A sustained self-evolution vehicle energy management method and system based on cooperation of deterministic working conditions and uncertain disturbances

Abstract

The present application provides a kind of based on the persistent self-evolution vehicle energy management method and system of certainty working condition and uncertainty disturbance coordination, belong to vehicle energy management and intelligent control technical field, the method of the present application is with stable basic strategy as main controller, generates basic energy management control instruction according to current vehicle state;On this basis, based on current certainty working condition identification result and prior information of road ahead constructs certainty correction term, for the feedforward correction of basic control instruction in combination with current working condition difference and future working condition transition trend;While constructing uncertainty disturbance correction term, when detecting driving style mutation, traffic state fluctuation or demand distribution offset, output limited residual compensation quantity, and through amplitude constraint, change rate constraint and automatic attenuation mechanism guarantee online collaborative stability.Finally, basic energy management control instruction, certainty correction term and uncertainty disturbance correction term are fused to obtain final energy management control instruction.

Description

Technical Field

[0001] This invention relates to the field of vehicle energy management and intelligent control technology, and more particularly to a continuous self-evolving vehicle energy management method and system based on the synergy of deterministic operating conditions and uncertain disturbances, applicable to energy optimization control of hybrid electric vehicles or multi-energy drive system vehicles in complex road environments. Background Technology

[0002] With the widespread application of hybrid vehicles in energy conservation and emission reduction, achieving efficient and stable energy distribution under complex and ever-changing road conditions has become a key issue in vehicle control. Existing energy management strategies mainly include rule-based methods and optimization-based methods. Rule-based strategies rely on experience and are difficult to adapt to complex and ever-changing operating environments; while optimization-based methods have high theoretical optimality, they usually rely on complete operating condition information and are difficult to meet real-time control requirements.

[0003] In recent years, data-driven intelligent energy management methods have gradually emerged, such as reinforcement learning methods, which can achieve adaptive control of complex systems to a certain extent. However, existing methods still have the following problems: (1) Most methods make decisions based only on the current state and lack effective use of future road information (such as slope changes), resulting in a significant lag in the control strategy; (2) Existing methods usually adopt a single strategy model, which is difficult to adapt to multiple typical working conditions at the same time, resulting in a decrease in control performance when switching working conditions; (3) In actual vehicle operation, there are uncertain disturbances such as changes in driving behavior and fluctuations in traffic conditions. Existing methods usually lack effective online adaptive update mechanisms, making it difficult to maintain optimal performance in long-term operation. (4) Although some methods introduce learning mechanisms, they fail to distinguish between deterministic operating condition changes and uncertain disturbances, resulting in chaotic control structures and insufficient interpretability.

[0004] Therefore, there is an urgent need for a vehicle energy management method that can simultaneously utilize deterministic operating conditions and forward-looking information, and possess continuous adaptive evolution capabilities, in order to improve the system's economy, robustness, and generalization ability. Summary of the Invention

[0005] To address the technical problems of insufficient utilization of future information, poor adaptability to operating conditions, and lack of continuous self-adaptation in existing technologies, this paper proposes a continuous self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances, thereby improving the energy utilization efficiency and long-term operating performance of vehicles under complex operating conditions.

[0006] The technical means employed in this invention are as follows: A continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances includes: S1. Obtain the current operating status information of the vehicle and generate basic energy management control commands from the stable basic strategy; S2. Identify the current deterministic operating condition mode based on the vehicle's recent operating data, and generate a current mode correction sub-item based on the current deterministic operating condition mode; S3. Obtain prior road information within a preset prediction distance range ahead of the vehicle, construct a forward-looking feature sequence, and generate a forward-looking attention correction sub-item based on the current vehicle state information and the forward-looking feature sequence; S4. Merge the current mode correction sub-item and the forward attention correction sub-item to obtain the deterministic correction item; S5. Detect whether an uncertainty disturbance has occurred based on the degree of deviation of the vehicle's current operating behavior from the current deterministic operating condition mode reference distribution, and generate an uncertainty disturbance correction term when the triggering condition is met; S6. Apply amplitude limits, rate of change limits, and automatic decay constraints to the uncertainty disturbance correction term after the disturbance ends. S7. By integrating the basic energy management control command, deterministic correction term, and uncertain disturbance correction term, the final energy management control command is obtained. S8. Determine the target output of the engine, motor and / or power battery according to the final energy management control command.

[0007] Further, step S1 includes: S11. Obtain the current operating status information of the vehicle, including vehicle speed, acceleration, drive power demand and power battery state of charge (SOC); S12. The acquired vehicle current operating status information is used to construct a current state vector, which is represented as:

[0008] in, For vehicle speed, For acceleration, For the current power requirements of the whole vehicle, This refers to the state of charge of the power battery. S13. The stable basic strategy generates basic energy management control commands based on the current state vector:

[0009] in, It is a type of rule-based strategy, optimization strategy, and reinforcement learning strategy; a basic energy management and control instruction. It can be a scalar or a vector, specifically representing one or more of the following: engine target power, engine target torque, motor target power, motor target torque, battery charging / discharging power, power distribution ratio, torque distribution ratio, or engine start / stop command.

[0010] Further, step S2 includes: S21. In the offline phase, acquire historical vehicle operation data, including driving condition characteristics, road slope and key variables related to vehicle load conditions, to form the original data sequence. S22. The parameters representing the driving condition characteristics are as follows:

[0011] in, For vehicle speed, For instantaneous acceleration, Average vehicle speed For maximum vehicle speed, For the standard deviation of vehicle speed, For the standard deviation of acceleration, This represents the percentage of idling time. , and These represent the percentages of low-speed, medium-speed, and high-speed ranges, respectively. S23. To extract the intrinsic features of the driving conditions, feature extraction is performed on the driving condition features in the original data sequence, and a feature compression method is used to map the high-dimensional driving condition features to a low-dimensional feature space:

[0012] in, This is the original driving condition characteristic sequence. Low-dimensional features; S24. In the low-dimensional feature space, density clustering method is used to perform cluster analysis on the samples to obtain several typical operating condition types, including urban congestion operating condition, urban smooth traffic operating condition, suburban operating condition and highway operating condition. S25, respectively, regarding road slope and vehicle load capacity Discretize and classify the slope levels. and load rating The formula is as follows:

[0013]

[0014] Among them, slope Grades 1, 2, and 3 represent flat roads, gentle slopes, and steep slopes, respectively; load capacity... Levels 1, 2, and 3 represent light load, standard load, and heavy load, respectively. S26. Construct a multi-dimensional working condition tag library by performing full permutations and combinations of basic working condition types, slope grades, and load grades:

[0015] S27, Labels for each operating condition mode Each energy management decision model is constructed separately, and independent decision networks are trained based on historical data samples belonging to the same operating condition mode label.

