An Internet of Things-based Urban Traffic Collaborative Intelligent Parking Management System

Abstract

This invention relates to the field of intelligent transportation technology and discloses an IoT-based urban traffic collaborative intelligent parking management system. The system includes a data acquisition and preprocessing module that captures vehicle parking behavior and parking space status data in real time, generating a standardized parking event stream through time series alignment and noise filtering; a feature extraction and anomaly detection module that extracts features of parking space occupancy duration fluctuations and vehicle parking frequency characteristics to determine the degree of abnormal deviation in parking behavior; a congestion prediction and modeling module that constructs a node congestion correlation model to predict potential traffic resistance in the surrounding road network; a resource allocation and strategy generation module that formulates a dynamic reallocation strategy for parking space resources based on potential traffic resistance, calculates a data migration urgency score, and generates an isolation configuration scheme that includes path planning and permission adjustment; and an execution and feedback monitoring module that continuously monitors changes in parking space usage patterns and locates abnormal behavior sources by comparing differences in access sequences before and after configuration.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to an Internet of Things-based urban traffic collaborative intelligent parking management system. Background Technology

[0002] Current urban parking management systems primarily employ fixed-point detection and independent guidance. Existing technologies for monitoring parking behavior largely rely on single sensors, failing to effectively correlate vehicle parking data with changes in parking space status. Data processing methods are simplistic, lacking sufficient time synchronization accuracy for multi-source heterogeneous data, and noise interference impacts data quality. Feature extraction dimensions are limited, and the temporal and frequency characteristics of parking behavior are not fully explored. Anomaly detection mechanisms are rigid, often using fixed thresholds, unable to adapt to dynamic changes in parking behavior across different areas. Congestion prediction models are linear, and the intrinsic correlation between parking anomalies and traffic flow is not deeply analyzed. Resource allocation strategies are static, lacking real-time traffic status support for parking space resource adjustments. Existing methods need to address key technical challenges such as multi-source data fusion, behavior pattern recognition, congestion correlation prediction, and dynamic resource allocation.

[0003] Traditional parking management systems suffer from significant shortcomings in collaborative management and intelligent decision-making. Sensor network coverage is incomplete, with data collection in key areas lacking. Time series alignment algorithms are simplistic, leading to large synchronization errors across different sampling frequencies. Fixed noise filtering parameters result in unsatisfactory processing of non-stationary signals. Feature extraction methods are limited, failing to capture the periodic and sudden characteristics of parking behavior. Historical behavior feature databases are outdated, failing to reflect the latest changes in parking demand. Anomaly deviation calculations are linearized, resulting in low accuracy in identifying complex anomaly patterns. Node positioning accuracy is insufficient, leading to inaccurate mapping between physical locations and logical regions. Congestion correlation models are idealized, neglecting the impact of road topology and signal control. Traffic resistance predictions have large biases, and the reliability of potential congestion assessments is insufficient. Resource allocation algorithms are inefficient, resulting in delayed response times for large-scale node scheduling. Migration urgency scoring criteria are subjective, and priority settings for important nodes are unreasonable. Isolation configuration scheme generation is rigid, leading to poor coordination between path planning and permission adjustments. A lack of execution effect monitoring mechanisms results in an incomplete feedback optimization loop for configuration adjustments. Summary of the Invention

[0004] The purpose of this invention is to provide an Internet of Things-based urban traffic collaborative intelligent parking management system to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an Internet of Things-based urban traffic cooperative intelligent parking management system, the system comprising:

[0006] The data acquisition and preprocessing module is used to capture vehicle parking behavior data and parking space status change signals in real time through a group of sensor devices deployed in urban parking areas, and to perform time series alignment and noise filtering on the captured multi-source data to generate a standardized parking event stream.

[0007] The feature extraction and anomaly detection module is used to perform pattern parsing on the standardized parking event stream, extract the fluctuation features of parking space occupancy time and vehicle parking frequency features, and determine the degree of abnormal deviation of the current area parking behavior by comparing with the historical normal behavior feature library.

[0008] The congestion prediction and modeling module is used to locate the physical nodes where high-frequency abnormal parking behavior occurs based on the abnormal deviation, and to build a node congestion correlation model by combining real-time traffic flow data to predict the potential traffic resistance of the road network around the node.

[0009] The resource allocation and strategy generation module is used to formulate a dynamic reallocation strategy for parking space resources based on the potential traffic resistance, calculate the data migration urgency score of different parking space nodes, and generate an isolation configuration scheme that includes path planning and permission adjustment instructions.

[0010] The execution and feedback monitoring module is used to continuously monitor changes in parking space usage patterns after the isolation configuration scheme is executed, and to locate abnormal behavior sources by comparing the differences in access sequences before and after the configuration takes effect.

