Generator power supply system and method for hybrid vehicle

Abstract

The invention relates to the technical field of hybrid electric vehicles, in particular to a hybrid electric vehicle type generator power supply system and method, and the system comprises a multi-source data processing module, a driving condition prediction module, an efficiency solving module and a database. The multi-source data processing module collects working condition, energy state and environment data, and obtains clean data through a variational mode decomposition and wavelet threshold denoising combined algorithm; the driving condition prediction module constructs a dynamic graph based on a residual image neural network, integrates adjacent vehicle motion and gradient influence prediction working conditions, constructs a first optimization function in combination with hybrid electric vehicle performance requirements, and generates target power instructions of an engine and a generator set through an optimization algorithm. And the efficiency solving module is used for solving a second optimization function taking the maximum total efficiency of the system as a target and outputting the optimal torque and the rotating speed set value of the engine in real time. The efficiency and working condition adaptability of the power generation system can be effectively improved, the fuel economy is optimized, and the service life of the battery is prolonged.

Description

Technical Field

[0001] This invention relates to the field of hybrid electric vehicle technology, and in particular to a generator power supply system and method for a hybrid electric vehicle. Background Technology

[0002] Hybrid electric vehicles, with their dual power source of gasoline and electricity, have become a core vehicle type for reducing fuel consumption and carbon emissions. The generator power supply system, as the core of energy distribution, directly impacts the vehicle's power performance, fuel economy, and battery life. However, existing systems face multi-dimensional technical bottlenecks, including: Existing prediction methods rely heavily on historical data, fail to establish dynamic correlations between workshops, roads, and cloud computing, and ignore key factors such as traffic light status and slope constraints. As a result, the prediction results deviate significantly from the actual driving conditions, leading to poor adaptability of engine and generator set target power commands and frequent occurrences of power excess or deficiency. Efficiency solutions often rely on offline calibrated MAP graph queries, which cannot respond to changes in driving conditions in real time. Some optimization algorithms only focus on engine efficiency without integrating generator efficiency for collaborative optimization, and the solution methods are prone to getting trapped in local optima, failing to maximize the overall system efficiency. Long-term use leads to high fuel consumption and accelerated battery degradation due to overcharging and discharging. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a generator power supply system and method for hybrid electric vehicles, thereby resolving the aforementioned technical problems in the prior art.

[0004] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows: A generator power supply system for a hybrid vehicle includes a multi-source data processing module, a driving condition prediction module, an efficiency calculation module, and a database.

[0005] The multi-source data processing module collects real-time data on vehicle operating conditions, energy state, and environment during the driving process. It then uses a joint algorithm combining variational mode decomposition and wavelet threshold denoising to denoise the raw operating condition data, obtaining clean data. The specific steps are as follows: Set the acquisition frequency of each vehicle sensor, including vehicle speed sensor, engine speed sensor, engine torque sensor, generator output power sensor, battery SOC sensor, and throttle opening sensor. Based on the acquisition frequency, collect vehicle speed, engine speed, engine torque, generator power, battery SOC, throttle opening, and ambient temperature, and integrate them into raw vehicle data. Based on the collected raw vehicle data, a variational optimization objective function for variational mode decomposition is constructed. The construction logic of this optimization objective function is as follows: ,in To obtain N IMF components from the decomposition, The center frequency of each IMF component, For the Dirac function, For convolution operators, Let be the impulse response of the filter, where c is the imaginary unit. The term is a complex exponential term, and t is a time point; The objective function is optimized by iteratively solving the method of alternating directions of multipliers, and by introducing Lagrange multipliers and quadratic penalty terms, and updating them alternately until the preset number of iterations is reached. The original vehicle data is decomposed into N IMF components from high frequency to low frequency. Obtain the continuous mean square error and Pearson correlation coefficient of each IMF component and the original vehicle data, and extract the preset mean square error threshold and preset Pearson coefficient threshold of the database. If the continuous mean square error of the IMF component is less than the preset mean square error threshold and the absolute value of the Pearson correlation coefficient is less than the preset Pearson coefficient threshold, it is classified as a high-frequency noise-dominated IMF, and the other IMF components are classified as useful signal-dominated IMFs. Wavelet packet decomposition is performed on the identified high-frequency noise-dominant IMF to obtain wavelet coefficients in different frequency bands. The wavelet coefficients are then processed by an improved threshold function. The processing logic is as follows: ,in For symbolic functions, This is the adaptive shrinkage term, which is the adaptive shrinkage amount based on the Sigmoid function. The threshold is used to obtain the denoising coefficients and perform inverse wavelet transform to obtain the denoised IMF components. The denoised IMF components and the useful signal-dominated IMF components are superimposed to reconstruct clean data.