[0016] in, Indicates the operating mode label. ; This indicates the decision-making strategy for the corresponding operating condition. Represents the current state vector of the vehicle; S28. Optimize and train each decision network according to the energy demand characteristics of different operating conditions, thereby improving the adaptability and control accuracy of the control strategy under different operating conditions, and forming a strategy set corresponding to the operating condition label library:

[0017] S29. During the online operation phase, acquire vehicle operation data over a continuous period of time and construct a real-time data sequence based on a sliding time window:

[0018] in, Road slope; For vehicle load capacity; S210. Driving condition characteristics in real-time data sequences Dimensionality reduction is performed to obtain real-time low-dimensional feature vectors. ; S211. Input the real-time low-dimensional feature vector into the working condition recognition model, and use the nearest neighbor template matching algorithm to calculate its relationship with the offline-built multi-dimensional working condition label library. The Euclidean distance between the reference feature vectors of each basic working condition type is used to select the category with the smallest distance to determine the current basic working condition type. :

[0019] S212, Real-time road slope and vehicle load capacity The slope grades are obtained by discretization using a grading function consistent with the offline stage. With load rating ; S213. Combine the current basic working condition type obtained from S211 and S212 with the slope level and load level to obtain the current working condition mode label. ; S214. After obtaining the current operating condition mode label, select the decision network corresponding to the current operating condition mode label from the policy set. The current state vector is then input into the decision network to generate the current mode correction term:

[0020] Further, step S3 includes: S31. Obtain the predicted distance ahead of the vehicle. Prior information about roads within the scope, including road slope, road curvature, and road segment type; S32, Predict the distance Classified by equidistant dispersion method T There are 1 sampling points, and the look-ahead distance for each sampling point is:

[0021] in, ; S33. Mapping the original road prior information from a spatial distribution to a sequence of forward-looking information with a temporal order. :

[0022]

[0023] in, Indicates the road slope; Indicates the curvature of the road; The characteristics of the road segment ahead are represented by numerical encoding or vectorization of the road segment type; S34. Encode the prospective information sequence in the time dimension, assigning each sampling point with corresponding temporal position information to obtain the prospective feature sequence. :

[0024] S35. Combine the encoded information with the look-ahead information sequence. Feature fusion is performed to obtain a complete look-ahead feature sequence. This enables the prospective feature sequence to effectively characterize the changing trends of future road conditions:

[0025] in, This represents a time series encoding function, implemented using absolute time step encoding, specifically as follows: , This is the time-series index of the current sampling point; This represents the total number of sampling points in the look-ahead sequence. S36. Select the vehicle state vector as follows:

[0026] in, Indicates the state of charge of the power battery; Indicates the current vehicle speed; Indicates the required power; S37. Transfer the vehicle state vector With prospective feature sequences Input attention calculation module to calculate the importance score of look-ahead information at each time step:

[0027] in, Indicates the first The attention score corresponding to each look-ahead feature is used to characterize the degree of correlation between the look-ahead information at that time step and the current vehicle state; Represents the current vehicle state vector; Indicates the first One forward-looking feature vector; , Both represent weight matrices; Represents the weight vector; Indicates the bias term; Represents a nonlinear activation function; S38. Normalize the attention score using the Softmax function to obtain the attention weights at each time step:

[0028] in, Indicates the first The attention weights corresponding to each time step are used to characterize the importance of the forward-looking information to the current decision. S39. Weight the look-ahead features according to the attention weights to obtain the fused look-ahead information representation:

[0029] in, This represents a fusion of forward-looking information, used to comprehensively reflect the impact of future road conditions on current control decisions. S310, Integrating and representing forward-looking information Current vehicle operating status information Input to parameterized nonlinear mapping function Generate forward attention correction sub-items:

[0030] in, This indicates the forward attention correction sub-item, used to reflect the impact of future road conditions on current control decisions; Represents the current vehicle state vector; This represents the fusion of forward-looking information. This represents a parameterized nonlinear mapping function used to establish the mapping relationship between vehicle state, look-ahead information, and control corrections.

[0031] Further, step S4 includes: S41. Add the current mode correction sub-item to the look-ahead attention engine power correction sub-item to obtain the deterministic correction term:

[0032] S42. To avoid power surges caused by deterministic correction terms, a limiting condition is applied to the deterministic correction terms:

[0033] in, and These are the lower and upper limits for the deterministic correction term, respectively.

[0034] Further, step S5 includes: S51. Construct the perturbation feature vector as follows:

[0035] in, For acceleration fluctuations, For pedal opening fluctuation, For power demand fluctuations, For the stop-and-go frequency, This refers to the behavioral deviation. S52. Calculate the deviation of the perturbation eigenvector using the following formula:

[0036] in, and These are the inverse matrices of the current model's reference mean and covariance matrix, respectively. S53. Construct the CUSUM change point detection statistics as follows:

[0037] When satisfied When an uncertainty disturbance occurs, it is determined that such disturbance has occurred; among which, Indicates a preset threshold; Indicates the drift compensation term; S54. An uncertainty perturbation correction term is generated from the perturbation correction model. The perturbation correction model adopts a reinforcement learning residual strategy and achieves online adaptation through meta-learning initialization and local fast update. During the online update process, catastrophic forgetting is suppressed through parameter importance constraints or elastic weight consolidation mechanisms. S55. Use historical data from multiple operating conditions in the cloud to perform meta-learning training on the disturbance correction model to obtain meta-initialization parameters. This endows the model with the ability to quickly generalize and adapt to unknown perturbation distributions; S56. During actual operation on the vehicle, initialize the model parameters. ; S57. When faced with specific uncertain disturbances, perform gradient iterations in one or a few steps based on local real-time data to quickly fine-tune the parameters to the optimal parameters under the working condition, so as to adapt to new scenario disturbances with less data and faster speed. S58. Calculate the adaptive learning rate using the following formula:

[0038] in, The initial learning rate, The attenuation coefficient is... This represents the cumulative number of updates. S59. Update the parameters of the perturbation correction model. The update formula is as follows:

[0039] in, express The current model parameters of the time-perturbation correction module; This indicates the new perturbation model parameters at the next time step after the update; The learning rate; For loss function, ,in, This represents the task loss function. , Instantaneous fuel consumption rate For reference SOC, This indicates a regularization term that reinforces the elastic weights. , For parameter importance weights, For reference parameters, This is the constraint strength coefficient.

