Parking lock remote control system based on cloud side cooperation

Abstract

The invention discloses a parking lock remote control system based on cloud edge collaboration, and the system comprises the following modules: a multi-source state preprocessing module which is used for collecting multi-source state data and generating a basic state vector; the four-way potential field construction module is used for constructing a four-way coupling control potential field constructor and generating a four-way control potential field vector; the path deduction module is used for executing the action execution path and generating a reverse anti-fact difference value; the control window generation module is used for generating a reverse anti-fact control window and issuing the reverse anti-fact control window to the edge collaborative gateway; the micro-action chain adjusting module is used for generating a micro-action chain and adjusting the execution amplitude, the execution duration and the step length of each micro-action unit; and the abnormal rollback updating module is used for updating a reverse anti-fact control window strategy and a four-way potential field weight parameter. According to the invention, through combination of cloud anti-fact deduction and edge reversible gradient adjustment, controllable, reversible and high-stability intelligent remote control of the parking lock in a complex environment is realized.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things and ground lock device management technology, and in particular to a remote control system for ground locks based on cloud-edge collaboration. Background Technology

[0002] Existing parking lock control technologies primarily employ local remote control or mobile terminal-based remote control, using simple up / down commands to enable vehicle occupancy or release. These methods largely rely on single sensor data, such as lock status or infrared detection results, for execution decisions. The accompanying control logic typically uses fixed threshold triggers or timed execution, lacking multi-dimensional state perception capabilities and intelligent judgment mechanisms. With the development of smart city construction, cloud platforms are gradually being integrated into parking management systems, supporting user reservations, information synchronization, and authorized control. However, existing cloud control methods are mostly command-driven structures, failing to form a true cloud-edge collaborative mechanism, exhibiting poor stability and recovery capabilities, especially in scenarios involving chain interruptions, sudden occupancy, or equipment malfunctions.

[0003] In complex scenarios, factors such as weak signals when vehicles approach, mechanical resistance of the parking lock, and uncertainties in the execution path often lead to command execution failures, status stagnation, or even false triggering, resulting in severe occupancy conflicts. Especially when an action cannot be rolled back due to insufficient power or communication disconnection after execution, the parking lock becomes uncontrollable, affecting subsequent scheduling and equipment safety. Existing control mechanisms cannot perceive the dynamic changes in the equipment's operating environment in real time, nor have they established a potential field quantification expression or potential field feedback mechanism, lacking a comprehensive judgment method based on reversibility, security, and occupancy status.

[0004] In summary, existing remote control technologies for ground locks have significant shortcomings in multi-state fusion judgment, reversibility assurance after action execution, flexible adaptability of the control link, and cloud-edge collaborative feedback control.

[0005] Therefore, how to provide a remote control system for ground locks based on cloud-edge collaboration is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a cloud-edge collaborative remote control system for parking locks. This invention constructs a four-way coupled control potential field constructor comprising an action potential field layer, a communication potential field layer, an occupancy potential field layer, and a reversibility potential field layer. It introduces a cloud-based reverse counterfactual path deduction mechanism to generate a reverse counterfactual control window with execution boundary constraints, and achieves dynamic adaptive execution of the parking lock's lifting and lowering actions at the edge through a micro-action chain driven by reversible gradients. This invention achieves deep fusion perception of multi-source states, real-time safety verification and reversibility assurance during execution, and possesses advantages such as high action reliability, strong anomaly handling capabilities, and significantly improved stability in weak communication and complex occupancy environments.

[0007] A remote control system for ground locks based on cloud-edge collaboration according to an embodiment of the present invention includes: The multi-source state preprocessing module collects multi-source state data related to the operation of the ground lock, preprocesses the multi-source state data, and generates a basic state vector. The four-way potential field construction module constructs a four-way coupled control potential field builder consisting of an action potential field layer, a communication potential field layer, an occupancy potential field layer, and a reversibility potential field layer. It performs potential field tensor quantization reconstruction processing on the basic state vector to obtain the four-way control potential field vector. The path deduction module uploads the four-way control potential field vector to the cloud control platform, and performs action execution path deduction, action execution rollback path deduction, and non-execution path deduction respectively, generating reverse counterfactual difference; The control window generation module generates a reverse counterfactual control window in the cloud based on the reverse counterfactual difference and the four-way control potential field vector. It performs reversibility constraint domain shaping processing to determine the action execution boundary and sends the shaped reverse counterfactual control window to the edge collaborative gateway. The micro-action chain adjustment module generates a micro-action chain based on the reverse counterfactual control window at the edge collaborative gateway. It decomposes the lifting and lowering action of the ground lock into multiple micro-action units. Before and after the execution of each micro-action unit, it re-acquires the four-way control potential field vector, calculates the invertibility gradient, and adjusts the execution amplitude, execution duration, and step size of the micro-action unit. In the abnormal rollback update module, if the reversible potential field quantity in the four-way control potential field vector is lower than the threshold or communication instability, abnormal resistance, or occupancy change occurs during the execution of the micro-action chain, the micro-action chain is terminated and the rollback chain reset lock is triggered. The execution trajectory and potential field disturbance data are reported, and the reverse counterfactual control window strategy and four-way potential field weight parameters are updated.

[0008] Optionally, the multi-source status data includes action status data, communication channel status data, occupancy status data, and energy status data.

[0009] Optionally, the preprocessing of multi-source state data to generate a basic state vector includes: performing noise reduction, formatting, time alignment, and normalization on the multi-source state data, and combining the processed data to form a basic state vector, which includes an action sub-vector, a communication sub-vector, an occupancy sub-vector, and an energy sub-vector.

