Multi-parameter collaborative optimization control method, system and equipment for commercial vehicle engine and medium

Abstract

The embodiment of the invention provides a multi-parameter collaborative optimization control method, system and device for a commercial vehicle engine and a medium, and belongs to the field of engine optimization. The method comprises the steps that vehicle operation data, driver operation data and environment parameters are collected and preprocessed to generate a structured data set; driving style features, road condition features and engine operation features are extracted, and feature vectors are generated; adopting a behavior classification model to classify the driving style of the driver to obtain a driving style label; based on the driving style label, the road condition characteristics and the engine operation characteristics, a dynamic strategy generation model is adopted to adjust fuel-saving control parameters; and generating evaluation feedback based on the evaluated fuel-saving effect, carrying out strategy iteration, generating an engine control strategy, and carrying out collaborative optimization on a plurality of control parameters of the engine. Style clustering is carried out on driving behaviors through deep learning, and based on the adaptive personalized control strategy, the oil-saving control problem under different driving habits is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of engine optimization, in particular to a multi-parameter collaborative optimization control method, system, device and medium for a commercial vehicle engine. BACKGROUND

[0002] Parameter optimization control of a commercial vehicle engine refers to dynamic adjustment of key parameters such as engine fuel injection, intake volume, and ignition timing by an electronic control unit (ECU) to achieve the best balance of power, economy, and emission performance. This technology relies on real-time collection of engine operating state data by sensors, combined with advanced control algorithms (such as PID, fuzzy control, or model predictive control), to accurately control air-fuel ratio, turbocharging pressure, and exhaust gas recirculation rate, thereby improving combustion efficiency, reducing fuel consumption and pollutant emissions. Parameter optimization control not only enhances the adaptability and stability of the engine under different operating conditions, but also meets increasingly stringent emission regulations (such as China VI and Euro VI standards). With the development of Internet of Vehicles and big data technology, intelligent optimization strategies based on cloud calibration and adaptive learning are gradually being applied, promoting the development of commercial vehicles towards high efficiency, greenness, and intelligence.

[0003] However, existing technologies mostly use fixed control logic or single parameter optimization strategies, which are difficult to adapt to differences in driving styles of different drivers and changes in complex road conditions, resulting in limited fuel saving effects under preset operating conditions. At the same time, there is a lack of dynamic feedback iteration mechanism based on actual operating data, and the system has weak adaptive ability, so the optimization performance may easily decay after long-term use. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide a multi-parameter collaborative optimization control method, system, device and medium for a commercial vehicle engine, which accurately clusters driving behaviors through deep learning and realizes individualized adaptation of control strategies based on this, effectively solving the problem of fuel saving control under different driving habits; by multi-dimensionally integrating driving style, real-time road conditions and engine operating parameters, the control strategy adapted to the current scene is dynamically generated using reinforcement learning, significantly improving the adaptability of the system to complex road conditions and changes in driving behavior; further, through multi-parameter collaborative optimization, global optimal matching of engine core control parameters is realized, breaking through the limitations of traditional single-parameter optimization. To achieve the above purpose, the embodiments of the present application provide a multi-parameter collaborative optimization control method for a commercial vehicle engine, comprising: Collecting vehicle operating data, driver operation data and environmental parameters, and preprocessing to generate a structured data set; Extracting driving style features, road condition features and engine operating features from the structured data set to generate a feature vector; Based on the feature vector, a pre-built behavior classification model is used to classify the driver's driving style and obtain driving style labels; Based on the driving style tags, road condition characteristics, and engine operating characteristics, a pre-built dynamic strategy is used to generate a model to adjust fuel-saving control parameters. The fuel-saving effect is evaluated based on the adjusted fuel-saving control parameters, evaluation feedback is generated, and strategy iteration is performed based on the evaluation feedback to generate an engine control strategy. The dynamic strategy generation model is built on the Actor-Critic network architecture. Based on the engine control strategy, multiple control parameters of the engine are optimized collaboratively.

[0005] Optionally, driving style features, road condition features, and engine operation features are extracted from the structured dataset to generate a feature vector, including: Based on the structured dataset, a three-level feature system is constructed; wherein: Driving style characteristics are calculated through a sliding time window, including the percentage of rapid accelerations, the frequency of rapid decelerations, and the rationality index of gear shifting. The road condition features are calculated by fusing GPS data and map information, and include average slope, road curvature and road surface smoothness indicators. Engine operating characteristics include idling time percentage and load rate distribution; wherein, the idling time percentage is the ratio of the cumulative time when the engine speed is below the idling threshold to the total operating time; Calculate the importance scores of driving style features, road condition features, and engine operation features to the driving style classification task, and select features that meet the contribution requirements of driving style classification to form a feature vector.

[0006] Optionally, based on the feature vector, a pre-built behavior classification model is used to classify the driver's driving style to obtain driving style labels, including: The feature vector is input into the encoding layer of a deep clustering network for nonlinear transformation and dimensionality reduction to obtain a low-dimensional embedding vector. In the clustering layer, a probability distribution corresponding to different driving style categories is generated based on the distance between the low-dimensional embedding vector and the preset cluster center. The driving style classification result is determined based on the probability distribution and output as a digital style label; Among them, the behavior classification model is built on a deep clustering network, including an encoding layer formed by a multi-layer fully connected neural network, a probability allocation based on softmax and a clustering layer with clustering loss as the objective function, and a self-supervised fine-tuning layer that introduces an auxiliary loss function. The deep clustering network is trained using a composite objective function that includes clustering loss and auxiliary loss to simultaneously optimize feature encoding and clustering assignment. The training process of the deep clustering network includes: using the K-means++ algorithm to pre-cluster the embedding vectors output by the encoding layer to initialize the cluster centers; then iteratively updating the network parameters of the encoding layer by minimizing the composite objective function and combining it with backpropagation; the updating of the network parameters and the correction of the cluster centers are carried out alternately until the deep clustering network reaches a convergent state.

