An unmanned parking control system and method

Abstract

This application relates to the field of parking management technology, and in particular to an unmanned parking control system and method. The proposed solution involves constructing a multi-granularity fuzzy rule base, combining vehicle parking error and gravitational potential energy deviation to correct platform position error, thus forming a mechanical error. This error is then dynamically updated based on the error distribution map to intelligently plan the next repositioning target position. Furthermore, an adaptive deformation-adaptive inference weight network structure and path intervention matrix are introduced to improve error correction accuracy and robustness. This method effectively solves the positioning deviation problem during platform repositioning and is applicable to various parking scenarios.

Description

Technical Field

[0001] This application relates to the field of parking management technology, and in particular to an unmanned parking control system and method. Background Technology

[0002] In existing two-level parking systems, vehicle repositioning typically relies on a linkage mechanism to move vehicles vertically or horizontally between upper and lower parking spaces to optimize space utilization. However, due to limitations in the positioning accuracy of the platform control, vehicle positioning errors often occur during the repositioning process. Over time, this can lead to mechanical fatigue or eccentric wear in the fixed docking area, affecting structural stability and operational safety.

[0003] In addition, existing technologies mostly use static preset control commands for platform positioning, lacking a feedback mechanism for actual parking status, changes in vehicle center of gravity, and historical error trends. This makes it difficult to dynamically correct each positioning operation, resulting in insufficient positioning accuracy, frequent vehicle position deviations, and in severe cases, even misalignment, jamming, or repeated calibration, which reduces system operating efficiency and user experience.

[0004] The above problems are particularly prominent in high-frequency parking scenarios with multiple shift operations. To solve these problems, this application designs an unmanned parking control system and method. Summary of the Invention

[0005] The technical problem this application aims to solve is to address the shortcomings of existing technologies by providing an unmanned parking control system and method. The method constructs a multi-granularity fuzzy rule base, combines vehicle parking error with gravitational potential energy deviation to correct platform position errors, thus forming mechanical errors. It then dynamically updates the system based on the error distribution map, intelligently planning the next repositioning target location. Furthermore, it introduces an adaptive deformation-based inference weight network structure and a path intervention matrix to improve error correction accuracy and robustness. This method effectively solves the positioning deviation problem during platform repositioning and is applicable to various parking scenarios.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] An unmanned parking control method is applied to a two-level multi-level parking system, which includes upper-level parking spaces and lower-level parking spaces. When a lower-level parking space is occupied, an empty parking space in the upper level is replaced with an empty parking space in the lower level through a linkage device. The method includes:

[0008] After the parking space repositioning operation is completed, calculate the position error value when the platform controlled by the linkage device reaches the repositioning target position;

[0009] The position error value is corrected by combining the identified vehicle parking error and gravitational potential energy deviation to obtain the mechanical error. The correction includes using the vehicle parking error and gravitational potential energy deviation as input variables for fuzzy inference, and inputting them into a fuzzy rule base including a coarse-grained rule subset, a medium-grained rule subset and a fine-grained rule subset respectively for multi-scale error correction inference.

[0010] The preset error distribution map is updated based on the mechanical error, wherein the error distribution map is used to calculate the target position of the next transposition and generate a transposition control command based on the target position.

[0011] The position error value of the platform controlled by the calculation linkage device when it reaches the target position includes:

[0012] Obtain the actual placement position after the parking space swapping operation is completed, and generate the current physical position of the platform;

[0013] Based on the transposition target position in the transposition control command, a transposition reference value is generated;

[0014] Acquire the platform's status data during the positioning process, wherein the status data includes the platform's acceleration, velocity curve, and transmission feedback information;

[0015] Based on the state data, the transposition reference value is compensated by a preset error estimation model to obtain the theoretical landing position after state compensation, wherein the error estimation model is constructed based on historical transposition state data;

[0016] The position error value is obtained by calculating the difference between the current physical position of the platform and the theoretical landing position after state compensation.

[0017] The correction of the position error value by combining the identified vehicle parking error and gravitational potential energy deviation includes:

[0018] According to the preset fuzzy rule base, the vehicle parking error and gravitational potential energy deviation are used as input variables for fuzzy inference. The position error value is corrected by the fuzzy rule base. The fuzzy rule base includes a first rule subset for coarse-grained error inference, a second rule subset for medium-grained error inference, and a third rule subset for fine-grained error inference.

[0019] The fuzzy rule base also includes an adaptively deformable inference weight network structure and a path intervention matrix. The correction of the position error value using the fuzzy rule base includes:

[0020] The input variables are respectively input into the first rule subset, the second rule subset, and the third rule subset to obtain multiple fuzzy correction output sets corresponding to each granularity rule subset;

[0021] Based on the input variables, the weight tensor nodes corresponding to each granularity rule subset are called in the inference weight network structure to calculate the activation weight coefficients, and the fuzzy correction output set is weighted to obtain the initial inference result.

[0022] Obtain vehicle status data during parking space swapping, determine the intervention variable associated with the input variable in the path intervention matrix, and select the corresponding intervention path based on the status value of the intervention variable;

[0023] The initial inference result is defuzzified according to the intervention rules of the intervention path to obtain a fuzzy correction amount. The fuzzy correction amount is then combined with the position error value to obtain the mechanical error.

[0024] The inference weight network structure includes a tensor structure, the dimensions of which correspond to the fuzziness level of the input variables and the activation intensity of the rules. The inference weight network structure is deformed using the current error distribution map, wherein:

[0025] Based on the distribution trend of mechanical errors recorded in historical transposition operations, the high-incidence error region and the stable error region in the error distribution map are determined;

[0026] For the high-incidence error region, the corresponding weight node is located in the tensor structure, and the corresponding weight node and adjacent node are stretched and deformed according to the entropy value of the error distribution map.

[0027] For the error-stable region, the corresponding weight nodes are located in the tensor structure, and the corresponding weight nodes and adjacent nodes are compressed and deformed according to the entropy value of the error distribution map.

[0028] The deformation is achieved by adjusting the weight coefficients, slope gradients, and activation domain widths of the weight nodes.

[0029] The path intervention matrix is ​​used to adjust the cross-path control relationship between input variables during fuzzy inference, wherein the intervention path includes at least one of the following:

[0030] When there is a geometric mismatch between the vehicle wheelbase and the platform limiting structure, the path intervention matrix controls the vehicle wheelbase variable to intervene in the inference path of vehicle parking error, and adjusts the activation probability, response amplitude and output priority of the corresponding rule.

[0031] When the gravitational potential energy deviation during the transposition operation is greater than the preset potential energy threshold, the path intervention matrix controls the gravitational potential energy deviation variable to intervene in the reasoning path of the gravitational potential energy deviation and adjusts the activation range of the corresponding rule, wherein the activation range adjustment includes compression or expansion.

[0032] When an asymmetric load distribution is detected during the transposition operation, the path intervention matrix controls the load distribution variables to simultaneously intervene in the reasoning path of vehicle parking error and gravitational potential energy deviation, calculates the path cross-linkage rate, and activates the backup fuzzy rule.