[0006] The driving condition prediction module establishes a dynamic graph of the workshop, road, and cloud topology through a residual graph neural network, incorporating the transmission of neighboring vehicle motion states and the influence of slope, to predict the driving condition results within a time range. Based on the driving condition results and the performance requirements of the hybrid vehicle, a first optimization function is constructed. The first optimization function is solved in a rolling manner to generate the target power command for the engine and generator set. The specific process is as follows: The residual graph neural network sets clean data as the vehicle node, obtains the total number of neighboring vehicles, the speed of neighboring vehicles, and the lateral distance between neighboring vehicles and the vehicle through navigation, vehicle networking and historical driving data, and sets them as neighboring vehicle nodes. It also obtains traffic light status, the distance between the vehicle and the traffic light and the remaining time of the traffic light, and sets them as traffic light nodes. Two edge types are set to represent the association between nodes, and their corresponding edge weights reflect the strength of the association. Specifically, edge type 1 represents the current vehicle node and its neighboring vehicle node, and the weight of the first edge is obtained. The calculation logic is as follows: ,in Let m be the vehicle speed at node m. For the vehicle's speed, is the normalization coefficient for the speed difference, and dc is the distance between the vehicle and node m. The maximum detection distance of the sensor is 2; edge type 2 represents the association between the vehicle node and the traffic light or ramp, and its calculation logic is as follows: ,in The distance between this vehicle and the traffic light. To determine the maximum recognition distance of traffic lights, a dynamic graph is constructed based on edge type and nodes. The dynamic graph is mapped to a unified dimension to obtain embedded features. The traffic wave propagation mechanism is embedded into the convolutional layer to obtain propagation features. Specifically, the traffic wave propagation mechanism is to pass on the motion features of neighboring vehicles through edge types, such as the deceleration of the preceding vehicle to the deceleration requirement of the current vehicle. The slope feature is directly transmitted to the embedded features of the current vehicle through the edge between the slope node and the current vehicle node. The propagation features are input into the fully connected layer and the working condition results within a preset time are output. The working condition results include predicted vehicle speed, predicted load power and predicted braking probability. The engine speed nc and engine torque Gs are acquired in real time, and the fuel consumption rate is obtained by fitting the engine speed and torque based on the engine universal characteristic curve. The calculation logic is as follows: Where w, x, and z are fitting coefficients, the battery charging current and degradation coefficient are obtained, and the battery health degradation rate is obtained based on the battery charging current and degradation coefficient, and the load power is predicted. and the target power of the engine and generator set The dynamic response deviation RU is obtained, and its calculation logic is as follows: The objective function is obtained by weighting the fuel consumption rate, battery health degradation rate, and power response deviation. The range of the operating condition results is then embedded into the objective function as a constraint to obtain the first optimization function. The whale population size is set to M. Each individual represents a set of optimization variables, including the target power of the engine and generator set, the target torque of the engine, the target speed of the engine, and the charging and discharging current of the battery. The minimum value of the first optimization function is set as the whale fitness value; the smaller the fitness value, the better the individual. The probability of the predation strategy is set to... ,when When the fitness value is less than 0.5, bubble web foraging is performed; otherwise, random search is performed. This process is repeated iteratively. When the difference between the optimal fitness values ​​of two adjacent iterations is less than or equal to the preset difference coefficient, the target power command for the engine and generator set is output.

[0007] The efficiency solution module receives the target power commands of the engine and generator set through the optimal operating point tracker and constructs a second optimization function. It then solves the second optimization function using the golden section search method, obtaining the optimal torque setpoint and optimal speed setpoint of the engine in real time. The specific steps are as follows: The optimal operating point tracker receives the target power commands from the engine and generator set, obtains the engine efficiency spectrum data preset in the database, and obtains the engine efficiency curve. The engine efficiency spectrum data is the core performance data of the engine power, essentially a correlation data between input parameters and output performance stored in a two-dimensional table format. Then, the generator efficiency and output power are fitted with a quadratic polynomial to obtain the generator efficiency curve. The engine efficiency curve and the generator efficiency curve are integrated to obtain the second optimization function. The search interval is initialized using the golden section search method to obtain the search start point and search end point. The first trial point and the second trial point are randomly set within the search interval, and the first target solution value and the second target solution value corresponding to the second optimization function at the trial point are calculated. The calculation steps are as follows: obtain the real-time speed value cr of the first trial point, and calculate the estimated engine torque value Te through the power balance constraint derivation formula. The calculation logic is as follows: Where 9550 is the conversion factor for speed, torque, and power. The target power of the engine and generator set output by the whale optimization algorithm. To preset the initial engine efficiency, The efficiency of the generator in converting the engine's mechanical power into electrical energy; based on real-time speed and torque estimates, the actual engine efficiency is obtained by querying engine efficiency map data through bilinear interpolation. The first objective solution value GM is calculated using the actual engine efficiency. The calculation logic is as follows: Then, calculate the second target solution value for the second test point using the same steps. The first and second target solutions are compared and processed in two cases: Case 1: If the first target solution is greater than the second target solution, the optimal solution is determined to be within the interval between the search start point and the second target solution, and the first trial point is retained for the next iteration, and a new first target solution is recalculated; Case 2: If the first target solution is less than or equal to the second target solution, the optimal solution is determined to be within the interval between the first trial point and the search end point, and the second trial point is retained for the next iteration, and a new second target solution is recalculated. Iteration is performed sequentially, and iteration stops when the interval width is less than the preset width threshold. The midpoint of the interval is taken as the optimal speed setting value. The optimal speed setting value is substituted into the power balance constraint formula, and the optimal torque setting value is obtained by interpolation using engine efficiency spectrum data.