[0010] Optionally, the four-directional potential field construction module includes: Construct a four-way coupled control potential field builder, initialize the action potential field layer, communication potential field layer, occupancy potential field layer and reversible potential field layer, and set a unified time base and sampling step size; Based on the four-way coupled control potential field builder, hierarchical potential field processing is performed on the basic state vector: The action potential field layer performs action differential spectrum extraction and drag gradient encoding on the action sub-vector, and outputs the action potential field sub-vector. The communication potential field layer performs joint stability coding of link delay and jitter and packet loss mutation masking on the communication subvector, and outputs the communication potential field subvector. The occupancy potential field layer performs spatiotemporal occupancy encapsulation of the lock-in state and the approach trajectory on the occupancy subvector, and outputs the occupancy potential field subvector. The reversible potential field layer performs energy margin, mechanical rebound and link maintenance capability reversibility margin estimation and back-off condition boundary calculation on the energy subvector, and outputs the reversible potential field subvector. Perform temporal alignment, scale registration, and energy level mapping on each potential field subvector to form four types of potential field subvector sets, resulting in the registered four types of potential field subvector sets. Potential field tensor quantization reconstruction is performed on the registered four types of potential field subvector sets: Pairwise interactive coupling is expanded to capture the direct coupling relationship between layers; ternary interactive coupling is aggregated to capture the joint effect across three layers; and quaternary interactive coupling is summarized to form an overall coupling representation. During the interactive unfolding process, redundant interactive components are eliminated, and constraint domain shaping is performed in the summarization stage to satisfy the boundary conditions that do not reduce monotonicity and invertibility, thus obtaining the four-way coupled intermediate representation. The four-way coupled intermediate representation is subjected to aggregation mapping processing. The four types of potential field interaction results are reorganized in a fixed dimension, the components are merged and the hierarchical projection is performed to generate a four-way control potential field vector. The four-way control potential field vector includes the action reachability potential field, the communication stability potential field, the occupancy security potential field, and the reversibility potential field.

[0011] Optionally, the path deduction module includes: Receive the four-way control potential field vector and establish the four-way potential field time series according to the timestamp, and determine the unified deduction time window, time step and boundary conditions; Action execution path deduction: Apply action load deformation scheduling step by step within a unified time window to dynamically adjust the action intensity and rhythm according to the reversible potential field quantity, enable communication occupancy pre-quota to limit the peak link occupancy and insert link stability checkpoints, and use occupancy boundary look-ahead encapsulation to extrapolate the time and suppress over-boundary of the occupancy potential field safety boundary to generate action execution path trajectory. Rollback path deduction after action execution: Based on the reversible potential field quantity, a rollback envelope constraint is generated. After the action completion node, the rollback envelope constraint is applied step by step to advance the rollback process. Simultaneously, the reversible energy ledger records energy consumption and recoverable margin. Communication is deployed to maintain sentinel verification to determine the link availability of key rollback nodes and generate the rollback path trajectory after action execution. Execution and non-execution path deduction: Within a unified time window, without applying any action, the four-way potential field time series is updated step by step according to the natural evolution process, including the evolution of communication fluctuations, occupancy changes and energy self-discharge, to obtain the non-execution path trajectory; Within a unified time window, the three trajectories are aligned in time and their lengths are consistent. The corresponding component differences between the rollback path trajectory after the action is executed and the non-executed path trajectory are compared step by step in chronological order to generate a difference sequence. The reverse counterfactual difference is obtained by aggregating the results through time-weighted accumulation, abnormal component removal, and interval pruning.

[0012] Optionally, the control window generation module includes: Receive the four-way control potential field vector and the inverse counterfactual difference in the cloud, and establish a unified time window, state threshold set and execution boundary index; Generate an initial reverse counterfactual control window, and filter time periods in chronological order that meet the following conditions: reversibility potential field quantity is not lower than the safety threshold, communication stability is not lower than the link threshold, occupancy security is in the safe zone, and action reachability meets the execution conditions. Record the time periods that pass the filter as candidate execution areas, and determine the allowed action type, maximum micro-action chain length, potential field disturbance upper limit value, and rollback trigger condition for each candidate execution area. Perform reversibility constraint domain shaping on the initial reverse counterfactual control window, including: Perform boundary compression on segments exhibiting a downward trend in reversibility; Insert link protection gaps in communication fluctuation sections; A safety buffer zone shall be set up for the critical section of the occupation; Perform adjacent merging and hole filling on discrete candidate execution regions; Slice the cross-boundary segments and remove sub-regions that do not satisfy monotonicity to obtain a shaped execution boundary set; The consistency and validity of the shaped execution boundary set are checked, and the final reverse counterfactual control window is generated and sent to the edge collaboration gateway.

[0013] Optionally, the micro-motion chain adjustment module includes: The edge collaborative gateway receives the reverse counterfactual control window, parses the time window, reversibility potential field threshold, communication stability threshold, occupancy security threshold, potential field disturbance upper limit, maximum chain length and rollback trigger condition, generates a micro-action chain initialization parameter set, and selects a micro-action template that matches the window constraints from the pulse packet template library. Based on the micro-action template, the lifting action of the ground lock is decomposed into multiple micro-action units. The initial execution sequence is established using a three-element description of amplitude, duration, and step distance. Potential field disturbance budget and local rollback sub-chain are allocated to each micro-action unit. Before each micro-action unit is executed, a four-way control potential field vector is obtained, and a double sampling closed loop is executed to obtain the change in potential field before and after execution. Based on the change in potential field before and after execution and the window threshold group, an invertibility gradient metric is constructed to shape the execution amplitude, execution duration and step size of the current micro-action unit. The micro-action unit is executed and the four-way control potential field vector is reacquired after completion. The execution amplitude, execution time and step size of the next micro-action unit are adjusted in linkage according to the continuous change of the invertibility gradient metric. When the communication stability or occupancy security is close to the threshold, the protection gap and check point are automatically inserted. During the execution of the micro-action chain, if the reversibility potential field is lower than the threshold or any of the conditions of communication stability, occupancy security, or potential field disturbance exceeds the limit, the current micro-action unit is immediately interrupted, the local rollback sub-chain is reset, and the micro-action execution parameters and potential field changes are recorded.