[0011] Preferably, the time series alignment and noise filtering processing of the captured multi-source data includes: receiving raw data streams with inconsistent timestamps from geomagnetic sensors, cameras, and payment terminals; segmenting and buffering the data streams using a sliding window mechanism; performing timestamp normalization processing on the data within each segment based on the GPS clock signal; applying a variance-based outlier detection algorithm to remove transient noise data caused by equipment failure; and extracting feature vectors from the processed data segments, wherein the feature vectors include the vehicle parking start time, duration, and interval between adjacent events.

[0012] Preferably, the extraction of parking space occupancy duration fluctuation features and vehicle parking frequency features includes: statistically analyzing the total occupancy duration sequence of a single parking space within a preset time period from the normalized parking event stream, calculating the variance and autocorrelation coefficient of the total occupancy duration sequence to characterize duration fluctuation; statistically analyzing the vehicle parking frequency sequence within the same period, calculating its coefficient of variation to characterize frequency stability; combining the duration fluctuation and frequency stability into a multi-dimensional feature vector, inputting it into a pre-trained autoencoder for feature dimensionality reduction, and obtaining low-dimensional dense features.

[0013] Preferably, the step of determining the abnormal deviation of the current area parking behavior by comparing with the historical normal behavior feature library includes: calling the historical normal parking behavior feature template stored in the distributed database, the template containing the baseline feature range of different time periods and different parking space types; calculating the Mahalanobis distance between the currently extracted low-dimensional dense feature and the corresponding template baseline feature, and mapping the distance value to a deviation score from zero to one hundred; when the deviation score exceeds the dynamic threshold, marking the parking space node as an abnormal deviation, and recording the abnormal time window.

[0014] Preferably, the construction of the node congestion correlation model includes: acquiring real-time traffic flow data provided by the municipal traffic management platform, establishing a road network grid with abnormal parking nodes as the center and a radiation radius of 500 meters; analyzing the average vehicle speed change rate and traffic flow density increment of each road segment within the grid during the abnormal time window; using a graph convolutional network to model the nonlinear relationship between the abnormality degree of parking nodes and the vehicle speed change rate of surrounding road segments, and outputting the influence weight of each node on the road network traffic resistance.

[0015] Preferably, the calculation of the data migration urgency score for different parking space nodes includes: determining the set of parking space nodes that need to be adjusted based on the influence weights; analyzing the amount of parking order data stored in the nodes within the set and the level of user sensitivity information; assessing the strength of data dependencies between nodes, with the dependency strength quantified by the weighted sum of the cross-node order association ratio and the synchronous access frequency; and combining the influence weights, data volume, and dependency strength, using a multilayer perceptron model to calculate the migration urgency score for each node, with a higher score indicating that it needs to be processed first.

[0016] Preferably, the step of generating an isolation configuration scheme that includes path planning and permission adjustment instructions includes: sorting nodes according to migration urgency scores and selecting the node with the highest score as the source node to be migrated; selecting a target storage node cluster in a low-load area, calculating the optimal data transmission path from the source node to the target cluster based on the Dijkstra algorithm, wherein the path must avoid the current network congestion links; and generating an isolation configuration scheme that includes the source node identifier, target cluster address, migration data volume, transmission path sequence, and execution time window.

[0017] Preferably, the continuous monitoring of parking space usage pattern changes after executing the isolation configuration scheme includes: collecting usage logs of affected parking spaces during the monitoring period after data migration is completed, extracting access time distribution, single access duration and concurrent access count; dynamically time-aligning the access pattern characteristics after migration with the baseline pattern of the same period before migration, and calculating the cumulative difference between the two at key pattern points.

[0018] Preferably, the step of locating the abnormal behavior source by comparing the access sequence differences before and after the configuration takes effect includes: when the cumulative difference value exceeds the adaptive threshold, tracing back the access event sequence with significant differences, parsing the operation source device number and user identifier; associating the access records of other adjacent nodes within the same time window to identify whether there is a collaborative abnormal pattern; and performing pattern matching between the abnormal access sequence and the system vulnerability feature library to determine the type of behavior source that caused the abnormality.

[0019] Preferably, the system further includes an adaptive response mechanism: automatically triggering a predefined response script based on the determined behavior source type. The script includes, but is not limited to, temporarily locking the parking space reservation permissions associated with the abnormal source, adjusting the parking space pricing coefficient in the abnormal area, sending a detailed diagnostic report to the operation and maintenance terminal, and prompting manual review of key decision points.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] The feature extraction and anomaly detection module performs pattern analysis on standardized parking event flows, extracting parking space occupancy duration fluctuations and vehicle parking frequency characteristics. It then compares these characteristics with a historical normal behavior feature database to determine the degree of deviation from current parking behavior. Pattern analysis employs time series analysis to identify the periodic, trend, and random components of parking events. Parking space occupancy duration fluctuations are calculated using statistical distribution and coefficient of variation to reflect the temporal regularity of parking space use. Vehicle parking frequency characteristics statistically analyze the number of parking events per unit time, characterizing the intensity of parking demand in the area. The historical normal behavior feature database is generated through machine learning algorithms, establishing standard behavioral patterns for different time periods and areas. Anomaly deviation calculation uses Mahalanobis distance or dynamic time warping algorithms to quantify the degree of difference between current behavior and historical patterns. The deviation threshold is dynamically adjusted based on regional characteristics to improve detection accuracy. The feature database supports online updates to adapt to the long-term evolution of urban parking patterns.

