Shared parking space management method and device based on Internet of Things, medium and product

Abstract

The invention discloses a shared parking space management method and device based on the Internet of Things, a medium and a product. The method comprises the following steps: assigning a target parking space for a parking reservation request, creating a digital twinborn model, and initializing the digital twinborn model to a reserved state; collecting a multi-channel time sequence vibration signal covering the target parking space area; performing joint time-frequency analysis on the multi-channel time sequence vibration signals to generate a space-time vibration energy distribution diagram; performing sequence pattern recognition on the space-time vibration energy distribution diagram, and analyzing a semantic event; a state updating instruction is sent to the digital twin model, and the state of the digital twin model is driven to be converted; when the state of the digital twin model is updated to a drive-in state, obtaining vehicle identity information, updating the state of the digital twin model to an occupation state, and recording an occupation starting moment; and when the state of the digital twin model is updated to a driving-off state, the parking duration is calculated, and a charging voucher is generated. By implementing the technical scheme, the problem of low reliability caused by a single sensing dimension is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet of Things, and in particular to a shared parking space management method, device, medium and product based on Internet of Things. BACKGROUND

[0002] In the existing shared parking management technology, in order to obtain the real-time occupancy state of the parking space, a single type of sensor based on Internet of Things is usually used for monitoring. For example, a common technical solution is to install an ultrasonic sensor directly above the parking space, to determine whether the parking space is occupied by measuring the distance change between the sensor and the ground or the top of the vehicle, and to send the occupied or idle signal to the cloud server. Another solution is to bury a geomagnetic sensor under the ground of the parking space, to sense the driving in and out of the vehicle by monitoring the disturbance of the vehicle chassis metal to the earth's magnetic field. In addition, some systems also use the method of installing a camera at a fixed position, to analyze the images captured by the back-end server to identify whether there is a vehicle outline in the parking space, so as to determine the occupancy state.

[0003] However, the above existing technology has the technical problem of insufficient accuracy in monitoring the state of the parking space in actual application. Since these technical solutions usually rely on a single source of information for judgment, their accuracy is easily disturbed by complex external environments. For example, ultrasonic sensors may misjudge due to obstacles such as rain, snow, and fallen leaves; geomagnetic sensors are easily affected by the surrounding strong electromagnetic environment or high chassis vehicles, resulting in detection failure; and the image analysis-based method will significantly decrease the recognition accuracy in conditions of poor light and vision such as night, bad weather, or camera obstruction. The low reliability caused by a single sensing dimension makes it impossible for the platform to obtain accurate parking space state data, thereby affecting the effectiveness of subsequent parking reservation and resource scheduling, and reducing the practical value of the entire shared parking system. SUMMARY

[0004] To solve the above technical problems, the present application provides a shared parking space management method, device, medium and product based on Internet of Things.

[0005] In the first aspect of the present application, a shared parking space management method based on Internet of Things is provided, which adopts the following technical solution: assign a target parking space to the received parking reservation request, create a digital twin model bound to the target parking space, and initialize the digital twin model to a reserved state; collect multi-channel time-series vibration signals covering the target parking space area in real time through a vibration sensor array deployed at the target parking space; generate a time-space vibration energy distribution map by jointly performing time-frequency analysis on the multi-channel time-series vibration signals; performing sequence pattern recognition on the spatio-temporal vibration energy distribution map to analyze a semantic event matching a preset vehicle behavior logic sequence; based on the semantic event, sending a state update instruction to the digital twin model to drive a state of the digital twin model to be converted, the state including an idle state, a reserved state, a driving-in state, an occupied state and a driving-off state; when the state of the digital twin model is updated from the reserved state to the driving-in state, triggering a preset image acquisition device to acquire vehicle identity information, and based on a verification result of the vehicle identity information and reserved vehicle information in the parking reservation request, updating the state of the digital twin model to the occupied state and recording an occupied start time; when the state of the digital twin model is updated to the driving-off state, calculating a parking duration and generating a billing voucher according to the occupied start time and an end time of the driving-off state.

[0006] By adopting the above technical solution, the accuracy and reliability of the parking space state monitoring are improved. Specifically, the vibration sensor is not sensitive to environmental interference such as rain, snow and fallen leaves, as it mainly perceives vibration characteristics generated by physical movement of the vehicle; the multi-channel signal and the spatio-temporal energy distribution map enhance the robustness of feature extraction, avoiding the defects of single sensor being easily affected by electromagnetic, light or shielding. The state conversion mechanism of the digital twin model combined with image verification ensures accurate matching of vehicle identity and occupancy state, effectively reducing misjudgment.

[0007] Optionally, the assigning a target parking space to the received parking reservation request and creating a digital twin model bound to the target parking space and initializing the digital twin model to a reserved state include: According to the health state of the vibration sensor array associated with each parking space in the preset parking space resource pool and the geographic location information of each parking space, a candidate parking space set meeting a preset condition is selected; According to the user preference or historical parking data in the parking reservation request, the target parking space is matched from the candidate parking space set for the parking reservation request; based on a preset digital twin model template, filling the static attributes of the target parking space, the data interface identifier of the vibration sensor array and the reservation information of the parking reservation request into the digital twin model template to create the digital twin model; setting the internal state variable of the digital twin model to the reserved state.

[0008] By adopting the above technical solutions, the reliability of the shared parking space allocation process and the user experience have been improved. Specifically, the health status of sensors is considered during the allocation stage, ensuring the reliability of subsequent status perception data from the source and avoiding misjudgments caused by sensor device malfunctions. At the same time, personalized matching is performed by combining geographical location and user preferences, improving the utilization efficiency of parking space resources and user satisfaction.

[0009] Optionally, generating a spatiotemporal vibration energy distribution map by performing joint time-frequency analysis on the multi-channel time-series vibration signals includes: Each channel of the multi-channel time-series vibration signal is processed by a preset wavelet transform algorithm to obtain a set of time-spectrum diagrams; For each time spectrum diagram in the set of time spectrum diagrams, multiple energy characteristic values ​​are calculated in a preset time-frequency analysis pane. Each energy characteristic value is combined with the corresponding preset spatial coordinates of the vibration sensor to obtain spatial energy data points; Based on the spatial energy data points, a preset spatial interpolation algorithm is used to calculate the energy value for each grid point in the preset two-dimensional spatial grid, thereby generating the spatiotemporal vibration energy distribution map.

[0010] By adopting the above technical solutions, the accuracy and robustness of vehicle behavior feature extraction are improved. Specifically, wavelet transform is used to process non-stationary temporal vibration signals, which can accurately capture transient features generated at different frequency bands when a vehicle passes by, overcoming the shortcomings of insufficient feature extraction by traditional single sensors in complex environments.

[0011] Optionally, the step of performing sequence pattern recognition on the spatiotemporal vibration energy distribution map and parsing out semantic events that match the preset vehicle behavior logic sequence includes: Extract multi-dimensional spatiotemporal feature vectors from the time series of the spatiotemporal vibration energy distribution map; The multi-dimensional spatiotemporal feature vector is input into a preset recurrent neural network model to generate a probability distribution of the multi-dimensional spatiotemporal feature vector belonging to each preset event category, wherein the preset event category is a classification target determined during the training of the recurrent neural network model; Based on the probability distribution, a preset event category corresponding to the highest probability value is determined as the initial identification event; The preliminary identification event is matched with the preset vehicle behavior logic sequence. When the match is successful, the preliminary identification event is identified as the semantic event.

[0012] By adopting the above technical solutions, a highly reliable vehicle behavior analysis framework was constructed, thereby achieving accurate and automated semantic-level understanding of parking space occupancy status. Specifically, by using recurrent neural networks to process the sequence features extracted from the spatiotemporal vibration energy distribution map, it can effectively learn and identify the temporal dependencies and dynamic patterns of vehicle events on vibration signals. Its powerful sequence modeling capability significantly improves the accuracy and generalization ability of event classification in complex environments, overcoming the poor robustness of traditional methods based on fixed thresholds or simple rules.