[0014] Optionally, the abnormal rollback update module includes: During the execution of the micro-action chain, the four-way control potential field vector is read in real time and the execution trajectory is cached. The read reversible potential field quantity is compared with the set threshold, and at the same time, it is determined whether the communication stability potential field quantity, the occupancy security potential field quantity, and the potential field disturbance quantity have exceeded the limit. When the reversibility potential field is below the threshold or any of the conditions of communication stability, occupancy security, or potential field disturbance exceeds the limit, an interrupt command is immediately issued to terminate the current micro-action unit. According to the rollback trigger conditions in the reverse counterfactual control window, the corresponding rollback action chain is invoked to execute the rollback micro-action unit until the ground lock is restored to a safe state. The system collects execution trajectory and potential field disturbance data, generates execution records, and uploads them to the cloud. These records are used to refresh the reverse counterfactual control window generation strategy and the four-way potential field weight parameters.

[0015] The beneficial effects of this invention are: This invention constructs a four-way coupled control potential field, unifying the expression of action state, communication link, occupancy environment, and reversibility characteristics into a potential field quantity. This enables the parking lock to possess comprehensive operational situation awareness capabilities before execution, overcoming the limitations of traditional methods relying on single sensor information or simple threshold judgments. The four-way control potential field vector obtained based on potential field tensor reconstruction can accurately characterize the executability and recoverability of the parking lock in complex environments, providing a reliable data foundation for subsequent cloud-based simulations and control window generation, fundamentally improving the accuracy and security of execution decisions.

[0016] This invention employs a reverse counterfactual path deduction mechanism to synchronously construct action execution paths, action rollback paths, and non-execution paths in the cloud. It generates a reverse counterfactual control window through interpolation, enabling intelligent shaping of the prediction and control boundaries of pre-action risks. It can predict whether the action can be recovered after execution, whether it will trigger link or occupy sudden changes, etc., before the execution command is issued. This avoids common problems in existing technologies such as inability to roll back, high resistance jamming, and link breakage and loss of control, making the execution of remote commands more deterministic and fault-tolerant.

[0017] At the edge, this invention introduces a reversible gradient-driven micro-action chain execution method, decomposing the traditional overall lifting action into multiple low-disturbance micro-action units, and dynamically adjusting the potential field before and after each micro-action execution. By adjusting the execution amplitude, duration, and step size in real time, the ground lock possesses adaptive adjustment capabilities, effectively coping with uncertainties such as communication fluctuations, resistance changes, and sudden occupancy. When the reversible potential field decreases or an abnormal state occurs, the system can immediately terminate execution and trigger a rollback chain reset, improving the operational stability and security of the ground lock in complex real-world scenarios. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of a cloud-edge collaborative remote control system for ground locks proposed in this invention; Figure 2 This is a schematic diagram of the process of a cloud-edge collaborative remote control system for ground locks proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 and Figure 2A cloud-edge collaborative remote control system for ground locks includes: The multi-source state preprocessing module collects multi-source state data related to the operation of the ground lock, preprocesses the multi-source state data, and generates a basic state vector. The four-way potential field construction module constructs a four-way coupled control potential field builder consisting of an action potential field layer, a communication potential field layer, an occupancy potential field layer, and a reversibility potential field layer. It performs potential field tensor quantization reconstruction processing on the basic state vector to obtain the four-way control potential field vector. The path deduction module uploads the four-way control potential field vector to the cloud control platform, and performs action execution path deduction, action execution rollback path deduction, and non-execution path deduction respectively, generating reverse counterfactual difference; The control window generation module generates a reverse counterfactual control window in the cloud based on the reverse counterfactual difference and the four-way control potential field vector. It performs reversibility constraint domain shaping processing to determine the action execution boundary and sends the shaped reverse counterfactual control window to the edge collaborative gateway. The micro-action chain adjustment module generates a micro-action chain based on the reverse counterfactual control window at the edge collaborative gateway. It decomposes the lifting and lowering action of the ground lock into multiple micro-action units. Before and after the execution of each micro-action unit, it re-acquires the four-way control potential field vector, calculates the invertibility gradient, and adjusts the execution amplitude, execution duration, and step size of the micro-action unit. In the abnormal rollback update module, if the reversible potential field quantity in the four-way control potential field vector is lower than the threshold or communication instability, abnormal resistance, or occupancy change occurs during the execution of the micro-action chain, the micro-action chain is terminated and the rollback chain reset lock is triggered. The execution trajectory and potential field disturbance data are reported, and the reverse counterfactual control window strategy and four-way potential field weight parameters are updated.

[0021] In this embodiment, the multi-source status data includes action status data, communication channel status data, occupancy status data, and energy status data.

[0022] In this embodiment, the preprocessing of multi-source state data to generate a basic state vector includes: performing noise reduction, formatting, timing alignment, and normalization on the multi-source state data, and combining the processed data to form a basic state vector, which includes an action sub-vector, a communication sub-vector, an occupancy sub-vector, and an energy sub-vector.

