A hybrid power distribution method for a hybrid tracked vehicle using a deep neural network

Abstract

The application provides a hybrid power tracked vehicle power distribution method fusing a deep neural network, which utilizes a GRU deep neural network to predict the output power of an engine-generator set, has the advantages of good fuel economy, stable state of charge and easy real vehicle application, etc.; in the method, the data set of characteristic parameters is reduced in dimension through a principal component analysis method, and then the reduced data is used as the input of the neural network, so that the calculation speed of the controller can be effectively improved; the application fully utilizes the neural network to extract a global optimal control rule and apply the rule to real-time energy management of the vehicle, and can provide a beneficial reference for designing a method of fusing a deep neural network to plan power distribution for various hybrid power vehicles not limited to tracked forms.

Description

Technical Field

[0001] This invention belongs to the field of hybrid vehicle energy management technology, specifically relating to a power distribution method for hybrid tracked vehicles that integrates deep neural networks. Background Technology

[0002] With the increasing application of hybrid electric drive technology in tracked vehicles, these vehicles, compared to traditional models, exhibit more diverse and complex operating conditions, higher fuel consumption, and often lack the ability to be externally charged during actual use. Therefore, they require significantly different characteristics, such as maintaining the state of charge (SOC) of the power battery within a reasonable range. Consequently, it is necessary to design energy management strategies suitable for hybrid tracked vehicles to improve fuel economy, increase driving range, and reduce emissions. Some existing energy management technologies for hybrid tracked vehicles employ dynamic programming-based hybrid system control strategies. These strategies can obtain optimal decisions based on known global operating information and global optimization of control and state variables, ensuring globally optimal control results. However, since this strategy relies on obtaining global vehicle driving information, it is difficult to balance optimality with real-time performance, making it unsuitable for actual vehicle operation. Therefore, there is an urgent need in this field for a new energy management method that can fully learn and utilize globally optimal control rules, improve computational speed, be suitable for real-time applications, and achieve the goals of energy saving and maintaining a stable SOC in hybrid tracked vehicles. Summary of the Invention

[0003] In view of this, and to address the technical problems existing in this field, the present invention provides a power distribution method for hybrid tracked vehicles that integrates deep neural networks, specifically including the following steps:

[0004] Step 1: Establish the vehicle dynamics model, power component model, and vehicle power balance model for the hybrid tracked vehicle.

[0005] Step 2: Collect historical real-world operating data for multiple hybrid tracked vehicles of the same model, including vehicle speed, acceleration, yaw rate, and yaw acceleration. For each model established in Step 1, sequentially set the cost function, state transition equation, and system constraints for the global optimal dynamic programming algorithm. Execute the global optimal dynamic programming algorithm offline using the operating data to obtain the optimal energy management control sequence for different operating conditions. The control variables in the control sequence include the engine's electronic throttle opening (thr), and the state variables are the battery's state of charge (SOC) and generator speed (ω). g ;

[0006] Step 3: Select multiple characteristic parameters that affect the power distribution of the hybrid tracked vehicle, and store the control sequence results obtained by the global optimal dynamic programming algorithm corresponding to these characteristic parameters; use principal component analysis to reduce the dimensionality of the selected characteristic parameters.

[0007] Step 4: Establish a real-time power distribution prediction model for the vehicle based on a deep neural network. The prediction model takes the principal component feature parameters obtained after dimensionality reduction in Step 3 as input and the target output power of the optimal engine-generator set as output. The deep neural network is trained using historical data corresponding to the principal component feature parameters.

[0008] Step 5: Apply the trained real-time power distribution prediction model online to the real-time energy management of the hybrid tracked vehicle, input the principal component feature parameters collected in real time, and predict the target output power curve of the vehicle's optimal engine-generator set.

[0009] Furthermore, the power system configuration of the hybrid tracked vehicle specifically adopts the form of full-wave rectification of the AC power generated by the engine-generator set into DC power, and the power battery is directly connected in parallel to the bus.

