Online evaluation method for driving intention recognition accuracy of intelligent connected HEV

Abstract

The application discloses an online evaluation method for driving intention recognition accuracy of intelligent network connection HEV, which is achieved by obtaining state data and driving intention of a current vehicle in a historical time period under an intelligent network connection environment, establishing a whole vehicle system multi-dimensional feature input matrix on a cloud server, constructing a whole vehicle twin reverse evaluation model to evaluate the driving intention recognized by the system, obtaining an evaluation factor output matrix, constructing a logic control rule to process the evaluation factor output matrix according to different rules, calculating the system fluctuation amount between the evaluation factors after processing to obtain a fluctuation vector set, and finally obtaining the recognition accuracy of the driving intention recognition system through functional calculation. The application can evaluate the driving intention recognition accuracy of the vehicle under the complex driving environment of the future intelligent network connection HEV, and provide an important guarantee for the reliability of the whole vehicle driving system and the driving safety.

Description

Technical Field

[0001] This invention belongs to the field of intelligent connected hybrid electric vehicle driving, specifically an online evaluation method for the accuracy of driving intention recognition in intelligent connected HEVs. Background Technology

[0002] Intelligent connected HEVs are new energy vehicles that utilize vehicle-to-everything (V2X) technology and computer system control to achieve assisted or automated driving. By integrating modern communication and network technologies, they enable intelligent information sharing among people, vehicles, roads, and the cloud, giving them functions such as complex environment perception, intelligent decision-making, and collaborative driving. In the complex driving environment of intelligent connected HEVs, accurately identifying the driver's intentions is crucial for improving the reliability and safety of the driving system and guiding the vehicle's human-like decision-making.

[0003] Research on driver intent recognition has matured, but increasingly complex traffic scenarios and more frequent driver intent shifts, especially in urban areas where frequent braking, acceleration, and lane changes by surrounding vehicles pose significant challenges to the accuracy of driver intent recognition systems. Furthermore, the data required for driver intent recognition systems is becoming increasingly complex, and their resistance to interference in complex scenarios is weak. This means that driver intent recognition remains at a shallow feature recognition stage, resulting in accuracy that fails to meet the requirements of practical applications.

[0004] Therefore, in the complex driving environment of intelligent connected HEVs, how to accurately evaluate the recognition accuracy of the driving intention recognition system by utilizing historical vehicle state data and the driving intentions identified by the system has become an urgent problem to be solved. This research is expected to provide important support for the future development of intelligent transportation, and at the same time, it has important theoretical significance and research value for the subsequent real-time control and energy management strategies of intelligent connected HEVs during vehicle operation. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this invention proposes an online evaluation method for the accuracy of driving intention recognition in intelligent connected HEVs. The aim is to evaluate the accuracy of the driving intention recognition system online in complex driving environments of intelligent connected HEVs, thereby providing an important basis for real-time control and subsequent energy management strategies of intelligent connected HEVs during driving.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The present invention provides an online evaluation method for the accuracy of driving intention recognition in intelligent connected hybrid electric vehicles (HEVs). This method is characterized by its application in multi-vehicle operation scenarios of intelligent connected hybrid electric vehicles, which consist of a cloud server and a driving intention recognition system, and is performed according to the following steps:

[0008] Step 1: Based on vehicle-to-vehicle (V2V) communication and vehicle-to-infrastructure (V2I) communication, obtain the current vehicle status data and the current vehicle's driving intention identified by the driving intention recognition system within the historical time period T, and upload them to the cloud server.

[0009] The cloud server integrates the current vehicle's state data from time t-1+1 to time t within the historical time period T, along with the driving intention, into a multi-dimensional feature input matrix P = [p t-l+1 ,p t-l+2 ,...,p t-l+i ,...,p t ], where p t-l+i p represents the feature parameter vector of the current vehicle at time t-l+i within the historical time period T. t Let represent the parameter vector of the current vehicle at time t within the historical time period T, and T t This represents the instantaneous engine torque of the vehicle at time t. This represents the instantaneous engine speed of the vehicle at time t. This represents the real-time fuel consumption of the vehicle at time t. This represents the motor output torque of the vehicle at time t. L represents the instantaneous motor speed of the vehicle at time t. t M t Let α represent the real-time motor power and real-time engine power of the vehicle at time t, respectively. t I represents the instantaneous engine throttle opening of the vehicle at time t. t Let I represent the real-time driving intention of the vehicle at time t. t The category label is Y m l represents the number of time points;

