Energy-saving optimization method for predicting transverse and longitudinal working conditions of hybrid electric vehicle

Abstract

The invention relates to an energy-saving optimization method for predicting transverse and longitudinal working conditions of a hybrid electric vehicle, and aims to efficiently utilize multi-source traffic information to realize energy-saving optimization. According to the method, a longitudinal vehicle speed prediction model and a transverse vehicle speed prediction model are trained, transverse and longitudinal working conditions are predicted, vehicle motion energy consumption, driving motor and steering motor energy consumption system regulation and control are considered, and an energy management strategy can respond to road environment changes in advance.

Description

Technical Field

[0001] The present disclosure relates to the field of energy-saving control of hybrid electric vehicles, and particularly to a method for predicting and optimizing energy-saving in the longitudinal and lateral working conditions of a hybrid electric vehicle. Background Art

[0002] The transportation field occupies an important position in terms of energy consumption and emissions. Hybrid electric vehicles (HEVs), with their low emissions or even zero emissions, utilization of renewable energy, and extensive application scenarios, have become an important direction for promoting the development of green transportation. However, their insufficient economy remains the main bottleneck for popularization. With the rapid development of connected technology and big data, the transportation system has gradually achieved interconnection and interoperability, and vehicles can receive and process more traffic information.

[0003] In the research on energy-saving technologies for hybrid electric vehicles, existing strategies usually focus on longitudinal drive energy consumption management and generally ignore the energy consumption problem of the lateral steering system. Especially in complex traffic scenarios, frequent lane changes of vehicles can lead to continuous accumulation of the power of the steering motor. Although its energy consumption is lower than that of the longitudinal drive system, long-term implementation of this strategy will result in a superimposed effect, thus significantly affecting the economy of the whole vehicle. Summary of the Invention

[0004] The purpose of the present disclosure is to propose an energy-saving strategy for hybrid electric vehicles considering longitudinal and lateral working condition prediction. By integrating actual driving data, a high-precision longitudinal vehicle speed prediction model and a lateral lane prediction model are constructed to focus on solving the problem of predicting the correlation between vehicle dynamic behavior and energy consumption. By using connected technology to obtain real-time traffic information and combining vehicle kinematic characteristics, the longitudinal vehicle speed change and lateral lane change trend in the future time domain are synchronously predicted, and based on the prediction results, the energy distribution between the drive motor and the steering motor is dynamically coordinated.

[0005] To achieve the above technical purpose, the present disclosure proposes a method for predicting and optimizing energy-saving in the longitudinal and lateral working conditions of a hybrid electric vehicle. The steps include: obtaining the longitudinal acceleration and speed, lateral and longitudinal positions of the target vehicle and surrounding vehicles as node features X ea_LSPM , obtaining the relative longitudinal acceleration and speed between each surrounding vehicle and the target vehicle, and the relative lateral and longitudinal positions as edge features E ea_LSPM , obtaining the current phase, remaining time of the traffic signal, and the lateral and longitudinal distances from the target vehicle as traffic signal node features X tl ; obtaining the lane where the target vehicle is located and its lateral speed, the lateral and longitudinal positions of the target vehicle, the lanes where the surrounding vehicles are currently located and their lateral speeds, and the lateral and longitudinal positions of each surrounding vehicle as node features X ea_LCPM , obtaining the relative lanes between each surrounding vehicle and the target vehicle and the relative lateral and longitudinal positions as edge features E ea_LCPM ; taking X ea_LSPM , E ea_LSPM and Xtl As the input of the trained longitudinal driving condition prediction model, the lateral speed is output, and X ea_LCPM , E ea_LCPM and X tl are used as the inputs of the trained lateral driving condition prediction model, and the longitudinal speed is output. The lateral driving condition prediction model and the longitudinal driving condition prediction model are constructed based on EGCN, LSTM, and MLP; based on the lateral speed and the longitudinal speed, by solving the total energy consumption objective function of the powertrain vehicle constructed based on model predictive control, the fuel cell power change rate is used as the control variable to allocate the required power of the hybrid vehicle for a period of time in the future; repeating the above process, the energy management during the driving of the hybrid vehicle is completed.

