An electric light truck city cruise control method and system

Abstract

The application provides an electric light truck urban cruise control method and system, which comprises the following steps: based on a federal learning architecture, the global parameters of an urban traffic cloud platform are updated cooperatively through a federal average algorithm; a working condition characteristic vector library covering global urban roads is constructed according to the global parameters, a migration learning algorithm is used to migrate the speed trajectory planning experience under the historical optimal working condition to the inside of the working condition characteristic vector library, and a source domain-target domain adaptation model is constructed in combination with the remaining battery capacity and power system state of the electric light truck; a speed trajectory initial framework corresponding to the electric light truck is output through the source domain-target domain adaptation model, and a corresponding optimization objective function is constructed according to the speed trajectory initial framework; preset constraint conditions are called out, and the optimal speed trajectory is calculated according to the constraint conditions and the optimization objective function, so that the corresponding urban cruise is completed according to the optimal speed trajectory. The application can effectively improve the cruise control efficiency of the electric light truck.

Description

Technical Field

[0001] This invention relates to the field of automotive technology, and in particular to a method and system for controlling urban cruise control of an electric light truck. Background Technology

[0002] With the rapid development of intelligent connected vehicle technology, energy saving and efficient operation have become the core requirements for technological innovation in electric light trucks. Electric light trucks are widely used in urban logistics and short-distance transportation, and their urban cruise control performance directly affects operating costs and traffic efficiency. Therefore, developing urban cruise control methods and systems for electric light trucks that are adapted to urban conditions has become an important focus of the industry.

[0003] Among them, existing cruise control and adaptive cruise control (ACC) systems have been widely used in high-speed constant speed conditions. However, the urban roads where electric light trucks mainly operate have characteristics such as frequent traffic lights, large fluctuations in traffic flow, and complex road topology. The control logic of traditional cruise systems cannot adapt to this dynamic scenario, resulting in limited energy-saving effect and control accuracy, making it difficult to meet the needs of urban operation.

[0004] Furthermore, a core deficiency in existing cruise control methods and systems lies in their failure to fully utilize the dynamic traffic light timings and refined road topology information contained in high-precision maps. This information is crucial for optimizing driving speed trajectories in urban conditions. Existing technologies, unable to effectively utilize this information, struggle to plan optimal speed trajectories, failing to avoid increased energy consumption due to frequent acceleration and deceleration, and reducing the communication efficiency of vehicle-to-infrastructure (V2I) communication. This hinders the widespread adoption of urban cruise control technology for electric light trucks. Summary of the Invention

[0005] Based on this, the purpose of the present invention is to provide an urban cruise control method and system for electric light trucks, so as to solve the problem that the existing technology cannot plan the optimal speed trajectory according to urban working conditions, which leads to a reduction in communication efficiency.

[0006] The first aspect of the present invention proposes:

[0007] A method for controlling urban cruise control of an electric light truck, wherein the method includes:

[0008] A federated learning architecture for the vehicle terminal of an electric light truck and the urban traffic cloud platform is constructed. Based on the federated learning architecture, the global parameters of the urban traffic cloud platform are updated collaboratively through a federated averaging algorithm.

[0009] Based on the global parameters, a working condition feature vector library covering the entire urban road network is constructed. Simultaneously, a transfer learning algorithm is used to transfer the speed trajectory planning experience under the historical best working conditions to the working condition feature vector library. Combined with the remaining battery power and power system status of the electric light truck, a source domain-target domain adaptation model is constructed.

[0010] The source-target domain adaptation model outputs an initial speed trajectory framework corresponding to the electric light truck, and the corresponding optimization objective function is constructed simultaneously based on the initial speed trajectory framework.

[0011] The system retrieves preset constraints and simultaneously calculates the optimal speed trajectory based on the constraints and the optimization objective function, thereby completing the corresponding urban cruise based on the optimal speed trajectory.

[0012] The beneficial effects of this invention are as follows: This technical solution constructs a federated learning architecture between an electric light truck onboard terminal and an urban traffic cloud platform, and relies on a federated averaging algorithm to achieve distributed collaborative updating of global parameters, effectively avoiding redundant overhead and bandwidth waste in centralized data transmission, and significantly improving communication efficiency; based on global parameters, a feature vector library of urban road conditions covering the entire domain is built, and historical optimal planning experience is imported by combining transfer learning, along with a source domain-target domain adaptation model constructed with remaining battery power and power system status, to accurately output and optimize the optimal speed trajectory, which not only solves the problem that existing technologies cannot adapt to complex urban conditions, but also ensures the stability and economy of electric light truck urban cruising.

[0013] Furthermore, the step of constructing the source domain-target domain adaptation model by combining the remaining battery power and power system status of the electric light truck includes:

[0014] The time-series variation curves of the remaining battery power and the system parameters of the power system of the electric light truck under different driving conditions are collected, and the time-series dependency relationship between the time-series variation curves and the system parameters is simultaneously mined through a long short-term memory network.

[0015] By introducing battery aging coefficient and power system health index, the time series data corresponding to the time series dependency relationship is subjected to feature enhancement processing to generate the corresponding target domain dynamic feature set.

[0016] Using the historical best operating condition data in the operating condition feature vector library as the source domain data, the source domain-target domain adaptation model is constructed synchronously based on the source domain data and the target domain dynamic feature set.

[0017] Furthermore, the step of constructing the source domain-target domain adaptation model based on the source domain data and the target domain dynamic feature set includes:

[0018] The source domain data is decomposed into a speed trajectory feature layer, an energy consumption feature layer, and an operating environment feature layer. Simultaneously, the target domain dynamic feature set is decomposed into a time-series power consumption feature layer, a power system state layer, and a health index layer. The correlation degree of each layer's features is calculated using mutual information entropy to select the optimal feature layer combination.

