Online energy management control system, method and apparatus

Abstract

The present disclosure provides an online energy management control system, method and device, and relates to the field of engineering machinery. A heuristic dynamic programming algorithm is used to intelligently learn the optimal power distribution of the engine and the battery pack in a rolling time domain, and output an optimal output power combination sequence. Under the premise of meeting the vehicle dynamics, real-time adjustment and online rolling optimization of the vehicle energy consumption in complex working conditions are realized, and the energy management control of, for example, a range-extending loader has better energy-saving robustness.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of construction machinery, and in particular to an online energy management control system, method and device. BACKGROUND

[0002] A loader is a kind of earthwork construction machinery widely used in highway, railway, building, water and electricity, port, and mine construction projects.

[0003] The range-extending loader has significant differences in structural form, working condition type, working load and environment from road vehicles, and the energy management method of road vehicles is difficult to adapt to the energy management needs of the range-extending loader under non-road complex working conditions.

[0004] As one of the important ways for loaders to achieve green and sustainable development, it is crucial to tap the energy-saving and emission-reducing potential of the range-extending loader to improve the energy consumption performance of the whole machine. SUMMARY

[0005] In the embodiments of the present disclosure, a heuristic dynamic programming algorithm is used to intelligently learn the optimal power distribution of the engine and the battery pack in the rolling time domain, and output an optimal output power combination sequence. Under the premise of meeting the vehicle dynamics, real-time adjustment and online rolling optimization of the vehicle energy consumption under complex working conditions are realized, which has better energy-saving robustness and can be applied to, for example, energy management control of the range-extending loader.

[0006] Some embodiments of the present disclosure propose an online energy management control system, comprising:

[0007] A data collector configured to collect working condition data in real time;

[0008] A predictor configured to predict a system state vector sequence and an input disturbance vector sequence within a prediction window according to the working condition data collected in real time;

[0009] A demand power calculation module configured to calculate a vehicle demand power sequence within the prediction window;

[0010] A vehicle dynamics model configured to output a system state vector at the next time according to the system state vector, the input disturbance vector and the control parameter vector at the previous time;

[0011] An energy management controller configured to calculate a power distribution combination sequence of the engine and the battery pack within the rolling time domain by using a heuristic dynamic programming algorithm according to the vehicle demand power sequence output by the demand power calculation module, the system state vector sequence and the input disturbance vector sequence output by the predictor, the vehicle dynamics model and the constructed rolling optimization objective function.

[0012] In some embodiments, the energy management controller comprises: a control network, a first evaluation network, a second evaluation network, and an energy management control module;

[0013] the control network is configured to calculate the allocated power of the engine at the prediction time according to the system state vector at the prediction time within the prediction window;

[0014] the first evaluation network is configured to calculate the cumulative energy consumption at the prediction time according to the system state vector at the prediction time within the prediction window;

[0015] the second evaluation network is configured to calculate the cumulative energy consumption at the next time of the prediction time according to the system state vector at the next time of the prediction time output by the vehicle dynamics model;

[0016] the energy management control module is configured to calculate the allocated power of the battery pack at the prediction time according to the allocated power of the engine at the prediction time and the whole vehicle demand power at the prediction time; calculate the instantaneous energy consumption at the prediction time according to the allocated power of the engine at the prediction time; calculate the energy consumption prediction error at the prediction time and the target energy consumption at the prediction time according to the instantaneous energy consumption at the prediction time, the cumulative energy consumption at the prediction time, and the cumulative energy consumption at the next time of the prediction time; and output the combination sequence of the allocated power of the engine at the prediction time and the allocated power of the battery pack at the prediction time when the energy consumption prediction error at the prediction time and the target energy consumption at the prediction time satisfy the rolling optimization objective function, as the optimal control sequence.

[0017] In some embodiments, the control network, the first evaluation network, and the second evaluation network are constructed based on a neural network.

[0018] In some embodiments, the energy management control module is configured to:

[0019] subtract the whole vehicle demand power at the prediction time from the product of the allocated power of the engine at the prediction time and the engine efficiency to obtain the allocated power of the battery pack at the prediction time; and / or

[0020] calculate the instantaneous energy consumption at the prediction time according to the allocated power of the engine at the prediction time, in combination with the fuel consumption rate of the engine, the fuel density, and the boundary function of the input disturbance.

[0021] In some embodiments, the energy management control module is configured to:

[0022] calculate the energy consumption prediction error at the prediction time according to the difference between the cumulative energy consumption at the prediction time and the cumulative energy consumption at the next time of the prediction time after the instantaneous energy consumption at the prediction time is acted on by a discount factor;

[0023] calculate the target energy consumption at the prediction time according to the sum of the cumulative energy consumption at the next time of the prediction time after the instantaneous energy consumption at the prediction time is acted on by a discount factor.

[0024] The combination sequence of the allocated power of the engine and the allocated power of the battery pack at the prediction time when the energy consumption prediction error at the prediction time is less than or equal to the target error and the target energy consumption at the prediction time reaches the minimum is output as the optimal control sequence.

[0025] In some embodiments, the energy management control module is configured to update the weight parameters of the control network, the first evaluation network and the second evaluation network according to the gradient descent method and execute the functions of the control network, the first evaluation network, the second evaluation network and the energy management control module again if the maximum number of iterations is not reached and if the energy consumption prediction error at the prediction time and the target energy consumption do not satisfy the rolling optimization objective function.

[0026] In some embodiments, the predictor comprises a system state predictor comprising a vehicle speed prediction module and / or a state of charge prediction module, and the system state vector sequence comprises the vehicle speed and the state of charge of the battery pack at each time within the prediction window.

[0027] The vehicle speed prediction module comprises an information update layer and a vehicle speed prediction layer, the information update layer is configured to predict the pedal opening and the vehicle acceleration at the next time according to the pedal opening and the vehicle acceleration at the previous time, and the vehicle speed prediction layer is configured to predict the vehicle speed at the next time according to the pedal opening, the vehicle acceleration and the pedal opening rate of change at the previous time, and the vehicle speed at the previous time; and / or

[0028] The state of charge prediction module is configured to predict the state of charge of the battery pack at the next time according to the state of charge of the battery pack at the previous time.