[0010] Step 2: The cloud server constructs a vehicle twin reverse evaluation model, including: an input layer, a feature reconstruction layer, and an evaluation factor output layer;

[0011] Step 2.1, Processing of the input layer:

[0012] Integrating the parameter vectors at each time step in the multidimensional feature input matrix P, we obtain the feature coefficient space matrix P′=[p′ t-l+1 ,p′ t-l+2 ,...,p′t-l+i ,...,p′ t ], where p′ t-l+i This represents the feature coefficient space vector of the current vehicle at time t-l+i;

[0013] Step 2.2: The feature reconstruction layer processes the feature coefficient space matrix P′ to obtain the feature coefficient reconstruction matrix E within the time period T;

[0014] Step 2.3, Processing of the evaluation factor output layer:

[0015] The evaluation factor output matrix is ​​obtained using the inverse function ψ(U×E+c). Where U is the set influence factor matrix, and c represents the loss compensation vector. Represents the output matrix of evaluation factors The set of driving intention evaluation factors corresponding to the nth time point, and Where, τ r,n This represents the r-th driving intention evaluation factor in the set of driving intention evaluation factors corresponding to the nth time point;

[0016] Step 3: Construct logical control rules and perform logical judgment processing on the output matrix of the evaluation factors for driving intention recognition;

[0017] Step 3.1: Normalize each evaluation factor corresponding to each time point in the evaluation factor output matrix to obtain the final evaluation factor preprocessing matrix. in, This represents the r-th driving intention evaluation factor in the set of driving intention evaluation factors corresponding to the n-th time point after normalization.

[0018] Step 3.2: Obtain the evaluation factor preprocessing matrix M using equation (4). Φ The characteristic range;

[0019]

[0020] In equation (4), β1(M) Φ ) represents the evaluation factor preprocessing matrix M Φ The norm, β ∞ (M) Φ ) represents the evaluation factor preprocessing matrix M Φ row norm, β2(M) Φ ) represents the evaluation factor preprocessing matrix M Φ spectral norm, Preprocessing matrix M for evaluation factors Φ The normal matrix, λ max |M| represents the largest eigenvalue of the matrix. Φ|1 represents the preprocessing matrix M of the evaluation factors Φ The maximum sum of the evaluation factors in each row of the table, |M Φ | ∞ M represents the preprocessing matrix of evaluation factors. Φ The maximum sum of the evaluation factors in each column, |M Φ |2 represents the evaluation factor preprocessing matrix M Φ The maximum absolute value of the singular values.

[0021] Determine whether the condition shown in equation (5) is true. If it is true, set M to... Φ Evaluation factor matrix for driving intention recognition If the error persists, proceed to step 4; otherwise, proceed to step 3.3.

[0022] argmin(β 1,2,∞ (M) Φ ))≤β(Μ Φ )≤argmax(β 1,2,∞ (M) Φ (5)

[0023] In equation (5), argmin(β) 1,2,∞ (M) Φ )) represents the evaluation factor preprocessing matrix M Φ The minimum value among the column norm, spectral norm, and row norm; argmax(β) 1,2,∞ (M) Φ )) represents the evaluation factor preprocessing matrix M Φ The maximum value among the column norm, spectral norm, and row norm;

[0024] Step 3.3: Define the evaluation factor volatility threshold Z for M. Φ Each evaluation factor in the matrix is ​​measured individually. If an evaluation factor exceeds its own fluctuation threshold, the corresponding evaluation factor is set to "0"; otherwise, the evaluation factor remains unchanged. This process ultimately yields the evaluation factor matrix for driving intention recognition.

[0025] Step 4: Calculate the evaluation factor matrix for driving intention recognition using equation (6). The r-th driving intention evaluation factor corresponding to the nth time point in the same row The r-th driving intention evaluation factor corresponding to the (n-1)-th time point The system fluctuation θ between r,n-1 Finally, the system fluctuation vector set Θ=(θ) is obtained. 1,1 ,θ 1,2 ,…,θ r,n-1 ,…,θ m,l-1 );

[0026]

[0027] In equation (6), and These represent the evaluation factor matrices for driver intent recognition. The maximum and minimum values ​​of the evaluation factors;

[0028] Step 5: Process the system fluctuation quantities in the system fluctuation vector set Θ using the Tanh activation function to obtain the enabled system fluctuation vector set. This represents the r-th driving intention evaluation factor corresponding to the n-th time point after processing. The r-th driving intention evaluation factor corresponding to the (n-1)-th time point System fluctuations between;

[0029] Step 6: Use equation (7) to obtain the final recognition accuracy of the current driving intention recognition system.

