Heavy vehicle dynamic shift control strategy based on driving intention recognition

Abstract

This invention belongs to the field of vehicle control technology, specifically a dynamic gear shifting control strategy for heavy vehicles based on driving intention recognition. It adopts a hierarchical control architecture: In the perception and recognition layer, vehicle operating state parameters are collected, smoothed through local weighted regression, and classified using K-means clustering. An LSTM is used to output the current driving intention category in real time. In the speed prediction and optimization layer, an LSTM network is used to predict the vehicle speed in the prediction time domain. Simultaneously, an objective function is constructed using power performance and economic performance as evaluation indicators. A greedy algorithm is used to search for and solve for the local optimum in the prediction time domain to obtain the optimized shifting speed. In the strategy adaptation layer, driving intention is combined with the optimized shifting speed. Adaptive shifting rules are formulated by dynamically allocating weights for power performance and economic performance. This invention can solve the problems of high computational load, high algorithm complexity, and difficulty in online solution during the optimization and control of economic performance and power performance.

Description

Technical Field

[0001] This invention belongs to the field of vehicle control technology, specifically a dynamic gear shifting control strategy for heavy vehicles based on driving intention recognition. Background Technology

[0002] Heavy-duty vehicles play a vital role in both transportation and industry, and with industrial development, the performance requirements for these vehicles are constantly increasing. Power performance and economic efficiency, as core performance indicators of heavy-duty vehicles, directly affect driving performance and operating costs. Gear shifting has a significant impact on power transmission and fuel consumption, while driving intention reflects the driver's needs for performance, fuel efficiency, or comfort. Combining driving intention with gear shift control strategies helps promote the intelligent development of vehicles.

[0003] In recent years, vehicle shift control strategies have become a research hotspot. Optimization-based shift point optimization control methods have emerged as a viable technical approach. Currently, the main methods used are static optimization and fuzzy model control, which suffer from problems such as high computational load, high algorithm complexity, and difficulty in online solution. Therefore, this paper employs a greedy algorithm to optimize shift points, decomposing the global optimal solution for driving economy and power performance into local optimal solutions for each subproblem. Finally, the obtained local optimal solutions are superimposed and considered as the global optimal solution, thereby achieving real-time online optimization of the shift control strategy.

[0004] Current patents, such as the invention patent with patent number CN202511051420.7, propose a shift control system and method based on vehicle speed prediction, and the invention patent with patent number CN202411064164.0, propose an AMT shift strategy selection method, system, device, and medium based on operating condition prediction. The former introduces a vehicle speed prediction module, which improves the accuracy of shift timing to some extent, but does not fully consider the driver's driving intentions, driving habits, and reaction capabilities. The latter's shift decision-making method based on operating condition prediction, while shortening shift time, relies on traditional control logic based on offline clustering of historical data and operating condition matching, resulting in an inability to dynamically adjust shifts. Therefore, it is necessary to propose a collaborative control strategy that integrates dynamic optimization, online prediction, and a balance between economy and power performance to overcome the shortcomings of existing technologies in terms of vehicle driving performance and operating costs. Summary of the Invention

[0005] This invention aims to propose a dynamic shift control strategy for heavy-duty vehicles based on driving intention recognition. This strategy employs a hierarchical control architecture: In the perception and recognition layer, vehicle operating state parameters are collected, data is smoothed using a local weighted regression algorithm, and a classification threshold is determined by combining K-means clustering. A long short-term memory neural network (LSTM) is used to construct a driving intention recognition model, outputting the current driving intention category in real time. In the speed prediction and optimization layer, the LSTM network is used to predict the vehicle speed in the prediction time domain. Simultaneously, an objective function is constructed using power performance and economic performance as evaluation indicators. A greedy algorithm is used to search for and solve for local optimal solutions in the prediction time domain, obtaining the power-oriented and economy-oriented shift speeds optimized by the greedy strategy. In the strategy adaptation layer, the real-time recognized driving intention is combined with the optimized shift speed. Adaptive shift rules are formulated by dynamically allocating the weights of power performance and economic performance.