[0029] In some embodiments, the predictor comprises an input disturbance predictor configured to predict the slope angle at the next time according to the slope angle at the previous time, and / or to predict the ambient temperature at the next time according to the ambient temperature at the previous time, and the input disturbance vector sequence comprises the slope angle and the ambient temperature at each time within the prediction window.

[0030] In some embodiments, the demand power calculation module is configured to calculate the vehicle demand power sequence according to the vehicle speed sequence in the system state vector sequence.

[0031] In some embodiments, the vehicle dynamics model is configured to output the vehicle speed and the state of charge of the battery pack at the next time according to the vehicle speed and the state of charge of the battery pack, the slope angle and the ambient temperature, and the allocated power of the engine and the battery pack at the previous time.

[0032] In some embodiments, the vehicle dynamics model is configured to determine whether a vehicle operation time meets a retraining time condition, and if so, update the vehicle dynamics model online according to historical driving condition data in the training window, and if not, maintain the vehicle dynamics model at the previous time.

[0033] In some embodiments, the data collector is configured to collect at least one of a pedal opening degree, an engine power, a battery pack power, a slope angle, an ambient temperature, a vehicle speed, a state of charge of the battery pack, and a vehicle acceleration in real time.

[0034] Some embodiments of the present disclosure propose an online energy management control method, comprising:

[0035] collecting driving condition data in real time;

[0036] predicting a system state vector sequence and an input disturbance vector sequence within a prediction window according to the driving condition data collected in real time;

[0037] calculating a whole vehicle demand power sequence within the prediction window;

[0038] outputting a system state vector at a next time according to a system state vector, an input disturbance vector and a control parameter vector at a previous time by using a vehicle dynamics model;

[0039] calculating a distributed power combination sequence of the engine and the battery pack in a rolling time domain by using a heuristic dynamic programming algorithm according to the whole vehicle demand power sequence, the system state vector sequence, the input disturbance vector sequence, the vehicle dynamics model and a constructed rolling optimization objective function.

[0040] In some embodiments, calculating the distributed power combination sequence of the engine and the battery pack in the rolling time domain by using the heuristic dynamic programming algorithm comprises:

[0041] calculating a distributed power of the engine at a prediction time within the prediction window according to a system state vector at the prediction time by using a control network;

[0042] calculating an accumulated energy consumption at the prediction time according to the system state vector at the prediction time by using a first evaluation network;

[0043] calculating an accumulated energy consumption at a next time of the prediction time according to a system state vector at the next time output by the vehicle dynamics model at the prediction time by using a second evaluation network;

[0044] According to the distribution power of the engine at the prediction moment and the demand power of the whole vehicle, the distribution power of the battery pack at the prediction moment is calculated; according to the distribution power of the engine at the prediction moment, the instantaneous energy consumption at the prediction moment is calculated; according to the instantaneous energy consumption at the prediction moment, the cumulative energy consumption at the prediction moment, and the cumulative energy consumption at the next moment of the prediction moment, the energy consumption prediction error and the target energy consumption at the prediction moment are calculated; when the energy consumption prediction error and the target energy consumption at the prediction moment satisfy the rolling optimization objective function, the combination sequence of the distribution power of the engine and the distribution power of the battery pack at the prediction moment is output as the optimal control sequence.

[0045] In some embodiments, the calculation of the combination sequence of the distribution power of the engine and the battery pack in the rolling time domain by using the heuristic dynamic programming algorithm further includes: if the maximum number of iterations is not reached and if the energy consumption prediction error and the target energy consumption at the prediction moment do not satisfy the rolling optimization objective function, the weight parameters of the control network, the first evaluation network, and the second evaluation network are updated according to the gradient descent method, and the process of determining the optimal control sequence is executed again.

[0046] In some embodiments, the method further includes: judging whether the running time of the vehicle satisfies the time condition for retraining, if yes, updating the vehicle dynamics model online according to the historical working condition data in the training window, and if no, keeping the vehicle dynamics model at the previous moment.

[0047] Some embodiments of the present disclosure provide an online energy management control device, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute an online energy management control method based on instructions stored in the memory.

[0048] Some embodiments of the present disclosure provide a computer readable storage medium having computer instructions stored thereon, the computer instructions being executed by a processor to implement the steps of an online energy management control method.

[0049] Some embodiments of the present disclosure provide a computer program product comprising computer instructions, the computer instructions being executed by a processor to implement the steps of an online energy management control method. BRIEF DESCRIPTION OF DRAWINGS

[0050] The drawings needed to be used in the following embodiments or related technical descriptions will be briefly introduced. According to the detailed description below with reference to the drawings, the present disclosure can be more clearly understood.

[0051] Obviously, the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor.

[0052] Figure 1A schematic diagram of an online energy management control system showing some embodiments of the present disclosure.

[0053] Figure 2 A schematic diagram of a pedal opening / vehicle acceleration prediction model showing some embodiments of the present disclosure.

[0054] Figure 3 A schematic diagram of a vehicle speed prediction model showing some embodiments of the present disclosure.

[0055] Figure 4 A schematic diagram of an RNN vehicle dynamics model showing some embodiments of the present disclosure.

[0056] Figure 5 A schematic diagram of an online energy management controller showing some embodiments of the present disclosure.

[0057] Figure 6 A schematic diagram of an RNN evaluation network showing some embodiments of the present disclosure.

[0058] Figure 7 A schematic diagram of an RNN control network showing some embodiments of the present disclosure.

[0059] Figure 8 A schematic diagram of an online energy management control method showing some embodiments of the present disclosure.

[0060] Figure 9 A schematic diagram of an online energy management control method showing some embodiments of the present disclosure.

[0061] Figure 10 A schematic diagram of an online energy management control apparatus showing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0062] It should be noted that the relative arrangement, numerical expressions, and numerical values of the components and steps set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

[0063] Those skilled in the art can understand that the terms "first", "second", etc. in the embodiments of the present disclosure are only used to distinguish different steps, devices or modules, and do not represent any specific technical meaning, nor do they represent the inevitable logical order between them.

