Hybrid vehicle adaptive energy management method
By combining the CNN-BiLSTM-Attention model and the ECMS control module, accurate prediction of the operating conditions of hybrid electric vehicles and optimization of the power system are achieved, solving the problem of inaccurate operating condition prediction in existing technologies and improving fuel economy and driving experience.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANHUI POLYTECHNIC UNIV
- Filing Date
- 2025-07-23
- Publication Date
- 2026-06-09
AI Technical Summary
Existing hybrid vehicle energy management systems are inadequate in terms of the accuracy and real-time performance of operating condition predictions, making it difficult to achieve optimal fuel economy and smooth power output.
The CNN-BiLSTM-Attention model combined with the dung beetle algorithm is used to optimize the prediction of operating conditions. The engine and motor torque are allocated through the ECMS control module. A total cost function is established to minimize fuel consumption. K-means clustering and multi-feature datasets are used for operating condition identification and power system optimization.
It improves the accuracy of operating condition prediction and fuel economy, ensures smooth power output and driving comfort, reduces fuel consumption and emissions, and adapts to different driving environments in terms of energy efficiency.
Smart Images
Figure CN122174352A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent management of new energy, specifically, this invention relates to an adaptive energy management method for hybrid electric vehicles. Background Technology
[0002] Hybrid electric vehicles, as an environmentally friendly and efficient mode of transportation, combine the advantages of internal combustion engines and electric motors, significantly improving fuel economy and reducing emissions, thus meeting the modern society's demand for energy conservation and emission reduction. Through intelligent energy management strategies (EMS), vehicles can optimize power output according to different driving conditions, reducing fuel consumption and enhancing the driving experience.
[0003] Hybrid electric vehicles (HEVs), as typical multi-power input and multi-objective control systems, involve highly complex processes such as the rational allocation of torque between the engine and motor, power parameter distribution, and regenerative braking. Currently, HEV energy management strategies are mainly divided into two categories: rule-based and optimization-based. Much current research incorporates deep learning models into optimization-based strategies to improve the adaptability of energy management systems, such as identifying and predicting road conditions, and implementing different energy management scheduling for different conditions. However, despite some technological advancements, the accuracy of condition prediction remains affected by multiple factors, and computational efficiency and real-time performance are key challenges in achieving precise energy management.
[0004] Chinese patent application number 202210359338.0 discloses a driving condition prediction method based on a Long Short-Term Memory (LSTM) model. However, relying solely on time and vehicle speed as features makes it difficult for the model to capture sufficient information, especially for complex nonlinear tasks during vehicle operation. Therefore, the driving condition prediction-based energy management strategy for hybrid electric vehicles still needs continuous optimization to improve its performance and adaptability in practical applications.
[0005] The aim is to provide an adaptive energy management method for hybrid electric vehicles, particularly regarding how to improve the accuracy of predicting the operating conditions of hybrid electric vehicles to achieve optimal fuel economy. Summary of the Invention
[0006] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention provides an adaptive energy management method for hybrid electric vehicles, with the aim of improving the accuracy of predicting the operating conditions of hybrid electric vehicles.
[0007] To achieve the above objectives, the technical solution adopted by the present invention is: an adaptive energy management method for hybrid electric vehicles, comprising the following steps:
[0008] Step 1): Construct a mathematical model of a hybrid electric vehicle with a P2 configuration;
[0009] Step 2): Organizing the vehicle speed prediction dataset under various working conditions;
[0010] Step 3): Extract kinematic parameters of multiple features from the vehicle speed dataset samples to obtain a multi-feature dataset;
[0011] Step 4): Divide the driving conditions into three different driving conditions to obtain the driving condition dataset. The three driving conditions include urban driving conditions, suburban driving conditions, and highway driving conditions.
[0012] Step 5): Use K-means clustering to cluster the driving condition dataset to obtain cluster labels for the three driving conditions, thus obtaining a dataset with driving condition labels;
[0013] Step 6): Import the dataset with the working condition label into the CNN-BiLSTM-Attention working condition prediction model, and use the dung beetle algorithm to optimize the first eight features of the CNN-BiLSTM-Attention working condition prediction model as input, and output the working condition prediction result.
[0014] Step 7): Using the ECMS control module, based on the operating condition identification results and the vehicle's required torque T... req Maximum engine torque T emax Engine minimum torque T emin Battery state of charge (SOC), motor speed (n) m The engine torque T is calculated by the ECMS control module using the equivalent factor s as input. e_out Motor torque T m_out and total fuel consumption total As output;
[0015] Step 8): The ECMS control module establishes the total cost function and determines the state variables and control variables, defining minimum fuel consumption as the optimization objective for hybrid vehicles.
