A robust HDP energy management system and method for extended-range loaders

Abstract

The present invention discloses a robust HDP energy management system and method for an extended-range loader. The system includes a control layer and an execution layer. The control layer includes an on-board data collector, a demand power mapping module, a vehicle system model, an optimization objective function construction module, a robust HDP energy management controller, an engine controller, and a BMS controller. The on-board data collector obtains the vehicle's system state parameters in real time. The demand power mapping module obtains the demand power of the drive motor. The vehicle system model uses the vehicle speed and battery SOC at the next moment as output. The robust HDP energy management controller calculates the power allocated to the engine and the power allocated to the battery pack. The engine controller is used to control the power output of the engine, and the BMS controller is used to control the power output of the battery pack. The present invention reduces the strategy's dependence on the analytical model, has good real-time performance, and the online update method of the vehicle system model improves the working condition adaptability and robustness of the energy management strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of off-road vehicle energy management, and in particular to a robust HDP energy management system and method for an extended-range loader. Background Art

[0002] As one of the important ways for traditional loaders to achieve energy conservation and emission reduction, it is particularly important to tap into the energy conservation and emission reduction potential of extended-range loaders for optimizing the energy consumption of the entire machine. However, the differences in the design ideas of energy management strategies lead to different energy consumption optimization effects. In recent years, many patents have proposed energy management strategies based on reinforcement learning and compared them with strategies based on rules and optimal control theory to verify their optimization effects. However, their offline calibration and training processes are difficult to fully reflect the actual working conditions, and the strategy relies on analytical models, resulting in poor robustness and adaptability to working conditions. In addition, the operating conditions of loaders are complex, changeable, and highly periodic, which poses new challenges to the robustness of the strategy.

[0003] Chinese patent application CN112498334A discloses a robust energy management method and system for an intelligent connected hybrid vehicle. The method comprises: obtaining energy-saving driving decisions based on human-machine collaboration and global and real-time operating condition updates based on intelligent networking; executing an energy management strategy based on deep reinforcement learning based on the energy-saving driving decisions and the global and real-time operating condition updates; applying a robustness correction to the energy management strategy based on deep reinforcement learning to obtain a corrected robust control strategy; and applying the corrected robust control strategy to the hybrid vehicle to obtain an energy allocation result for the hybrid vehicle. The robust energy management method for an intelligent connected hybrid vehicle proposed in this invention effectively addresses the robustness issues of the deep reinforcement learning energy management strategy and improves the energy efficiency of hybrid vehicles in a connected environment, thus possessing significant engineering application value.

[0004] Chinese patent application CN116699991A discloses a fuzzy energy management method based on dynamic robust multi-objective optimization, which belongs to the field of automation technology. The present invention improves the multi-objective optimization algorithm NSGA-III from the two perspectives of fitness function and selection mechanism, and proposes an improved algorithm NSGA-III-SD. At the same time, the dynamic robust evaluation method based on prediction is integrated into the fuzzy EMS framework based on NSGA-III-SD, thereby improving the robustness of the controller under various working conditions. The dynamic robust multi-objective optimization fuzzy energy management strategy based on RNSGA-III-SD of the present invention not only meets the real-time power demand and reduces the energy consumption of the system, but also improves the fluctuation of the output current of the lithium battery under different driving conditions, which is of great significance to prolonging its service life.

[0005] Chinese patent application CN112026744A discloses a DQN variant-based energy management method for a series-parallel hybrid electric vehicle (SHEV) system. This method, which belongs to the field of SHEV technology, can improve training convergence speed and vehicle fuel economy. The method includes: establishing a SHEV model and obtaining environmental parameters that influence the energy management strategy, including road slope and vehicle mass; utilizing a dynamic programming (DP) algorithm to solve for the optimal energy management strategy, storing the experience in an optimal experience pool (OEB), and then employing a DuelingDQN strategy training model in conjunction with hybrid experience replay (HER) technology to obtain a trained deep reinforcement learning agent for energy management of the SHEV under different operating conditions. The HER technology and DQN variant Dueling architecture constructed in this invention can effectively improve training convergence speed, vehicle fuel economy, and algorithm robustness.