[0006] To solve the above technical problems, the present invention is implemented using the following technical solution:

[0007] A mechatronics-integrated composite anti-lock braking coordination control strategy for commercial vehicles includes the following steps:

[0008] S1: In the perception and recognition layer, vehicle operating status parameters are collected, data is smoothed through a local weighted regression algorithm, and a classification threshold is determined by combining K-means clustering. A driving intention recognition model is constructed using a long short-term memory neural network (LSTM) to output the current driving intention category in real time.

[0009] S2: In the speed prediction and optimization layer, the LSTM network is used to predict the vehicle speed in the prediction time domain. At the same time, the objective function is constructed with power performance and economic performance as evaluation indicators. The greedy algorithm is used to search for and solve the local optimal solution in the prediction time domain to obtain the optimized power-oriented shift speed and economy-oriented shift speed.

[0010] S3: In the adaptive strategy layer, the driving intentions identified in real time by S1 are combined with the shift speeds optimized by S2. By dynamically allocating the weights of power performance and economic performance, adaptive shift rules are formulated.

[0011] S4: Through the coordinated control of intent recognition and shift speed optimization, the shift points of heavy vehicles are dynamically adjusted in real time to maximize fuel economy while ensuring vehicle power performance.

[0012] In the preferred technical solution, the design process of the sensing and recognition layer in step S1 is specifically as follows:

[0013] S11: Select vehicle speed, accelerator pedal opening, and brake pedal opening as feature parameters of driver intent, and set the data window length to 10. Use a cubic function as a weighting function to smooth the vehicle speed data to remove random noise. The formula for the weighting function is:

[0014] (1)

[0015] In the formula, This indicates the data sampling point that needs to be filtered, i.e., vehicle speed;

[0016] S12: The K-means algorithm based on Euclidean distance is used to cluster the acceleration of different driving intentions to determine the classification threshold between different driving intentions. Specifically, driving intentions are divided into five categories: rapid acceleration, acceleration, cruising, deceleration and emergency braking.

[0017] (2)

[0018] In the formula, y represents the acceleration of the car; y represents the cluster center. This represents the number of feature parameters for the cluster centers; here, acceleration and deceleration are clustered once each, so n is 1.

[0019] S13 constructs an LSTM driving intention recognition model, which includes vehicle speed. Accelerator pedal opening Brake pedal opening The time series sample vector is used as the model's current time step. Input :

[0020] (3)

[0021] S14: Use the forget gate to determine the state information of the units that need to be forgotten:

[0022] (4)

[0023] Use the input gate to select the information to be stored in the new cell state. and create new candidate unit states. :

[0024] (5)

[0025] Using the aforementioned information, the old cell state Update to new unit status ;

[0026] Update and output the hidden state at the current time step. :

[0027] (6)

[0028] In the formula, For activation functions; It is the hyperbolic tangent function; This is the corresponding weight matrix; For the corresponding bias term; This is the hidden state from the previous time step;

[0029] S15: Pass the output of the LSTM network layer into the fully connected layer to output the probability of all possible target categories, and use the softmax function to perform classification output:

[0030] (7)

[0031] S16: To avoid instability caused by a single time step, the model's time step is set to 0.5s, the time window is set to 1s, the main driving intentions within 5 consecutive time steps are selected to represent the driving intentions of the entire time window, and the category with the highest probability is used as the real-time driving intention output.

[0032] In the preferred technical solution, the design process of the velocity prediction and optimization layer in step S2 is specifically as follows:

[0033] S21: The total time it takes for a vehicle to accelerate from a standstill to a target speed is used as an indicator of power performance evaluation. Its objective function is:

[0034] (8)

[0035] S22: Using total fuel consumption during vehicle acceleration as an economic performance evaluation indicator. Its objective function is:

[0036] (9)

[0037] In the formula, Number of gears; This refers to engine torque; For effective fuel consumption efficiency;

[0038] S23: Using a greedy algorithm, minimize the time domain of the vehicle speed predicted by LSTM. and To determine the direction of the shift, we progressively search for local optima between adjacent gears at different throttle openings, and obtain the optimized power-oriented shift speeds for each. Speed ​​of shifting gears with economic orientation .