[0064] It should also be understood that in the embodiments of the present disclosure, "a plurality of" can mean two or more, and "at least one" can mean one, two or more.

[0065] It should also be understood that for any component, data or structure mentioned in the embodiments of the present disclosure, unless specifically limited or given a contrary implication by the context or prior art, it can be understood as one or more in general.

[0066] In addition, the term "and / or" in the present disclosure is merely a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in the present disclosure generally represents an "or" relationship between the front and rear associated objects.

[0067] It should also be understood that the description of the various embodiments of the present disclosure focuses on the differences between the various embodiments, and the same or similar parts can be referred to each other, and for the sake of brevity, will not be repeated.

[0068] At the same time, it should be understood that, for the sake of description, the size of each part shown in the drawings is not drawn in accordance with the actual proportional relationship.

[0069] The following description of at least one example embodiment is merely illustrative in nature and is in no way limiting to the disclosure and its application or uses.

[0070] Techniques, methods, and devices known to those of ordinary skill in the relevant art can not be discussed in detail, but where appropriate, the described techniques, methods, and devices should be considered as part of the specification.

[0071] It should be noted that similar reference numbers and letters refer to similar items in the following drawings, and therefore, once an item is defined in one drawing, it need not be discussed further in subsequent drawings.

[0072] In addition, in order to avoid obscuring the present disclosure due to unnecessary details, only the processing steps and / or device structures closely related to the scheme according to the present disclosure are shown in the drawings, and other details not closely related to the present disclosure are omitted. It should be noted that similar reference numbers and letters in the drawings indicate similar items, and therefore, once an item is defined in one drawing, it need not be discussed further in subsequent drawings.

[0073] The working principle of the range extender (electric) loader is that in pure electric mode, the vehicle can rely on the power provided by the battery pack to drive the motor to run, meet the loading, transportation and other operation requirements, when the battery pack is insufficient, the engine on the vehicle is started to drive the generator to generate electricity, charge the battery pack or directly provide power for the motor, thereby extending the cruising range of the vehicle.

[0074] This embodiment utilizes a heuristic dynamic programming (HDP) algorithm to intelligently learn the optimal power allocation between the engine and battery pack within the rolling time domain, outputting an optimal output power combination sequence. While satisfying the vehicle's overall power performance, it achieves real-time adjustment and online rolling optimization of vehicle energy consumption under complex operating conditions, exhibiting better energy-saving robustness. This method can be applied to energy management control of, for example, range-extended (electric) loaders. In subsequent embodiments, the range-extended (electric) loader will be referred to simply as a vehicle.

[0075] Figure 1 Schematic diagrams of online energy management and control systems according to some embodiments of this disclosure are shown. Figure 1 As shown, the system in this embodiment includes a control layer and an execution layer. The control layer mainly includes a data acquisition unit (also called an on-board data acquisition unit or operating condition data acquisition unit), a system state predictor, an input disturbance predictor, a demand power calculation module, a vehicle dynamics model, a rolling optimization objective function construction module, an energy management controller, an engine controller, and a BMS (Battery Management System) controller. The data acquisition unit acquires real-time operating condition data of the vehicle, such as pedal opening p and engine power P. eg Battery (pack) power P batt The system state predictor includes vehicle speed prediction and SoC prediction modules, which together generate a system state vector (sequence) containing vehicle speed and SoC. The vehicle speed prediction module is used to predict vehicle speed and includes an information update layer and a vehicle speed prediction layer. A backpropagation neural network (BPNN) is used to build the information update layer to update the information required for vehicle speed prediction, and a recurrent neural network (RNN) is used to build the vehicle speed prediction layer. The SoC prediction module is used to predict the SoC, which can be implemented using a BPNN or similar neural network. The input disturbance predictor uses a BPNN or similar neural network to predict input disturbances such as slope angle and ambient temperature, and outputs an input disturbance vector (sequence) containing slope angle and ambient temperature. The demand power calculation module is used to calculate the total vehicle demand power P. d The system outputs the total vehicle power demand (sequence); the vehicle dynamics model is constructed using neural networks such as RNNs, where the engine power P at the previous moment is... eg Battery power P batt, vehicle speed v, battery SoC, slope angle θ, ambient temperature T as the input of the RNN network, the vehicle speed and battery SoC at the next moment as the output of the RNN network. The energy management controller calculates the vehicle demand power sequence output by the demand power calculation module, the system state vector sequence output by the system state predictor, the input disturbance vector sequence output by the input disturbance predictor, the vehicle dynamics model, and the constructed rolling optimization objective function, and calculates the distribution power combination sequence (optimal output power combination sequence) of the engine and the battery pack in the rolling time domain by using the heuristic dynamic programming algorithm. The engine controller is used to control the power output of the engine according to the distribution power of the engine in the optimal output power combination sequence output by the energy management controller. The BMS controller controls the power output of the battery pack according to the distribution power of the battery pack in the optimal output power combination sequence output by the energy management controller.

[0076] The execution layer mainly includes an engine, a generator, a battery pack, a rectifier, an inverter, a drive assembly, a gearbox, a running system, and an operation system. The engine and the generator are not directly connected to the drive assembly, which greatly reduces the impact and vibration of the load mutation on the engine. The generator drives the drive assembly through rectification by the rectifier and inversion by the inverter, and can charge the battery pack at the same time. The battery pack plays a role in peak shaving, and works with the engine to drive the drive assembly at high power demand, and stores the recovered energy during braking; the drive assembly includes a motor controller and a drive motor, which accurately outputs a driving torque according to the instruction signal of the motor controller to control the vehicle running and operation. The gearbox is a mechanical device with adjustable speed ratio and motion direction composed of a plurality of gear pairs. The running system is used to control the vehicle running. The operation system is used to control the vehicle operation.

[0077] The system state predictor includes a vehicle speed prediction module and a SoC prediction module. The vehicle speed prediction module includes an information update layer and a vehicle speed prediction layer, and a neural network such as BPNN is used to construct the information update layer to update the information required for vehicle speed prediction, and then a neural network such as RNN is used to construct the vehicle speed prediction layer to realize vehicle speed prediction. In order to realize the vehicle speed prediction in the prediction window, the predicted pedal opening Pedal opening rate Vehicle acceleration information needs to be obtained in advance. The information update layer takes a BPNN network structure of 1-5-1 (1 / 5 / 1 respectively represents the number of input layer neurons, the number of hidden layer neurons, and the number of output layer neurons) as an example, constructs the prediction models of Figure 2 and as shown in the accompanying drawings.