[0040] Further, step S6 includes: S61. Apply amplitude constraints to the uncertainty disturbance correction term as follows:

[0041] in, For amplitude constraints; Adjust the upper limit to account for uncertainty disturbances; Lower bound adjusted for uncertainty disturbances; S62. Apply a rate-of-change constraint to the uncertainty disturbance correction term, as follows:

[0042] in, Constrained by rate of change; The upper limit of the rate of change is adjusted for uncertainty disturbances; S63. When the disturbance detection returns to the normal range, the uncertainty disturbance correction term automatically decays according to the following rule:

[0043] in, This is the attenuation coefficient.

[0044] Further, step S7 includes: S71. Define the disturbance gating factor ,as follows:

[0045] S72. The basic energy management control command, deterministic correction term, and uncertain disturbance correction term are fused using an incremental superposition method to obtain the final energy management control command:

[0046] in, This indicates basic energy management control commands. Represents a deterministic correction term. This represents the uncertainty disturbance correction term.

[0047] Further, step S8 includes: According to the final energy management control command Based on the current vehicle operating status, power system component constraints, and energy distribution principles, the target output of the engine, motor, and / or power battery is further determined.

[0048] This invention also provides a continuously self-evolving vehicle energy management system based on the aforementioned method for continuous self-evolving vehicle energy management based on the coordination of deterministic operating conditions and uncertain disturbances. The system includes: a vehicle state acquisition and basic strategy module, a current pattern recognition and correction module, a look-ahead information processing and correction module, a deterministic correction and fusion module, an uncertain disturbance detection and correction module, a disturbance constraint module, a final control quantity generation module, and a target output determination module, wherein: The vehicle status acquisition and basic strategy module is used to acquire the current operating status information of the vehicle and generate basic energy management control commands from the stable basic strategy. The current mode recognition and correction module is used to identify the current deterministic operating condition mode based on the vehicle's recent operating data, and generate a current mode correction sub-item based on the current deterministic operating condition mode. The forward information processing and correction module is used to acquire prior road information within a preset prediction distance range ahead of the vehicle, construct a forward feature sequence, and generate a forward attention correction sub-item based on the current vehicle state information and the forward feature sequence. The deterministic correction fusion module is used to fuse the current mode correction sub-item and the look-ahead attention correction sub-item to obtain a deterministic correction item; The uncertainty disturbance detection and correction module is used to detect whether an uncertainty disturbance has occurred based on the degree of deviation of the vehicle's current operating behavior from the current deterministic operating condition mode reference distribution, and to generate an uncertainty disturbance correction item when the triggering condition is met. The disturbance constraint module is used to apply amplitude limits, rate of change limits, and automatic decay constraints to the uncertainty disturbance correction term after the disturbance ends. The final control quantity generation module is used to integrate the basic energy management control command, the deterministic correction term, and the uncertainty disturbance correction term to obtain the final energy management control command; The target output determination module is used to determine the target output of the engine, motor and / or power battery according to the final energy management control command.

[0049] Compared with the prior art, the present invention has the following advantages: 1. The continuous self-evolving vehicle energy management method based on the coordination of deterministic operating conditions and uncertain disturbances provided by the present invention constructs a stable basic strategy as the main controller, and superimposes deterministic correction terms and uncertain disturbance correction terms on this basis, realizing hierarchical coordinated processing of current mode differences, future operating condition changes and sudden disturbances, and improving control robustness in complex road environments without destroying the stability of the basic strategy.

[0050] 2. The continuous self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances provided by this invention generates a current mode correction sub-item based on the current deterministic operating condition identification result, and constructs a forward attention correction sub-item by combining the prior information of the road ahead. This achieves full utilization of the deterministic operating condition rules and feedforward prediction of future operating condition change trends, effectively solving the control lag problem caused by the existing technology making decisions only based on the current state and lacking effective utilization of future road information.

[0051] 3. The continuous self-evolving vehicle energy management method based on the coordination of deterministic operating conditions and uncertain disturbances provided by this invention introduces an uncertain disturbance detection mechanism. Based on the degree of deviation of the vehicle's current operating behavior from the current deterministic operating condition mode reference distribution, it detects uncertain disturbances such as sudden changes in driving style, fluctuations in traffic conditions, or shifts in demand distribution in real time. It also ensures online coordination stability through amplitude constraints, rate of change constraints, and automatic decay mechanisms. This effectively solves the problem that existing technologies lack an effective online adaptive update mechanism and are difficult to maintain optimal performance in long-term operation.

[0052] 4. The continuous self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances provided by this invention realizes the disturbance correction module by combining reinforcement learning residual strategy and meta-learning rapid adaptation strategy. In the online update process, catastrophic forgetting is suppressed by parameter importance constraints or elastic weight consolidation mechanism, realizing the continuous self-evolution capability of adapting to new scenario disturbances with "less data and faster speed". It effectively solves the problems of existing technologies failing to distinguish between deterministic operating condition changes and uncertain disturbances, and having chaotic control structures and insufficient interpretability.

[0053] 5. The continuous self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances provided by this invention achieves dynamic weight allocation between different control components through the synergistic fusion of deterministic correction terms and uncertain disturbance correction terms, using incremental superposition or gating modulation. This realizes the hierarchical decoupling processing of deterministic operating condition laws and uncertain disturbances, thereby improving fuel economy, SOC maintenance capability, control smoothness, and long-term operational robustness.