[0023] In this embodiment, the four-directional potential field construction module includes: Construct a four-way coupled control potential field builder, initialize the action potential field layer, communication potential field layer, occupancy potential field layer and reversible potential field layer, and set a unified time base and sampling step size; Based on the four-way coupled control potential field builder, hierarchical potential field processing is performed on the basic state vector: The action potential field layer performs action differential spectrum extraction and drag gradient encoding on the action subvector, outputting the action potential field subvector, specifically: Within a fixed time window, the angle, speed, and motor current of the action sub-vector are discretely sampled, and the first and second order changes are calculated and normalized to form a differential spectrum feature describing the instantaneous dynamic changes. By combining real-time mechanical resistance, load current and ground inclination, the gradient of the resistance change curve over time is calculated, and the gradient value is segmented and quantized into three levels of codes: safety, warning and danger, and mapped into vector components. The differential spectrum features are concatenated, normalized, and compressed with the resistance gradient encoding to output a motion potential field sub-vector that represents the action accessibility and resistance trend. The communication potential field layer performs joint stability coding of link delay and jitter and packet loss mutation masking on the communication subvector, and outputs the communication potential field subvector, specifically: Record the round-trip delay sequence and instantaneous jitter amplitude of the communication subvector within a continuous sampling window, calculate the weighted moving average and variance of the two, and generate a normalized score vector representing the link stability based on the statistical interval mapping. A sliding median filter is applied to the real-time packet loss rate sequence to detect abrupt changes that exceed the background noise threshold. A dynamic masking mask is used to reduce the weight of the impact of abrupt changes on the stability score, thus avoiding misjudging short-term spikes as long-term instability. The joint stability score is concatenated with the masked packet loss features, and after normalization and compression, the communication potential field subvector is output. The occupancy potential field layer performs spatiotemporal occupancy encapsulation of the lock-in state and approach trajectory on the occupancy subvector, and outputs the occupancy potential field subvector, specifically: The pressure sensor is used to sample the force changes of the lock at high speed, detect the continuous pressure time and pressure amplitude, compare with the threshold library to generate three-level state labels of no pressure - light pressure - heavy pressure, and convert them into the pressure component of the occupied sub-vector; Ultrasonic waves are used to extract the vehicle's leading edge trajectory coordinates within a fixed time window, and the trajectory velocity, acceleration and orientation angle are calculated and mapped into a two-dimensional spatiotemporal feature matrix of distance-velocity. The matrix is ​​then encoded according to three modes: close-range rapid approach, slow approach and far away. After aligning the pressure component and the proximity trajectory encoding on the time axis, spatiotemporal encapsulation is performed, and weighted fusion is used to generate the occupancy safety score vector. After normalization and compression, the occupancy potential field sub-vector is output. The reversible potential field layer performs energy margin, mechanical rebound, and link maintenance capability reversibility margin estimation, as well as back-off condition boundary calculation, on the energy subvector, outputting a reversible potential field subvector, specifically: Monitor the battery voltage curve and instantaneous current, evaluate the remaining available energy by referring to the historical discharge model, calculate the percentage of energy reserve based on the power consumption required to complete one full rise and fall and one rollback, and map the results into three levels of energy safety, warning and undervoltage components. Record the motor rebound angle and resistance recovery curve during the execution gap of the micro-motion chain, combine the rebound amplitude and resistance attenuation rate to evaluate the mechanical elastic recovery capability, and generate easy-rebound-normal-difficult-rebound graded codes to reflect the level of mechanical self-reset. By combining the link stability score and the channel hold-up time during the rollback phase of the remaining energy prediction, the minimum energy and minimum stability threshold required to complete the rollback under the worst link condition are calculated to form the rollback condition boundary. The energy component, bounce coding and rollback condition boundary are fused by normalization to output the invertible potential field sub-vector. Each potential field subvector undergoes temporal alignment, scale registration, and energy level mapping to form four types of potential field subvector sets, resulting in four registered potential field subvector sets. Specifically, the temporal alignment, scale registration, and energy level mapping are performed on each potential field subvector as follows: The four types of potential field sub-vectors are resampled using a unified time base of the edge gateway, and missing segments are interpolated and redundant segments are pruned so that each sub-vector gradually corresponds on the same time axis. After alignment, each subvector is scaled according to its physical quantity range. For angle, current, voltage, RSSI, and distance, interval linear compression or logarithmic compression is applied respectively to make all components fall into a consistent metric space. Based on the dynamic safety threshold, link stability classification, occupancy risk level and reversibility margin level, energy level mapping is performed on the registered components to map continuous values ​​into multi-level energy level symbols, ultimately forming four types of potential field sub-vector sets that are time-aligned, scale-consistent and energy level unified.

[0024] Potential field tensor quantization reconstruction is performed on the registered four types of potential field subvector sets: Pairwise interactive coupling is expanded to capture the direct coupling relationship between layers; ternary interactive coupling is aggregated to capture the joint effect across three layers; and quaternary interactive coupling is summarized to form an overall coupling representation. During the interactive unfolding process, redundant interactive components are eliminated. In the summarization phase, constraint domain shaping is performed to satisfy the boundary conditions that do not reduce monotonicity and reversibility, resulting in a four-way coupled intermediate representation. The constraint domain shaping is specifically performed as follows: Boundary contraction is performed on over-limit components: components that exceed the safety limits for action, communication, occupancy, or reversibility are truncated to the allowable range based on historical safety thresholds; In the section where the reversible potential field may drop, a risk buffer zone is inserted: the adjacent components are interpolated and stretched and the rollback margin is increased so that the local sudden drop is smoothly absorbed. Gradient smoothing is applied to the component sequences after contraction and buffering: the change curves that eliminate sudden increases and decreases and maintain monotonically non-increasing risk are eliminated, so that the four types of potential field characteristics continuously and smoothly satisfy the executable and reversible boundary conditions in the time domain. The aggregation mapping process is performed on the four-way coupled intermediate representation to generate a four-way control potential field vector by reorganizing the four types of potential field interaction results according to a fixed dimension, merging the components, and projecting them in the order of hierarchical level. The four-way control potential field vector includes the action reachability potential field, the communication stability potential field, the occupancy security potential field, and the reversibility potential field. Specifically, the aggregation mapping process on the four-way coupled intermediate representation is as follows: The intermediate representation of four-way coupling is reorganized according to the fixed order of four dimensions: action-communication-occupancy-reversibility. The high-order interaction information is split into four equal-length feature segments. Each segment retains only the components directly related to the target potential field quantity and removes duplicate or null components. Subsequently, component merging is performed on each feature segment. First, the difference in dimensions is eliminated by normalization of the maximum-minimum interval. Then, weights are assigned based on the historical execution success rate and safety factor. The same type of components are weighted and summed and the average deviation is calculated to obtain four initial potential field scores: action, communication, occupation, and reversibility. Finally, the four initial potential field scores are projected into a five-level security level range within the 0-1 interval. Scores below the warning threshold are improved by exponential buffering, and scores above the upper limit of the security threshold are compressed. The final output is a four-way control potential field vector consisting of the action reachability potential field, the communication stability potential field, the occupancy security potential field, and the reversibility potential field.