[0078]

[0079] ​where X and Y represent the input and output of the BPNN network, respectively; W1 and W2 represent the weight vectors of the input to the hidden layer and the hidden layer to the output layer, respectively; θ1 and θ2 represent the threshold vectors of the hidden layer and the output layer, respectively; f and h represent the activation functions of the hidden layer and the output layer, respectively, for example, the tansig function (a kind of nonlinear activation function) and the purelin function (a kind of linear activation function) are selected, respectively. If the BPNN network is used to predict and The above variables can be further represented as:

[0080]

[0081] where, when the BPNN network input X = p, the output When the BPNN network input X = a, the output p and represent the pedal opening degree of the input and the pedal opening degree of the predicted output of the BPNN network, respectively, the pedal opening degree of the input is usually the pedal opening degree at the previous moment, and the pedal opening degree of the predicted output is usually the pedal opening degree at the next moment. a and represent the vehicle acceleration of the input and the vehicle acceleration of the output of the BPNN network, respectively, the vehicle acceleration of the input is usually the vehicle acceleration at the previous moment, and the vehicle acceleration of the output is usually the vehicle acceleration at the next moment.

[0082] Then, the weight and threshold of formula (1) are trained using the collected historical pedal opening degree and acceleration data information, and based on this, and The prediction model is determined.

[0083] Pedal opening degree change rate can be directly obtained from the following formula:

[0084]

[0085] That is, the pedal opening degree at the next moment Subtract the pedal opening degree at the previous moment to obtain the pedal opening degree change rate

[0086] Next, the vehicle speed, the pedal opening degree, the pedal opening degree change rate, and the acceleration, i.e., [v(k), p(k), Δp(k), a(k)] T are taken as inputs, and the predicted vehicle speed As output, an RNN is used to construct a vehicle speed prediction model, where time k+1 is the next time step after time k. The vehicle speed prediction model predicts the vehicle speed at the next time step based on the pedal opening, vehicle acceleration, and rate of change of pedal opening at the previous time step, as well as the vehicle speed at the previous time step. Here, a three-layer RNN structure in the form of "4-5-1" (4 / 5 / 1 representing the number of neurons in the input layer, hidden layer, and output layer, respectively) is used as an example, as shown in the attached diagram. Figure 3 As shown, a vehicle speed prediction model is constructed:

[0087]

[0088] Among them, s k and These represent the input and output values ​​of the hidden layer nodes, respectively; k σ represents the input to the output layer node; g and σ represent the activation functions of the hidden layer node and the output layer node, respectively, for example, choosing the tansig and purelin functions; U 54 W 55 V 15 Let k,t represent the weight matrix from the input layer node to the hidden layer node, the weight matrix between hidden layer nodes, and the weight matrix from the hidden layer node to the output layer node, respectively. f -N represent discrete time, start time, and end time, respectively. Using historical operating condition information (vehicle speed, pedal opening, acceleration) collected by the data acquisition device and the pedal opening change rate obtained from equation (3), the gradient descent method is used to refine U in equation (4). 54 W 55 and V 15 Perform training updates until the error is correct. Once the set value is reached, a trained vehicle speed prediction model is obtained, where N represents the time window. v j Let J and Y represent the predicted and actual vehicle speeds at time j, respectively.

[0089] The SoC prediction module utilizes neural networks such as BPNN to predict the state of charge (SOC) of the battery pack at the next time step based on the SOC of the battery pack at the previous time step. For example, the SoC prediction module employs a 1-5-1 BPNN network (e.g., ...). Figure 2 (As shown) is used to construct the system, taking the battery pack SoC as input, to predict... As the output, X = SoC, Then, the weights and thresholds of equation (1) are trained using the historical battery SoC information collected by the data acquisition device to obtain the SoC prediction model.

[0090] The input perturbation predictor, constructed using a neural network, predicts the slope angle at the next time step based on the slope angle at the previous time step, and / or predicts the ambient temperature at the next time step based on the ambient temperature at the previous time step. For example, the input perturbation predictor is constructed using a BPNN network, taking the collected slope angle θ or ambient temperature T as input, and the predicted slope angle... or ambient temperature As the output, i.e., X = θ, Or X = T, Then, using the historical slope angle and ambient temperature information collected by the data acquisition device, the weights and thresholds in formula (1) are trained to obtain... and Predictive models.

[0091] According to x(k+1)=f(x(k),u(k),ω(k),k),k∈[k0,k f Therefore, the calculation logic of the power calculation module can be expressed as follows:

[0092]

[0093] Here, the function f represents the ability to determine x(k+1) based on x(k), u(k), ω(k), and k, thus obtaining equation (5). k, k0, and k f These represent discrete time points, the start time, and the end time, respectively. The system state vector x includes vehicle speed v, state of charge (SoC), and motor speed n. m , because n m It can be obtained from v, therefore x = [v, SoC] T v(k) represents v at time k, SoC(k) represents SoC at time k, and so on. Other similar notations will not be elaborated here. The control parameter vector u includes the engine power P. e and battery pack power P b That is, u = [P] e ,P b ] T ω represents the input disturbance vector, which includes the slope angle θ and the ambient temperature T, i.e., ω = [θ, T]. T P wh P d P e P b P wor P m η m η gb η g and η tThese represent wheel-end power, total vehicle power requirement, allocated engine power, allocated battery pack power, operating system power, running system power, drive system efficiency, gearbox efficiency, generator efficiency, and transmission system efficiency, respectively. f ,r wh These represent the overall speed ratio of the transmission system and the tire radius, respectively. v ,g,θ,ρ r C D ,A,ρ a ,v,n m T and V represent vehicle mass, gravitational acceleration, slope angle, rolling resistance coefficient, air drag coefficient, frontal area, air density, vehicle speed, motor speed, and ambient temperature, respectively. b ,R b and C Ah These represent the battery pack's voltage, internal resistance, and capacity, respectively. Based on the v(k+1) formula in equation (5), P can be calculated. wh Then P wh Substituting into equation (6) will allow us to calculate P. d .