[0007] Optionally, the step of adjusting fuel-saving control parameters using a pre-built dynamic strategy generation model based on the driving style label, road condition characteristics, and engine operating characteristics includes: The driving style label is encoded into a style feature vector, and then fused with road condition features and engine operation features to construct a state vector. The state vector is input into the Actor network, and the probability distribution of the continuous action space is output. The probability distribution of the continuous action space includes the action probability distribution of engine injection timing, intake air volume and ignition advance angle. The value of the state vector and its corresponding action is evaluated simultaneously using a Critic network. A multi-objective weighted reward function is used for strategy optimization. This reward function includes at least a basic fuel-saving reward item and a driving smoothness reward item. A style adaptation factor that matches the driving style and a road condition weight factor that adapts to real-time road conditions are introduced. A comprehensive reward signal is generated by weighted fusion to guide the output action of the Actor network, thereby generating adjustment instructions for the fuel-saving control parameters.

[0008] Optionally, the step of evaluating the fuel-saving effect based on the adjusted fuel-saving control parameters, generating evaluation feedback, and iterating the strategy based on the evaluation feedback includes: The weight ratio of style adaptation factor and road condition weight factor in the reward function is adjusted based on the actual fuel saving rate from the evaluation feedback. Using the reward signal output by the adjusted reward function as the optimization objective, the temporal difference between the objective value and the current value estimate is calculated. The parameters of the Critic network are periodically updated in the opposite direction of the value estimation gradient. Based on the gradient of the temporal difference and the policy probability, the parameters of the Actor network are updated synchronously to achieve closed-loop iterative optimization of the control policy.

[0009] Optionally, the style adaptation factor is set with different values ​​according to the driving style category, and the road condition weight factor is calculated based on road curvature and slope. Optionally, the iterative update process for the behavior classification model and the dynamic policy generation model includes: A training dataset is constructed based on historical evaluation data and newly added driving cases. When the cumulative number of samples in the training dataset does not reach a preset threshold, incremental training is performed on the behavior classification model and the dynamic policy generation model. When the preset threshold is reached, full retraining is performed on the behavior classification model and the dynamic policy generation model. The model update process includes the following stages executed sequentially: The first stage updates the behavior classification model by using transfer learning to freeze the underlying parameters of the pre-trained behavior classification model and fine-tune the parameters of the top classification layer. The second stage updates the dynamic policy generation model by fixing the underlying weights of the Actor network and adjusting the parameters of its upper-layer network. In this second stage, an early stopping strategy is adopted, and training is terminated when the validation set loss no longer decreases for several consecutive rounds. The updated behavior classification model and dynamic strategy generation model will be deployed in stages. First, they will be tested on a small number of vehicles. If no abnormalities occur within the preset verification period and the performance indicators meet expectations, they will be pushed to all vehicles. In addition, a version rollback mechanism will be retained during the deployment process. If the fuel saving rate of the new model decreases by more than the preset tolerance threshold, the system will automatically switch to the old version model.

[0010] The second invention also provides a multi-parameter collaborative optimization control system for a commercial vehicle engine, comprising: The data acquisition unit is used to collect vehicle operation data, driver operation data and environmental parameters, and to preprocess them to generate structured datasets. The feature extraction unit is used to extract driving style features, road condition features, and engine operation features from the structured dataset and generate feature vectors. The behavior classification unit is used to classify the driver's driving style based on the feature vector using a pre-built behavior classification model to obtain a driving style label; The parameter adjustment unit is used to adjust the fuel-saving control parameters based on the driving style label, road condition characteristics and engine operating characteristics using a pre-built dynamic strategy generation model. The strategy generation unit is used to evaluate the fuel-saving effect based on the adjusted fuel-saving control parameters, generate evaluation feedback, and perform strategy iteration based on the evaluation feedback to generate an engine control strategy. The dynamic strategy generation model is built on the Actor-Critic architecture. The parameter optimization unit is used to collaboratively optimize multiple control parameters of the engine according to the engine control strategy.

[0011] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the multi-parameter collaborative optimization control method for commercial vehicle engines described above.

[0012] Fourthly, the present invention also provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the multi-parameter collaborative optimization control method for commercial vehicle engines described above. Through the above technical solutions, deep learning is used to accurately cluster driving behaviors by style, and personalized adaptation of control strategies is achieved based on this, effectively solving the fuel-saving control problem under different driving habits. By integrating driving style, real-time road conditions and engine operating parameters in multiple dimensions, reinforcement learning is used to dynamically generate control strategies adapted to the current scenario, significantly improving the system's adaptability to complex road conditions and changes in driving behavior. Furthermore, through a multi-parameter collaborative optimization mechanism, the global optimal matching of the engine's core control parameters is achieved, breaking through the limitations of traditional single-parameter optimization. The system also constructs a complete closed-loop feedback iteration mechanism, continuously correcting the strategy model based on actual fuel-saving effects, enabling the system to continuously evolve, and ultimately achieving synergistic improvement and continuous optimization of fuel-saving performance and driving smoothness in long-term use.