[0025] In this embodiment, the path deduction module includes: Receive the four-way control potential field vector and establish the four-way potential field time series according to the timestamp, and determine the unified deduction time window, time step and boundary conditions; Action execution path derivation: Within a unified time window, action load deformation scheduling is applied step-by-step to dynamically adjust the action intensity and rhythm according to the reversible potential field. Communication occupancy pre-quota limits the peak link occupancy and links stability checkpoints are inserted. Occupancy boundary look-ahead encapsulation is used to extrapolate the safety boundary of the occupancy potential field and suppress out-of-bounds errors, generating the action execution path trajectory, where: Action load deformation scheduling refers to performing segmented scaling and rhythm adjustment on the action load curve based on the real-time changes of the reversible potential field within the simulation window. When the reversible potential field decreases, the motor pulse amplitude is automatically reduced, the duration is shortened, or a delay is inserted; conversely, the load is restored or increased. The pre-allocation of communication occupancy quota refers to the pre-allocation of communication link occupancy quota for each time step. The link stability potential field is used as the upper limit to limit the number of packets sent and the total number of bytes during peak periods. If the real-time link load is close to the quota limit, unnecessary data interaction is suspended and a link stability checkpoint is inserted. Only after the check passes can the simulation continue. Occupancy boundary look-ahead encapsulation refers to short-term prediction of vehicle approach trajectory and locking trend. The prediction result is used to generate a time extrapolation boundary with the current occupancy safety threshold. When the extrapolation boundary is about to intersect with the threshold, a safety buffer is automatically inserted into the extrapolation trajectory and the action intensity is reduced. If the extrapolation boundary exceeds the limit, the action extrapolation is paused until the occupancy risk is reduced. Post-action rollback path deduction: Based on the reversible potential field, a rollback envelope constraint is generated. After the action completion node, the rollback envelope constraint is applied step-by-step to advance the rollback process. Simultaneously, the reversible energy ledger records energy consumption and recoverable margin. Communication is deployed to maintain sentinel verification and determine the link availability of key rollback nodes. The rollback path trajectory after action execution is generated. Specifically, the generation of the rollback envelope constraint is as follows: Extract the reversible potential field at the action completion node, and combine it with the remaining battery capacity, mechanical resistance trend and link stability threshold to determine the rollback start safety zone and target stop position, forming the rollback displacement-time initial boundary curve; Based on the initial boundary curve, the maximum allowable energy consumption, maximum allowable resistance, and minimum communication bandwidth of the entire rollback process are calculated. A three-dimensional limit surface of energy, resistance, and communication is constructed. Boundary contraction is performed on the over-limit section to obtain the multi-dimensional safety domain of the rollback envelope. Expand the multidimensional safety domain by time step to generate a rollback envelope constraint sequence that includes upper limits for displacement progress, energy consumption, resistance, and communication. Execution / Non-execution path deduction: Within a unified time window, without applying any action, the four-way potential field time series is updated step-by-step according to the natural evolution process, including the evolution processing of communication fluctuations, occupancy changes, and energy self-discharge, to obtain the non-execution path trajectory. Specifically, updating the four-way potential field time series step-by-step according to the natural evolution process involves: At each time step, the current RSSI and jitter amplitude are first read, and the increase or decrease difference is calculated compared with the previous time step to directly correct the communication stability potential field. Based on the vehicle distance, speed and locking status detected in real time by ultrasound or radar, the occupied safety potential field quantity is recursively deduced according to the principle that a shorter distance means an increased risk and a slower speed means a decreased risk. Read the battery voltage, current and resistance recovery value, subtract the reversible potential field quantity according to the self-discharge quantity per unit time, and simultaneously adjust the action reachability potential field quantity, and sequentially complete the gradual evolution and update of the four-way control potential field vector without action intervention. Within a unified time window, the three trajectories are aligned in time and their lengths are consistent. The corresponding component differences between the rollback path trajectory after the action is executed and the non-executed path trajectory are compared step by step in chronological order to generate a difference sequence. The reverse counterfactual difference is obtained by aggregating the results through time-weighted accumulation, abnormal component removal, and interval pruning.

[0026] In this embodiment, the control window generation module includes: Receive the four-way control potential field vector and the inverse counterfactual difference in the cloud, and establish a unified time window, state threshold set and execution boundary index; Generate an initial reverse counterfactual control window, and filter time periods in chronological order that meet the following conditions: reversibility potential field quantity is not lower than the safety threshold, communication stability is not lower than the link threshold, occupancy security is in the safe zone, and action reachability meets the execution conditions. Record the time periods that pass the filter as candidate execution areas, and determine the allowed action type, maximum micro-action chain length, potential field disturbance upper limit value, and rollback trigger condition for each candidate execution area. Perform reversibility constraint domain shaping on the initial reverse counterfactual control window, including: Perform boundary compression on segments exhibiting a downward trend in reversibility; Insert link protection gaps in communication fluctuation sections; A safety buffer zone shall be set up for the critical section of the occupation; Perform adjacent merging and hole filling on discrete candidate execution regions, specifically as follows: All candidate execution zones are arranged in chronological order, and zones with adjacent time intervals less than the shortest start interval of a single micro-action are directly merged into a continuous zone. For gaps within the merged section, if the gap duration is shorter than the minimum protection gap and the potential field quantities on both sides meet the window threshold, then the gaps are filled inward in a linear expansion manner to form a complete executable region. For isolated narrow holes or fragmented areas that still exist after filling, if the length is less than the minimum execution time of the micro-action chain, the entire area is absorbed into the adjacent segment; otherwise, it is discarded, thus completing the merging of adjacent areas and filling of holes. Slice the cross-boundary segments and remove sub-regions that do not satisfy monotonicity to obtain a shaped execution boundary set; The shaped execution boundary set undergoes consistency and validity checks to generate a final reverse counterfactual control window, which is then distributed to the edge collaboration gateway. Specifically, the consistency and validity checks on the shaped execution boundary set are performed as follows: For each boundary segment, the reversibility potential field threshold, communication stability threshold, and occupancy security threshold are checked simultaneously. Any segment with any index below the lower limit is removed from the set. Perform a continuity check on the remaining segments, merge segments with adjacent intervals shorter than the shortest cycle of a single micro-action, and delete segments whose length is shorter than the shortest execution time of the micro-action chain. Perform rollback reachability checks on a segment-by-segment basis to confirm that the rollback can be completed within the maximum chain length without exceeding the potential field perturbation limit. Only segments that pass the checks are retained to generate the final control window.