[0019] The optimal feature layer is combined into the input domain adaptation network, and the contrastive loss function of transfer learning is used to minimize the distribution difference between domains, learn domain-invariant features, and output the initial adaptation model.

[0020] The model parameters of the initial adaptation model are locally updated and globally optimized using a federated averaging algorithm. The model is iteratively verified until the adaptation accuracy of the model in the entire scenario meets the requirements, so as to generate the source domain-target domain adaptation model.

[0021] Furthermore, the step of outputting the initial framework of the speed trajectory corresponding to the electric light truck through the source domain-target domain adaptation model includes:

[0022] A bidirectional attention mechanism is embedded in the feature interaction layer of the source-target domain adaptation model so that the feature interaction layer outputs the corresponding key features according to the working condition feature vector library.

[0023] A bidirectional feature mapping matrix is ​​constructed based on the key features, and the bidirectional feature mapping matrix is ​​iteratively corrected simultaneously to output the corresponding target feature mapping matrix.

[0024] The initial framework of the velocity trajectory is generated based on the target feature mapping matrix.

[0025] Furthermore, the step of generating the initial framework of the velocity trajectory based on the target feature mapping matrix includes:

[0026] The intersection dimension between the source domain historical trajectory features and the target domain actual working condition features in the target feature mapping matrix is ​​detected, and the dynamic topology features of urban roads are introduced as an enhancement dimension to construct a feature enhancement vector.

[0027] The initial velocity time series sequence is generated based on the feature enhancement vector, and the initial velocity time series sequence is simultaneously subjected to dynamic feasibility pre-verification to output several candidate velocity time series sequences that meet the dynamic constraints, forming a candidate trajectory pool.

[0028] The vehicle-mounted terminal collects traffic light timing data within a preset distance in front of the vehicle, and simultaneously calculates the target speed timing sequence in the candidate trajectory pool that has the highest matching degree with the traffic light timing data, so as to construct the initial framework of the speed trajectory based on the target speed timing sequence.

[0029] Furthermore, the step of calculating the optimal velocity trajectory based on the constraints and the objective function includes:

[0030] The federated learning architecture aggregates global working condition data and static road data, and combines graph neural networks to mine the correlation between working condition features and constraints, so as to generate a working condition-constraint correlation graph.

[0031] The optimal energy consumption range is obtained by solving the working condition-constraint correlation graph using the aforementioned objective function.

[0032] The basic speed range is verified to generate the optimal speed trajectory.

[0033] Furthermore, the step of verifying the basic speed range to generate the optimal speed trajectory includes:

[0034] The federated learning architecture aggregates multi-source validation data, and simultaneously uses an attention mechanism to filter out strongly correlated data subsets. Combined with the working condition-constraint correlation graph, a corresponding three-dimensional validation model is constructed.

[0035] The basic velocity range is iteratively narrowed through the three-dimensional verification model to output the corresponding intermediate velocity range;

[0036] The optimal speed trajectory is generated by calculating the speed interval corresponding to the intermediate speed interval using the interval contraction algorithm and simultaneously fitting each of the speed interval optimal solutions into a continuous speed trajectory.

[0037] The second aspect of the present invention proposes:

[0038] An urban cruise control system for electric light trucks, wherein the system comprises:

[0039] The module is used to construct a federated learning architecture between the vehicle terminal of the electric light truck and the urban transportation cloud platform, and synchronously update the global parameters of the urban transportation cloud platform based on the federated learning architecture through a federated averaging algorithm.

[0040] The migration module is used to construct a working condition feature vector library covering the entire urban road network based on the global parameters, and simultaneously use a transfer learning algorithm to transfer the speed trajectory planning experience under the historical best working conditions to the internal working condition feature vector library. Combined with the remaining battery power and power system status of the electric light truck, a source domain-target domain adaptation model is constructed.

[0041] The output module is used to output an initial speed trajectory framework corresponding to the electric light truck through the source domain-target domain adaptation model, and simultaneously construct a corresponding optimization objective function based on the initial speed trajectory framework;

[0042] The calculation module is used to retrieve preset constraints and simultaneously calculate the optimal speed trajectory based on the constraints and the optimization objective function, so as to complete the corresponding urban cruise based on the optimal speed trajectory.

[0043] Furthermore, the migration module is specifically used for:

[0044] The time-series variation curves of the remaining battery power and the system parameters of the power system of the electric light truck under different driving conditions are collected, and the time-series dependency relationship between the time-series variation curves and the system parameters is simultaneously mined through a long short-term memory network.

[0045] By introducing battery aging coefficient and power system health index, the time series data corresponding to the time series dependency relationship is subjected to feature enhancement processing to generate the corresponding target domain dynamic feature set.

[0046] Using the historical best operating condition data in the operating condition feature vector library as the source domain data, the source domain-target domain adaptation model is constructed synchronously based on the source domain data and the target domain dynamic feature set.

[0047] Furthermore, the migration module is specifically used for:

[0048] The source domain data is decomposed into a speed trajectory feature layer, an energy consumption feature layer, and an operating environment feature layer. Simultaneously, the target domain dynamic feature set is decomposed into a time-series power consumption feature layer, a power system state layer, and a health index layer. The correlation degree of each layer's features is calculated using mutual information entropy to select the optimal feature layer combination.

[0049] The optimal feature layer is combined into the input domain adaptation network, and the contrastive loss function of transfer learning is used to minimize the distribution difference between domains, learn domain-invariant features, and output the initial adaptation model.

[0050] The model parameters of the initial adaptation model are locally updated and globally optimized using a federated averaging algorithm. The model is iteratively verified until the adaptation accuracy of the model in the entire scenario meets the requirements, so as to generate the source domain-target domain adaptation model.