[0016] The ECMS control module includes an engine fuel consumption calculation module, a constraint condition judgment module, and a torque distribution module. In step 8), the constraint condition judgment module and the torque distribution module establish the total cost function and determine the state variables and control variables based on the Pontryagin minimum principle.
[0017] In step 8), the total cost function is established as follows: t0,t f m represents the start and end times. f (T e (t), where t) represents the engine fuel consumption rate, f(SOC) (tf) ) is the penalty function related to SOC at the end of the loop.
[0018] In step 8), the constraints on the variables are as follows:
[0019] SOC L SOC H These are the minimum and maximum values of SOC, respectively.
[0020] Based on the Pontryagin minimum principle, the Hamiltonian function that minimizes equivalent fuel consumption is established as follows:
[0021] H(SOC (t) Te (t) , λ (t) ,t)=m f (T e(，t) ,t)+λ (t) SOC (t)
[0022] Where I is the battery pack current (A); U OC P is the open-circuit voltage (V) of the battery pack. bat (t) represents the battery pack discharge power (W), E bat (t) represents the battery pack energy (kJ), S (t) Hlhv is the equivalent factor, and Hlhv is the calorific value of the fuel.
[0023] The mathematical model of the hybrid electric vehicle includes an engine module, a motor module, and a power battery. When modeling the engine module, the output speed, output torque, and instantaneous fuel consumption rate of the engine at different operating points are obtained through experimental calibration. Based on the experimental data, the universal characteristic curve of the engine is obtained. The modeling method of the motor module is the same as that of the engine module, which is based on the motor efficiency MAP diagram for interpolation processing. The motor power or torque is used as the control input. The motor module can be used as both a drive motor and a generator.
[0024] Based on the functional relationship between the motor efficiency and speed / torque in the aforementioned motor efficiency MAP diagram, the motor efficiency η is calculated to be... m =f 插值关系 (T m ,n m The power P of the motor m The calculation is as follows:
[0025] The time motor is used as a drive motor, T m When the value is less than 0, the motor is used as a generator.
[0026] When modeling the power battery module, R is selected. int The internal resistance model is used as a power battery model, Ubat =U oc -(I g +I m )R int U oc It is the battery terminal voltage, U oc It is the battery open-circuit voltage, I g It is the generator current, I m It is the drive motor current; the SOC value of the power battery is estimated by the ampere-hour integration method, I m This refers to the motor output current; SOC t and SOC int These are the current SOC value and the initial SOC value of the power battery, respectively; Q bat For power battery capacity; I m and SOC t The calculation is as follows:
[0027]
[0028] In step 2), the driving conditions of heavy-duty vehicles in China in 2019 are selected as the original data. In order to ensure that there are enough samples to train the CNN-BiLSTM-Attention driving condition prediction model, the overlapping sampling method is used on the original dataset to select kinematic segments at different ΔT time intervals and accumulate the number of training samples.
[0029] In step 5), the dataset X = {x1, x2, ..., x} n}, where each data point x i =(x i1 ,x i2 …,x i8 The clustering method contains 8 feature parameters. The purpose of clustering is to divide these data points into k=3 clusters to represent three operating conditions: urban, suburban and highway.
[0030] In step 5), the cluster centers are initialized by randomly selecting k data points as the initial cluster centers μ1, μ2…μ k ; Calculate the distance between each data point and the cluster center, Where x il μ represents the l-th feature of the i-th data point. jl Represent the i-th feature of the j-th cluster; assign each data point to the nearest cluster based on distance, where each data point x i Assigned to cluster C j , For each cluster C j Update the cluster center to the mean of all data points in the cluster. Where |C j | is cluster Cj The number of data points in the middle, x i These are data points within a cluster; finally, the distance between the points and the cluster center is repeatedly calculated and updated until the cluster center no longer changes.
[0031] The adaptive energy management method for hybrid electric vehicles of the present invention has the following beneficial effects:
[0032] 1. Improve fuel economy. By accurately predicting real-time operating conditions, this invention can flexibly adjust the power distribution strategy of hybrid vehicles according to different operating conditions (urban suburbs, highways), thereby achieving optimal power (torque) distribution between the engine and the electric motor, minimizing fuel consumption, and improving fuel economy.
[0033] 2. Accurate Operating Condition Prediction. This invention employs a CNN-BiLSTM-Attention model to predict operating conditions based on multi-feature data such as vehicle speed. Furthermore, it utilizes the Dung Beetle Algorithm (DBO) to optimize the learning rate, the number of neurons in the BiLSTM, the key values of the attention mechanism, and the regularization parameters. Compared to single LSTM or RBF vehicle speed prediction methods, this method can more accurately identify the current operating condition, ensuring more reasonable energy management decisions.