[0008] This invention also provides a generator power supply method for hybrid electric vehicles, the specific steps of which are as follows: Step 1: Multi-source data processing: By collecting real-time data on vehicle operating conditions, energy status, and environment during the driving process, the original operating condition data is denoised using a joint algorithm of variational mode decomposition and wavelet threshold denoising to obtain clean operating condition data. Step 2, Operating Condition Prediction: A dynamic graph of the workshop, road and cloud topology is established through a residual graph neural network, incorporating the motion state transmission of neighboring vehicles and the influence of slope, to predict the operating condition results within the time range. Based on the operating condition results and the performance requirements of the hybrid vehicle, a first optimization function is constructed, and the first optimization function is solved in a rolling manner to generate the target power command of the engine and generator set. Step 3, Efficiency Solution: Receive the target power commands of the engine and generator set through the optimal operating point tracker and construct the second optimization function. Solve the second optimization function using the golden section search method to obtain the optimal torque setpoint and optimal speed setpoint of the engine in real time.

[0009] Compared with the prior art, the beneficial effects of the present invention are as follows: The efficiency solution module breaks through the limitations of traditional offline MAP graph queries, integrates engine and generator efficiencies to construct a second optimization function, and uses the golden section search method to accurately locate the global optimal operating point, reducing efficiency losses caused by local optimization and keeping the overall system efficiency at a high level. This optimization logic can reduce the engine's running time in the inefficient range, reduce battery damage due to overcharging and discharging, and balance fuel economy and battery life.

[0010] The driving condition prediction module constructs a dynamic graph through a residual graph neural network, incorporating factors from actual driving scenarios such as the motion transmission of neighboring vehicles, traffic light status, and slope constraints into the prediction logic, making the driving condition prediction results more consistent with the real driving conditions. Then, the Whale Optimization Algorithm solves the first multi-objective optimization function, and the generated engine and generator set target power commands can better adapt to the current driving conditions, reducing the situation of power excess. Attached Figure Description

[0011] Figure 1 This is a module connection diagram of the present invention.

[0012] Figure 2 This is a diagram illustrating the method steps of the present invention. Detailed Implementation

[0013] The technical solution of the present invention will be further described in a non-limiting manner below with reference to the accompanying drawings and specific embodiments.

[0014] Example 1: As Figure 1 As shown, this embodiment provides a generator power supply system for a hybrid vehicle, including a multi-source data processing module, a driving condition prediction module, an efficiency solution module, and a database.

[0015] The multi-source data processing module collects real-time data on vehicle operating conditions, energy state, and environment during the driving process. It then uses a joint algorithm combining variational mode decomposition and wavelet threshold denoising to denoise the raw operating condition data, obtaining clean data. The specific steps are as follows: Set the acquisition frequency of each vehicle sensor, including vehicle speed sensor, engine speed sensor, engine torque sensor, generator output power sensor, battery SOC sensor, and throttle opening sensor. Based on the acquisition frequency, collect vehicle speed, engine speed, engine torque, generator power, battery SOC, throttle opening, and ambient temperature, and integrate them into raw vehicle data. Based on the collected raw vehicle data, a variational optimization objective function for variational mode decomposition is constructed. The construction logic of this optimization objective function is as follows: ,in To obtain N IMF components from the decomposition, The center frequency of each IMF component, For the Dirac function, For convolution operators, Let be the impulse response of the filter, where c is the imaginary unit. The term is a complex exponential term, and t is a time point; The objective function is optimized by iteratively solving the method of alternating directions of multipliers, and by introducing Lagrange multipliers and quadratic penalty terms, and updating them alternately until the preset number of iterations is reached. The original vehicle data is decomposed into N IMF components from high frequency to low frequency. Obtain the continuous mean square error and Pearson correlation coefficient of each IMF component and the original vehicle data, and extract the preset mean square error threshold and preset Pearson coefficient threshold of the database. If the continuous mean square error of the IMF component is less than the preset mean square error threshold and the absolute value of the Pearson correlation coefficient is less than the preset Pearson coefficient threshold, it is classified as a high-frequency noise-dominated IMF, and the other IMF components are classified as useful signal-dominated IMFs. Wavelet packet decomposition is performed on the identified high-frequency noise-dominant IMF to obtain wavelet coefficients in different frequency bands. The wavelet coefficients are then processed by an improved threshold function. The processing logic is as follows: ,in For symbolic functions, This is the adaptive shrinkage term, which is the adaptive shrinkage amount based on the Sigmoid function. The threshold is used to obtain the denoising coefficients, and an inverse wavelet transform is performed to obtain the denoised IMF components. The denoised IMF components and the IMF components dominated by the useful signal are superimposed to reconstruct clean data. It should be noted that this step achieves a balance between hard and soft thresholds by using variational mode decomposition and wavelet thresholding in conjunction with a continuously differentiable improved wavelet threshold function. This avoids the constant deviation of the soft threshold and suppresses the oscillation of the hard threshold. A dual criterion of continuous mean square error and Pearson correlation coefficient is introduced. For example, in an intense driving cycle, some high-frequency IMFs may have a high Pearson correlation coefficient because they contain real driving information and are therefore retained. This ensures the context awareness capability of the denoising process and is suitable for processing real car data with different driving styles.

