Energy management method and system for a vehicle

Abstract

The application discloses an energy management method and system of a vehicle, and relates to the technical field of new energy vehicle power control. In view of the problem that the existing energy management strategy is prone to causing power supply and power limitation of the vehicle due to road condition prediction error, the method obtains future driving prediction information and a prediction confidence index in the framework based on model prediction control. When optimizing the objective function, the system establishes a negative correlation dynamic adjustment relationship between the penalty weight of the penalty of the state parameter of the energy storage unit deviating from the preset track and the size of the prediction confidence index. When the road condition uncertainty increases and the prediction confidence decreases, the system automatically increases the power supply weight, and drives the optimization solving process to deviate to the instruction range increasing system to generate power in advance for defensive power reserve. The application supports deployment in the architecture containing the pluggable range increasing device, completely resolves the power supply anchoring risk in the bottom algorithm, and realizes the extreme optimization of global energy efficiency in multiple scenes.

Description

Technical Field

[0001] This invention relates to the fields of new energy vehicle powertrain, vehicle energy management system and intelligent control algorithm technology, and in particular to a vehicle energy management method and system. Background Technology

[0002] As the global automotive industry continues its shift towards new energy vehicles, the market penetration rate of electric vehicles is constantly increasing. To overcome the "range anxiety" pain point of pure electric vehicles during long-distance travel, plug-in hybrid electric vehicles (PHEVs) and range-extended electric vehicles (REEVs) have been widely adopted as parallel technological approaches. The vehicle's energy management strategy (EMS) is the core brain that determines the overall energy consumption level and power stability of these hybrid vehicles.

[0003] Through in-depth research and analysis of existing technologies, the following two major technical pain points were identified in current vehicle energy management technologies:

[0004] First, at the algorithmic control level, advanced energy management systems are increasingly adopting Model Predictive Control (MPC) algorithms based on predictive driving. MPC algorithms rely on predictions of future road conditions (such as vehicle speed and gradient) to plan the power distribution between the engine and battery in advance. However, real-world traffic environments are extremely complex and unpredictable, leading to frequent deviations in the algorithm's predictions. If environmental predictions are inaccurate (such as a sudden drop in speed due to congestion), traditional MPC algorithms, if still relying on optimistic planning, can excessively deplete battery power, easily causing the battery's state of charge to fall below the safe minimum, resulting in severely limited vehicle power or even power interruption (battery failure). Current technologies generally lack adaptive fault-tolerant underlying mechanisms to address "predictive uncertainty."

[0005] Secondly, at the hardware structure level, the range-extending systems (including internal combustion engines, generators, and fuel tanks) of the vast majority of existing models are fixed installations. Since most users spend over 90% of their daily driving time on short-distance urban commutes, pure electric drive is sufficient. This means that the vehicle must carry the heavy "dead weight" of the range-extending system for extended periods during daily driving. This not only violates the original intention of energy conservation and emission reduction but also occupies a significant amount of interior storage space. Summary of the Invention

[0006] To address the shortcomings of existing energy management strategies, such as the lack of a fault-tolerant mechanism for predictive uncertainties, which can easily lead to power outages and breakdowns, and the high daily energy consumption due to the fixed weight of hardware, this invention provides a vehicle energy management method and system. The aim is to completely avoid the risk of power outages under unforeseen operating conditions by introducing a negative correlation game mechanism between prediction confidence and risk penalty weights. Simultaneously, it is compatible with pluggable range extender hardware, achieving a perfect fusion of daily weight reduction in pure electric mode and long-distance efficient hybrid operation.

[0007] To solve the above-mentioned technical problems, in a first aspect, the present invention provides the following technical solution:

[0008] A vehicle energy management method is applied to a vehicle including a drive system, an energy storage unit electrically connected to the drive system, and a range extender system for charging the energy storage unit. The method includes: acquiring predictive information of future driving conditions and a prediction confidence index characterizing the reliability of the predictive information; determining control commands for the range extender system by optimizing an objective function in each control time domain based on a model predictive control method; wherein the objective function includes at least a penalty term for penalizing deviations of the state parameters of the energy storage unit from a preset reference trajectory, and the penalty weight of the penalty term is negatively correlated with the magnitude of the prediction confidence index; and driving the prime mover in the range extender system to adjust its output power and / or controlling the charging and discharging power of the energy storage unit based on the optimized control commands, so as to minimize the overall energy consumption of the vehicle while maintaining the energy safety threshold of the energy storage unit.

[0009] By innovatively introducing a "prediction confidence index" into the underlying objective function of Model Predictive Control (MPC) and establishing a "negative correlation" between it and the penalty weight for deviation of state parameters, this invention achieves risk game theory at the algorithmic level. When the prediction is highly reliable, the system pursues extremely low energy consumption; when facing complex and unknown road conditions (low confidence), the system automatically increases the power conservation weight, favoring a conservative and defensive approach. Furthermore, by directly and strongly binding the abstract optimization results to the physical actions of the driving prime mover and the control energy storage unit, the effective execution of the control method on the physical system is ensured, completely eliminating the risk of power failure caused by blindly consuming electricity.