[0022] The congestion prediction and modeling module locates physical nodes where high-frequency abnormal parking behavior occurs based on anomaly deviation. It then constructs a node congestion correlation model using real-time traffic flow data to predict the potential traffic resistance of the surrounding road network. Node localization is achieved through a geographic information system, accurately mapping the spatial location of anomalies. Real-time traffic flow data comes from multiple sources, including geomagnetic coils, video detection, and floating car sensors. The congestion correlation model employs a graph neural network structure, where nodes represent road intersections and edges represent road segment connections. Model training considers historical congestion propagation patterns and real-time traffic conditions to improve prediction accuracy. Traffic resistance calculation is based on multiple dimensions, including travel time, vehicle speed, and queue length. Resistance prediction differentiates between different time periods and weather conditions, enabling personalized assessments. Prediction results are visualized to assist traffic management personnel in decision-making.

[0023] The resource allocation and strategy generation module formulates a dynamic reallocation strategy for parking space resources based on potential traffic resistance, calculates the data migration urgency score for different parking space nodes, and generates an isolation configuration scheme that includes path planning and permission adjustment instructions. The dynamic reallocation strategy employs an optimization algorithm to minimize traffic impact while meeting parking demand. The migration urgency score considers node importance, anomaly severity, and the scope of traffic impact. The score calculation uses a multi-attribute decision-making method to balance the weights of various indicators. The path planning algorithm considers real-time traffic conditions to provide vehicles with the optimal driving route. Permission adjustment instructions control parking space usage permissions, limiting the spread of abnormal parking behavior. The isolation configuration scheme includes detailed parameters such as execution time, scope of effect, and recovery conditions. The scheme generation process is automated and supports batch processing of large-scale nodes.

[0024] After executing the isolation configuration scheme, the execution and feedback monitoring module continuously monitors changes in parking space usage patterns, locating abnormal behavior sources by comparing access sequence differences before and after the configuration takes effect. The monitoring process employs real-time stream processing technology to track changes in parking space status. Usage pattern analysis includes key indicators such as parking duration distribution, turnover rate, and occupancy rate. Access sequence difference detection uses sequence alignment algorithms to identify abnormal access patterns. Abnormal behavior source localization is achieved through source tracing analysis to determine the origin of abnormal parking behavior. Feedback data records the configuration execution effect, including the degree of traffic improvement and user satisfaction. Monitoring results are used to optimize model parameters, forming a closed-loop control. Through the synergistic effect of behavioral feature extraction, abnormal deviation judgment, congestion correlation prediction, and dynamic resource allocation, the coordinated optimization of urban parking and transportation systems is achieved. Multi-source data provides the basis for decision-making, behavioral analysis identifies abnormal patterns, congestion prediction assesses traffic impact, and resource allocation enables precise management. This integrated approach significantly improves the intelligence level of urban parking management and traffic operation efficiency. Attached Figure Description

[0025] Figure 1 This is a schematic diagram illustrating the working principle of the Internet of Things-based urban traffic collaborative intelligent parking management system described in this invention.

[0026] Figure 2 A flowchart for multi-source data time series alignment and noise filtering;

[0027] Figure 3 A flowchart for determining abnormal deviations in parking behavior;

[0028] Figure 4 A graph showing the correlation between the impact weight of abnormal nodes and road network and traffic flow density;

[0029] Figure 5 Heatmap distribution of parking space node data migration urgency rating. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Please see Figure 1 This invention provides an IoT-based urban traffic collaborative intelligent parking management system. The system includes: a data acquisition and preprocessing module, a feature extraction and anomaly detection module, a congestion prediction and modeling module, a resource allocation and strategy generation module, and an execution and feedback monitoring module. The data acquisition and preprocessing module captures vehicle parking behavior data and parking space status change signals in real time through a cluster of sensors deployed in urban parking areas. It performs time-series alignment and noise filtering on the captured multi-source data to generate a standardized parking event stream. The feature extraction and anomaly detection module performs pattern analysis on the standardized parking event stream, extracting parking space occupancy duration fluctuation features and vehicle parking frequency features. It also determines the degree of abnormal deviation in current area parking behavior by comparing it with a historical normal behavior feature database. The congestion prediction and modeling module locates the physical nodes where high-frequency abnormal parking behavior occurs based on the abnormal deviation degree. It constructs a node congestion correlation model based on real-time traffic flow data to predict the potential traffic resistance of the surrounding road network. The resource allocation and strategy generation module formulates a dynamic reallocation strategy for parking space resources based on potential traffic resistance, calculates the data migration urgency score for different parking space nodes, and generates an isolation configuration scheme containing path planning and permission adjustment instructions. After the execution and feedback monitoring module executes the isolation configuration scheme, it continuously monitors changes in parking space usage patterns and identifies abnormal behavior sources by comparing the differences in access sequences before and after the configuration takes effect.