[0094] The vehicle dynamics model, constructed using a neural network, outputs the vehicle speed and battery state of charge (SOC) for the next moment based on the vehicle speed, battery state of charge (SOC), gradient angle, ambient temperature, and power distribution between the engine and battery pack at the previous time step. For example, the vehicle dynamics model might be constructed using an RNN, with the control parameter vector u(k), system state vector x(k), and input disturbance vector ω(k) at time k, i.e., [u(k), x(k), ω(k)]. T As input to the RNN network, the system state vector at time k+1 As the output of an RNN network, for example, a three-layer RNN structure in the form of "6-5-2" (6 / 5 / 2 representing the number of neurons in the input layer, hidden layer, and output layer, respectively) can be used to construct an RNN vehicle dynamics model, as shown in the attached figure. Figure 4 As shown:

[0095]

[0096] In the formula, m h1 (k) is the input to the hidden layer of the RNN network; m h2 (k) is the output of the hidden layer of the RNN network. The bipolar sigmoid function and the purelin function are selected as the transfer functions of the hidden layer and the output layer, respectively. W m1 W m2 W mrespectively represent the weight matrix from input layer to hidden layer, the weight matrix from hidden layer to output layer, and the weight matrix from hidden layer to hidden layer. The RNN vehicle dynamics model adopts gradient descent method with momentum to train the weights until the model error E m reaches the set target value ε m , i.e.

[0097]

[0098] wherein, represents the true value of vehicle speed and state of charge, represents the predicted value of vehicle speed and state of charge.

[0099] The specific training process of the RNN vehicle dynamics model is as follows:

[0100]

[0101] wherein q is the number of online training, and W represents one of W m1 , W m2 , W m . l m and a respectively represent the model network learning factor and momentum factor. Based on formula (9) and the historical working condition information collected by the data collector, W m1 , W m2 , W m of the vehicle dynamics model network are trained to obtain the trained vehicle dynamics model with input disturbance processing capability. At the same time, in order to further improve the adaptability of the vehicle dynamics model to the working condition and the robustness of the strategy, when the vehicle running time , the vehicle dynamics model will be updated by online training, wherein is the training window size, and n is the number of online training.

[0102] The energy consumption optimization objective function J * (τ) can be represented as:

[0103]

[0104] wherein J(τ) represents the total energy consumption of the vehicle at time τ (i.e. the cumulative energy consumption at time τ), the optimization objective of the energy consumption is to minimize (min) the total energy consumption of the vehicle, γ ∈ (0, 1] is a discount factor, and N represents the prediction window size. JMx(τ+1)O can be simply denoted as J(τ+1). F(x(k+N)) ≥ 0 represents the terminal cost from k+N to ∞. The instantaneous energy consumption function Q f can be solved according to the following formula:

[0105]

[0106] Here, Te and n e These represent engine torque and speed, respectively. ρ represents the engine's fuel consumption rate. f This represents the fuel density, ΔT is the time step (e.g., 1 second), and η represents a positive constant, ω. M (k) represents the bounding function of the input perturbation ω(k), satisfying ||ω(k)|| ≤ ω M (k), and ω M (0) = 0. Q f (x(k),u(k),ω(k),k) is abbreviated as Q f (k).

[0107] Based on the vehicle demand power obtained from the demand power calculation module, the vehicle speed and SoC sequence predicted by the system state predictor, the slope angle and ambient temperature sequence predicted by the input disturbance predictor, the vehicle dynamics model, and the constructed rolling optimization objective function, an online energy management controller is constructed, as shown in the appendix. Figure 5 As shown.

[0108] In the design of the energy management controller, the design of the evaluation network (including the first evaluation network and the second evaluation network) and the control network are crucial. The internal network parameters of the first evaluation network (evaluation network 1) and the second evaluation network (evaluation network 2) are identical and are trained and updated synchronously. Here, the evaluation network, for example, adopts a 2-5-1 RNN structure, as shown in the attached diagram. Figure 6 As shown, the bipolar sigmoid function and the purelin function are selected as the transfer functions of the hidden and output layers, respectively. Accordingly, in the prediction time domain [k, k+N-1], letting τ ∈ [k, k+N-1], the transfer relation of the evaluation network can be expressed as:

[0109]

[0110] Among them, W c1 W c and W c2 These are the weight matrices from the input layer to the hidden layer, from the hidden layer to the hidden layer, and from the hidden layer to the output layer, respectively; c h1 and c h2 These represent the input and output of the hidden layer of the evaluation network, respectively; i represents the iteration number; Let represent the total energy consumption of the vehicle in the i-th iteration at time τ.

[0111] The learning objective of the evaluation network is to minimize the energy consumption prediction error. Right now

[0112]

[0113] Here, ε c is a set target error. The gradient descent method is used to train and update W c1 , W c and W c2 , and the specific update process is as follows:

[0114]

[0115]

[0116] wherein l c ∈(0, 1] represents a learning goal. respectively represent the weight matrices of W c1 , W c , W c2 at the i-th iteration, respectively represent the weight matrices of W c1 , W c , W c2 at the (i+1)-th iteration, respectively represent the change amounts of the weight matrices . represents the transpose of the weight matrix , x T represents the transpose of the state vector x, c h2 T (τ-1) represents the transpose of the hidden layer output c h2 (τ-1) at the (τ-1)-th time, c h2 T (τ) represents the transpose of the hidden layer output c h2 (τ) at the τ-th time.

[0117] Referring to equation (10), the control network aims to minimize the energy consumption for a determined demand power P d (τ), the allocated engine power P e and the battery pack power P b are not independent of each other, and their relationship satisfies the following equation:

[0118] P b (τ) = P d (τ) - P e (τ) · η g (15)

[0119] Therefore, the control vector u(τ) can be represented as [P e (τ), P d (τ) - P e (τ) · η g ] TTherefore, the control network here also chooses the RNN structure of 2-5-1, as shown in Fig. 2. Figure 7 Similarly, the transfer functions of the hidden layer and the output layer are chosen as the Bipolar sigmoid function and the purelin function, respectively.

