Full-active suspension system control method based on rule learning
By using a rule-based learning method to mine the mode switching boundary of the suspension system, the problem of precise control of the suspension system under complex operating conditions is solved, the adaptability and real-time performance of the suspension system are realized, the comfort and stability of the vehicle are improved, and energy consumption is reduced.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEFEI UNIV OF TECH
- Filing Date
- 2026-01-30
- Publication Date
- 2026-05-12
AI Technical Summary
Existing suspension systems struggle to achieve precise multi-mode switching under complex and varied driving conditions, impacting passenger comfort and vehicle stability. Furthermore, existing switching rules are highly subjective and have ambiguous boundaries, posing potential safety risks.
A rule-based learning approach is adopted, which uses the k-means algorithm to cluster vehicle states and combines it with the RIPPER algorithm to extract rules, forming clear mode switching boundaries. Switching rules between semi-active and fully active suspension modes are designed, and the adaptive and real-time control of the suspension system is achieved by utilizing the inherent correlation between vehicle state and operating conditions.
It improves the accuracy of suspension mode adaptation, reduces sudden changes in vehicle posture, enhances the comfort and handling stability of the vehicle in complex road conditions, reduces energy consumption, enhances the adaptability and interpretability of the suspension system, and ensures the timeliness and safety of control response.
Smart Images
Figure FT_1 
Figure FT_2 
Figure FT_3
Abstract
Description
Technical Field
[0001] This invention relates to the fields of rule learning and suspension multi-mode switching, specifically to a rule-based fully active suspension system control method. Background Technology
[0002] During operation, vehicles often face complex and ever-changing external conditions, experiencing multi-frequency and time-varying vibrations from the road surface. This places extremely high demands on the ride comfort and handling stability of the suspension system. Although some progress has been made in suspension control technology research—for example, optimal control algorithms such as model predictive control (MPC) have shown some effectiveness in multivariable suspension optimization—limitations remain in the field of multi-mode switching control.
[0003] The development of suspension systems has significantly improved vehicle handling capabilities on various terrains. A good suspension system can improve vehicle ride comfort and handling stability. Commonly used damping systems are divided into passive suspension, semi-active suspension, and fully active suspension. Passive suspension is simple in structure and low in cost, but its parameters cannot be adjusted, limiting its damping effect. Fully active suspension can generate forces that are adjustable in both magnitude and direction, offering good damping performance, but its widespread adoption is hindered by high energy consumption, complex structure, and high cost. Semi-active suspension offers performance between that of passive and fully active suspension, but the limited range of mechanical properties of its actuators restricts its optimal control performance across the entire frequency range.
[0004] Due to the complexity and variability of vehicle operating conditions, the requirements for suspension performance vary significantly under different conditions. Existing control solutions typically employ a single control strategy or simple multi-mode switching logic to address vibration issues, resulting in an inability to adapt to dynamic changes across multiple operating conditions. Current suspension multi-mode switching largely relies on manually preset fixed rules and thresholds. Rule design is highly subjective and lacks interpretability, making it difficult to cover complex operating conditions involving dynamic coupling such as road surface type and vehicle speed. Furthermore, existing suspension multi-mode switching logic only sets switching conditions based on a single state parameter, such as the vehicle acceleration threshold, failing to explore the intrinsic relationship between multi-dimensional vehicle states and operating conditions, and thus failing to form clear and interpretable mode switching boundaries. This leads to vague switching decision criteria and low logical transparency, making it difficult to accurately match the suspension performance requirements under different operating conditions. Furthermore, unclear boundaries can easily cause deviations in mode switching timing, affecting not only passenger comfort but also posing potential risks to vehicle system stability and passenger safety. Summary of the Invention
[0005] The present invention addresses the shortcomings of the existing technology by proposing a rule-based learning-based fully active suspension system control method. This method aims to achieve suspension system mode switching and optimal control, and enhance the adaptability, interpretability, and real-time performance of the suspension system, thereby optimizing the comfort and handling stability of the suspension under complex operating conditions.