[0016] The driving condition prediction module establishes a dynamic graph of the workshop, road, and cloud topology through a residual graph neural network, incorporating the transmission of neighboring vehicle motion states and the influence of slope, to predict the driving condition results within a time range. Based on the driving condition results and the performance requirements of the hybrid vehicle, a first optimization function is constructed. The first optimization function is solved in a rolling manner to generate the target power command for the engine and generator set. The specific process is as follows: The residual graph neural network sets clean data as the vehicle node. It obtains the total number of neighboring vehicles, the speed of neighboring vehicles, and the lateral distance between neighboring vehicles and the vehicle through navigation, vehicle networking, and historical driving data, and sets them as neighboring vehicle nodes. It obtains traffic light status, the distance between the vehicle and the traffic light, and the remaining time of the traffic light, and sets them as traffic light nodes. It should be noted that the traffic light status is 0=red, 1=green, and 2=yellow. The slope gradient and slope length are set as slope nodes. Two edge types are set to represent the association between nodes, and their corresponding edge weights reflect the strength of the association. Specifically, edge type 1 represents the current vehicle node and its neighboring vehicle node, and the weight of the first edge is obtained. The calculation logic is as follows: ,in Let m be the vehicle speed at node m. For the vehicle's speed, is the normalization coefficient for the speed difference, and dc is the distance between the vehicle and node m. The maximum detection distance of the sensor is 2; edge type 2 represents the association between the vehicle node and the traffic light or ramp, and its calculation logic is as follows: ,in The distance between this vehicle and the traffic light. To determine the maximum recognition distance of traffic lights, a dynamic graph is constructed based on edge type and nodes. It should be noted that the preset time is based on the sensor data to update node attributes and edge weights. If a node is out of the detection range (such as when the vehicle in front leaves), the corresponding node and edge are removed. Newly detected entities are added with nodes and edges to ensure that the topology reflects the actual scene in real time. The dynamic graph is mapped to a unified dimension to obtain embedded features. The traffic wave propagation mechanism is embedded into the convolutional layer to obtain propagation features. Specifically, the traffic wave propagation mechanism is to pass on the motion features of neighboring vehicles through edge types, such as the deceleration of the preceding vehicle to the deceleration requirement of the current vehicle. The slope feature is directly transmitted to the embedded features of the current vehicle through the edge between the slope node and the current vehicle node. The propagation features are input into the fully connected layer and the working condition results within a preset time are output. The working condition results include predicted vehicle speed, predicted load power and predicted braking probability. It should be noted that the neural network was trained based on 1,000 sets of real-world test data from multiple scenarios, such as urban congestion, highway cruising, and mountain road climbing. The loss function was mean squared error, the optimizer was Adam, and there were several training epochs. The engine speed nc and engine torque Gs are acquired in real time, and the fuel consumption rate is obtained by fitting the engine speed and torque based on the engine universal characteristic curve. The calculation logic is as follows: Where w, x, and z are fitting coefficients, the battery charging current and degradation coefficient are obtained, and the battery health degradation rate is obtained based on the battery charging current and degradation coefficient, and the load power is predicted. and the target power of the engine and generator set The dynamic response deviation RU is obtained, and its calculation logic is as follows: The objective function is obtained by weighting the fuel consumption rate, battery health degradation rate, and power response deviation. The range of the operating condition results is then embedded into the objective function as a constraint to obtain the first optimization function. It should be noted that the following section uses a whale optimization algorithm to simulate whales' prey encirclement, bubble-net foraging, and random search behaviors to solve the first optimization problem, obtaining the optimal target power command and related parameters. Specifically, the whale population size is set to M, and each individual represents a set of optimization variables, including the target power of the engine and generator set, the target torque of the engine, the target speed of the engine, and the charging and discharging current of the battery. The minimum value of the first optimization function is the whale fitness value; the smaller the fitness value, the better the individual. The probability of the predation strategy is set to... ,when When the fitness value is less than 0.5, bubble web foraging is performed; otherwise, random search is performed. This process is repeated iteratively. When the difference between the optimal fitness values ​​of two adjacent iterations is less than or equal to the preset difference coefficient, the target power command for the engine and generator set is output.