[0033] The steps for identifying vehicle parking errors include:

[0034] Acquire vehicle image data and platform status information before the parking space swapping operation is initiated, wherein the vehicle image data includes an image of the vehicle's parking position on the platform, and the platform status information includes the limiting structure, load sensing data, and platform attitude information;

[0035] Based on the vehicle image data, the relative position between the vehicle tires and the limiting structure is detected by an image recognition algorithm, and the offset of the vehicle on the platform is calculated, wherein the offset includes lateral offset and longitudinal offset.

[0036] Based on the load sensing data, the position of the vehicle's center of gravity on the platform is determined and compared with the preset center point of the platform structure to obtain the center of gravity offset value.

[0037] Based on the platform attitude information, calculate the local torque generated by the vehicle on the platform;

[0038] The vehicle parking error is calculated by combining the offset, the center of gravity offset value, and the local torque.

[0039] The preset error distribution map is updated based on the mechanical error, including:

[0040] Obtain the target parking space involved in the current parking space swapping operation, and obtain the error distribution map corresponding to the target parking space;

[0041] In the corresponding error distribution map, the corresponding platform placement coordinates and the mechanical error of the current repositioning operation are recorded to obtain new error samples;

[0042] The newly added error samples are fused with the historical error data in the error distribution map to update the mean error, error fluctuation amplitude, and statistical density index corresponding to the target parking space.

[0043] The distribution density of the newly added error samples in the error distribution map is calculated based on the mean error, the amplitude of error fluctuation, and the statistical density index. The error distribution map is then reconstructed based on the distribution density, wherein the regional reconstruction includes adjusting the aggregation domain range and granularity density of the error points.

[0044] The calculation of the target position for the next transposition includes:

[0045] In the error distribution map corresponding to the target parking space, the location region is determined according to the aggregation domain range and particle size density;

[0046] Based on the location region, the target position for the swap is determined using a particle swarm optimization algorithm.

[0047] An unmanned parking control system, the system comprising:

[0048] The platform control module is used to control the linkage device in the double-layer three-dimensional parking system to perform parking space relocation operation, and after the relocation operation is completed, it collects the actual landing position of the platform, the target position of the relocation, and the status data during the relocation process, and calculates the position error value when the platform reaches the target position of the relocation.

[0049] An error correction module is used to perform fuzzy inference correction on the position error value based on the identified vehicle parking error and gravitational potential energy deviation to obtain the mechanical error. The error correction module includes a fuzzy rule base, an inference weight network structure, and a path intervention matrix.

[0050] The map update and control module is used to update the error distribution map of the corresponding target parking space according to the mechanical error, calculate the target position of the next repositioning based on the error distribution map, and generate repositioning control commands for the next round of repositioning operation.

[0051] Compared with the prior art, the beneficial effects of this application are:

[0052] This application constructs a closed-loop control mechanism based on error feedback, integrating vehicle parking error, gravitational potential energy deviation, and platform state data to dynamically correct platform positioning errors. It also introduces an error distribution map to accumulate and learn the distribution of historical error data, thereby achieving intelligent optimization of the target position during repositioning. Furthermore, through the synergistic effect of fuzzy inference logic, an adaptive deformation inference weighted network structure, and a path intervention matrix, the correction strategy during repositioning can adaptively adjust according to the actual state, effectively avoiding mechanical fatigue accumulation points and reducing the risk of repeated error accumulation. Attached Figure Description

[0053] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0054] Figure 1 This is a schematic diagram illustrating the parking space swapping principle in an embodiment of this application;

[0055] Figure 2 This is a schematic diagram illustrating the error principle of an embodiment of this application;

[0056] Figure 3 This is a flowchart illustrating an unmanned parking control method according to an embodiment of this application;

[0057] Figure 4 This is a schematic diagram of the process for correcting position error values ​​in an embodiment of this application. Detailed Implementation

[0058] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0059] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0060] This application applies to automated multi-level parking systems with multi-platform collaborative positioning structures, precise positioning requirements, and sensitivity to error accumulation. It is particularly suitable for parking systems with two-level structures that require platform repositioning to improve parking efficiency. These parking systems are widely deployed in high-density urban areas, underground spaces, and intelligent transportation facilities. The repeated lifting and precise positioning of the platforms during operation places extremely high demands on the precision of mechanical control.

[0061] Application scenarios include, but are not limited to:

[0062] The upper-level parking spaces are synchronously exchanged with the lower-level parking spaces through a linkage mechanism. The system needs to interface with the platform to perform micron-level precision control during each exchange process.

[0063] The platform support structure is rigid and linked. Affected by temperature drift, inertial loading and long-term mechanical fatigue, its positioning error exhibits a slow accumulation and spatial concentration effect.

[0064] Due to the limited execution accuracy of the platform control system, the actual position after each control command is executed deviates from the theoretical target, forming error hotspots or mechanical dead zones, which affect the stable operation of the system.

[0065] Multiple parking spaces repeatedly using the same platform or path area leads to the accumulation of single-point impacts, and the system needs a decentralized strategy to offset long-term concentrated loads.

[0066] The method provided in this application does not depend on a specific type of robotic arm, parking space assembly structure, or linkage track form, nor does it require the vehicle size, mass, or platform material to remain unchanged. Instead, it proposes a general error dispersion and predictive control logic for automated transfer platforms with common operational bottlenecks.

[0067] The applicable scenarios have at least one of the following common characteristics:

[0068] The controlled target exhibits long-period repetitive motion, with fixed or highly concentrated landing points;

[0069] Control commands cannot fully cover the inertial errors, load disturbances, or asymmetric responses of the drive train;

[0070] Errors themselves are statistical, clustered, and path-dependent, and cannot be eliminated by a single correction;

[0071] The system needs to maintain millimeter-level alignment accuracy, but the platform's operation is limited by mechanical aging or environmental disturbances.

[0072] Taking a two-level multi-level parking system as an example, you can refer to... Figure 1 To understand, Figure 1 This is a schematic diagram of the parking space swapping principle in an embodiment of this application. The double-layer three-dimensional parking system is mainly suitable for enclosed or semi-enclosed parking space scenarios with upper and lower double-layer structures, and is especially suitable for the dynamic allocation of limited parking space resources in urban high-density three-dimensional parking garages.

[0073] Figure 1 A two-level parking system is shown, including upper-level parking spaces, lower-level parking spaces, and a transmission structure between the two.

[0074] Figure 1 The diagram further illustrates a movable platform structure with the upper parking space configured as an unloaded state, while the lower parking space is a conventional movable platform structure for carrying vehicles. The transmission structure is used to drive the upper parking space downwards when it detects that the lower parking space is occupied, so as to perform a swap operation with the lower parking space.

[0075] Understandable Figure 1 This is just an example. In actual applications, both the upper and lower parking spaces can include multiple movable platform structures. The transmission structure is not only used to swap the upper parking spaces when the lower parking spaces are fully occupied, but also to swap the platform corresponding to the upper parking space with the platform in the lower parking space that is occupied when a user needs to use the vehicle in the upper parking space.

[0076] Figure 1The diagram further illustrates that during the repositioning process, the upper parking space moves vertically downwards under the drive of the transmission structure and completes the docking operation at a preset control position. The docking process requires mechanical limiting engagement between the platform edge and the parking space contact surface to ensure the stability of the platform after repositioning.