[0054] In summary, by applying the technical solution of this invention, the problems in existing technologies—such as control lag due to decision-making based solely on the current state, the inability of a single strategy model to adapt to various typical operating conditions, the lack of an effective online adaptive update mechanism, and the inability to distinguish between deterministic operating condition changes and uncertain disturbances leading to chaotic control structures—can all be effectively solved by the hierarchical and collaborative architecture of stable basic strategy, deterministic correction term, and uncertain disturbance correction term proposed in this invention. Therefore, the technical solution of this invention solves the problems of insufficient utilization of future information, poor adaptability to operating conditions, and lack of continuous adaptive capability in existing technologies.

[0055] Based on the above reasons, this invention can be widely applied in the fields of energy management and control of hybrid electric vehicles, plug-in hybrid electric vehicles, and multi-energy drive system vehicles. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart of the method of the present invention.

[0058] Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0059] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0060] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product or device.

[0061] This embodiment takes a P2 hybrid electric vehicle as an example and illustrates the control of the engine under target operating conditions.

[0062] like Figure 1 As shown, this invention provides a continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances, comprising: S1. Obtain the current operating status information of the vehicle and generate basic energy management control commands from the stable basic strategy; S2. Identify the current deterministic operating condition mode based on the vehicle's recent operating data, and generate a current mode correction sub-item based on the current deterministic operating condition mode; S3. Obtain prior road information within a preset prediction distance range ahead of the vehicle, construct a forward-looking feature sequence, and generate a forward-looking attention correction sub-item based on the current vehicle state information and the forward-looking feature sequence; S4. Merge the current mode correction sub-item and the forward attention correction sub-item to obtain the deterministic correction item; S5. Detect whether an uncertainty disturbance has occurred based on the degree of deviation of the vehicle's current operating behavior from the current deterministic operating condition mode reference distribution, and generate an uncertainty disturbance correction term when the triggering condition is met; S6. Apply amplitude limits, rate of change limits, and automatic decay constraints to the uncertainty disturbance correction term after the disturbance ends. S7. By integrating the basic energy management control command, deterministic correction term, and uncertain disturbance correction term, the final energy management control command is obtained. S8. Determine the target output of the engine, motor and / or power battery according to the final energy management control command.

[0063] In a specific implementation, as a preferred embodiment of the present invention, step S1 includes: S11. Obtain the current operating status information of the vehicle, including vehicle speed, acceleration, drive power demand and power battery state of charge (SOC); S12. The acquired vehicle current operating status information is used to construct a current state vector, which is represented as:

[0064] in, For vehicle speed, For acceleration, For the current power requirements of the whole vehicle, This refers to the state of charge of the power battery. S13. The stable basic strategy generates basic energy management control commands based on the current state vector:

[0065] in, It is a type of rule-based strategy, optimization strategy, and reinforcement learning strategy; a basic energy management and control instruction. It can be a scalar or a vector, specifically representing one or more of the following: engine target power, engine target torque, motor target power, motor target torque, battery charging / discharging power, power distribution ratio, torque distribution ratio, or engine start / stop command.

[0066] In this embodiment, the stable base strategy adopts the parameterized equivalent cost minimization strategy ECMS, which specifically includes: Within the feasible power set, a candidate engine power set is constructed, and the motor power corresponding to each candidate scheme is determined according to the power balance relationship, as follows:

[0067] in, Indicates engine power; The equivalent cost function for each candidate solution is calculated as follows:

[0068] in, Indicates engine power as Instantaneous fuel consumption rate at that time As an equivalent factor, For charging and discharging power, This is the SOC deviation weighting coefficient. This represents the actual state of charge (SOC) of the power battery. This serves as a reference target value for the state of charge of the power battery. The candidate engine power that minimizes the equivalent cost function is selected as the target power of the base engine.

[0069] The target power of the base engine is used as the base energy management control command, that is... .

[0070] In a specific implementation, as a preferred embodiment of the present invention, step S2 includes: S21. In the offline phase, acquire historical vehicle operation data, including driving condition characteristics, road slope and key variables related to vehicle load conditions, to form the original data sequence. S22. The parameters representing the driving condition characteristics are as follows:

[0071] in, For vehicle speed, For instantaneous acceleration, Average vehicle speed For maximum vehicle speed, For the standard deviation of vehicle speed, For the standard deviation of acceleration, This represents the percentage of idling time. , and These represent the percentages of low-speed, medium-speed, and high-speed ranges, respectively. S23. To extract the intrinsic features of the driving conditions, feature extraction is performed on the driving condition features in the original data sequence, and a feature compression method is used to map the high-dimensional driving condition features to a low-dimensional feature space:

[0072] in, This is the original driving condition characteristic sequence. Low-dimensional features; S24. In the low-dimensional feature space, density clustering method is used to perform cluster analysis on the samples to obtain several typical operating condition types, including urban congestion operating condition, urban smooth traffic operating condition, suburban operating condition and highway operating condition. S25, respectively, regarding road slope and vehicle load capacity Discretize and classify the slope levels. and load rating The formula is as follows:

[0073]

[0074] Among them, slope Grades 1, 2, and 3 represent flat roads, gentle slopes, and steep slopes, respectively; load capacity... Levels 1, 2, and 3 represent light load, standard load, and heavy load, respectively. S26. Construct a multi-dimensional working condition tag library by performing full permutations and combinations of basic working condition types, slope grades, and load grades:

[0075] in, Indicates the type of driving condition; Indicates the slope grade; Indicates the load capacity rating; S27, Labels for each operating condition mode Each energy management decision model is constructed separately, and independent decision networks are trained based on historical data samples belonging to the same operating condition mode label.

