Hybrid power system energy management algorithm

Abstract

The invention relates to the technical field of energy management, in particular to a hybrid power system energy management algorithm, which comprises a working condition prediction module, a dynamic weight distribution module and a hierarchical execution module, and is characterized in that the working condition prediction module is used for fusing vehicle networking data and vehicle-mounted sensor data and generating a multi-dimensional working condition prediction result in a future time period; the dynamic weight distribution module calculates dynamic weight coefficients of different optimization targets in real time based on the working condition prediction result, and generates an optimization problem description file containing weight coefficient constraint conditions; a rolling optimization layer unit in the hierarchical execution module receives the optimization problem description file and solves a global optimal power distribution reference in a fixed time window; the dynamic compensation layer unit generates a compensation correction item for dynamically adjusting a weight coefficient according to a deviation value between a global optimal power distribution reference and a current vehicle state; and the real-time control layer unit converts the compensation correction item into a power distribution instruction, and synchronously updates the weight coefficient of the dynamic weight distribution module.

Description

Technical Field

[0001] The present invention relates to the technical field of energy management, and in particular to an energy management algorithm for a hybrid power system. Background Art

[0002] The energy management strategy of traditional hybrid power systems has long been limited by the inherent defects of the technical framework. At the data perception level, traditional solutions mostly use local sensor networks (such as vehicle speed and engine speed) to judge working conditions, lacking global perception of the vehicle operating environment (such as road slope and traffic flow density) and driver behavior (such as acceleration habits and braking frequency), resulting in insufficient adaptability of energy distribution strategies to real driving scenarios. At the optimization algorithm level, existing methods are generally based on static weights or preset rules for limited working conditions, which makes it difficult to cope with the coupling effects of time-varying factors such as battery aging and ambient temperature changes on multi-objective optimization (fuel efficiency, emission control, and power response), causing the system to easily fall into local optimal solutions under complex working conditions. From the perspective of control architecture, there is a lag in information interaction between the planning layer (such as model-based predictive control) and the execution layer (such as the torque distribution module) of the traditional hierarchical strategy, and the mismatch between the instruction refresh frequency and the execution cycle under dynamic conditions is prominent, making it difficult to balance the real-time and robustness of energy flow regulation.

[0003] In addition, with the intelligent upgrade and networking development of new energy vehicles, the technical bottlenecks exposed by traditional systems in vehicle-road-cloud collaborative decision-making and multi-energy coupling dynamic optimization have become key obstacles to improving the energy efficiency of hybrid vehicles throughout their life cycle. Summary of the invention

[0004] The object of the present invention is to provide a hybrid power system energy management algorithm to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a hybrid power system energy management algorithm, including a working condition prediction module, a dynamic weight allocation module and a hierarchical execution module, wherein:

[0006] The working condition prediction module is used to integrate the Internet of Vehicles data and the vehicle-mounted sensor data, and generate multi-dimensional working condition prediction results for future periods;

[0007] The dynamic weight allocation module calculates the dynamic weight coefficients of different optimization objectives in real time based on the working condition prediction results, and generates an optimization problem description file containing weight coefficient constraint conditions;

[0008] The layered execution module comprises a rolling optimization layer unit, a dynamic compensation layer unit and a real-time control layer unit, wherein:

[0009] The rolling optimization layer unit receives the optimization problem description file and solves the global optimal power allocation benchmark within a fixed time window;

[0010] The dynamic compensation layer unit generates a compensation correction term including a weight coefficient for dynamically adjusting the weight coefficient according to the deviation between the global optimal power allocation reference and the current vehicle state;

[0011] The real-time control layer unit converts the compensation correction term into a power distribution instruction and synchronously updates the weight coefficient of the dynamic weight distribution module; through the actual execution feedback of the power distribution instruction and the dynamic weight re-update mechanism, a closed-loop dynamic correction link of the multi-objective optimization strategy is formed to ensure the real-time performance of the energy distribution strategy.

[0012] As a further improvement of the technical solution, the operating condition prediction module specifically includes:

[0013] Obtaining Internet of Vehicles data in a fixed area ahead through Internet of Everything technology, the data including the road slope change rate and traffic light timing;

[0014] The millimeter-wave radar is used to detect the position and speed of obstacles to generate a motion trajectory, the visual sensor is used to identify the curvature of the lane line, and the Kalman filter algorithm is used to fuse the trajectory and curvature to form the vehicle-mounted sensor data;

[0015] A spatiotemporal correlation matrix is ​​constructed to fuse the Internet of Vehicles data and the vehicle-mounted sensor data, wherein the spatiotemporal correlation matrix includes current environmental information and future road condition predictions.