[0006] The aforementioned robust energy management patents primarily target road vehicles. Their offline calibration and training processes struggle to fully reflect real-world operating conditions, and their strategies rely on analytical models, resulting in poor robustness and adaptability to operating conditions. However, extended-range loaders are tightly coupled electromechanical, hydraulic, and mechanical systems. The complex transient operating conditions and input disturbances experienced during actual operation can easily cause time-varying model parameters and state instability, reducing the effectiveness and feasibility of existing energy management strategies. Summary of the Invention

[0007] In order to adapt to the harsh operating environment and complex and changeable working conditions of extended-range loaders, the present invention takes into account input disturbances such as ambient temperature, operating mechanism load, and slope during the loader operation, and proposes a robust HDP energy management system and method for extended-range loaders to improve the adaptability and robustness of the energy management strategy under different working conditions, thereby optimizing the consistency of the energy consumption of the entire vehicle.

[0008] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0009] A robust HDP energy management system for extended-range loaders includes a control layer and an execution layer. The control layer includes an on-board data collector, a demand power mapping module, a vehicle system model, an optimization objective function building module, a robust HDP energy management controller, an engine controller, and a BMS controller. HDP stands for heuristic dynamic programming.

[0010] The vehicle-mounted data collector obtains the vehicle's system status parameters in real time, including pedal opening p, engine power P eg , battery power P batt , vehicle speed v, battery SOC (State of Charge), ambient temperature T, workload Pload , slope angle θ;

[0011] The demand power mapping module is based on the pedal opening p collected by the vehicle data collector, and obtains the demand power P of the drive motor according to the calibrated pedal opening p and demand power mapping curve. d ;

[0012] The vehicle system model is based on the engine power P collected at the current moment eg , battery power P batt , vehicle speed v, battery SOC, ambient temperature T, workload P load , slope angle θ as input, and vehicle speed and battery SOC at the next moment as output;

[0013] The robust HDP energy management controller obtains the required power P according to the required power mapping module. d , the output of the vehicle system model and the optimization objective function constructed by the optimization objective function construction module, and use HDP to calculate the power P allocated to the engine e And the power P allocated to the battery pack b ;

[0014] The engine controller is used to control the power output of the engine, and the BMS controller is used to control the power output of the battery pack;

[0015] The execution layer includes the engine, generator, battery pack, rectifier, inverter, drive assembly, and gearbox. The output of the BMS controller is connected to the battery pack, the output of the engine controller is connected to the engine, the engine and the generator are mechanically connected, and the engine and the generator form a range extender. The generator is electrically connected to the battery pack through the rectifier, the battery pack and the rectifier are electrically connected to the inverter, the inverter and the drive assembly are electrically connected, the drive assembly is mechanically connected to the gearbox, and the gearbox is mechanically connected to the travel system and the operating system.

[0016] Preferably, the vehicle system model is constructed using RNN (Recurrent neural network), where the system state x includes the vehicle speed v and the battery pack SOC, and the control quantity u includes the power P allocated to the engine. e And the power P allocated to the battery pack b , that is, x = [v, SOC] T , u=[P e ,P b ] T ; Input disturbance is represented by ω, including ambient temperature T, workload P load , slope angle θ, that is, ω=[T,P load ,θ]; take [u(k) T ,x(k) T ,ω(k)]T As the RNN network input, the state quantity at the next moment As the RNN network output, the RNN vehicle system model is constructed:

[0017]

[0018] Where m h1 (k) is the hidden layer input of the RNN network; m h2 (k) is the output of the hidden layer of the RNN network, i represents the number of iterations, and the maximum number of iterations is set to N; k represents the current moment, k-1 represents the previous moment, and k+1 represents the next moment; Tansig and purelin functions are selected as the transfer functions of the hidden layer and output layer respectively, W m1 、W m2 、W m They represent the weight matrix from the input layer to the hidden layer, the weight matrix from the hidden layer to the output layer, and the weight matrix from the hidden layer to the hidden layer respectively; the weight training of the RNN-based vehicle system model adopts the gradient descent method until the error Reaching the set value, where The specific training process of weights is as follows:

[0019]

[0020] Where, l m represents the learning goal of the model network, Represents the weight matrix W from the hidden layer to the output layer of the i-th iteration m2 The transposed matrix of the vehicle system model network is W based on the historical operating condition information collected by the vehicle data collector. m1 、W m2 、W m Training is performed to obtain a trained vehicle system model; at the same time, in order to improve the adaptability of the vehicle system model to the working conditions and the robustness of the strategy, when the vehicle is running for a long time, When , the vehicle system model will be updated through online training, is the training window size, n is the number of online training times;

[0021] The energy consumption optimization objective function is expressed as:

[0022]

[0023] Among them, γ∈(0,1] is a discount factor, which means that the farther away from the current moment, the smaller the impact of the corresponding instantaneous energy consumption on the total energy consumption estimated at the current moment, k0 represents the initial moment, k f Indicates the end time, Q f (x(k),u(k),ω(k)) is the instantaneous energy consumption function, which can be abbreviated as Qf (k), is solved by the following formula:

[0024]

[0025] in, represents the fuel consumption rate of the engine, ρ f is the fuel density, Δt is the time step, ω M (k) is the bounding function of the known input perturbation ω(k), ||ω(k)||≤ω M (k), η is a positive constant.

[0026] Preferably, the robust HDP energy management controller includes a control network, an evaluation network 1, an evaluation network 2, and a vehicle system model network;

[0027] The control network is used to estimate the power P allocated to the engine according to the current vehicle state [v(k), SOC(k)] e (k), and based on P d (k) = P b (k)+P e (k)·η eg and the P obtained by the demand power mapping module d (k), calculate P b (k), η eg Indicates the range extender's power generation efficiency;

[0028] The vehicle system model network is used to calculate the vehicle state [v(k), SOC(k)], control quantity [P e (k),P b (k)], input disturbance [T(k),P load (k),θ(k)] generates the vehicle state at the next moment Evaluation network 1 and evaluation network 2 are used to evaluate the vehicle state at the current moment [v(k), SOC(k)] and the estimated vehicle state at the next moment Estimate and In this process, both the evaluation network and the execution network use their respective learning objectives E c (k) and E a (k) to update and adjust the network parameters and finally obtain the optimal control quantity And output to the engine controller and BMS controller for execution;

[0029] In the design process of the robust HDP energy management controller, due to the presence of input disturbances, the internal parameters of the evaluation network and the control network are changing at every moment. The internal network parameters of evaluation network 1 and evaluation network 2 are exactly the same and are trained and updated synchronously. Tansig and purelin functions are used as transfer functions of the hidden layer and the output layer. Accordingly, the transfer relationship of the evaluation network is expressed as:

[0030]

[0031] Among them, W c1 、W c2 、W c are the weight matrices from input layer to hidden layer, hidden layer to output layer, and hidden layer to hidden layer respectively; c h1 and c h2 Represent the input and output of the hidden layer respectively; i represents the number of iterations; represents the energy consumption of the vehicle at the i-th iteration at time k;

[0032] The learning goal of the evaluation network is to minimize Right now

[0033]

[0034] Among them, ε c Is to set the target error; use the gradient descent method to train and update W c1 、W c2 、W c The specific update process is as follows:

[0035]

[0036] l c ∈(0,1] represents the learning goal;

[0037] Control the network to minimize For the goal, To perform an estimate of the output of the network, make ε a is the target error; for a certain required power P d (k), engine power P e and battery pack power P b They are not independent of each other, and their relationship satisfies:

[0038] P d (k) = P b (k)+P e (k)·η eg (8)

[0039] Among them, ηeg represents the power generation efficiency of the range extender; therefore, the control vector u(k) is expressed as [P e (k),P d (k)-P e (k)·η eg ] T , the transfer functions of the hidden layer and the output layer are selected as tansig and purelin functions respectively, and the corresponding transfer relationship of the control network is expressed as:

[0040]

[0041] Among them, W a1 、W a2 and W a are the weight matrices from input layer to hidden layer, hidden layer to output layer, and hidden layer to hidden layer respectively; a h1 and a h2 Represent the input and output of the hidden layer respectively; use the gradient descent method to train and update W a1 、W a2 and W a The specific update process is as follows:

[0042]

[0043] l a ∈(0,1] represents the learning goal.