[0076] in, Indicates the operating mode label. ; This indicates the decision-making strategy for the corresponding operating condition. Represents the current state vector of the vehicle; S28. Optimize and train each decision network according to the energy demand characteristics of different operating conditions, thereby improving the adaptability and control accuracy of the control strategy under different operating conditions, and forming a strategy set corresponding to the operating condition label library:

[0077] S29. During the online operation phase, acquire vehicle operation data over a continuous period of time and construct a real-time data sequence based on a sliding time window:

[0078] in, Road slope; For vehicle load capacity; S210. Driving condition characteristics in real-time data sequences Dimensionality reduction is performed to obtain real-time low-dimensional feature vectors. ; S211. Input the real-time low-dimensional feature vector into the working condition recognition model, and use the nearest neighbor template matching algorithm to calculate its relationship with the offline-built multi-dimensional working condition label library. The Euclidean distance between the reference feature vectors of each basic working condition type is used to select the category with the smallest distance to determine the current basic working condition type. :

[0079] S212, Real-time road slope and vehicle load capacity The slope grades are obtained by discretization using a grading function consistent with the offline stage. With load rating ; S213. Combine the current basic working condition type obtained from S211 and S212 with the slope level and load level to obtain the current working condition mode label. ; S214. After obtaining the current operating condition mode label, select the decision network corresponding to the current operating condition mode label from the policy set. The current state vector is then input into the decision network to generate the current mode correction term: .

[0080] In a specific implementation, as a preferred embodiment of the present invention, step S3 includes: S31. Obtain the predicted distance ahead of the vehicle. Prior road information within the scope includes road slope, road curvature, and road segment type; in other embodiments, it may also include speed limit information, traffic signal information, or other relevant features.

[0081] S32, Predict the distance Classified by equidistant dispersion method There are 1 sampling points, and the look-ahead distance for each sampling point is:

[0082] in, ; S33. Mapping the original road prior information from a spatial distribution to a sequence of forward-looking information with a temporal order. :

[0083]

[0084] in, Indicates the road slope; Indicates the curvature of the road; The characteristics of the road segment ahead are represented by numerical encoding or vectorization of the road segment type; S34. Encode the prospective information sequence in the time dimension, assigning each sampling point with corresponding temporal position information to obtain the prospective feature sequence. :

[0085] S35. Combine the encoded information with the look-ahead information sequence. Feature fusion is performed to obtain a complete look-ahead feature sequence. This enables the prospective feature sequence to effectively characterize the changing trends of future road conditions:

[0086] in, This represents a time series encoding function, implemented using absolute time step encoding, specifically as follows: , This is the time-series index of the current sampling point; This represents the total number of sampling points in the look-ahead sequence. S36. Select the vehicle state vector as follows:

[0087] in, Indicates the state of charge of the power battery; Indicates the current vehicle speed; The state variable represents the required power; in this embodiment, the state variable can comprehensively characterize the vehicle's current energy state and load demand, providing a basis for subsequent energy management and control. In other embodiments, the vehicle state vector can also be expanded to include variables such as motor speed, engine speed, or acceleration.

[0088] S37. Transfer the vehicle state vector With prospective feature sequences Input attention calculation module to calculate the importance score of look-ahead information at each time step:

[0089] in, Indicates the first The attention score corresponding to each look-ahead feature is used to characterize the degree of correlation between the look-ahead information at that time step and the current vehicle state; Represents the current vehicle state vector; Indicates the first One forward-looking feature vector; , Both represent weight matrices; Represents the weight vector; Indicates the bias term; Represents a nonlinear activation function; S38. Normalize the attention score using the Softmax function to obtain the attention weights at each time step:

[0090] in, Indicates the first The attention weights corresponding to each time step are used to characterize the importance of the forward-looking information to the current decision. S39. Weight the look-ahead features according to the attention weights to obtain the fused look-ahead information representation:

[0091] in, This represents a fusion of forward-looking information, used to comprehensively reflect the impact of future road conditions on current control decisions. S310, Integrating and representing forward-looking information Current vehicle operating status information Input to parameterized nonlinear mapping function Generate forward attention correction sub-items:

[0092] in, This indicates the forward attention correction sub-item, used to reflect the impact of future road conditions on current control decisions; Represents the current vehicle state vector; This represents the fusion of forward-looking information. This represents a parameterized nonlinear mapping function used to establish the mapping relationship between vehicle state, look-ahead information, and control corrections.

[0093] In this embodiment, the nonlinear mapping function A feedforward neural network is used to construct a nonlinear mapping relationship between the current vehicle state and forward-looking information, specifically including: The current vehicle state vector is concatenated with the look-ahead fusion representation to obtain the input vector of the mapping function. :

[0094] Input the mapping function into the vector Input to a parameterized nonlinear mapping function for nonlinear transformation:

[0095] in, This represents the intermediate representation after nonlinear transformation by the activation function. Represents a non-linear activation function. Represents the weight matrix. Represents the bias vector; The intermediate representation after nonlinear transformation by activation function The input is subjected to a nonlinear transformation via a parameterized nonlinear mapping function to generate a forward-looking attention engine power correction sub-term:

[0096] in, This indicates the generated forward-looking attention engine power correction sub-item, which is used to adjust the basic control strategy and characterizes the feedforward correction requirement of the current basic engine power due to future operating condition transition trends. Represents the weight matrix. This represents the bias vector.

[0097] In a specific implementation, as a preferred embodiment of the present invention, step S4 includes: S41. Add the current mode correction sub-item to the look-ahead attention engine power correction sub-item to obtain the deterministic correction term:

[0098] S42. To avoid power surges caused by deterministic correction terms, a limiting condition is applied to the deterministic correction terms:

[0099] in, and These are the lower and upper limits of the deterministic engine power correction term, respectively.