[0006] The present invention adopts the following technical solution to solve the technical problem: The characteristic of the rule-based learning-based fully active suspension system control method of the present invention is that it is carried out in the following steps: Step 1: Input the road surface excitation into the vehicle dynamics model, obtain the relevant state parameters of the vehicle dynamics model, and form a vehicle state dataset. ;in, Indicates the first A sequence of vehicle state response parameters, and , express The Middle Vehicle status, express The total number of vehicle statuses. This represents the total number of vehicle state response parameter sequences; Define the mode of the fully active suspension system as follows: If the fully active suspension system is set to semi-active suspension mode, then let If the fully active suspension system is in fully active suspension mode, then let ; make middle The first vehicle state response parameter sequence The optimal fully active suspension system mode corresponding to the vehicle state is denoted as: Thus, the optimal fully active suspension system mode sequence corresponding to each vehicle state is obtained. ; Step 2: Use the k-means algorithm to analyze the first... A sequence of vehicle state response parameters Clustering is performed to obtain Clustering results ;in, Indicates the first A cluster of vehicle state response parameter sequences Indicate the number of clusters after clustering; and The corresponding vehicle state response parameter sequence number j and cluster number u are used as cluster class labels, denoted as... This yields a set of cluster labels; The maximum and minimum values of each vehicle state response parameter sequence are used as... The clustering range; Step 3: Divide the pattern sequence Matching with the set of cluster labels to form a dataset containing cluster labels and patterns. ; Step 4: Use the RIPPER algorithm for rule learning to... Extract the potential rules from the data to form the final rule set. ;in, Indicates the first Optimization rules, This represents the default rule, and N represents the total number of optimization rules. Step 5: Obtain the fully active suspension system data before time t. Historical patterns And calculate the suspension pattern change between two adjacent historical patterns, if If all suspension mode changes remain constant, then the switching signal at time t will be output. Otherwise, the switching signal at time t will be output. ; Step 6: Obtain the first digit at time t. The vehicle state response parameter sequence of the th th Vehicle status and with After mapping the cluster range, with Perform matching to determine the suspension mode at time t. Suspension mode after time t ; Step 7: Combining and and Output the final demand mode of the fully active suspension system This enables control of the fully active suspension system.
[0007] The characteristic of the rule-based fully active suspension system control method described in this invention is that step 4 is performed as follows: Step 4.1: Based on the set extraction rules: ",from Different rules are extracted to form a preliminary rule set. ,in, Indicates the first Preliminary rules, Indicates the total number of preliminary rules; Step 4.2: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require Using the rule as a positive example, The rules are used as counterexamples to calculate the performance metrics of the initial rules, which are then used to evaluate the initial rule set. Pruning optimization is performed to form an optimization rule set. , ; Step 4.3: Set a default rule: ", used for processing Cluster tags not covered in the original text are used to form the final rule set. Among them, conditions express Cluster tags not covered in the code.
[0008] Furthermore, step 7 includes: if and If they are the same, then it's incorrect. Switch, directly As the final demand model Output to suspension controller; otherwise, when At that time, As the final demand model Output to suspension controller; when At that time, As the final demand model Output to the suspension controller.
[0009] The present invention provides an electronic device, including a memory and a processor, characterized in that the memory is used to store a program supporting the processor in performing the method described therein, and the processor is configured to execute the program stored in the memory.
[0010] The present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program is executed by a processor to perform the steps of the method described thereon.