[0006] In the above technical solution, the input data of the longitudinal driving condition prediction model and the lateral driving condition prediction model are normalized. The normalization process includes using the z-score standardization method to convert the values into a distribution with a mean of 0 and a standard deviation of 1.

[0007] In the above technical solution, the specific structures of the lateral driving condition prediction model and the longitudinal driving condition prediction model are: three layers of EGCN, one layer of LSTM, and multiple layers of MLP are stacked.

[0008] In the above technical solution, the total energy consumption objective function is the sum of the costs brought by the hydrogen consumption of the fuel cell and the equivalent hydrogen consumption of the battery and the costs brought by the aging and degradation of the fuel cell and the battery, and its solution is to minimize the total energy consumption objective function.

[0009] To achieve the above technical purpose, the present disclosure proposes a computer-readable storage medium storing a computer program that can be loaded and executed by a processor to perform any of the above methods.

[0010] To achieve the above technical purpose, the present disclosure proposes a hybrid vehicle energy management system. The system includes a first acquisition module, a second acquisition module, and a prediction and allocation module; wherein: the first acquisition module is configured to acquire the longitudinal acceleration and speed, lateral and longitudinal positions of the target vehicle and surrounding vehicles as the node feature X ea_LSPM , acquire the relative longitudinal acceleration and speed between each surrounding vehicle and the target vehicle, and the relative lateral and longitudinal positions as the edge feature E ea_LSPM , and acquire the current phase, remaining time of the traffic signal, and the lateral and longitudinal distances from the target vehicle as the traffic signal node feature X tl ; the second acquisition module is configured to acquire the lane where the target vehicle is located and the lateral speed, the lateral and longitudinal positions of the target vehicle, the lanes where the surrounding vehicles are currently located and the lateral speeds, and the lateral and longitudinal positions of each surrounding vehicle as the node feature X ea_LCPM, obtain the relative lanes and relative longitudinal and lateral positions of each surrounding vehicle and the target vehicle as the edge feature E ea_LCPM ; the prediction and allocation module is configured to use X ea_LSPM , E ea_LSPM and X tl as the inputs of the trained longitudinal driving condition prediction model, output the lateral speed, and use X ea_LCPM , E ea_LCPM and X tl as the inputs of the trained lateral driving condition prediction model, output the longitudinal speed. The lateral driving condition prediction model and the longitudinal driving condition prediction model are constructed based on EGCN, LSTM, and MLP; the prediction module is configured to, based on the lateral speed and the longitudinal speed, by solving the total energy consumption objective function of the power vehicle constructed based on model predictive control, use the fuel cell power change rate as the control variable to allocate the required power of the hybrid vehicle for a future period of time; the system realizes the energy management of the hybrid vehicle by repeatedly executing the above modules during the driving process of the hybrid vehicle.

[0011] Technical effects of the present disclosure: Different from the prior art's reliance on driving condition optimization, this solution focuses on the construction of prediction models and data-driven decision-making mechanisms. By predicting the longitudinal and lateral motion requirements of the vehicle, it realizes the coordinated control of key energy consumption components, and systematically reduces the comprehensive energy consumption on the premise of ensuring driving safety. Description of the Drawings

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0013] Figure 1 , schematic diagram of the technical route process.

[0014] Figure 2 , EGCN network model structure diagram. Detailed Embodiments

[0015] The English term explanations in this case are as follows:

[0016] EGCN: A new deep learning framework for graph-structured data. Its core theory lies in explicitly modeling the dynamics and complexity of the relationships between nodes, and is particularly suitable for modeling spatio-temporal data with multi-element interactions in traffic scenarios. Different from the traditional graph convolutional network (GCN) that only relies on the adjacency matrix, EGCN uses edge attributes as high-dimensional feature vectors and dynamically generates edge weights through a learnable edge encoder. The state transition process of EGCN can be expressed as:

[0017]

[0018] Among them, Xl-1 represents the node feature at the previous moment, represents the "edge feature" corresponding to the k-th node feature, W l is the weight matrix, B 1 represents the bias term.

[0019] LSTM: Long Short-Term Memory, is an improved recurrent neural network (RNN), specifically designed to address the problem of vanishing or exploding gradients that traditional RNNs encounter when processing long sequence data. Its core feature is to achieve long-term storage and selective transmission of information through memory cells (cell state) and gating mechanisms (forget gate, input gate, output gate).

