A fuel cell vehicle energy management strategy integrating traffic perception

Abstract

This invention provides an energy management strategy for fuel cell vehicles that integrates traffic perception. It constructs a future speed prediction model based on a Transformer network, effectively representing the dynamic interaction between the vehicle and surrounding multiple vehicles. Compared to traditional speed prediction methods based on historical data or single-vehicle information, it exhibits higher accuracy and stability. Furthermore, a deep ensemble model structure is introduced for speed prediction, enabling the quantification of prediction uncertainty. Combined with a dynamic threshold mechanism, it achieves better prediction credibility discrimination and adaptive output adjustment, significantly improving the robustness of the method in complex traffic scenarios. Based on the predicted future speed sequence, a reinforcement learning energy management strategy is constructed, enabling early perception and rapid response to future traffic changes, and also demonstrating significant advantages in fuel economy.

Description

Technical Field

[0001] This invention belongs to the field of energy management technology for intelligent connected hybrid electric vehicles, and specifically relates to an energy management strategy for fuel cell vehicles that integrates traffic perception. Background Technology

[0002] For hybrid power systems, energy management strategy, as the core element that coordinates and determines the output power of each energy unit, determines the overall energy consumption and economic performance of the vehicle. With the development of related technologies, traditional rule-based or optimization-based energy management methods are gradually becoming insufficient to meet current needs. Recently emerging reinforcement learning-based energy management strategies have been widely deployed due to their advantages such as model independence, strong self-learning ability, and realistic environmental representation. However, this method relies excessively on current and historical operating data and lacks a good predictive mechanism for short-term future information, resulting in shortcomings in real-time performance and limiting energy-saving performance. Summary of the Invention

[0003] In view of this, and to address the technical problems existing in this field, the present invention provides an energy management strategy for fuel cell vehicles that integrates traffic perception, specifically including the following steps:

[0004] Step 1: Obtain traffic perception information through vehicle sensors and vehicle-to-everything (V2X) communication equipment, collect the speed and relative distance of at least six adjacent vehicles (front left, front right, front right, rear left, rear right) around the main vehicle in real time, and construct a traffic state input vector;

[0005] Step 2: Construct a speed prediction model containing several sub-models based on the Transformer algorithm. Use the traffic state input vector obtained in Step 1 as the model input and the predicted future speed sequence of the main vehicle as the model output. Each sub-model is trained independently. Calculate the mean and variance of the future vehicle speed sequence prediction results obtained by each sub-model using the deep ensemble method, and determine the uncertainty quantification index of the model prediction based on this.

[0006] Step 3: Set the corresponding threshold for the uncertainty quantification index and compare the two to determine whether the current prediction result is credible. If the result is credible, directly output the prediction result of the speed prediction model. If the result is not credible, perform exponential smoothing prediction to output a credible prediction result.

[0007] Step 4: Define the system state space based on the future speed sequence and vehicle energy state, define the action space based on the fuel cell system power, and define the corresponding reward function considering the energy consumption of fuel cell and power battery. On this basis, build an energy management optimizer based on deep reinforcement learning, so that it can select the corresponding fuel cell power control action according to the predicted future speed sequence and the current vehicle energy state, obtain the corresponding reward and enter the next state.

[0008] Step 5: Train the reinforcement learning-based energy management optimizer on historical operating data, and update the optimizer network parameters using the corresponding action-reward evaluation mechanism to make the output fuel cell control commands tend to be optimal.

[0009] Step 6: Deploy the trained energy management optimizer in an online application.

[0010] Furthermore, the traffic state input vector x constructed in step 1 t The specific format is as follows:

[0011] x t =[v0,v1,…,v6,d1,…,d6]

[0012] In the formula, v0 represents the current speed of the main vehicle; v i (i = 1, ..., 6) represents the current speed of the i-th adjacent vehicle; d i (i = 1, ..., 6) represents the relative distance between the i-th adjacent vehicle and the main vehicle.

[0013] Furthermore, the speed prediction model established in step 2 specifically utilizes the following multi-head attention mechanism to dynamically process traffic scene information:

[0014]

[0015] Q = XW Q K = XW K V = XW V

[0016] X = [x1, x2, ..., x T ]

[0017] In the formula, Q is the query vector matrix; K is the key vector matrix; V is the value vector matrix; W Q W K W V All parameters are trainable; X is the input sequence, specifically consisting of traffic state input vectors x at T time steps. t composition; Used to measure the similarity between each time step; Used for scaling, avoiding the numerical dimension d kExcessive size leads to gradient instability; the result obtained after performing softmax normalization is used as the weight, and thus the final representation of each time step is obtained.