[0013] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a multi-parameter collaborative optimization control method for a commercial vehicle engine provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a multi-parameter collaborative optimization control system for a commercial vehicle engine provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0015] Various embodiments of this disclosure will be described more fully in the following detailed description. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0016] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions or operations and do not limit the addition of one or more functions or operations. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a specific feature, number, step, operation, or combination of the foregoing and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, or combinations of the foregoing, or the possibility of adding one or more features, numbers, steps, operations, or combinations of the foregoing.

[0017] In various embodiments of this disclosure, the expression "or" or "at least one of A and / or B" includes any combination or all combinations of the words listed simultaneously. For example, the expression "A or B" or "at least one of A and / or B" may include A, may include B, or may include both A and B.

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 The diagram shows a flowchart of a multi-parameter collaborative optimization control method for a commercial vehicle engine in a specific embodiment, including the following execution steps: Step 100: Collect vehicle operation data, driver operation data, and environmental parameters, and preprocess them to generate a structured dataset. Specifically, the onboard terminal data acquisition module records driver operation behavior and vehicle driving data in real time, including accelerator pedal opening, braking frequency, gear shifting timing, engine speed, vehicle speed, fuel consumption, and external environmental parameters, providing raw input for subsequent data preprocessing.

[0020] In a preferred embodiment, the vehicle-mounted terminal acquisition module establishes real-time communication with the engine ECU, transmission, ABS system, and environmental sensors via the vehicle's CAN bus interface, acquiring raw data at a frequency of 10 times per second. Specific acquisition parameters include accelerator pedal opening (0%–100%), brake pedal trigger frequency (times / minute), gear shift timing, engine speed range (800–2500 r / min), vehicle speed (0–120 km / h), instantaneous fuel consumption rate (g / L), and external environmental parameters such as ambient temperature (-20–40℃) and atmospheric pressure (80–110 kPa). The data acquisition device uses an industrial-grade STM32H743 MCU chip to ensure stability under operating conditions of -40–85℃. Acquired data is uploaded to a cloud server in real-time via a 5G module, and simultaneously backed up on a local SD card with a storage capacity of 128 GB. The timestamp accuracy of the sampled data reaches the millisecond level.

[0021] In one specific implementation, after performing step 100, a further step is performed to remove outliers, fill in missing values, and standardize the collected data to form a structured dataset. In a preferred embodiment, a distributed computing architecture is employed, with preliminary processing completed at edge nodes. Outlier removal utilizes a modified IQR method, setting dynamic thresholds for each parameter; for example, the outlier threshold for engine speed is the mean ± 3 times the standard deviation. Missing value imputation employs an LSTM-based time series prediction model, imputing data with no more than 5 consecutive missing sampling points. Standardization uses the Z-score method, mapping all features to the 0-1 range. The preprocessing workflow sets a processing window of 5 minutes, applying a moving average filter to the data within the window, with a filter window size of 10 sampling points. The processed structured dataset is transmitted to the feature extraction module via the MQTT protocol, with data transmission latency controlled within 200ms.

[0022] Step 101: Extract driving style features, road condition features, and engine operation features from the structured dataset to generate feature vectors. Specifically, when executing step 101, the following steps can be performed: Based on the structured dataset, a three-level feature system is constructed; wherein: Driving style features are calculated using a sliding time window, including the percentage of rapid accelerations, frequency of rapid decelerations, and gear shift rationality indicators. Road condition features are calculated by fusing GPS data and map information, including average gradient, road curvature, and road surface smoothness indicators. Engine operation features include the percentage of idling time and load rate distribution; the percentage of idling time is the ratio of the cumulative time the engine speed is below the idling threshold to the total operating time. The importance scores of driving style features, road condition features, and engine operation features to the driving style classification task are calculated, and features that meet the contribution requirements of driving style classification are selected to form a feature vector.

[0023] In a preferred embodiment, a three-level feature system is constructed based on preprocessed data. Driving style features are calculated using a sliding time window of 5 minutes, including the percentage of rapid accelerations (defined as the ratio of the number of times the throttle opening change rate exceeds 5% / second to the total number of samples), the frequency of rapid decelerations (the frequency of events where the brake pedal opening change rate exceeds 8% / second), and the shift rationality index (the sum of squares of the deviations between the actual shift speed and the optimal economic speed). Road condition features are calculated by fusing GPS data and high-precision maps, including average gradient (the ratio of road segment elevation difference to road segment length), road curvature (calculated using the radius of the fitted circle from five adjacent GPS points), and road surface smoothness index (based on the root mean square value of suspension vibration sensors). Engine operation features include the percentage of idling time (the ratio of the cumulative time below 850 r / min to the total operating time) and load distribution (statistically calculating the percentage of operating time in different load ranges of 20%–40%, 40%–60%, and 60%–80%). All features are dimensionality reduced using the ReliefF feature selection algorithm, retaining the top 30 important features to form the final feature vector.

[0024] Step 102: Based on the feature vector, the driver's driving style is classified using a pre-built behavior classification model to obtain a driving style label.

[0025] Specifically, when executing step 102, the following steps can be performed: S1020: Input the feature vector into the encoding layer of the deep clustering network for nonlinear transformation and dimensionality reduction to obtain a low-dimensional embedding vector.

[0026] S1021: In the clustering layer, based on the distance between the low-dimensional embedding vector and the preset cluster center, a probability distribution corresponding to different driving style categories is generated.