[0010] For this type of vehicle, the following type of vehicle dynamics model is specifically established in step one:

[0011]

[0012] Where δ represents the rotational mass conversion factor, and m represents the vehicle's curb weight. v and v represent longitudinal acceleration and velocity, respectively; F1 and F2 represent the traction forces provided by the drive motors on both sides to the drive wheels; Fr1 and Fr2 represent the longitudinal ground resistance forces that the vehicle needs to overcome on the left and right sides, respectively; I z The moment of inertia of the vehicle, w and These represent the vehicle's yaw rate and yaw acceleration, respectively; B represents the track center distance; M represents... r F represents the steering resistance torque acting on the vehicle. drive F represents the driving force. f and F w f and C represent the rolling resistance and air resistance that the vehicle needs to overcome, respectively. D Let g and μ represent the rolling resistance coefficient and air resistance coefficient of the vehicle, respectively; g is the acceleration due to gravity; A is the equivalent frontal area of ​​the vehicle; l is the ground contact length of the track; and μ is the steering resistance coefficient of the vehicle.

[0013] The power component models include an engine fuel consumption lookup table model, an engine-generator set model, a power battery model, and a drive motor model; the specific form of the engine fuel consumption lookup table model is as follows:

[0014]

[0015] Among them, b e n represents the engine's fuel consumption rate. e T is the engine speed. e This is the engine torque. This refers to the engine fuel consumption per unit time.

[0016] The engine-generator model specifically includes the equations for uncontrolled rectified voltage and electromagnetic torque, which are as follows:

[0017]

[0018] Among them, U g The DC voltage ω output by the generator g For generator torque, I g P is the DC current output by the generator. g T represents the power output of the generator. g For electromagnetic torque, K e K is the equivalent electromotive force coefficient of the generator. x This is the equivalent impedance coefficient of the generator;

[0019] and engine output torque T e and generator electromagnetic torque T g The dynamic equilibrium formula is in the following form:

[0020]

[0021] Among them, i e-g J is the transmission ratio from the engine to the generator. e J is the engine's moment of inertia. g Let t be the moment of inertia of the generator, d be the time variable, and d be the differential operator. Based on the DC-side voltage and electromagnetic torque equations after uncontrolled rectification of the engine-generator set, the target speed n of the generator is... g The calculation formula is:

[0022]

[0023] The specific form of the power battery model is as follows:

[0024]

[0025] Among them, U b V is the load voltage of the power battery. oc I represents the open-circuit voltage of the battery. b R is the battery output current. int P represents the internal resistance of the battery. bThe SOC represents the battery's state of charge, which is the output power of the power battery. initial C represents the initial value of the state of charge. b Indicates the battery pack capacity;

[0026] The specific form of the drive motor model is as follows:

[0027] η m =η m (n m T m )

[0028] Where, η m For the output efficiency of the drive motor, n m and T m These are the speed and torque of the drive motor, respectively.

[0029] The specific form of the vehicle power balance model is as follows:

[0030]

[0031] Among them, P d The total power requirement is given by η1 and η2, which are the efficiencies of the left and right drive motors, respectively. T1 and T2 are the torques of the left and right drive motors, respectively. i0 is the side reduction ratio, r is the radius of the drive wheel, and η... T This refers to the efficiency of the transmission system.

[0032] Furthermore, the specific form of the cost function for the globally optimal dynamic programming algorithm set in step two is as follows:

[0033]

[0034] Where k represents the discrete time and N is the control duration;

[0035] The specific form of the state transition equation is as follows:

[0036]

[0037] The system's constraints specifically include:

[0038]

[0039] Wherein, ΔSOC max ω is the maximum state-of-charge deviation value. g_min and ω g_max T represents the minimum and maximum speeds of the generator. g_min and T g_max For the minimum and maximum torque of the generator, thr min and thr max For the minimum and maximum throttle opening, I b_maxThe maximum battery current to ensure safe battery operation, Δω max For the maximum rate of change of rotational speed, I g_max The maximum generator output current is given by SOC(final) and SOC(0), which are the values ​​of the state of charge at the final and zero time points, respectively.