[0077] Understandable Figure 1 The transmission structure described herein is merely a schematic device to illustrate the linkage action during the parking space relocation process. This application does not limit the specific implementation of the transmission structure during implementation. The linkage structure may take, but is not limited to, the following forms:

[0078] For example, vertical lifting can be achieved by using a chain motor or screw electric cylinder mounted on the garage column, or by using a hydraulic cylinder, pneumatic cylinder, or electric slide rail device as the drive actuator; in terms of transmission path design, various methods such as scissor lifting structure, guide rail lifting structure, or rotating arm lifting structure can be used to achieve displacement control of the upper parking space according to the spatial layout.

[0079] In some alternative embodiments, in compact parking space layouts, precise lifting and limiting docking during the repositioning process can also be achieved through a crank-slider mechanism, a synchronous belt pulley system, or a multi-point servo linkage structure. This application focuses on error identification, correction, and control strategy optimization during the repositioning control process. As a basic actuator, the transmission structure's specific mechanical implementation can be flexibly selected according to the usage scenario, space conditions, and equipment configuration. These are conventional methods that can be understood and replaced by those skilled in the art, and will not be elaborated upon here.

[0080] Taking docking error as an example, please refer to Figure 2 To understand, Figure 2 This is a schematic diagram illustrating the error principle of an embodiment of this application, used to explain the problem of inconsistent docking positions caused by structural errors during platform repositioning.

[0081] Figure 2 The diagram shows a two-level, multi-level parking system with two platform structures, each used to enable vertical movement during the parking space relocation process.

[0082] Figure 2 This further illustrates that during the repositioning operation, the transmission structure precisely aligns with the target parking space under system control commands. The alignment point refers to the designed point in three-dimensional space between the platform structure and the bottom limiting structure of the target parking space, typically including multiple mechanical elements such as horizontal positioning, vertical support, and structural limiting. The alignment relationship depends on the platform's transmission accuracy and structural matching, and is usually achieved through mechanical guide rails, slots, tapered limiting blocks, or edge positioning blocks.

[0083] It is understandable that, due to limitations in the response accuracy of the transmission system, the dynamic changes in the platform's load distribution, and the slight inertial offsets that may occur during the platform's movement, there may be a slight error in the actual landing point when the platform finally reaches the target position. This error is called the docking error, which manifests as a relative misalignment between the platform and the parking space on the docking surface. For example, the platform may not be perfectly aligned with the slot or limiting structure of the parking space, resulting in overlap, misalignment, or incomplete engagement.

[0084] Those skilled in the art will understand that docking errors are cumulative over a long period and concentrated in a specific location. If each repositioning operation is performed with a fixed path and a fixed target point, the error will be repeatedly applied to the same structural node, which can easily lead to local wear, structural deformation or loosening of contact, and reduce the long-term stability of the system.

[0085] Next, with reference to the accompanying drawings, a method for unmanned parking control provided by an embodiment of this application will be described. Figure 3 The method shown is applied to a two-level multi-level parking system, where the two-level structure includes upper-level parking spaces and lower-level parking spaces. The upper-level parking spaces can be switched positions via a linkage platform structure. When all lower-level parking spaces are detected to be occupied, the control logic initiates a switching process, scheduling available spaces in the upper-level parking space and replacing them one-to-one with lower-level parking spaces via a linkage device. The switching operation is completed through a combination of lifting and horizontal movement. The linkage platform, as the actuator, needs precise control of its position throughout the process to avoid structural interference or vehicle misalignment due to accumulated errors. The method includes:

[0086] S1: After the parking space repositioning operation is completed, calculate the position error value when the platform controlled by the linkage device reaches the repositioning target position;

[0087] Specifically, positional error not only reflects the error transmission inherent in the mechanical mechanism itself, but is also affected by speed disturbances, acceleration deviations, and transmission system backlash during the execution process.

[0088] In this embodiment, to improve the accuracy of error identification, the platform's running trajectory is fitted and compensated by fusing state feedback. The state data includes, but is not limited to, acceleration change curves, limit feedback signals, and horizontal displacement trajectories. An error estimation model is constructed by combining the state records of historical position switching operations to establish a dynamic deviation relationship between the expected position and the actual response.

[0089] S2: The position error value is corrected by combining the identified vehicle parking error and gravitational potential energy deviation to obtain the mechanical error;

[0090] Specifically, during the repositioning process, since the actual parking position of the vehicle is not always strictly centered and the center of gravity distribution is also somewhat offset, the load distribution of the platform structure is uneven during operation, which in turn causes deformation of the transmission path or micro-displacement response under off-center load.

[0091] In this embodiment, fuzzy rule inference logic uses vehicle parking error and gravitational potential energy deviation as input variables, which are then input into a fuzzy rule base comprising coarse-grained, medium-grained, and fine-grained rule subsets. Multi-layered fuzzy inference extracts correction suggestions for the position error. An adaptively deformable weighted network structure and path intervention matrix guide the inference path, assigning weights to the output results of each rule for calculation, thereby outputting a precise error correction amount.

[0092] Understandably, the aforementioned correction logic can reduce error uncertainty caused by differences in vehicle conditions and improve the relevance and interpretability of the correction output. In particular, during repeated platform operation, it can suppress systematic deviations caused by structural coupling errors.

[0093] S3: Update the preset error distribution map according to the mechanical error, wherein the error distribution map is used to calculate the target position of the next transposition, and generate a transposition control command according to the target position.

[0094] In this embodiment, the error distribution map serves not only as a tool for recording historical errors but also as a crucial foundation for subsequent target point generation strategies. Before repositioning, the system proactively accesses the map, determines whether the target parking space is located at a high-risk error point based on its error distribution, and generates micro-offsets based on the map to adjust the target point at the millimeter level. This proactive strategy avoids mechanical dead zones or eccentric points, significantly extending structural lifespan, reducing error accumulation, and improving long-term operational stability.

[0095] Before elaborating on the specific technical content of this application, it is necessary to further emphasize the following:

[0096] In a double-layered, multi-level parking structure, the parking space repositioning operation relies heavily on the precise docking of the platform structure. However, due to limitations in the repetitive motion accuracy of the actuators, load disturbances, and external environmental influences, the actual landing point exhibits a negligible micro-error in each operation. Although this error is typically controlled within millimeters, because the platform always docks at the same target point, this repetitive micro-error accumulates at certain spatial coordinates, ultimately leading to localized stress imbalances in the docking structure, resulting in mechanical dead zones or error eccentricities. This structural imbalance directly affects the platform's long-term structural stability, limit assembly accuracy, and vehicle parking safety.

[0097] It is understandable that the unmanned parking control method proposed in this embodiment does not attempt to compress the error range by improving the machining accuracy of the mechanical structure itself or the feedback frequency of the control unit. Instead, it actively identifies and avoids high-error-occurrence areas by dynamically adjusting the logic and feedback learning logic, thereby realizing an error dispersion strategy for the repositioning control logic. After the platform repositioning is completed, the comprehensive positioning error during this operation is identified based on factors such as the vehicle's parking status and transmission path disturbances. This error is then bound to the platform coordinates and written into the error distribution map of the corresponding target parking space, thus forming a spatial error information map with historical traceability.