[0013] Optionally, matching the preliminary identification event with the preset vehicle behavior logic sequence, and determining the preliminary identification event as the semantic event when a match is successful, includes: Based on the preset vehicle behavior logic sequence and the current state of the digital twin model, the logical validity of the preliminary identification event is verified. Based on the occurrence time of the preliminary identified event and the start time of the current state, the time difference is calculated, and the time difference is checked for time compliance according to the preset time window associated with the preliminary identified event. Extract the energy features of the spatiotemporal vibration energy distribution map corresponding to the preliminary identification event, and perform energy conformity verification on the energy features based on the preset energy transition features associated with the preliminary identification event; The matching is considered successful when the results of the logical validity check, the time compliance check, and the energy compliance check are all passed.

[0014] By adopting the above technical solutions, a multi-dimensional, high-confidence event confirmation system was constructed, thereby improving the accuracy of semantic event parsing and the system's anti-interference capability. Specifically, logical validity verification ensures the rationality of the event sequence and the state transition of the digital twin model, preventing logical chaos in the state machine from the business process level; time compliance verification filters out occasional false alarms caused by signal delays or instantaneous disturbances through a preset time window, enhancing the accuracy of timing judgment; and energy compliance verification performs secondary confirmation from the physical characteristic level, effectively identifying interference from non-vehicle vibrations (such as pedestrians passing by).

[0015] Optionally, the step of calculating the parking duration and generating a billing voucher based on the occupancy start time and the departure end time includes: From the semantic events that trigger the state update of the digital twin model to the departure state, identify the event of the vehicle completely leaving the vehicle, and determine the time of occurrence of the event of the vehicle completely leaving the vehicle as the end time; The parking duration is calculated based on the start and end times of occupancy, and the pre-designed fee rule engine is invoked to calculate the parking fee to be paid based on the parking duration and the time-based fee rate of the target parking space. The occupancy start time, the departure end time, the parking duration, and the parking fee to be paid are integrated into structured billing data, and the billing voucher is generated based on the structured billing data.

[0016] By adopting the above technical solution, high accuracy and efficiency in parking duration calculation and fee generation are achieved. Specifically, the end time of the occupancy period is accurately determined based on semantic events (vehicle completely leaving), combined with the start time of occupancy driven by vibration sensing. This ensures the objectivity and accuracy of the timing basis, fundamentally avoiding timing disputes that may be caused by misjudgment of status in traditional methods. By calling the pre-designed fee rule engine for automatic calculation, not only are errors and delays that may be caused by manual intervention eliminated, but the uniformity and transparency of the billing standard are also guaranteed.

[0017] Optionally, the method further includes: In response to receiving an appeal request for the billing voucher, the system retrieves the occupancy start time, the departure end time, the vehicle identity information associated with the billing voucher, and the spatiotemporal vibration energy distribution map sequence generated between the occupancy start time and the departure end time. Based on the occupancy start time, a first spatiotemporal vibration energy distribution map is matched and extracted from the spatiotemporal vibration energy distribution map sequence, and the occupancy start time, the vehicle identity information, and the first spatiotemporal vibration energy distribution map are constructed into a first event evidence group; Based on the end time of the departure state, a second spatiotemporal vibration energy distribution map is matched and extracted from the spatiotemporal vibration energy distribution map sequence, and the end time of the departure state, the vehicle identity information, and the second spatiotemporal vibration energy distribution map are constructed into a second event evidence group. An event retrospective report is generated based on the first event evidence group and the second event evidence group.

[0018] By adopting the above technical solutions, the credibility of the system and the user experience have been effectively improved. Specifically, when a user disputes the billing, the system can quickly retrieve and associate key time information, vehicle identity, and the most crucial original perception data—a sequence of spatiotemporal vibration energy distribution maps—transforming abstract timing nodes into concrete, visual physical vibration evidence. This intuitive presentation based on the vibration energy distribution map also greatly enhances the persuasiveness and efficiency of the appeal process, resolving billing disputes arising from status perception controversies.

[0019] A second aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any of the foregoing.

[0020] A third aspect of this application provides a computer-readable storage medium storing instructions that, when executed, perform the method described in any of the preceding descriptions.

[0021] A fourth aspect of this application provides a computer program product that, when run on an electronic device, causes the electronic device to perform the method as described in any of the preceding claims.

[0022] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: By deeply integrating a vibration sensor array with a digital twin model, a shared parking management method with multi-dimensional perception, intelligent decision-making, and full-process traceability was constructed. This method utilizes joint time-frequency analysis to transform vibration signals into a spatiotemporal energy distribution map, and accurately analyzes vehicle behavior semantics through sequence pattern recognition and a triple verification mechanism, achieving precise management throughout the entire lifecycle from reservation, entry, occupancy to departure. It effectively overcomes the technical shortcomings of traditional single-sensor systems, which are susceptible to environmental interference, and improves the accuracy and reliability of parking space status monitoring. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the system architecture of an embodiment of a shared parking space management method based on the Internet of Things according to this application; Figure 2 This is a flowchart illustrating a shared parking space management method based on the Internet of Things disclosed in an embodiment of this application; Figure 3 This is another flowchart illustrating a shared parking space management method based on the Internet of Things disclosed in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application.

[0024] Explanation of reference numerals in the attached figures: 100, System architecture; 101, First terminal device; 102, Second terminal device; 103, Third terminal device; 104, Network; 105, Server; 401, Processor; 402, Communication bus; 403, User interface; 404, Network interface; 405, Memory. Detailed Implementation

[0025] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification 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.

[0026] Figure 1 This is a schematic diagram of the system architecture of an embodiment of a shared parking space management method based on the Internet of Things according to this application; like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. The network 104 serves as the medium for providing communication links between the terminal devices 101, 102, and 103 and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables. Users can use the terminal devices 101, 102, and 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications may be installed on the terminal devices 101, 102, and 103, such as model training applications, video recognition applications, web browser applications, and social media platform software.

[0027] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with displays, including but not limited to smartphones, tablets, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III) players, MP4 (Moving Picture Experts Group Audio Layer IV) players, laptops, and desktop computers, etc. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices. They can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services) or as a single software program or software module. No specific limitations are imposed here.

[0028] This embodiment discloses a shared parking space management method based on the Internet of Things. Figure 2 This is a flowchart illustrating an IoT-based shared parking space management method disclosed in an embodiment of this application. Figure 2 As shown, the method includes the following steps: S201. Assign a target parking space to the received parking reservation request, create a digital twin model bound to the target parking space, and initialize the digital twin model to the reserved state. A parking reservation request is not a simple signal, but a structured data packet, typically containing the user's unique identifier, the desired start and end times of parking, and the vehicle's identification information (such as license plate number). The digital twin model is not a physical entity, but a virtual software object or data structure dynamically generated on a cloud server, corresponding one-to-one with the physical parking space. This model can be understood as a "proxy" for the physical parking space in the digital world. It not only encapsulates the space's static attributes (such as number, geographical coordinates, and size), but more importantly, it establishes a data link to the space's physical sensing devices (i.e., a vibration sensor array) and carries all the dynamic information of the reservation (such as the user ID, vehicle information, and reservation time period).

[0029] Optionally, assigning a target parking space to the received parking reservation request, creating a digital twin model bound to the target parking space, and initializing the digital twin model to the reserved state includes: filtering a set of candidate parking spaces that meet preset conditions based on the health status of the vibration sensor array associated with each parking space in a preset parking space resource pool and the geographical location information of each parking space; matching the target parking space for the parking reservation request from the set of candidate parking spaces based on user preferences or historical parking data in the parking reservation request; filling the digital twin model template with the static attributes of the target parking space, the data interface identifier of the vibration sensor array, and the reservation information of the parking reservation request to create the digital twin model; and setting the internal state variables of the digital twin model to the reserved state.

[0030] Specifically, upon receiving a parking reservation request, the server first initiates a rigorous screening process to select a preliminary set of candidate parking spaces that meet the criteria from the entire parking lot's resource pool. The core of this process is a double verification. The first verification is device health: for each parking space, the operational status of its associated vibration sensor array is checked. This status is not simply online or offline, but a comprehensive health score. For example, this score considers factors such as the sensor's remaining battery power (e.g., required to be above 20%), wireless signal strength (e.g., value must be greater than -90dBm), and the continuity and completeness of recent data reports. Any parking space with a health score below a preset threshold (e.g., 80 points) is excluded to prevent subsequent service issues caused by sensor malfunctions. The second verification is geographic location: the coordinates of the target location specified in the reservation request are parsed, and a spatial index (e.g., R-tree) is used to quickly query all available parking spaces within a certain distance (e.g., 800 meters) with healthy sensors. Only parking spaces that pass both of these checks can be included in the final set of candidate parking spaces for further refined matching.