[0040] Furthermore, in step three, 11 characteristic parameters affecting the power distribution of hybrid tracked vehicles (speed, acceleration, yaw rate, yaw acceleration, total vehicle power demand, rate of change of total vehicle power demand, SOC deviation, generator speed, engine torque, battery current, and bus voltage) are specifically selected for principal component analysis. Through dimensionality reduction, the following mapping relationship between the principal component features and the original data is obtained:

[0041] F i =a 1i X1+a 2i X2 + ... + a ni X n

[0042] Among them, F i The i-th principal component is a weighted combination of n observed variables; the calculated weight coefficients need to be saved, and the principal component feature parameters are used as the prediction input of the deep neural network.

[0043] Furthermore, in step four, a gated recurrent unit (GRU) deep neural network is specifically used to build a real-time power distribution prediction model for the vehicle.

[0044] Furthermore, when applying the trained real-time power allocation prediction model online, the predicted engine-generator set target output power curve is smoothed using a real-time wavelet filter, and the processed engine-generator set target power P is optimized. g The following constraints must be met:

[0045]

[0046] Among them, P req P is the total power demand of the vehicle. b_max P represents the maximum charge and discharge power of the power battery. eng_min P is the minimum power of the engine. eng_max This refers to the engine's maximum power.

[0047] Using the target power P of the engine-generator set g The output power P of the power battery was calculated. b :

[0048] P b =P d -P g .

[0049] Furthermore, steps two through four are repeated periodically to update the real-time power allocation prediction model.

[0050] The power distribution method for hybrid tracked vehicles fused with deep neural networks provided by the present invention offers at least the following advantages over existing technologies:

[0051] (1) This method uses a GRU deep neural network to predict the output power of the engine-generator set, which has the advantages of good fuel economy, stable charge state and easy application in real vehicles.

[0052] (2) In this method, principal component analysis is used to reduce the dimensionality of the dataset of feature parameters, and then the dimensionality-reduced data is used as the input of the neural network, which can effectively improve the calculation speed of the controller.

[0053] (3) This invention makes full use of neural networks to extract global optimal control rules and apply them to real-time energy management of vehicles, which can provide a useful reference for designing a method for power allocation that integrates deep neural networks for various hybrid vehicles, not limited to tracked vehicles. Attached Figure Description

[0054] Figure 1 This is an overall flowchart of the method provided by the present invention;

[0055] Figure 2 This is the overall architecture diagram of power distribution for hybrid tracked vehicles that incorporate deep neural networks. Detailed Implementation

[0056] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The power distribution method for hybrid tracked vehicles fused with deep neural networks provided by this invention, such as... Figure 1 As shown, the specific steps include:

[0058] Step 1: Establish the vehicle dynamics model, power component model, and vehicle power balance model for the hybrid tracked vehicle.

[0059] Step 2: Collect historical real-world operating data for multiple hybrid tracked vehicles of the same model, including vehicle speed, acceleration, yaw rate, and yaw acceleration. For each model established in Step 1, sequentially set the cost function, state transition equation, and system constraints for the global optimal dynamic programming algorithm. Execute the global optimal dynamic programming algorithm offline using the operating data to obtain the optimal energy management control sequence for different operating conditions. The control variables in the control sequence include the engine's electronic throttle opening (thr), and the state variables are the battery's state of charge (SOC) and generator speed (ω). g ;

[0060] Step 3: Select multiple characteristic parameters that affect the power distribution of the hybrid tracked vehicle, and store the control sequence results obtained by the global optimal dynamic programming algorithm corresponding to these characteristic parameters; use principal component analysis to reduce the dimensionality of the selected characteristic parameters.

[0061] Step 4: Establish a real-time power distribution prediction model for the vehicle based on a deep neural network. The prediction model takes the principal component feature parameters obtained after dimensionality reduction in Step 3 as input and the target output power of the optimal engine-generator set as output. The deep neural network is trained using historical data corresponding to the principal component feature parameters.

[0062] Step 5: Apply the trained real-time power distribution prediction model online to the real-time energy management of the hybrid tracked vehicle, input the principal component feature parameters collected in real time, and predict the target output power curve of the vehicle's optimal engine-generator set.

[0063] In a preferred embodiment of the present invention, the power system configuration of the hybrid tracked vehicle specifically adopts the form of full-wave rectification of the AC power generated by the engine-generator set into DC power, and the power battery being directly connected in parallel to the bus.