[0120] In the prediction time domain τ∈[k, k+N-1], the transfer relationship of the control network can be expressed as:

[0121]

[0122] where W a1 , W a and W a2 are the weight matrices from the input layer to the hidden layer, from the hidden layer to the hidden layer, and from the hidden layer to the output layer, respectively; a h1 and a h2 represent the input and the output of the hidden layer of the control network, respectively. Here, the gradient descent method is also used to train and update W a1 , W a and W a2 , and the specific update process is as follows:

[0123]

[0124]

[0125] where l a ∈(0, 1] represents the learning goal. T represents the transpose symbol, represent the weight matrices of W a1 , Q a and W a2 in the i-th iteration, represent the weight matrices of W a1 , Q a and W a2 in the (i+1)-th iteration, represent the change amounts of the weight matrices in the i-th iteration. represents the total energy consumption of the vehicle in the i-th iteration at time τ+1, P e i (τ) represents the engine power allocated in the i-th iteration at time τ, represents the transpose of the weight matrix , W m1u = W m1 (:, 1) represents the first column of the weight matrix W m1 , represents the transpose of the weight matrix , c h2 (τ+1) represents the output of the hidden layer of the evaluation network at time τ+1, and Ah2 t a(τ-1) represents the hidden layer output a at time τ-1 h2 a(τ-1) represents the hidden layer output a at time τ-1 a(τ) represents the hidden layer output a at time τ h2 a(τ) represents the hidden layer output a at time τ

[0126] Based on the proposed online energy management control system, the embodiment further proposes an online energy management control method. As shown in the figure, the online energy management control method of the embodiment comprises the following steps. Figure 8

[0127] In step 81, real-time working condition data is collected. For example, real-time working condition data is collected by using a data collector.

[0128] In step 82, according to the real-time collected working condition data, the system state vector sequence and the input disturbance vector sequence within the prediction window are predicted. For example, the system state vector sequence within the prediction window is predicted by using a system state predictor; the input disturbance vector sequence within the prediction window is predicted by using an input disturbance predictor.

[0129] In step 83, the whole vehicle demand power sequence within the prediction window is calculated. For example, the whole vehicle demand power sequence within the prediction window is calculated by using a demand power calculation module.

[0130] In step 84, according to the system state vector, the input disturbance vector and the control parameter vector at the previous time, the system state vector at the next time is output by using the vehicle dynamics model.

[0131] Before using the vehicle dynamics model, it is judged whether the vehicle running time meets the time condition of retraining. If yes, the vehicle dynamics model is updated by online training according to the historical working condition data within the training window; if no, the vehicle dynamics model at the previous time is kept.

[0132] In step 85, according to the whole vehicle demand power sequence, the system state vector sequence and the input disturbance vector sequence, the vehicle dynamics model and the constructed rolling optimization objective function, the distribution power combination sequence of the engine and the battery pack in the rolling time domain is calculated by using a heuristic dynamic programming algorithm. For example, the distribution power combination sequence of the engine and the battery pack in the rolling time domain is calculated by using an energy management control module and a heuristic dynamic programming algorithm.

[0133] ​Using a control network, the engine's allocated power at a predicted time is calculated based on the system state vector at a predicted time within the prediction window. Using a first evaluation network, the cumulative energy consumption at a predicted time is calculated based on the system state vector at a predicted time within the prediction window. Using a second evaluation network, the cumulative energy consumption at the next predicted time is calculated based on the system state vector at the next predicted time output by the vehicle dynamics model. The battery pack's allocated power at the predicted time is calculated based on the engine's allocated power and the vehicle's required power. The instantaneous energy consumption at the predicted time is calculated based on the engine's allocated power. The energy consumption prediction error and target energy consumption at the predicted time are calculated based on the instantaneous energy consumption, cumulative energy consumption, and cumulative energy consumption at the next predicted time. The optimal control sequence is output as the combination sequence of the engine's allocated power and the battery pack's allocated power at the predicted time when the energy consumption prediction error and target energy consumption satisfy the rolling optimization objective function. For example, the optimal control sequence is output as the combination sequence of engine power allocation and battery pack power allocation at the prediction time when the energy consumption prediction error at the prediction time is less than or equal to the target error and the target energy consumption at the prediction time is minimized. If the maximum number of iterations is not reached and if the energy consumption prediction error and the target energy consumption at the prediction time do not satisfy the rolling optimization objective function, the weight parameters of the control network, the first evaluation network, and the second evaluation network are updated according to the gradient descent method, and the process of determining the optimal control sequence is executed again.

[0134] This embodiment also proposes an online energy management and control method. For example... Figure 9 As shown, the online energy management control method of this embodiment includes the following steps. Here, it is assumed that the current time is k, and τ∈[k,k+N-1]).

[0135] Step 91: Use a data acquisition device to obtain real-time pedal opening p(k) and engine power P. eg (k) Battery pack power P batt The system uses operating condition information such as slope angle θ(k), vehicle speed v(k), vehicle acceleration a(k), SoC(k), and ambient temperature T(k) to predict the vehicle speed sequence using the vehicle speed prediction module in the system state predictor. The SoC sequence is predicted using the SoC prediction module in the system state predictor. This allows us to determine the system state vector sequence within the current prediction window [k, k+N-1]. Predict using the input disturbance predictor This allows us to determine the input perturbation vector sequence within the current prediction window [k, k+N-1].

[0136] Step 92: judge whether is satisfied, where is the training window size, n is the number of online training times; if satisfied, use the working condition data information in the training window and formula (9) to perform online training and update of the vehicle dynamics model, and let n = n + 1; if not satisfied, continue to use the vehicle dynamics model at the last time.

[0137] That is, when the vehicle running time reaches , the vehicle dynamics model is trained and updated online, otherwise the vehicle dynamics model at the last time is continued to be used.

[0138] Step 93: based on the vehicle speed sequence , the vehicle demand power sequence P d (τ) is calculated by using formulas (5) and (6) d (k), P d (k+1), …, P d (k+N-1)}.