[0020] MLP: Multilayer Perceptron, a basic form of artificial neural network (ANN), belonging to a type of feedforward neural network. An MLP consists of multiple neurons (or nodes), which are arranged in a hierarchical structure, including an input layer, hidden layers, and an output layer. Neurons between layers are connected by weights, and information propagates forward from the input layer to the output layer in sequence, without feedback connections.

[0021] MPC: Model Predictive Control, an advanced control strategy. Its core idea is to online solve a finite-horizon open-loop optimization problem at each sampling moment based on the current measurement information, and apply the first element of the obtained control sequence to the controlled object. At the next sampling moment, repeat the above process, refresh the optimization problem with the new measurement value and solve it again.

[0022] LSPM: Lateral vehicle speed prediction model.

[0023] LCPM: Longitudinal vehicle speed prediction model.

[0024] See Figure 1 For illustration, this solution involves several parts including vehicle state and multi-source traffic information processing, longitudinal vehicle speed prediction model, lateral vehicle speed prediction model, and power distribution framework based on MPC. Specifically as follows.

[0025] (I) Lateral and longitudinal vehicle speed prediction models

[0026] Based on EGCN, establish lateral and longitudinal vehicle speed prediction models, taking traffic elements in local traffic flow as nodes, and constructing a local traffic space topology graph. See Figure 2. Input the state characteristics of the target vehicle and surrounding vehicles in the input time series and the state of the intersection signal lights, and output the future vehicle speed of the target vehicle.

[0027] The horizontal and vertical working condition prediction model framework consists of three layers of EGCN, one layer of LSTM and multiple layers of MLP, which realizes the spatio-temporal feature extraction and fusion of multi-source traffic information and vehicle state data to predict the future vehicle speed. The model input is a spatial topology graph centered on the target vehicle, including neighbor vehicles and the front signal light node. The features of each node include vehicle state and multi-source traffic information. The three layers of EGCN are responsible for extracting spatial features, modeling the dynamic interaction relationship between the target vehicle and neighbor vehicles and signal light nodes, inputting graph structure data at each time step and outputting the high-dimensional spatial feature representation of the target vehicle. The first layer of EGCN calculates the edge weight e through the edge encoder ij , aggregates the information of neighbor nodes, and generates the preliminary spatial features of the target vehicle

[0028]

[0029] The second layer of EGCN further extracts high-order spatial features to capture more complex interaction relationships:

[0030]

[0031] The third layer of EGCN outputs the final spatial feature representation of the target vehicle for input to the LSTM layer:

[0032]

[0033] LSTM is responsible for extracting time features, inputting the time series features output by the EGCN layer into LSTM to capture the long-term dependence relationship of vehicle speed changes:

[0034]

[0035] MLP is responsible for dimensionality reduction of features and outputting the final prediction value. It inputs the time features output by LSTM into multiple layers of MLP, gradually reduces the dimension and maps it to the vehicle speed prediction value:

[0036]

[0037] (2) Vehicle state and multi-source traffic information processing

[0038] Real-time parameters such as vehicle drive and steering motor torque, speed, power source status, acceleration, and heading angle collected by on-vehicle sensors, as well as vehicle position information in traffic flow obtained by GPS, combined with multi-dimensional data such as vehicle distance, signal light phase, and road topology, can accurately characterize the coupled energy consumption characteristics of longitudinal drive and lateral steering in scenarios such as rapid acceleration and frequent lane changes. Using the vehicle operation dataset collected through experiments, a database is constructed to calibrate the target vehicle status, surrounding vehicle status, and intersection signal light status. Public vehicle operation datasets can also be directly used. The actual vehicle operation data and traffic flow information can truly reflect the dynamic characteristics and complex interaction relationships in the traffic environment, and can reduce the idealization deviation of simulation modeling.

[0039] First, normalize the collected vehicle status and traffic information data to eliminate the dimensional differences between features, making the numerical ranges of different features tend to be consistent, thereby accelerating model convergence and improving prediction accuracy; and use the z-score standardization method to convert the numerical values into a distribution with a mean of 0 and a standard deviation of 1. Then, perform structured processing respectively to construct a traffic topology map dataset with spatio-temporal correlation characteristics.