[0064] For this type of vehicle, the following type of vehicle dynamics model is specifically established in step one:

[0065]

[0066] Where δ represents the rotational mass conversion factor, and m represents the vehicle's curb weight. v and v represent longitudinal acceleration and velocity, respectively; F1 and F2 represent the traction forces provided by the drive motors on both sides to the drive wheels; Fr1 and Fr2 represent the longitudinal ground resistance forces that the vehicle needs to overcome on the left and right sides, respectively; I z The moment of inertia of the vehicle, w and These represent the vehicle's yaw rate and yaw acceleration, respectively; B represents the track center distance; M represents... r F represents the steering resistance torque acting on the vehicle.drive F represents the driving force. f and F w f and C represent the rolling resistance and air resistance that the vehicle needs to overcome, respectively. D Let g and μ represent the rolling resistance coefficient and air resistance coefficient of the vehicle, respectively; g is the acceleration due to gravity; A is the equivalent frontal area of ​​the vehicle; l is the ground contact length of the track; and μ is the steering resistance coefficient of the vehicle.

[0067] The power component models include an engine fuel consumption lookup table model, an engine-generator set model, a power battery model, and a drive motor model; the specific form of the engine fuel consumption lookup table model is as follows:

[0068]

[0069] Among them, b e n represents the engine's fuel consumption rate. e T is the engine speed. e This is the engine torque. This refers to the engine fuel consumption per unit time.

[0070] The engine-generator model specifically includes the equations for uncontrolled rectified voltage and electromagnetic torque, which are as follows:

[0071]

[0072] Among them, U g The DC voltage ω output by the generator g For generator torque, I g P is the DC current output by the generator. g T represents the power output of the generator. g For electromagnetic torque, K e K is the equivalent electromotive force coefficient of the generator. x This is the equivalent impedance coefficient of the generator;

[0073] and engine output torque T e and generator electromagnetic torque T g The dynamic equilibrium formula is in the following form:

[0074]

[0075] Among them, i e-g J is the transmission ratio from the engine to the generator. e J is the engine's moment of inertia. g Let t be the moment of inertia of the generator, d be the time variable, and d be the differential operator. Based on the DC-side voltage and electromagnetic torque equations after uncontrolled rectification of the engine-generator set, the target speed n of the generator is... g The calculation formula is:

[0076]

[0077] The specific form of the power battery model is as follows:

[0078]

[0079] Among them, U b V is the load voltage of the power battery. oc I represents the open-circuit voltage of the battery. b R is the battery output current. int P represents the internal resistance of the battery. b The SOC represents the battery's state of charge, which is the output power of the power battery. initiai C represents the initial value of the state of charge. b Indicates the battery pack capacity;

[0080] The specific form of the drive motor model is as follows:

[0081] η m =η m (n m T m )

[0082] Where, η m For the output efficiency of the drive motor, n m and T m These are the speed and torque of the drive motor, respectively.

[0083] The specific form of the vehicle power balance model is as follows:

[0084]

[0085] Among them, P d The total power requirement is given by η1 and η2, which are the efficiencies of the left and right drive motors, respectively. T1 and T2 are the torques of the left and right drive motors, respectively. i0 is the side reduction ratio, r is the radius of the drive wheel, and η... T This refers to the efficiency of the transmission system.

[0086] In a preferred embodiment of the present invention, the cost function of the globally optimal dynamic programming algorithm set in step two has the following specific form:

[0087]

[0088] Where k represents the discrete time and N is the control duration;

[0089] The specific form of the state transition equation is as follows:

[0090]

[0091] The system's constraints specifically include:

[0092]

[0093] Wherein, ΔSOC max ω is the maximum state-of-charge deviation value. g_min and ω g_max T represents the minimum and maximum speeds of the generator. g_min and T g_max For the minimum and maximum torque of the generator, thr min and thr max For the minimum and maximum throttle opening, I b_max The maximum battery current to ensure safe battery operation, Δω max For the maximum rate of change of rotational speed, I g_max The maximum generator output current is given by SOC(final) and SOC(0), which are the values ​​of the state of charge at the final and zero time points, respectively.