[0016] As a further improvement of the technical solution, the spatiotemporal correlation matrix generates a multi-dimensional working condition prediction result through long short-term memory network model analysis, specifically including:

[0017] Road slope change rate, traffic flow and congestion probability, traffic light timing and obstacle movement trajectory;

[0018] The road slope change rate includes real-time slope data within a fixed distance in the future;

[0019] The traffic flow and congestion probability reflect the traffic density and start-stop frequency;

[0020] The traffic light timing is used to predict the acceleration and deceleration requirements of vehicles;

[0021] The obstacle motion trajectory includes the position, speed and trajectory of the obstacle within the prediction range.

[0022] As a further improvement of the present technical solution, the optimization objectives in the dynamic weight allocation module include fuel economy, battery life maintenance and power response requirements, and based on the optimization objectives and operating condition prediction results, the fuel saving rate Fs, the battery state of charge fluctuation suppression coefficient Sc and the power response delay penalty term Dp are calculated.

[0023] As a further improvement of the technical solution, the calculation process of the fuel saving rate Fs specifically includes:

[0024] Obtain fuel consumption benchmark value based on flat road and uniform speed conditions;

[0025] Calculate additional load fuel consumption based on the road grade change rate;

[0026] Estimate the number of starts and stops and additional fuel consumption based on traffic light timing;

[0027] The percentage difference between the actual predicted fuel consumption and the benchmark value is calculated comprehensively as the fuel saving rate.

[0028] As a further improvement of the technical solution, the calculation process of the battery state of charge fluctuation suppression coefficient Sc specifically includes:

[0029] Estimate the average acceleration and deceleration frequency based on traffic flow;

[0030] Predict the number of emergency events based on obstacle movement trajectories;

[0031] A battery state of charge variation sequence model is established and the inverse of the standard deviation is calculated, and the battery state of charge fluctuation suppression coefficient is obtained through normalization processing.

[0032] As a further improvement of the technical solution, the calculation process of the power response delay penalty term Dp specifically includes:

[0033] Estimate acceleration demand based on congestion probability;

[0034] Set target response time thresholds for emergency conditions;

[0035] When the actual response time exceeds the target response time threshold, the square of the difference is calculated as the penalty term.

[0036] As a further improvement of the technical solution, the dynamic weight allocation module uses a deep deterministic policy gradient algorithm to calculate the weight coefficient, and its reward function R is:

[0037] R=w1Fs+w2Sc+w3Dp, w1 represents the fuel economy weight coefficient; w2 represents the battery life maintenance weight coefficient; w3 represents the power response penalty weight coefficient.

[0038] As a further improvement of the technical solution, the rolling optimization layer unit uses a quadratic programming algorithm to process nonlinear optimization problems, specifically including:

[0039] Transform the fuel-battery balance constraint into a convex optimization problem;

[0040] The optimal reference value sequence of engine output power, motor torque distribution and battery charging and discharging power in the future period is calculated as the global optimal power distribution benchmark.

[0041] As a further improvement of the present technical solution, when the dynamic compensation layer unit generates the compensation correction term, it obtains the current vehicle status information in real time through the on-board sensor data and the Internet of Vehicles data, calculates the deviation between the current status parameters and the global optimal power allocation benchmark; generates fine-tuning instructions for the engine power, motor torque and battery charging and discharging power; the real-time control layer unit converts the compensation correction term into a specific power distribution instruction and transmits it to the instruction execution system, collects the actual operating data after the execution of the instruction, and determines whether to trigger the weight coefficient to be updated again based on the difference between the execution result and the expected target.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] This hybrid system energy management algorithm effectively overcomes the defects of traditional technologies by building a full-link closed-loop decision-making system of "prediction-optimization-execution". The operating condition prediction module significantly improves the accuracy of operating condition prediction and provides a forward-looking decision-making basis for energy allocation. The dynamic weight allocation module realizes adaptive adjustment of multi-objective optimization weights based on real-time operating condition characteristics, effectively balancing the coupling relationship between fuel economy, emission performance and power.