[0010] In a preferred embodiment of the present invention, the penalty weight of the penalty term is obtained by multiplying a basic penalty weight by a dynamic adjustment factor. This dynamic adjustment factor increases linearly or non-linearly as the prediction confidence index decreases, so that when the prediction confidence index decreases, the optimization process of the objective function tends to suppress the decline in the state parameters of the energy storage unit. This feature enables a confidence-based smooth transition control logic, avoiding abrupt changes in system operating conditions.

[0011] In a preferred embodiment of the present invention, the penalty term is a quadratic penalty term of the state parameters, and the corresponding penalty weight is determined through a specific functional relationship that includes a risk aversion intensity coefficient and a prediction confidence index. By introducing a specific nonlinear mathematical constraint function, the optimization solver is able to accurately quantify the risk and improve the solution efficiency.

[0012] In a preferred embodiment of the present invention, the objective function is solved within a rolling prediction time domain to obtain a sequence of control commands covering the future prediction time domain; and only the first control command at the current moment is extracted and executed. This rolling optimization feature enables the system to continuously correct historical prediction deviations using the latest vehicle state.

[0013] As a preferred embodiment of the present invention, switching is performed between the energy management method and at least one rule-based basic energy management method. This feature enriches the vehicle's operating modes, giving the user absolute control in deterministic short-distance or off-road conditions.

[0014] As a preferred embodiment of the present invention, the rule-based basic energy management method includes at least one of a pure electric priority mode, a battery hold mode, and a forced charging mode. This feature ensures optimal vehicle adaptation in highly differentiated scenarios such as urban commuting and off-road driving in mountainous areas.

[0015] As a preferred embodiment of the present invention, multi-source travel information is collected through an onboard sensor network, and the multi-source travel information is input into a pre-trained deep learning model based on a long short-term memory network or a gated recurrent unit network to output prediction information and confidence index. This feature greatly improves the accuracy and robustness of medium- and long-distance traffic flow prediction.

[0016] As a preferred embodiment of the present invention, when the prediction confidence index is higher than a set threshold and the prediction information indicates that the vehicle is about to enter a long downhill section, the state parameters of the energy storage unit are actively consumed to a set low-level node. This active "emptying" feature enables the system to recover downhill braking energy with 100% zero waste, significantly reducing overall energy consumption.

[0017] In a preferred embodiment of the present invention, operational feature data is uploaded to a cloud server via an onboard communication unit. The model parameters are then globally iteratively updated based on a reinforcement learning algorithm and downloaded for replacement. This feature endows the vehicle with a "self-evolving" capability, making it increasingly intelligent with use and extending the lifespan of the hardware.

[0018] As a preferred embodiment of the present invention, the objective function further includes a fuel consumption term characterizing fuel economy, and / or an operating efficiency term constraining the range extender system to operate within its high-efficiency range. This feature mathematically constrains the range extender to always operate near its optimal combustion point, while also ensuring smooth NVH (noise, vibration, and harshness) performance.

[0019] Secondly, the present invention provides the following technical solution:

[0020] An energy management system for a vehicle is applied to a vehicle including a drive system, an energy storage unit electrically connected to the drive system, and a range extender system for charging the energy storage unit. The system includes: a trip information sensing module configured to acquire predicted information of future driving conditions and a prediction confidence index characterizing the reliability of the predicted information; a control strategy decision module configured to determine the control command of the range extender system in each control time domain by optimizing an objective function based on a model predictive control method; wherein the objective function includes a penalty term for penalizing the deviation of the state parameters of the energy storage unit from a preset reference trajectory, and the penalty weight of the penalty term is negatively correlated with the magnitude of the prediction confidence index; and a dynamic execution module configured to receive the control command and generate physical electrical signals for driving hardware actuators to adjust the output power of the range extender system.

[0021] This invention provides a closed-loop control architecture system corresponding to the aforementioned management method, which facilitates the modular deployment of this solution in the vehicle's existing electronic and electrical architecture (such as the VCU vehicle controller) and improves the system's integration response speed.

[0022] As a preferred embodiment of the system of the present invention, the range extender system is a pluggable range extender device, comprising: an internal combustion engine, a generator connected to the internal combustion engine, a fuel tank, and a controller unit; the pluggable range extender device is mechanically fixed and electrically connected to the vehicle's energy storage unit and the energy management system through an integrated vehicle interface. By limiting the application environment to a pluggable architecture, the present invention completely eliminates the hardware dead weight of urban commuting, realizing a perfect multi-purpose vehicle form of "daily pure electric unloading for weight reduction, and long-distance hybrid loading for range extension".