[0040] Specifically, considering the characteristics of EGCN, the input end of the network requires a dataset in graph structure. Therefore, construct a graph structure data with the target vehicle as the central node of the graph structure, other surrounding vehicles as neighbor nodes, and at the same time convert the real-time phase status and remaining duration of the traffic lights at the front intersection into discrete time series codes, and establish directional connection edges with the central vehicle as independent signal light nodes to represent the impact of signal constraints on longitudinal driving behavior. In the time dimension, continuously intercept traffic scene snapshots at a fixed sampling interval (such as 1 s) to form a time-series graph structure data stream, and each time-step graph node retains the absolute timestamp and relative time-series offset.

[0041] The above spatio-temporal fusion modeling method enables the dataset to simultaneously have the ability to finely depict the local spatial topology and the continuity of global temporal evolution, reflecting the longitudinal and lateral working condition change trends of vehicles through the distribution of neighbor vehicle nodes, and the associated impact of the signal light node cycle change on the traffic flow passing mode.

[0042] Separate the structured vehicle operation status information and multi-source traffic information data of the traffic flow it is in to establish datasets for training longitudinal vehicle speed prediction and training lateral lane change behavior:

[0043]

[0044] Among them, H LSPM and H LCPM respectively represent the expected input node feature matrices of LSPM and LCPM, and Represent the output of LSPM and LCPM respectively: future longitudinal and lateral speeds.

[0045] The node feature matrix of the target vehicle and surrounding vehicles of LSPM can be expressed as:

[0046]

[0047] Among them, X ego represents the node feature of the target vehicle, and X aroi (i = 1, 2,..., n) represents the node feature of the i-th surrounding vehicle, and represent the longitudinal acceleration and speed, lateral and longitudinal positions of the target vehicle respectively; and represent the longitudinal acceleration and speed, lateral and longitudinal positions of the i-th surrounding vehicle respectively.

[0048] The relative features of the target vehicle and surrounding vehicle nodes characterize the interaction between nodes, and based on this, the edge feature matrix of the target vehicle and surrounding vehicles is constructed:

[0049]

[0050] Among them, the relative feature of the target vehicle itself is 1, fully retaining the target vehicle feature. and represent the relative longitudinal acceleration and speed of the i-th surrounding vehicle and the target vehicle, as well as the relative lateral and longitudinal positions respectively.

[0051] The signal light node feature can be expressed as:

[0052]

[0053] Among them, Ph tl , t tl , represent the current phase, remaining time of the signal light, and the lateral and longitudinal distances from the target vehicle respectively.

[0054] The expected input node feature matrix of LSPM can be expressed as:

[0055] H LSPM = [X ea_LSPM , E ea_LSPM , X tl

[0056] The output of LSPM is the future vehicle speed of the target vehicle The LSPM dataset can be expressed as:

[0057]

[0058] To characterize the spatial characteristics of lateral vehicle lane changes, the input data needs to include target vehicle characteristics, surrounding vehicle characteristics, and signal light characteristics. Among them, the node feature matrices of the target vehicle and surrounding vehicles in the LCPM can be expressed as:

[0059]

[0060] Among them, and represent the lateral speed of the target vehicle in the current lane, and represent the current lane and lateral speed of the i-th surrounding vehicle.

[0061] Edge feature matrix of the target vehicle and surrounding vehicles:

[0062]

[0063] Among them, represents the relative lane of the i-th surrounding vehicle to the target vehicle. The upper limit of the lane data where the vehicle is located depends on the number of lanes on the current road. For example, if there are 4 lanes on the current road, then from the rightmost lane to the leftmost lane are 1, 2, 3, 4 in sequence. The positive and negative values of the relative lane represent which lane to the left or right of the target vehicle the i-th surrounding vehicle is located in.

[0064] The desired input node feature matrix of the LCPM can be expressed as:

[0065] H LCPM =[X ea_LCPM , E ea_LCPM , X tl

[0066] The output of the LCPM is the future lateral speed of the target vehicle The LCPM dataset can be expressed as:

[0067]

[0068] (III) Power Allocation Framework Based on MPC

[0069] Taking into account the economic impact brought by the energy consumption of the longitudinal drive motor and the lateral steering motor, a multi-objective optimization objective function for the energy consumption model of the longitudinal and lateral power systems is constructed.