[0094] In a preferred embodiment of the present invention, step three specifically selects 11 characteristic parameters affecting the power distribution of hybrid tracked vehicles (speed, acceleration, yaw rate, yaw acceleration, total vehicle power demand, rate of change of total vehicle power demand, SOC deviation, generator speed, engine torque, battery current, and bus voltage) to perform principal component analysis. The following mapping relationship between the principal component features and the original data is obtained through dimensionality reduction:

[0095] F i =a 1i X1+a 2i X2 + ... + a ni X n

[0096] Among them, F i The i-th principal component is a weighted combination of n observed variables; the calculated weight coefficients need to be saved, and the principal component feature parameters are used as the prediction input of the deep neural network.

[0097] In a preferred embodiment of the present invention, step four specifically employs a gated recurrent unit (GRU) deep neural network to build a real-time power distribution prediction model for the vehicle.

[0098] When applying the trained real-time power allocation prediction model online, the predicted engine-generator target output power curve is smoothed using a real-time wavelet filter, and the processed engine-generator target power P is optimized. g The following constraints must be met:

[0099]

[0100] Among them, P req P is the total power demand of the vehicle. b_max P represents the maximum charge and discharge power of the power battery. eng_min P is the minimum power of the engine. eng_max This refers to the engine's maximum power.

[0101] Using the target power P of the engine-generator set g The output power P of the power battery was calculated. b :

[0102] P b =P d -P g .

[0103] In a preferred embodiment of the invention, steps two through four are periodically repeated to update the real-time power distribution prediction model. The overall architecture diagram of the power distribution of a hybrid tracked vehicle incorporating a deep neural network is shown below. Figure 2 As shown.

[0104] It should be understood that the sequence number of each step in the embodiments of the present invention does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0105] 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 power distribution method for hybrid tracked vehicles incorporating deep neural networks, characterized in that: Specifically, the following steps are included: Step 1: Establish a vehicle dynamics model, a power component model, and a vehicle power balance model for the hybrid tracked vehicle. Specifically, the power system configuration of the hybrid tracked vehicle adopts the form of full-wave rectification of the AC power generated by the engine-generator set into DC power, and the power battery is directly connected in parallel to the bus. For this type of vehicle, the following type of vehicle dynamics model is established: in, This represents the rotational mass conversion factor. Indicates the vehicle's curb weight. and These represent longitudinal acceleration and velocity, respectively. and These represent the traction force provided by the two drive motors to the drive wheel. and These represent the longitudinal resistance to ground movement that the vehicle needs to overcome on the left and right sides, respectively. This represents the vehicle's moment of inertia. and These represent the vehicle's yaw rate and yaw acceleration, respectively. Indicates the center distance of the tracks. This represents the steering resistance torque acting on the vehicle. Indicates driving force. and These represent the rolling resistance and air resistance that the vehicle needs to overcome, respectively. and These represent the rolling resistance coefficient and the air resistance coefficient of the vehicle, respectively. It is the acceleration due to gravity. This indicates the vehicle's equivalent frontal area. Indicates the ground contact length of the track. This indicates the vehicle's steering resistance coefficient; The power component models include an engine fuel consumption lookup table model, an engine-generator set model, a power battery model, and a drive motor model; the specific form of the engine fuel consumption lookup table model is as follows: in, This refers to the engine's fuel consumption rate. This refers to the engine's rotational speed. For engine output torque, This refers to the engine fuel consumption per unit time. The engine-generator model specifically includes the equations for uncontrolled rectified voltage and electromagnetic torque, which are as follows: in, This is the DC voltage output by the generator. For generator speed, This refers to the DC current output by the generator. This refers to the power output of the generator. For the electromagnetic torque of the generator, This is the equivalent electromotive force coefficient of the generator. This is the equivalent impedance coefficient of the generator; and engine output torque and generator electromagnetic torque The dynamic equilibrium formula is in the following form: in, This refers to the transmission ratio from the engine to the generator. This is the engine's moment of inertia. Let be the moment of inertia of the generator. t For time variables, d The differential operator; based on the DC-side voltage and electromagnetic torque equations after uncontrolled rectification of the engine-generator set, the target speed of the generator is... The calculation formula is: The specific form of the power battery model is as follows: in, This refers to the load voltage of the power battery. This indicates the open-circuit voltage of the battery. For battery output current, Indicates the battery's internal resistance. For the output power of the power battery, SOC Indicates the state of charge of the battery. Indicates the initial value of the state of charge. Indicates the battery pack capacity; The specific form of the drive motor model is as follows: in, η m To improve the output efficiency of the drive motor, and These are the speed and torque of the drive motor, respectively. The specific form of the vehicle power balance model is as follows: in, For the power required by the whole vehicle, and The efficiency of the left and right drive motors. and This represents the torque of the left and right drive motors. For the side reduction ratio, The radius of the driving wheel, For the efficiency of the transmission system; Step 2: Collect historical real-world operating data for multiple hybrid tracked vehicles of the same model, including vehicle speed, acceleration, yaw rate, and yaw acceleration. For each model established in Step 1, sequentially set the cost function, state transition equation, and system constraints for the global optimal dynamic programming algorithm. Execute the global optimal dynamic programming algorithm offline using the operating data to obtain the optimal energy management control sequence for different operating conditions. The control variables in the control sequence include: the engine's electronic throttle opening. The state variable is the state of charge of the power battery. and generator speed ; Step 3: Select multiple characteristic parameters that affect the power distribution of the hybrid tracked vehicle, and store the control sequence results obtained by the global optimal dynamic programming algorithm corresponding to these characteristic parameters; use principal component analysis to reduce the dimensionality of the selected characteristic parameters. Step 4: Build a real-time power distribution prediction model for the vehicle based on a deep neural network. The prediction model takes the principal component feature parameters obtained after dimensionality reduction in Step 3 as input and the target output power of the optimal engine-generator set as output. Train the deep neural network using historical data corresponding to the principal component feature parameters. Step 5: Apply the trained real-time power distribution prediction model online to the real-time energy management of the hybrid tracked vehicle, input the principal component feature parameters collected in real time, and predict the target output power curve of the vehicle's optimal engine-generator set.