[0032] Example 1: See Figure 2In practical implementation, the data acquisition and preprocessing module is responsible for processing raw data streams from various sensing devices deployed in urban parking areas, including geomagnetic sensors, cameras, and payment terminals. Due to differences in their operating principles and internal clocks, these devices generate raw data streams with inconsistent timestamps. The data acquisition and preprocessing module needs to receive this data and standardize it. The processing employs a sliding window mechanism to segment and buffer continuously arriving data streams. Each sliding window covers a data sample within a preset duration. Using a high-precision GPS clock signal as a unified benchmark, the data buffered within the sliding window undergoes timestamp normalization, thereby eliminating time deviations between different source data. Based on time series alignment, a variance-based outlier detection algorithm is applied to scan the normalized data segments. This algorithm calculates the variance of data points within the sliding window and identifies transient data points that significantly deviate from the main variance range. These points typically originate from momentary equipment failures or external interference and are judged as noise data and discarded. It is understandable that data segments that have undergone timestamp normalization and noise filtering have higher consistency and reliability. Subsequently, core feature vectors describing parking events are extracted from each purified data segment. The feature vectors include the vehicle parking start time, the vehicle parking duration, and the interval between adjacent parking events.

[0033] The feature extraction and anomaly detection module performs deep pattern analysis on the standardized parking event stream generated after the preprocessing. Its core task is to extract features that reflect parking behavior patterns. Specifically, from the standardized parking event stream after time series alignment and noise filtering, for each parking space entity, the module statistically analyzes its total occupancy duration within a preset time period. It calculates the variance of this total occupancy duration sequence to quantify its dispersion and its autocorrelation coefficient to characterize the correlation of the sequence at different time lags. The variance and autocorrelation coefficient together constitute an indicator of the volatility of parking space occupancy duration. Simultaneously, it statistically analyzes the sequence of vehicle stops at the same parking space within the same preset time period and calculates the coefficient of variation (the ratio of standard deviation to mean) of this sequence to characterize the stability of vehicle stop frequency. It can be understood that combining the calculated duration volatility indicator and frequency stability indicator forms a multi-dimensional feature vector. While this vector contains important information, it may suffer from redundancy and high dimensionality. To this end, multidimensional feature vectors are input into a pre-trained autoencoder model. The autoencoder compresses the high-dimensional input data into low-dimensional dense features through the encoder part, and then attempts to reconstruct the original input through the decoder part. The training objective is to minimize the reconstruction error, thereby ensuring that the low-dimensional dense features can retain the main information of the original multidimensional feature vectors.

[0034] In some embodiments, the size of the sliding window needs to balance real-time requirements and data stability. A window that is too small may cause statistical features to be overly sensitive to instantaneous fluctuations, while a window that is too large will introduce response latency. Optionally, the duration of the sliding window can be set to five or ten minutes and dynamically adjusted according to the network conditions and computing resources of the actual application scenario. The specific implementation of the variance-based outlier detection algorithm involves calculating the variance of all data points within the current sliding window relative to the window mean, then setting a multiple threshold based on the variance, and marking data points whose deviation from the mean exceeds the threshold multiple of the variance as outliers and removing them.

[0035] Optionally, during the feature extraction stage, the selection of a preset time period is crucial for the effectiveness of the features. This period needs to match the tidal patterns of urban traffic or the periodicity of commercial activities. For example, the preset time period can be set to one hour, a specific time period within a day, or a specific number of days within a week to more accurately capture the cyclical patterns of parking behavior. The training of the autoencoder is an offline process, requiring the collection of historical parking data over a sufficiently long period, covering different seasons, weather conditions, weekdays, and holidays, to ensure that the learned low-dimensional dense features are broadly representative. In practical deployment, the trained autoencoder model parameters are embedded into the feature extraction and anomaly detection modules for online feature dimensionality reduction of the real-time data stream. Low-dimensional dense features, as direct input to subsequent anomaly detection algorithms, have a dimensionality far lower than the original multidimensional feature vectors. This not only reduces the computational burden but also often improves the accuracy of anomaly detection due to the removal of irrelevant noise.

[0036] Example 2: See Figure 3 In practical implementation, the feature extraction and anomaly detection module calls historical normal parking behavior feature templates stored in a distributed database. These templates are pre-established by analyzing long-term historical data and include baseline feature ranges for different time periods and parking space types. The Mahalanobis distance between the currently extracted low-dimensional dense features and the corresponding template baseline features is calculated. The Mahalanobis distance calculation considers the covariance structure of the feature vectors, effectively measuring the deviation of the current features from historical normal patterns in a multivariate space. The calculated Mahalanobis distance value is mapped to a deviation score range of zero to one hundred using a linear or non-linear scaling function, ensuring uniformity and comparability of the scores. When the deviation score exceeds a dynamic threshold, the system marks the parking space node as having an abnormal deviation state and accurately records the time window of the anomaly's occurrence and duration. The dynamic threshold is not a fixed value but is adaptively adjusted based on the distribution of deviation scores in historical data from the same period. For example, it can be set to a high-order value of the historical quantile to cope with inherent traffic flow fluctuations at different times.