[0034] 3. Optimized Energy Management Strategy. This invention introduces an instantaneous equivalent fuel minimum control strategy, which can intelligently adjust the working state of the power system while ensuring the overall vehicle power performance. It can distribute torque according to different operating conditions, smoothly adjust the output torque of the engine and motor, avoid the power output fluctuations that may occur under traditional energy management strategies, and improve driving smoothness and comfort.
[0035] 4. This invention can precisely control the working mode of the power system, effectively reduce fuel consumption, and thus reduce vehicle exhaust emissions, which helps to reduce greenhouse gas and pollutant emissions and has significant environmental benefits.
[0036] 5. By accurately predicting vehicle operating conditions, the invention can adapt to different driving environments. Whether it is complex urban traffic conditions or long-distance driving in suburbs and on highways, it can rationally allocate the power system according to actual needs, effectively improving the energy efficiency of hybrid vehicles under different operating conditions. Attached Figure Description
[0037] This manual includes the following figures, which illustrate the following:
[0038] Figure 1 This is a structural framework diagram of an adaptive energy management strategy for hybrid electric vehicles based on operating condition prediction.
[0039] Figure 2This is a schematic diagram of the CNN-BiLSTM-Attention working condition prediction model structure;
[0040] Figure 3 It is the vehicle speed curve of the working condition dataset;
[0041] Figure 4 This is the result of working condition cluster analysis;
[0042] Figure 5(a) is the error curve between the actual value and the predicted value of the working condition prediction output;
[0043] Figure 5(b) is a bar chart of the mean absolute error, mean absolute percentage error and root mean square error of the prediction model;
[0044] Figure 6 This is a MAP diagram of the instantaneous fuel consumption of the engine of the vehicle model shown in this patent;
[0045] Figure 7 This is the battery SOC change simulated under WLTC conditions using the method described in this patent;
[0046] Figure 8 This is the simulated engine landing point under WLTC conditions using the method described in this patent. Detailed Implementation
[0047] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, in order to help those skilled in the art to have a more complete, accurate and in-depth understanding of the concept and technical solutions of the present invention, and to facilitate its implementation.
[0048] This invention provides an adaptive energy management method for hybrid electric vehicles, comprising the following steps:
[0049] Step 1): Construct a hybrid electric vehicle model with a P2 configuration;
[0050] Step 2): Organizing the vehicle speed prediction dataset under various working conditions;
[0051] Step 3): Extract kinematic parameters of multiple features from the vehicle speed dataset samples to obtain a multi-feature dataset;
[0052] Step 4): Divide the driving conditions into three different driving conditions to obtain the driving condition dataset. The three driving conditions include urban driving conditions, suburban driving conditions, and highway driving conditions.
[0053] Step 5): Use K-means clustering to cluster the driving condition dataset to obtain cluster labels for the three driving conditions, thus obtaining a dataset with driving condition labels;
[0054] Step 6): Import the dataset with the working condition label into the CNN-BiLSTM-Attention working condition prediction model, and use the dung beetle algorithm to optimize the first eight features of the CNN-BiLSTM-Attention working condition prediction model as input, and output the working condition prediction result.
[0055] Step 7): Using the ECMS control module, based on the operating condition identification results and the vehicle's required torque T... req Maximum engine torque T emax Engine minimum torque T emin Battery state of charge (SOC), motor speed (n) m The engine torque T is calculated by the ECMS control module using the battery equivalent factor s as input. e_out Motor torque T m_out and total fuel consumption total As output;
[0056] Step 8): The ECMS control module establishes the total cost function and determines the state variables and control variables, defining minimum fuel consumption as the optimization objective for hybrid vehicles.
[0057] The ECMS control module includes an engine fuel consumption calculation module, a constraint condition judgment module, and a torque distribution module. In step 8), the constraint condition judgment module and the torque distribution module establish the total cost function and determine the state variables and control variables based on the Pontryagin minimum principle.
[0058] In step 8), the total cost function is established as follows: t0,t f m represents the start and end times. f (T e (t), where t) represents the engine fuel consumption rate, f(SOC) (tf) ) is the penalty function related to SOC at the end of the loop.
[0059] In step 8), the constraints on the variables are as follows:
[0060] SOC L SOC H These are the minimum and maximum values of SOC, respectively.
[0061] Based on the Pontryagin minimum principle, the Hamiltonian function that minimizes equivalent fuel consumption is established as follows:
[0062] H(SOC (t) ,Te (t) ,λ (t) ,t)=m f (T e(t),t)+λ (t) SOC (t)
[0063] Where I is the battery pack current (A); U OC P is the open-circuit voltage (V) of the battery pack. bat (t) represents the battery pack discharge power (W), E bat (t) represents the battery pack energy (kJ), S (t) Hlhv is the battery equivalence factor, and Hlhv is the fuel calorific value.