[0070] The longitudinal drive demand power is calculated according to the longitudinal dynamics formula:

[0071]

[0072] Among them, F t ​Denote the driving force of the vehicle as f d Denote the rolling resistance coefficient as ρ, the air density as ρ, and C d Denote the air resistance coefficient as C, the frontal area as A, and v V as the predicted longitudinal vehicle speed, θ as the road slope angle, and m as the vehicle mass.

[0073] The required power of the steering motor can be expressed as:

[0074]

[0075] where k s is determined by the steering geometry (kingpin offset, suspension lever arm, rack mechanism, etc.), v H is the predicted lateral vehicle speed, L f is the distance from the center of mass to the front axle, ω is the yaw rate, Cαf is the front wheel cornering stiffness, and Lf is the distance from the front axle to the center of mass.

[0076] The total required power of the powertrain is satisfied by the fuel cell output power P fc and the battery system output power P bat and is expressed as:

[0077]

[0078] where η fc , η bat , η dcdc , η D and η S are the efficiencies of the fuel cell system, battery system, DCDC converter, drive motor, and steering motor, respectively.

[0079] Establish a multi-objective optimization function considering energy consumption and powertrain degradation:

[0080] J goat = min ∑(C fc + C bat + c cost_f ; c + C costbat )

[0081] The longitudinal and lateral required power of the vehicle is mainly satisfied by the drive motor and the steering motor. The power of the drive motor can be expressed as:

[0082] P D = T D n D

[0083] where T D is the output torque of the drive motor, and n D is the rotational speed of the drive motor.

[0084] The longitudinal driving demand power of the vehicle is expressed as:

[0085]

[0086]

[0087] Wherein, F t represents the driving force of the vehicle, f d represents the rolling resistance coefficient, ρ represents the air density, C d represents the air resistance coefficient, A represents the frontal area, v V is the longitudinal vehicle speed, θ represents the road gradient angle, and m is the vehicle mass.

[0088] The output power of the steering motor is expressed as:

[0089] P S = T S n S + P nom

[0090] Wherein, n S is the steering motor speed, and P nom represents the stable output torque of the steering motor when the driver does not steer, which is used to ensure the stability, comfort and assist at any time.

[0091] The demand power of the vehicle steering motor:

[0092]

[0093] Therefore, the total demand power of the power system can be expressed as:

[0094]

[0095] The power actually output by the power system to overcome the longitudinal and lateral driving resistances is:

[0096] P m = P D + P S

[0097] The total demand power P req puts forward a demand for the hybrid power (energy storage), which is satisfied by the output power P fc of the fuel cell and the output power P bat of the battery system, and is expressed as:

[0098] P req = (P fc η fc + P bat η bat ) η dcdc η D ηS

[0099] Among them, η fc 、η bat 、η dcdc 、η D and η S are the efficiencies of the fuel cell system, battery system, DCDC converter, drive motor, and steering motor, respectively.

[0100] Establish a target optimization function. Considering the total energy consumption of the power system comprehensively, the target function can be expressed as:

[0101] J goat = min ∑(C fc + C bat + C cost_fc + C cost_bat )

[0102] Among them, C fc and C bat represent the costs brought by the hydrogen consumption of the fuel cell and the equivalent hydrogen consumption of the battery, and C cost_fc and C cost_bat represent the costs brought by the aging degradation of the fuel cell and the battery.

[0103] The state and control variables of the power distribution system can be expressed as:

[0104] S = [v V , v H , ω, SOC, P fc

[0105]

[0106] Among them, the control variable is the power change rate of the fuel cell. The state transition process can be expressed as:

[0107]

[0108] During the power distribution process, physical quantity boundaries need to be set to avoid generating physical quantities that do not meet the vehicle state constraints. The control constraints of the power distribution can be expressed as:

[0109] T D_min ≤ T D ≤ T D_max

[0110] n D_min ≤ n D ≤ n D_max

[0111] T S_min ≤ T S ≤ T S_max ​

[0112] n S_min n ≤ S ≤ n S_max

[0113] 0 ≤ P fc_min ≤ P fc ≤ P fc_max

[0114] P bat_min ≤ P bat ≤ P bat_max

[0115] 0 ≤ SOC min ≤ SOC ≤ SOC max ≤ 1

[0116] During the vehicle driving process, by using the trained LSPM and LCPM, the longitudinal and lateral vehicle speeds for a future period are predicted according to the current vehicle state and multi-source traffic information, and the data is transmitted to the vehicle dynamics model in the MPC power distribution framework to allocate the future required power. Combining MPC rolling optimization and feedback correction, and based on the dynamic programming solver, the left and right power allocation sequences for a future period are obtained to provide an optimized energy management decision for the vehicle.