[0011] Compared with existing technologies, the beneficial effects of this invention are reflected in: 1. This invention utilizes the RIPPER algorithm for rule learning to deeply explore the intrinsic relationship between vehicle state and operating conditions, forming clear and interpretable mode switching boundaries. This solves the problems of strong subjectivity and ambiguous boundaries in existing switching rules, improves the adaptation accuracy of suspension modes under complex operating conditions, effectively reduces sudden changes in vehicle posture caused by unclear switching boundaries, and improves the overall performance of the vehicle in complex road conditions, such as comfort, operation stability, and adaptability. 2. This invention uses the k-means algorithm to perform cluster analysis on the vehicle state set, providing clusters and cluster ranges for rule extraction, making the final mode switching rules more intuitive and reliable, and effectively avoiding switching decision deviations caused by data anomalies; 3. The rule-based switching control strategy designed in this invention clearly defines the switching rules between semi-active and fully active modes. The rule learning process involves a small amount of computation, which is conducive to achieving real-time mode switching and ensuring the timeliness of control response. 4. The rule-based learning-based fully active suspension system control method provided by this invention has better vibration reduction effect than passive and semi-active systems, and its energy consumption is lower than that of fully active systems. It is stable and reliable in operation, inheriting the low energy consumption advantage of semi-active systems and combining the high-performance vibration reduction characteristics of fully active systems. It has a wider range of applications and provides strong technical support for the comfort, economy and safety of vehicles under complex road conditions. Attached Figure Description
[0012] Figure 1 This is a flowchart illustrating the rule set generation process for the fully active suspension system of this invention. Figure 2 This is a flowchart of the rule extraction process for the RIPPER algorithm. Figure 3 This is a schematic diagram of the control process of the rule-based fully active suspension system of the present invention; Figure 4 This is a flowchart of the steady-state switching module of the rule-based fully active suspension system of the present invention; Figure 5 This is a flowchart of the rule-based learning-based fully active suspension system switching method of the present invention. Detailed Implementation
[0013] In this embodiment, considering the different requirements for suspension performance under various driving conditions, the suspension operating modes are divided into semi-active and fully active modes. A multi-mode switching control strategy based on rule sets is designed to address the characteristics of each suspension mode. Corresponding switching rules and switching parameter thresholds are established, and a rule-learning-based fully active suspension system control method is designed. This method aims to identify the switching boundaries between different suspension modes, thereby enhancing the adaptiveness, interpretability, and real-time performance of the suspension system, and optimizing the comfort and handling stability of the suspension under complex conditions. Specifically, as... Figure 3 As shown, the method is performed according to the following steps: Step 1: Input the road surface excitation into the vehicle dynamics model, obtain the relevant state parameters of the vehicle dynamics model, and form a vehicle state dataset. ;in, Indicates the first A sequence of vehicle state response parameters, and , express The Middle Vehicle status, express The total number of vehicle statuses. This represents the total number of vehicle state response parameter sequences.
[0014] Define the mode of the fully active suspension system as follows: If the fully active suspension system is set to semi-active suspension mode, then let If the fully active suspension system is in fully active suspension mode, then let ; make middle The first vehicle state response parameter sequence The optimal fully active suspension system mode corresponding to the vehicle state is denoted as: Thus, the optimal fully active suspension system mode sequence corresponding to each vehicle state is obtained. .
[0015] This example uses a nine-degree-of-freedom vehicle dynamics model to generate a vehicle state dataset. The system includes vehicle speed, longitudinal acceleration, vertical acceleration, roll acceleration, and pitch acceleration. The total number of vehicle state response parameter sequences is 6 columns, and the total number of vehicle states is 18,000. The MPC control algorithm is used to obtain the mode sequence of the fully active suspension system corresponding to each vehicle state. A total of 18,000 fully active suspension systems; Initialize t=0, .
[0016] Step 2: Use the k-means algorithm to analyze the first... A sequence of vehicle state response parameters Clustering is performed to obtain Clustering results ;in, Indicates the first A cluster of vehicle state response parameter sequences Indicate the number of clusters after clustering; and The corresponding vehicle state response parameter sequence number j and cluster number u are used as cluster class labels, denoted as... This yields a set of cluster labels; The maximum and minimum values of each vehicle state response parameter sequence are used as... The clustering range.
[0017] The number of clusters after clustering in this example The driving speed is divided into four categories, the longitudinal acceleration of the vehicle body is divided into four categories, the vertical acceleration of the vehicle body is divided into four categories, the roll angle acceleration is divided into four categories, and the pitch angle acceleration is divided into four categories. For example, the cluster labels for driving speed are 11, 12, 13, and 14.
[0018] Step 3: Divide the pattern sequence Matching with the set of cluster labels to form a dataset containing cluster labels and patterns. In this example, the dataset It includes cluster tags and patterns, for example, the suspension patterns corresponding to the cluster tag "11^21^31^41^51^61". .