[0044] In addition, the hierarchical execution module significantly shortens the control command response time and improves execution stability through the collaborative mechanism of rolling optimization and dynamic compensation. This technical system comprehensively improves the real-time decision-making ability and global optimization effect of the hybrid system, and provides a solution with strong engineering feasibility for the energy management of new energy vehicles, promoting the industry to continue to evolve towards intelligence and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the overall module of the present invention;

[0046] Figure 2 It is a schematic diagram of the hierarchical execution module unit of the present invention.

[0047] In the figure: 100, working condition prediction module; 200, dynamic weight allocation module; 300, layered execution module; 301, rolling optimization layer unit; 302, dynamic compensation layer unit; 303, real-time control layer unit. DETAILED DESCRIPTION

[0048] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Next, see Figure 1 The present invention provides a technical solution: a hybrid power system energy management algorithm, including a working condition prediction module 100, a dynamic weight allocation module 200 and a hierarchical execution module 300.

[0050] The working condition prediction module 100 is used to integrate the Internet of Vehicles data and the vehicle sensor data, and generate multi-dimensional working condition prediction results for the future period, specifically including:

[0051] Through V2X (Vehicle-to-Everything) communication technology (i.e. Internet of Everything technology), the Internet of Vehicles data within a fixed area ahead (e.g. 5 kilometers) is obtained. The Internet of Vehicles data includes but is not limited to the road slope change rate and traffic light timing. The Internet of Vehicles data can provide detailed environmental information about the road section the vehicle is about to travel. V2X represents the information exchange between the vehicle and the outside world, which is also called the Internet of Vehicles technology.

[0052] Using multiple on-board sensors, including millimeter-wave radar and visual sensors, real-time environmental information around the current vehicle is collected. The position and speed of obstacles are detected by millimeter-wave radar to generate obstacle motion trajectories. The curvature of lane lines is identified by visual sensors. The Kalman filter algorithm is used to fuse the obstacle motion trajectory and lane curvature as on-board sensor data.

[0053] The Internet of Vehicles data and vehicle sensor information are integrated to construct a spatiotemporal correlation matrix, which contains not only the real-time information of the current vehicle surroundings, but also the road conditions that may be encountered in the future.

[0054] The long short-term memory network model is used to analyze the spatiotemporal correlation matrix to generate multi-dimensional working condition prediction results for future time periods. The model can predict the future working condition change trend based on historical data and current data, and generate multi-dimensional working condition prediction results, which include road slope change rate, traffic flow and congestion probability, traffic light timing and obstacle movement trajectory; the road slope change rate represents the real-time change data of the road slope within a fixed distance (for example, 5 kilometers) in the future, which directly affects the engine load and energy recovery potential; traffic flow and congestion probability represent the traffic density and congestion probability of the future road section, which determines the vehicle start-stop frequency and power demand fluctuation; traffic light timing represents the phase switching time of the traffic light ahead, which is used to predict the vehicle acceleration and deceleration requirements; obstacle movement trajectory represents the position, speed and trajectory of the obstacle within the prediction range, which affects the possibility of emergency braking or lane change of the vehicle;

[0055] The generated operating condition prediction results will be passed to the dynamic weight allocation module 200 for subsequent optimization calculations.

[0056] The dynamic weight allocation module 200 calculates the dynamic weight coefficients of different optimization objectives in real time based on the working condition prediction results, and generates an optimization problem description file containing weight coefficient constraint conditions. The optimization objectives include fuel economy, battery life maintenance and power response requirements, specifically including:

[0057] The dynamic weight allocation module 200 needs to consider multiple optimization objectives at the same time, mainly including fuel economy, battery life maintenance and power response requirements, among which fuel economy improves the overall efficiency of the system by reducing fuel consumption; battery life maintenance prolongs the service life of the battery by reasonably controlling the charging and discharging state of the battery; and power response requirements ensure that the vehicle can respond in time when needed and provide sufficient power support;

[0058] The dynamic weight allocation module 200 calculates the percentage difference between the fuel consumption reference value and the actual predicted fuel consumption according to the road slope change rate and the traffic light timing in the working condition prediction result as the fuel saving rate Fs. The specific calculation process is as follows:

[0059] Under the condition of driving at a constant speed on a flat road (e.g. 60km / h), obtain the fuel consumption benchmark value under this working condition through the standard test data or historical data of the engine;