[0031] Furthermore, after forming a high-quality set of candidate parking spaces, a personalized matching and ranking process based on multi-dimensional scoring is executed to match the user with the best parking space. For each parking space in the candidate set, a scoring calculation program is initiated. This program comprehensively considers multiple factors and assigns different weights. For example, the overall score = (weight A × user preference matching score) + (weight B × historical behavior matching score) - (weight C × price cost score). The user preference matching score is calculated based on the degree of consistency between the user's explicit settings in their account (such as preference for proximity to elevators or need for charging stations) and the parking space attributes. The historical behavior matching score is obtained by analyzing the user's recent parking records (such as selecting the third basement level 15 out of the last 20 parkings), awarding bonus points to parking spaces that match the user's hidden habits. The price cost score converts the parking space rate into a negative score. After scoring and ranking all candidate parking spaces, the one with the highest score is ultimately determined as the target parking space for this reservation. In the event of a tie, priority will be given to parking spaces with shorter walking distances to the user's final destination (such as a specific store in a shopping mall).

[0032] Furthermore, once a unique target parking space is selected, the server immediately shapes it in the digital world, creating a dedicated, real-time virtual instance based on a standardized digital twin model template. This creation process is akin to filling out a detailed file. First, a new data object is cloned from the template (a predefined JSON (JavaScript Object Notation) structure) and assigned a globally unique identifier. Next, using the target parking space's ID, all its static attributes are retrieved from the database, such as the number B2-C-08, precise latitude and longitude coordinates, dimensions of 2.4 meters x 5.3 meters, load capacity, etc., and these are filled into the corresponding fields of the new object. Further, the data interface identifier for the vibration sensor array is entered. This can be a specific MQTT (Message Queuing Telemetry Transport) subscription topic (e.g., parkinglot / area_C / spot_08 / vibration_stream) or a dedicated WebSocket connection address, forming the lifeline between the virtual model and physical sensing. Finally, the reservation information for this parking reservation request, including dynamic business data such as user ID, license plate number, and reservation start and end times, will also be written into this data object.

[0033] Furthermore, after the digital twin model is successfully created, a final crucial operation is performed to make it officially effective and ready for operation: setting its internal state variable to the reserved state. Technically, this involves writing the predefined value RESERVED to a specific field of the model's data object (e.g., a field named current_status). This action triggers a series of chain reactions. First, in the underlying resource management database, an exclusive lock is added to the target parking space for the specified reservation period to ensure it is not reassigned. Second, a status change event message is published to the message bus. The user's mobile application subscribes to this message and immediately refreshes the interface upon receiving it, displaying a confirmation message of successful reservation and a parking space navigation entry. More importantly, the internal logic of the digital twin model instance is activated; it immediately enters an active listening mode, continuously receiving and analyzing vibration signals from corresponding physical sensors through its data interface, ready to identify events consistent with vehicle entry behavior characteristics.

[0034] S202. By deploying a vibration sensor array in the target parking space, multi-channel time-series vibration signals covering the target parking space area are collected in real time. In a preferred embodiment, the vibration sensor array may consist of a set (e.g., 9 or 16) of highly sensitive piezoelectric accelerometers. During deployment, these sensors are pre-embedded in a near-uniform grid (e.g., a 3x3 or 4x4 matrix) approximately 5-10 cm below the asphalt or concrete surface structure of the target parking space. This burial method not only protects the sensors from vehicle traffic and environmental erosion but also ensures good mechanical coupling between the sensors and the ground, enabling accurate pickup of subtle surface vibrations caused by vehicle pressure, tire rolling, or engine idling. These sensors are connected via shielded cables to a local data acquisition (DAQ) unit located near the parking space. This DAQ unit incorporates a multi-channel synchronous analog-to-digital converter capable of synchronously sampling and quantizing the analog signals from all sensor channels at a relatively high, preset sampling frequency (e.g., 1 kHz), thereby generating a digitized multi-channel time-series vibration signal. In another alternative embodiment, the vibration sensor array comprises multiple independent, battery-powered wireless sensor nodes, each integrating a low-power microelectromechanical system (MEMS) triaxial accelerometer. The deployment of these nodes can be more flexible. For example, they can be surface-mounted and fixed to several key locations in the parking space, such as the center and four corners, forming a non-uniform sparse array. Unlike the continuous acquisition in the previous embodiment, the real-time acquisition here employs a sleep-wake-up energy-saving strategy. Under normal circumstances, all sensor nodes are in a low-power sleep or listening mode, detecting vibration energy at a very low frequency to see if it exceeds a very low wake-up threshold. Once the vibration energy of any node exceeds this threshold (which usually means that a vehicle is approaching or entering the parking space), the node will immediately broadcast a wake-up signal to all other nodes in the array and the main controller via wireless communication (e.g., ZigBee). Upon receiving the signal, the entire array will immediately switch to a high-frequency synchronous sampling mode, acquiring multi-channel temporal vibration signals completely within a preset period of time (e.g., 5 seconds).

[0035] S203. By performing joint time-frequency analysis on the multi-channel time-series vibration signals, a spatiotemporal vibration energy distribution map is generated. In a preferred implementation, the system performs windowing and overlapping segmentation on the one-dimensional time-series vibration signal input to each channel of the sensor array using a window function such as a Hamming window. Next, a Fast Fourier Transform (FFT) is performed on each short-time signal segment to transform it from the time domain to the frequency domain and obtain its spectral information. Subsequently, the system performs energy integration on the spectrum of each segment within a preset target frequency band closely related to vehicle vibration characteristics (e.g., 0-100Hz) to calculate a scalar energy value for each channel at each time point. Thus, at any given analysis time, the system obtains the energy values ​​corresponding to N (positive integers, greater than or equal to 1) sensors. The system takes N energy values ​​at the physical location and uses these N energy values ​​and their known sensor spatial coordinates as control points. It then calculates and generates a two-dimensional dense energy distribution grid covering the entire parking space area using a spatial interpolation algorithm (such as inverse distance weighted interpolation or bilinear interpolation). This grid is then visualized as a pseudo-color heatmap, where color changes (e.g., from blue to red) intuitively correspond to the intensity of vibration energy. This presents local high-energy areas generated by vehicle tires or engines as clear "bright spots" or "hot spots," providing a high-quality, high signal-to-noise ratio decision-making basis for subsequent parking space status assessment.

[0036] Optionally, the step of generating a spatiotemporal vibration energy distribution map by performing joint time-frequency analysis on the multi-channel time-series vibration signals includes: processing each of the multi-channel time-series vibration signals using a preset wavelet transform algorithm to obtain a set of time-spectrum diagrams; calculating multiple energy feature values ​​for each time-spectrum diagram in the set within a preset time-frequency analysis pane; combining each energy feature value with the preset spatial coordinates of the corresponding vibration sensor to obtain spatial energy data points; and using a preset spatial interpolation algorithm based on the spatial energy data points to calculate energy values ​​for each grid point in a preset two-dimensional spatial grid, thereby generating the spatiotemporal vibration energy distribution map.

[0037] In this embodiment, the preferred preset wavelet transform algorithm is the Continuous Wavelet Transform (CWT), as it exhibits excellent time-frequency localization characteristics when processing non-stationary signals such as vehicle vibrations. Specifically, the system independently executes CWT for the signal channel corresponding to each sensor in the sensor array. Before execution, a mother wavelet needs to be preset; for example, the Morlet wavelet or Paul wavelet, which is well-suited for analyzing transient impact signals, can be selected. Simultaneously, a scale parameter sequence needs to be defined, which directly corresponds to the frequency range to be analyzed. For example, the scale sequence can be set to cover a frequency range of 10Hz to 200Hz, effectively encompassing the characteristic vibration frequencies generated by key behaviors such as low-speed vehicle driving, tire friction, engine idling, and door opening / closing. After CWT processing of each signal, the output is a two-dimensional matrix, i.e., the time-spectrum diagram of that channel (often referred to as a scale diagram in this scenario). The row axes represent time, and the column axes represent scale (frequency). Each element in the matrix is ​​the wavelet coefficient modulus at that frequency point at that moment, characterizing the intensity of the vibration energy. By combining the time-spectrum graphs generated from all channels, we obtain the aforementioned time-spectrum graph set.