[0064] The mathematical model of the hybrid electric vehicle includes an engine module, a motor module, and a power battery. When modeling the engine module, the output speed, output torque, and instantaneous fuel consumption rate of the engine at different operating points are obtained through experimental calibration. Based on the experimental data, the universal characteristic curve of the engine is obtained. The modeling method of the motor module is the same as that of the engine module, which is based on the motor efficiency MAP diagram for interpolation processing. The motor power or torque is used as the control input. The motor module can be used as both a drive motor and a generator.
[0065] Based on the functional relationship between the motor efficiency and speed / torque in the aforementioned motor efficiency MAP diagram, the motor efficiency η is calculated to be... m =f 插值关系 (T m ,n m The power P of the motor m The calculation is as follows:
[0066] The time motor is used as a drive motor, T m When the value is less than 0, the motor is used as a generator.
[0067] When modeling the power battery module, R is selected. int The internal resistance model is used as a power battery model, U bat =U oc -(I g +I m )R int U oc It is the battery terminal voltage, U oc It is the battery open-circuit voltage, I g It is the generator current, I m It is the drive motor current; the SOC value of the power battery is estimated by the ampere-hour integration method, I m This is the motor's output current;
[0068] SOC t and SOC int These are the current SOC value and the initial SOC value of the power battery, respectively; Qbat For power battery capacity; I m and SOC t The calculation is as follows:
[0069]
[0070] In step 2), the 2019 China Light Vehicle Driving Condition (CLTC) is selected as the original data. To ensure that there are enough samples for training the CNN-BiLSTM-Attention condition prediction model, the overlapping sampling method is used on the original dataset to select kinematic segments at different ΔT time intervals and accumulate the number of training samples.
[0071] In step 5), the dataset X = {x1, x2, ..., x} n}, where each data point x i =(x i1 ,x i2 …,x i8 The clustering method contains 8 feature parameters. The purpose of clustering is to divide these data points into k=3 clusters to represent three operating conditions: urban, suburban and highway.
[0072] In step 5), the cluster centers are initialized by randomly selecting k data points as the initial cluster centers μ1, μ2…μ k ; Calculate the distance between each data point and the cluster center, Where x il μ represents the l-th feature of the i-th data point. jl Represent the i-th feature of the j-th cluster; assign each data point to the nearest cluster based on distance, where each data point x i Assigned to cluster C j , For each cluster C j Update the cluster center to the mean of all data points in the cluster. Where |C j | is cluster C j The number of data points in the middle, x i These are data points within a cluster; finally, the distance between the points and the cluster center is repeatedly calculated and updated until the cluster center no longer changes.
[0073] Example
[0074] The purpose of this invention is to improve the accuracy of operating condition prediction by employing a CNN-BiLSTM-Attention model with an attention mechanism and using the equivalent fuel minimum control strategy (ECMS) to rationally allocate the torque of the engine and electric motor based on the predicted operating conditions, thereby achieving optimal fuel economy.
[0075] This invention provides an adaptive energy management method for hybrid electric vehicles based on operating condition prediction, comprising the following steps:
[0076] Step 1): Construct a mathematical model of a P2 hybrid electric vehicle, consisting of modules such as the engine module (FC), electric motor module (MT), and power battery (EQ). The vehicle's required torque T... req Engine output torque T e Motor output torque T m Engine speed n e Motor speed n m The actual speed of the vehicle, v actual And battery state of charge (SOC) and actual state of charge (SOC) actual .
[0077] Step 2): Organizing the vehicle speed prediction dataset under operating conditions. The 2019 driving conditions of light vehicles in China were selected as the original data. To ensure a sufficient number of samples for training the CNN-BiLSTM-Attention model, overlapping sampling was used on the original dataset to select kinematic segments at different time intervals (ΔT), accumulating the number of training samples.
[0078] Step 3): Extract multi-feature kinematic parameters from the obtained vehicle speed dataset samples. Introducing multi-dimensional features provides more driving information, enhances prediction accuracy, and better captures complex driving patterns. Single vehicle speed data may be affected by sensor errors or environmental factors, resulting in noise. Multi-feature datasets, by providing multiple perspectives, can reduce the impact of noise from a single feature on the model's prediction results. Even if the measurement value of one feature is incorrect, other features can still provide useful information, enhancing the model's robustness.
[0079] The selected driving characteristic parameters are: average vehicle speed v a Average acceleration a acc Average deceleration a dec Parking time ratio t, maximum speed v max Maximum acceleration acc max Maximum deceleration dec max The eight characteristic parameters are the vehicle speed (excluding idle speed) and the cycle speed v0.