2. The method as described in claim 1, characterized in that: The cost function of the globally optimal dynamic programming algorithm set in step two is as follows: in, k Representing discrete time, N To control the duration; The specific form of the state transition equation is as follows: The system's constraints specifically include: in, This is the maximum state-of-charge deviation value. and These are the minimum and maximum speeds of the generator. and For the minimum and maximum torque of the generator, and For minimum and maximum throttle opening, The maximum battery current to ensure safe battery operation. The maximum rate of change of rotational speed. This is the maximum generator output current. and These are the values ​​of the state of charge at the final time and at time 0, respectively.

3. The method as described in claim 2, characterized in that: Step three specifically selects 11 characteristic parameters that affect the power distribution of hybrid tracked vehicles: speed, acceleration, yaw rate, yaw acceleration, total vehicle power demand, and rate of change of total vehicle power demand. SOC Principal component analysis was performed on the deviation values, generator speed, engine torque, battery current, and bus voltage. Dimensionality reduction yielded the following mapping relationship between the principal component features and the original data: in, For the first i Each principal component is a principal component. n The weighted combination of the observed variables; the calculated weight coefficients need to be saved, and the principal component feature parameters are used as the prediction input of the deep neural network.

4. The method as described in claim 1, characterized in that: In step four, a gated recurrent unit deep neural network is used to build a real-time power distribution prediction model for the vehicle.

5. The method as described in claim 1, characterized in that: When applying the trained real-time power allocation prediction model online, the predicted engine-generator set target output power curve is smoothed using a real-time wavelet filter, and the processed engine-generator set target power is optimized. The following constraints must be met: in, This represents the total power required by the vehicle. This refers to the maximum charge and discharge power of the power battery. This is the engine's minimum power. This refers to the engine's maximum power. Using engine-generator set target power The output power of the power battery is calculated. : 。 6. The method as described in claim 1, characterized in that: Steps two through four are repeated periodically to update the real-time power distribution prediction model.

Images