[0027] Preferred category determination rules: Aggressive type: rapid acceleration percentage > 20%, shift speed deviation > 300 rpm, idle speed percentage < 10%. Stable type: rapid acceleration percentage < 5%, shift speed deviation < 100 rpm, idle speed percentage > 20%. Intermediate type is determined by softmax probability allocation (e.g., probability > 0.7 is used to determine the corresponding category).

[0028] S1022: Determine the driving style classification result based on the probability distribution and output it as a digital style label.

[0029] The behavior classification model is built on a deep clustering network, which includes an encoding layer formed by a multi-layer fully connected neural network, a probability allocation based on softmax and a clustering layer with clustering loss as the objective function, and a self-supervised fine-tuning layer that introduces an auxiliary loss function.

[0030] The deep clustering network is trained using a composite objective function that includes clustering loss and auxiliary loss to simultaneously optimize feature encoding and clustering assignment.

[0031] The training process of the deep clustering network includes: using the K-means++ algorithm to pre-cluster the embedding vectors output by the encoding layer to initialize the cluster centers; then iteratively updating the network parameters of the encoding layer by minimizing the composite objective function and combining it with backpropagation; the updating of the network parameters and the correction of the cluster centers are carried out alternately until the deep clustering network reaches a convergent state.

[0032] For example, the clustering results generate driver style labels (such as "aggressive" or "stationary") and encode them as style feature vectors z∈Rk (when k=2, aggressive z=[1,0], stationary z=[0,1]), which are used as inputs to the dynamic policy generation model to achieve personalized adaptation of control policies.

[0033] Step 103: Based on the driving style tags, road condition characteristics, and engine operating characteristics, a pre-built dynamic strategy is used to generate a model to adjust the fuel-saving control parameters.

[0034] The dynamic strategy generation model is built on the Actor-Critic network architecture.

[0035] Specifically, when executing step 103, the following steps can be performed: S1030: Encode the driving style label into a style feature vector, and fuse it with road condition features and engine operation features to construct a state vector.

[0036] S1031: Input the state vector into the Actor network and output the probability distribution of the continuous action space.

[0037] The probability distribution of the continuous action space includes the probability distribution of actions related to engine fuel injection timing, intake air volume, and ignition advance angle.

[0038] S1032: Use the Critic network to simultaneously evaluate the value of the state vector and the corresponding action.

[0039] S1033: Use a multi-objective weighted reward function for policy optimization.

[0040] The reward function includes at least a basic fuel-saving reward item and a driving smoothness reward item, and introduces a style adaptation factor that matches the driving style and a road condition weight factor that adapts to real-time road conditions. The reward signal is generated by weighted fusion to guide the output action of the Actor network, thereby generating adjustment instructions for the fuel-saving control parameters.

[0041] In a preferred embodiment, a dynamic policy generation model is constructed based on the Actor-Critic architecture, encoding driver style labels (such as aggressive / stationary) into style feature vectors z∈R. k (where k is the number of style categories), and real-time road condition data (including slope θ, road curvature κ, real-time vehicle speed v, traffic flow density ρ) and engine parameters are fused into a state vector st=[z,θ,κ,v,ρ,n,p] (n is the engine speed, p is the current injection pressure).

[0042] The Actor network employs a 3-layer fully connected structure, taking the state vector st as input and outputting the continuous action space a. t =[τ, ,λ](τ is the fuel injection timing, The probability distribution of intake air volume (where λ is the ignition advance angle); the Critic network synchronously evaluates the value V(s) of state-action pairs. t ,a t The reward function is designed as a multi-objective optimization: Basic terms: (c is the fuel consumption coefficient, (Instantaneous fuel consumption rate), and introduces a style adaptation factor ωstyle (aggressive driver ω style =0.3, stable driver ω style =0.7) and road condition complexity weight ωroad=1+a·κ+b·θ, where a is the road curvature weight coefficient and b is the slope weight coefficient. Final reward R t =ω style ·R fuel +ω road ·R smooth ; (R smooth(The reward is for smooth driving and is inversely proportional to acceleration fluctuations). During model training, a policy iteration is triggered every 100 decision steps, dynamically adjusting the reward function weights based on the actual fuel savings feedback. ; and The average fuel consumption before and after optimization are respectively; the reward function weights are dynamically adjusted: if Δη < 5%, then ω is increased. style Weight; if Δη > 15%, then decrease ω. road To balance dynamic performance, the Critic network parameters w and Actor network policy parameters θ are updated using a temporal difference algorithm to achieve closed-loop iterative optimization of the policy; the policy iterative update formula is: In the formula: θ represents the Actor network policy parameters; This represents the Actor network policy parameters updated at time t+1, determined by... Calculated using an iterative formula, The gradient operator with respect to the policy parameter θ is used to solve... The partial derivative with respect to θ guides the direction of parameter updates, and α represents the policy learning rate (initial value 0.001, dynamically adjusted with Δη, where Δη represents the actual fuel saving rate). Represents the policy function (given state) Time action (probability distribution) The reward signal at time t is dynamically generated based on the real-time fuel saving rate Δη, where the higher the fuel saving rate, the greater the reward value. The action variable selected at time t is the adjustment amount of the engine control parameters (such as the fuel injection timing correction value, the intake air volume adjustment step size, etc.). The state vector at time t contains the driver's driving style label, real-time road condition features (such as gradient and vehicle speed), and engine operating parameters (such as speed and fuel consumption). The state vector at time t+1 represents the action being performed. The new state that the system transitions to reflects the impact of actions on the vehicle's operating state. w represents the weight parameters of the Critic network, used to fit the state value function. The value estimation error is minimized during the update, where γ represents the discount factor (a fixed value of 0.95). This represents the Critic network value function (with parameter w).