[0039] In the preferred technical solution, the design process of the strategy adaptive layer in step S3 is specifically as follows:

[0040] S31: When the control system receives a driving intention output from layer S1 and identifies it as "rapid acceleration," it prioritizes power optimization, with economy optimization as a secondary consideration, taking into account the overall shift speed. The calculation is as follows:

[0041] (10)

[0042] S32: When the driving intention is recognized as normal "acceleration", the overall shift speed The calculation is as follows:

[0043] (11)

[0044] S33: When the driving intention is recognized as "cruising," "decelerating," or "emergency braking," the power demand is reduced, prioritizing economy and overall shift speed. Directly adopt economy-oriented shift speed:

[0045] (12)

[0046] S34: Based on the above calculation, the optimal overall shift speed is matched and compared with the real-time vehicle speed, and the final upshift or downshift control command is executed.

[0047] Compared with the prior art, the advantages of the present invention are:

[0048] 1. The dynamic shift control strategy for heavy vehicles based on driving intention recognition described in this invention adopts a hierarchical collaborative control architecture, which decouples intention recognition and strategy optimization in layers, making the control clearer and facilitating targeted tuning and porting.

[0049] 2. The dynamic shift control strategy for heavy vehicles based on driving intention recognition described in this invention integrates driving intention and greedy algorithm optimization to achieve online dynamic optimization of shift speed, thereby improving acceleration performance while reducing fuel consumption and taking into account multiple objectives.

[0050] 3. The dynamic gear shifting control strategy for heavy vehicles based on driving intention recognition described in this invention can solve the problems of large computational load, high algorithm complexity, and difficulty in online solution when optimizing and controlling economy and power. It decomposes the global optimal solution into the local optimal solutions of each subproblem, and then superimposes them as the global optimal solution, which can effectively reduce algorithm complexity and improve computation speed. Attached Figure Description

[0051] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0052] Figure 1 This is a flowchart of a dynamic gear shifting control strategy for heavy vehicles based on driving intention recognition, as described in this invention.

[0053] Figure 2 This is a flowchart of K-means clustering in a dynamic shift control strategy for heavy vehicles based on driving intention recognition, as described in this invention.

[0054] Figure 3 This is a flowchart of the greedy algorithm in the dynamic gear shifting control strategy for heavy vehicles based on driving intention recognition as described in this invention.

[0055] Figure 4 This is a diagram of an adaptive shift control strategy in a dynamic shift control strategy for heavy vehicles based on driving intention recognition, as described in this invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0057] The invention will now be further described with reference to the accompanying drawings.

[0058] See Figure 1 This invention provides a dynamic gear shifting control strategy for heavy vehicles based on driving intention recognition, specifically including the following steps:

[0059] S1: In the perception and recognition layer, vehicle operating status parameters are collected, data is smoothed through a local weighted regression algorithm, and a classification threshold is determined by combining K-means clustering. A driving intention recognition model is constructed using a long short-term memory neural network (LSTM) to output the current driving intention category in real time.

[0060] S2: In the speed prediction and optimization layer, the LSTM network is used to predict the vehicle speed in the prediction time domain. At the same time, the objective function is constructed with power performance and economic performance as evaluation indicators. The greedy algorithm is used to search for and solve the local optimal solution in the prediction time domain to obtain the optimized power-oriented shift speed and economy-oriented shift speed.

[0061] S3: In the strategy adaptive layer, the driving intention identified in real time by S1 is combined with the shift speed optimized by S2, and adaptive shift rules are formulated by dynamically allocating the weights of power performance and economic performance.

[0062] S4: Through the coordinated control of intent recognition and shift speed optimization, the shift points of heavy vehicles are dynamically adjusted in real time to maximize fuel economy while ensuring vehicle power performance.