[0038] Furthermore, the time-frequency analysis pane here is a rectangular area defined on the time-frequency spectrum, defining the specific time and frequency range of interest. This pane is sliding, moving forward along the time axis in fixed steps (e.g., 100 milliseconds). The pane size is a configurable key parameter; for example, the time width can be set to 500 milliseconds, and the frequency range can be set to cover 15Hz to 50Hz. This combination of parameters is designed to specifically capture persistent low-frequency vibration energy corresponding to low-speed vehicle movement or idling. Once the analysis pane is determined, multiple energy characteristic values ​​are calculated within the area covered by the pane to more comprehensively describe the energy situation within that pane. One specific implementation involves calculating at least the following three characteristic values: first, the energy mean, which is the average of the squared moduli of all wavelet coefficients within the pane, representing the average vibration intensity during that time period; second, the energy peak, which is the maximum value of the squared moduli of the wavelet coefficients within the pane, representing the most intense instantaneous vibration intensity; and third, the energy centroid frequency, which is the frequency-weighted average value weighted by energy, indicating the frequency point where the energy is most concentrated. Through this step, the originally complex time-spectrum diagram is condensed into a series of energy feature vectors that vary over time and contain multiple dimensions.

[0039] Furthermore, each calculated energy feature value needs to be combined with the corresponding preset spatial coordinates of the vibration sensor to obtain a set of spatial energy data points. This is a data association step, the purpose of which is to provide input data with geographic location labels for subsequent spatial interpolation. During system initialization or sensor deployment, the precise two-dimensional coordinates (x, y) of each vibration sensor have been measured and recorded in the system's configuration file. These coordinates are usually based on a local coordinate system with the center of the parking space or a corner point as the origin. At any analysis time t, the system obtains the energy feature vector Vi(t) corresponding to each sensor (e.g., sensor i) from the previous step. Then, the system binds this energy vector with the sensor's coordinates (xi, yi) to form a spatial energy data point Pi(t) = {(xi, yi), Vi(t)}. For example, if a single energy mean is used as the feature value, the data point simplifies to {(xi, yi), Ei(t)}. After processing all N sensors, at time t, the system obtains a set of N data points {P1(t), P2(t), ..., PN(t)}.

[0040] Furthermore, based on this set of discrete spatial energy data points, the system will employ a preset spatial interpolation algorithm to calculate the energy value for each grid point in a preset two-dimensional spatial grid, ultimately generating a spatiotemporal vibration energy distribution map for that moment. First, the system will create a fine two-dimensional grid in virtual space that perfectly corresponds to the physical dimensions of the target parking space, for example, a grid with a resolution of 100x200. Then, a preset spatial interpolation algorithm is selected; for example, a robust and relatively simple scheme is the Inverse Distance Weighted Interpolation (IDW) method. For any grid point Pj in the grid whose energy value is to be determined, the IDW algorithm will traverse all N spatial energy data points Pi. For each Pi, the algorithm calculates the Euclidean distance d(j,i) between Pj and Pi, and calculates a weight w(i) based on this distance, typically w(i) = 1 / d(j,i). p (p is a power parameter, usually taken as 2). Finally, the energy value of point Pj is calculated as the weighted average of the energy values ​​of all sensors: E(j) = Σ[w(i) × Ei] / Σ[w(i)], where E(j) is the energy value of grid point Pj to be calculated; Ei is the known energy value of the i-th spatial energy data point, that is, the energy value obtained after measurement and processing by the i-th sensor; w(i) is the weight corresponding to the i-th spatial energy data point. After applying this calculation to each point in the grid, a complete energy matrix is ​​obtained. Rendering this matrix, for example in the form of a heatmap, where points with high energy are displayed in red and points with low energy are displayed in blue, generates a visually intuitive spatiotemporal vibration energy distribution map for that moment.

[0041] S204. Perform sequence pattern recognition on the spatiotemporal vibration energy distribution map and parse out semantic events that match the preset vehicle behavior logic sequence; Sequence pattern recognition is an advanced data analysis technique that focuses not on the state of individual data points, but on how data points (in this application, the spatiotemporal vibration energy distribution map of each frame) evolve over time into a sequence with specific patterns. This technique aims to identify a complete process conforming to a specific script from a continuous, dynamically changing energy map video, rather than simply identifying the features of a single frame. This script is the so-called pre-defined vehicle behavior logic sequence. It is not a simple template, but an abstract, phased logical model used to describe the typical evolutionary trajectory of a specific vehicle behavior (such as entering or leaving a parking space) at the vibration energy level. For example, the logical sequence of a vehicle entering a parking space can be pre-defined as follows: First stage, a high-energy cluster appears at the edge of the parking space entrance; second stage, the energy cluster moves into the parking space; third stage, the energy cluster stops moving and may be accompanied by a widening of its shape (vehicle comes to a stop); fourth stage, a continuous low-frequency vibration energy field covering the entire parking space appears (engine idling). When a sequence pattern recognition algorithm finds that the observed energy map sequence successfully matches the preset logical sequence, its output is a semantic event. Semanticization refers to successfully transforming the complex, purely digital signal patterns at the underlying level into a high-level label with clear business meaning or human-understandable meaning. For example, a successfully matched vibration data stream can be interpreted as EVENT_VEHICLE_ENTERED or that a vehicle has stopped.

[0042] Optionally, the step of performing sequence pattern recognition on the spatiotemporal vibration energy distribution map to parse out semantic events that match a preset vehicle behavior logic sequence includes: extracting multi-dimensional spatiotemporal feature vectors from the time series of the spatiotemporal vibration energy distribution map; inputting the multi-dimensional spatiotemporal feature vectors into a preset recurrent neural network model to generate a probability distribution of the multi-dimensional spatiotemporal feature vectors belonging to each preset event category, wherein the preset event category is a classification target determined during the training of the recurrent neural network model; determining the preset event category corresponding to the highest probability value as a preliminary identification event based on the probability distribution; matching the preliminary identification event with the preset vehicle behavior logic sequence, and when the match is successful, determining the preliminary identification event as the semantic event.

[0043] One specific implementation scheme is based on image morphological analysis. For each frame in the energy map time series, the system first executes an adaptive thresholding algorithm to segment the energy... Figure TwoThe data is divided into foreground (regions with significant energy) and background. Next, for the identified foreground region (which may consist of one or more connected components), a series of features are calculated to form the feature vector of the frame. These dimensions may specifically include: 1) Global energy features: the total energy, mean energy, and variance of the entire image; 2) Spatial location and distribution features: the centroid coordinates (x, y) of the region with the strongest energy (main energy cluster), the offset of the energy centroid relative to the center of the parking space, and the orientation angle; 3) Morphological features: the area, perimeter, and compactness (area / perimeter²) of the main energy cluster, the aspect ratio of the circumscribed rectangle, and the principal axis orientation angle calculated by principal component analysis (PCA) (representing the vehicle's orientation); 4) Multi-objective features: the number of connected components (e.g., when a vehicle is parked, the four tires may form multiple independent energy hotspots). Combining these dozens of feature values ​​constitutes a multi-dimensional spatiotemporal feature vector that comprehensively describes the state of the vibration energy field at that moment. As another feasible implementation, the system can employ a grid-based feature extraction method. This approach divides each frame's energy map into a fixed, coarse grid (e.g., a 5x10 grid). Then, for each grid cell, the average or maximum energy value within it is calculated. The calculation results for all grid cells are then concatenated in a predetermined order (e.g., row-by-row scanning) to form a fixed-length vector. The advantage of this method is its simplicity and fixed feature dimensions, making it well-suited for direct input into certain types of neural networks.