[0051] Furthermore, the output module is specifically used for:

[0052] A bidirectional attention mechanism is embedded in the feature interaction layer of the source-target domain adaptation model so that the feature interaction layer outputs the corresponding key features according to the working condition feature vector library.

[0053] A bidirectional feature mapping matrix is ​​constructed based on the key features, and the bidirectional feature mapping matrix is ​​iteratively corrected simultaneously to output the corresponding target feature mapping matrix.

[0054] The initial framework of the velocity trajectory is generated based on the target feature mapping matrix.

[0055] Furthermore, the output module is specifically used for:

[0056] The intersection dimension between the source domain historical trajectory features and the target domain actual working condition features in the target feature mapping matrix is ​​detected, and the dynamic topology features of urban roads are introduced as an enhancement dimension to construct a feature enhancement vector.

[0057] The initial velocity time series sequence is generated based on the feature enhancement vector, and the initial velocity time series sequence is simultaneously subjected to dynamic feasibility pre-verification to output several candidate velocity time series sequences that meet the dynamic constraints, forming a candidate trajectory pool.

[0058] The vehicle-mounted terminal collects traffic light timing data within a preset distance in front of the vehicle, and simultaneously calculates the target speed timing sequence in the candidate trajectory pool that has the highest matching degree with the traffic light timing data, so as to construct the initial framework of the speed trajectory based on the target speed timing sequence.

[0059] Furthermore, the calculation module is specifically used for:

[0060] The federated learning architecture aggregates global working condition data and static road data, and combines graph neural networks to mine the correlation between working condition features and constraints, so as to generate a working condition-constraint correlation graph.

[0061] The optimal energy consumption range is obtained by solving the working condition-constraint correlation graph using the aforementioned objective function.

[0062] The basic speed range is verified to generate the optimal speed trajectory.

[0063] Furthermore, the calculation module is specifically used for:

[0064] The federated learning architecture aggregates multi-source validation data, and simultaneously uses an attention mechanism to filter out strongly correlated data subsets. Combined with the working condition-constraint correlation graph, a corresponding three-dimensional validation model is constructed.

[0065] The basic velocity range is iteratively narrowed through the three-dimensional verification model to output the corresponding intermediate velocity range;

[0066] The optimal speed trajectory is generated by calculating the speed interval corresponding to the intermediate speed interval using the interval contraction algorithm and simultaneously fitting each of the speed interval optimal solutions into a continuous speed trajectory.

[0067] The third aspect of the present invention proposes:

[0068] A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the electric light truck urban cruise control method as described above.

[0069] The fourth aspect of the present invention proposes:

[0070] A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the electric light truck urban cruise control method as described above.

[0071] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0072] Figure 1 A flowchart of the urban cruise control method for electric light trucks provided in the first embodiment of the present invention;

[0073] Figure 2 This is a structural block diagram of an electric light truck urban cruise control system provided in the third embodiment of the present invention.

[0074] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0075] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0076] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0077] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0078] Please see Figure 1 The figure shows the electric light truck urban cruise control method provided in the first embodiment of the present invention. The electric light truck urban cruise control method provided in this embodiment can accurately plan the cruise speed of the electric light truck, thereby improving the control efficiency.

[0079] Specifically, this embodiment provides:

[0080] A method for controlling urban cruise control of an electric light truck, wherein the method includes:

[0081] Step S10: Construct a federated learning architecture between the vehicle terminal of the electric light truck and the urban transportation cloud platform, and synchronously update the global parameters of the urban transportation cloud platform based on the federated learning architecture through a federated averaging algorithm.

[0082] It should be noted that, firstly, urban cruising of electric light trucks needs to cover the entire urban road conditions. The operating condition data of a single vehicle is limited, and the onboard data contains private information (such as driving trajectory and energy consumption data). Therefore, a federated learning architecture is constructed between the onboard terminal and the urban traffic cloud platform. Specifically, federated learning allows each onboard terminal to participate in model training without disclosing the original data, taking into account both data privacy protection and the collaborative value of data across the entire region. The global parameters of the cloud platform are updated collaboratively through a federated averaging algorithm: this algorithm weights and averages the local model parameters of each onboard terminal to generate global model parameters adapted to the urban operating conditions across the entire region, laying the foundation for subsequent modeling of the features of the entire region's operating conditions.

[0083] Step S20: Construct a working condition feature vector library covering the entire urban road network based on the global parameters. Simultaneously, use a transfer learning algorithm to transfer the speed trajectory planning experience under the historical best working conditions to the working condition feature vector library. Combine the remaining battery power and power system status of the electric light truck to construct a source domain-target domain adaptation model.

[0084] It should be noted that, secondly, a feature vector library covering the entire urban road network (including features such as speed, traffic density, and traffic light cycle for conditions such as smooth traffic, congestion, and intersections) is constructed based on global parameters. Traditional cruise control is difficult to reuse historical optimal cruise experience (such as the optimal speed trajectory for a certain road segment). Therefore, a transfer learning algorithm is used to transfer the speed trajectory planning experience under historical optimal conditions to the feature vector library. At the same time, a source domain-target domain adaptation model is constructed by combining the remaining battery charge (SOC) and power system status (such as motor speed and controller temperature) of the electric light truck: the source domain is the historical optimal condition data, and the target domain is the real-time status data of the current vehicle. The adaptation model can solve the problem of mismatch between historical experience and current vehicle status, and realize the accurate reuse of experience.

[0085] Step S30: Output the initial speed trajectory framework corresponding to the electric light truck through the source domain-target domain adaptation model, and simultaneously construct the corresponding optimization objective function based on the initial speed trajectory framework;

[0086] It should be noted that, next, the source domain-target domain adaptation model outputs an initial framework of speed trajectory that matches the current electric light truck. Specifically, the initial framework determines the approximate range and trend of the cruising speed (such as deceleration before intersections and constant speed on unobstructed road sections). Based on the initial framework, an optimization objective function is constructed, and the optimization objective focuses on the core requirements of urban cruising for electric light trucks (such as minimizing energy consumption, maximizing traffic efficiency, and optimizing driving smoothness).