[0043] In a preferred embodiment, the dynamic policy generation model employs a deep reinforcement learning Actor-Critic architecture. The Actor network is a 3-layer fully connected neural network with an input layer dimension equal to the state vector dimension of 35, and hidden layers containing 128 and 64 neurons respectively, using the ReLU activation function. The output layer generates a probability distribution of the continuous action space through Softmax activation. The Critic network is a 2-layer fully connected structure with the same input layer dimension as the Actor network, 256 hidden layers, and uses the LeakyReLU activation function. It outputs the value assessment of state-action pairs. Model training uses the Asynchronous Advantage Actor-Critic algorithm (A3C), with four parallel training threads. Each thread interacts independently with the environment, and the experience replay pool capacity is set to 10,000 samples, employing a priority-based experience replay mechanism. The reward function is calculated every 1 second. Each policy iteration is triggered when 100 decision steps have been accumulated or when the evaluation feedback signal from S8 is received. During the iteration process, the initial value of the learning rate is set to 0.001. When the actual fuel saving rate Δη is less than 5%, the learning rate is automatically increased by 1.2 times. When Δη is greater than 15%, the learning rate is reduced to 0.8 times.

[0044] For example, driver style labels (e.g., aggressive / moderate) are encoded into style feature vectors z∈Rk (k being the number of style categories), and fused with real-time road condition data (slope θ, road curvature κ, vehicle speed v, traffic density ρ) and engine parameters (speed n, injection pressure p) to form a state vector: st=[z,θ,κ,v,ρ,n,p]. The reward function is optimized using multi-objectives, designed as: Rt=ωstyle·Rfuel+ωroad·Rsmooth, with the basic fuel-saving term Rfuel: Rfuel= c·mf (c is the fuel consumption coefficient, mf is the instantaneous fuel consumption rate), directly reflects fuel economy; Style adaptation factor ωstyle: aggressive driver ωstyle=0.3 (emphasizing power), moderate driver ωstyle=0.7 (emphasizing fuel economy); Road condition complexity weight ωroad: ωroad=1+0.5κ+0.3θ (κ is road curvature, θ is slope, the higher the complexity, the greater the weight); Driving smoothness reward Rsmooth: inversely proportional to acceleration fluctuation, ensuring driving comfort. Dynamic weight adjustment based on fuel saving rate: using the actual fuel saving rate Δη (Δη=(fuel consumption before optimization)) as feedback. Adjusting the weight of ωstyle (both before and after optimization): If Δη < 5% (poor fuel saving effect), increase the weight of ωstyle to enhance style adaptation; if Δη > 15% (excessive fuel saving affects power), decrease the weight of ωroad to balance power and smoothness.

[0045] In a preferred embodiment, the multi-parameter co-optimizer is built based on the Model Predictive Control (MPC) framework, with an optimization cycle of 200ms. Optimization variables include injection timing (adjustment range: 10°-30°CA before top dead center), intake air volume (control accuracy ±0.5g / kg), and ignition advance angle (adjustment step size: 0.5°CA). The optimization objective function focuses on minimizing fuel consumption while simultaneously satisfying emission constraints (NOx emissions ≤5g / kW·h) and power performance constraints (torque response delay ≤300ms). The optimization algorithm employs a modified Particle Swarm Optimization (PSO) algorithm with a population size of 30 particles, a maximum iteration count of 20, and a dynamic inertia weight (initial value 0.9, linearly decreasing to 0.4 with each iteration). During optimization, a forward simulation of the engine's physical model is performed to predict engine performance parameters for the next 5 steps. The simulation model accuracy error is controlled within 3%, and the optimization results are transmitted to the engine ECU via real-time Ethernet.

[0046] In a preferred embodiment, the engine electronic control unit is equipped with an external signal processing module, which adopts a dual-core processor architecture. The main core is a 32-bit PowerPC processor with a main frequency of 800MHz, which is responsible for the real-time execution of optimization parameters, while the slave core is responsible for data acquisition and communication.

[0047] Step 104: Evaluate the fuel-saving effect based on the adjusted fuel-saving control parameters, generate evaluation feedback, and perform strategy iteration based on the evaluation feedback to generate an engine control strategy.

[0048] Specifically, when executing step 104, the following steps can be performed: S1040: Using the actual fuel-saving rate from the evaluation feedback, adjust the weight ratio of the style adaptation factor and the road condition weight factor in the reward function.

[0049] Specifically, the style adaptation factor is set with different values ​​according to the driving style category, and the road condition weight factor is calculated based on road curvature and slope. In a preferred embodiment, the fuel-saving effect evaluation employs a double-blind comparative experimental design, using the same driving route before and after the optimization control as a control group. Evaluation indicators include average fuel consumption rate (L / 100km), comparison difference, fuel saving rate (calculated as the difference in fuel consumption before and after optimization divided by the fuel consumption before optimization), and emission indicators (monitoring the mass concentrations of NOx, HC, and PM, in mg / m³). The evaluation period is set to be completed within 5 minutes after the end of a single trip. Data processing uses Welch's t-test to verify the statistical significance of the fuel-saving effect, with a significance level set at 0.05. The evaluation report is stored in a distributed database using an HBase architecture, with a data retention period of one year. Simultaneously, key evaluation indicators such as the fuel saving rate Δη and emission improvement rate are pushed to the strategy for correction via a message queue. The trigger condition is that the strategy correction process is forcibly initiated when Δη < 3%.