[0044] Furthermore, the system inputs the time series composed of the multi-dimensional spatiotemporal feature vectors extracted in real time in the previous step into a pre-trained Recurrent Neural Network (RNN) model. Its task is to generate in real time the probability distribution of the sequence belonging to each pre-defined event category at the current moment. The reason for choosing RNNs and their advanced variants (such as Long Short-Term Memory (LSTM) or Gated Recurrent Units (GRUs) is that they are specifically designed to process and remember the temporal dependencies in sequence data, effectively capturing the dynamic evolution patterns of vibration features during vehicle driving, parking, and other behaviors. The pre-defined event categories here are intermediate-layer, atomic-level behavior classifications determined during the model training phase through learning from a large amount of labeled data. For example, they can be finely divided into: wheel crossing (entering), vehicle translation, front and rear fine-tuning (parking), engine idling (stationary), door opening and closing vibration, personnel movement interference, and pure background noise. During runtime, the feature vector sequence is fed into the RNN model frame by frame, sequentially. The model's internal memory units continuously update their hidden states, fusing historical information to understand the current input. Finally, the model's output layer (usually a Softmax activation function layer) outputs a probability distribution vector for the latest input sequence. Each element of this vector corresponds to the probability of a predefined event category, for example, P(t) = {p(wheel crossing the line): 0.85, p(vehicle translation): 0.05, p(idle): 0.10, ...}. This quantitatively describes which atomic behavior the currently observed vibration mode is most likely to belong to. As an alternative, besides RNNs / LSTMs / GRUs, Temporal Convolutional Networks (TCNs) can also be used. TCNs use causal convolutions and dilated convolutions, enabling parallel processing of sequence data, faster training speeds, and a very large receptive field (i.e., the ability to recall long historical information) through stacked convolutional layers. They outperform traditional RNNs on many sequence modeling tasks.

[0045] Furthermore, the system determines a preliminary identification event based on the real-time probability distribution output by the RNN model in the previous step. The most direct implementation is to use the maximum a posteriori probability criterion, that is, simply select the event category with the highest probability value as the preliminary identification result at the current moment. For example, if the probability distribution is {p(wheel crossing the line): 0.85, ...}, then the preliminary identification event is wheel crossing the line. However, to improve the robustness of the system, a preferred implementation is to introduce a dual-threshold confidence mechanism. The system presets a high-confidence threshold T_high (e.g., 0.8) and a low-confidence threshold T_low (e.g., 0.5). Only when the probability of a certain event category first exceeds T_high will the system switch the preliminary identification event to that category. Thereafter, as long as the probability of that category is not lower than T_low, the system maintains the identification result unchanged. This mechanism can effectively filter out noise caused by signal fluctuations and frequent jumps near the decision boundary, making the output preliminary identification event sequence smoother and more stable.

[0046] Furthermore, in the first implementation, the logical model is concretized as a finite state machine (FSM), which predefines a unique state transition path conforming to a specific parking behavior (such as angled parking): when a continuous sequence of wheel-over-line events is input, the state machine transitions from idle to entering; subsequently, a sequence of vehicle body translation events transitions it to adjustment; finally, only when a continuous sequence of static vibration events is input does the state machine enter the final state of being stopped, thus determining a successful match. In the second alternative implementation, this logical matching process is implemented through a rule-based expert system containing a series of IF-THEN logical rules, such as setting the rule: IF (current state is adjustment) AND (continuous static signal received) THEN (update state to stopped). This approach allows administrators to more flexibly and dynamically add, delete, or modify the behavior recognition logic to adapt to different scenarios.

[0047] Optionally, matching the preliminary identification event with the preset vehicle behavior logic sequence, and determining the preliminary identification event as the semantic event when the match is successful, includes: performing a logical validity check on the preliminary identification event based on the preset vehicle behavior logic sequence and the current state of the digital twin model; calculating the time difference based on the occurrence time of the preliminary identification event and the start time of the current state, and performing a time compliance check on the time difference based on a preset time window associated with the preliminary identification event; extracting the energy features of the spatiotemporal vibration energy distribution map corresponding to the preliminary identification event, and performing an energy compliance check on the energy features based on a preset energy transition feature associated with the preliminary identification event; and determining that the match is successful when the results of the logical validity check, the time compliance check, and the energy compliance check are all passed.

[0048] In one specific embodiment, logical validity verification serves as the first layer of filtering, its core function being to ensure that any preliminary identification event must conform to the logical context of the current vehicle state. The system first reads the current state of the digital twin model corresponding to the target parking space, such as idle, reserved, or occupied. Simultaneously, the system's embedded vehicle behavior logic sequence (e.g., a finite state machine model) explicitly defines which subsequent events are legal and predictable in each state. For example, when the digital twin model is in the idle state, a legal preliminary identification event can only be a wheel crossing the line (entering); while a door opening or engine shutdown event is considered logically invalid in this state because there cannot be anyone inside the vehicle or the engine cannot be running.

[0049] Furthermore, events that pass the logical verification will proceed to the time compliance verification stage. This step aims to confirm whether the time of the event's occurrence falls within the expected causal chain. The system extracts the precise occurrence time of the initially identified event and reads the start time of the current state from the digital twin model, subtracting the two to obtain a time difference. For each valid state-event combination, the system pre-defines a reasonable time window, which defines the normal time range for the event to occur (e.g., a minimum and a maximum value). For example, when a vehicle transitions from the entering state to the attitude adjustment state, the initial event of vehicle translation should typically occur within 1 to 5 seconds after the start of the entering action. If a vehicle translation event occurs outside this time window, such as 30 seconds after the start of the entering action, the system will determine that its timing is inconsistent, considering it potentially an isolated disturbance unrelated to the current parking situation.

[0050] Furthermore, the system retrieves the spatiotemporal vibration energy distribution map corresponding to the moment the initially identified event occurred, and extracts key energy features, such as the peak value of the total energy, the geometric shape of the energy distribution (point, line, or area), and the rate of energy change over time (i.e., transition features). These features are compared with a pre-calibrated library of preset energy transition feature templates associated with the event type. For example, the energy template for a car door closing event should be a short-duration, high-amplitude local energy pulse, while the energy template for an engine idling event should be a continuous, low-amplitude, globally stable vibration. Only when the similarity between the extracted energy features and the preset templates exceeds a certain threshold is the energy conformity check considered passed. This greatly enhances the ability to distinguish subtle differences in event types and prevents vibrations with inconsistent physical characteristics from being misjudged as the target event.

[0051] In other embodiments, instead of the aforementioned vibration sensor array-based approach, a multi-sensor data fusion-based approach can be used. This approach utilizes various heterogeneous sensors (e.g., ultrasonic sensors, geomagnetic sensors, and video acquisition devices) deployed around the target parking space, and applies DS evidence theory to intelligently fuse multi-source data. This allows for highly reliable determination of the actual occupancy status of the parking space, thereby generating corresponding semantic events. Specifically, the method may include the following processing flow: First, the algorithm performs independent preprocessing and preliminary status determination on the raw data from each sensor. For the ultrasonic sensor installed above the parking space, its ranging result Dus can be calculated using the following formula (Equation 1): Dus = v × t / 2 - ε (Equation 1), where v is the real-time sound velocity (which can be compensated for based on temperature), t is the echo time, and ε is the system's inherent correction coefficient. By comparing Dus with a preset vacant parking space height threshold, a preliminary parking space occupancy determination can be obtained. For the geomagnetic sensor buried below the parking space, its geomagnetic disturbance determination result Smg can be obtained using the following formula (Equation 2): (Equation 2), where Bx, By, and Bz are the current measured values ​​of the three-axis geomagnetic field, Bx0, By0, and Bz0 are the initial reference values ​​when the parking space is vacant, and θmg is the geomagnetic disturbance judgment threshold. This formula determines whether the intensity of the magnetic field change caused by the vehicle's metal exceeds the threshold. For video acquisition devices, pre-trained deep learning object detection algorithms (such as YOLO or Faster R-CNN) can be used to process the video stream, directly outputting the judgment result Svideo of the parking space occupancy status and the corresponding confidence level Cvideo. After obtaining the preliminary judgment results of each single sensor, this embodiment uses DS evidence theory to perform multi-sensor data fusion. This process first establishes a recognition framework θ={A, ¬A}, where A represents the focal element "parking space occupied" and ¬A represents the focal element "parking space vacant". Then, the preliminary judgment results of each sensor are converted into their respective basic probability allocation functions (BPA or mass function). For example, for a video sensor, its BPA can be set as: m3(A) = Cvideo (when Svideo is occupied), m3(¬A) = Cvideo (when Svideo is idle), and m3(θ) = 1 - Cvideo represents uncertainty. The BPA of ultrasonic and geomagnetic sensors can be generated by a preset mapping function based on the degree to which their measured values ​​deviate from the threshold. Subsequently, the evidence from the three sensors is fused using the Dempster synthesis rule, where the confidence level m(A) of the synthesized parking space occupancy can be calculated by the following formula (Equation 3): (Equation 3), where m1, m2, and m3 are the basic probability assignment functions (BPA) for the ultrasonic, geomagnetic, and video sensors, respectively. Xi, Yj, and Zk are the focal elements corresponding to these three BPAs, i.e., all non-empty subsets of the recognition frame θ = {A, ¬A}. In this embodiment, the specific values ​​can be {A}, {¬A}, or {A, ¬A} (i.e., θ itself), where K is the conflict coefficient, derived from the formula... The calculation measures the degree of conflict between the various pieces of evidence. Finally, by comparing the synthesized confidence level m(A) or m(¬A) with the decision threshold θd, a high-confidence final decision is made, and this decision (e.g., occupied or idle) is used as a semantic event to drive the digital twin model to perform the corresponding state transition. To further improve the robustness of the system under different environmental conditions, this embodiment may also include an adaptive threshold calibration mechanism. For example, the decision threshold θmg of the geomagnetic sensor can be adjusted at each time step t using the formula... Dynamic updates are performed, where θ t-1 It is the threshold of the previous time step, α is a smoothing factor between 0 and 1, and E iThis consists of a set of sensor characteristic values ​​(such as geomagnetic disturbance intensity) of samples recently confirmed by the system as true negatives (i.e., confirmed vacancy), allowing the judgment threshold to automatically adapt to changes in the environmental background magnetic field. Through the above-mentioned multi-sensor data fusion and adaptive calibration method, this application can also achieve the technical effect of accurately and reliably analyzing the parking space occupancy status.