[0087] Step S40: Invoke the preset constraints and simultaneously calculate the optimal speed trajectory based on the constraints and the optimization objective function, so as to complete the corresponding urban cruise based on the optimal speed trajectory.

[0088] It should be noted that, finally, preset constraints (such as maximum motor speed, battery discharge power limit, urban road speed limit, and longitudinal acceleration constraint) are invoked, and the objective function is solved under the constraints to obtain the optimal speed trajectory. The power system of the electric light truck is controlled according to the optimal speed trajectory to complete urban cruising and achieve a multi-objective balance of "energy consumption-efficiency-smoothness".

[0089] Second Embodiment

[0090] Furthermore, the step of constructing the source domain-target domain adaptation model by combining the remaining battery power and power system status of the electric light truck includes:

[0091] The time-series variation curves of the remaining battery power and the system parameters of the power system of the electric light truck under different driving conditions are collected, and the time-series dependency relationship between the time-series variation curves and the system parameters is simultaneously mined through a long short-term memory network.

[0092] By introducing battery aging coefficient and power system health index, the time series data corresponding to the time series dependency relationship is subjected to feature enhancement processing to generate the corresponding target domain dynamic feature set.

[0093] Using the historical best operating condition data in the operating condition feature vector library as the source domain data, the source domain-target domain adaptation model is constructed synchronously based on the source domain data and the target domain dynamic feature set.

[0094] It should be noted that, firstly, the remaining battery charge and the state of the power system are key dynamic factors affecting the cruising performance of electric light trucks (e.g., energy consumption needs to be reduced when the SOC is low, and power needs to be limited when the motor is at high temperature). Therefore, the time-series change curves of the remaining battery charge under different driving conditions (e.g., the rate of SOC decrease under congested conditions) and power system parameters (e.g., motor torque, controller voltage, and cooling system temperature) are collected. Since these data have strong time-series correlations (e.g., an increase in motor speed will lead to an increase in controller temperature, which in turn affects battery discharge efficiency), a Long Short-Term Memory (LSTM) network is used to mine the time-series dependencies. Specifically, LSTM is good at capturing the dynamic correlations of long-term time-series data and can accurately depict the coupling law of "charge change - power parameters".

[0095] Secondly, the aging degree of the battery in electric light trucks (such as the decrease in battery capacity after one year of use) and the health of the power system (such as the reduction in efficiency due to motor wear) will affect the actual cruising performance. Traditional models do not consider these factors and are prone to adaptation deviations. Therefore, a battery aging coefficient (such as the capacity decay coefficient calculated based on the number of cycles) and a power system health index (such as the motor efficiency decay rate) are introduced. The time-series data corresponding to the time-series dependencies are subjected to feature enhancement processing (such as fusing the aging coefficient with the SOC time-series data and fusing the health index with the motor parameters) to generate a dynamic feature set of the target domain. Specifically, feature enhancement can make the features of the target domain more accurately reflect the actual performance status of the current vehicle.

[0096] Finally, using the historical best operating condition data in the operating condition feature vector library as the source domain data (including the historical best speed trajectory and the corresponding energy consumption data of the operating condition), and combining it with the target domain dynamic feature set, a source domain-target domain adaptation model is constructed: the model learns the feature mapping relationship between the source domain and the target domain, adapts the historical best speed trajectory experience to the current vehicle state, provides accurate model support for the subsequent generation of the initial speed trajectory framework, and connects with the transfer learning logic mentioned above.

[0097] Furthermore, the step of constructing the source domain-target domain adaptation model based on the source domain data and the target domain dynamic feature set includes:

[0098] The source domain data is decomposed into a speed trajectory feature layer, an energy consumption feature layer, and an operating environment feature layer. Simultaneously, the target domain dynamic feature set is decomposed into a time-series power consumption feature layer, a power system state layer, and a health index layer. The correlation degree of each layer's features is calculated using mutual information entropy to select the optimal feature layer combination.

[0099] The optimal feature layer is combined into the input domain adaptation network, and the contrastive loss function of transfer learning is used to minimize the distribution difference between domains, learn domain-invariant features, and output the initial adaptation model.

[0100] The model parameters of the initial adaptation model are locally updated and globally optimized using a federated averaging algorithm. The model is iteratively verified until the adaptation accuracy of the model in the entire scenario meets the requirements, so as to generate the source domain-target domain adaptation model.

[0101] It should be noted that, firstly, both the source and target domain data contain multi-dimensional features, some of which have little impact on speed trajectory planning (such as irrelevant environmental noise features). Therefore, the source domain data is decomposed into a speed trajectory feature layer (such as historical speed sequences and acceleration changes), an energy consumption feature layer (such as energy consumption per unit mileage), and an operating environment feature layer (such as traffic density and road slope). The target domain dynamic feature set is decomposed into a time-series energy consumption feature layer (such as SOC time-series curves and discharge rate), a power system state layer (such as motor parameters and controller status), and a health index layer (such as aging coefficient and health score). Mutual information entropy is used to calculate the correlation between features in each layer (such as the correlation between the speed trajectory feature layer and the time-series energy consumption feature layer, reflecting the impact of speed changes on energy consumption), and the optimal feature layer combination is selected (such as retaining feature layers with a correlation ≥ 0.6). Specifically, feature selection can eliminate redundant features, reduce model computational complexity, and improve adaptation efficiency.