[0017] The efficiency solution module receives the target power commands of the engine and generator set through the optimal operating point tracker and constructs a second optimization function. It then solves the second optimization function using the golden section search method, obtaining the optimal torque setpoint and optimal speed setpoint of the engine in real time. The specific steps are as follows: It should be noted that the efficiency solution module dynamically solves and tracks the optimal combination of engine torque and engine speed that maximizes the overall efficiency of the entire system based on the power command of the driving condition prediction module. The optimal operating point tracker receives the target power commands from the engine and generator set, obtains the engine efficiency spectrum data preset in the database, and obtains the engine efficiency curve. The engine efficiency spectrum data is the core performance data of the engine power, essentially a correlation data between input parameters and output performance stored in a two-dimensional table format. Then, the generator efficiency and output power are fitted with a quadratic polynomial to obtain the generator efficiency curve. The engine efficiency curve and the generator efficiency curve are integrated to obtain the second optimization function. The search interval is initialized using the golden section search method to obtain the search start point and search end point. The first trial point and the second trial point are randomly set within the search interval, and the first target solution value and the second target solution value corresponding to the second optimization function at the trial point are calculated. The calculation steps are as follows: obtain the real-time speed value cr of the first trial point, and calculate the estimated engine torque value Te through the power balance constraint derivation formula. The calculation logic is as follows: Where 9550 is the conversion factor for speed, torque, and power. The target power of the engine and generator set output by the whale optimization algorithm. To preset the initial engine efficiency, The efficiency of the generator in converting the engine's mechanical power into electrical energy; based on real-time speed and torque estimates, the actual engine efficiency is obtained by querying engine efficiency map data through bilinear interpolation. The first objective solution value GM is calculated using the actual engine efficiency. The calculation logic is as follows: Then, calculate the second target solution value for the second test point using the same steps. The first and second target solutions are compared and processed in two cases: Case 1: If the first target solution is greater than the second target solution, the optimal solution is determined to be within the interval between the search start point and the second target solution, and the first trial point is retained for the next iteration, and a new first target solution is recalculated; Case 2: If the first target solution is less than or equal to the second target solution, the optimal solution is determined to be within the interval between the first trial point and the search end point, and the second trial point is retained for the next iteration, and a new second target solution is recalculated. Iteration is performed sequentially, and iteration stops when the interval width is less than the preset width threshold. The midpoint of the interval is taken as the optimal speed setting value. The optimal speed setting value is substituted into the power balance constraint formula, and the optimal torque setting value is obtained by interpolation using engine efficiency spectrum data.

[0018] Example 2: As Figure 2 As shown, this embodiment of the invention provides a generator power supply method for a hybrid vehicle, the specific steps of which are as follows: Step 1: Multi-source data processing: By collecting real-time data on vehicle operating conditions, energy status, and environment during the driving process, the original operating condition data is denoised using a joint algorithm of variational mode decomposition and wavelet threshold denoising to obtain clean operating condition data. Step 2, Operating Condition Prediction: A dynamic graph of the workshop, road and cloud topology is established through a residual graph neural network, incorporating the motion state transmission of neighboring vehicles and the influence of slope, to predict the operating condition results within the time range. Based on the operating condition results and the performance requirements of the hybrid vehicle, a first optimization function is constructed, and the first optimization function is solved in a rolling manner to generate the target power command of the engine and generator set. Step 3, Efficiency Solution: Receive the target power commands of the engine and generator set through the optimal operating point tracker and construct the second optimization function. Solve the second optimization function using the golden section search method to obtain the optimal torque setpoint and optimal speed setpoint of the engine in real time.

[0019] The variational mode decomposition optimization objective function, improved wavelet threshold processing function, edge weight calculation logic, fuel consumption rate fitting formula, power response deviation calculation formula, and engine torque estimation formula involved in this invention have all undergone dimensionless processing, retaining only the numerical values ​​for calculation. Dimensionless processing can employ conventional techniques in the field of hybrid vehicle data processing, such as standardization and normalization; this invention does not limit this to a single method. The formulas are obtained through real-world data collection in various scenarios, including urban congestion, highway cruising, and mountain road climbing, combined with software simulation and iterative optimization. This ensures accurate alignment with the parameter correlation patterns of "sensor data - operating condition characteristics - energy distribution" during vehicle operation. The preset parameters in the formulas (such as IMF component classification threshold, whale optimization population size, and golden section search interval threshold) can be flexibly set by those skilled in the art based on specific application conditions such as vehicle configuration, sensor accuracy, and driving scenarios.