[0018] Furthermore, in step 2, the mean and variance of the vehicle speed sequence predicted by the sub-model are calculated using the deep ensemble method according to the following formula:

[0019]

[0020] Where N is the number of sub-models; The velocity prediction result of the nth sub-model at time t+h; μ h The mean of the prediction results at time t+h; For the prediction variance, H is the prediction time step window length, and U is the uncertainty quantification index.

[0021] Furthermore, step 3 specifically sets the threshold δ for the uncertainty quantification index. t The format is as follows:

[0022]

[0023] in, tk is the historical mean uncertainty from time tk to time t; β is an empirically set scaling factor used to adjust the prediction tolerance; k is the historical sliding window length.

[0024] The uncertainty quantification index is compared with the threshold δ. t Compare, if U≤δ t If the current prediction result is deemed reliable, the future velocity sequence prediction result of the velocity prediction model is directly output; if U>δ t If the current prediction result is deemed unreliable, then the following exponential smoothing is performed to output the future velocity:

[0025]

[0026] in, The future velocity predicted by exponential smoothing; v t The current speed of the main vehicle is α; α∈(0,1) is a smoothing coefficient used to control the weight of the current speed and the historical trend.

[0027] Furthermore, step 4 specifically defines the following state space:

[0028]

[0029] in, For the predicted future velocity sequence; a t P is the current acceleration. fc,t ΔP represents the power output of the current fuel cell system.fc,t =P fc,t -P fc,t-1 For fuel cell power change rate; SOC t This represents the current state of charge of the power battery.

[0030] Define the following action space:

[0031]

[0032] in, Control actions for energy management strategies; These are the minimum and maximum limits for fuel cell power, respectively.

[0033] And design the following reward function:

[0034]

[0035] Where, r t The reward value at the current moment; SOC t This refers to the current state of charge (SOC) of the power battery. ref The target state of charge is used as a reference; ω1 and ω2 are the weight coefficients of each term in the reward function.

[0036] The energy management strategy for fuel cell vehicles that integrates traffic perception provided by the present invention constructs a future speed prediction model based on a Transformer network. This model can effectively characterize the dynamic interaction between the vehicle and surrounding multiple vehicles, exhibiting higher accuracy and stability compared to traditional speed prediction methods based on historical data or single-vehicle information. Furthermore, a deep integrated model structure is introduced for speed prediction, enabling the quantification of prediction uncertainty. Combined with a dynamic threshold mechanism, it achieves better prediction credibility discrimination and adaptive output adjustment, significantly improving the robustness of the method in complex traffic scenarios. The reinforcement learning energy management strategy constructed based on the predicted future speed sequence enables early perception and rapid response to future traffic changes, also demonstrating significant advantages in fuel economy. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the framework of the method provided by the present invention. Detailed Implementation

[0038] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] The energy management strategy for fuel cell vehicles that integrates traffic perception provided by this invention, such as Figure 1 As shown, the specific steps include:

[0040] Step 1: Obtain traffic perception information through vehicle sensors and vehicle-to-everything (V2X) communication equipment, collect the speed and relative distance of at least six adjacent vehicles (front left, front right, front right, rear left, rear right) around the main vehicle in real time, and construct a traffic state input vector;

[0041] Step 2: Construct a speed prediction model containing several sub-models based on the Transformer algorithm. Use the traffic state input vector obtained in Step 1 as the model input and the predicted future speed sequence of the main vehicle as the model output. Each sub-model is trained independently. Calculate the mean and variance of the future vehicle speed sequence prediction results obtained by each sub-model using the deep ensemble method, and determine the uncertainty quantification index of the model prediction based on this.

[0042] Step 3: Set the corresponding threshold for the uncertainty quantification index and compare the two to determine whether the current prediction result is credible. If the result is credible, directly output the prediction result of the speed prediction model. If the result is not credible, perform exponential smoothing prediction to output a credible prediction result.