[0052] S205. Based on the semantic event, send a state update instruction to the digital twin model to drive the state of the digital twin model to change. The state includes idle state, reserved state, driven-in state, occupied state and driven-out state. Specifically, the state is a predefined finite set that can fully describe the life cycle of a parking space. In a preferred embodiment of this application, it may include, but is not limited to: idle state (indicating that there is currently no vehicle in the parking space and it is available for use), reserved state (indicating that the parking space has been reserved by a specific user through the platform and is in a reserved state), entering state (indicating that the system has detected that a vehicle is entering the parking space), occupied state (indicating that the vehicle has been stably parked in the parking space), and leaving state (indicating that the system has detected that the parked vehicle is leaving the parking space).

[0053] In a straightforward implementation, the system can send instructions directly to a specific endpoint of the digital twin platform via a synchronous API (Application Programming Interface). The request body includes the target twin ID and the new state (e.g., occupancy state). Upon receiving the request, the server immediately updates its backend database. As an alternative, the system can employ an asynchronous, decoupled architecture based on message queues. The event processing module publishes messages containing event information to specific topics, while the digital twin service, as a subscriber, autonomously performs state transitions upon receiving the messages. This publish / subscribe model enhances the system's resilience and scalability. Furthermore, in scenarios with higher requirements for data credibility and traceability, this state transition logic can be encapsulated in a blockchain smart contract. In this case, a semantic event triggers a transaction, calling the state update function in the contract. Once the transaction is confirmed by network consensus, the twin's state undergoes an immutable atomic update on the distributed ledger, providing a permanent audit record for all state changes.

[0054] S206. When the state of the digital twin model is updated from the reserved state to the driven state, a preset image acquisition device is triggered to obtain vehicle identity information, and based on the verification result of the vehicle identity information and the reserved vehicle information in the parking reservation request, the state of the digital twin model is updated to the occupied state and the occupancy start time is recorded. In one specific embodiment, the system is triggered when the digital twin model detects a vehicle entering (i.e., the status changes from reserved to entered). The system immediately sends a command to a pre-installed image acquisition device (typically a high-definition network camera) at or above the parking space entrance to capture an image of the vehicle's front or rear. After the image is acquired, the system uses an integrated automatic license plate recognition engine to perform optical character recognition processing, thereby extracting the vehicle's identity information, i.e., its license plate number string. Subsequently, the system rigorously compares the extracted license plate number with the reserved vehicle information recorded in the parking reservation request corresponding to the parking space. If the two match perfectly, the system determines that the verification is successful and immediately performs two key actions: first, it officially updates the status of the parking space's digital twin model to occupied; second, it records the current precise system timestamp as the start time of the parking occupancy, thus initiating the billing cycle.

[0055] In another alternative embodiment, the vehicle identity verification process is more intelligent and multi-dimensional. Upon triggering image acquisition in the entry state, the system not only attempts to identify the license plate number but also concurrently invokes a trained computer vision model (e.g., a convolutional neural network) to analyze other visual features of the vehicle, such as brand logo, model, and color. In this scheme, vehicle identity information is a feature vector containing multiple dimensions, including license plate, color, and model. The verification process correspondingly becomes a multi-factor compliance scoring mechanism: the system comprehensively compares the features identified in the image with the multi-dimensional information provided by the user during reservation (possibly through uploaded vehicle photos or manual selection). The verification result is no longer a simple yes / no, but a similarity score. Only when this comprehensive score exceeds a preset confidence threshold (e.g., the license plate partially matches but the color and model highly match), will the system determine successful verification, update the digital twin model to the occupied state, and record the start time.

[0056] To further improve the utilization efficiency of parking resources and establish a fair reservation mechanism, a preferred embodiment of this application also provides refined management of overdue scenarios in the reserved state. Specifically, when the digital twin model is initialized to the reserved state, the system not only records the reservation information but also associates it with the reservation start time in the reservation request. The system starts a timer bound to the digital twin model and presets a reservation retention period or grace period (e.g., 15 minutes). If, after adding the grace period to the reservation start time, the state of the digital twin model corresponding to the target parking space still has not changed from the reserved state to the entered state (i.e., the vibration sensor array has not detected a semantic event of vehicle entry), the system determines that a reservation breach event has occurred. In response to this event, the system will execute the preset timeout handling logic: First, it will automatically reset the status of the digital twin model from the reserved state to the idle state, thereby releasing the parking space resource in time and adding it back to the allocable resource pool for other users to reserve or use; Second, optionally, the system can generate a reservation penalty billing voucher for the defaulting user according to preset business rules to compensate for the opportunity cost caused by the invalid occupation of the parking space.

[0057] S207. When the state of the digital twin model is updated to the departure state, the parking duration is calculated and a billing voucher is generated based on the occupancy start time and the departure state end time.

[0058] In one specific implementation, the system initiates the billing process when the state of the digital twin model is updated to the departed state. The system precisely captures the moment when the vehicle completely leaves the parking space, causing the departed state to end and transition to the next state (such as the idle state), and identifies this as the "end time of the departed state." Subsequently, the system extracts the previously recorded "occupancy start time" and calculates the precise total parking duration based on these two time points. Based on this parking duration, the system can invoke a preset billing rule (such as time-based rates, tiered pricing, etc.) to calculate the parking fee payable. Finally, the system integrates key information such as the occupancy start time, the end time of the departed state, the parking duration, and the final fee into a structured billing voucher.

[0059] Optionally, calculating the parking duration and generating a billing voucher based on the occupancy start time and the departure end time includes: identifying a vehicle complete departure event from the semantic events that trigger the state update of the digital twin model to the departure state, and determining the occurrence time of the vehicle complete departure event as the end time; calculating the parking duration based on the occupancy start time and the end time, and calling a pre-designed billing rule engine to calculate the parking fee to be paid based on the parking duration and the time-based rate of the target parking space; integrating the occupancy start time, the departure end time, the parking duration, and the parking fee to be paid into structured billing data, and generating the billing voucher based on the structured billing data.

[0060] As a preferred refinement of the aforementioned billing process, this application provides a more precise definition of the method for determining the end time of the departure state to ensure absolute fairness and accuracy in billing. Specifically, the system does not simply use the moment when the digital twin model transitions from the departure state to the idle state as the end point, but rather delves into the underlying semantic events that trigger this state transition for analysis. From the data carrier of the semantic event that triggered the transition to the departure state, the system uses an event parsing engine to identify and extract a sub-event with clear physical meaning—the vehicle completely departing event. The determination of this event can be based on the fusion analysis of multiple sensor data. For example, in one implementation, the system continuously monitors visual sensor data above the parking space and underground vibration sensor data. When the visual algorithm confirms that the vehicle's geometric outline has completely moved out of the preset virtual parking space frame, and the vibration sensor reading remains below the static environmental noise threshold for a preset time window (e.g., 3 seconds), the system determines that the vehicle has completely departed event. The high-precision system timestamp at the time of this event is then determined as the final end time of this billing.