[0100] In a specific implementation, as a preferred embodiment of the present invention, step S5 includes: S51. Construct the perturbation feature vector as follows:

[0101] in, For acceleration fluctuations, For pedal opening fluctuation, For power demand fluctuations, For the stop-and-go frequency, This refers to the behavioral deviation. S52. Calculate the deviation of the perturbation eigenvector using the following formula:

[0102] in, and These are the inverse matrices of the current model's reference mean and covariance matrix, respectively. S53. Construct the CUSUM change point detection statistics as follows:

[0103] When satisfied When an uncertainty disturbance occurs, it is determined that such disturbance has occurred; among which, Indicates a preset threshold; Indicates the drift compensation term; S54. An uncertainty perturbation correction term is generated from the perturbation correction model. The perturbation correction model adopts a reinforcement learning residual strategy and achieves online adaptation through meta-learning initialization and local fast update. During the online update process, catastrophic forgetting is suppressed through parameter importance constraints or elastic weight consolidation mechanisms. In this embodiment, the perturbation correction model adopts the TD3 residual strategy and outputs an uncertainty perturbation correction term:

[0104] in, This is a perturbation correction strategy. Parameters can be updated online; S55. Use historical data from multiple operating conditions in the cloud to perform meta-learning training on the disturbance correction model to obtain meta-initialization parameters. This endows the model with the ability to quickly generalize and adapt to unknown perturbation distributions; S56. During actual operation on the vehicle, initialize the model parameters. ; S57. When faced with specific uncertain disturbances, perform gradient iterations in one or a few steps based on local real-time data to quickly fine-tune the parameters to the optimal parameters under the working condition, so as to adapt to new scenario disturbances with less data and faster speed. S58. Calculate the adaptive learning rate using the following formula:

[0105] in, The initial learning rate, The attenuation coefficient is... This represents the cumulative number of updates. S59. Update the parameters of the perturbation correction model. The update formula is as follows:

[0106] in, express The current model parameters of the time-perturbation correction module; This indicates the new perturbation model parameters at the next time step after the update; The learning rate; For loss function, ,in, This represents the task loss function. , Instantaneous fuel consumption rate For reference SOC, This indicates a regularization term that reinforces the elastic weights. , For parameter importance weights, For reference parameters, This is the constraint strength coefficient.

[0107] In a specific implementation, as a preferred embodiment of the present invention, step S6 includes: S61. Apply amplitude constraints to the uncertainty disturbance correction term as follows:

[0108] in, For amplitude constraints; Upper limit for engine power disturbance correction; Lower limit for engine power disturbance correction; S62. Apply a rate-of-change constraint to the uncertainty disturbance correction term, as follows:

[0109] in, Constrained by rate of change; This is the upper limit of the rate of change for engine power disturbance correction; S63. When the disturbance detection returns to the normal range, the uncertainty disturbance correction term automatically decays according to the following rule:

[0110] in, This is the attenuation coefficient.

[0111] In a specific implementation, as a preferred embodiment of the present invention, step S7 includes: S71. Define the disturbance gating factor ,as follows:

[0112] S72. By using an incremental superposition method to fuse the basic energy management control command, deterministic correction term, and uncertainty disturbance correction term, the final engine target power is obtained as follows:

[0113] In a specific implementation, as a preferred embodiment of the present invention, step S8 includes: According to the final energy management control command Based on the current vehicle operating status, power system component constraints, and energy distribution principles, the target output of the engine, motor, and / or power battery is further determined.

[0114] In this embodiment, the modified control command is the engine power, so the target power of the motor is determined according to the power balance relationship:

[0115] in, This indicates the required power.

[0116] like Figure 2 As shown, this invention provides a continuously self-evolving vehicle energy management system based on the aforementioned method for continuous self-evolving vehicle energy management based on deterministic operating conditions and uncertain disturbances. The system includes: a vehicle state acquisition and basic strategy module, a current pattern recognition and correction module, a look-ahead information processing and correction module, a deterministic correction and fusion module, an uncertain disturbance detection and correction module, a disturbance constraint module, a final control quantity generation module, and a target output determination module, wherein: The vehicle status acquisition and basic strategy module is used to acquire the current operating status information of the vehicle and generate basic energy management control commands from the stable basic strategy. The current mode recognition and correction module is used to identify the current deterministic operating condition mode based on the vehicle's recent operating data, and generate a current mode correction sub-item based on the current deterministic operating condition mode. The forward information processing and correction module is used to acquire prior road information within a preset prediction distance range ahead of the vehicle, construct a forward feature sequence, and generate a forward attention correction sub-item based on the current vehicle state information and the forward feature sequence. The deterministic correction fusion module is used to fuse the current mode correction sub-item and the look-ahead attention correction sub-item to obtain a deterministic correction item; The uncertainty disturbance detection and correction module is used to detect whether an uncertainty disturbance has occurred based on the degree of deviation of the vehicle's current operating behavior from the current deterministic operating condition mode reference distribution, and to generate an uncertainty disturbance correction item when the triggering condition is met. The disturbance constraint module is used to apply amplitude limits, rate of change limits, and automatic decay constraints to the uncertainty disturbance correction term after the disturbance ends. The final control quantity generation module is used to integrate the basic energy management control command, the deterministic correction term, and the uncertainty disturbance correction term to obtain the final energy management control command; The target output determination module is used to determine the target output of the engine, motor and / or power battery according to the final energy management control command.

[0117] The embodiments of the present invention are described simply because they correspond to those in the embodiments above. For any similarities, please refer to the descriptions in the embodiments above, which will not be elaborated here.

[0118] In summary, this invention achieves layered and coordinated processing of current mode differences, future operating condition changes, and sudden disturbances without compromising the stability of the basic strategy. This improves fuel economy, SOC maintenance capability, control smoothness, and long-term operational robustness in complex road environments, and is applicable to energy management and control of hybrid electric vehicles, plug-in hybrid electric vehicles, and multi-energy drive system vehicles.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances, characterized in that, include: S1. Obtain the current operating status information of the vehicle and generate basic energy management control commands from the stable basic strategy; S2. Identify the current deterministic operating condition mode based on the vehicle's recent operating data, and generate a current mode correction sub-item based on the current deterministic operating condition mode; S3. Obtain prior road information within a preset prediction distance range ahead of the vehicle, construct a forward-looking feature sequence, and generate a forward-looking attention correction sub-item based on the current vehicle state information and the forward-looking feature sequence; S4. Merge the current mode correction sub-item and the forward attention correction sub-item to obtain the deterministic correction item; S5. Detect whether an uncertainty disturbance has occurred based on the degree of deviation of the vehicle's current operating behavior from the current deterministic operating condition mode reference distribution, and generate an uncertainty disturbance correction term when the triggering condition is met; S6. Apply amplitude limits, rate of change limits, and automatic decay constraints to the uncertainty disturbance correction term after the disturbance ends. S7. By integrating the basic energy management control command, deterministic correction term, and uncertain disturbance correction term, the final energy management control command is obtained. S8. Determine the target output of the engine, motor and / or power battery according to the final energy management control command.