[0060] The additional load of each road section is calculated based on the road slope change rate in the working condition prediction results; if the slope is large, the vehicle needs more fuel to overcome gravity, thereby increasing fuel consumption;

[0061] Predict the number of starts and stops at each traffic light based on the traffic light sequence, and estimate the additional fuel consumption for each start and stop; each start and stop consumes additional fuel, so this additional fuel consumption needs to be added to the total fuel consumption;

[0062] The effects of slope and start-stop are combined to obtain the actual predicted fuel consumption;

[0063] The percentage difference between the fuel consumption benchmark value and the actual predicted fuel consumption is taken as the fuel saving rate Fs.

[0064] According to the traffic flow and obstacle information in the working condition prediction results, the battery state of charge fluctuation suppression coefficient Sc is calculated; the specific calculation process is as follows:

[0065] According to the traffic flow in the working condition prediction results, the average acceleration and deceleration frequency of the vehicle is estimated; the greater the traffic flow, the more frequent the vehicle acceleration and deceleration, resulting in an increase in the number of battery charge and discharge times;

[0066] Predict the number of emergency braking or lane change events based on the obstacle movement trajectory in the obstacle information; each emergency braking or lane change will cause high-power discharge or charging of the battery, thereby increasing the fluctuation of the battery's state of charge;

[0067] According to the acceleration and deceleration frequency and the number of emergency events, a battery state of charge change sequence model for the future period is established; the change of the battery state of charge is calculated, and the inverse of its standard deviation is calculated and normalized to the [0,1] interval as the battery state of charge fluctuation suppression coefficient Sc.

[0068] According to the obstacle information and congestion probability in the working condition prediction results, the power response delay penalty term Dp is calculated. The specific calculation process is as follows:

[0069] According to the congestion probability, the average vehicle speed and acceleration demand are estimated. In congestion, the vehicle may need to accelerate or decelerate frequently, which increases the response time of the power system.

[0070] If an obstacle cut-in risk is detected, an emergency torque request is triggered; at this time, the power system needs to respond quickly and provide sufficient torque to avoid a collision; the response time of the power system depends on the performance of the motor and the battery voltage; the default target response time threshold is 100 milliseconds; if an emergency condition is predicted (such as an obstacle less than 10 meters away), the target response time threshold is shortened to 50 milliseconds;

[0071] If the actual response time is greater than the target response time threshold, the difference between the two is calculated and squared as the power response delay penalty term Dp. The larger the penalty term, the more serious the power response delay. If the actual response time is less than or equal to the target response time threshold, the power response delay penalty term Dp is 0.

[0072] The dynamic weight allocation module 200 uses a deep deterministic policy gradient algorithm to update the weight coefficient calculation strategy. The specific steps are as follows:

[0073] Define a reward function R to evaluate the effect of the current weight setting. The reward function R is expressed as:

[0074] R=w1Fs+w2Sc+w3Dp, where Fs is the fuel saving rate, calculated as the percentage difference between the actual predicted fuel consumption and the benchmark value; Sc is the battery state of charge fluctuation suppression coefficient, calculated as the normalized value of the inverse of the battery state of charge standard deviation; Dp is the power response delay penalty term, calculated as the square of the excess of the actual response time over the target response time threshold; w1 represents the fuel economy weight coefficient; w2 represents the battery life maintenance weight coefficient; w3 represents the power response penalty weight coefficient.

[0075] Through the deep deterministic policy gradient algorithm, the weight coefficient is dynamically adjusted and the weight calculation strategy is continuously updated to maximize the value of the reward function R;

[0076] To ensure that the optimization problem is feasible, the dynamic weight allocation module 200 defines the following constraints:

[0077] Normalization constraint: w1+w2+w3=1, and 0≤wi≤1;

[0078] Emergency working condition constraint: When the obstacle approaches the critical distance for emergency braking, the power response penalty weight coefficient w3 is automatically increased to the safety response threshold;

[0079] Traffic-related constraints: When the traffic density reaches the road load warning value, the battery life maintenance weight coefficient w2 is forced to rise to the congestion condition benchmark value;

[0080] Fuel-battery balance constraint: The ratio of the fuel economy weight coefficient w1 to the battery life maintenance weight coefficient w2 shall not exceed the maximum balance coefficient preset by the system.