[0080] Step 4): Classify driving conditions. Divide the multi-feature dataset obtained above into three different driving conditions: urban, suburban, and highway. Urban driving conditions are affected by urban traffic, with more starts and stops, and lower vehicle speeds; suburban driving conditions have moderate vehicle speeds and fewer starts and stops; highway driving conditions involve stable high-speed driving with almost no stops. Different labels are used to represent different driving conditions: "1" represents urban, "2" represents suburban, and "3" represents highway.
[0081] Step 5): Cluster the above work condition dataset using K-means clustering to obtain cluster labels for the three work conditions, which are then used for work condition prediction in CNN-BiLST-M Atte. Based on the dataset X = {x1, x2, ..., x...} described above... n}, where each data point x i =(x i1 ,x i2 …,x i8 The clustering method contains 8 feature parameters. The purpose of clustering is to divide these data points into k=3 clusters to represent three operating conditions: urban, suburban and highway.
[0082] Furthermore, initialize cluster centers by randomly selecting k data points as initial cluster centers μ1, μ2…μ k ; Calculate the distance between each data point and the cluster center, Where x il μ represents the l-th feature of the i-th data point. jl Represent the i-th feature of the j-th cluster; assign each data point to the nearest cluster based on distance, where each data point x i Assigned to cluster C j , For each cluster C j Update the cluster center to the mean of all data points in the cluster. Where |C j | is cluster C j The number of data points in the middle, x i These are data points within a cluster. Finally, the distance between each point and the cluster center is repeatedly calculated and updated until the cluster center no longer changes.
[0083] Step 6): Import the dataset with working condition labels obtained above into the CNN-BiLSTM-Attention working condition prediction model, and use the dung beetle algorithm to optimize the first eight features of the prediction model as input and the last column of working condition labels as output.
[0084] Step 7): The ECMS control module, consisting of an engine fuel consumption calculation module, a constraint condition judgment module, and a torque distribution module, uses the operating condition identification results to determine the vehicle's required torque T. req Engine maximum torque T emax Engine minimum torque T emin Battery state of charge (SOC), motor speed n m The battery equivalent factor s is taken as input, and the output is the engine torque T obtained by the ECMS calculation module. e_out Motor torque T m_out and total fuel consumption totalAs output.
[0085] Furthermore, the engine fuel consumption calculation module in ECMS is based on the engine universal characteristic diagram, which is essentially determined by the engine speed n. e Engine torque T e A contour map consisting of three elements: fuel consumption, and water content, is generated by inputting n. e and T e The corresponding engine fuel consumption can be obtained by referring to the table. e .
[0086] Step 8): The constraint judgment module and torque distribution module establish the total cost function and determine the state variables and control variables based on the Pontryagin Minimum Principle (PMP). Minimizing fuel consumption is defined as the optimization objective of the hybrid vehicle, and the cost function is established as follows: t0,t f m represents the start and end times. f (T e (t), where t) represents the engine fuel consumption rate, f(SOC) (tf) `()` is the penalty function related to SOC at the end of the loop. The constraints on the variables are as follows:
[0087] SOC L SOC H These are the minimum and maximum values of SOC, respectively.
[0088] Furthermore, based on PMP, the Hamiltonian function that minimizes equivalent fuel consumption is established as follows:
[0089] H(SOC (t) ,Te (t) ,λ (t) ,t)=m f (T e(t) ,t)+λ (t) SOC (t)
[0090]
[0091] Where I is the battery pack current (A); U OC P is the battery open-circuit voltage (V). bat (t) represents the battery discharge power (W), E bat (t) represents the battery pack energy (kJ), S (t) Hlhv is the battery equivalence factor, and Hlhv is the fuel calorific value.
[0092] Figure 1The diagram shows the structural framework of an adaptive energy management strategy for hybrid electric vehicles based on driving condition prediction. Driving condition data is collected, combined into a driving condition information set, and an overlap sampling method is used to increase the number of sample points. Eight kinematic features, as shown in Table 1, are selected.
[0093] Table 1 Kinematic characteristic parameters
[0094]
[0095]
[0096] K-means clustering was used to cluster the feature-rich driving condition dataset, obtaining cluster labels for three driving conditions. A column of driving condition labels was added to the dataset as output, resulting in a dataset with driving condition labels. The clustering results are as follows: Figure 4 As shown.
[0097] Figure 2 The model shown is a working condition prediction model built based on the CNN-BiLSTM-Attention deep learning model. The input is the dataset with working condition labels obtained in step 5), and the output is the working condition prediction result.