[0098] Furthermore, the error distribution map relied upon in this application is not constructed based on ideal modeling or preset thresholds, but is dynamically generated through position offset results from actual vehicle operating data. Its core is not to establish a standard position template, but to grasp the density clustering state and dynamic changing trend of errors in spatial coordinates, thereby inferring the load-sensitive area and structural adaptability area of ​​the docking position. In the new round of repositioning control, the control logic generates a finely adjusted repositioning target position based on the spatial clustering characteristics given by the error distribution map, ensuring that the actual landing point of each platform docking is spatially discrete. This discreteness does not cause mechanical misalignment problems; instead, it helps to evenly distribute the structural impact caused by micro-errors to multiple points on the platform structure, slowing down the rate of local structural fatigue.

[0099] Next, we will further elaborate on the part of the method of this application regarding the position error value.

[0100] Understandably, during platform repositioning, the sources of positional deviation are diverse. These include long-term accumulated factors contributing to stability, such as geometric errors in the linkage structure, connection clearances, and assembly tolerances, as well as motion disturbances caused by instantaneous fluctuations in the platform's state, such as inertial impacts during acceleration and deceleration, transmission lag, and the influence of ambient temperature on the material's coefficient of thermal expansion. These disturbances possess a degree of randomness and vary with each operating state. Directly incorporating them into the error distribution map would distort the map, rendering subsequent repositioning strategies meaningless.

[0101] In other words, the purpose of error compensation is not to eliminate the mechanical error itself, but to strip away the non-structural deviations in the motion control process, thereby more accurately extracting the mechanical error related to the transposition structure itself.

[0102] In one example, the specific steps of S1 are as follows:

[0103] S1.1: Obtain the actual parking space repositioning location after the operation is completed, and generate the current physical location of the platform;

[0104] Specifically, the purpose of this step is to obtain the final physical position coordinates of the platform after the platform completes the repositioning operation, so as to compare the error with the theoretical landing position later.

[0105] In some optional embodiments, the platform's current physical position is determined by combining the position encoder output, the boundary sensor trigger status, and the trajectory feedback signal. The position encoder provides real-time feedback on the position scale of the linkage transmission structure, the boundary sensor identifies whether the platform is fully in position, and the trajectory feedback signal provides a time series of position changes during the repositioning process. By fusing these three data points, the platform's precise position coordinates within the current cycle can be obtained. These position coordinates are absolute position data in a two-dimensional or three-dimensional structural coordinate system, and the units can be set to millimeters or micrometers depending on the platform control system.

[0106] S1.2: Generate a transposition reference value based on the transposition target position in the transposition control command;

[0107] In this embodiment, the positioning reference value is based on the target landing position defined in the positioning control command, specifically including the center point of the parking space structure coordinates, the platform motion endpoint control quantity, and the control time window parameter. The target landing position is obtained by modeling the parking space layout and represents the coordinates of a fixed node in the three-dimensional structure. The platform motion endpoint control quantity is given by the platform controller to ensure that the transmission structure completes its displacement action within the command time window. The control time window parameter is used to limit the execution interval of the platform positioning process, ensuring that the target position remains consistent within the control rhythm.

[0108] S1.3: Acquire the platform's status data during the repositioning process, wherein the status data includes platform acceleration, velocity curve, and transmission feedback information;

[0109] Specifically, the purpose of this step is to provide dynamic data support for the subsequent error estimation model, using kinematic parameters during platform operation to evaluate the actual response state of control execution. State changes during the transposition process will directly affect the stability and accuracy of the platform when it reaches the endpoint.

[0110] In this embodiment, the status data acquisition consists of a variety of sensors installed on the linkage device, the limit structure, and the platform body, including but not limited to a triaxial accelerometer, a speed closed-loop control feedback module, a load torque sensor, and a limit position detector.

[0111] Furthermore, acceleration data is used to determine the impact intensity and dynamic stability of the platform during the repositioning process, velocity curves are used to evaluate the smoothness of the platform's running trajectory and whether there is overshoot or pullback, and transmission feedback information reflects the accuracy of the linkage device's execution, such as motor rotation angle and chain tension.

[0112] In some optional embodiments, to avoid misjudgment due to a single data anomaly, the state data is processed by multi-source fusion, and a weighted fusion-based method is used to analyze data from different dimensions simultaneously, thereby improving the reliability of the data.

[0113] S1.4: Based on the state data, the transposition reference value is compensated by a preset error estimation model to obtain the theoretical landing position after state compensation, wherein the error estimation model is constructed based on historical transposition state data;

[0114] Specifically, the goal of this step is to incorporate the actual operating state during the transposition process into the adjustment process of the target position. Through an error estimation model, deviations caused by factors such as control accuracy, load disturbances, or path damping are dynamically compensated, so that the final theoretical landing position used for comparison is closer to the position that the platform should reach in the current state.

[0115] In this embodiment, the error estimation model is a fitted model built based on historical transposition state data. A mapping relationship is established between the state data and the placement deviation through multivariate nonlinear regression. The input variables include the state data. The model can be trained based on historical transposition state data using the least squares method or gradient optimization strategy, which will not be elaborated upon here.

[0116] S1.5: Calculate the difference between the current physical position of the platform and the theoretical landing position after state compensation to obtain the position error value.

[0117] Next, we will further elaborate on the part of the method in this application regarding the correction of position error values.

[0118] Please see Figure 4 , Figure 4 This is a schematic diagram of the process for correcting position error values ​​in an embodiment of this application.

[0119] In this embodiment, the correction process is implemented through fuzzy inference logic. The purpose of the correction process is to further refine the positional error generated during the transposition process into offsets caused by controllable factors, and to identify and correct them through inference logic, so that the final mechanical error is more representative and repeatable. The introduction of fuzzy inference logic effectively solves the problems of nonlinear relationships, strong boundary ambiguity, and unstable fluctuation range of unavoidable error disturbances in actual operation.

[0120] Furthermore, the fuzzy inference logic includes a preset fuzzy rule base, which takes the pre-measured vehicle parking error and gravitational potential energy deviation as input variables. The position error value is corrected by the fuzzy rule base. The vehicle parking error refers to the lateral offset and longitudinal offset of the vehicle relative to the platform's limiting structure when the vehicle is actually parked on the platform, as well as the resulting load center of gravity offset and local torque. It is usually obtained by combining image recognition, limit sensing, and load sensing. The gravitational potential energy deviation refers to the difference in gravitational potential energy caused by the change in platform height during the repositioning of the platform and the vehicle. This deviation reflects the asymmetric deformation or load disturbance that may be brought about by the docking of heavy-duty structures with high and low platforms.

[0121] Furthermore, the fuzzy rule base includes a first subset of rules for coarse-grained error inference, a second subset of rules for medium-grained error inference, and a third subset of rules for fine-grained error inference. It also includes an adaptively deformable inference weight network structure and a path intervention matrix. The adaptively deformable inference weight network structure is used to adjust the weight coefficients of different inference paths in real time according to the error distribution trend in the historical error distribution map, thereby improving the inference sensitivity in high-error-occurrence areas and optimizing the inference accuracy in error-stable areas. The path intervention matrix is ​​used to dynamically regulate the cross-relationship between input variables, solving the problem of coupling effects of some non-independent factors (such as the linkage between load distribution and potential energy deviation) in the inference path. The path intervention rules limit the influence boundary and avoid inference interference.