2. The continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances as described in claim 1, characterized in that, Step S1 includes: S11. Obtain the current operating status information of the vehicle, including vehicle speed, acceleration, drive power demand and power battery state of charge (SOC); S12. The acquired vehicle current operating status information is used to construct a current state vector, which is represented as: in, For vehicle speed, For acceleration, For the current power requirements of the whole vehicle, This refers to the state of charge of the power battery. S13. The stable basic strategy generates basic energy management control commands based on the current state vector: in, It is a type of rule-based strategy, optimization strategy, and reinforcement learning strategy; a basic energy management and control instruction. It can be a scalar or a vector, specifically represented as one or more of the following: engine target power, engine target torque, motor target power, motor target torque, battery charging / discharging power, power distribution ratio, torque distribution ratio, or engine start / stop command; the stable basic strategy relies only on the current state to make decisions, maintains the stability of the structure and parameters during online operation, and does not perform rapid online reconfiguration updates.

3. The continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances as described in claim 1, characterized in that, Step S2 includes: S21. In the offline phase, acquire historical vehicle operation data, including driving condition characteristics, road slope and key variables related to vehicle load conditions, to form the original data sequence. S22. The parameters representing the driving condition characteristics are as follows: in, For vehicle speed, For instantaneous acceleration, Average vehicle speed For maximum vehicle speed, For the standard deviation of vehicle speed, For the standard deviation of acceleration, This represents the percentage of idling time. , and These represent the percentages of low-speed, medium-speed, and high-speed ranges, respectively. S23. To extract the intrinsic features of the driving conditions, feature extraction is performed on the driving condition features in the original data sequence, and a feature compression method is used to map the high-dimensional driving condition features to a low-dimensional feature space: in, This is the original driving condition characteristic sequence. Low-dimensional features; S24. In the low-dimensional feature space, density clustering method is used to perform cluster analysis on the samples to obtain several typical operating condition types, including urban congestion operating condition, urban smooth traffic operating condition, suburban operating condition and highway operating condition. S25, respectively, regarding road slope and vehicle load capacity Discretize and classify the slope levels. and load rating The formula is as follows: Among them, slope Grades 1, 2, and 3 represent flat roads, gentle slopes, and steep slopes, respectively; load capacity... Levels 1, 2, and 3 represent light load, standard load, and heavy load, respectively. S26. Construct a multi-dimensional working condition tag library by performing full permutations and combinations of basic working condition types, slope grades, and load grades: in, Indicates the type of driving condition; Indicates the slope grade; Indicates the load capacity rating; S27, Labels for each operating condition mode Each energy management decision model is constructed separately, and independent decision networks are trained based on historical data samples belonging to the same operating condition mode label. in, Indicates the operating mode label. ; This indicates the decision-making strategy for the corresponding operating condition. Represents the current state vector of the vehicle; S28. Optimize and train each decision network according to the energy demand characteristics of different operating conditions, thereby improving the adaptability and control accuracy of the control strategy under different operating conditions, and forming a strategy set corresponding to the operating condition label library: S29. During the online operation phase, acquire vehicle operation data over a continuous period of time and construct a real-time data sequence based on a sliding time window: in, Road slope; For vehicle load capacity; S210. Driving condition characteristics in real-time data sequences Dimensionality reduction is performed to obtain real-time low-dimensional feature vectors. ; S211. Input the real-time low-dimensional feature vector into the working condition recognition model, and use the nearest neighbor template matching algorithm to calculate its relationship with the offline-built multi-dimensional working condition label library. The Euclidean distance between the reference feature vectors of each basic working condition type is used to select the category with the smallest distance to determine the current basic working condition type. : S212, Real-time road slope and vehicle load capacity The slope grades are obtained by discretization using a grading function consistent with the offline stage. With load rating ; S213. Combine the current basic working condition type obtained from S211 and S212 with the slope level and load level to obtain the current working condition mode label. ; S214. After obtaining the current operating condition mode label, select the decision network corresponding to the current operating condition mode label from the policy set. The current state vector is then input into the decision network to generate the current mode correction term: 。 4. The continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances as described in claim 1, characterized in that, Step S3 includes: S31. Obtain the predicted distance ahead of the vehicle. Prior information about roads within the scope, including road slope, road curvature, and road segment type; S32, Predict the distance Classified by equidistant dispersion method There are 1 sampling points, and the look-ahead distance for each sampling point is: in, ; S33. Mapping the original road prior information from a spatial distribution to a sequence of forward-looking information with a temporal order. : in, Indicates the road slope; Indicates the curvature of the road; The characteristics of the road segment ahead are represented by numerical encoding or vectorization of the road segment type; S34. Encode the prospective information sequence in the time dimension, assigning each sampling point with corresponding temporal position information to obtain the prospective feature sequence. : S35. Combine the encoded information with the look-ahead information sequence. Feature fusion is performed to obtain a complete look-ahead feature sequence. This enables the prospective feature sequence to effectively characterize the changing trends of future road conditions: in, This represents a time series encoding function, implemented using absolute time step encoding, specifically as follows: , This is the time-series index of the current sampling point; This represents the total number of sampling points in the look-ahead sequence. S36. Select the vehicle state vector as follows: in, Indicates the state of charge of the power battery; Indicates the current vehicle speed; Indicates the required power; S37. Transfer the vehicle state vector With prospective feature sequences Input attention calculation module to calculate the importance score of look-ahead information at each time step: in, Indicates the first The attention score corresponding to each look-ahead feature is used to characterize the degree of correlation between the look-ahead information at that time step and the current vehicle state; Represents the current vehicle state vector; Indicates the first One forward-looking feature vector; , Both represent weight matrices; Represents the weight vector; Indicates the bias term; Represents a nonlinear activation function; S38. Normalize the attention score using the Softmax function to obtain the attention weights at each time step: in, Indicates the first The attention weights corresponding to each time step are used to characterize the importance of the forward-looking information to the current decision. S39. Weight the look-ahead features according to the attention weights to obtain the fused look-ahead information representation: in, This represents a fusion of forward-looking information, used to comprehensively reflect the impact of future road conditions on current control decisions. S310, Integrating and representing forward-looking information Current vehicle operating status information Input to parameterized nonlinear mapping function Generate forward attention correction sub-items: in, This indicates the forward attention correction sub-item, used to reflect the impact of future road conditions on current control decisions; Represents the current vehicle state vector; This represents the fusion of forward-looking information. This represents a parameterized nonlinear mapping function used to establish the mapping relationship between vehicle state, look-ahead information, and control corrections.