[0019] Step 4: Use the RIPPER algorithm for rule learning to... Extract the potential rules from the data to form the final rule set. ;in, Indicates the first Optimization rules, This represents the default rule, and N represents the total number of optimization rules, such as... Figure 2 As shown.
[0020] Step 4.1: Extract rules based on the set cluster tags and patterns: ",from Different rules are extracted to form a preliminary rule set. ,in, Indicates the first Preliminary rules, This indicates the total number of initial rules; the rule set extraction process is as follows: Figure 1 As shown.
[0021] Step 4.2: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require Using the rule as a positive example, The rules are used as counterexamples to calculate the performance metrics of the initial rules, which are then used to evaluate the initial rule set. Pruning optimization is performed to form an optimization rule set. , The performance metrics of the algorithm are shown in equation (1): (1) In equation (1), For performance metrics; , These represent the number of positive and negative examples covered by the rule, respectively. , These represent the number of positive and negative examples in the dataset, respectively.
[0022] Step 4.3: Set a default rule: ", used for processing Cluster tags not covered in the original text are used to form the final rule set. Among them, conditions express Cluster tags not covered in the code.
[0023] In this example, a total of 46 preliminary rules are generated for the preliminary rule set. After pruning and optimization, there are 29 rules, ultimately forming readable rules in the form of "IF…,THEN…". An example of the generated rules is shown below: Rule 1: IF 11^21^31^41^51^61, THEN ; Rule 2: IF 13^22^31^42^53^62, THEN .
[0024] Step 5: Switch to steady-state module to obtain the fully active suspension system before time t. Historical patterns And calculate the suspension pattern change between two adjacent historical patterns, if If all suspension mode changes remain constant, then the switching signal at time t will be output. Otherwise, the switching signal at time t will be output. ,like Figure 4 As shown.
[0025] Step 5.1: Define the suspension mode at time t. ,history The suspension modes are respectively represented as follows: Define the set of suspension mode changes , ,in express The Middle The amount of change in each suspension mode -1 indicates the total number of suspension mode changes; input the suspension mode. Send suspension mode switching signal Output to the system stability assessment module.
[0026] Step 5.2: The stability module will evaluate the stability of the input suspension pattern. If it is continuous... The input remains unchanged, that is Then output switching signal The signal is sent to the suspension system mode switching module to change the suspension mode. If it is continuously... During this input, the suspension mode changes, namely If the system is unstable, the output switching signal will be changed. To the suspension system mode switching module; in this example, five historical modes of the fully active suspension system before time t are obtained, namely... .
[0027] Step 6: Obtain the first digit at time t. The vehicle state response parameter sequence of the th th Vehicle status and with After mapping the cluster range, with Perform matching to determine the suspension mode at time t. Suspension mode after time t .
[0028] Step 7: Suspension system switching module integration and and Output the final demand mode of the fully active suspension system To achieve control of the fully active suspension system, such as Figure 5 As shown.
[0029] Step 7.1: Determine the suspension mode obtained in Step 6. If the current suspension mode... With subsequent suspension modes If the modes are the same, there is no need to switch the suspension system mode; the output suspension system requirement mode is... To the suspension controller, i.e. ; Step 7.2: If the current suspension mode With subsequent suspension modes If they are different, a suspension system mode switch is required. This is determined by the steady-state module in step 5, when the output... At this time, the suspension system switches modes, and the output suspension system demand mode is: To the suspension controller, i.e. This enables the switching of modes in the fully active suspension system and controls the fully active suspension system; if the steady-state module in step 5 determines that the suspension mode has changed, it outputs... If the system is unstable, the suspension system will not switch modes, and the output suspension system demand mode will be... To the suspension controller, i.e. This enables control of the fully active suspension system.