[0098] In step 7), when using the ECMS control module, the battery equivalent factor s is used as input, and different equivalent factors are set for different operating conditions, such as... Figure 1 As shown, an equivalent factor S is set for suburban operating conditions. ubran An equivalent factor S is set for urban working conditions. city An equivalent factor S is set for high-speed operating conditions. high Suburban driving conditions refer to the driving conditions of a car on suburban roads, urban driving conditions refer to the driving conditions of a car on urban roads, and highway driving conditions refer to the driving conditions of a car on highways.
[0099] Finally, the CNN-BiLSTM-Attention working condition prediction model was combined with the ECMS control module to execute this energy management strategy in a P2 configuration hybrid electric vehicle model.
[0100] Starting from step 5 above, a mathematical model of a hybrid electric vehicle with a certain P2 configuration is used as the experimental vehicle for this energy management strategy. The implementation process is as follows.
[0101] Engine module modeling. Through experimental calibration, the engine's output speed, output torque, and instantaneous fuel consumption rate at different operating points were obtained. Based on the experimental data, the following was derived: Figure 6 The universal characteristic curve of the engine, and the fuel consumption curve represent the points where the instantaneous fuel consumption rate of the engine is the same at different operating points.
[0102] Motor module modeling. The method for motor modeling is the same as for the engine, based on interpolation of the motor efficiency MAP. The required motor power or torque can be used as a control input. The motor used can be used as either a motor or a generator. The established motor characteristic diagram is shown below. Figure 7 As shown in the diagram. Based on the functional relationship between the motor's speed and torque in the MAP diagram, the motor efficiency η is calculated. m =f 插值关系 (T m ,n m The power P of the motor m The calculation is as follows:
[0103] The time motor is used as an electric motor, T m When the value is less than 0, the motor is used as a generator.
[0104] Modeling the power battery module. Select R. int The internal resistance model is used as a power battery model, U bat =U oc -(I g +I m )R int U bat It is the battery terminal voltage, U oc It is the battery open-circuit voltage I g , is the generator current, I m This is the drive motor current. The battery's SOC is estimated using the ampere-hour integration method, I m This refers to the motor output current; SOC t and SOC int These are the current SOC value and the initial SOC value of the power battery, respectively; Q bat Battery capacity; I m and SOC t The calculation is as follows:
[0105]
[0106] Calculation of vehicle torque requirements. Based on the vehicle's longitudinal dynamics equations, calculate the vehicle's torque requirements. F t =F f +F i +F w +F j , where F t F f F i F w F j These are traction force, rolling resistance, gradient resistance, air resistance, and acceleration resistance, F. f F i F w F jThe calculation is as follows:
[0107]
[0108] Where f is the road friction coefficient, θ is the road slope angle, and C d ρ is the air drag coefficient, ρ is the air density, and σ is the air drag coefficient. m To increase the drag coefficient, combining the above formulas, we can obtain the required torque at the wheel end as T. req =(F f +F i +F w +F j )r.
[0109] like Figure 3 As shown, four road spectra—CYC_WVUCITY, CYC_UDDS, CYC_EUDC, and CYC_MANHATTAN—were used as the original datasets for prediction. The vehicle speed segment sampling frequency was set to 1Hz, covering various typical driving behaviors such as acceleration, deceleration, and idling. Overlap sampling was used to increase the data volume, with window lengths set to 25s, 50s, and 100s. Eight features, including average vehicle speed and average acceleration, were calculated for each time period. Min-Max normalization was performed on each feature, mapping all feature values to [0,1] to eliminate the influence of units and remove null (Nan) values.
[0110] K-means clustering was used to cluster the multi-feature data. Initial cluster centers were randomly selected, and the number of clusters was set to 3, corresponding to urban, suburban, and highway driving conditions, respectively. After clustering, the statistical distribution of each cluster was calculated. The clustering results are shown below. Figure 4 As shown.
[0111] The clustered data from the previous step is formatted and organized. The features of each time segment are combined into a time series matrix of the form (N, T, F), where N is the number of samples, T is the time step, and F is the feature dimension. For example... Figure 2 The CNN module is used to extract the spatial distribution features of the data. The CNN convolutional layer uses a one-dimensional convolutional kernel (1D-CNN) to perform convolution operations on multi-feature data of the time series.
[0112] The convolution kernel slides to extract the correlation between features. The formula for calculating convolution is: Where W is the kernel weight, b is the bias, and k is the kernel size. CNN pooling layers reduce feature dimensionality after convolution using max pooling, reducing data redundancy while preserving salient features, improving model computational efficiency, and reducing the risk of overfitting. For example... Figure 2The BiLSTM module is used to capture time series dependencies. BiLSTM considers both forward and backward dependencies of the time series. Forward LSTM processes historical information of the time series; backward LSTM captures the influence of future information. The hidden state at each time step is given by equation h. t =σ(W h ·[h t-1 ,x t ]+b h ), where h t It is in a hidden state, x t It is the input at the current time step, W h and b h These are weights and biases. Finally, the spatial features extracted by the CNN are used as time-series data and input into the BiLSTM. After modeling the dynamic changes of the time series, a hidden state matrix of the time series is output, which serves as a time-dependent representation of the vehicle's driving characteristics.