5. The continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances as described in claim 1, characterized in that, Step S4 includes: S41. Add the current mode correction sub-item to the look-ahead attention engine power correction sub-item to obtain the deterministic correction term: S42. To avoid power surges caused by deterministic correction terms, a limiting condition is applied to the deterministic correction terms: in, and These are the lower and upper limits for the deterministic correction term, respectively.

6. The continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances as described in claim 1, characterized in that, Step S5 includes: S51. Construct the perturbation feature vector as follows: in, For acceleration fluctuations, For pedal opening fluctuation, For power demand fluctuations, For the stop-and-go frequency, This refers to the behavioral deviation. S52. Calculate the deviation of the perturbation eigenvector using the following formula: in, and These are the inverse matrices of the current model's reference mean and covariance matrix, respectively. S53. Construct the CUSUM change point detection statistics as follows: When satisfied When an uncertainty disturbance occurs, it is determined that such disturbance has occurred; among which, Indicates a preset threshold; Indicates the drift compensation term; S54. An uncertainty perturbation correction term is generated from the perturbation correction model. The perturbation correction model adopts a reinforcement learning residual strategy and achieves online adaptation through meta-learning initialization and local fast update. During the online update process, catastrophic forgetting is suppressed through parameter importance constraints or elastic weight consolidation mechanisms. S55. Use historical data from multiple operating conditions in the cloud to perform meta-learning training on the disturbance correction model to obtain meta-initialization parameters. This endows the model with the ability to quickly generalize and adapt to unknown perturbation distributions; S56. During actual operation on the vehicle, initialize the model parameters. ; S57. When faced with specific uncertain disturbances, perform gradient iterations in one or a few steps based on local real-time data to quickly fine-tune the parameters to the optimal parameters under the working condition, so as to adapt to new scenario disturbances with less data and faster speed. S58. Calculate the adaptive learning rate using the following formula: in, The initial learning rate, The attenuation coefficient is... This represents the cumulative number of updates. S59. Update the parameters of the perturbation correction model. The update formula is as follows: in, express The current model parameters of the time-perturbation correction module; This indicates the new perturbation model parameters at the next time step after the update; The learning rate; For loss function, ,in, This represents the task loss function. , Instantaneous fuel consumption rate For reference SOC, This indicates a regularization term that reinforces the elastic weights. , For parameter importance weights, For reference parameters, This is the constraint strength coefficient.

7. The continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances as described in claim 1, characterized in that, Step S6 includes: S61. Apply amplitude constraints to the uncertainty disturbance correction term as follows: in, For amplitude constraints; Adjust the upper limit to account for uncertainty disturbances; Lower bound adjusted for uncertainty disturbances; S62. Apply a rate-of-change constraint to the uncertainty disturbance correction term, as follows: in, Constrained by rate of change; The upper limit of the rate of change is adjusted for uncertainty disturbances; S63. When the disturbance detection returns to the normal range, the uncertainty disturbance correction term automatically decays according to the following rule: in, This is the attenuation coefficient.

8. The continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances as described in claim 1, characterized in that, Step S7 includes: S71. Define the disturbance gating factor ,as follows: S72. By integrating the basic energy management control command, deterministic correction term, and uncertainty disturbance correction term, the final energy management control command is obtained: in, This indicates basic energy management control commands. Represents a deterministic correction term. This represents the uncertainty disturbance correction term.

9. The continuously self-evolving vehicle energy management method based on the synergy of deterministic operating conditions and uncertain disturbances as described in claim 1, characterized in that, Step S8 includes: According to the final energy management control command Based on the current vehicle operating status, power system component constraints, and energy distribution principles, the target output of the engine, motor, and / or power battery is further determined.

10. A continuously self-evolving vehicle energy management system based on the co-existence of deterministic operating conditions and uncertain disturbances, implemented according to any one of claims 1-9, characterized in that, include: The system comprises the following modules: vehicle status acquisition and basic strategy module, current pattern recognition and correction module, look-ahead information processing and correction module, deterministic correction and fusion module, uncertainty disturbance detection and correction module, disturbance constraint module, final control quantity generation module, and target output determination module. The vehicle status acquisition and basic strategy module is used to acquire the current operating status information of the vehicle and generate basic energy management control commands from the stable basic strategy. The current mode recognition and correction module is used to identify the current deterministic operating condition mode based on the vehicle's recent operating data, and generate a current mode correction sub-item based on the current deterministic operating condition mode. The forward information processing and correction module is used to acquire prior road information within a preset prediction distance range ahead of the vehicle, construct a forward feature sequence, and generate a forward attention correction sub-item based on the current vehicle state information and the forward feature sequence. The deterministic correction fusion module is used to fuse the current mode correction sub-item and the look-ahead attention correction sub-item to obtain a deterministic correction item; The uncertainty disturbance detection and correction module is used to detect whether an uncertainty disturbance has occurred based on the degree of deviation of the vehicle's current operating behavior from the current deterministic operating condition mode reference distribution, and to generate an uncertainty disturbance correction item when the triggering condition is met. The disturbance constraint module is used to apply amplitude limits, rate of change limits, and automatic decay constraints to the uncertainty disturbance correction term after the disturbance ends. The final control quantity generation module is used to integrate the basic energy management control command, the deterministic correction term, and the uncertainty disturbance correction term to obtain the final energy management control command; The target output determination module is used to determine the target output of the engine, motor and / or power battery according to the final energy management control command.

Images