[0020] The technical solutions of this invention can be implemented through software, hardware, firmware, or any combination thereof. When implemented in hardware, the sensor units (vehicle speed, rotational speed, torque, etc.) of the multi-source data processing module can be directly integrated into the vehicle's existing sensing system, and the computing units of the driving condition prediction module and efficiency calculation module can be deployed based on the onboard multi-core main control MCU. When implemented in software, it can be based on the vehicle's embedded system or cloud server, and can be fully or partially embodied as a computer program product. The computer program product includes one or more computer instructions or computer programs, which are stored in a computer-readable storage medium that complies with vehicle data security specifications. When the computer device (which can be an onboard dedicated control computer, a cloud server cluster, or an MCU processor) loads and executes the computer instructions or computer programs, it performs all or part of the full process or core functions described in this invention embodiment. The computer instructions can be transmitted from one computer-readable storage medium to another via the onboard CAN bus or wireless communication methods (such as vehicle networking, infrared, microwave, etc.), or from a cloud data center or vehicle server to the onboard control terminal.

[0021] In this embodiment of the invention, the sequence numbers of each core process, such as multi-source data acquisition and denoising, operating condition prediction and optimization, and optimal operating point solution, are only used to distinguish different operation links and do not represent the order of execution. The execution order of each process should be based on its functional logic (such as data processing should precede operating condition prediction and efficiency solution should be based on the target power command) and should not constitute any limitation on the actual application process of this embodiment of the invention.

[0022] Those skilled in the art should understand that the joint denoising unit in the multi-source data processing module, the Res-GNN dynamic graph construction and whale optimization solution unit in the driving condition prediction module, the golden section search unit in the efficiency solution module, and related algorithm steps described in the embodiments of the present invention can be implemented through electronic hardware (such as dedicated ASIC chips) or a combination of computer software and electronic hardware (such as algorithms deployed on FPGA chips). Depending on the application scenario of the technical solution (such as different displacement hybrid vehicle models, different road condition requirements) and design constraints (such as hardware cost, real-time requirements), those skilled in the art can adopt different implementation methods to complete the functions, but their implementation should not exceed the protection scope of the present invention.

[0023] The system device described in this embodiment of the invention is logically divided into "modules and functional units," and can be adjusted according to the vehicle hardware deployment requirements in actual implementation. For example, the preprocessing unit of the multi-source data processing module and the driving condition prediction module can be integrated into the vehicle local control terminal, or some optimization functions of the driving condition prediction module can be migrated to the cloud server. Some non-core auxiliary functions can be ignored according to actual needs, or can be executed in conjunction with the vehicle's existing control system (such as engine ECU and battery management system). The coupling, direct coupling, or communication connection between the modules and units can be achieved through a dedicated vehicle communication interface, which can be an electrical connection (such as data cable transmission), a mechanical connection (such as the physical interface between the sensor and the control terminal), or other connection forms adapted to the vehicle scenario.

[0024] The units described as separate components (such as the Res-GNN computation unit of the driving condition prediction module and the optimal operating point tracking unit of the efficiency solution module) can be physically separate components (such as independent computing chips and dedicated sensors) or physically non-separate components (such as different functional modules integrated into the same MCU). The components shown as units (such as the wavelet packet decomposition unit of the multi-source data processing module) can be physical units (such as dedicated signal processing modules) or non-physical units (such as algorithm modules deployed on virtual servers), meaning they can be located in one place (such as the vehicle engine compartment electronic control box) or distributed across multiple network units (such as vehicle-mounted and cloud-based collaboration). Depending on actual needs such as driving safety, data processing efficiency, and cost control, some or all of the units can be selected to achieve the objectives of the embodiments of this invention.

[0025] In the embodiments of the present invention, each functional unit can be integrated into a processing unit (such as an integrated vehicle control module), or each unit can exist physically separately (such as an independent sensor acquisition unit), or two or more units can be integrated into one unit (such as integrating the edge weight calculation and dynamic graph construction unit into the same computing module).

[0026] If the aforementioned functions (such as joint denoising, Res-GNN condition prediction, and golden section search for optimal operating points) are implemented as software functional units and integrated as independent modules into the hybrid vehicle control system, they can be stored in a computer-readable storage medium adapted to the vehicle environment. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, can be embodied as a computer software product, stored in a computer-readable storage medium conforming to vehicle data security standards, containing several instructions for causing a computer device (which may be an on-board control computer, cloud server, MCU processor, etc.) to execute all or part of the steps of the methods described in the embodiments of this invention (such as IMF component classification, dynamic graph construction, optimization function solving, and optimal speed and torque calculation). The computer-readable storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and vehicle-specific solid-state storage media, etc., which are media capable of storing program code and adapted to the vehicle operating environment.