[0061] After accurately determining the start and end times of parking, the system initiates an automated fee calculation process. First, the billing module calculates the total parking duration, for example, in seconds, based on the recorded start time and the precise end time determined in the previous step. Next, instead of directly multiplying the duration by a fixed price, the system invokes a pre-configured, highly flexible billing rule engine. This engine is a standalone software module or microservice whose core function is to dynamically apply complex billing strategies based on the input parking duration, parking space information, and current time. For example, the engine first queries the backend database for the time-based rate policy for the area to which the parking space belongs, based on the target parking space ID associated with the billing voucher. This policy can define very complex rules, such as 10 yuan per hour from 8 am to 10 pm on weekdays; 2 yuan per hour at other times; free for the first 30 minutes; a maximum daily charge of 80 yuan; or a flat rate of 5 yuan per hour on weekends and holidays. The billing rules engine parses these rules, precisely divides the total parking time into different rate periods, calculates the fees for each period, and then adds them up to arrive at the final parking fee to be paid.

[0062] Furthermore, once the parking fee to be paid is calculated, the system enters the final stage of generating and solidifying the billing voucher. The core task of this step is to integrate all the key information of this parking service into a standardized, structured billing data. In a preferred embodiment, this structured billing data is represented as a data object in JSON or XML format. This object clearly contains key fields such as a unique bill ID, parking space number, identified license plate number, occupancy start time accurate to the second or millisecond, departure end time, parking duration expressed in a human-readable format (e.g., XX hours XX minutes) and standard units (e.g., seconds), a description of the applied billing rules, and the parking fee to be paid, including the currency unit and numerical value. Subsequently, the system formally generates the billing voucher based on this complete and formatted structured billing data. There are several ways to implement this generation process: one way is to create a new, unmodifiable order record in a local or cloud-based transaction database; another way is to call an external financial or ticketing system API, pass in the bill data as a parameter, and have the external system generate an electronic invoice that complies with tax regulations; yet another way is to record the hash value of this bill data on the blockchain, forming an immutable public certificate.

[0063] Figure 3 This is another flowchart illustrating an IoT-based shared parking space management method disclosed in an embodiment of this application, as shown below. Figure 3 As shown, the method includes the following steps: S301. In response to receiving an appeal request for the billing voucher, retrieve the occupancy start time, the departure end time, the vehicle identity information associated with the billing voucher, and the spatiotemporal vibration energy distribution map sequence generated between the occupancy start time and the departure end time. Specifically, the system receives an appeal request for a specific generated billing voucher. This request can be received in several ways: First, the user, through a mobile application linked to their account, clicks the "Disagree with Bill" or "Appeal" button on the parking bill details page. This action generates an appeal request containing a specific billing voucher ID and sends it to the backend server via API. Second, the user submits an appeal through offline customer service channels (such as telephone or service desk). Customer service personnel manually trigger the appeal process after finding the corresponding billing voucher in the backend management system. Regardless of the method, once the system recognizes the appeal request, it immediately initiates a comprehensive evidence retrieval process, indexed by the unique identifier of the billing voucher (e.g., bill ID or transaction serial number). The core task of this process is to aggregate all records related to the parking event from different data storage layers.

[0064] The evidence retrieval process can be further divided into two parallel sub-processes. First, the system initiates a query to the main business database or transaction database. This query uses the billing voucher ID as the primary key to quickly and accurately retrieve the core structured billing information, including the start time of occupancy, the end time of departure, and vehicle identification information (such as license plate number) determined by the vehicle identification system. This information forms the basis of the billing voucher and is the primary object of appeal review. Simultaneously, or immediately following, the system performs a more complex data retrieval operation, namely, retrieving the original sensor evidence—the spatiotemporal vibration energy distribution map sequence. Since this data sequence is typically large, it is not stored in the main business database. A preferred storage and retrieval scheme is to store these distribution maps as files with timestamps and billing voucher IDs (such as binary files, image files, or serialized data objects) in a high-capacity object storage service (such as AWS S3). During retrieval, the system uses the billing credential ID as the path or tag, and the aforementioned start time of occupancy and end time of departure as time range filters to batch retrieve all distribution map files generated within this time period. Alternatively, this distribution map data can be streamed in real-time to a dedicated time-series database (such as InfluxDB) and tagged with the billing credential ID. During retrieval, the system initiates a range query request to this database to obtain all data points with the specified ID within a specific time window.

[0065] S302. Based on the occupancy start time, match and extract the first spatiotemporal vibration energy distribution map from the spatiotemporal vibration energy distribution map sequence, and construct the occupancy start time, the vehicle identity information and the first spatiotemporal vibration energy distribution map into a first event evidence group; In this embodiment, the system will perform precise matching and extraction within the retrieved spatiotemporal vibration energy distribution map sequence based on the occupancy start time obtained from the billing voucher, in order to locate the original evidence representing the key action of the vehicle's first entry into the parking space, namely the first spatiotemporal vibration energy distribution map. This matching process can be implemented in several ways. One implementation is based on timestamp-based nearest neighbor matching: the system uses the high-precision timestamp of the occupancy start time as a benchmark, traverses each map in the distribution map sequence, calculates the time difference between its own generation timestamp and the benchmark time, and selects the map with the smallest absolute time difference as the first spatiotemporal vibration energy distribution map. To improve accuracy, a very small time tolerance window (e.g., ±100 milliseconds) can be set, prioritizing matching within this window.

[0066] Furthermore, upon successfully extracting the first spatiotemporal vibration energy distribution map, the system will immediately proceed to the next step: logically binding it with other key information to construct a clearly structured and complete set of first event evidence. The essence of this construction process is to create a standardized data object to present the complete and irrefutable picture of the vehicle entry event. Specifically, the system will create a data structure containing three core elements: first, the start time of the occupancy as the timestamp of the event; second, the vehicle identification information (such as the license plate number) retrieved from the fare collection voucher as the subject of the event; and third, the newly extracted first spatiotemporal vibration energy distribution map (which can be a Base64 encoded string of an image or a secure URL (Uniform Resource Locator) pointing to the image file) as physical evidence of the event.

[0067] S303. Based on the end time of the departure state, match and extract the second spatiotemporal vibration energy distribution map from the spatiotemporal vibration energy distribution map sequence, and construct the end time of the departure state, the vehicle identity information, and the second spatiotemporal vibration energy distribution map into a second event evidence group. In this embodiment, the system matches and extracts data from the same spatiotemporal vibration energy distribution map sequence retrieved in step S301 based on the end time of the departure state obtained from the billing voucher, which serves as the billing endpoint, to obtain a second spatiotemporal vibration energy distribution map that represents the complete departure of the vehicle from the parking space. This matching and extraction process can also be implemented using various specific technical solutions. The first solution is approximate matching based on timestamps: the system uses the end time of the departure state as a query benchmark, traverses the distribution map sequence, and locates the distribution map that is closest in time by calculating the difference between the timestamp of each map generation and the benchmark time. This solution is simple and direct, and can effectively reflect the physical state at the moment the vehicle leaves. A second, more preferred solution is precise retrieval based on event tags: in the initial stage of parking space status monitoring, when the system determines that the parking space status has changed from a departing state to an vacant state, the generated distribution map will be automatically appended with a specific event tag, such as "event_type": "exit_complete" or "state_transition": "leaving_to_idle". Therefore, during the appeal process, the system can directly retrieve this specific tag from the metadata of the graph sequence, thereby finding the "second spatiotemporal vibration energy distribution map" in one step without ambiguity, which is more efficient and accurate.

[0068] After successfully locating and extracting the second spatiotemporal vibration energy distribution map using any of the above methods, the system immediately performs a construction operation, packaging it with other relevant information to form a structured second event evidence set. This construction process maintains consistency and symmetry with the process of generating the first event evidence set in step S302, and its purpose is to create a data object that can fully present the entire picture of the vehicle departure event. Specifically, the system generates a new data structure and encapsulates the following three core elements: First, the end time of the departure state is used as the authoritative timestamp of the event; second, the same vehicle identity information (such as license plate number) is used as the subject of the event, which further confirms that the departing vehicle and the entering vehicle are the same subject; third, the newly extracted second spatiotemporal vibration energy distribution map is used as the physical layer original evidence of the event.