[0043] Step 4: Define the system state space based on the future speed sequence and vehicle energy state, define the action space based on the fuel cell system power, and define the corresponding reward function considering the energy consumption of fuel cell and power battery. On this basis, build an energy management optimizer based on deep reinforcement learning, so that it can select the corresponding fuel cell power control action according to the predicted future speed sequence and the current vehicle energy state, obtain the corresponding reward and enter the next state.

[0044] Step 5: Train the reinforcement learning-based energy management optimizer on historical operating data, and update the optimizer network parameters using the corresponding action-reward evaluation mechanism to make the output fuel cell control commands tend to be optimal.

[0045] Step 6: Deploy the trained energy management optimizer in an online application.

[0046] In a preferred embodiment of the present invention, the traffic state input vector x constructed in step 1 t The specific format is as follows:

[0047] x t =[v0,v1,…,v6,d1,…,d6]

[0048] In the formula, v0 represents the current speed of the main vehicle; v i (i = 1, ..., 6) represents the current speed of the i-th adjacent vehicle; di (i = 1, ..., 6) represents the relative distance between the i-th adjacent vehicle and the main vehicle.

[0049] In a preferred embodiment of the present invention, the speed prediction model established in step 2 specifically utilizes the following multi-head attention mechanism to dynamically process traffic scene information:

[0050]

[0051] Q = XW Q K = XW K V = XW V

[0052] X = [x1, x2, ..., x T ]

[0053] In the formula, Q is the query vector matrix; K is the key vector matrix; V is the value vector matrix; W Q W K W V All parameters are trainable; X is the input sequence, specifically consisting of traffic state input vectors x at T time steps. t composition; Used to measure the similarity between each time step; Used for scaling, avoiding the numerical dimension d k Excessive size leads to gradient instability; the result obtained after performing softmax normalization is used as the weight, and thus the final representation of each time step is obtained.

[0054] In a preferred embodiment of the present invention, in step 2, the mean and variance of the vehicle speed sequence predicted by the sub-model are calculated using a deep integration method according to the following formula:

[0055]

[0056]

[0057] Where N is the number of sub-models; The velocity prediction result of the nth sub-model at time t+h; μ h The mean of the prediction results at time t+h; For the prediction variance, H is the prediction time step window length, and U is the uncertainty quantification index.

[0058] In a preferred embodiment of the present invention, step 3 specifically sets the threshold δ of the uncertainty quantification index. t The format is as follows:

[0059]

[0060] in, tk is the historical mean uncertainty from time tk to time t; β is an empirically set scaling factor used to adjust the prediction tolerance; k is the historical sliding window length.

[0061] The uncertainty quantification index is compared with the threshold δ. t Compare, if U≤δ t If the current prediction result is deemed reliable, the future velocity sequence prediction result of the velocity prediction model is directly output; if U>δ t If the current prediction result is deemed unreliable, then the following exponential smoothing is performed to output the future velocity:

[0062]

[0063] in, The future velocity predicted by exponential smoothing; v t The current speed of the main vehicle is α; α∈(0,1) is a smoothing coefficient used to control the weight of the current speed and the historical trend.

[0064] In a preferred embodiment of the present invention, step 4 specifically defines the following state space:

[0065]

[0066] in, For the predicted future velocity sequence; a t P is the current acceleration. fc,t ΔP represents the power output of the current fuel cell system. fc,t =P fc,t -P fc,t-1 For fuel cell power change rate; SOC t This represents the current state of charge of the power battery.

[0067] Define the following action space:

[0068]

[0069] in, Control actions for energy management strategies; These are the minimum and maximum limits for fuel cell power, respectively.

[0070] And design the following reward function:

[0071]

[0072] Where, r t The reward value at the current moment; SOC t This refers to the current state of charge (SOC) of the power battery. refThe target state of charge is used as a reference; ω1 and ω2 are the weight coefficients of each term in the reward function.