[0113] The attention mechanism enhances the model's focus on key features by calculating the importance weights of the hidden states. The weight calculation formula is as follows: Where, α t W is the attention weight at time step t. a It is a learnable weight matrix. The global feature vector output by the Attention mechanism combines the weight information of temporal and spatial features, providing a more accurate input for the final working condition classification.
[0114] The feature vector output by the Attention mechanism is fed into the fully connected layer, mapping it to the target classification space. The Softmax activation function is used to calculate the probability of each class.
[0115]
[0116] Among them, W c and b c These are the weights and biases of category c.
[0117] The final classification result outputs the predicted category of the vehicle's driving condition (urban, suburban, or highway), with corresponding labels 1, 2, and 3, respectively.
[0118] In this embodiment of the invention, to achieve the optimal distribution of torque between the hybrid vehicle's engine and motor, and combining the aforementioned operating condition prediction and equivalent fuel minimization control strategy, with the goal of minimizing the overall energy consumption of the vehicle, step 8) includes the following steps:
[0119] Step 801): Data preprocessing and input. Input is as follows: Figure 6The system includes engine speed, torque data, and corresponding fuel consumption data. An interpolation function is used to fit the fuel consumption within the torque range, ensuring that fuel consumption under various operating conditions can be quickly retrieved from the table during actual operation. This is implemented in the code as: `fuel_lookup=@(Te_value)interp1(Te,fuel_consumption,Te_value,'linear','extrap')`.
[0120] Use 'extrap' to handle torque values that are out of range;
[0121] Step 802): Calculate the current engine fuel consumption. Using the fuel_lookup function, calculate the fuel consumption based on the input engine speed n. e Calculate the engine's fuel consumption. Implement this in the code:
[0122] engine_fuel_consumption=fuel_lookup(ne);
[0123] Step 803): Define the optimization objective. Based on the engine's current torque and speed, look up the corresponding fuel consumption in a table; introduce the battery equivalence factor s to quantify the impact of motor torque on energy consumption into a fuel equivalence form, thereby uniformly evaluating the energy consumption of the engine and motor. The Hamiltonian function takes the following form:
[0124]
[0125] in, It refers to the engine fuel consumption rate; implemented in the code:
[0126] H=@(Te_value,SOC_value,Tm_value)(fuel_lookup(Te_value)+(s*(Tm_value / ne)));
[0127] Step 804): Optimize engine and motor torque. Based on the vehicle torque requirement T... req Given the known conditions, by optimizing the engine torque T e and motor torque T m The distribution ensures that the total torque meets the following constraints: T req =T e +T m At the same time, the torque of the engine and the electric motor must meet their physical range: T e ∈[T min ,T max ],T m ∈[T m,min ,T m,max ].
[0128] In this embodiment of the invention, the optimization process for engine and motor torque employs an exhaustive search method:
[0129] 1. Traverse the engine torque T e and motor torque T m All possible combinations;
[0130] 2. For each combination, calculate the corresponding total fuel consumption;
[0131] 3. Record the torque combination that minimizes total fuel consumption as the optimal solution.
[0132] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution; or the direct application of the inventive concept and technical solution to other situations without modification, are all within the protection scope of the present invention.
Claims
1. An adaptive energy management method for hybrid electric vehicles, characterized in that, Including the following steps: Step 1): Construct a mathematical model of a hybrid electric vehicle with a P2 configuration; Step 2): Organizing the vehicle speed prediction dataset under various working conditions; Step 3): Extract kinematic parameters of multiple features from the vehicle speed dataset samples to obtain a multi-feature dataset; Step 4): Divide the driving conditions into three different driving conditions to obtain the driving condition dataset. The three driving conditions include urban driving conditions, suburban driving conditions, and highway driving conditions. Step 5): Use K-means clustering to cluster the driving condition dataset to obtain cluster labels for the three driving conditions, thus obtaining a dataset with driving condition labels; Step 6): Import the dataset with the working condition label into the CNN-BiLSTM-Attention working condition prediction model, and use the dung beetle algorithm to optimize the first eight features of the CNN-BiLSTM-Attention working condition prediction model as input, and output the working condition prediction result. Step 7): Using the ECMS control module, based on the operating condition identification results and the vehicle's required torque T... req Maximum engine torque T emax Engine minimum torque T emin Battery state of charge (SOC), motor speed (n) m The engine torque T is calculated by the ECMS control module using the equivalent factor s as input. e_out Motor torque T m_out and total fuel consumption total As output; Step 8): The ECMS control module establishes the total cost function and determines the state variables and control variables, defining minimum fuel consumption as the optimization objective for hybrid vehicles.