[0023] Compared with the prior art, the present invention has the following substantial features and significant beneficial effects:

[0024] 1. A groundbreaking "risk-adaptive fallback" defense mechanism: This innovatively incorporates "prediction confidence level" as a feedback variable into the optimization control objective. Faced with sudden congestion that cannot be accurately predicted in the real world, the algorithm automatically increases the power conservation weight. This is equivalent to buying "insurance" for the uncertain future, completely eliminating serious safety hazards such as over-discharge of the power battery and power limitation caused by prediction errors.

[0025] 2. Extreme optimization and dynamic allocation of global energy consumption: By integrating multi-source information sensing and long-term time-domain planning, this system abandons short-sighted rule-based strategies. When predicting with high confidence, it performs deep discharge "emptying" based on the full-stroke potential energy; verified, this can effectively reduce the combined energy consumption of oil and electricity by 5% to 15%.

[0026] 3. Breaking physical boundaries and adapting to pluggable architecture for burden-free travel: The system's algorithm is perfectly compatible with the intervention of pluggable range extenders. This allows the vehicle to operate as a pure electric vehicle on a daily basis, reducing its weight by hundreds of kilograms; for long-distance travel, adding a module instantly transforms it into an all-weather hybrid vehicle, truly achieving a balance between energy efficiency and anxiety-free travel. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments disclosed in this invention, the accompanying drawings of the embodiments will be briefly described below. These drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention.

[0028] Figure 1 This is a logic flowchart of the vehicle energy management method provided in an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of the control architecture and module interaction of the vehicle energy management system provided in an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions (including preferred technical solutions) of the present invention will be further described in detail below with reference to the accompanying drawings and by way of listing some optional embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0031] Example 1

[0032] This embodiment focuses on illustrating the backbone signal flow and control architecture of the energy management method at the vehicle controller level. For example... Figure 1 As shown, this method is executed periodically in a real-time operating system with a preset discrete time step (e.g., 10 to 100 milliseconds).

[0033] At the start of each control cycle, the system performs forward sensing. It reads predicted information about future driving conditions via the bus and simultaneously extracts a "prediction confidence index" that characterizes the reliability of this information. This index is a normalized continuous scalar; the lower the value, the higher the uncertainty of the future traffic environment (such as congestion, accidents, lane changes).

[0034] Subsequently, the system enters the core solution phase based on Model Predictive Control (MPC). At the current moment, the algorithm establishes a prediction time domain of length N for the future. The system establishes an objective function containing multiple costs, and before each iteration, it uses the currently acquired prediction confidence index to reshape the penalty weights of the deviation terms of the energy storage unit (such as the power battery) state parameters in the objective function in real time. Under the constraints of satisfying the system's physical limits (such as the maximum torque response rate of the prime mover and the maximum charge / discharge current of the energy storage unit), the nonlinear optimization solver is invoked to solve for the sequence of control commands that minimizes the total cost within the prediction time domain.

[0035] To overcome the mismatch in the system's mathematical model and unobservable external disturbances, this method strictly implements the "rolling optimization" mechanism in MPC theory. The system discards subsequent calculated sequences, extracting only the first control command at the current moment and converting it into actual physical electrical signals (such as torque request messages or PWM pulse width modulation signals), which are then sent to the underlying actuators of the range extender system and energy storage unit. In the next control cycle, the system restarts the closed-loop game process based on the latest actual vehicle state as the initial state. This architecture completely establishes the link from the upper-level abstract mathematical game to the lower-level mechanical power response.

[0036] Example 2

[0037] This embodiment takes a progressive approach to deeply analyze the underlying mathematical logic of the objective function construction and dynamic weight adjustment mechanism.

[0038] In the control time domain N, the discrete functional form of the multi-objective cost function J constructed by the system is as follows:

[0039]

[0040] To clearly illustrate the specific physical and engineering significance of this formula, the parameters are defined in the following table:

[0041]

[0042] Among them: the first item is the fuel consumption item, w fuel For constant weights, Let be the fuel consumption rate of the prime mover at step k, designed to constrain global fuel economy. The third term is the operating efficiency term, where λ is a constant weight, and P... eng,opt The optimal thermal efficiency point power for the range extender system is designed to constrain the prime mover to operate in the high-efficiency region through quadratic penalty, thereby reducing NVH (noise and vibration).

[0043] The second term is the penalty for the energy storage unit's state parameters (specifically, the state of charge (SOC(k)) deviating from the reference trajectory SOC) in this example. ref (k) is the core penalty term. In this embodiment, the penalty weight wSOC Defined as a time-varying parameter, it is related to the prediction confidence index C ( Establish the following negative correlation dynamic adjustment function:

[0044]

[0045] In this formula, w SOC,base The basic penalty weight is α, which is the risk aversion intensity coefficient.