[0081] The dynamic weight allocation module 200 encapsulates the reward function, weight coefficients and constraints into an optimization problem description file.

[0082] See also Figure 2 , the layered execution module 300 includes a rolling optimization layer unit 301, a dynamic compensation layer unit 302 and a real-time control layer unit 303, wherein:

[0083] The rolling optimization layer unit 301 in the layered execution module 300 receives the optimization problem description file and solves the global optimal power allocation benchmark within a fixed time window, specifically including:

[0084] The quadratic programming algorithm is used to handle the nonlinear optimization problem in the optimization problem description file, that is, to handle the fuel-battery balance constraint, and the problem is converted into convex optimization using quadratic programming to ensure fast solution;

[0085] Under the constraints in the optimization problem description file, the rolling optimization layer unit 301 combines the operating condition prediction results to calculate the global optimal power allocation benchmark, including the optimal benchmark value sequence of engine output power, motor torque distribution and battery charging and discharging power in future time periods, to achieve comprehensive optimization of energy consumption, life and performance.

[0086] The dynamic compensation layer unit 302 in the hierarchical execution module 300 generates a compensation correction item for dynamically adjusting the weight coefficient according to the deviation between the global optimal power allocation reference and the current vehicle state, specifically including:

[0087] Through the vehicle sensor data and the Internet of Vehicles data, the dynamic compensation layer obtains the current vehicle status information in real time, including but not limited to vehicle speed, acceleration, battery charge status, engine load, etc.;

[0088] Compare the current vehicle state with the global optimal power allocation benchmark and calculate the deviation of each parameter, for example, calculate the difference between the current engine output power and the benchmark value, the difference between the current motor torque and the benchmark value, etc.;

[0089] Based on the deviation, the dynamic compensation layer generates compensation correction items, which are intended to adjust the global optimal power allocation benchmark to make it more in line with the actual operating conditions of the current vehicle; the compensation correction items include fine-tuning of the engine output power, motor torque and battery charging and discharging power to ensure that the system can maintain optimal performance under actual driving conditions; and are used to dynamically adjust the weight coefficients.

[0090] The real-time control layer unit 303 in the hierarchical execution module 300 converts the compensation correction term into a power distribution instruction and synchronously updates the weight coefficient of the dynamic weight distribution module 200, specifically including:

[0091] Convert the compensation correction item into a specific power distribution instruction, including specific values ​​of engine output power, motor torque, and battery charging and discharging power;

[0092] The generated power distribution command is sent to the command execution system of the vehicle, including the engine, motor and battery management system; the actual execution result of the power distribution command, that is, the vehicle network data and the vehicle sensor data after the actual execution, is collected, and based on the difference between the actual execution result and the expected target, it is determined whether to execute the re-update of the weight coefficient in the dynamic weight distribution module 200. If an update is required, the role of the compensation correction term in indirectly adjusting the weight coefficient in the dynamic weight distribution module 200 is also explained.

[0093] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A hybrid power system energy management algorithm, characterized in that: The system comprises a working condition prediction module (100), a dynamic weight allocation module (200) and a hierarchical execution module (300), wherein: The operating condition prediction module (100) is used to integrate Internet of Vehicles data and vehicle-mounted sensor data, and generate multi-dimensional operating condition prediction results for future periods; The dynamic weight allocation module (200) calculates the dynamic weight coefficients of different optimization objectives in real time based on the working condition prediction results, and generates an optimization problem description file containing weight coefficient constraint conditions; The layered execution module (300) comprises a rolling optimization layer unit (301), a dynamic compensation layer unit (302) and a real-time control layer unit (303), wherein: The rolling optimization layer unit (301) receives the optimization problem description file and solves the global optimal power allocation benchmark within a fixed time window; The dynamic compensation layer unit (302) generates a compensation correction term including a weight coefficient for dynamically adjusting the weight coefficient according to the deviation between the global optimal power allocation reference and the current vehicle state; The real-time control layer unit (303) converts the compensation correction term into a power distribution instruction and synchronously updates the weight coefficient of the dynamic weight distribution module (200); through the actual execution feedback of the power distribution instruction and the dynamic weight re-update mechanism, a closed-loop dynamic correction link of the multi-objective optimization strategy is formed to ensure the real-time performance of the energy distribution strategy.