[0139] Step 94: within the prediction window τ ∈ [k, k+N-1], based on the system state vector [v(τ), SoC(τ)] T , the evaluation network 1 and the control network respectively calculate and P e i (τ) by using formulas (12) and (16); then based on P e i (τ) and P d (τ), P b i (τ) is calculated by using formula (15).

[0140] Step 95: based on [v(τ), SoC(τ)] T , [P e i (τ), P b i (τ)z T and [θ(k), T(k)] T , P is calculated by using the vehicle dynamics model, i.e. formula (7). On this basis, the evaluation network 2 estimates based on and according to P e i (τ), Q f (τ) is calculated by using formula (11).

[0141] Step 96: according to Qf (τ), and E c (τ) is calculated f (τ) + γ · wherein E c (τ) is calculated according to equation (13).

[0142] Step 97: judge whether E c (τ) ≤ ε c and the minimum is reached or the iteration number i > N. If these conditions are satisfied, output the optimal control sequence {[P e * (τ), P b * (τ)] and continue to execute step 99; otherwise, execute step 98.

[0143] Step 98: adjust the weights of the evaluation network and the control network according to equations (14) and (17) respectively using E c (τ) and then return to step 94, i = i + 1.

[0144] Step 99: if τ ≤ k + N - 1, let τ = τ + 1, and then return to step 94; otherwise, output the optimal control sequence {[P e * (k), P b * (k)], [P e * (k+1), P b * (k+1)], …, [P e * (k+N-1), P b * (k+N-1)]} in the prediction window [k, k + N - 1], and output the optimal control sequence {[P e * (k), P b * (k)] to the range extender loader for actual execution, and the vehicle generates new system state and input disturbance information, let the time k = k + 1, and return to execute step 91.

[0145] With the continuous driving of the range extender loader (vehicle), the above process continues to roll until the vehicle stops running.

[0146] Figure 10 A schematic diagram of an online energy management control device according to some embodiments of the present disclosure is shown.

[0147] As Figure 10 shown, the online energy management control device 1000 of this embodiment includes a memory 1010 and a processor 1020 coupled to the memory 1010, and the processor 1020 is configured to execute the online energy management control method (not described here again) in various embodiments based on instructions stored in the memory 1010.

[0148] The online energy management control device 1000 can also include an input / output interface 1030, a network interface 1040, a storage interface 1050, etc. These interfaces 1030, 1040, 1050 and the memory 1010 and the processor 1020 can be connected through a bus 1060, for example.

[0149] The memory 1010 can include a system memory, a fixed non-volatile storage medium, etc., for example. The system memory stores, for example, an operating system, an application program, a boot loader (Boot Loader) and other programs, etc.

[0150] The processor 1020 can be implemented in the form of a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic device, a discrete gate or transistor, etc. discrete hardware component.

[0151] The input / output interface 1030 provides a connection interface for display, mouse, keyboard, touch screen and other input / output devices. The network interface 1040 provides a connection interface for various networking devices. The storage interface 1050 provides a connection interface for SD card, U disk and other external storage devices. The bus 1060 can use any of a variety of bus structures. For example, the bus structure includes but is not limited to industry standard architecture (Industry Standard Architecture, ISA) bus, micro channel architecture (Micro Channel Architecture, MCA) bus, peripheral component interconnect (Peripheral Component Interconnect, PCI) bus.

[0152] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more (non-transitory) computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, cloud storage, etc.) containing computer program code. A computer program product should be understood as a software product that primarily implements its solution through a computer program.

[0153] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0154] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0155] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0156] The above description is only a preferred embodiment of this disclosure and is not intended to limit this disclosure. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the protection scope of this disclosure.

Claims

1. An online energy management and control system, comprising: The data acquisition unit is configured to collect operating condition data in real time; The predictor is configured to predict the system state vector sequence and the input disturbance vector sequence within the prediction window based on real-time acquired operating condition data. The demand power calculation module is configured to calculate the vehicle demand power sequence within the prediction window; The vehicle dynamics model is configured to output the system state vector at the next moment based on the system state vector, input disturbance vector, and control parameter vector at the previous moment. The energy management controller is configured to calculate the power allocation sequence of the engine and battery pack in the rolling time domain using a heuristic dynamic programming algorithm, based on the vehicle demand power sequence output by the demand power calculation module, the system state vector sequence and input disturbance vector sequence output by the predictor, the vehicle dynamics model, and the constructed rolling optimization objective function. The energy management controller includes: a control network, a first evaluation network, a second evaluation network, and an energy management control module; The control network is configured to calculate the engine's allocated power at a prediction time based on the system state vector at a prediction time within the prediction window; The first evaluation network is configured to calculate the cumulative energy consumption at a prediction time based on the system state vector at a certain prediction time within the prediction window. The second evaluation network is configured to calculate the cumulative energy consumption at the next time step based on the system state vector at the next time step of the predicted time step output by the vehicle dynamics model. The energy management control module is configured to calculate the battery pack allocation power at the predicted time by subtracting the product of the engine allocation power and engine efficiency at the predicted time from the vehicle demand power at the predicted time; calculate the instantaneous energy consumption at the predicted time based on the engine allocation power at the predicted time, combined with the engine's fuel consumption rate, fuel density, and the boundary function of the input disturbance; calculate the energy consumption prediction error and target energy consumption at the predicted time based on the instantaneous energy consumption, the cumulative energy consumption at the predicted time, and the cumulative energy consumption at the next predicted time; and output the optimal control sequence as the combination sequence of the engine allocation power and battery pack allocation power at the predicted time when the energy consumption prediction error and target energy consumption satisfy the rolling optimization objective function.

2. The online energy management and control system according to claim 1, wherein, The control network, the first evaluation network, and the second evaluation network are constructed based on neural networks.

3. The online energy management and control system according to claim 1, wherein, The energy management control module is configured as follows: The energy consumption prediction error at the prediction time is calculated by subtracting the instantaneous energy consumption at the prediction time from the cumulative energy consumption at the prediction time and the cumulative energy consumption at the next time after the discount factor is applied. The target energy consumption at the prediction time is calculated by summing the instantaneous energy consumption at the prediction time and the cumulative energy consumption at the next prediction time after the discount factor is applied. The optimal control sequence is output as the combination sequence of the engine's allocated power and the battery pack's allocated power at the prediction time when the energy consumption prediction error at the prediction time is less than or equal to the target error and the target energy consumption at the prediction time reaches its minimum.