[0046] When the prediction is extremely reliable (C→1), the adjustment factor (1-C) approaches 0, w SOC The system collapses to its base value. At this point, the objective function J grants the optimizer significant power allocation authority, allowing SOC(k) to fluctuate considerably in the prediction time domain, so that the system can seek the theoretical minimum of the first term (fuel consumption) through deep discharge or delayed charging.

[0047] Conversely, when encountering information blind spots or divergent prediction results (C→0), the multiplier factor (1-C) increases significantly, leading to w SOC The cost increases several times or even tens of times. Under this extremely high penalty constraint, any mathematical solution that leads to a decrease in SOC(k) during the solver's optimization process will cause the total cost J to increase dramatically. Therefore, the solver is forced to output control commands that favor the immediate activation of the range extender system or even high-power generation. This mechanism mathematically prevents the risk of power limitation caused by excessive power consumption, achieving a defensive bottom-line control for vehicle energy management.

[0048] Example 3

[0049] This embodiment details the techniques for acquiring travel information and generating prediction confidence indices.

[0050] The onboard sensor network is configured to capture real-time, multi-dimensional trip data. Sensing sources include not only onboard radar and vision sensors, but also deeply integrated road curvature and slope sequences extracted from high-resolution maps (HD Maps), average traffic speed heatmaps of main roads downloaded from the cloud, and traffic light phase timing (SPaT) information provided by vehicle-to-everything (V2X) communication. Furthermore, the system simultaneously extracts the power spectral density of the driver's historical accelerator pedal opening to characterize individual driving styles.

[0051] The aforementioned multidimensional heterogeneous time-series data is fed into a deep learning model deployed on an onboard computing platform. This model is preferably a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU) network to effectively capture the long-term temporal dependencies of the sequence data. To output an accurate "prediction confidence index," the deep learning model employs Bayesian variational inference or Monte Carlo Dropout techniques at the output layer to perform multiple forward random samplings and calculate the variance of the multiple predicted trajectory results. This variance is then processed by a nonlinear inverse mapping function and normalized to a confidence index within the range of 0 to 1. This process accurately quantifies the data uncertainty caused by sensor noise and the cognitive uncertainty caused by insufficient model training.

[0052] Example 4

[0053] This embodiment covers control strategies for various special operating conditions and the expansion of basic rule patterns.

[0054] In the first high-confidence specific scenario (such as a long downhill section on a highway), when the prediction confidence index is higher than a set threshold (e.g., 0.90), and the terrain data clearly indicates an extremely long downhill section ahead, the system actively activates a "storage clearing" mechanism: the system prioritizes and fully utilizes the energy storage unit to drive the vehicle, actively reducing the state of charge of the energy storage unit to a set low level (e.g., 20%) before reaching the start of the downhill section. This strategy completely releases the battery's chemical energy storage space, allowing 100% absorption of the regenerative braking energy during subsequent long downhill sections, avoiding the waste of mechanical braking energy caused by a fully charged battery.

[0055] In the second low-confidence scenario (such as a sudden large-scale traffic jam), when radar or V2V networks detect multiple vehicles braking suddenly ahead and the duration of the congestion is unknown, the confidence index plummets. At this point, the aforementioned negative correlation adjustment mechanism is triggered, and the system enters a defensive posture, forcing the range extender system to generate electricity in its high-efficiency zone, raising the state of charge to a high level (e.g., above 55%) in advance. This ensures sufficient power to cope with frequent starts and stops during prolonged congestion and avoids frequent and inefficient start-stops of the prime mover in low-speed congestion.

[0056] Furthermore, the system supports seamless switching between rule-based basic energy management methods. Based on user commands, the system can lock into "pure electric priority mode" (completely cutting off the range extender's start-up enable signal as long as the battery level is above the lower limit, achieving quiet commuting with zero fuel consumption in urban areas) or switch to "forced charging mode" (ignoring current energy consumption optimization, forcing the range extender to operate at full load until the battery level reaches 80%) to meet the high-voltage power reserve requirements before extreme off-road extrication.

[0057] Example 5

[0058] This embodiment illustrates a cloud-based collaborative evolution architecture that ensures energy efficiency does not decline throughout the vehicle's lifespan.

[0059] The system deploys a cloud-based iterative mechanism for reinforcement learning (RL) based on Markov decision processes. During vehicle operation, the onboard communication unit continuously de-identifies and compresses vehicle operation feature data before uploading it to the cloud server.

[0060] A reinforcement learning architecture for policy evolution was constructed within the computing cluster of cloud servers. The state space includes features such as vehicle speed, acceleration, state of charge, temperature, and confidence level; the action space includes discrete start-stop commands for the prime mover and continuous power output coefficients; and the reward function is configured to positively incentivize lower fuel and electricity consumption, impose extremely stringent one-sided penalties for battery levels falling below a safe threshold, and penalize frequent and drastic power fluctuations to balance NVH (noise, vibration, and harshness).