[0044] The present invention also provides a robust HDP energy management method for an extended-range loader, comprising the following steps:

[0045] S1: Use the vehicle data collector to obtain the real-time pedal opening p(k) and engine power P eg , battery pack power P batt , speed v(k), battery pack SOC(k), ambient temperature T(k), workload P load (k), slope angle θ(k) working condition information; the demand power mapping module obtains the demand power P of the drive motor based on the collected pedal opening p(k) information and the calibrated pedal opening and demand power mapping curve. d (k);

[0046] S2: Judgment Is it satisfied? If so, use the training window The vehicle system model is updated online based on the working condition data information and formula (2), and n=n+1 is set; if it is not satisfied, the vehicle system model of the previous moment is continued to be used;

[0047] S3: Based on the current state vector [v(k), SOC(k)] T, the evaluation network 1 and the execution network use equations (5) and (9) to calculate and Then based on and P d (k), calculated using formula (8)

[0048] S4: Based on [v(k), SOC(k)] T 、 and [T(k),P load (k),θ(k)], and the vehicle system model (1) is used to calculate On this basis, the evaluation network 2 is used to estimate And according to Q is calculated using formula (4) f (k);

[0049] S5: According to Q f (k), and Calculate E c (k); According to Q f (k) and Calculate E a (k);

[0050] S6: Judge E c (k)≤ε c And E a (k)≤ε a Whether it is satisfied or the number of iterations i>N; if these conditions are met, the Considered as optimal output And output to the actual extended-range loader for execution, the vehicle then generates a new state, and sets time k=k+1, and returns to step S1; otherwise, executes step S7;

[0051] S7: According to formula (7) and (10), use E c (k) and E a (k) Adjust the weights of the evaluation network and the execution network, and then return to step S3, i=i+1; until the vehicle stops running.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] The present invention takes into account the ambient temperature, operating mechanism load, and slope input disturbances during the loader's operation, and proposes a robust HDP energy management system and method for extended-range loaders. The invention uses RNN to establish a vehicle system model, and on this basis, constructs a robust HDP energy management controller. Through reinforcement learning of the evaluation network and control network in the HDP method, the optimal solution of the engine and battery pack power at the current moment is achieved to obtain the optimal control quantity under the current power demand. The energy management method designed by the present invention reduces the strategy's dependence on the analytical model, and the strategy has good real-time performance. In addition, the online update method of the vehicle system model improves the adaptability and robustness of the energy management strategy to working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 Diagram of the robust HDP energy management system for extended-range loaders.

[0056] Figure 2 It is the RNN vehicle system model.

[0057] Figure 3 Design diagram for the robust HDP energy management controller.

[0058] Figure 4 The diagram of the evaluation network structure.

[0059] Figure 5 This is the control network structure diagram. DETAILED DESCRIPTION

[0060] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0061] like Figure 1 As shown, a robust HDP energy management system for extended-range loaders includes a control layer and an execution layer. The control layer includes an on-board data collector, a demand power mapping module, a vehicle system model, an optimization objective function building module, a robust HDP energy management controller, an engine controller, and a BMS controller.