[0027] The above are merely preferred embodiments of the present invention, and the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the multi-source data joint denoising-Res-GNN working condition prediction-whale optimization and golden section search collaborative optimization techniques disclosed in the present invention, such as module function optimization, algorithm parameter adjustment, and hardware interface adaptation, should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A generator power supply system for a hybrid vehicle, comprising a multi-source data processing module, a driving condition prediction module, an efficiency calculation module, and a database, characterized in that: The multi-source data processing module collects real-time data on vehicle operating conditions, energy status, and environment during the driving process. It then uses a combined algorithm of variational mode decomposition and wavelet threshold denoising to denoise the original operating condition data, resulting in clean operating condition data. The driving condition prediction module establishes a dynamic graph of the workshop, road and cloud topology through residual graph neural network, incorporates the transmission of the motion state of neighboring vehicles and the influence of slope, predicts the working condition results within the time range, constructs the first optimization function based on the working condition results and the performance requirements of hybrid electric vehicles, and solves the first optimization function by the whale optimization algorithm to generate the target power command of the engine and generator set. The efficiency solution module receives the target power commands of the engine and generator set through the optimal operating point tracker and constructs a second optimization function. It then solves the second optimization function using the golden section search method to obtain the optimal torque setpoint and optimal speed setpoint of the engine in real time.

2. The generator power supply system for a hybrid vehicle according to claim 1, characterized in that, The efficiency solving module receives the target power commands of the engine and generator set through the optimal operating point tracker and constructs a second optimization function. It then solves the second optimization function using the golden section search method to obtain the optimal torque setpoint and optimal speed setpoint of the engine in real time. The specific steps are as follows: The optimal operating point tracker receives the target power command of the engine and generator set, obtains the engine efficiency spectrum data preset in the database, obtains the engine efficiency curve, and then performs a quadratic polynomial fitting between the generator efficiency and the output power to obtain the generator efficiency curve. The engine efficiency curve and the generator efficiency curve are integrated to obtain the second optimization function. The search interval is initialized by the golden section search method to obtain the search start point and search end point. The first trial point and the second trial point are randomly set within the search interval, and the first objective solution value and the second objective solution value corresponding to the second optimization function of the trial point are calculated. The calculation steps are as follows: obtain the real-time speed value of the first trial point, and calculate the estimated value of engine torque through the power balance constraint derivation formula. Based on real-time speed and torque estimates, the actual engine efficiency is obtained by querying engine efficiency map data through bilinear interpolation. The first target solution value is then calculated using the actual engine efficiency. The second target solution value for the second test point is then calculated using the same steps. The first and second target solutions are compared and processed in two cases: Case 1: If the first target solution is greater than the second target solution, the optimal solution is determined to be within the interval between the search start point and the second target solution, and the first trial point is retained for the next iteration, and a new first target solution is recalculated; Case 2: If the first target solution is less than or equal to the second target solution, the optimal solution is determined to be within the interval between the first trial point and the search end point, and the second trial point is retained for the next iteration, and a new second target solution is recalculated. Iteration is performed sequentially, and iteration stops when the interval width is less than the preset width threshold. The midpoint of the interval is taken as the optimal speed setting value. The optimal speed setting value is substituted into the power balance constraint formula, and the optimal torque setting value is obtained by interpolation using engine efficiency spectrum data.

3. The generator power supply system for a hybrid vehicle according to claim 1, characterized in that, The multi-source data processing module collects real-time data on the vehicle's operating conditions, energy status, and environment during its operation. The specific steps are as follows: The sampling frequency of each vehicle sensor is set. The sensors include vehicle speed sensor, engine speed sensor, engine torque sensor, generator output power sensor, battery SOC sensor, and throttle opening sensor. Based on the sampling frequency, vehicle speed, engine speed, engine torque, generator power, battery SOC, throttle opening, and ambient temperature are collected and integrated into raw vehicle data.

4. The generator power supply system for a hybrid vehicle according to claim 3, characterized in that, The multi-source data processing module uses a joint algorithm of variational mode decomposition and wavelet threshold denoising to denoise the original operating condition data, obtaining clean data. Specifically: Based on the collected raw vehicle data, a variational optimization objective function of variational mode decomposition is constructed. The optimization objective function is solved iteratively by the alternating direction method of multipliers. Lagrange multipliers and a quadratic penalty term are introduced and alternately updated until the preset number of iterations is reached. The raw vehicle data is decomposed into N IMF components from high frequency to low frequency. Obtain the continuous mean square error and Pearson correlation coefficient of each IMF component and the original vehicle data, and extract the preset mean square error threshold and preset Pearson coefficient threshold of the database. If the continuous mean square error of the IMF component is less than the preset mean square error threshold and the absolute value of the Pearson correlation coefficient is less than the preset Pearson coefficient threshold, it is classified as a high-frequency noise-dominated IMF, and the other IMF components are classified as useful signal-dominated IMFs. Wavelet packet decomposition is performed on the identified high-frequency noise-dominant IMF to obtain wavelet coefficients in different frequency bands. The wavelet coefficients are processed by improving the threshold function to obtain denoised coefficients, and inverse wavelet transform is performed to obtain denoised IMF components. The denoised IMF components are superimposed with the useful signal-dominant IMF components to reconstruct clean data.