[0061] Through billions of virtual simulations conducted in the cloud via an algorithmic agent, the cloud server can train new policy neural network parameters that can achieve higher cumulative rewards, tailored to the specific hardware aging conditions of a vehicle (such as increased battery internal resistance) and the driver's specific driving habits. Subsequently, the vehicle communication unit securely downloads the updated model parameter matrix from the cloud and silently replaces the original policy in the vehicle's control policy decision module via OTA (Over-The-Air) technology, realizing the dynamic self-healing and evolution of the vehicle's control brain.

[0062] Example 6

[0063] This embodiment specifically illustrates the system-level architecture and physical implementation of the hardware interface for implementing the aforementioned control method.

[0064] like Figure 2 As shown, this invention provides a vehicle energy management system. Within the software domain architecture, the system is divided into a trip information perception module, a control strategy decision-making module, and a dynamic execution module.

[0065] At the physical execution level, in addition to conventional fixed-assembly range extender systems, this invention further provides a decoupled "pluggable range extender". This device is a separate, highly integrated hardware module, with an internal combustion engine, generator, fuel tank, and controller unit that performs AC / DC rectification compactly arranged inside the housing.

[0066] The physical connection of this pluggable range extender relies on an integrated vehicle interface assembly. This assembly includes rails and self-locking mechanical connectors installed in vehicle storage spaces (such as the trunk floor), as well as an electrical quick-connect interface with a high-voltage safety interlock mechanism. When the device is pushed into the trunk, it instantly achieves high-impact-resistant mechanical anchoring, while the electrical interface safely connects the device's high-voltage output to the high-voltage DC bus of the vehicle's energy storage unit, and the low-voltage communication pins simultaneously wake up the control strategy decision module.

[0067] In addition, the pluggable range extender's housing is equipped with a human-machine interface and an independent operating interface. The human-machine interface includes a start / stop switch and status indicator lights; the independent operating interface is an independent waterproof standard power output socket (such as a 220V AC or DC fast charging interface). When users remove the device from their vehicles for short-distance commutes in urban areas to reduce vehicle weight, the device can directly power other external electrical equipment (such as power tools, outdoor appliances, emergency power supply for their own shops, or charging other powered vehicles) through the independent operating interface. In this case, the device can operate as an independent micro generator set, greatly expanding the product's application scenarios.

[0068] In this configuration, the energy management system of the present invention possesses dynamic device discovery capabilities. When no hardware access is detected, the vehicle operates as a pure electric vehicle, significantly reducing daily load power consumption and freeing up physical space. Once the hardware is inserted and a handshake is successful, the energy management system automatically loads a confidence-based MPC control algorithm, taking over the power generation equipment for all-weather, long-duration hybrid scheduling. This deep collaborative design of software and hardware revolutionarily achieves optimal performance solutions under various vehicle attribute scenarios.

[0069] Example 7

[0070] This embodiment details the three basic and parallel operating modes provided by the control strategy decision module when the vehicle is equipped with the pluggable range extender.

[0071] One mode is the pure electric priority mode, specifically designed for short-distance urban commuting or severe traffic congestion. The underlying control logic in this mode is as follows: the system monitors the state of charge (SOC) of the energy storage unit (such as the battery pack) in real-time at high frequency. If the SOC parameter exceeds a set lower threshold (e.g., 20%), the control strategy decision module will cut off the range extender's start-up enable signal from the underlying software, forcing the prime mover into a completely dormant state. At this time, the vehicle exhibits the full physical characteristics of a pure electric vehicle, completely avoiding the inefficiency of the internal combustion engine in low-speed congestion, drastically reducing urban driving costs, and providing passengers with an absolutely quiet and smooth experience.

[0072] The second is the battery hold mode (automatic mode), which is the core strategy for long-distance high-speed cruising. The system pre-calibrates a target reference state parameter (e.g., a target state of charge of 30%). When the energy storage unit has sufficient charge, it drives the discharge. When the state parameter drops to near the target value, the control strategy decision module instructs the range extender system to smoothly intervene. Under stable load conditions such as cruise control, the internal combustion engine is strictly anchored to its optimal economic efficiency point (BSFC) on its universal characteristic diagram, and the excess power generated is rectified and charged into the energy storage unit, allowing the battery to slowly recover. When encountering rapid acceleration for overtaking or steep uphill climbs, the system triggers dynamic parallel logic, and the range extender system and the energy storage unit simultaneously deliver peak power to the drive system, ensuring the ultimate response of the vehicle's power.