[0062] The vehicle system model is constructed using RNN, where the system state x includes the vehicle speed v and the battery pack SOC, and the control quantity u includes the power P allocated to the engine. e And the power P allocated to the battery pack b , that is, x = [v, SOC]T , u=[P e ,P b ] T ; Input disturbance is represented by ω, including ambient temperature T, workload P load , slope angle θ, that is, ω=[T,P load ,θ]; take [u(k) T ,x(k) T ,ω(k)] T As the RNN network input, the state quantity at the next moment As the RNN network output, such as Figure 2 As shown, the RNN vehicle system model is constructed:

[0063]

[0064] Where m h1 (k) is the hidden layer input of the RNN network; m h2 (k) is the output of the hidden layer of the RNN network, i represents the number of iterations, and the maximum number of iterations is set to N; k represents the current moment, k-1 represents the previous moment, and k+1 represents the next moment; Tansig and purelin functions are selected as the transfer functions of the hidden layer and output layer respectively, W m1 、W m2 、W m They represent the weight matrix from the input layer to the hidden layer, the weight matrix from the hidden layer to the output layer, and the weight matrix from the hidden layer to the hidden layer respectively; the weight training of the RNN-based vehicle system model adopts the gradient descent method until the error Reaching the set value, where The specific training process of weights is as follows:

[0065]

[0066] Where, l m represents the learning goal of the model network, Represents the weight matrix W from the hidden layer to the output layer of the i-th iteration m2 The transposed matrix of the vehicle system model network is W based on the historical operating condition information collected by the vehicle data collector. m1 、W m2 、W m Training is performed to obtain a trained vehicle system model; at the same time, in order to improve the adaptability of the vehicle system model to the working conditions and the robustness of the strategy, when the vehicle is running for a long time, When , the vehicle system model will be updated through online training, is the training window size, n is the number of online training times;

[0067] The energy consumption optimization objective function is expressed as:

[0068]

[0069] Among them, γ∈(0,1] is a discount factor, which means that the farther away from the current moment, the smaller the impact of the corresponding instantaneous energy consumption on the total energy consumption estimated at the current moment, k0 represents the initial moment, k f Indicates the end time, Q f (x(k),u(k),ω(k)) is the instantaneous energy consumption function, which can be abbreviated as Q f (k), is solved by the following formula:

[0070]

[0071] in, represents the fuel consumption rate of the engine, ρ f is the fuel density, Δt is the time step, ω M (k) is the bounding function of the known input perturbation ω(k), ||ω(k)||≤ω M (k), η is a positive constant.

[0072] The vehicle-mounted data collector obtains the vehicle's system status parameters in real time, including pedal opening p, engine power P eg , battery power P batt , vehicle speed v, battery SOC, ambient temperature T, workload P load , slope angle θ;

[0073] The demand power mapping module is based on the pedal opening p collected by the vehicle data collector, and obtains the demand power P of the drive motor according to the calibrated pedal opening p and demand power mapping curve. d ;

[0074] The vehicle system model is based on the engine power P collected at the current moment eg , battery power P batt , vehicle speed v, battery SOC, ambient temperature T, workload P load , slope angle θ as input, and vehicle speed and battery SOC at the next moment as output;

[0075] like Figure 3 As shown, the robust HDP energy management controller obtains the required power P according to the required power mapping module. d , the output of the vehicle system model and the optimization objective function constructed by the optimization objective function construction module, and use HDP to calculate the power P allocated to the engine e And the power P allocated to the battery pack b The robust HDP energy management controller includes a control network, an evaluation network 1, an evaluation network 2, and a vehicle system model network.

[0076] The control network is used to estimate the power P allocated to the engine according to the current vehicle state [v(k), SOC(k)] e (k), and based on P d (k) = P b (k)+P e (k)·η eg and the P obtained by the demand power mapping module d (k), calculate P b (k), η eg Indicates the range extender's power generation efficiency;

[0077] The vehicle system model network is used to calculate the vehicle state [v(k), SOC(k)], control quantity [P e (k),P b (k)], input disturbance [T(k),P load (k),θ(k)] generates the vehicle state at the next moment Evaluation network 1 and evaluation network 2 are used to evaluate the vehicle state at the current moment [v(k), SOC(k)] and the estimated vehicle state at the next moment Estimate and In this process, both the evaluation network and the execution network use their respective learning objectives E c (k) and E a (k) to update and adjust the network parameters and finally obtain the optimal control quantity And output to the engine controller and BMS controller for execution;