[0030]

[0031] In equation (7), c1 represents the correction coefficient.

[0032] The online evaluation method for the accuracy of driving intention recognition in an intelligent connected HEV, as described in this invention, is also characterized in that step 2.2 includes:

[0033] Step 2.2.1: Preprocess the eigenvalue space matrix P′ to obtain the preprocessed eigenvalue space matrix. in, Let represent the preprocessed feature coefficient space vector of the current vehicle at time t-l+i, and f t-l+i Let represent the offset of the current vehicle at time t-l+i, and obtain it from equation (1):

[0034]

[0035] Step 2.2.2: The feature reconstruction layer uses equation (2) to calculate the reconstruction correlation coefficient g of the current vehicle at time t-l+i. t-l+i and weighted ratio ε t-l+i :

[0036]

[0037] Step 2.2.3: The feature reconstruction layer uses equation (3) to calculate the feature coefficient reconstruction vector e of the current vehicle at time t-l+i. t-l+iFinally, the feature coefficient reconstruction matrix E = [e] for the time period T is obtained. t-l+1 ,e t-l+2 ,e t-l+i ,…,e t ];

[0038]

[0039] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the online evaluation method, and the processor is configured to execute the program stored in the memory.

[0040] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to perform the steps of the online evaluation method.

[0041] Compared with existing technologies, the beneficial effects of the present invention are as follows:

[0042] 1. This invention designs a vehicle twin feature parameter inverse evaluation model. Based on the vehicle's state data and driving intentions over a historical time period, a feature coefficient matrix is ​​formed. After feature parameter reconstruction, an evaluation factor output matrix is ​​obtained using an inverse function. This provides a means to obtain an evaluation factor set for evaluating driving intentions in complex driving environments of intelligent connected HEVs.

[0043] 2. This invention proposes a method for quantitatively evaluating the accuracy of driving intention recognition. By constructing a logical control rule, the output matrix of the evaluation factors for driving intention recognition is logically judged and processed. Then, the system fluctuation between each node of the evaluation factor output matrix is ​​calculated to obtain a set of fluctuation vectors. Finally, the recognition accuracy of the final driving intention recognition system is obtained by functional calculation, thus realizing accurate online evaluation of the accuracy of driving intention recognition in complex driving environments of intelligent connected HEVs. Attached Figure Description

[0044] Figure 1 This is a flowchart of the whole vehicle twin reverse evaluation model of the present invention;

[0045] Figure 2 This is a basic flowchart of the online evaluation method for the accuracy of driving intention recognition in intelligent connected HEVs according to the present invention. Detailed Implementation

[0046] In this embodiment, an online evaluation method for the accuracy of driving intention recognition in an intelligent connected HEV is applied to a multi-vehicle operation scenario of an intelligent connected hybrid electric vehicle consisting of a cloud server and a driving intention recognition system. This method evaluates the accuracy of driving intention recognition online to improve the safety and reliability of the driving system and is significant for guiding the human-like decision-making of the vehicle's driving system. First, V2V and V2I methods are used to acquire the current vehicle's state data within a historical time period. This data is then integrated with the driving intentions recognized by the driving system to form a multi-dimensional feature input matrix for the entire vehicle system. This matrix is ​​then input into a vehicle twin inverse evaluation model to obtain an evaluation factor output matrix, such as... Figure 1 As shown. Based on this, logical control rules are constructed to process the output matrix of the evaluation factors according to different rules. After processing, the system fluctuation between evaluation factors is calculated to obtain a set of fluctuation vectors. Finally, the recognition accuracy of the driving intention recognition system is calculated using a functional formula. Figure 2 As shown, specifically, the steps are as follows:

[0047] 1. An online evaluation method for the accuracy of driving intention recognition in an intelligent connected hybrid electric vehicle (HEV), characterized in that it is applied to a multi-vehicle operation scenario of an intelligent connected hybrid electric vehicle consisting of a cloud server and a driving intention recognition system, and is performed according to the following steps:

[0048] Step 1: Based on vehicle-to-vehicle (V2V) communication and vehicle-to-infrastructure (V2I) communication, obtain the current vehicle status data and the current vehicle's driving intention identified by the driving intention recognition system within the historical time period T, and upload them to the cloud server.