[0037] After receiving anomaly deviation markers and anomaly time window information from the feature extraction and anomaly detection module, the congestion prediction and modeling module begins constructing a node congestion correlation model. This module obtains real-time traffic flow data from the municipal traffic management platform via a data interface. This data typically includes information such as average vehicle speed and traffic density. A geospatial road network grid with a radius of 500 meters is established, centered on the parking nodes marked as having anomaly deviations. This grid covers surrounding roads potentially affected by abnormal parking behavior. The module analyzes the rate of change of average vehicle speed and the increase in traffic density for each road segment within the grid during the recorded anomaly time window. The rate of change of average vehicle speed refers to the percentage change in the current average vehicle speed relative to the average vehicle speed during a baseline period before the anomaly occurred, while the increase in traffic density is the difference between the current density and the baseline density. A graph convolutional network (GCNN) is used to model the complex nonlinear relationship between the anomaly degree of parking nodes and the speed change rate of surrounding road segments. In this model, the road network mesh is abstracted as a graph structure, where nodes represent road intersections or key points of road segments, edges represent road links, and node features can include historical vehicle speed, real-time speed change rate, distance to the abnormal parking node, etc., while edge weights can reflect road grade or traffic capacity. The GCNN effectively captures spatial dependencies by iteratively transmitting and aggregating feature information among its neighboring nodes. The output of the GCNN model is the influence weight of each abnormal parking node on the traffic resistance of the surrounding road network. The influence weight quantifies the contribution of the abnormal parking behavior of the node to the smoothness of the road network. The calculation of the influence weight can be understood as being modeled based on the following relationship:

[0038]

[0039] in: Indicates the first The impact weight of each abnormal parking node This represents the activation function. This indicates the nodes in the road network diagram structure. A set of influential adjacent road segments or nodes. Represents a node With adjacent entities Topological distance or actual path distance between them Representing adjacent entities The abnormal deviation score or its derived features at the location. Representing adjacent entities The rate of change of average vehicle speed at that location It is the set of learnable parameters of the model, the function This represents the feature transformation function learned by the graph convolutional network.

[0040] In some embodiments, the establishment of historical normal parking behavior feature templates is a continuously updated process. The system periodically incorporates newly generated parking behavior data that has been verified to fall within the normal range into the template library and recalculates the baseline feature range, enabling the templates to adapt to the long-term evolution of urban traffic patterns. The radiation radius of the road network grid, 500 meters, is a configurable parameter that can be differentiated based on urban road density and regional functions. In areas with sparse road networks, the radiation radius can be appropriately expanded, while in areas with dense road networks, it can be appropriately reduced. It is understood that training the graph convolutional network requires a large amount of historical data as support. The training data should include records of various known abnormal parking events and their actual impact on traffic resistance in the surrounding road network. Through supervised learning, the model learns the mapping relationship from input features to output targets. The trained graph convolutional network model is deployed in the congestion prediction and modeling module to assess the impact of real-time abnormal parking events.

[0041] See Figure 4 This graph embodies the core feature association logic of the congestion prediction and modeling module: the number of abnormal nodes corresponds to the physical node scale of high-frequency abnormal parking behavior, which is a quantitative result of locating the physical nodes where high-frequency abnormal parking behavior occurs; the road network influence weight is the contribution of a node to the traffic resistance of the surrounding road network, matching the node influence weight index output by the graph convolutional network; and the traffic flow density is the real-time state parameter of the increase in traffic flow density of road segments within the road network grid. The correlation of these indicators is the input feature carrier for modeling the nonlinear relationship between the abnormality of parking nodes and the rate of change of road segment speed: the positive correlation between the number of abnormal nodes and the road network influence weight fits the core logic that abnormal parking behavior exacerbates road network resistance; the gradient distribution of traffic flow density corresponds to the analysis dimension of traffic state fluctuations within the abnormal time window. The association results of this graph directly support the weight calculation of the node congestion association model, provide data basis for the priority determination of influence weights in the resource allocation and strategy generation module, and serve as the technical presentation carrier of the association logic between abnormal nodes, traffic state, and road network resistance in the congestion prediction process.