2. The hybrid power system energy management algorithm according to claim 1, characterized in that: The operating condition prediction module (100) specifically comprises: Obtaining Internet of Vehicles data in a fixed area ahead through Internet of Everything technology, wherein the Internet of Vehicles data includes a road slope change rate and a traffic light timing sequence; The millimeter-wave radar is used to detect the position and speed of obstacles to generate a motion trajectory, the visual sensor is used to identify the curvature of the lane line, and the Kalman filter algorithm is used to fuse the trajectory and curvature to form the vehicle-mounted sensor data; A spatiotemporal correlation matrix is ​​constructed to fuse the Internet of Vehicles data and the vehicle-mounted sensor data, wherein the spatiotemporal correlation matrix includes current environmental information and future road condition predictions.

3. The hybrid power system energy management algorithm according to claim 2, characterized in that: The spatiotemporal correlation matrix generates multi-dimensional working condition prediction results through long short-term memory network model analysis, specifically including: Road slope change rate, traffic flow and congestion probability, traffic light timing and obstacle movement trajectory; The road slope change rate includes real-time slope data within a fixed distance in the future; The traffic flow and congestion probability reflect the traffic density and start-stop frequency; The traffic light timing is used to predict the acceleration and deceleration requirements of vehicles; The obstacle motion trajectory includes the position, speed and trajectory of the obstacle within the prediction range.

4. The hybrid power system energy management algorithm according to claim 1, characterized in that: The optimization objectives in the dynamic weight allocation module (200) include fuel economy, battery life maintenance and power response requirements, and based on the optimization objectives and the operating condition prediction results, the fuel saving rate Fs, the battery state of charge fluctuation suppression coefficient Sc and the power response delay penalty term Dp are calculated.

5. The hybrid power system energy management algorithm according to claim 4, characterized in that: The calculation process of the fuel saving rate Fs specifically includes: Obtain fuel consumption benchmark value based on flat road and uniform speed conditions; Calculate additional load fuel consumption based on the road grade change rate; Estimate the number of starts and stops and additional fuel consumption based on traffic light timing; The percentage difference between the actual predicted fuel consumption and the benchmark value is calculated comprehensively as the fuel saving rate.

6. The hybrid power system energy management algorithm according to claim 4, characterized in that: The calculation process of the battery state of charge fluctuation suppression coefficient Sc specifically includes: Estimate the average acceleration and deceleration frequency based on traffic flow; Predict the number of emergency events based on obstacle movement trajectories; A battery state of charge variation sequence model is established and the inverse of the standard deviation is calculated, and the battery state of charge fluctuation suppression coefficient is obtained through normalization processing.

7. The hybrid power system energy management algorithm according to claim 4, characterized in that: The calculation process of the power response delay penalty term Dp specifically includes: Estimate acceleration demand based on congestion probability; Set target response time thresholds for emergency conditions; When the actual response time exceeds the target response time threshold, the square of the difference is calculated as the power response delay penalty term Dp.

8. The hybrid power system energy management algorithm according to claim 1, characterized in that: The dynamic weight allocation module (200) uses a deep deterministic policy gradient algorithm to calculate the weight coefficient, and its reward function R is: R=w1Fs+w2Sc+w3Dp, w1 represents the fuel economy weight coefficient; w2 represents the battery life maintenance weight coefficient; w3 represents the power response penalty weight coefficient.

9. The hybrid power system energy management algorithm according to claim 1, characterized in that: The rolling optimization layer unit (301) uses a quadratic programming algorithm to process nonlinear optimization problems, specifically including: Transform the fuel-battery balance constraint into a convex optimization problem; The optimal reference value sequence of engine output power, motor torque distribution and battery charging and discharging power in the future period is calculated as the global optimal power distribution benchmark.

10. The hybrid power system energy management algorithm according to claim 1, characterized in that: When the dynamic compensation layer unit (302) generates the compensation correction item, it obtains the current vehicle status information in real time through the vehicle-mounted sensor data and the vehicle network data, calculates the deviation between the current status parameters and the global optimal power allocation benchmark; generates fine-tuning instructions for the engine power, motor torque and battery charging and discharging power; the real-time control layer unit (303) converts the compensation correction item into a specific power distribution instruction and transmits it to the instruction execution system, collects the actual operation data after the instruction is executed, and determines whether to trigger the weight coefficient to be updated again based on the difference between the execution result and the expected target.

Images