[0049] The cloud server integrates the current vehicle's state data from time t-1+1 to time t within the historical time period T, along with the driving intention, into a multi-dimensional feature input matrix P = [p t-l+1 ,p t-l+2 ,...,p t-l+i ,...,p t ], where p t-l+i p represents the feature parameter vector of the current vehicle at time t-l+i within the historical time period T. t Let represent the parameter vector of the current vehicle at time t within the historical time period T, and T t This represents the instantaneous engine torque of the vehicle at time t. This represents the instantaneous engine speed of the vehicle at time t. This represents the real-time fuel consumption of the vehicle at time t. This represents the motor output torque of the vehicle at time t. L represents the instantaneous motor speed of the vehicle at time t. t Mt Let α represent the real-time motor power and real-time engine power of the vehicle at time t, respectively. t I represents the instantaneous engine throttle opening of the vehicle at time t. t Let I represent the real-time driving intention of the vehicle at time t. t The category label is Y m l represents the number of time points;

[0050] Step 2: The cloud server constructs a vehicle twin reverse evaluation model, including: an input layer, a feature reconstruction layer, and an evaluation factor output layer;

[0051] Step 2.1, Processing of the input layer:

[0052] Step 2.1.1: Integrate the parameter vector at each time step in the multidimensional feature input matrix P to obtain the feature coefficient space matrix P′=[p′ t-l+1 ,p′ t-l+2 ,...,p′ t-l+i ,...,p′ t ], where p′ t-l+i This represents the feature coefficient space vector of the current vehicle at time t-l+i;

[0053] Step 2.2, Processing of the feature reconstruction layer:

[0054] Step 2.2.1: Preprocess the eigenvalue space matrix P′ to obtain the preprocessed eigenvalue space matrix. in, Let represent the preprocessed feature coefficient space vector of the current vehicle at time t-l+i, and f t-l+i Let represent the offset of the current vehicle at time t-l+i, and obtain it from equation (1):

[0055]

[0056] Step 2.2.2: The feature reconstruction layer uses equation (2) to calculate the reconstruction correlation coefficient g of the current vehicle at time t-l+i. t-l+i and weighted ratio ε t-l+i :

[0057]

[0058] Step 2.2.3: The feature reconstruction layer uses equation (3) to calculate the feature coefficient reconstruction vector e of the current vehicle at time t-l+i. t-l+i Finally, the feature coefficient reconstruction matrix E = [e] for the time period T is obtained. t-l+1 ,e t-l+2,e t-l+i ,…,e t ];

[0059]

[0060] Step 2.3, Processing of the evaluation factor output layer:

[0061] Step 2.3.1: Obtain the evaluation factor output matrix using the inverse function ψ(U×E+c). Where U is the set influence factor matrix, and c represents the loss compensation vector. Represents the output matrix of evaluation factors The set of driving intention evaluation factors corresponding to the nth time point, and Where, τ r,n This represents the r-th driving intention evaluation factor in the set of driving intention evaluation factors corresponding to the nth time point;

[0062] In this embodiment, l = 8 is chosen, and the evaluation factor output layer is defined as consisting of 8×8 nodes. The output matrix of the current vehicle in the output layer is: in Let τ represent the set of perturbation factors corresponding to the i-th node of the output layer, where τ 1,i τ represents the battery charging and discharging efficiency at the corresponding position under the i-th node. 2,i τ represents the rate of change of brake thermal decay at the corresponding position under the i-th node. 3,i τ represents the change in road surface geometry at the corresponding location of the i-th node. 4,i τ represents the battery temperature variation factor at the corresponding position under the i-th node. 5,i τ represents the volatility of sentiment intensity at the corresponding position under the i-th node. 6,i τ represents the intentional transformation mutation factor at the corresponding position under the i-th node. 7,i τ represents the sentiment metric extension factor at the corresponding position under the i-th node. 8,i Let c represent the emotion recognition fatigue coefficient at the corresponding position under the i-th node, and take c = (0.2, 0.5, 0.23, 0.8, 1, 1.3, 2, 3). T ;

[0063] Step 3: Construct logical control rules and perform logical judgment processing on the output matrix of the evaluation factors for driving intention recognition;

[0064] Step 3.1: Normalize each evaluation factor corresponding to each time point in the evaluation factor output matrix to obtain the final evaluation factor preprocessing matrix. in, This represents the r-th driving intention evaluation factor in the set of driving intention evaluation factors corresponding to the n-th time point after normalization.