[0042] Example 3: In specific implementation, the resource allocation and strategy generation module determines the set of parking space nodes that need adjustment based on the impact weights output by the congestion prediction and modeling module. Nodes with higher impact weights are prioritized for inclusion in the set to be adjusted. The module analyzes the amount of parking order data and the user sensitivity level stored in each node within the set of parking space nodes to be adjusted. The amount of parking order data refers to the total storage size of data such as currently valid reservation orders and historical parking records stored on the node. The user sensitivity level is determined by classifying the personal information contained in the orders according to data classification standards, such as dividing them into different levels like public, internal, and sensitive. The strength of data dependencies between nodes is assessed. The dependency strength is quantified by the weighted sum of the cross-node order association ratio and the synchronous access frequency. The cross-node order association ratio refers to the proportion of orders involving multiple nodes to the total number of orders. The synchronous access frequency refers to the average number of times the system needs to access multiple associated nodes simultaneously to complete a certain business operation per unit time. Combining the impact weights, parking order data volume, and data dependency strength, a multilayer perceptron model is used to calculate the migration urgency score for each node. As we can understand, the migration urgency score is a core output of the comprehensive decision-making process; a higher score indicates that the data at that node needs to be migrated first to alleviate local congestion. Migration Urgency Score The calculation can be expressed as:

[0043] in: Representing the The urgency score for relocating each parking space node. This represents the influence weight of the node provided by the congestion prediction and modeling module. This represents the amount of parking order data stored in this node. This represents the average level of user sensitivity information in the data stored on this node. This represents the strength of the data dependency between this node and other nodes. The set of trainable parameters representing a multilayer perceptron model. This represents the forward propagation computation process of a multilayer perceptron. Through supervised learning, the multilayer perceptron model learns how to optimally combine these heterogeneous input features from historical operational data to output a score reflecting the overall urgency.

[0044] Nodes are ranked based on migration urgency scores, and the node with the highest score is selected as the source node to be migrated. The specific number can be dynamically determined based on the current overall system load and available resources. Target storage node clusters are selected from pre-defined low-load areas within the system. These target storage node clusters should have sufficient storage space, computing resources, and network bandwidth. The optimal data transmission path from each source node to the selected target storage node cluster is calculated using the Dijkstra algorithm. The Dijkstra algorithm traverses the network topology, calculates the total cost of all possible paths from the source node to the target cluster, and selects the path with the lowest total cost as the optimal path. The path cost is mainly determined based on factors such as link bandwidth, real-time transmission latency, and current link utilization, and the path must avoid currently congested links identified by the network monitoring system. An isolation configuration scheme is generated, including source node identifiers, target cluster addresses, migration data volume, transmission path sequence, and execution time windows. The execution time window is typically selected during off-peak business hours to minimize the impact on normal parking operations. In some embodiments, the quantitative calculation of data dependency strength can be more refined, for example, setting the weight of the cross-node order association ratio. The weight of synchronous access frequency is Then the strength of the dependency relationship :

[0045]

[0046] in: It is a node The cross-node order association ratio, It is a node The frequency and weight of synchronous access involved and Adjustments can be made based on the importance of the business.

[0047] Optionally, the multilayer perceptron model can be designed to include an input layer, several hidden layers, and an output layer. The input layer neurons correspond to four input features, and the output layer consists of one neuron, outputting a transfer urgency score. The hidden layers use non-linear activation functions to capture complex interactions between features. Model training requires a large amount of labeled data, namely the feature vectors of each node in historical scenarios and their corresponding "true" urgency levels or scores, determined by operations experts or post-event assessments based on actual impact. It is understood that the Dijkstra algorithm requires real-time or near-real-time network topology information as input, provided by the underlying network management system. The generation of isolation configuration schemes is an automated process; the schemes exist in a structured data format and are passed to the execution and feedback monitoring module for execution. In some embodiments, the selection of the target storage node cluster may be more than one, and decisions can be made based on data affinity strategies or load balancing strategies. The transmission path sequence explicitly indicates the address of each network device that the data packet needs to pass through, ensuring the determinism of the data transmission path.

[0048] See Figure 5 This graph carries the core quantitative results of the resource allocation and strategy generation module, including migration urgency scores. The parking space node grid corresponding to the rows and columns in the graph is a spatial mapping carrier of the set of parking space nodes to be adjusted. High-scoring areas in the graph correspond to priority nodes with high migration urgency scores, which is the direct basis for selecting the highest-scoring node as the source node to be migrated; low-scoring areas are the screening reference dimensions for target storage nodes in low-load areas of the module. The distribution results of this graph directly support the resource allocation process of node sorting - source node selection - target cluster matching. It is a key data presentation format for the implementation of dynamic reallocation strategies within the module, and provides a spatial reference for node priority for the generation of subsequent isolation configuration schemes.

[0049] Example 4: In the specific implementation, the execution and feedback monitoring module begins collecting usage logs of parking space nodes affected by the migration operation during the monitoring period after the data migration is completed. These logs record every access event to the parking space data. Key access pattern features are extracted from these logs, mainly including access time distribution, single access duration, and concurrent access count. Access time distribution refers to the frequency of access events occurring in different time periods; single access duration refers to the duration from initiating an access request to receiving a complete data response; and concurrent access count refers to the number of access requests occurring simultaneously at a specific point in time. The access pattern features observed during the monitoring period after migration are dynamically time-warped and aligned with the baseline pattern established in a comparable period before migration. The dynamic time warping algorithm can overcome the slight scaling and distortion that may exist on the time axis before and after migration, finding the optimal correspondence between the two sequences. The cumulative difference between the migrated sequence and the baseline sequence at the key pattern points determined by dynamic time warping is calculated. The cumulative difference reflects the overall degree of change in access behavior patterns. Refer to Table 1, which shows a comparison of some access pattern features of a representative parking space node before and after migration.