[0078] In the design process of the robust HDP energy management controller, due to the existence of input disturbances, the internal parameters of the evaluation network and the control network are changing at each moment; the internal network parameters of the evaluation network 1 and the evaluation network 2 are exactly the same and are trained and updated synchronously; the evaluation network adopts a 2-5-1 RNN structure, such as Figure 4 As shown in Figure 2, Tansig and purelin functions are used as transfer functions of the hidden layer and the output layer. Accordingly, the transfer relationship of the evaluation network is expressed as:

[0079]

[0080] Among them, W c1 、W c2 、W c are the weight matrices from input layer to hidden layer, hidden layer to output layer, and hidden layer to hidden layer respectively; c h1 and c h2 Represent the input and output of the hidden layer respectively; i represents the number of iterations; represents the energy consumption of the vehicle at the i-th iteration at time k;

[0081] The learning goal of the evaluation network is to minimize Right now

[0082]

[0083] Among them, ε c Is to set the target error; use the gradient descent method to train and update W c1 、W c2 、W c The specific update process is as follows:

[0084]

[0085] l c ∈(0,1] represents the learning goal;

[0086] Control the network to minimize For the goal, To perform an estimate of the output of the network, make ε a is the target error; for a certain required power P d (k), engine power P e and battery pack power P b They are not independent of each other, and their relationship satisfies:

[0087] P d (k) = P b (k)+P e (k)·η eg (8)

[0088] Among them, η eg represents the power generation efficiency of the range extender; therefore, the control vector u(k) is expressed as [P e (k),P d (k)-P e (k)·η eg ] T , the control network also chooses the 2-5-1 RNN structure, such as Figure 5 As shown; the transfer functions of the hidden layer and the output layer are selected as tansig and purelin functions respectively, and the corresponding transfer relationship of the control network is expressed as:

[0089]

[0090] Among them, W a1 、W a2 and W a are the weight matrices from input layer to hidden layer, hidden layer to output layer, and hidden layer to hidden layer respectively; a h1and a h2 Represent the input and output of the hidden layer respectively; use the gradient descent method to train and update W a1 、W a2 and W a The specific update process is as follows:

[0091]

[0092] l a ∈(0,1] represents the learning goal.

[0093] The engine controller is used to control the power output of the engine, and the BMS controller is used to control the power output of the battery pack;

[0094] The execution layer includes the engine, generator, battery pack, rectifier, inverter, drive assembly, and gearbox. The output of the BMS controller is connected to the battery pack, the output of the engine controller is connected to the engine, the engine and the generator are mechanically connected, and the engine and the generator form a range extender. The generator is electrically connected to the battery pack through the rectifier, the battery pack and the rectifier are electrically connected to the inverter, the inverter and the drive assembly are electrically connected, the drive assembly is mechanically connected to the gearbox, and the gearbox is mechanically connected to the travel system and the operating system; the engine is not directly connected to the drive assembly, which greatly reduces the impact and vibration of the sudden load change on the engine. The generator drives the assembly after rectification by the rectifier and inversion by the inverter, and can also charge the battery pack; the battery pack plays the role of peak shaving and valley filling. When there is high power demand, it drives the assembly together with the range extender and stores the recovered energy during braking.

[0095] The present invention also provides a robust HDP energy management method for an extended-range loader, comprising the following steps:

[0096] S1: Use the vehicle data collector to obtain the real-time pedal opening p(k) and engine power P eg , battery pack power P batt , speed v(k), battery pack SOC(k), ambient temperature T(k), workload P load (k), slope angle θ(k) working condition information; the demand power mapping module obtains the demand power P of the drive motor based on the collected pedal opening p(k) information and the calibrated pedal opening and demand power mapping curve. d (k);

[0097] S2: Judgment Is it satisfied? If so, use the training window The vehicle system model is updated online based on the working condition data information and formula (2), and n=n+1 is set; if it is not satisfied, the vehicle system model of the previous moment is continued to be used;

[0098] S3: Based on the current state vector [v(k), SOC(k)] T , the evaluation network 1 and the execution network use equations (5) and (9) to calculate and Then based on and P d (k), calculated using formula (8)