[0102] Secondly, the optimal feature layer is combined into the input domain adaptation network, and the contrastive loss function of transfer learning is used to minimize the distribution difference between domains: the contrastive loss function brings the feature distributions of the source domain and the target domain closer together, allowing the model to learn common features that are not affected by domain differences (such as the common pattern of "low energy consumption speed range"), and outputs the initial adaptation model. Specifically, the initial model solves the distribution offset problem between the source domain and the target domain, and realizes the initial transfer of historical experience.

[0103] Finally, to ensure the model adapts to all urban conditions (such as congestion characteristics in different areas and differences in traffic light cycles), the federated learning architecture described above is combined with the federated averaging algorithm to update the parameters of the initial adapted model: each vehicle terminal updates the local parameters of the model based on local operating data, and the cloud platform aggregates the local parameters to obtain the globally optimized parameters; the model's adaptation accuracy in all scenarios (such as trajectory adaptation error in different road sections and different SOC states) is iteratively verified until the accuracy meets the preset requirements (such as error ≤ 5%), and finally a source domain-target domain adapted model is generated. Specifically, federated optimization allows the model to take into account both the personalized state of a single vehicle and the universality of all operating conditions, improving the model's generalization ability.

[0104] Furthermore, the step of outputting the initial framework of the speed trajectory corresponding to the electric light truck through the source domain-target domain adaptation model includes:

[0105] A bidirectional attention mechanism is embedded in the feature interaction layer of the source-target domain adaptation model so that the feature interaction layer outputs the corresponding key features according to the working condition feature vector library.

[0106] A bidirectional feature mapping matrix is ​​constructed based on the key features, and the bidirectional feature mapping matrix is ​​iteratively corrected simultaneously to output the corresponding target feature mapping matrix.

[0107] The initial framework of the velocity trajectory is generated based on the target feature mapping matrix.

[0108] It should be noted that, firstly, the feature interaction layer of the source-target domain adaptation model needs to focus on the most critical operating condition features for the current cruise (such as the traffic light cycle and traffic density of the current road segment, rather than the operating condition features of other areas). Therefore, a bidirectional attention mechanism is embedded in the feature interaction layer: bidirectional attention can simultaneously focus on the historical key features of the source domain (such as the speed adjustment features corresponding to congestion conditions in the historical optimal trajectory) and the current key features of the target domain (such as the current SOC and motor status), so that the feature interaction layer outputs targeted key features. Specifically, the attention mechanism avoids interference from irrelevant features and improves the accuracy of feature interaction.

[0109] Secondly, a bidirectional feature mapping matrix is ​​constructed based on key features: the matrix depicts the bidirectional mapping relationship between historical key features of the source domain and current key features of the target domain (such as the mapping between speed adjustment features under historical congestion conditions and SOC features under current congestion conditions); by iteratively correcting the mapping matrix (such as adjusting the mapping weights according to actual driving data to reduce mapping errors), the target feature mapping matrix is ​​output. Specifically, iterative correction can make the feature mapping more in line with the actual cruising needs of the current vehicle and avoid mismatch between historical experience and current conditions.

[0110] Finally, an initial framework for the speed trajectory is generated based on the target feature mapping matrix: the framework transforms the historical best speed trajectory experience into an initial speed sequence that adapts to the current vehicle state and operating conditions (e.g., when the current SOC is low, the uniform speed in the initial framework is 5 km / h lower than the historical best value), providing a basic framework for the subsequent construction of the objective function.

[0111] Furthermore, the step of generating the initial framework of the velocity trajectory based on the target feature mapping matrix includes:

[0112] The intersection dimension between the source domain historical trajectory features and the target domain actual working condition features in the target feature mapping matrix is ​​detected, and the dynamic topology features of urban roads are introduced as an enhancement dimension to construct a feature enhancement vector.

[0113] The initial velocity time series sequence is generated based on the feature enhancement vector, and the initial velocity time series sequence is simultaneously subjected to dynamic feasibility pre-verification to output several candidate velocity time series sequences that meet the dynamic constraints, forming a candidate trajectory pool.

[0114] The vehicle-mounted terminal collects traffic light timing data within a preset distance in front of the vehicle, and simultaneously calculates the target speed timing sequence in the candidate trajectory pool that has the highest matching degree with the traffic light timing data, so as to construct the initial framework of the speed trajectory based on the target speed timing sequence.

[0115] It should be noted that, firstly, the source domain historical trajectory features and the target domain actual operating condition features in the target feature mapping matrix have an overlap (e.g., both contain the "intersection operating condition" feature), and the overlap dimension is the core basis for reusing historical experience; at the same time, the dynamic topology of urban roads (e.g., intersection spacing, number of lanes, turning radius) will affect speed trajectory planning (e.g., road sections with large turning radii can maintain higher speeds, while short intersection spacing requires frequent speed adjustments). Therefore, this feature is introduced as an enhancement dimension to construct a feature enhancement vector. Specifically, the enhancement vector integrates common features and scene-specific features, making the initial framework more in line with the actual urban roads.

[0116] Secondly, an initial speed time series is generated based on the feature enhancement vector (e.g., "0-500m: 30km / h → 500-600m (before the intersection): decelerate to 10km / h → 600-1000m: accelerate to 35km / h"). Since the initial sequence may exceed the vehicle's power capacity (e.g., excessive acceleration causing the motor power to exceed the limit), a power feasibility pre-verification is required: based on the power system parameters of the electric light truck (e.g., maximum torque, maximum power), the feasibility of the acceleration / deceleration requirements of each speed segment in the initial sequence is verified, infeasible sequences are eliminated, and candidate speed time series sequences that meet the power constraints are output to form a candidate trajectory pool. Specifically, the pre-verification ensures the engineering feasibility of the initial framework and avoids invalid calculations in subsequent optimization.