[0065] Step 3.2: Obtain the evaluation factor preprocessing matrix M using equation (4). Φ The range of features;

[0066]

[0067] In equation (4), β1(M) Φ ) represents the evaluation factor preprocessing matrix M Φ The norm, β ∞ (M) Φ ) represents the evaluation factor preprocessing matrix M Φ row norm, β2(M) Φ ) represents the evaluation factor preprocessing matrix M Φ spectral norm, Preprocessing matrix M for evaluation factors Φ The normal matrix, λ max |M| represents the largest eigenvalue of the matrix. Φ |1 represents the preprocessing matrix M of the evaluation factors Φ The maximum sum of the evaluation factors in each row of the table, |M Φ | ∞ M represents the preprocessing matrix of evaluation factors. Φ The maximum sum of the evaluation factors in each column, |M Φ |2 represents the evaluation factor preprocessing matrix M Φ The maximum absolute value of the singular values.

[0068] Determine whether the condition shown in equation (5) is true. If it is true, set M to... Φ Evaluation factor matrix for driving intention recognition If the error persists, proceed to step 4; otherwise, proceed to step 3.3.

[0069] argmin(β 1,2,∞ (M) Φ ))≤β(Μ Φ )≤argmax(β 1,2,∞ (M) Φ (5)

[0070] In equation (5), argmin(β) 1,2,∞ (M) Φ )) represents the evaluation factor preprocessing matrix M Φ The minimum value among the column norm, spectral norm, and row norm; argmax(β) 1,2,∞ (M) Φ )) represents the evaluation factor preprocessing matrix M ΦThe maximum value among the column norm, spectral norm, and row norm;

[0071] Step 3.3: Define the evaluation factor volatility threshold Z for M. Φ Each evaluation factor in the matrix is ​​measured individually. If an evaluation factor exceeds its own fluctuation threshold, the corresponding evaluation factor is set to "0"; otherwise, the evaluation factor remains unchanged. This process ultimately yields the evaluation factor matrix for driving intention recognition. In this embodiment, Z = 1.

[0072] Step 4: Calculate the evaluation factor matrix for driving intention recognition using equation (6). The r-th driving intention evaluation factor corresponding to the nth time point in the same row The r-th driving intention evaluation factor corresponding to the (n-1)-th time point The system fluctuation θ between r,n-1 Finally, the system fluctuation vector set Θ=(θ) is obtained. 1,1 ,θ 1,2 ,…,θ r,n-1 ,…,θ m,l-1 );

[0073]

[0074] In equation (6), and These represent the evaluation factor matrices for driver intent recognition. The maximum and minimum values ​​of the evaluation factors;

[0075] Step 5: Process the system fluctuation quantities of the system fluctuation vector set Θ using the Tanh activation function to obtain the enabled system fluctuation vector set. This represents the r-th driving intention evaluation factor corresponding to the n-th time point after processing. The r-th driving intention evaluation factor corresponding to the (n-1)-th time point System fluctuations between;

[0076] Step 6: Use equation (7) to obtain the final recognition accuracy of the current driving intention recognition system.

[0077]

[0078] In equation (7), c1 represents the correction coefficient; in this embodiment, c1 = 0.8.