[0099] S4: Based on [v(k), SOC(k)] T 、 and [T(k),P load (k),θ(k)], and the vehicle system model (1) is used to calculate On this basis, the evaluation network 2 is used to estimate And according to Q is calculated using formula (4) f (k);

[0100] S5: According to Q f (k), and Calculate E c (k); According to Q f (k) and Calculate E a (k);

[0101] S6: Judge E c (k)≤ε c And E a (k)≤ε a Whether it is satisfied or the number of iterations i>N; if these conditions are met, the Considered as optimal output And output to the actual extended-range loader for execution, the vehicle then generates a new state, and sets time k=k+1, and returns to step S1; otherwise, executes step S7;

[0102] S7: According to formula (7) and (10), use E c (k) and E a (k) Adjust the weights of the evaluation network and the execution network, and then return to step S3, i=i+1; until the vehicle stops running.

[0103] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A robust HDP energy management system for extended-range loaders, characterized by: It includes the control layer and the execution layer; the control layer includes the vehicle data collector, the demand power mapping module, the vehicle system model, the optimization objective function construction module, the robust HDP energy management controller, the engine controller and the BMS controller; The on-board data collector obtains the vehicle's system status parameters in real time, including the pedal opening , engine power , battery power , vehicle speed , battery SOC, ambient temperature , workload , slope angle ; The demand power mapping module is based on the pedal opening collected by the vehicle data collector , according to the calibrated pedal opening Map the curve with the required power to get the required power of the drive motor ; The vehicle system model is based on the engine power collected at the current moment , battery power , vehicle speed , battery SOC, ambient temperature , workload , slope angle As input, the vehicle speed and battery SOC at the next moment are output; The robust HDP energy management controller obtains the required power according to the required power mapping module , the output of the vehicle system model and the optimization objective function constructed by the optimization objective function construction module, and the HDP is used to calculate the power allocated to the engine and power allocated to the battery pack ; The engine controller is used to control the power output of the engine, and the BMS controller is used to control the power output of the battery pack; The execution layer includes the engine, generator, battery pack, rectifier, inverter, drive assembly, and gearbox. The output of the BMS controller is connected to the battery pack, the output of the engine controller is connected to the engine, the engine and generator are mechanically connected, and the engine and generator form a range extender. The generator is electrically connected to the battery pack through the rectifier, the battery pack and rectifier are electrically connected to the inverter, the inverter is electrically connected to the drive assembly, the drive assembly is mechanically connected to the gearbox, and the gearbox is mechanically connected to the travel system and the operating system. The vehicle system model is constructed using RNN, where the system state quantity Including vehicle speed and battery pack SOC, control quantity Including the power allocated to the engine and power allocated to the battery pack ,Right now , ; Input disturbance Indicates, including ambient temperature , workload , slope angle ,Right now ;by As the RNN network input, the state quantity at the next moment As the RNN network output, the RNN vehicle system model is constructed: （1） Where, Is the hidden layer input of the RNN network; is the output of the hidden layer of the RNN network, Indicates the number of iterations, set the maximum number of iterations to N ; Indicates the current moment, Indicates the previous moment, Represents the next moment; Tansig and purelin functions are selected as the transfer functions of the hidden layer and output layer respectively. 、 、 They represent the weight matrix from the input layer to the hidden layer, the weight matrix from the hidden layer to the output layer, and the weight matrix from the hidden layer to the hidden layer respectively; the weight training of the RNN-based vehicle system model adopts the gradient descent method until the error Reaching the set value, where , the specific training process of weights is as follows: （2） Where, represents the learning goal of the model network, Indicates the i The weight matrix from the hidden layer to the output layer of the iteration The transposed matrix of the vehicle system model network is based on the historical operating condition information collected by the vehicle data collector. 、 、 Training is performed to obtain a trained vehicle system model; at the same time, in order to improve the adaptability of the vehicle system model to the working conditions and the robustness of the strategy, when the vehicle is running for a long time, When , the vehicle system model will be updated through online training, is the training window size, is the number of online training; The energy consumption optimization objective function is expressed as: （3） in, It is a discount factor, which means that the farther away from the current moment, the smaller the impact of the corresponding instantaneous energy consumption on the total energy consumption estimated at the current moment. represents the initial moment, Indicates the end time, is the instantaneous energy consumption function, which can be simplified as , solved by the following formula: （4） in, Indicates the fuel consumption rate of the engine, is the fuel density, is the time step, is the input disturbance Known bound functions, , is a positive constant.