[0117] Finally, urban cruise needs to avoid frequent starts and stops caused by traffic lights (to improve efficiency and energy consumption). Therefore, traffic light timing data (such as current light color and remaining duration) within a preset distance in front of the vehicle is collected through onboard terminals (such as cameras and vehicle networking modules). The matching degree between each sequence in the candidate trajectory pool and the traffic light timing data is calculated (e.g., a high matching degree means that the vehicle arrives at the intersection just as the light turns green and there is no need to stop). The target speed timing sequence with the highest matching degree is selected. An initial framework for the speed trajectory is constructed based on this sequence. Specifically, traffic light matching makes the initial framework more in line with the characteristics of urban traffic flow, thereby improving the efficiency of cruise.

[0118] Furthermore, the step of calculating the optimal velocity trajectory based on the constraints and the objective function includes:

[0119] The federated learning architecture aggregates global working condition data and static road data, and combines graph neural networks to mine the correlation between working condition features and constraints, so as to generate a working condition-constraint correlation graph.

[0120] The optimal energy consumption range is obtained by solving the working condition-constraint correlation graph using the aforementioned objective function.

[0121] The basic speed range is verified to generate the optimal speed trajectory.

[0122] It should be noted that, firstly, urban operating conditions are complex and diverse (e.g., different speed limits on different road sections, and different traffic flow constraints under different levels of congestion). Constraints need to be precisely matched with specific operating conditions. Therefore, a federated learning architecture is used to aggregate operating condition data across the entire domain (real-time operating condition data from each vehicle terminal) and static road data (e.g., road speed limits, intersection locations, and number of lanes). Graph Neural Networks (GNNs) are used to mine the correlation between operating condition features and constraints. Specifically, GNNs are good at processing structured data and can construct a correlation graph of "operating condition nodes - constraint nodes" to generate an operating condition-constraint correlation graph (e.g., "smooth road section" operating condition is associated with "speed limit 60km / h, maximum acceleration 2m / s²" constraint, and "congested road section" is associated with "speed limit 30km / h, maximum acceleration 1m / s²" constraint).

[0123] Secondly, the objective function is optimized with "minimizing energy consumption" as the core (the core requirement of urban cruising for electric light trucks). A multi-objective function is constructed by combining traffic efficiency (such as the shortest travel time) and driving smoothness (such as the minimum acceleration fluctuation). Based on the working condition-constraint correlation graph, the constraints corresponding to the current working condition are substituted into the objective function to solve for the optimal basic speed range for energy consumption (such as 35-45 km / h for the current unobstructed road section). Specifically, the basic speed range takes into account both the optimization objective and the working condition constraints, providing a range basis for subsequent accurate trajectory generation.

[0124] Finally, the basic speed range is verified: whether the speed within the range meets the constraints of the power system (such as whether the motor power exceeds the limit), traffic rules (such as whether it complies with the speed limit), and energy consumption (such as whether the energy consumption per unit mileage is within the optimal range); speed segments that do not meet the constraints are eliminated, the basic speed range is corrected, and a preliminary range of the optimal speed trajectory is generated to connect with the subsequent detailed verification steps.

[0125] Furthermore, the step of verifying the basic speed range to generate the optimal speed trajectory includes:

[0126] The federated learning architecture aggregates multi-source validation data, and simultaneously uses an attention mechanism to filter out strongly correlated data subsets. Combined with the working condition-constraint correlation graph, a corresponding three-dimensional validation model is constructed.

[0127] The basic velocity range is iteratively narrowed through the three-dimensional verification model to output the corresponding intermediate velocity range;

[0128] The optimal speed trajectory is generated by calculating the speed interval corresponding to the intermediate speed interval using the interval contraction algorithm and simultaneously fitting each of the speed interval optimal solutions into a continuous speed trajectory.

[0129] It should be noted that, firstly, to ensure the comprehensiveness of the verification, multi-source verification data (such as cruise data of other similar electric light trucks under the same operating conditions, historical traffic flow data from the urban traffic cloud platform, and historical energy consumption data of the vehicle itself) are aggregated through a federated learning architecture; an attention mechanism is used to filter out strongly relevant subsets of data (such as verification data consistent with the current vehicle model and current operating conditions) to avoid interference from irrelevant data; and a three-dimensional verification model (with the dimensions of "speed-energy consumption-operating conditions") is constructed by combining the operating condition-constraint correlation graph. Specifically, the three-dimensional model can comprehensively verify the feasibility of the basic speed range under different energy consumption and different operating condition details.

[0130] Secondly, the basic speed range is iteratively narrowed through the three-dimensional verification model: for example, if the verification finds that the energy consumption of the 42-45km / h segment in the basic speed range of 35-45km / h is significantly increased, the range is narrowed to 35-42km / h, and the intermediate speed range is output. Specifically, the iterative narrowing makes the speed range more accurate and further improves the energy consumption optimization effect.

[0131] Finally, an interval contraction algorithm (such as the golden section method) is used to calculate the optimal speed range solution within the intermediate speed range (e.g., the optimal speed point is 38 km / h, the acceleration range is 35-38 km / h, and the deceleration range is 38-35 km / h). Each optimal solution is fitted into a continuous speed trajectory according to the time series (ensuring smooth acceleration and avoiding abrupt changes), generating the optimal speed trajectory. Specifically, the continuous trajectory can directly drive the power control system of the electric light truck, achieving smooth, efficient, and low-energy-consumption urban cruising, and completing the closed loop of the entire control process.

[0132] Please see Figure 2 The third embodiment of the present invention provides:

[0133] An urban cruise control system for electric light trucks, wherein the system comprises:

[0134] The module is used to construct a federated learning architecture between the vehicle terminal of the electric light truck and the urban transportation cloud platform, and synchronously update the global parameters of the urban transportation cloud platform based on the federated learning architecture through a federated averaging algorithm.