[0079] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0080] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. An online evaluation method for the accuracy of driving intention recognition in an intelligent connected HEV, characterized in that, It is applied in multi-vehicle operation scenarios of intelligent connected hybrid vehicles consisting of cloud servers and driving intention recognition systems, and is carried out in accordance with the following steps: Step 1: Obtain historical data based on vehicle-to-vehicle (V2V) communication and vehicle-to-infrastructure (V2I) communication. The current vehicle status data within the time period and the driving intention of the current vehicle identified by the driving intention recognition system are uploaded to the cloud server; The cloud server will record the history The current vehicle's state data from time t-l+1 to time t, along with the driving intention, are integrated into a multi-dimensional feature input matrix. ,in, Representing history The current vehicle during the time period The feature parameter vector at time t, Representing history The current vehicle during the time period The parameter vector at time step, and , Indicates the current vehicle is Instantaneous engine torque at a given moment Indicates the current vehicle is The instantaneous engine speed at a given moment. Indicates the current vehicle is Real-time fuel consumption at any given moment. Indicates the current vehicle is The motor output torque at any given moment. Indicates the current vehicle is The instantaneous motor speed at a given moment. , These respectively indicate the current vehicle is in Real-time motor power and real-time engine power at any given moment. Indicates the current vehicle is The instantaneous engine throttle opening at any given moment. Indicates the current vehicle is Real-time driving intentions at any moment, The category label is , Indicates the number of time points; Step 2: The cloud server constructs a vehicle twin reverse evaluation model, including: an input layer, a feature reconstruction layer, and an evaluation factor output layer; Step 2.1, Processing of the input layer: For multidimensional feature input matrix Integrating the parameter vector at each time step yields the eigenvalue space matrix. ,in, Indicates the current vehicle is The feature coefficient space vector at time t; Step 2.2: The feature reconstruction layer modulates the feature coefficient space matrix. Processing is performed to obtain eigencoefficient reconstruction matrix within a time period ; Step 2.3, Processing of the evaluation factor output layer: Using the inverse function Obtain the evaluation factor output matrix ,in, For the set influence factor matrix, Represents the loss compensation vector. Represents the evaluation factor output matrix The The set of driving intention assessment factors corresponding to each time point, and ,in, Indicates the first The first time point in the set of driving intention assessment factors One driving intention assessment factor; Step 3: Construct logical control rules and perform logical judgment processing on the output matrix of the evaluation factors for driving intention recognition; Step 3.1: Normalize each evaluation factor corresponding to each time point in the evaluation factor output matrix to obtain the final evaluation factor preprocessing matrix. ,in, This indicates the first normalized number. The first time point in the set of driving intention assessment factors One driving intention assessment factor; Step 3.2: Obtain the evaluation factor preprocessing matrix using equation (4). The characteristic range; (4) In equation (4), Represents the preprocessing matrix of evaluation factors The norm of the sequence, Represents the preprocessing matrix of evaluation factors row norm, Represents the preprocessing matrix of evaluation factors spectral norm, Preprocessing matrix for evaluation factors The normal matrix, Represents the largest eigenvalue of the matrix. Represents the preprocessing matrix of evaluation factors The maximum sum of the evaluation factors in each row. Represents the preprocessing matrix of evaluation factors The maximum sum of the evaluation factors in each column. Represents the preprocessing matrix of evaluation factors The maximum value of the absolute value of the singular values; Determine whether the condition shown in equation (5) is true. If it is true, then... Evaluation factor matrix for driving intention recognition If yes, proceed to step 4; otherwise, proceed to step 3.

3. (5) In equation (5), Represents the preprocessing matrix of evaluation factors The minimum value among the column norm, spectral norm, and row norm; Represents the preprocessing matrix of evaluation factors The maximum value among the column norm, spectral norm, and row norm; Step 3.3: Define the volatility threshold of the evaluation factor. ,right Each evaluation factor in the matrix is ​​measured individually. If an evaluation factor exceeds its own fluctuation threshold, the corresponding evaluation factor is set to "0"; otherwise, the evaluation factor remains unchanged. This process ultimately yields the evaluation factor matrix for driving intention recognition. ; Step 4: Calculate the evaluation factor matrix for driving intention recognition using equation (6). The nth time point corresponding to the same row Driving intention assessment factors The corresponding time point n-1 Driving intention assessment factors System fluctuations between Finally, the system fluctuation vector set is obtained. ; (6) In equation (6), and These represent the evaluation factor matrices for driver intent recognition. The maximum and minimum values ​​of the evaluation factors; Step 5: Apply the Tanh activation function to the system fluctuation vector set. The system fluctuations in the data are processed to obtain the enabled system fluctuation vector set. , This represents the nth time point after processing. Driving intention assessment factors The corresponding time point n-1 Driving intention assessment factors System fluctuations between; Step 6: Use equation (7) to obtain the final recognition accuracy of the current driving intention recognition system. : (7) In equation (7), This represents the correction factor.

2. The online evaluation method for the accuracy of driving intention recognition in an intelligent connected HEV according to claim 1, characterized in that, Step 2.2 includes: Step 2.2.1: Analyze the eigenvalue space matrix. Preprocessing is performed to obtain the preprocessed eigenvalue space matrix. ,in, Indicates the current vehicle is The feature coefficient space vector after preprocessing at time t, and , This indicates that the current vehicle is in The offset at time t is obtained from equation (1): (1) Step 2.2.2: The feature reconstruction layer uses equation (2) to calculate the current vehicle's position in the [missing information]. Reconstruction correlation coefficient at time 1 and weighted ratio : (2) Step 2.2.3: The feature reconstruction layer uses equation (3) to calculate the current vehicle's position in the [missing information]. The feature coefficient reconstruction vector at time 1 Finally obtained eigencoefficient reconstruction matrix within a time period ; (3)。 3. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store programs that support the processor in executing the online evaluation method of claim 1 or 2, and the processor is configured to execute the programs stored in the memory.

4. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the steps of the online evaluation method of claim 1 or 2.

Images