[0063] S11: Select vehicle speed, accelerator pedal opening, and brake pedal opening as feature parameters of driver intent, and set the data window length to 10. Use a cubic function as a weighting function to smooth the vehicle speed data to remove random noise. The formula for the weighting function is:

[0064] (1)

[0065] In the formula, This indicates the data sampling point that needs to be filtered, i.e., vehicle speed;

[0066] See Figure 2 The K-means clustering process in step S1 is specifically as follows:

[0067] S12: The K-means algorithm based on Euclidean distance is used to cluster the acceleration of different driving intentions to determine the classification threshold between different driving intentions. Specifically, driving intentions are divided into five categories: rapid acceleration, acceleration, cruising, deceleration and emergency braking.

[0068] (2)

[0069] In the formula; y represents the acceleration of the car; y represents the cluster center. This represents the number of feature parameters for the cluster centers; here, acceleration and deceleration are clustered once each, so n is 1.

[0070] S13: Construct an LSTM driving intention recognition model, including vehicle speed Accelerator pedal opening Brake pedal opening The time series sample vector is used as the model's current time step. Input :

[0071] (3)

[0072] S14: Use the forget gate to determine the state information of the units that need to be forgotten:

[0073] (4)

[0074] Use the input gate to select the information to be stored in the new cell state. and create new candidate unit states. :

[0075] (5)

[0076] Using the aforementioned information, the old cell state Update to new unit status ;

[0077] Update and output the hidden state at the current time step. :

[0078] (6)

[0079] In the formula, For activation functions; It is the hyperbolic tangent function; This is the corresponding weight matrix; For the corresponding bias term; This is the hidden state from the previous time step;

[0080] S15: Pass the output of the LSTM network layer into the fully connected layer to output the probability of all possible target categories, and use the softmax function to perform classification output:

[0081] (7)

[0082] S16: To avoid instability caused by a single time step, the model's time step is set to 0.5s, the time window is set to 1s, the main driving intentions within 5 consecutive time steps are selected to represent the driving intentions of the entire time window, and the category with the highest probability is used as the real-time driving intention output.

[0083] See Figure 3 The greedy algorithm process in step S2 is as follows:

[0084] S21: The total time it takes for a vehicle to accelerate from a standstill to a target speed is used as an indicator of power performance evaluation. Its objective function is:

[0085] (8)

[0086] S22: Using total fuel consumption during vehicle acceleration as an economic performance evaluation indicator. Its objective function is:

[0087] (9)

[0088] In the formula, Number of gears; This refers to engine torque; For effective fuel consumption efficiency;

[0089] S23: Using a greedy algorithm, minimize the time domain of the vehicle speed predicted by LSTM. and To determine the direction of the shift, we progressively search for local optima between adjacent gears at different throttle openings, and obtain the optimized power-oriented shift speeds for each. Speed ​​of shifting gears with economic orientation .

[0090] See Figure 4 The adaptive shift control process in step S3 is specifically as follows:

[0091] S31: When the control system receives a driving intention output from layer S1 and identifies it as "rapid acceleration," it prioritizes power optimization, with economy optimization as a secondary consideration, taking into account the overall shift speed. The calculation is as follows:

[0092] (10)

[0093] S32: When the driving intention is recognized as normal "acceleration", the overall shift speed The calculation is as follows:

[0094] (11)

[0095] S33: When the driving intention is recognized as "cruising," "decelerating," or "emergency braking," the power demand is reduced, prioritizing economy and overall shift speed. Directly adopt economy-oriented shift speed:

[0096] (12)

[0097] S34: Based on the above calculation, the optimal overall shift speed is matched and compared with the real-time vehicle speed, and the final upshift or downshift control command is executed.

[0098] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic gear shifting control strategy for heavy-duty vehicles based on driving intention recognition, characterized in that, Includes the following steps: S1: In the perception and recognition layer, vehicle operating status parameters are collected, data is smoothed through a local weighted regression algorithm, and a classification threshold is determined by combining K-means clustering. A driving intention recognition model is constructed using a long short-term memory neural network (LSTM) to output the current driving intention category in real time. S2: In the speed prediction and optimization layer, the LSTM network is used to predict the vehicle speed in the prediction time domain. At the same time, the objective function is constructed with power performance and economic performance as evaluation indicators. The greedy algorithm is used to search for and solve the local optimal solution in the prediction time domain to obtain the optimized power-oriented shift speed and economy-oriented shift speed. S3: In the strategy adaptive layer, the driving intention identified in real time by S1 is combined with the shift speed optimized by S2, and adaptive shift rules are formulated by dynamically allocating the weights of power performance and economic performance. S4: Through the coordinated control of intent recognition and shift speed optimization, the shift points of heavy vehicles are dynamically adjusted in real time to maximize fuel economy while ensuring vehicle power performance.