5. The generator power supply system for a hybrid vehicle according to claim 1, characterized in that, The driving condition prediction module establishes a dynamic graph of the workshop, road, and cloud topology using a residual graph neural network, incorporating the transmission of adjacent vehicle motion states and the influence of slope, to predict the driving condition results within a time range, specifically as follows: The residual graph neural network sets clean data as the vehicle node, obtains the total number of neighboring vehicles, the speed of neighboring vehicles, and the lateral distance between neighboring vehicles and the vehicle through navigation, vehicle networking and historical driving data, and sets them as neighboring vehicle nodes. It also obtains traffic light status, the distance between the vehicle and the traffic light and the remaining time of the traffic light, and sets them as traffic light nodes. Two edge types are set to represent the relationship between nodes. Specifically, edge type 1 is the relationship between the current vehicle node and the neighboring vehicle node, and the weight of the first edge is obtained. Edge type 2 is the relationship between the current vehicle node and traffic lights or ramps, and the weight of the second edge is obtained. A dynamic graph is constructed based on the edge types and nodes. The dynamic graph is mapped to a unified dimension to obtain embedded features. The traffic wave propagation mechanism is embedded into the convolutional layer to obtain propagation features. Specifically, the traffic wave propagation mechanism transmits the motion features of neighboring vehicles through edge types. The slope features are directly transmitted to the embedded features of the vehicle through the edge between the slope node and the vehicle node. The propagation features are input into the fully connected layer and the operating conditions results within a preset time are output. The operating conditions results include predicted vehicle speed, predicted load power, and predicted braking probability.

6. The generator power supply system for a hybrid vehicle according to claim 5, characterized in that, The driving condition prediction module constructs a first optimization function based on the operating condition results and the performance requirements of the hybrid vehicle, and solves the first optimization function in a rolling manner to generate the target power command for the engine and generator set, specifically: The engine speed and torque are acquired in real time, and the fuel consumption rate is obtained by fitting the engine speed and torque based on the universal characteristic curve of the engine. Then, the battery charging current and attenuation coefficient are acquired, and the battery health attenuation rate is obtained based on the battery charging current and attenuation coefficient. The power response deviation is obtained by predicting the load power and the target power of the engine and generator set. The fuel consumption rate, battery health attenuation rate and power response deviation are weighted and calculated to obtain the objective function. The range value of the working condition result is embedded as a constraint condition into the objective function to obtain the first optimization function.

7. The generator power supply system for a hybrid vehicle according to claim 6, characterized in that, The driving condition prediction module performs a rolling solution to the first optimization function to generate target power commands for the engine and generator set, specifically: The whale population size is set, and each individual represents a set of optimization variables, including target power of the engine and generator set, target engine torque, target engine speed, and battery charging / discharging current. The minimum value of the first optimization function is set as the whale fitness value; the smaller the fitness value, the better the individual. The probability of the predation strategy is set as follows. ,when When the fitness value is less than 0.5, bubble web foraging is performed; otherwise, random search is performed. This process is repeated iteratively. When the difference between the optimal fitness values ​​of two adjacent iterations is less than or equal to the preset difference coefficient, the target power command for the engine and generator set is output.

8. A generator power supply method for a hybrid vehicle, employing the generator power supply system for a hybrid vehicle as described in any one of claims 1-7, characterized in that, include: Step 1: Multi-source data processing: By collecting real-time data on vehicle operating conditions, energy status, and environment during the driving process, the original operating condition data is denoised using a joint algorithm of variational mode decomposition and wavelet threshold denoising to obtain clean operating condition data. Step 2, Operating Condition Prediction: A dynamic graph of the workshop, road and cloud topology is established through a residual graph neural network, incorporating the motion state transmission of neighboring vehicles and the influence of slope, to predict the operating condition results within the time range. Based on the operating condition results and the performance requirements of the hybrid vehicle, a first optimization function is constructed, and the first optimization function is solved in a rolling manner to generate the target power command of the engine and generator set. Step 3, Efficiency Solution: Receive the target power commands of the engine and generator set through the optimal operating point tracker and construct the second optimization function. Solve the second optimization function using the golden section search method to obtain the optimal torque setpoint and optimal speed setpoint of the engine in real time.

Images