[0073] Thirdly, there is the forced charging mode (mountain mode), designed as a defensive reserve strategy for extreme off-road conditions or areas without refueling facilities. When the driver activates this mode via command, the control strategy decision module ignores the fact that the current driving resistance is extremely low, forcing the range extender system to abandon the pursuit of economic optimization and directly jump to a high-speed, high-power state to continuously generate electricity. The system rapidly charges a massive amount of electrical energy into the energy storage unit, forcibly raising and stabilizing the state parameters in a high range (e.g., 60% to 80%). This strategy ensures that when the vehicle subsequently enters muddy mountain roads or extreme cross-axle terrain, the drive system always has peak discharge voltage and torque burst power, completely eliminating the risk of "rollback" or power limitation due to undervoltage in the energy storage unit.

[0074] Example 8

[0075] This embodiment takes the algorithm data flow as the main line and deeply analyzes the execution architecture of the adaptive energy management method based on model predictive control (MPC).

[0076] The core calculations for energy management are completed within the vehicle control domain. Once the vehicle starts and the pluggable range extender is detected as online, the trip information sensing module begins operation. This module breaks through the limitations of a single sensor, acquiring multi-dimensional heterogeneous trip data through the fusion of onboard antennas and environmental sensors. Sensing sources include positioning and terrain gradient extracted from the onboard high-precision GPS, forward-looking traffic light phases captured via vehicle-to-everything (V2X) communication, and long-distance traffic congestion heatmaps transmitted from the cloud.

[0077] The trip information perception module embeds a deep learning prediction model based on a Long Short-Term Memory (LSTM) network. This model takes the aforementioned multi-source heterogeneous data and the driver's recent historical accelerator pedal action sequences as input feature vectors. After nonlinear hidden layer operations, it outputs a refined speed trajectory (i.e., prediction information) that foresees the future (e.g., the next 5 kilometers). Simultaneously, through an internal uncertainty assessment mechanism, the model outputs a prediction confidence index ranging from 0 to 1. This index precisely quantifies the system's confidence in its understanding of the future traffic environment.

[0078] Next, the control strategy decision module constructs a multi-objective cost function based on a rolling time-domain framework. The system dynamically reshapes the penalty weights of the energy storage unit state parameters in the objective function using negative correlation logic, based on real-time acquired prediction confidence indices. Within an extremely short control cycle, the optimization solver searches for a sequence of control commands that minimizes the total functional within the prediction window, spanning a vast action space. The dynamic execution module extracts only the current-time solution of this sequence, converts it into physical electrical signals (such as torque request commands), and sends them to the range extender system's controller unit, thereby precisely controlling the prime mover's fuel injection ignition and the generator's anti-drag torque.

[0079] Example 9

[0080] This embodiment illustrates the specific physical execution process of the energy management method of the present invention in squeezing the global energy consumption limit under specific macroscopic deterministic road conditions.

[0081] When a vehicle is traveling on a highway, the trip information perception module, combined with a high-precision map, learns that a long, continuous downhill section is about to begin a few kilometers ahead. Due to favorable weather conditions, simple road conditions, and stable vehicle-to-everything (V2X) signal, the prediction confidence index output by the deep learning model climbs to an extremely high level (e.g., above 0.95).

[0082] Due to the extremely high confidence level, the dynamic adjustment logic within the control strategy decision module compresses the penalty weight for state parameters deviating from the preset trajectory to an extremely low level. After evaluating the global cost, the optimization solver decisively makes a "proactive energy storage" decision: the system immediately issues a shutdown command to stop the range extender system's power generation, and on the straight section of road before reaching the downhill starting point, it forcibly and aggressively calls the energy storage unit to output high power to drive the vehicle, actively and rapidly consuming the current state of charge from a high level (e.g., 50%) to a set low level (e.g., 25%).

[0083] When the vehicle actually enters a long downhill section, its gravitational potential energy generates significant anti-drag braking power. Thanks to the large chemical energy storage space pre-released by the system before the descent, the regenerative braking energy during the downhill process is 100% absorbed by the energy storage unit, completely avoiding energy waste and thermal degradation risks caused by being forced to use mechanical braking when the battery is fully charged. Engineering verification has shown that this high-confidence active scheduling strategy can significantly improve overall energy utilization.

[0084] Example 10

[0085] This embodiment illustrates how the underlying algorithm of the present invention implements a defense mechanism to ensure vehicle power safety when faced with a surge in prediction uncertainty.

[0086] Imagine a vehicle traveling normally on an urban expressway, with no traffic expected. However, the vehicle's sensor network suddenly receives a V2X warning signal indicating that multiple vehicles are braking suddenly hundreds of meters ahead. The system determines that a chain collision is highly likely ahead, causing unexpected and severe congestion. However, the duration and extent of this congestion are completely unknown. This significant information blind spot causes the prediction confidence index output by the deep learning model to plummet to an extremely low level (e.g., below 0.35).