[0117] Through the description of the above embodiments, those skilled in the art can clearly understand that a corresponding system can be implemented according to the method of the present disclosure. Exemplarily, a hybrid vehicle energy management system, the system includes a first acquisition module, a second acquisition module, and a prediction and allocation module; wherein: the first acquisition module is configured to acquire the longitudinal acceleration and speed, lateral and longitudinal positions of the target vehicle and surrounding vehicles as node features X ea_LSPM , acquire the relative longitudinal acceleration and speed of each surrounding vehicle and the target vehicle, and the relative lateral and longitudinal positions as edge features E ea_LSPM , acquire the current phase, remaining time of the traffic signal, and the lateral and longitudinal distances from the target vehicle as traffic signal node features X tl ; the second acquisition module is configured to acquire the lane where the target vehicle is located and the lateral speed, the lateral and longitudinal positions of the target vehicle, the lanes where the surrounding vehicles are currently located and the lateral speeds, and the lateral and longitudinal positions of each surrounding vehicle as node features X ea_LCPM , acquire the relative lanes of each surrounding vehicle and the target vehicle and the relative lateral and longitudinal positions as edge features E ea_LCPM ; the prediction and allocation module is configured to use X ea_LSPM , E ea_LSPM and X tl as the input of the trained longitudinal driving condition prediction model, output the lateral speed, and use X ea_LCPM , E ea_LCPM and X tlAs the input of the trained lateral driving condition prediction model, the longitudinal speed is output. The lateral driving condition prediction model and the longitudinal driving condition prediction model are constructed based on EGCN, LSTM, and MLP. A prediction module is configured to, based on the lateral speed and the longitudinal speed, by solving the total energy consumption objective function of a power vehicle constructed based on model predictive control, use the fuel cell power change rate as a control variable to allocate the required power of the hybrid vehicle for a future period of time. The system realizes the energy management of the hybrid vehicle by repeatedly executing the above modules during the driving process of the hybrid vehicle.

[0118] Through the description of the above embodiments, those skilled in the art can clearly understand that the method of the present disclosure can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by dedicated hardware including application-specific integrated circuits, dedicated CPUs, dedicated memories, dedicated components, etc. Generally, functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structures for implementing the same function can also be various, such as analog circuits, digital circuits, or dedicated circuits. However, in more cases for the present disclosure, software program implementation is a better implementation manner.

[0119] Although the embodiments of the present disclosure have been described above in conjunction with the accompanying drawings, the present disclosure is not limited to the above specific embodiments and application fields. The above specific embodiments are merely illustrative and guiding, rather than restrictive. Under the inspiration of this specification and without departing from the scope protected by the claims of the present disclosure, those of ordinary skill in the art can also make many forms, and all of these fall within the scope of protection of the present disclosure.

Claims

1. A transverse and longitudinal working condition prediction energy-saving optimization method for a hybrid vehicle, characterized in that Including: Obtain the longitudinal acceleration, speed, lateral and longitudinal positions of the target vehicle and surrounding vehicles as the node feature X ea_LSPM , obtain the relative longitudinal acceleration and speed of each surrounding vehicle with respect to the target vehicle, as well as the relative lateral and longitudinal positions as the edge feature E ea_LSPM , obtain the current phase, remaining time of the traffic signal, as well as the lateral and longitudinal distances from the target vehicle as the traffic signal node feature X tl ; Obtain the lane where the target vehicle is located, the lateral speed of the target vehicle, the lateral and longitudinal positions of the target vehicle, the lanes where the surrounding vehicles are currently located and their lateral speeds, and the lateral and longitudinal positions of each surrounding vehicle as the node feature X ea_LCPM , and obtain the relative lanes of each surrounding vehicle and the target vehicle, as well as the relative lateral and longitudinal positions as the edge feature Ee a_LCPM ; Take X ea_LSPM , E ea_LSPM and X tl as the inputs of the trained longitudinal driving condition prediction model, and output the lateral speed. Take X ea_LCPM , E ea_LCPM and X tl as the inputs of the trained lateral driving condition prediction model, and output the longitudinal speed. The lateral driving condition prediction model and the longitudinal driving condition prediction model are constructed based on EGCN, LSTM, and MLP; Based on the lateral speed and longitudinal speed, by solving the total energy consumption objective function of a power vehicle constructed based on model predictive control, taking the fuel cell power change rate as the control variable, and distributing the required power of the hybrid vehicle over a future period of time; Repeat the above process to complete the energy management during the driving of the hybrid vehicle.