[0050] Table 1: Comparison of Access Pattern Characteristics Before and After Parking Space Node A Migration

[0051] Feature indicators Baseline pattern before migration (same time period) Post-migration observation mode (monitoring cycle) Differences after normalization Peak periods of access time distribution 09:00-10:00 (35% of visits) 09:30-10:30 (40% of visits) Time offset 0.5 hours, quantity increase 5%. Average duration per access (milliseconds) 120 180 +60 Maximum concurrent access 25 35 +10

[0052] When the calculated cumulative difference value exceeds the system's set adaptive threshold, the system initiates a backtracking analysis process to locate the source of the abnormal behavior. The adaptive threshold is not fixed but dynamically calculated based on the statistical distribution of cumulative differences over historically stable periods; for example, it can be set as the historical mean plus three standard deviations. Backtracking analysis targets access event sequences that exhibit significant differences after dynamic time-warping alignment, analyzing the device IDs of the operation originating from these access events and the user IDs of the users initiating the operations. It correlates and queries access records of other adjacent or related nodes within the same abnormal time window to identify cross-node, temporally or pattern-wise coordinated abnormal access patterns; for example, multiple related user IDs initiating similar abnormal queries from devices in different geographical locations at similar times. The identified abnormal access sequences are then matched against a pre-built vulnerability signature database, which includes known attack patterns, system vulnerability exploitation characteristics, and unauthorized operation patterns. Through pattern matching, the type of behavioral source causing the observed anomalies is determined, such as crawling behavior originating from a specific device, abuse of user credentials, or a new type of unknown attack probe.

[0053] It is understandable that the cumulative difference value The calculation can be based on the differences between sequence points after dynamic time warping and alignment, and one implementation method is as follows:

[0054]

[0055] in: Represents the cumulative difference value. This represents the length of the sequence after dynamic time-normalized alignment. Representing the The importance weight of each alignment point It is a distance function used to calculate the distance after migration. eigenvectors With regularized path Mapping the corresponding pre-migration baseline feature vector The differences between them, feature vectors It includes the access time distribution component, the single access duration component, and the concurrent access number component. This represents the index of the sequence found by the dynamic time warping algorithm after the migration. Mapping to the pre-migration sequence index.

[0056] In some embodiments, the length of the monitoring period needs to be matched with the scale of the data migration and the scope of business impact. For migration operations with a large impact, the monitoring period may be set to 24 hours or longer, representing the index of the sequence found by the dynamic time warping algorithm after migration, to ensure that the complete access pattern is captured. It is understood that the application of the dynamic time warping algorithm allows the comparison process to tolerate non-linear distortions of the time axis that may occur before and after migration due to slight changes in network latency or natural fluctuations in business, thus focusing more on identifying genuine behavioral anomalies. Maintaining the vulnerability signature database is an ongoing process that requires continuous supplementation and revision based on newly discovered threat intelligence and system updates. Optionally, the determination of the behavioral source type can introduce a confidence score. When the matching confidence score is below a certain threshold, the abnormal sequence can be marked as an "unknown suspicious pattern" and submitted to a security analyst for further manual review. Analyzing the access records of adjacent nodes helps distinguish between localized, sporadic anomalies and organized, coordinated attacks, the latter typically exhibiting higher harm and requiring a more urgent response.

[0057] Example 5: In specific implementation, the system automatically triggers a predefined response script based on the type of behavior source determined by the execution and feedback monitoring module. The response script is a pre-written set of instructions designed to take corresponding control and mitigation measures against different types of security threats or abnormal behaviors. The response script contains various operation instructions, including but not limited to temporarily locking the parking space reservation permissions associated with the abnormal source, adjusting the parking space pricing coefficient for the abnormal area, and sending a detailed diagnostic report to the maintenance terminal with a prompt for manual review of key decision points. Temporarily locking the parking space reservation permissions associated with the abnormal source means that when the behavior source is identified as a malicious user or a stolen account, the system automatically suspends the user's or its associated device's ability to reserve all or some parking spaces in the system, preventing further malicious parking space occupation or data crawling. Adjusting the parking space pricing coefficient for the abnormal area means that when abnormal behavior causes systemic access pressure or resource competition in a certain area, the system can automatically increase the temporary usage rate of parking spaces in that area based on a dynamic pricing model, using economic levers to suppress potential abuse or channel excessive demand. Sending a detailed diagnostic report to the operations and maintenance terminal and prompting for manual review of key decision points means that the system will generate a structured report that summarizes the entire chain of information from anomaly detection to source location, and push it to the operations and maintenance personnel's console through a message queue or API interface. The report will clearly mark the decision-making links that require human expert intervention, such as whether to permanently block a new type of suspicious pattern with low matching confidence.