[0087] Under this extreme condition, the control strategy decision module responds instantly, triggering the "risk avoidance mode." Based on the logic of the negative correlation function, the penalty weight in the objective function is amplified exponentially. Faced with extremely high penalties for deviations from the state of charge, the optimization solver resolutely rejects any control action that would lead to a further decrease in charge.

[0088] Even when the vehicle is already crawling at low speed with minimal drive power demand, the control strategy decision module will still forcibly activate the range extender system and constrain the prime mover to operate at full speed within its optimal efficiency curve range. The vast amount of electrical energy generated by the range extender system, aside from a very small portion used to propel the vehicle forward slowly, is largely forcibly stored as a "defensive reserve" in the energy storage unit, rapidly raising the state of charge from its current low level (e.g., 30%) to a safe, high level (e.g., above 45%). The physical significance of this strategy is that it utilizes short-term fuel consumption to reserve sufficient electrical energy for potential hours of unpredictable traffic congestion, completely avoiding the risk of breakdown due to depleted battery power, while also preventing the severe mechanical vibrations and NVH deterioration caused by repeated start-stop cycles in congested traffic.

[0089] Example 11

[0090] This embodiment elaborates on the cloud collaboration and parameter self-evolution framework built by the vehicle to cope with hardware aging and individual differences among drivers.

[0091] Since the factory static calibration of energy management strategies cannot cover the degradation characteristics throughout the vehicle's entire life cycle, this invention deeply integrates a vehicle-cloud collaborative mechanism. During the vehicle's daily operation, the onboard communication unit continuously uploads massive amounts of anonymized driving data to a remote cloud server via a long-term connection.

[0092] In the cloud computing cluster, a high-order reinforcement learning evolution framework based on Markov decision processes was deployed, and its core mathematical elements are rigorously defined as follows:

[0093] First is the State Space, which serves as the boundary for the intelligent agent to perceive the environment. It not only includes the vehicle's current absolute physical parameters (such as the state of charge of the energy storage unit, vehicle speed, ambient temperature, etc.), but more importantly, it extracts feature parameters that reflect the driver's behavioral profile (such as the root mean square error of the rate of change of the accelerator pedal opening, which is used to accurately quantify the aggressiveness or mildness of the driving style).

[0094] Secondly, there is the Action Space, which defines the physical execution domain that the energy management system can intervene in. It covers a mixture of discrete and continuous commands, including the target state of charge setpoint in the next control time domain, the start-stop Boolean commands issued to the prime mover of the range extender system, and the continuous power allocation ratio coefficient between the range extender system and the energy storage unit.

[0095] Finally, there is the reward function, which serves as the sole rule guiding the algorithm to converge in the right direction. It is configured to simultaneously balance three major control objectives: positive rewards to drive lower total fuel and electricity consumption (economy objective), extreme one-sided penalties to absolutely prevent the state of charge from falling below the safe lower limit (power safety objective), and moderate penalties to suppress frequent start-stop of the prime mover and drastic power-torque jumps (smoothness objective).

[0096] Based on the above framework, the cloud server endlessly iterates and tries and fails under billions of simulated conditions. When a new model parameter matrix that can obtain higher cumulative discount rewards is trained for a specific vehicle, the cloud server silently sends the updated model parameters through the vehicle communication unit using OTA (Over-The-Air) technology, replacing the old strategy in the vehicle's control strategy decision module, thus giving the vehicle the lifelong evolution capability of "understanding the driver better and better the more it is driven, and consuming less energy the more it is driven".

[0097] It will be readily understood by those skilled in the art that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, combinations, substitutions, improvements, etc., made under the spirit and principles of the present invention are included within the protection scope of the present invention.

Claims

1. A vehicle energy management method, applied to a vehicle including a drive system, an energy storage unit electrically connected to the drive system, and a range extender system for charging the energy storage unit, characterized in that, The method includes: Obtain predictive information about future driving conditions, and a prediction confidence index that characterizes the reliability of the predictive information; Based on the model predictive control method, in each control time domain, the control command of the range extender system is determined by optimizing an objective function; wherein, the objective function includes at least a penalty term for penalizing the deviation of the state parameters of the energy storage unit from a preset reference trajectory, and the penalty weight of the penalty term is negatively correlated with the magnitude of the prediction confidence index; Based on the optimized control commands, the prime mover in the range extender system is driven to adjust its output power, and / or the charging and discharging power of the energy storage unit is controlled, so as to minimize the overall energy consumption of the vehicle while maintaining the safe energy threshold of the energy storage unit.

2. The method according to claim 1, characterized in that, The negative correlation between the penalty weight of the penalty item and the magnitude of the prediction confidence index includes: The penalty weight of the penalty term is obtained by multiplying the basic penalty weight by a dynamic adjustment factor. The dynamic adjustment factor increases linearly or non-linearly as the prediction confidence index decreases, so that when the prediction confidence index decreases, the optimization process of the objective function tends to suppress the decline of the state parameters of the energy storage unit.