[0122] In some optional embodiments, the construction of the fuzzy rule base includes the following aspects:

[0123] In the first aspect, the construction of each rule subset is accomplished by combining rule extraction and fuzzy classification. Specifically, based on a large number of historical error samples, the changing trends between input variables and positional offsets are analyzed using a fuzzy clustering algorithm to extract sample clusters with regular characteristics. Then, combined with manually set boundary conditions and weight constraints, these are summarized into "fuzzy IF-THEN rules". Each rule includes the definition of a fuzzy linguistic value, its corresponding membership function, and the range of output quantity control.

[0124] For example, a coarse-grained rule might look like this: "When the vehicle's lateral offset is large and the gravity deviation is medium, then the position correction is relatively large," where "large," "medium," and "relatively large" are fuzzy linguistic terms, modeled using triangular or trapezoidal membership functions. Each rule is labeled with its corresponding domain of application, confidence coefficient, and priority indicator to facilitate rapid activation and fusion during subsequent inference.

[0125] Secondly, the fuzzy rule base also includes intermediate inference logic, which serves as the key logic between rule triggering and output fusion. This intermediate inference logic is implemented through a fuzzy inference engine and includes modules for rule activation detection, fuzzy value combination, conflict rule handling, and output fusion.

[0126] Understandably, to avoid output conflicts between multiple rules, a weighted average method with weight fusion is used, combined with the inference weight network structure to dynamically calculate the activation intensity of each rule.

[0127] In this embodiment, the inference weight network structure is in tensor form and contains three dimensions: input variable level, rule subset number, and error space index.

[0128] Furthermore, the value of each node in the tensor represents the credibility of a rule output in the current context. The node values ​​are automatically adjusted according to the error distribution map to achieve adaptive deformation of the inference intensity for different parking spaces or usage states, thereby enabling fuzzy inference to have historical memory and dynamic adjustment capabilities.

[0129] Thirdly, to address the issues of cross-interference and causal inversion among input variables, a path intervention matrix is ​​configured in the fuzzy rule base to dynamically manage the inference path regulation relationships among input variables. The path intervention matrix has a sparse connection structure and records the intervention trigger conditions, path priorities, and cross-linkage factors among input variables.

[0130] For example, when features such as abnormal vehicle wheelbase, severe center of gravity shift, or platform asymmetry are detected, the path intervention matrix activates specific intervention paths, guiding the inference engine into backup rule channels and appropriately adjusting the amplitude and response domain of the inference output. For instance, when the gravitational potential energy deviation exceeds a preset potential energy threshold, the path intervention matrix activates a gravity deviation enhancement path, increasing the activation probability of relevant rules and suppressing other interfering paths.

[0131] In one example, the specific steps of S2 are as follows:

[0132] S2.1: Input the input variables into the first rule subset, the second rule subset, and the third rule subset respectively to obtain multiple fuzzy correction output sets corresponding to each granularity rule subset;

[0133] Specifically, the core objective of this step is to design multi-layered fuzzy inference channels to address deviations at different scales and in different forms, thereby achieving error correction with greater discriminative and context-aware capabilities. Since vehicle parking errors and gravitational potential energy deviations in real-world scenarios may be influenced by factors such as platform attitude, load distribution, and vehicle shape, a single-granularity rule set cannot fully cover all inference paths and is prone to recognition jitter or correction fluctuations at boundary or critical states. Therefore, fuzzy rules are divided into three categories: coarse-grained rule subsets, medium-grained rule subsets, and fine-grained rule subsets, respectively undertaking the functions of large-amplitude error response, neutral range correction, and fine-grained disturbance adjustment. These three rule sets complement each other, can be executed in parallel, and updated independently, forming a dynamic and collaborative fuzzy response mechanism.

[0134] In this embodiment, the vehicle parking error and gravitational potential energy deviation in the input variables are first fuzzified. Fuzzification can be performed using trigonometric membership functions or Gaussian membership functions, which will not be elaborated upon here.

[0135] Furthermore, vehicle parking errors are fuzzy-leveled based on longitudinal offset, lateral offset, and tire limit relationships, while gravitational potential energy deviation is fuzzified based on the calculation results of center of gravity height difference, platform tilt angle, and load center. Each input variable is mapped to a membership space, where each level corresponds to a fuzzy label.

[0136] Furthermore, the fuzzified combination of input variables is used as a condition term in three sets of fuzzy rule subsets. The rules in each rule subset are expressed in the classic "IF-THEN" form.

[0137] For example, the rule body is shown below:

[0138] Example rule for the first subset of rules: If the longitudinal offset of the vehicle is "high" and the gravitational potential energy deviation is "extremely high", then the position error correction value is "strong backward pull".

[0139] Example rule for the second subset of rules: If the lateral offset of the vehicle is "medium" and the gravitational potential energy deviation is "medium", then the position error correction value is "smooth rightward shift";

[0140] Example rule for the third subset: If the vehicle's longitudinal offset is "low" and the gravitational potential energy deviation is "small", then the position error correction value is "small left correction".

[0141] Understandably, each subset is executed through an independent fuzzy inference engine, employing a min-max inference method for rule activation and fusion of the output fuzzy set. The activation degree of each rule is calculated using the membership function of the fuzzy input condition, ultimately generating a corresponding fuzzy corrected output set, which represents the possible distribution range of the corrected values ​​in the output space.

[0142] S2.2: Based on the input variables, the weight tensor nodes corresponding to each granularity rule subset are called in the inference weight network structure to calculate the activation weight coefficients, and the fuzzy correction output set is weighted to obtain the initial inference result;

[0143] Specifically, since the importance of rule outputs at different granularities varies dynamically under different operating conditions, fixed-weight fusion is prone to inference bias. Therefore, this application introduces a deformable weight tensor structure to dynamically weight the outputs of each rule subset to obtain a fusion result with stronger global adaptability.

[0144] In this embodiment, the weighted tensor nodes construct an index space based on three dimensions: the fuzziness level of the input variable, the regional weights in the historical error distribution map, and the subset response frequency. A set of weighted entropy values ​​is formed by recording the error types and regional distributions in historical permutation operations, which are used to adjust the response slope and activation width of the nodes in the tensor structure. During actual inference, the current input variable will be matched with a corresponding activation node in the tensor network, and the fusion coefficients of each granular subset will be calculated through its output, thereby completing the weighting of the fuzzy correction output set.

[0145] Furthermore, the inference weight network structure includes a tensor structure, the dimension of which corresponds to the fuzziness level of the input variables and the activation intensity of the rules, and the inference weight network structure is deformed by the current error distribution map.

[0146] In this embodiment, to enhance the adaptability of the inference structure, the node weights of the tensor structure can be deformed based on the current error distribution map. The error distribution map consists of mechanical error samples from multiple historical transposition operations, recording the error amplitude and error change trend corresponding to the placement positions of each platform. Based on this map, high-incidence error regions and stable error regions can be identified. Using this as a basis, the weight nodes associated with high-incidence error regions in the tensor structure are stretched, i.e., the response weights are increased, while the nodes associated with stable error regions are compressed, i.e., the weight amplitude is reduced or the activation domain is narrowed. The weight deformation process can be achieved by adjusting the node weights, response slopes, and activation domain widths, dynamically improving the accuracy of fuzzy inference results without changing the original rule structure.