[0069] S304. Based on the first event evidence group and the second event evidence group, generate an event retrospective report.

[0070] The specific implementation methods for generating this event retrospective report can be diverse to adapt to different application scenarios and display requirements. The first and most common implementation is for the system to generate a dynamic webpage or a specific mobile application interface. This interface typically uses a contrasting layout, such as left-right or top-bottom columns. One side is clearly marked as entry evidence, clearly listing all the contents of the first event evidence group: the start time of occupancy, vehicle identification information, and, most importantly, rendering the included first spatiotemporal vibration energy distribution map as a visual heatmap or pseudo-color image. The other side is symmetrically marked as departure evidence, displaying all the contents of the second event evidence group, including the rendered second spatiotemporal vibration energy distribution map. A second alternative implementation is for the system to generate a standard format document file, such as a PDF report. This approach is particularly suitable for formal occasions where evidence needs to be archived, sent via email, or printed. In this approach, the system calls a document generation library and fills in the content according to a preset template. The first page of the report can be a summary of the appeal (such as the bill number and appeal time), and subsequent pages detail the entry and departure events respectively. Each page contains the event timestamp, vehicle information, and an image generated from spatiotemporal vibration energy distribution data, along with a legend (e.g., the range of energy intensity represented by the color).

[0071] This embodiment also discloses an electronic device, as shown in the reference. Figure 4 The electronic device may include: at least one processor 401, at least one communication bus 402, a user interface 403, a network interface 404, and at least one memory 405. The communication bus 402 is used to enable communication between these components. The user interface 403 may include a display screen or a camera; optionally, the user interface 403 may also include a standard wired interface or a wireless interface. The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0072] The processor 401 may include one or more processing cores. The processor 401 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 405, and by calling data stored in memory 405. Optionally, the processor 401 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 401 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem.

[0073] In some embodiments of this application, a computer-readable storage medium is provided, including instructions that, when executed on an electronic device, cause the electronic device to perform an Internet of Things-based shared parking space management method according to an embodiment of this application.

[0074] In some embodiments of this application, a computer program product is also provided, which, when run on an electronic device, causes the electronic device to execute an Internet of Things-based shared parking space management method according to an embodiment of this application.

[0075] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

Claims

1. A method for managing shared parking spaces based on Internet of Things, characterized in that, Applied to a server, the method comprises: assigning a target parking space for a received parking reservation request, creating a digital twin model bound to the target parking space, initializing the digital twin model to a reserved state; collecting a multi-channel time-series vibration signal covering the target parking space area in real time through a vibration sensor array deployed at the target parking space; generating a time-space vibration energy distribution map by jointly analyzing the time-frequency of the multi-channel time-series vibration signal; performing sequence pattern recognition on the time-space vibration energy distribution map to analyze a semantic event matching a preset vehicle behavior logic sequence; sending a state update instruction to the digital twin model based on the semantic event to drive the state of the digital twin model to change, the state including an idle state, a reserved state, a driving-in state, an occupied state and a driving-off state; when the state of the digital twin model is updated from the reserved state to the driving-in state, triggering a preset image acquisition device to obtain vehicle identity information, and based on the verification result of the vehicle identity information and the reservation vehicle information in the parking reservation request, updating the state of the digital twin model to the occupied state and recording the starting time of occupation; when the state of the digital twin model is updated to the driving-off state, calculating the parking duration and generating a billing voucher according to the starting time of occupation and the end time of the driving-off state.

2. The method of claim 1, wherein, The method for assigning a target parking space for a received parking reservation request, creating a digital twin model bound to the target parking space, and initializing the digital twin model to a reserved state comprises: According to the health status of the vibration sensor array associated with each parking space in the preset parking space resource pool and the geographical location information of each parking space, a candidate parking space set meeting the preset condition is selected; According to the user preference or historical parking data in the parking reservation request, the target parking space is matched from the candidate parking space set for the parking reservation request; Based on the preset digital twin model template, the static attributes of the target parking space, the data interface identifier of the vibration sensor array and the reservation information of the parking reservation request are filled into the digital twin model template to create the digital twin model; The internal state variable of the digital twin model is set to the reserved state.

3. The method of claim 1, wherein, The method for generating a time-space vibration energy distribution map by jointly analyzing the time-frequency of the multi-channel time-series vibration signal comprises: processing each time-series vibration signal in the multi-channel time-series vibration signal through a preset wavelet transform algorithm to obtain a set of time-frequency spectrum graphs; For each time-frequency spectrum graph in the set of time-frequency spectrum graphs, a plurality of energy characteristic values are calculated in a preset time-frequency analysis window; Combining each energy characteristic value with the preset spatial coordinates of the corresponding vibration sensor to obtain a spatial energy data point; Based on the spatial energy data point, a preset spatial interpolation algorithm is used to calculate the energy value for each grid point in the preset two-dimensional spatial grid to generate the time-space vibration energy distribution map.

4. The method of claim 1, wherein, The sequence pattern recognition on the spatio-temporal vibration energy distribution map includes: extracting a multi-dimensional spatio-temporal feature vector from a time sequence of the spatio-temporal vibration energy distribution map; inputting the multi-dimensional spatio-temporal feature vector into a preset recurrent neural network model to generate a probability distribution of the multi-dimensional spatio-temporal feature vector belonging to each preset event category, the preset event category being a classification target determined when the recurrent neural network model is trained; determining, according to the probability distribution, a preset event category corresponding to a highest probability value as a preliminary identified event; matching the preliminary identified event with the preset vehicle behavior logic sequence, and determining the preliminary identified event as the semantic event when the matching is successful.

5. The method of claim 4, wherein, The matching of the preliminary identified event with the preset vehicle behavior logic sequence, and the determination of the preliminary identified event as the semantic event when the matching is successful include: performing a logic validity check on the preliminary identified event according to the preset vehicle behavior logic sequence and a current state of the digital twin model; calculating a time difference based on a time of occurrence of the preliminary identified event and a starting time of the current state, and performing a time compliance check on the time difference according to a preset time window associated with the preliminary identified event; extracting an energy feature of the spatio-temporal vibration energy distribution map corresponding to the preliminary identified event, and performing an energy compliance check on the energy feature according to a preset energy transition feature associated with the preliminary identified event; determining that the matching is successful when the results of the logic validity check, the time compliance check, and the energy compliance check are all passed.

6. The method of claim 1, wherein, The calculation of the parking duration and the generation of the billing voucher according to the occupancy starting time and the ending time of the leaving state include: identifying a vehicle completely leaving event from the semantic event that triggers the state update of the digital twin model to the leaving state, and determining a time of occurrence of the vehicle completely leaving event as the ending time; calculating a parking duration according to the occupancy starting time and the ending time, and calling a preset billing rule engine to calculate a parking fee to be paid based on the parking duration and a time period rate of the target parking space; integrating the occupancy starting time, the ending time of the leaving state, the parking duration, and the parking fee to be paid into structured billing data, and generating the billing voucher based on the structured billing data.

7. The method of claim 6, wherein, The method further includes: in response to receiving a complaint request for the billing voucher, calling the occupancy starting time, the ending time of the leaving state, the vehicle identity information, and the sequence of the spatio-temporal vibration energy distribution maps generated between the occupancy starting time and the ending time associated with the billing voucher. According to the occupancy starting moment, a first spatio-temporal vibration energy distribution map is matched and extracted from the sequence of spatio-temporal vibration energy distribution maps, and the occupancy starting moment, the vehicle identity information and the first spatio-temporal vibration energy distribution map are configured into a first event evidence group; According to the end moment of the driving-off state, a second spatio-temporal vibration energy distribution map is matched and extracted from the sequence of spatio-temporal vibration energy distribution maps, and the end moment of the driving-off state, the vehicle identity information and the second spatio-temporal vibration energy distribution map are configured into a second event evidence group; Based on the first event evidence group and the second event evidence group, an event backtracking report is generated.

8. An electronic device, comprising: The electronic device comprises a processor, a memory, a user interface and a network interface, the memory is configured to store instructions, the user interface and the network interface are configured to communicate with other devices, and the processor is configured to execute the instructions stored in the memory to enable the electronic device to perform the method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores instructions which, when executed, perform the method according to any one of claims 1-7.

10. A computer program product, characterised in that, The computer program product, when running on the electronic device, enables the electronic device to perform the method according to any one of claims 1-7.

Images