[0050] S1041: Using the reward signal output by the adjusted reward function as the optimization objective, calculate the temporal difference between the objective value and the current value estimate, periodically update the parameters of the Critic network in the opposite direction of the value estimate gradient, and synchronously update the parameters of the Actor network based on the gradient of the temporal difference and the policy probability, thereby realizing the closed-loop iterative optimization of the control policy.

[0051] Step 105: Based on the engine control strategy, perform coordinated optimization of multiple control parameters of the engine.

[0052] In one specific approach, the multi-parameter collaborative optimization control method for commercial vehicle engines further includes the following steps: SA: A training dataset is constructed based on historical evaluation data and newly added driving cases. When the cumulative sample size of the training dataset does not reach a preset threshold, incremental training is performed on the behavior classification model and the dynamic policy generation model. When the preset threshold is reached, full retraining is performed on the behavior classification model and the dynamic policy generation model.

[0053] SB: The model update process consists of the following stages executed sequentially.

[0054] SB0: The first stage updates the behavior classification model by using transfer learning to freeze the underlying parameters of the pre-trained behavior classification model and fine-tune the parameters of the top classification layer.

[0055] SB1: The second stage updates the dynamic policy generation model, fixes the underlying weights of the Actor network, and adjusts the parameters of its upper-layer network; wherein, in the second stage, an early stopping strategy is adopted, and training is terminated when the validation set loss no longer decreases for several consecutive rounds.

[0056] SB2: The updated behavior classification model and dynamic strategy generation model will be deployed in stages and first tested on a small number of vehicles. If no anomalies are found within the preset verification period and the performance indicators meet expectations, the model will be pushed to all vehicles. In addition, a version rollback mechanism will be retained during the deployment process. If the fuel saving rate of the new model decreases by more than the preset tolerance threshold, the system will automatically switch to the old version model.

[0057] In a preferred embodiment, the model iteration update process employs an incremental learning mechanism, automatically starting the update program every Sunday at 3:00 AM. The training dataset includes accumulated historical evaluation data and newly added driving cases. Full retraining is triggered when the cumulative sample size reaches 10,000 trip data points; otherwise, incremental training is performed. The model update process consists of two phases. First, the behavior classification model is updated using a transfer learning method, freezing the first 8 layers of the pre-trained BERT model and only fine-tuning the classification head, with a learning rate of 0.0005 and 20 training epochs. Second, the strategy generation model is updated, fixing 60% of the weights at the bottom layer of the Actor network and only adjusting the parameters of the upper layers, employing an early stopping strategy, terminating training when the validation set loss no longer decreases for 5 consecutive epochs. The updated model is deployed through A / B testing, first covering 20% ​​of vehicles for 24 hours of verification without any anomalies before full rollout. A rollback mechanism is retained during model deployment; if the fuel-saving rate of the new model decreases by more than 2%, it automatically switches to the old version.

[0058] In summary, the beneficial effects of this application are: 1. By using a deep learning model to accurately cluster driver behavior according to driving style, personalized adaptation of control strategies is achieved. The weights of the reinforcement learning reward function are dynamically adjusted based on driving style labels, enabling aggressive drivers to reduce fuel consumption while ensuring power requirements are met, and further optimizing the economic operating range for moderate drivers, effectively improving fuel-saving control accuracy under different driving habits. Combining real-time road condition features with multi-dimensional fusion of engine operating parameters, the reinforcement learning model can dynamically generate control strategies adapted to the current scenario, enhancing the system's adaptability to complex road conditions and changes in driving behavior.

[0059] 2. The multi-parameter collaborative optimization mechanism achieves globally optimal matching of the engine's core control parameters. By coupling and adjusting injection timing, intake air volume, and ignition advance angle, it overcomes the limitations of traditional single-parameter optimization, significantly improving engine operating efficiency. The system introduces a closed-loop feedback iteration mechanism, continuously refining the strategy model using actual fuel-saving effect evaluation data. Combined with regular historical data training and updates, it continuously strengthens the model's adaptability to diverse driving scenarios and vehicle states, achieving continuous optimization of fuel-saving performance and driving smoothness over long-term use.

[0060] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0061] like Figure 2 As shown, the following are embodiments of the multi-parameter collaborative optimization control system for commercial vehicle engines provided in this disclosure. These embodiments belong to the same inventive concept as the multi-parameter collaborative optimization control methods for commercial vehicle engines described above. For details not described in detail in the embodiments of the multi-parameter collaborative optimization control system for commercial vehicle engines, please refer to the embodiments of the multi-parameter collaborative optimization control methods for commercial vehicle engines described above.

[0062] A multi-parameter collaborative optimization control system for commercial vehicle engines, including: The data acquisition unit is used to collect vehicle operation data, driver operation data and environmental parameters, and to preprocess them to generate structured datasets. The feature extraction unit is used to extract driving style features, road condition features, and engine operation features from the structured dataset and generate feature vectors. The behavior classification unit is used to classify the driver's driving style based on the feature vector using a pre-built behavior classification model to obtain a driving style label; The parameter adjustment unit is used to adjust the fuel-saving control parameters based on the driving style label, road condition characteristics and engine operating characteristics using a pre-built dynamic strategy generation model. The strategy generation unit is used to evaluate the fuel-saving effect based on the adjusted fuel-saving control parameters, generate evaluation feedback, and perform strategy iteration based on the evaluation feedback to generate an engine control strategy. The dynamic strategy generation model is built on the Actor-Critic architecture. The parameter optimization unit is used to collaboratively optimize multiple control parameters of the engine according to the engine control strategy.