[0135] The migration module is used to construct a working condition feature vector library covering the entire urban road network based on the global parameters, and simultaneously use a transfer learning algorithm to transfer the speed trajectory planning experience under the historical best working conditions to the internal working condition feature vector library. Combined with the remaining battery power and power system status of the electric light truck, a source domain-target domain adaptation model is constructed.

[0136] The output module is used to output an initial speed trajectory framework corresponding to the electric light truck through the source domain-target domain adaptation model, and simultaneously construct a corresponding optimization objective function based on the initial speed trajectory framework;

[0137] The calculation module is used to retrieve preset constraints and simultaneously calculate the optimal speed trajectory based on the constraints and the optimization objective function, so as to complete the corresponding urban cruise based on the optimal speed trajectory.

[0138] Furthermore, the migration module is specifically used for:

[0139] The time-series variation curves of the remaining battery power and the system parameters of the power system of the electric light truck under different driving conditions are collected, and the time-series dependency relationship between the time-series variation curves and the system parameters is simultaneously mined through a long short-term memory network.

[0140] By introducing battery aging coefficient and power system health index, the time series data corresponding to the time series dependency relationship is subjected to feature enhancement processing to generate the corresponding target domain dynamic feature set.

[0141] Using the historical best operating condition data in the operating condition feature vector library as the source domain data, the source domain-target domain adaptation model is constructed synchronously based on the source domain data and the target domain dynamic feature set.

[0142] Furthermore, the migration module is specifically used for:

[0143] The source domain data is decomposed into a speed trajectory feature layer, an energy consumption feature layer, and an operating environment feature layer. Simultaneously, the target domain dynamic feature set is decomposed into a time-series power consumption feature layer, a power system state layer, and a health index layer. The correlation degree of each layer's features is calculated using mutual information entropy to select the optimal feature layer combination.

[0144] The optimal feature layer is combined into the input domain adaptation network, and the contrastive loss function of transfer learning is used to minimize the distribution difference between domains, learn domain-invariant features, and output the initial adaptation model.

[0145] The model parameters of the initial adaptation model are locally updated and globally optimized using a federated averaging algorithm. The model is iteratively verified until the adaptation accuracy of the model in the entire scenario meets the requirements, so as to generate the source domain-target domain adaptation model.

[0146] Furthermore, the output module is specifically used for:

[0147] A bidirectional attention mechanism is embedded in the feature interaction layer of the source-target domain adaptation model so that the feature interaction layer outputs the corresponding key features according to the working condition feature vector library.

[0148] A bidirectional feature mapping matrix is ​​constructed based on the key features, and the bidirectional feature mapping matrix is ​​iteratively corrected simultaneously to output the corresponding target feature mapping matrix.

[0149] The initial framework of the velocity trajectory is generated based on the target feature mapping matrix.

[0150] Furthermore, the output module is specifically used for:

[0151] The intersection dimension between the source domain historical trajectory features and the target domain actual working condition features in the target feature mapping matrix is ​​detected, and the dynamic topology features of urban roads are introduced as an enhancement dimension to construct a feature enhancement vector.

[0152] The initial velocity time series sequence is generated based on the feature enhancement vector, and the initial velocity time series sequence is simultaneously subjected to dynamic feasibility pre-verification to output several candidate velocity time series sequences that meet the dynamic constraints, forming a candidate trajectory pool.

[0153] The vehicle-mounted terminal collects traffic light timing data within a preset distance in front of the vehicle, and simultaneously calculates the target speed timing sequence in the candidate trajectory pool that has the highest matching degree with the traffic light timing data, so as to construct the initial framework of the speed trajectory based on the target speed timing sequence.

[0154] Furthermore, the calculation module is specifically used for:

[0155] The federated learning architecture aggregates global working condition data and static road data, and combines graph neural networks to mine the correlation between working condition features and constraints, so as to generate a working condition-constraint correlation graph.

[0156] The optimal energy consumption range is obtained by solving the objective function based on the working condition-constraint correlation graph.

[0157] The basic speed range is verified to generate the optimal speed trajectory.

[0158] Furthermore, the calculation module is specifically used for:

[0159] The federated learning architecture aggregates multi-source validation data, and simultaneously uses an attention mechanism to filter out strongly correlated data subsets. Combined with the working condition-constraint correlation graph, a corresponding three-dimensional validation model is constructed.

[0160] The basic velocity range is iteratively narrowed through the three-dimensional verification model to output the corresponding intermediate velocity range;

[0161] The optimal speed trajectory is generated by calculating the speed interval corresponding to the intermediate speed interval using the interval contraction algorithm and simultaneously fitting each of the speed interval optimal solutions into a continuous speed trajectory.

[0162] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the electric light truck urban cruise control method as described above.

[0163] The fifth embodiment of the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the electric light truck urban cruise control method as described above.

[0164] In summary, the electric light truck urban cruise control method and system provided by the above embodiments of the present invention can accurately plan the cruise speed of the electric pickup truck, thereby improving control efficiency.