[0147] In some optional embodiments, the process of inference weight network structure deformation includes the following three dimensions:

[0148] In the first dimension, based on the distribution trend of mechanical errors recorded in historical transposition operations, the high-incidence error region and the stable error region in the error distribution map are determined;

[0149] Specifically, the error distribution map uses the actual positioning coordinates of the parking space platform during each repositioning operation as an index to record the mechanical error values ​​and their statistical changes at the corresponding positions, such as the mean, variance, confidence interval, and density of the errors. Through spatial clustering and density analysis of historical error samples, areas with frequent abnormalities or large fluctuations in mechanical error values ​​can be marked on the coordinate map, defined as high-incidence error areas; conversely, areas with small changes in mechanical error values, concentrated distribution, and long-term stability can be identified as error-stable areas.

[0150] In the second dimension, for the high-incidence error region, the corresponding weight node is located in the tensor structure, and the corresponding weight node and adjacent node are stretched and deformed according to the entropy value of the error distribution map.

[0151] In this embodiment, stretching deformation refers to enhancing the weight value, activation function slope, or range of a certain weight node to improve the expressive power of the rule during inference. For example, when a specific combination of fuzzy levels is required for a location in a high-error region, its mapped node position in the tensor structure can be identified. By increasing the weight coefficient of that node or widening its activation interval, the weight of the corresponding fuzzy rule output in that region can be increased. The adjustment of adjacent nodes can employ a Gaussian adjacency weight decay mechanism to ensure that their weights change in a moderate and coordinated manner, avoiding abrupt rule outputs. This enhances the focus and sensitivity of the rule response in abnormal regions, facilitating the precise capture of extreme errors.

[0152] In the third dimension, for the error-stable region, the corresponding weight nodes are located in the tensor structure, and the corresponding weight nodes and adjacent nodes are compressed and deformed according to the entropy value of the error distribution map.

[0153] In this embodiment, compression deformation aims to reduce the sensitivity and activation range of the weighted node within this region, thereby preventing excessive intervention by fuzzy rules in stable regions where no response is required. Weight reduction can be achieved by narrowing the node's activation domain width, lowering the response slope, or directly reducing its weight coefficient. A gradient reduction mechanism can also be used for its adjacent nodes to converge the inference output fluctuations in this region as a whole. During operation, this effectively avoids unnecessary corrections caused by minor perturbations or measurement errors within the stable region, thus reducing redundant computation and the risk of miscompensation, and improving the robustness and stability of the correction logic.

[0154] S2.3: Obtain vehicle status data during the parking space swapping operation, determine the intervention variable associated with the input variable in the path intervention matrix, and select the corresponding intervention path according to the status value of the intervention variable;

[0155] Specifically, abnormal states such as structural mismatch, center of gravity shift, and load asymmetry may occur during vehicle repositioning. Although these variables are not direct input variables, they can affect the contextual logic of error correction. To enhance the responsiveness of the inference process to actual states, some state data is embedded into the inference path through a path intervention matrix during fuzzy inference, enabling proactive adjustment of the inference chain.

[0156] In this embodiment, the path intervention matrix has several preset intervention paths, each path is associated with one or more state variables, and when the value of the intervention variable meets the set conditions, the path switching or additional correction logic is triggered.

[0157] In one example, the intervention path includes at least one of the following:

[0158] When there is a geometric mismatch between the vehicle wheelbase and the platform limiting structure, the path intervention matrix controls the vehicle wheelbase variable to intervene in the inference path of vehicle parking error, and adjusts the activation probability, response amplitude and output priority of the corresponding rule.

[0159] Understandably, by increasing the activation probability of the vehicle wheelbase variable in relevant fuzzy rules, its response to fuzzy outputs is enhanced, and its priority is increased to give it greater decision-making power in rule conflicts. The process is implemented based on a configurable criterion function, ensuring that intervention is only performed when truly necessary, avoiding ineffective intervention.

[0160] When the gravitational potential energy deviation during the transposition operation is greater than the preset potential energy threshold, the path intervention matrix controls the gravitational potential energy deviation variable to intervene in the reasoning path of the gravitational potential energy deviation and adjusts the activation range of the corresponding rule, wherein the activation range adjustment includes compression or expansion.

[0161] When an asymmetric load distribution is detected during the transposition operation, the path intervention matrix controls the load distribution variables to simultaneously intervene in the reasoning path of vehicle parking error and gravitational potential energy deviation, calculates the path cross-linkage rate, and activates the backup fuzzy rule.

[0162] Understandably, the activation factor of the input membership degree corresponding to the variable is simultaneously increased on the relevant fuzzy rules in the two paths, and the path cross-linkage rate is solved by matrix calculation, that is, the number and degree of rules that change synchronously due to the change of the variable's state in the two paths are calculated, thereby identifying the dominance of the variable in the current inference chain; if the linkage rate exceeds the preset threshold, a set of preset backup fuzzy rules is activated to handle scenarios with asymmetric interference and compound bias, so as to improve the coverage of the overall inference model.

[0163] S2.4: Defuzzify the initial inference result according to the intervention rules of the intervention path to obtain a fuzzy correction amount, and synthesize the fuzzy correction amount with the position error value to obtain the mechanical error;

[0164] Specifically, the initial inference result is in the form of a fuzzy set, which needs to be defuzzified and converted into specific numerical values ​​to form the final correction quantity and be used for subsequent control command generation.

[0165] In this embodiment, a correction surface is first constructed based on the final output rules in the intervention path. The activation center and weight boundary of the fuzzy quantity are extracted, and the value of the correction quantity is calculated using a centroid weighting method. During the defuzzification process, the correction magnitude and direction are adjusted in conjunction with the fuzziness level of the input variables, ensuring that the correction quantity maintains the same direction as the original position error in structure, thus preventing overcompensation or cancellation. Subsequently, this correction quantity is synthesized with the position error value calculated in S1 to finally obtain the mechanical error corresponding to the current operation of the platform.

[0166] The synthesis method can be either vector superposition or nonlinear fusion, depending on the differences in platform motion characteristics and error manifestation.

[0167] In one example, the steps for identifying the vehicle parking error include:

[0168] Acquire vehicle image data and platform status information before the parking space swapping operation is initiated, wherein the vehicle image data includes an image of the vehicle's parking position on the platform, and the platform status information includes the limiting structure, load sensing data, and platform attitude information;

[0169] Based on the vehicle image data, the relative position between the vehicle tires and the limiting structure is detected by an image recognition algorithm, and the offset of the vehicle on the platform is calculated, wherein the offset includes lateral offset and longitudinal offset.

[0170] Based on the load sensing data, the position of the vehicle's center of gravity on the platform is determined and compared with the preset center point of the platform structure to obtain the center of gravity offset value.

[0171] Based on the platform attitude information, calculate the local torque generated by the vehicle on the platform;

[0172] The vehicle parking error is calculated by combining the offset, the center of gravity offset value, and the local torque.

