identifying vehicle driving conditions based on long short-term memory neural network according to claim 1 is characterized in that: In S8, the operating condition classification includes traffic jam condition, low-speed condition, medium-speed condition and high-speed condition. Corresponding operating condition control strategies are established according to the driving conditions. The corresponding operating condition control strategies are traffic jam condition control strategy, low-speed condition control strategy, medium-speed condition control strategy, high-speed condition control strategy and general ordinary control strategy.

3. The method for identifying vehicle driving conditions based on long short-term memory neural network according to claim 2 is characterized in that: The process of using the driving condition classifier to identify the vehicle's driving condition online and establishing the corresponding condition control strategy according to the driving condition is as follows: First, determine whether it is a high-speed operating condition. If it is a high-speed operating condition, execute the high-speed operating condition control strategy, otherwise continue to determine; Secondly, determine whether it is a medium-speed condition. If it is a medium-speed condition, execute the medium-speed condition control strategy, otherwise continue to determine; Determine again whether it is a low-speed operating condition. If it is a low-speed operating condition, execute the low-speed operating condition control strategy, otherwise continue to determine; Finally, determine whether it is a traffic jam condition. If it is a traffic jam condition, execute the traffic jam condition control strategy, otherwise execute the general ordinary control strategy.

4. The method for identifying vehicle driving conditions based on long short-term memory neural network according to claim 1 is characterized in that: In S1, different roads include urban roads, provincial roads, national roads and expressways.

5. The method for identifying vehicle driving conditions based on long short-term memory neural network according to claim 1 is characterized in that: In S2, the analyzed and screened data are randomly divided into two groups according to the ratio of 70% and 30%, and the 70% group is used as the training data set, and the 30% group is used as the verification data set.

6. The method for identifying vehicle driving conditions based on long short-term memory neural network according to claim 1, characterized in that: The hidden state formula is: h t =tanh(c t )*ott; Among them, the operator * is a matrix operation, h t represents the hidden state at the current moment, c t Represents the cell state at the current moment; Cell state formula: c t＝ gtt*itt+ftt*c t- 1 ； Among them, c t Represents the cell state at the current moment, c t-1 Represents the cell state at the previous moment.

Images