4. The online energy management and control system according to claim 3, wherein, The energy management control module is configured as follows: If the maximum number of iterations is not reached and the energy consumption prediction error and target energy consumption at the prediction time do not meet the rolling optimization objective function, update the weight parameters of the control network, the first evaluation network, and the second evaluation network according to the gradient descent method, and execute the functions of the control network, the first evaluation network, the second evaluation network, and the energy management control module again.

5. The online energy management and control system according to any one of claims 1-4, wherein, The predictor includes a system state predictor, which includes a vehicle speed prediction module and / or a state of charge prediction module. The system state vector sequence includes the vehicle speed and the state of charge of the battery pack at each time point within the prediction window. The vehicle speed prediction module includes an information update layer and a vehicle speed prediction layer. The information update layer is constructed using a neural network and is configured to predict the pedal opening and vehicle acceleration at the next moment based on the pedal opening and vehicle acceleration at the previous moment. The vehicle speed prediction layer is constructed using a neural network and is configured to predict the vehicle speed at the next moment based on the pedal opening, vehicle acceleration, and rate of change of pedal opening at the previous moment, as well as the vehicle speed at the previous moment; and / or The state of charge prediction module, constructed using a neural network, is configured to predict the state of charge of the battery pack at the next moment based on the state of charge of the battery pack at the previous moment.

6. The online energy management and control system according to any one of claims 1-4, wherein, The predictor includes an input perturbation predictor, constructed using a neural network, configured to predict the slope angle at the next moment based on the slope angle at the previous moment, and / or to predict the ambient temperature at the next moment based on the ambient temperature at the previous moment. The input perturbation vector sequence includes the slope angle and ambient temperature at each moment within the prediction window.

7. The online energy management and control system according to any one of claims 1-4, wherein, The power demand calculation module is configured to calculate the vehicle power demand sequence based on the vehicle speed sequence in the system state vector sequence.

8. The online energy management and control system according to any one of claims 1-4, wherein, The vehicle dynamics model, constructed using a neural network, is configured to output the vehicle speed and battery pack state of charge at the next moment based on the vehicle speed and battery pack state of charge at the previous moment, the slope angle and ambient temperature, and the power distribution between the engine and battery pack.

9. The online energy management and control system according to any one of claims 1-4, wherein, The vehicle dynamics model is configured to determine whether the vehicle running time meets the time condition for retraining. If it does, the vehicle dynamics model is trained and updated online based on historical operating condition data within the training window. If it does not meet the condition, the vehicle dynamics model from the previous moment is retained.

10. The online energy management and control system according to any one of claims 1-4, wherein, The data acquisition device is configured to collect at least one of the following in real time: pedal opening, engine power, battery pack power, slope angle, ambient temperature, vehicle speed, battery pack state of charge, and vehicle acceleration.

11. An online energy management and control method, comprising: Real-time acquisition of operating condition data; Based on real-time collected operating data, predict the system state vector sequence and input disturbance vector sequence within the prediction window; Calculate the vehicle power demand sequence within the prediction window; Using the vehicle dynamics model, the system state vector for the next moment is output based on the system state vector, input disturbance vector, and control parameter vector of the previous moment. Based on the vehicle's power demand sequence, system state vector sequence, input disturbance vector sequence, vehicle dynamics model, and the constructed rolling optimization objective function, a heuristic dynamic programming algorithm is used to calculate the power allocation combination sequence of the engine and battery pack in the rolling time domain. Among them, the power allocation combination sequence of the engine and battery pack calculated using a heuristic dynamic programming algorithm in the rolling time domain includes: Using the control network, the engine's allocated power at a prediction time is calculated based on the system state vector at a prediction time within the prediction window. Using the first evaluation network, the cumulative energy consumption at a prediction time is calculated based on the system state vector at a certain prediction time within the prediction window; Using the second evaluation network, the cumulative energy consumption at the next time step is calculated based on the system state vector at the next time step of the predicted time step output by the vehicle dynamics model. Based on the engine's allocated power and the vehicle's required power at the predicted time, the product of the engine's allocated power and engine efficiency at the predicted time is subtracted from the vehicle's required power at the predicted time to calculate the battery pack's allocated power. Based on the engine's allocated power at the predicted time, combined with the engine's fuel consumption rate, fuel density, and the boundary function of the input disturbance, the instantaneous energy consumption at the predicted time is calculated. Based on the instantaneous energy consumption at the predicted time, the cumulative energy consumption at the predicted time, and the cumulative energy consumption at the next predicted time, the energy consumption prediction error and the target energy consumption at the predicted time are calculated. The optimal control sequence is output as the combination sequence of the engine's allocated power and the battery pack's allocated power at the predicted time when the energy consumption prediction error and the target energy consumption satisfy the rolling optimization objective function.

12. The online energy management and control method according to claim 11, wherein, The heuristic dynamic programming algorithm used in the rolling time domain to calculate the power allocation combination sequence of the engine and battery pack also includes: If the maximum number of iterations is not reached and if the energy consumption prediction error and target energy consumption at the prediction time do not meet the rolling optimization objective function, update the weight parameters of the control network, the first evaluation network, and the second evaluation network according to the gradient descent method, and execute the process of determining the optimal control sequence again.

13. The online energy management and control method according to claim 11, further comprising: Determine whether the vehicle's running time meets the time condition for retraining. If it does, train and update the vehicle dynamics model online based on historical operating condition data within the training window. If it does not meet the condition, maintain the vehicle dynamics model from the previous moment.

14. An online energy management and control device, comprising: Memory; And a processor coupled to the memory, the processor being configured to execute the online energy management control method of any one of claims 11-13 based on instructions stored in the memory.

15. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the steps of the online energy management control method according to any one of claims 11-13.

16. A computer program product comprising computer instructions that, when executed by a processor, implement the steps of the online energy management control method according to any one of claims 11-13.

Images