[0173] Specifically, identifying vehicle parking errors is crucial for accurately assessing the spatial deviation between the vehicle's actual parking state on the platform and its ideal design state. This provides a data foundation for subsequent position error correction and repositioning control. As a passive load, the vehicle's parking state can change randomly due to driving behavior, the performance of the limiting devices, and the platform's rigidity response. Failure to identify such errors in advance will directly lead to inaccurate platform repositioning references, resulting in cumulative mechanical offsets and affecting the overall accuracy and stability of unmanned parking control.

[0174] In this embodiment, vehicle image data and platform status information are first acquired before the parking space swapping operation is initiated. The image data is collected by a top- or side-mounted visual sensing module, including a sequence of original image frames consisting of the vehicle's four-wheel positioning status and platform boundary features. The platform status information is acquired through built-in attitude sensors, limit detectors, and load sensors, including but not limited to contact status signals of the limit structures, load distribution vectors measured by sensor nodes, and platform surface tilt angle data. The image frames and status data are synchronized using timestamps to ensure the construction of a complete initial parking state description vector.

[0175] Furthermore, for vehicle image data, a deep learning-based contour recognition algorithm is used for processing. Combining convolutional neural networks and image edge enhancement technology, the outer contour of the vehicle tire is separated from the image, and the minimum distance between the tire boundary and the limiting structure is extracted. The lateral and longitudinal offsets of the vehicle in the platform coordinate system are calculated by bidirectional vector ranging.

[0176] In this embodiment, load sensing data is collected through multi-point piezoelectric pressure sensors located beneath the platform. The sensor array can sense the distribution center of gravity of the vehicle load on the platform plane in real time. The center point of the platform structure is a reference point automatically calculated and preset based on the platform's geometric boundary parameters. By inputting the load distribution matrix fed back by the sensors into the bidirectional cosine center of gravity calculation model, the offset of the vehicle's center of gravity relative to the platform center can be determined. The result is defined as the center of gravity offset value, which is used for subsequent mechanical analysis and reasoning correction.

[0177] Furthermore, platform attitude information is acquired through a high-precision IMU module, including pitch angle, roll angle, and local vibration frequency characteristics. Combined with the center of gravity offset value, the local torque caused by asymmetrical load distribution or limit support deviation can be derived. The torque model is constructed using a simplified model of equivalent rigid body eccentric force, ignoring small dynamic disturbances and considering only the torque contribution under steady-state distribution. The direction and magnitude of the torque value will be used to determine whether there is structural asymmetric compression of the platform structure by the vehicle, serving as auxiliary variables in the fuzzy inference logic for rule activation.

[0178] In one example, updating a preset error distribution map based on the mechanical error includes:

[0179] S3.1: Obtain the target parking space involved in the current parking space swapping operation, and obtain the error distribution map corresponding to the target parking space;

[0180] S3.2: In the corresponding error distribution map, record the corresponding platform placement coordinates and the mechanical error of the current repositioning operation to obtain the new error sample;

[0181] Specifically, the purpose of this step is to construct structured new error samples so that they can be incorporated into the error distribution map for incremental updates. The new error samples include the actual landing coordinates of the platform during the transposition operation (obtained through position sensors and image recognition feedback) and the mechanical error values ​​corrected based on fuzzy inference.

[0182] In this embodiment, to ensure the computability of the samples, the error samples are stored in the form of a structure, including spatial ternary coordinates, an error vector, a collection timestamp, and a transposition status label. The error vector contains the numerical magnitude and directional components of the synthesized error, which are used for subsequent distribution evolution modeling.

[0183] S3.3: The newly added error samples are fused with the historical error data in the error distribution map to update the mean error, error fluctuation amplitude and statistical density index corresponding to the target parking space;

[0184] In this embodiment, a time-decay weighted average strategy is adopted for updating the mean error, giving higher weight to recent samples; the error fluctuation amplitude is dynamically calculated based on the standard deviation or the maximum-minimum difference to reflect the uncertainty of the error distribution; the statistical density index is estimated by the number of samples per unit area to reflect the degree of error clustering in the region in history.

[0185] S3.3: Calculate the distribution density of the newly added error samples in the error distribution map based on the mean error, error fluctuation amplitude and statistical density index, and reconstruct the error distribution map based on the distribution density, wherein the region reconstruction includes adjusting the aggregation domain range and granularity density of the error points;

[0186] S3.4: In the error distribution map corresponding to the target parking space, determine the location region based on the aggregation domain range and particle size density;

[0187] Specifically, the purpose of this step is to adjust the division of error regions based on the distribution of newly added error samples in the error distribution map, so as to classify and label stable and abnormal regions subsequently. Distribution density reflects the clustering of errors; high density often indicates the risk of repeatability or structural errors.

[0188] In this embodiment, spatial density clustering is performed on the map grid, and a dynamic region sliding window statistical strategy is used to detect high-error point clusters, thereby adjusting the original grid aggregation domain range accordingly. If local error fluctuations are significant, the granularity density is reduced to achieve a more refined data representation; conversely, adjacent low-density regions are merged to reduce model complexity.

[0189] S3.5: Determine the target position for the swap using a particle swarm optimization algorithm based on the stated position region;

[0190] Specifically, the final step is to use the constructed graph structure as the search space and employ intelligent algorithms to predict and replan the target position for the next repositioning. The particle swarm optimization algorithm combines error aggregation domains, directional trends, and probability weights for low-error regions during the search process to ensure a higher success rate for the selected target points.

[0191] In this embodiment, each location region in the error distribution map is considered as the corresponding position of the search particle. Using the historical error mean as the basis for the fitness function, the particle velocity and position orientation are iteratively updated to ultimately determine the local optimal target point. The fitness evaluation criterion not only considers the minimum position error but also comprehensively assesses the error fluctuation risk and regional stability, thereby outputting the optimal coordinates of the repositioning control target point.

[0192] In one example, this application embodiment provides an unmanned parking control system, the system comprising:

[0193] The platform control module is used to control the linkage device in the double-layer three-dimensional parking system to perform parking space relocation operation, and after the relocation operation is completed, it collects the actual landing position of the platform, the target position of the relocation, and the status data during the relocation process, and calculates the position error value when the platform reaches the target position of the relocation.

[0194] An error correction module is used to perform fuzzy inference correction on the position error value based on the identified vehicle parking error and gravitational potential energy deviation to obtain the mechanical error. The error correction module includes a fuzzy rule base, an inference weight network structure, and a path intervention matrix.

[0195] The map update and control module is used to update the error distribution map of the corresponding target parking space according to the mechanical error, calculate the target position of the next repositioning based on the error distribution map, and generate repositioning control commands for the next round of repositioning operation.