[0027] In this embodiment, the micro-motion chain adjustment module includes: The edge collaborative gateway receives the reverse counterfactual control window, parses the time window, reversibility potential field threshold, communication stability threshold, occupancy safety threshold, potential field disturbance upper limit, maximum chain length and rollback trigger condition, generates a micro-action chain initialization parameter set, selects a micro-action template that matches the window constraints from the pulse packet template library, where the pulse packet template library refers to a set of standardized micro-action templates that are pre-set locally on the edge collaborative gateway and can be dynamically updated by the cloud. Each template encapsulates the motor pulse amplitude, execution duration, step increment / decrement rules, real-time sampling frequency, safety threshold and corresponding rollback sub-chain identifier parameters; Based on the micro-action template, the lifting action of the ground lock is decomposed into multiple micro-action units. The initial execution sequence is established using a three-element description of amplitude, duration, and step distance. Potential field disturbance budget and local rollback sub-chain are allocated to each micro-action unit. Before each micro-action unit is executed, a four-way control potential field vector is obtained, and a double-sampling closed loop is executed to obtain the change in potential field before and after execution. Based on the change in potential field before and after execution and the window threshold group, a reversibility gradient metric is constructed to shape the execution amplitude, execution duration, and step size of the current micro-action unit. The window threshold group is a set of constraint limits issued by the cloud when generating the reverse counterfactual control window. It includes four types of values: the minimum safe value of the reversibility potential field, the minimum available value of the communication stability potential field, the minimum allowable value of the occupancy security potential field, and the maximum allowable amplitude of the potential field disturbance. These values ​​are used to compare the change in potential field during the execution of the micro-action chain and determine whether the current micro-action unit needs to be adjusted or terminated. The micro-action unit is executed and the four-way control potential field vector is reacquired after completion. The execution amplitude, execution time and step size of the next micro-action unit are adjusted in linkage according to the continuous change of the invertibility gradient metric. When the communication stability or occupancy security is close to the threshold, the protection gap and check point are automatically inserted. During the execution of the micro-action chain, if the reversibility potential field is lower than the threshold or any of the conditions of communication stability, occupancy security, or potential field disturbance exceeds the limit, the current micro-action unit is immediately interrupted, the local rollback sub-chain is reset, and the micro-action execution parameters and potential field changes are recorded.

[0028] In this embodiment, the abnormal rollback update module includes: During the execution of the micro-action chain, the four-way control potential field vector is read in real time and the execution trajectory is cached. The read reversible potential field quantity is compared with the set threshold, and at the same time, it is determined whether the communication stability potential field quantity, the occupancy security potential field quantity, and the potential field disturbance quantity have exceeded the limit. When the reversibility potential field is below the threshold or any of the conditions of communication stability, occupancy security, or potential field disturbance exceeds the limit, an interrupt command is immediately issued to terminate the current micro-action unit. According to the rollback trigger conditions in the reverse counterfactual control window, the corresponding rollback action chain is invoked to execute the rollback micro-action unit until the ground lock is restored to a safe state. The system collects execution trajectory and potential field disturbance data, generates execution records, and uploads them to the cloud. These records are used to refresh the reverse counterfactual control window generation strategy and the four-way potential field weight parameters.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to a densely populated residential area. This area has roads less than 5.2 meters wide, high vehicle density, and insufficient base station coverage. Problems frequently arise around the parking locks, including unstable communication signals, sensor interference from approaching vehicles, and increased resistance due to rain. These issues cause traditional remote control methods for parking locks to frequently experience operation failures, jamming, and misjudgments of occupancy. Before the pilot test, the parking locks performed approximately 260 actions per day, with nearly 5% of these actions timing out, failing, or failing to roll back. This severely impacted parking efficiency and resulted in significant manual maintenance pressure.

[0030] In the pilot area, 20 parking spaces were selected as test subjects. Each parking lock terminal was equipped with a control unit containing a multi-source state acquisition device, a four-way potential field construction module, a micro-action chain execution unit, and an edge collaborative gateway. During operation, the parking lock continuously collects data such as action status, motor current, mechanical resistance, battery remaining power, communication link quality, vehicle approach trajectory, and occupancy status. After preprocessing, a basic state vector is generated. The four-way coupled control potential field constructor of this invention runs at the edge, inputting the basic state vector to the action potential field layer, communication potential field layer, occupancy potential field layer, and reversibility potential field layer. It then generates a four-way control potential field vector through potential field tensor reconstruction processing, enabling the parking lock to possess complete capabilities for execution reachability, communication stability, occupancy risk security, and action reversibility assessment.

[0031] For example, around 17:20 on May 14, 2025, a brief period of communication congestion occurred in the pilot area, with RSSI values ​​fluctuating between -86dBm and -92dBm. At this time, a user initiated a lock-down command via their mobile phone. The system first uploaded the four-directional control potential field vector of the lock to the cloud. The cloud control platform simultaneously performed action execution path simulation, action execution rollback path simulation, and no-execution path simulation, covering a 1.8-second time window. During the simulation, the system detected that if the action was executed around 17:20:08, the reversible potential field would drop to a low value during the rollback phase due to insufficient communication stability, posing a risk of not being able to roll back in time. Therefore, the cloud generates a reverse counterfactual control window and performs reversibility constraint domain shaping processing, shortening the executable time period from the originally planned 17:20:06—17:20:10 and shifting it to the right to 17:20:09—17:20:11. At the same time, the maximum potential field perturbation, the maximum micro-action chain length, and the rollback trigger condition are defined in the window.

[0032] Upon receiving the control window, the edge collaborative gateway does not immediately execute the complete locking action. Instead, it generates a chain of micro-actions based on the window constraints, breaking down the locking action into multiple low-amplitude, low-disturbance micro-action units. Before and after each micro-action, the system re-acquires the four-way control potential field vector and calculates the reversibility gradient. When the reversibility gradient shows a decreasing trend, it automatically reduces the amplitude or shortens the execution time. When the potential field disturbance approaches the limit, a protection gap is inserted to ensure that the entire action process always meets the reversibility and safety requirements. Taking the action of the day as an example, the entire locking action was broken down into 6 micro-actions. When executing the third micro-action, the occupancy of the potential field decreased briefly due to the vehicle's approach. The system automatically shortened the execution time and adjusted the step size, successfully avoiding the risk of erroneous locking.