2. The adaptive energy management method for hybrid electric vehicles according to claim 1, characterized in that, The ECMS control module includes an engine fuel consumption calculation module, a constraint condition judgment module, and a torque distribution module. In step 8), the constraint condition judgment module and the torque distribution module establish the total cost function and determine the state variables and control variables based on the Pontryagin minimum principle.
3. The adaptive energy management method for hybrid electric vehicles according to claim 2, characterized in that, In step 8), the total cost function is established as follows: t0,t f m represents the start and end times. f (T e (t), where t) represents the engine fuel consumption rate, f(SOC) (tf) ) is the penalty function related to SOC at the end of the loop.
4. The adaptive energy management method for hybrid electric vehicles according to claim 3, characterized in that, In step 8), the constraints on the variables are as follows: SOC L SOC H These are the minimum and maximum values of SOC, respectively. Based on the Pontryagin minimum principle, the Hamiltonian function that minimizes equivalent fuel consumption is established as follows: Where I is the battery pack current (A); U OC P is the open-circuit voltage (V) of the battery pack. bat (t) represents the battery pack discharge power (W), E bat (t) represents the battery pack energy (kJ), S (t) Hlhv is the equivalent factor, and Hlhv is the calorific value of the fuel.
5. The adaptive energy management method for hybrid electric vehicles according to any one of claims 1 to 4, characterized in that, The mathematical model of the hybrid electric vehicle includes an engine module, a motor module, and a power battery. When modeling the engine module, the output speed, output torque, and instantaneous fuel consumption rate of the engine at different operating points are obtained through experimental calibration. Based on the experimental data, the universal characteristic curve of the engine is obtained. The modeling method of the motor module is the same as that of the engine module, which is based on the motor efficiency MAP diagram for interpolation processing. The motor power or torque is used as the control input. The motor module can be used as both a drive motor and a generator.
6. The adaptive energy management method for hybrid electric vehicles according to claim 5, characterized in that, Based on the functional relationship between the motor efficiency and speed / torque in the aforementioned motor efficiency MAP diagram, the motor efficiency η is calculated to be... m =f 插值关系 (T m ,n m The power P of the motor m The calculation is as follows: T m When the value is >0, the motor is used as a drive motor. m When the value is less than 0, the motor is used as a generator.
7. The adaptive energy management method for hybrid electric vehicles according to claim 5, characterized in that, When modeling the power battery module, R is selected. in The internal resistance model is used as a power battery model, U bat =U oc -(I g +I m )R int U oc It is the battery terminal voltage, U oc It is the battery open-circuit voltage, I g It is the generator current, I m It is the drive motor current; the SOC value of the power battery is estimated by the ampere-hour integration method, I m This refers to the motor output current; SOC t and SOC int These are the current SOC value and the initial SOC value of the power battery, respectively. Q bat For power battery capacity; I m and SOC t The calculation is as follows:
8. The adaptive energy management method for hybrid electric vehicles according to any one of claims 1 to 7, characterized in that, In step 2), the driving conditions of heavy-duty vehicles in China in 2019 are selected as the original data. In order to ensure that there are enough samples to train the CNN-BiLSTM-Attention driving condition prediction model, the overlapping sampling method is used on the original dataset to select kinematic segments at different ΔT time intervals and accumulate the number of training samples.
9. The adaptive energy management method for hybrid electric vehicles according to any one of claims 1 to 7, characterized in that, In step 5), the dataset X = {x1, x2, ..., x} n }, where each data point x i =(x i1 ,x i2 …,x i8 The clustering method contains 8 feature parameters. The purpose of clustering is to divide these data points into k=3 clusters to represent three operating conditions: urban, suburban and highway.
10. The adaptive energy management method for hybrid electric vehicles according to claim 9, characterized in that, In step 5), the cluster centers are initialized by randomly selecting k data points as the initial cluster centers μ1, μ2…μ k ; Calculate the distance between each data point and the cluster center, Where x il μ represents the l-th feature of the i-th data point. jl Represent the i-th feature of the j-th cluster; assign each data point to the nearest cluster based on distance, where each data point x i Assigned to cluster C j , For each cluster C j Update the cluster center to the mean of all data points in the cluster. Where |C j | is cluster C j The number of data points in the middle, x i These are data points within a cluster; finally, the distance between the points and the cluster center is repeatedly calculated and updated until the cluster center no longer changes.