[0165] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0166] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0167] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0168] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0169] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0170] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for controlling urban cruise control of an electric light truck, characterized in that, The method includes: A federated learning architecture for the vehicle terminal of an electric light truck and the urban traffic cloud platform is constructed. Based on the federated learning architecture, the global parameters of the urban traffic cloud platform are updated collaboratively through a federated averaging algorithm. Based on the global parameters, a working condition feature vector library covering the entire urban road network is constructed. Simultaneously, a transfer learning algorithm is used to transfer the speed trajectory planning experience under the historical best working conditions to the working condition feature vector library. Combined with the remaining battery power and power system status of the electric light truck, a source domain-target domain adaptation model is constructed. The source-target domain adaptation model outputs an initial speed trajectory framework corresponding to the electric light truck, and the corresponding optimization objective function is constructed simultaneously based on the initial speed trajectory framework. The preset constraints are invoked, and the optimal speed trajectory is calculated simultaneously based on the constraints and the optimization objective function, so as to complete the corresponding urban cruise according to the optimal speed trajectory; The steps for constructing the source domain-target domain adaptation model by combining the remaining battery power and power system status of the electric light truck include: The time-series variation curves of the remaining battery power and the system parameters of the power system of the electric light truck under different driving conditions are collected, and the time-series dependency relationship between the time-series variation curves and the system parameters is simultaneously mined through a long short-term memory network. By introducing battery aging coefficient and power system health index, the time series data corresponding to the time series dependency relationship is subjected to feature enhancement processing to generate the corresponding target domain dynamic feature set. Using the historical best working condition data in the working condition feature vector library as the source domain data, the source domain-target domain adaptation model is constructed synchronously based on the source domain data and the target domain dynamic feature set. The step of constructing the source domain-target domain adaptation model based on the source domain data and the target domain dynamic feature set includes: The source domain data is decomposed into a speed trajectory feature layer, an energy consumption feature layer, and an operating environment feature layer. Simultaneously, the target domain dynamic feature set is decomposed into a time-series power consumption feature layer, a power system state layer, and a health index layer. The correlation degree of each layer's features is calculated using mutual information entropy to select the optimal feature layer combination. The optimal feature layer is combined into the input domain adaptation network, and the contrastive loss function of transfer learning is used to minimize the distribution difference between domains, learn domain-invariant features, and output the initial adaptation model. The model parameters of the initial adaptation model are locally updated and globally optimized using a federated averaging algorithm. The model is iteratively verified until the adaptation accuracy of the model in the entire scenario meets the requirements, so as to generate the source domain-target domain adaptation model.

2. The urban cruise control method for electric light trucks according to claim 1, characterized in that, The step of outputting the initial framework of the velocity trajectory corresponding to the electric light truck through the source domain-target domain adaptation model includes: A bidirectional attention mechanism is embedded in the feature interaction layer of the source-target domain adaptation model so that the feature interaction layer outputs the corresponding key features according to the working condition feature vector library. A bidirectional feature mapping matrix is ​​constructed based on the key features, and the bidirectional feature mapping matrix is ​​iteratively corrected simultaneously to output the corresponding target feature mapping matrix. The initial framework of the velocity trajectory is generated based on the target feature mapping matrix.

3. The urban cruise control method for electric light trucks according to claim 2, characterized in that, The step of generating the initial framework of the velocity trajectory based on the target feature mapping matrix includes: The intersection dimension between the source domain historical trajectory features and the target domain actual working condition features in the target feature mapping matrix is ​​detected, and the dynamic topology features of urban roads are introduced as an enhancement dimension to construct a feature enhancement vector. The initial velocity time series sequence is generated based on the feature enhancement vector, and the initial velocity time series sequence is simultaneously subjected to dynamic feasibility pre-verification to output several candidate velocity time series sequences that meet the dynamic constraints, forming a candidate trajectory pool. The vehicle-mounted terminal collects traffic light timing data within a preset distance in front of the vehicle, and simultaneously calculates the target speed timing sequence in the candidate trajectory pool that has the highest matching degree with the traffic light timing data, so as to construct the initial framework of the speed trajectory based on the target speed timing sequence.

4. The urban cruise control method for electric light trucks according to claim 1, characterized in that, The step of calculating the optimal velocity trajectory based on the constraints and the objective function includes: The federated learning architecture aggregates global working condition data and static road data, and combines graph neural networks to mine the correlation between working condition features and constraints, so as to generate a working condition-constraint correlation graph. The optimal energy consumption range is obtained by solving the working condition-constraint correlation graph using the aforementioned objective function. The basic speed range is verified to generate the optimal speed trajectory.

5. The urban cruise control method for electric light trucks according to claim 4, characterized in that, The step of verifying the basic speed range to generate the optimal speed trajectory includes: The federated learning architecture aggregates multi-source validation data, and simultaneously uses an attention mechanism to filter out strongly correlated data subsets. Combined with the working condition-constraint correlation graph, a corresponding three-dimensional validation model is constructed. The basic velocity range is iteratively narrowed through the three-dimensional verification model to output the corresponding intermediate velocity range; The optimal speed trajectory is generated by calculating the speed interval corresponding to the intermediate speed interval using the interval contraction algorithm and simultaneously fitting each of the speed interval optimal solutions into a continuous speed trajectory.

6. An urban cruise control system for electric light trucks, characterized in that, For implementing the urban cruise control method for electric light trucks as described in any one of claims 1 to 5, the system comprises: The module is used to construct a federated learning architecture between the vehicle terminal of the electric light truck and the urban transportation cloud platform, and synchronously update the global parameters of the urban transportation cloud platform based on the federated learning architecture through a federated averaging algorithm. The migration module is used to construct a working condition feature vector library covering the entire urban road network based on the global parameters, and simultaneously use a transfer learning algorithm to transfer the speed trajectory planning experience under the historical best working conditions to the internal working condition feature vector library. Combined with the remaining battery power and power system status of the electric light truck, a source domain-target domain adaptation model is constructed. The output module is used to output an initial speed trajectory framework corresponding to the electric light truck through the source domain-target domain adaptation model, and simultaneously construct a corresponding optimization objective function based on the initial speed trajectory framework; The calculation module is used to retrieve preset constraints and simultaneously calculate the optimal speed trajectory based on the constraints and the optimization objective function, so as to complete the corresponding urban cruise based on the optimal speed trajectory.

7. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the electric light truck urban cruise control method as described in any one of claims 1 to 5.

8. A readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the electric light truck urban cruise control method as described in any one of claims 1 to 5.

Images