[0196] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for unmanned parking control, applied to a double-deck stereoscopic parking system, characterized in that, The double-layer stereo parking system includes upper-layer parking spaces and lower-layer parking spaces, and when the lower-layer parking spaces are in an occupied state, idle parking spaces in the upper-layer parking spaces are replaced with the lower-layer parking spaces one by one through linkage devices, and the method comprises the following steps: When the parking space replacement operation is completed, a position error value of a platform controlled by the linkage device reaching a replacement target position is calculated; The position error value is corrected in combination with a recognized vehicle parking error and a gravity potential deviation to obtain a mechanical error, wherein the correction comprises taking the vehicle parking error and the gravity potential deviation as input variables of fuzzy reasoning, and inputting the input variables into a fuzzy rule base comprising a coarse-grained rule subset, a medium-grained rule subset and a fine-grained rule subset for multi-scale error correction reasoning, and the correction of the position error value in combination with the recognized vehicle parking error and the gravity potential deviation comprises the following steps: According to a preset fuzzy rule base, the vehicle parking error and the gravity potential deviation are taken as input variables of fuzzy reasoning, and the position error value is corrected through the fuzzy rule base, wherein the fuzzy rule base comprises a first rule subset for coarse-grained error reasoning, a second rule subset for medium-grained error reasoning and a third rule subset for fine-grained error reasoning; The fuzzy rule base further comprises an adaptive morphing reasoning weight network structure and a path intervention matrix, and the correction of the position error value through the fuzzy rule base comprises the following steps: The input variables are input into the first rule subset, the second rule subset and the third rule subset respectively to obtain a plurality of fuzzy correction output sets corresponding to each rule subset; According to the input variables, a weight tensor node corresponding to each rule subset is called in the reasoning weight network structure, an activation weight coefficient is calculated, and the fuzzy correction output sets are weighted and calculated to obtain an initial reasoning result; Vehicle state data in the parking space replacement operation is acquired, an intervention variable associated with the input variables is determined in the path intervention matrix, and a corresponding intervention path is selected according to a state value of the intervention variable; The initial reasoning result is defuzzified according to an intervention rule of the intervention path to obtain a fuzzy correction amount, and the fuzzy correction amount is combined with the position error value to obtain a mechanical error; According to the mechanical error, a preset error distribution map is updated, wherein the error distribution map is used to calculate a replacement target position of the next replacement, and a replacement control instruction is generated according to the replacement target position.

2. The unmanned parking control method according to claim 1, wherein The calculation of the position error value of the platform controlled by the linkage device reaching the replacement target position comprises the following steps: An actual landing position after the completion of the parking space replacement operation is acquired to generate a current physical position of the platform; A replacement reference value is generated according to the replacement target position in the replacement control instruction; State data of the platform in the replacement process is acquired, wherein the state data comprises platform acceleration, speed curve and transmission feedback information; According to the state data, the transposition reference value is compensated by a preset error estimation model to obtain a state-compensated theoretical landing position, wherein the error estimation model is constructed based on historical transposition state data; A difference between the current physical position of the platform and the state-compensated theoretical landing position is calculated to obtain a position error value.

3. The unmanned parking control method according to claim 1, wherein The inference weight network structure includes a tensor structure, dimensions of the tensor structure correspond to fuzzy levels of input variables and rule activation intensity, and the inference weight network structure is deformed by a current error distribution map, wherein: According to a distribution trend of mechanical errors recorded in historical transposition operations, error high-incidence areas and error stable areas in the error distribution map are determined; For the error high-incidence areas, corresponding weight nodes are located in the tensor structure, and the corresponding weight nodes and adjacent nodes are stretched according to the entropy value of the error distribution map; For the error stable areas, corresponding weight nodes are located in the tensor structure, and the corresponding weight nodes and adjacent nodes are compressed according to the entropy value of the error distribution map; Wherein, the deformation is operated by adjusting the weight coefficient value, slope gradient and activation domain width of the weight nodes.

4. The unmanned parking control method according to claim 1, wherein The path intervention matrix is used to adjust the cross-path regulation relationship between input variables in the fuzzy inference process, wherein the intervention path at least includes one of the following: When there is a geometric mismatch state between the vehicle wheelbase and the platform limiting structure, the path intervention matrix controls the vehicle wheelbase variable to intervene in the inference path of the vehicle parking error, adjusts the activation probability, response amplitude and output priority of the corresponding rule; When the gravity potential energy deviation in the transposition operation process is greater than a preset potential energy threshold, the path intervention matrix controls the gravity potential energy deviation variable to intervene in the inference path of the gravity potential energy deviation, and adjusts the activation range of the corresponding rule, wherein the activation range adjustment includes compression or widening; When an asymmetric load distribution state is detected in the transposition operation process, the path intervention matrix controls the load distribution variable to intervene in the inference paths of the vehicle parking error and the gravity potential energy deviation at the same time, calculates the path cross-linkage rate, and activates the standby fuzzy rule.

5. The unmanned parking control method according to claim 1, wherein The vehicle parking error identification step includes: Obtaining vehicle image data and platform state information before the parking space transposition operation is initiated, wherein the vehicle image data includes a parking position image of the vehicle on the platform, and the platform state information includes limiting structure, load sensing data and platform attitude information; According to the vehicle image data, the relative position between the vehicle tire and the limiting structure is detected by an image recognition algorithm, and the offset amount of the vehicle on the platform is calculated, wherein the offset amount includes a lateral offset amount and a longitudinal offset amount; According to the load sensing data, the center of gravity position of the vehicle on the platform is determined and compared with the preset center point of the platform structure to obtain a center of gravity offset value; According to the platform attitude information, the local torque generated by the vehicle on the platform is calculated; The vehicle parking error is calculated in combination with the offset amount, the center of gravity offset value and the local torque.

6. The unmanned parking control method according to claim 1, wherein According to the mechanical error, a preset error distribution atlas is updated, including: Obtaining a target parking space involved in the current transposition operation to obtain an error distribution atlas corresponding to the target parking space; In the corresponding error distribution atlas, record the platform landing coordinates and the mechanical error of the current transposition operation to obtain a new error sample; Fuse the new error sample with historical error data in the error distribution atlas to update the error mean, error fluctuation amplitude and statistical density index corresponding to the target parking space; According to the error mean, error fluctuation amplitude and statistical density index, calculate the distribution density of the new error sample in the error distribution atlas, and according to the distribution density, reconstruct the region of the error distribution atlas, wherein the region reconstruction includes adjusting the aggregation domain range and granularity density of error points.

7. The unmanned parking control method according to claim 6, wherein The calculation of the transposition target position of the next transposition includes: In the error distribution atlas corresponding to the target parking space, determine the position region according to the aggregation domain range and granularity density; According to the position region, determine the transposition target position by particle swarm algorithm.

8. An unmanned parking control system for implementing an unmanned parking control method according to any one of claims 1 to 7, characterized in that, The system includes: A platform control module for controlling the linkage device in the double-layer stereoscopic parking system to perform the parking space transposition operation, and collecting the actual landing position of the platform, the transposition target position and the state data in the transposition process after the transposition operation is completed, and calculating the position error value of the platform when reaching the transposition target position; An error correction module for correcting the position error value based on the identified vehicle parking error and gravitational potential deviation to obtain the mechanical error, wherein the error correction module includes a fuzzy rule base, an inference weight network structure and a path intervention matrix; An atlas updating and control module for updating the error distribution atlas of the corresponding target parking space according to the mechanical error, and calculating the transposition target position of the next transposition based on the error distribution atlas to generate a transposition control instruction for the next round of transposition operation.

Images