2. The dynamic gear shifting control strategy for heavy vehicles based on driving intention recognition according to claim 1, characterized in that, The design process of the perception and recognition layer in step S1 is as follows: S11: Select vehicle speed, accelerator pedal opening, and brake pedal opening as feature parameters of driver intent, and set the data window length to 10. Use a cubic function as a weighting function to smooth the vehicle speed data to remove random noise. The formula for the weighting function is: （1） In the formula, This indicates the data sampling point that needs to be filtered, i.e., vehicle speed; S12: The K-means algorithm based on Euclidean distance is used to cluster the acceleration of different driving intentions to determine the classification threshold between different driving intentions. Specifically, driving intentions are divided into five categories: rapid acceleration, acceleration, cruising, deceleration and emergency braking. （2） In the formula, y represents the acceleration of the car; y represents the cluster center. This represents the number of feature parameters for the cluster centers; here, acceleration and deceleration are clustered once each, so n is 1. S13: Construct an LSTM driving intention recognition model, including vehicle speed Accelerator pedal opening Brake pedal opening The time series sample vector is used as the model's current time step. Input : （3） S14: Use the forget gate to determine the state information of the units that need to be forgotten: （4） Use the input gate to select the information to be stored in the new cell state. and create new candidate unit states. : （5） Using the aforementioned information, the old cell state Update to new unit status ; Update and output the hidden state at the current time step. : （6） In the formula, For activation functions; It is the hyperbolic tangent function; This is the corresponding weight matrix; For the corresponding bias term; This is the hidden state from the previous time step; S15: Pass the output of the LSTM network layer into the fully connected layer to output the probability of all possible target categories, and use the softmax function to perform classification output: （7） S16: To avoid instability caused by a single time step, the model's time step is set to 0.5s, the time window is set to 1s, the main driving intentions within 5 consecutive time steps are selected to represent the driving intentions of the entire time window, and the category with the highest probability is used as the real-time driving intention output.

3. The dynamic gear shifting control strategy for heavy vehicles based on driving intention recognition according to claim 1, characterized in that, The design process of the velocity prediction and optimization layer in step S2 is as follows: S21: The total time it takes for a vehicle to accelerate from a standstill to a target speed is used as an indicator of power performance evaluation. Its objective function is: （8） S22: Using total fuel consumption during vehicle acceleration as an economic performance evaluation indicator. Its objective function is: （9） In the formula, Number of gears; This refers to engine torque; For effective fuel consumption efficiency; S23: Using a greedy algorithm, minimize the time domain of the vehicle speed predicted by LSTM. and To determine the direction of the shift, we progressively search for local optima between adjacent gears at different throttle openings, and obtain the optimized power-oriented shift speeds for each. Speed ​​of shifting gears with economic orientation .

4. The dynamic gear shifting control strategy for heavy vehicles based on driving intention recognition according to claim 1, characterized in that, The design process of the policy adaptation layer in step S3 is as follows: S31: When the control system receives a driving intention output from layer S1 and identifies it as "rapid acceleration," it prioritizes power optimization, with economy optimization as a secondary consideration, taking into account the overall shift speed. The calculation is as follows: （10） S32: When the driving intention is recognized as normal "acceleration", the overall shift speed The calculation is as follows: （11） S33: When the driving intention is recognized as "cruising," "decelerating," or "emergency braking," the power demand is reduced, prioritizing economy and overall shift speed. Directly adopt economy-oriented shift speed: （12） S34: Based on the above calculation, the optimal overall shift speed is matched and compared with the real-time vehicle speed, and the final upshift or downshift control command is executed.

Images