[0033] Table 1. Statistics on pilot projects for residential parking areas

[0034] Table 1 shows that this invention demonstrates outstanding performance in terms of the stability of the ground lock execution. During the pilot period, the cumulative number of ground lock executions increased from 5280 to 5428, and the success rate increased from 95.1% to 99.1%. This fully demonstrates that the four-way coupled control potential field is more accurate in judging the reachability of execution and can effectively reduce failures caused by changes in resistance, link jitter, or distortion of occupancy information. Especially in terms of rollback capability, the traditional method failed to rollback 11 times, while this invention achieved 0 times, completely eliminating the high-risk problem of unrecoverable failures. This demonstrates the key role of the reverse counterfactual control window in action boundary prediction and decision constraints.

[0035] This invention also achieves significant results in adapting to complex operating environments. Before the pilot test, the number of operation interruptions caused by link fluctuations reached 42 times per month, while after adopting this invention, it dropped to 18 times per month, a reduction of more than half. This indicates that the communication potential field and window shaping mechanism can effectively avoid the execution risks under unstable links. The average execution time was shortened from 2.05 seconds to 1.82 seconds, and the number of false occupancy actions decreased from 7 to 1. This reflects that the occupancy potential field layer has more accurate spatiotemporal encapsulation of vehicle approach trajectory and locking state, reducing the probability of false triggering and enabling the ground lock to have stronger adaptability and stability in complex scenarios.

[0036] In terms of potential field adjustment and micro-motion chain execution capabilities, this invention also demonstrates significant advantages. After pilot testing, the average length of the micro-motion chain was 5.6 segments, indicating that the system can break down the overall action into multiple controllable small-amplitude movements, achieving finer-grained dynamic adjustment. Simultaneously, the number of potential field disturbances exceeding the limit was only 2, both of which were promptly captured by the system and triggered the rollback chain, effectively preventing the spread of anomalies. The reversible potential field values ​​before and after execution were 0.83 and 0.81, respectively, with minimal difference, indicating that the execution of the micro-motion chain has a very low impact on the overall reversibility of the parking lock. This invention achieves fully controllable, reversible, and adaptive remote parking lock control capabilities, providing a safer and more reliable control solution for high-density parking scenarios.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A remote control system for ground locks based on cloud-edge collaboration, characterized in that, include: The multi-source state preprocessing module collects multi-source state data related to the operation of the ground lock, preprocesses the multi-source state data, and generates a basic state vector. The four-way potential field construction module constructs a four-way coupled control potential field builder consisting of an action potential field layer, a communication potential field layer, an occupancy potential field layer, and a reversibility potential field layer. It performs potential field tensor quantization reconstruction processing on the basic state vector to obtain the four-way control potential field vector. The path deduction module uploads the four-way control potential field vector to the cloud control platform, and performs action execution path deduction, action execution rollback path deduction, and non-execution path deduction respectively, generating reverse counterfactual difference; The control window generation module generates a reverse counterfactual control window in the cloud based on the reverse counterfactual difference and the four-way control potential field vector. It performs reversibility constraint domain shaping processing to determine the action execution boundary and sends the shaped reverse counterfactual control window to the edge collaborative gateway. The micro-action chain adjustment module generates a micro-action chain based on the reverse counterfactual control window at the edge collaborative gateway. It decomposes the lifting and lowering action of the ground lock into multiple micro-action units. Before and after the execution of each micro-action unit, it re-acquires the four-way control potential field vector, calculates the invertibility gradient, and adjusts the execution amplitude, execution duration, and step size of the micro-action unit. In the abnormal rollback update module, if the reversible potential field quantity in the four-way control potential field vector is lower than the threshold or communication instability, abnormal resistance, or occupancy change occurs during the execution of the micro-action chain, the micro-action chain is terminated and the rollback chain reset lock is triggered. The execution trajectory and potential field disturbance data are reported, and the reverse counterfactual control window strategy and four-way potential field weight parameters are updated.

2. The remote control system for ground locks based on cloud-edge collaboration according to claim 1, characterized in that, The multi-source status data includes action status data, communication channel status data, occupancy status data, and energy status data.

3. The remote control system for ground locks based on cloud-edge collaboration according to claim 1, characterized in that, The preprocessing of multi-source state data to generate a basic state vector includes: performing noise reduction, formatting, timing alignment, and normalization on the multi-source state data, and combining the processed data to form a basic state vector, which includes an action sub-vector, a communication sub-vector, an occupancy sub-vector, and an energy sub-vector.

4. The remote control system for ground locks based on cloud-edge collaboration according to claim 1, characterized in that, The four-directional potential field construction module includes: Construct a four-way coupled control potential field builder, initialize the action potential field layer, communication potential field layer, occupancy potential field layer and reversible potential field layer, and set a unified time base and sampling step size; Based on the four-way coupled control potential field builder, hierarchical potential field processing is performed on the basic state vector: The action potential field layer performs action differential spectrum extraction and drag gradient encoding on the action sub-vector, and outputs the action potential field sub-vector. The communication potential field layer performs joint stability coding of link delay and jitter and packet loss mutation masking on the communication subvector, and outputs the communication potential field subvector. The occupancy potential field layer performs spatiotemporal occupancy encapsulation of the lock-in state and the approach trajectory on the occupancy subvector, and outputs the occupancy potential field subvector. The reversible potential field layer performs energy margin, mechanical rebound and link maintenance capability reversibility margin estimation and back-off condition boundary calculation on the energy subvector, and outputs the reversible potential field subvector. Perform temporal alignment, scale registration, and energy level mapping on each potential field subvector to form four types of potential field subvector sets, resulting in the registered four types of potential field subvector sets. Potential field tensor quantization reconstruction is performed on the registered four types of potential field subvector sets: Pairwise interactive coupling is expanded to capture the direct coupling relationship between layers; ternary interactive coupling is aggregated to capture the joint effect across three layers; and quaternary interactive coupling is summarized to form an overall coupling representation. During the interactive unfolding process, redundant interactive components are eliminated, and constraint domain shaping is performed in the summarization stage to satisfy the boundary conditions that do not reduce monotonicity and invertibility, thus obtaining the four-way coupled intermediate representation. The four-way coupled intermediate representation is subjected to aggregation mapping processing. The four types of potential field interaction results are reorganized in a fixed dimension, the components are merged and the hierarchical projection is performed to generate a four-way control potential field vector. The four-way control potential field vector includes the action reachability potential field, the communication stability potential field, the occupancy security potential field, and the reversibility potential field.