[0073] It should be understood that the sequence number of each step in the embodiments of the present invention does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An energy management strategy for fuel cell vehicles that integrates traffic perception, specifically including the following steps: Step 1: Obtain traffic perception information through vehicle-mounted sensors and vehicle-to-everything (V2X) communication equipment, collect the speed and relative distance of at least six adjacent vehicles around the main vehicle in real time, and construct a traffic state input vector; Step 2: Construct a speed prediction model containing several sub-models based on the Transformer algorithm. Use the traffic state input vector obtained in Step 1 as the model input and the predicted future speed sequence of the main vehicle as the model output. Each sub-model is trained independently. Calculate the mean and variance of the future vehicle speed sequence prediction results obtained by each sub-model using the deep ensemble method, and determine the uncertainty quantification index of the model prediction based on this. Step 3: Set the corresponding threshold for the uncertainty quantification index and compare the two to determine whether the current prediction result is credible. If the result is credible, directly output the prediction result of the speed prediction model. If the result is not credible, perform exponential smoothing prediction to output a credible prediction result. Step 4: Define the system state space based on the future speed sequence and vehicle energy state, define the action space based on the fuel cell system power, and define the corresponding reward function considering the energy consumption of fuel cell and power battery. On this basis, build an energy management optimizer based on deep reinforcement learning, so that it can select the corresponding fuel cell power control action according to the predicted future speed sequence and the current vehicle energy state, obtain the corresponding reward and enter the next state. Step 5: Train the energy management optimizer using historical operating conditions, and update the optimizer network parameters using the corresponding action-reward evaluation mechanism to make the output fuel cell control commands tend to be optimal. Step 6: Deploy the trained energy management optimizer in an online application.

2. The fuel cell vehicle energy management strategy integrating traffic perception as described in claim 1, characterized in that: The traffic state input vector x constructed in step 1 t The specific format is as follows: x t =[v0,v1,…,v6,d1,…,d6] In the formula, v0 represents the current speed of the main vehicle; v i (i = 1, ..., 6) represents the current speed of the i-th adjacent vehicle; d i (i = 1, ..., 6) represents the relative distance between the i-th adjacent vehicle and the main vehicle.

3. The fuel cell vehicle energy management strategy integrating traffic perception as described in claim 1, characterized in that: The speed prediction model established in step 2 specifically utilizes the following multi-head attention mechanism to dynamically process traffic scene information: Q=XW Q K=XW K V=XW V X=[x1,x2,...,x T ] In the formula, Q is the query vector matrix; K is the key vector matrix; V is the value vector matrix; W Q W K W V All parameters are trainable; X is the input sequence, specifically consisting of traffic state input vectors x at T time steps. t composition; Used to measure the similarity between each time step; Used for scaling, avoiding the numerical dimension d k Excessive size leads to gradient instability; the result obtained after performing softmax normalization is used as the weight, and thus the final representation of each time step is obtained.

4. The fuel cell vehicle energy management strategy integrating traffic perception as described in claim 1, characterized in that: In step 2, the mean and variance of the vehicle speed sequence predicted by the sub-model are calculated using the deep ensemble method according to the following formula: Where N is the number of sub-models; The velocity prediction result of the nth sub-model at time t+h; μ h The mean of the prediction results at time t+h; For the prediction variance, H is the prediction time step window length, and U is the uncertainty quantification index.

5. The fuel cell vehicle energy management strategy integrating traffic perception as described in claim 1, characterized in that: Step 3 specifically sets the threshold δ for the uncertainty quantification index. t The format is as follows: in, tk is the historical mean uncertainty from time tk to time t; β is an empirically set scaling factor used to adjust the prediction tolerance; k is the historical sliding window length. The uncertainty quantification index is compared with the threshold δ. t Compare, if U≤δ t If the current prediction result is deemed reliable, the future velocity sequence prediction result of the velocity prediction model is directly output; if U>δ t If the current prediction result is deemed unreliable, then the following exponential smoothing is performed to output the future velocity: in, The future velocity predicted by exponential smoothing; v t The current speed of the main vehicle is α; α∈(0,1) is a smoothing coefficient used to control the weight of the current speed and the historical trend.

6. The fuel cell vehicle energy management strategy integrating traffic perception as described in claim 1, characterized in that: Step 4 defines the following state space: in, For the predicted future velocity sequence; a t P is the current acceleration. fc,t This represents the power output of the current fuel cell system; αP fc,t =P fc,t -P fc,t-1 For fuel cell power change rate; SOC t This represents the current state of charge of the power battery. Define the following action space: in, Control actions for energy management strategies; These are the minimum and maximum limits for fuel cell power, respectively. And design the following reward function: Where, r t The reward value at the current moment; SOC t This refers to the current state of charge (SOC) of the power battery. ref The target state of charge is used as a reference; ω1 and ω2 are the weight coefficients of each term in the reward function.

Images