[0063] Figure 3 This is a schematic diagram of the hardware structure of an electronic device that implements various embodiments of the present invention.

[0064] The multi-parameter collaborative optimization control method for commercial vehicle engines provided in this application can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiments of this invention does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the electronic device includes, but is not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0065] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0066] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device. In other embodiments of this application, the electronic device may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0067] A processor may include one or more processing units, such as: a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0068] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0069] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0070] An external storage interface (ESI) can be used to connect external memory cards, such as microSD cards, to expand the storage capacity of electronic devices. The external memory card communicates with the processor through the ESI to perform data storage functions, such as saving music and video files on the external memory card.

[0071] Internal memory can be used to store computer executable program code, which includes instructions. The processor executes various functional applications and data processing of electronic devices by running the instructions stored in internal memory. Internal memory can include a program storage area and a data storage area. Internal memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash storage (UFS), etc.

[0072] Wireless communication functionality in electronic devices can be achieved through antennas, wireless communication modules, modem processors, and baseband processors.

[0073] Wireless communication modules can provide solutions for wireless communication applications in electronic devices, including wireless local area networks (WLANs) (such as wireless fidelity (Wi-Fi) networks), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), and infrared (IR) technologies.

[0074] Electronic devices can implement audio functions through audio modules, speakers, receivers, microphones, headphone jacks, and application processors.

[0075] Electronic devices can achieve shooting functions through ISPs, cameras, video codecs, GPUs, displays, and application processors.

[0076] Electronic devices can achieve display functions through GPUs, displays, and application processors.

[0077] A GPU is a microprocessor for image processing, connected to the display screen and application processor. GPUs are used to perform mathematical and geometric calculations for graphics rendering. A processor may include one or more GPUs, which execute program instructions to generate or modify display information.

[0078] A display screen is used to display images, videos, etc. A display screen includes a display panel.

[0079] The storage medium provided in this application stores a program product capable of implementing a multi-parameter collaborative optimization control method for commercial vehicle engines.

[0080] A multi-parameter collaborative optimization control method for commercial vehicle engines includes: collecting vehicle operation data, driver operation data, and environmental parameters, and preprocessing them to generate a structured dataset; extracting driving style features, road condition features, and engine operation features from the structured dataset to generate feature vectors; classifying the driver's driving style using a pre-built behavior classification model based on the feature vectors to obtain driving style labels; adjusting fuel-saving control parameters using a pre-built dynamic strategy generation model based on the driving style labels, road condition features, and engine operation features; evaluating the fuel-saving effect based on the adjusted fuel-saving control parameters, generating evaluation feedback, and iterating the strategy according to the evaluation feedback to generate an engine control strategy, wherein the dynamic strategy generation model is built based on an Actor-Critic network architecture; and collaboratively optimizing multiple control parameters of the engine according to the engine control strategy.

[0081] In some possible implementations, the subject matter of this disclosure, namely, the multi-parameter collaborative optimization control method and system for commercial vehicle engines, can be implemented as a program product comprising program code. When the program product is run on a terminal device, the program code is used to cause the terminal device to perform the steps described in the "Exemplary Methods" section above, according to various exemplary embodiments of this disclosure.

[0082] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0083] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-parameter collaborative optimization control method for a commercial vehicle engine, characterized in that, include: Collect vehicle operation data, driver operation data, and environmental parameters, and preprocess them to generate a structured dataset; Driving style features, road condition features, and engine operation features are extracted from the structured dataset to generate feature vectors; Based on the feature vector, a pre-built behavior classification model is used to classify the driver's driving style and obtain driving style labels; Based on the driving style tags, road condition characteristics, and engine operating characteristics, a pre-built dynamic strategy generation model is used to adjust fuel-saving control parameters. The dynamic strategy generation model is built on the Actor-Critic network architecture. The fuel-saving effect is evaluated based on the adjusted fuel-saving control parameters, evaluation feedback is generated, and the strategy is iterated based on the evaluation feedback to generate an engine control strategy. Based on the engine control strategy, multiple control parameters of the engine are optimized collaboratively.

2. The multi-parameter collaborative optimization control method for commercial vehicle engines according to claim 1, characterized in that, Driving style features, road condition features, and engine operation features are extracted from the structured dataset to generate feature vectors, including: Based on the structured dataset, a three-level feature system is constructed; wherein: Driving style characteristics are calculated through a sliding time window, including the percentage of rapid accelerations, the frequency of rapid decelerations, and the rationality index of gear shifting. The road condition features are calculated by fusing GPS data and map information, and include average slope, road curvature and road surface smoothness indicators. Engine operating characteristics include idling time percentage and load rate distribution; wherein, the idling time percentage is the ratio of the cumulative time when the engine speed is below the idling threshold to the total operating time; Calculate the importance scores of driving style features, road condition features, and engine operation features to the driving style classification task, and select features that meet the contribution requirements of driving style classification to form a feature vector.