5. A remote control system for ground locks based on cloud-edge collaboration according to claim 1, characterized in that, The path deduction module includes: Receive the four-way control potential field vector and establish the four-way potential field time series according to the timestamp, and determine the unified deduction time window, time step and boundary conditions; Action execution path deduction: Apply action load deformation scheduling step by step within a unified time window to dynamically adjust the action intensity and rhythm according to the reversible potential field quantity, enable communication occupancy pre-quota to limit the peak link occupancy and insert link stability checkpoints, and use occupancy boundary look-ahead encapsulation to extrapolate the time and suppress over-boundary of the occupancy potential field safety boundary to generate action execution path trajectory. Rollback path deduction after action execution: Based on the reversible potential field quantity, a rollback envelope constraint is generated. After the action completion node, the rollback envelope constraint is applied step by step to advance the rollback process. Simultaneously, the reversible energy ledger records energy consumption and recoverable margin. Communication is deployed to maintain sentinel verification to determine the link availability of key rollback nodes and generate the rollback path trajectory after action execution. Execution and non-execution path deduction: Within a unified time window, without applying any action, the four-way potential field time series is updated step by step according to the natural evolution process, including the evolution of communication fluctuations, occupancy changes and energy self-discharge, to obtain the non-execution path trajectory; Within a unified time window, the three trajectories are aligned in time and their lengths are consistent. The corresponding component differences between the rollback path trajectory after the action is executed and the non-executed path trajectory are compared step by step in chronological order to generate a difference sequence. The reverse counterfactual difference is obtained by aggregating the results through time-weighted accumulation, abnormal component removal, and interval pruning.

6. The remote control system for ground locks based on cloud-edge collaboration according to claim 1, characterized in that, The control window generation module includes: Receive the four-way control potential field vector and the inverse counterfactual difference in the cloud, and establish a unified time window, state threshold set and execution boundary index; Generate an initial reverse counterfactual control window, and filter time periods in chronological order that meet the following conditions: reversibility potential field quantity is not lower than the safety threshold, communication stability is not lower than the link threshold, occupancy security is in the safe zone, and action reachability meets the execution conditions. Record the time periods that pass the filter as candidate execution areas, and determine the allowed action type, maximum micro-action chain length, potential field disturbance upper limit value, and rollback trigger condition for each candidate execution area. Perform reversibility constraint domain shaping on the initial reverse counterfactual control window, including: Perform boundary compression on segments exhibiting a downward trend in reversibility; Insert link protection gaps in communication fluctuation sections; A safety buffer zone shall be set up for the critical section of the occupation; Perform adjacent merging and hole filling on discrete candidate execution regions; Slice the cross-boundary segments and remove sub-regions that do not satisfy monotonicity to obtain a shaped execution boundary set; The consistency and validity of the shaped execution boundary set are checked, and the final reverse counterfactual control window is generated and sent to the edge collaboration gateway.

7. A remote control system for ground locks based on cloud-edge collaboration according to claim 1, characterized in that, The micro-action chain adjustment module includes: The edge collaborative gateway receives the reverse counterfactual control window, parses the time window, reversibility potential field threshold, communication stability threshold, occupancy security threshold, potential field disturbance upper limit, maximum chain length and rollback trigger condition, generates a micro-action chain initialization parameter set, and selects a micro-action template that matches the window constraints from the pulse packet template library. Based on the micro-action template, the lifting action of the ground lock is decomposed into multiple micro-action units. The initial execution sequence is established using a three-element description of amplitude, duration, and step distance. Potential field disturbance budget and local rollback sub-chain are allocated to each micro-action unit. Before each micro-action unit is executed, a four-way control potential field vector is obtained, and a double sampling closed loop is executed to obtain the change in potential field before and after execution. Based on the change in potential field before and after execution and the window threshold group, an invertibility gradient metric is constructed to shape the execution amplitude, execution duration and step size of the current micro-action unit. The micro-action unit is executed and the four-way control potential field vector is reacquired after completion. The execution amplitude, execution time and step size of the next micro-action unit are adjusted in linkage according to the continuous change of the invertibility gradient metric. When the communication stability or occupancy security is close to the threshold, the protection gap and check point are automatically inserted. During the execution of the micro-action chain, if the reversibility potential field is lower than the threshold or any of the conditions of communication stability, occupancy security, or potential field disturbance exceeds the limit, the current micro-action unit is immediately interrupted, the local rollback sub-chain is reset, and the micro-action execution parameters and potential field changes are recorded.

8. The remote control system for ground locks based on cloud-edge collaboration according to claim 1, characterized in that, The abnormal rollback update module includes: During the execution of the micro-action chain, the four-way control potential field vector is read in real time and the execution trajectory is cached. The read reversible potential field quantity is compared with the set threshold, and at the same time, it is determined whether the communication stability potential field quantity, the occupancy security potential field quantity, and the potential field disturbance quantity have exceeded the limit. When the reversibility potential field is below the threshold or any of the conditions of communication stability, occupancy security, or potential field disturbance exceeds the limit, an interrupt command is immediately issued to terminate the current micro-action unit. According to the rollback trigger conditions in the reverse counterfactual control window, the corresponding rollback action chain is invoked to execute the rollback micro-action unit until the ground lock is restored to a safe state. The system collects execution trajectory and potential field disturbance data, generates execution records, and uploads them to the cloud. These records are used to refresh the reverse counterfactual control window generation strategy and the four-way potential field weight parameters.

Images