[0030] In summary, the method of this invention is applied to vehicles containing semi-active and fully active suspension modes. Road surface excitation is input into the vehicle dynamics model to obtain relevant state parameters, forming a vehicle state dataset. Simultaneously, the patterns of the fully active suspension system corresponding to each vehicle state after optimal control are collected, forming a pattern sequence of the optimal fully active suspension system for each vehicle state. The k-means algorithm is used to perform cluster analysis on the vehicle state dataset, and each cluster is labeled to form a dataset containing cluster labels and patterns. Rule extraction and pruning optimization are performed based on the RIPPER algorithm in rule learning, and default rules are set to ensure coverage of all sample data, ultimately outputting the corresponding rule set. Subsequently, the real-time vehicle state signal is matched with the rule set, and the steady-state module and suspension mode switching module determine the current semi-active / fully active suspension mode to switch to. The suspension mode and control commands are input into the actuator control module to complete precise control of the suspension system.
[0031] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.
[0032] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.
Claims
1. A rule-based learning-based fully active suspension system control method, characterized in that, The procedure is as follows: Step 1: Input the road surface excitation into the vehicle dynamics model, obtain the relevant state parameters of the vehicle dynamics model, and form a vehicle state dataset. ;in, Indicates the first A sequence of vehicle state response parameters, and , express The Middle Vehicle status, express The total number of vehicle statuses. This represents the total number of vehicle state response parameter sequences; Define the mode of the fully active suspension system as follows: If the fully active suspension system is set to semi-active suspension mode, then let If the fully active suspension system is in fully active suspension mode, then let ; make middle The first vehicle state response parameter sequence The optimal fully active suspension system mode corresponding to the vehicle state is denoted as: Thus, the optimal fully active suspension system mode sequence corresponding to each vehicle state is obtained. ; Step 2: Use the k-means algorithm to analyze the first... A sequence of vehicle state response parameters Clustering is performed to obtain Clustering results ;in, Indicates the first A cluster of vehicle state response parameter sequences, Indicate the number of clusters after clustering; and The corresponding vehicle state response parameter sequence number j and cluster number u are used as cluster class labels, denoted as... This yields a set of cluster labels; The maximum and minimum values of each vehicle state response parameter sequence are used as... The clustering range; Step 3: Divide the pattern sequence Matching with the set of cluster labels to form a dataset containing cluster labels and patterns. ; Step 4: Use the RIPPER algorithm for rule learning to... Extract the potential rules from the data to form the final rule set. ;in, Indicates the first Optimization rules, This represents the default rule, and N represents the total number of optimization rules. Step 5: Obtain the fully active suspension system data before time t. Historical patterns And calculate the suspension pattern change between two adjacent historical patterns, if If all suspension mode changes remain constant, then the switching signal at time t will be output. Otherwise, the switching signal at time t will be output. ; Step 6: Obtain the first value at time t. The vehicle state response parameter sequence of the th th Vehicle status and with After mapping the cluster range, with Perform matching to determine the suspension mode at time t. Suspension mode after time t ; Step 7: Combining and and Output the final demand mode of the fully active suspension system This enables control of the fully active suspension system.
2. The rule-based learning-based fully active suspension system control method according to claim 1, characterized in that, Step 4 is performed as follows: Step 4.1: Based on the set extraction rules: ",from Extract different rules to form a preliminary rule set. ,in, Indicates the first Preliminary rules, Indicates the total number of preliminary rules; Step 4.2: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] Using the rule as a positive example, The rules are used as counterexamples to calculate the performance metrics of the initial rules, which are then used to evaluate the initial rule set. Pruning optimization is performed to form an optimization rule set. , ; Step 4.3: Set a default rule: ", used for processing Cluster tags not covered in the original text are used to form the final rule set. Among them, conditions express Cluster tags not covered in the code.
3. The rule-based learning-based fully active suspension system control method according to claim 1, characterized in that, Step 7 includes: If and If they are the same, then it's incorrect. Switch, directly As the final demand model Output to suspension controller; otherwise, when At that time, As the final demand model Output to suspension controller; when At that time, As the final demand model Output to the suspension controller.
4. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports a processor in executing the method of any one of claims 1-3, the processor being configured to execute the program stored in the memory.
5. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the steps of the method according to any one of claims 1-3.