2. The method according to claim 1, characterized in that, The input data of the longitudinal driving condition prediction model and the lateral driving condition prediction model are normalized. The normalization process includes using the Z-score standardization method to convert the numerical values into a distribution with a mean of 0 and a standard deviation of 1.

3. The method according to claim 1, wherein The specific structures of the lateral driving condition prediction model and the longitudinal driving condition prediction model are: stacked three-layer EGCN, one-layer LSTM, and multiple-layer MLP.

4. The method according to claim 1, characterized in that, The total energy consumption objective function is the sum of the costs brought by the hydrogen consumption of the fuel cell and the equivalent hydrogen consumption of the battery and the costs brought by the aging and degradation of the fuel cell and the battery. Its solution is to minimize the total energy consumption objective function.

5. A computer-readable storage medium, characterized in that: A computer program capable of being loaded and executed by a processor, such as any one of the methods in claims 1 to 4, is stored.

6. A hybrid vehicle energy management system, characterized in that, The system includes a first acquisition module, a second acquisition module, and a prediction and distribution module; The first acquisition module is configured to acquire the longitudinal acceleration, speed, lateral and longitudinal positions of the target vehicle and surrounding vehicles as node feature X ea_LSPM , acquire the relative longitudinal acceleration and speed of each surrounding vehicle with respect to the target vehicle, and the relative lateral and longitudinal positions as edge feature E ea_LSPM , acquire the current phase, remaining time of the traffic signal, and the lateral and longitudinal distances from the target vehicle as traffic signal node feature X tl ; The second acquisition module is configured to acquire the lane where the target vehicle is located and the lateral speed, the lateral and longitudinal positions of the target vehicle, the lanes where the surrounding vehicles are currently located and their lateral speeds, and the lateral and longitudinal positions of each surrounding vehicle as the node feature X ea_LCPM , and acquire the relative lanes between each surrounding vehicle and the target vehicle and the relative lateral and longitudinal positions as the edge feature E ea_LCPM ; The prediction and allocation module is configured to use X ea_LSPM , E ea_LSPM and X tl as the inputs of the trained longitudinal driving condition prediction model, and output the lateral speed. Use X ea_LCPM , E ea_LCPM and X tl as the inputs of the trained lateral driving condition prediction model, and output the longitudinal speed. The lateral driving condition prediction model and the longitudinal driving condition prediction model are constructed based on EGCN, LSTM, and MLP; A prediction module configured to, based on the lateral speed and longitudinal speed, by solving the total energy consumption objective function of a power vehicle constructed based on model predictive control, take the fuel cell power change rate as the control variable, and distribute the required power of the hybrid vehicle over a future period of time; The system realizes the energy management of the hybrid vehicle by repeatedly executing the above modules during the driving of the hybrid vehicle.

7. The system according to claim 6, wherein: The input data of the longitudinal driving condition prediction model and the lateral driving condition prediction model are normalized. The normalization process includes using the Z-score standardization method to convert the numerical values into a distribution with a mean of 0 and a standard deviation of 1.

8. The system according to claim 6, wherein: The specific structures of the lateral driving condition prediction model and the longitudinal driving condition prediction model are: stacked three-layer EGCN, one-layer LSTM, and multiple-layer MLP.

9. The system according to claim 6, wherein: The total energy consumption objective function is the sum of the costs brought by the hydrogen consumption of the fuel cell and the equivalent hydrogen consumption of the battery and the costs brought by the aging and degradation of the fuel cell and the battery. Its solution is to minimize the total energy consumption objective function.

Images