2. The robust HDP energy management system for extended-range loaders according to claim 1, characterized in that: The robust HDP energy management controller includes a control network, an evaluation network 1, an evaluation network 2, and a vehicle system model network; The control network is used to control the vehicle status at the current moment Estimated power allocated to the engine , and based on and the demand power mapping module , calculate , Indicates the range extender's power generation efficiency; The vehicle system model network is used to calculate the vehicle status at the current moment , control quantity , input disturbance Generate the vehicle state at the next moment , evaluation network 1 and evaluation network 2 are used to evaluate the vehicle status at the current moment and the estimated vehicle state at the next moment Estimate and In this process, both the evaluation network and the execution network use their respective learning objectives and To update and adjust the network parameters, and finally obtain the optimal control quantity , and output to the engine controller and BMS controller for execution; In the design process of the robust HDP energy management controller, due to the presence of input disturbances, the internal parameters of the evaluation network and the control network are changing at every moment. The internal network parameters of evaluation network 1 and evaluation network 2 are exactly the same and are trained and updated synchronously. Tansig and purelin functions are used as transfer functions of the hidden layer and the output layer. Accordingly, the transfer relationship of the evaluation network is expressed as: （5） in, 、 、 are the weight matrices from the input layer to the hidden layer, from the hidden layer to the output layer, and from the hidden layer to the hidden layer; and Represent the input and output of the hidden layer respectively; Indicates the number of iterations; express k Moment i The energy consumption of the vehicle in the next iteration; The learning goal of the evaluation network is to minimize ,Right now （6） in, Is to set the target error; use gradient descent method to train and update 、 、 The specific update process is as follows: （7） Indicate learning objectives; Control the network to minimize As the goal, To perform an estimate of the output of the network, ;make , is the target error; for a certain required power , engine power and battery pack power They are not independent of each other, and their relationship satisfies: （8） in, represents the range extender power generation efficiency; therefore, the control vector Expressed as , the transfer functions of the hidden layer and the output layer are selected as tansig and purelin functions respectively, and the corresponding transfer relationship of the control network is expressed as: （9） in, 、 and are the weight matrices from the input layer to the hidden layer, from the hidden layer to the output layer, and from the hidden layer to the hidden layer; and Represent the input and output of the hidden layer respectively; use the gradient descent method to train and update 、 and The specific update process is as follows: （10） Indicates learning objectives.

3. An energy management method based on the robust HDP energy management system according to claim 1, characterized in that: The steps include: S1: Use the vehicle data collector to obtain real-time pedal opening , engine power , battery pack power ,speed , battery pack , ambient temperature , workload , slope angle Working condition information; the demand power mapping module is based on the collected pedal opening The required power of the drive motor is obtained by mapping the pedal opening and the required power to the calibrated pedal opening and required power curve. ; S2: Judgment Is it satisfied? If so, use the training window The working condition data information and formula (2) are used to perform online training and update the vehicle system model, and ; If not satisfied, continue to use the vehicle system model of the previous moment; S3: Based on the current state vector , the evaluation network 1 and the execution network use equations (5) and (9) to calculate and ; then based on and , calculated using formula (8) ; S4: Based on 、 and , calculated using the vehicle system model On this basis, the evaluation network 2 is used to estimate ; and according to , calculated using formula (4) ; S5: According to , and Calculated ;according to 、 and Calculated ; S6: Judgment and Whether it is satisfied or the number of iterations If these conditions are met, the Considered as optimal output , and output to the actual extended range loader for execution, the vehicle then generates a new state and makes the time , return to step S1; Otherwise, execute step S7; S7: According to formulas (7) and (10), respectively use and To adjust the weights of the evaluation network and the execution network, then return to step S3, ; until the vehicle stops.

Images