3. The method according to claim 2, characterized in that, The penalty term is a quadratic penalty term of the state parameter, and its corresponding penalty weight w SOC This is determined through the following specific functional relationship: Among them, w SOC,base The basic penalty weight is α, which is a risk aversion strength coefficient greater than zero, and C is the prediction confidence index with a value between [0,1].

4. The method according to claim 1, characterized in that, The steps for determining the control commands for the range extender system using the model predictive control method include: Solve the objective function within the rolling prediction time domain to obtain a sequence of control commands covering the future prediction time domain; and Only the first control instruction at the current moment in the control instruction sequence is extracted and executed, and the above steps are repeated in the next control time domain based on the updated vehicle state.

5. The method according to claim 1, characterized in that, The method further includes: Switching between the energy management method and at least one rule-based basic energy management method.

6. The method according to claim 5, characterized in that, The rule-based basic energy management method includes at least one of pure electric priority mode, power retention mode, and forced charging mode.

7. The method according to claim 1, characterized in that, The steps for obtaining prediction information and prediction confidence index for future driving conditions include: Multi-source trip information is collected through vehicle-mounted sensor networks, including high-precision map data, real-time traffic information, vehicle-to-everything (V2X) information, and historical driving behavior data of vehicles. The multi-source travel information is input into a pre-trained deep learning model based on a long short-term memory network or a gated recurrent unit network to output the prediction information and the prediction confidence index.

8. The method according to claim 1, characterized in that, When the prediction confidence index is higher than a set threshold, and the prediction information indicates that the vehicle is about to enter a long downhill section, the method further includes: The vehicle is driven by the electric power of the energy storage unit, so as to actively consume the state parameters of the energy storage unit to a set low node, and reserve energy storage space for downhill braking energy recovery.

9. The method according to claim 1, characterized in that, The method also includes a co-evolution step between the vehicle and the cloud server: The vehicle's operational characteristic data is uploaded to a remote cloud server via the vehicle-mounted communication unit, so that the cloud server can perform global iterative updates of the model parameters of the multi-objective cost function based on a reinforcement learning algorithm. as well as The vehicle-mounted communication unit downloads updated model parameters from the cloud server to replace the original control strategy on the vehicle.

10. The method according to claim 1, characterized in that, The objective function also includes a fuel consumption term to characterize fuel economy, and / or an operating efficiency term to constrain the range extender system to operate in the high-efficiency range.

11. An energy management system for a vehicle, applied to a vehicle comprising a drive system, an energy storage unit electrically connected to the drive system, and a range extender system for charging the energy storage unit, characterized in that, The system includes: The trip information perception module is configured to acquire prediction information of future driving conditions and a prediction confidence index that characterizes the reliability of the prediction information. The control strategy decision module is configured to use a model predictive control method to determine the control command of the range extender system by optimizing an objective function in each control time domain. The objective function includes a penalty term for penalizing the deviation of the state parameters of the energy storage unit from a preset reference trajectory, and the penalty weight of the penalty term is negatively correlated with the magnitude of the prediction confidence index. The dynamic execution module is configured to receive the control commands and generate physical electrical signals for driving the hardware actuators to adjust the output power of the range extender system.

12. The system according to claim 11, characterized in that, The objective function also includes a first penalty term representing fuel economy and a second penalty term constraining the range extender system to operate within the optimal efficiency curve range; the control strategy decision module is configured to increase the weight of the penalty term for the state parameter deviating from the preset reference trajectory when the prediction confidence index decreases.

13. The system according to claim 11, characterized in that, The control strategy decision module is further configured to switch between an adaptive energy management mode based on the model predictive control method and a rule-based basic energy management mode according to preset conditions or user instructions.

14. The system according to claim 13, characterized in that, The rule-based basic energy management model includes: In pure electric priority mode, the start signal of the range extender system is cut off as long as the energy storage unit's charge level is above a lower threshold. Forced charging mode: In this mode, the range extender system is forced to continuously generate electricity at high power until the energy storage unit reaches a preset high level range.

15. The system according to claim 11, characterized in that, The system also includes an in-vehicle communication unit configured to establish a communication connection with a remote cloud server to enable over-the-air download and update of model parameters.

16. The system according to claim 11, characterized in that, The state parameter of the energy storage unit is the battery state of charge.

17. The system according to claim 11, characterized in that, The range extender system is a pluggable range extender, which includes: an internal combustion engine, a generator connected to the internal combustion engine, a fuel tank, and a controller unit; the pluggable range extender is mechanically fixed and electrically connected to the vehicle's energy storage unit and the energy management system through an integrated vehicle interface.

Images