3. The multi-parameter collaborative optimization control method for commercial vehicle engines according to claim 1, characterized in that, Based on the feature vectors, a pre-built behavior classification model is used to classify the driver's driving style, resulting in driving style labels, including: The feature vector is input into the encoding layer of a deep clustering network for nonlinear transformation and dimensionality reduction to obtain a low-dimensional embedding vector. In the clustering layer, a probability distribution corresponding to different driving style categories is generated based on the distance between the low-dimensional embedding vector and the preset cluster center. The driving style classification result is determined based on the probability distribution and output as a digital style label; Among them, the behavior classification model is built on a deep clustering network, including an encoding layer formed by a multi-layer fully connected neural network, a probability allocation based on softmax and a clustering layer with clustering loss as the objective function, and a self-supervised fine-tuning layer that introduces an auxiliary loss function. The deep clustering network is trained using a composite objective function that includes clustering loss and auxiliary loss to simultaneously optimize feature encoding and clustering assignment. The training process of the deep clustering network includes: using the K-means++ algorithm to pre-cluster the embedding vectors output by the encoding layer to initialize the cluster centers; then iteratively updating the network parameters of the encoding layer by minimizing the composite objective function and combining it with backpropagation; the updating of the network parameters and the correction of the cluster centers are carried out alternately until the deep clustering network reaches a convergent state.

4. The multi-parameter collaborative optimization control method for commercial vehicle engines according to claim 1, characterized in that, The step of adjusting fuel-saving control parameters using a pre-built dynamic strategy generation model based on the driving style tags, road condition characteristics, and engine operating characteristics includes: The driving style label is encoded into a style feature vector, and then fused with road condition features and engine operation features to construct a state vector. The state vector is input into the Actor network, and the probability distribution of the continuous action space is output. The probability distribution of the continuous action space includes the action probability distribution of engine injection timing, intake air volume and ignition advance angle. The value of the state vector and its corresponding action is evaluated simultaneously using a Critic network. A multi-objective weighted reward function is used for strategy optimization. This reward function includes at least a basic fuel-saving reward item and a driving smoothness reward item. A style adaptation factor that matches the driving style and a road condition weight factor that adapts to real-time road conditions are introduced. A comprehensive reward signal is generated by weighted fusion to guide the output action of the Actor network, thereby generating adjustment instructions for the fuel-saving control parameters.

5. The multi-parameter collaborative optimization control method for commercial vehicle engines according to claim 4, characterized in that, The process of evaluating fuel-saving effects based on adjusted fuel-saving control parameters, generating evaluation feedback, and iterating strategies based on the evaluation feedback includes: The weight ratio of style adaptation factor and road condition weight factor in the reward function is adjusted based on the actual fuel saving rate from the evaluation feedback. Using the reward signal output by the adjusted reward function as the optimization objective, the temporal difference between the objective value and the current value estimate is calculated. The parameters of the Critic network are periodically updated in the opposite direction of the value estimation gradient. Based on the gradient of the temporal difference and the policy probability, the parameters of the Actor network are updated synchronously to achieve closed-loop iterative optimization of the control policy.

6. The multi-parameter collaborative optimization control method for commercial vehicle engines according to claim 5, characterized in that, The style adaptation factor is set with different values ​​according to the driving style category, and the road condition weight factor is calculated based on road curvature and slope.

7. The multi-parameter collaborative optimization control method for commercial vehicle engines according to claim 1, characterized in that, The iterative update process for the behavior classification model and the dynamic policy generation model includes: A training dataset is constructed based on historical evaluation data and newly added driving cases. When the cumulative number of samples in the training dataset does not reach a preset threshold, incremental training is performed on the behavior classification model and the dynamic policy generation model. When the preset threshold is reached, full retraining is performed on the behavior classification model and the dynamic policy generation model. The model update process includes the following stages executed sequentially: The first stage updates the behavior classification model by using transfer learning to freeze the underlying parameters of the pre-trained behavior classification model and fine-tune the parameters of the top classification layer. The second stage updates the dynamic policy generation model by fixing the underlying weights of the Actor network and adjusting the parameters of its upper-layer network. In this second stage, an early stopping strategy is adopted, and training is terminated when the validation set loss no longer decreases for several consecutive rounds. The updated behavior classification model and dynamic strategy generation model will be deployed in stages. First, they will be tested on a small number of vehicles. If no abnormalities occur within the preset verification period and the performance indicators meet expectations, they will be pushed to all vehicles. In addition, a version rollback mechanism will be retained during the deployment process. If the fuel saving rate of the new model decreases by more than the preset tolerance threshold, the system will automatically switch to the old version model.

8. A multi-parameter collaborative optimization control system for a commercial vehicle engine, characterized in that, include: The data acquisition unit is used to collect vehicle operation data, driver operation data and environmental parameters, and to preprocess them to generate structured datasets. The feature extraction unit is used to extract driving style features, road condition features, and engine operation features from the structured dataset and generate feature vectors. The behavior classification unit is used to classify the driver's driving style based on the feature vector using a pre-built behavior classification model to obtain a driving style label; The parameter adjustment unit is used to adjust the fuel-saving control parameters based on the driving style label, road condition characteristics and engine operating characteristics using a pre-built dynamic strategy generation model. The strategy generation unit is used to evaluate the fuel-saving effect based on the adjusted fuel-saving control parameters, generate evaluation feedback, and perform strategy iteration based on the evaluation feedback to generate an engine control strategy. The dynamic strategy generation model is built on the Actor-Critic architecture. The parameter optimization unit is used to collaboratively optimize multiple control parameters of the engine according to the engine control strategy.

9. An electronic device 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 program, it implements the steps of the multi-parameter collaborative optimization control method for commercial vehicle engines as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-parameter collaborative optimization control method for commercial vehicle engines as described in any one of claims 1 to 7.

Images