[0058] The execution logic of the response script can be formally represented as a rule-based mapping function. Assume... This represents the set of response actions ultimately executed by the system, the specific content of which depends on the detected behavior source type. The decision can be made based on the severity of the abnormality. Perform parameterized adjustments. Response action set. The generation can be described as:

[0059]

[0060] in: Represents the set of response actions that will ultimately be executed. This represents the type of behavior source determined by the execution and feedback monitoring module. It is a mapping function that maps behavior to the source type. Mapped to a predefined set of response action template identifiers corresponding to that type. . This is related to identifiers The corresponding specific response action execution function accepts two parameters: the severity of the exception. and the parameter set specific to this type of response action. .function The execution result is a specific, system-manageable operation instruction. (Symbol) This means merging the specific operation instructions generated by all triggered response action templates into a complete set of response actions. .

[0061] In some embodiments, predefined response scripts are stored in a dedicated policy library, and each script is bound to one or more behavior source type tags. It is understood that the duration of temporary access lockout can vary depending on the severity of the anomaly. The correlation is positive; for minor anomalies, the lockout time may only be a few hours, while for severe anomalies confirmed as malicious attacks, the lockout time may be extended to several days or require manual unlocking. The adjustment of the parking space pricing coefficient can also be based on the severity. The pricing coefficient is dynamically calculated based on real-time demand pressure. For example, when the severity is high and real-time access volume far exceeds the normal baseline, the increase will be greater. Detailed diagnostic reports need to be readable and actionable, including an anomaly timeline, a list of affected parking space nodes, detailed information on the identified behavioral sources (such as user IDs and device fingerprints), the matched vulnerability characteristic patterns, and a list of automated response measures already implemented by the system. Optionally, key decision points prompting manual review can be set as an interactive workflow. After receiving the prompt, operations personnel can directly view the relevant evidence chain on the console and select operations such as "confirm execution," "reject," or "escalate processing." In some embodiments, for situations where the behavioral source type is determined to be "unknown" or has low matching confidence, the response script may only include sending the diagnostic report and prompting manual review, without immediately executing operations that directly affect users, such as permission modifications or pricing adjustments, to avoid misjudgment. The system supports version management and canary release of response scripts, allowing administrators to update and test the scripts to ensure the effectiveness and security of the response strategy.

[0062] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An IoT-based urban traffic cooperative intelligent parking management system, characterized in that, The system includes the following modules: The data acquisition and preprocessing module is used to capture vehicle parking behavior data and parking space status change signals in real time through a group of sensor devices deployed in urban parking areas, and to perform time series alignment and noise filtering on the captured multi-source data to generate a standardized parking event stream. The feature extraction and anomaly detection module is used to perform pattern parsing on the standardized parking event stream, extract the fluctuation features of parking space occupancy time and vehicle parking frequency features, and determine the degree of abnormal deviation of the current area parking behavior by comparing with the historical normal behavior feature library. The congestion prediction and modeling module is used to locate the physical nodes where high-frequency abnormal parking behavior occurs based on the abnormal deviation, and to build a node congestion correlation model by combining real-time traffic flow data to predict the potential traffic resistance of the road network around the node. The resource allocation and strategy generation module is used to formulate a dynamic reallocation strategy for parking space resources based on the potential traffic resistance, calculate the data migration urgency score of different parking space nodes, and generate an isolation configuration scheme that includes path planning and permission adjustment instructions. The execution and feedback monitoring module is used to continuously monitor changes in parking space usage patterns after the isolation configuration scheme is executed, and to locate abnormal behavior sources by comparing the differences in access sequences before and after the configuration takes effect.

2. The IoT-based urban traffic cooperative intelligent parking management system according to claim 1, characterized in that, The time-series alignment and noise filtering of the captured multi-source data includes: receiving raw data streams with inconsistent timestamps from geomagnetic sensors, cameras, and payment terminals; segmenting and buffering the data streams using a sliding window mechanism; normalizing the timestamps of the data within each segment using GPS clock signals as a reference; applying a variance-based outlier detection algorithm to remove transient noise data caused by equipment failures; and extracting feature vectors from the processed data segments, whereby the feature vectors include the vehicle's parking start time, duration, and interval between adjacent events.

3. The IoT-based urban traffic cooperative intelligent parking management system according to claim 2, characterized in that, The extraction of parking space occupancy duration fluctuation features and vehicle parking frequency features includes: statistically analyzing the total occupancy duration sequence of a single parking space within a preset time period from the normalized parking event stream, calculating the variance and autocorrelation coefficient of the total occupancy duration sequence to characterize duration fluctuation; statistically analyzing the vehicle parking frequency sequence within the same period, calculating its coefficient of variation to characterize frequency stability; combining the duration fluctuation and frequency stability into a multi-dimensional feature vector, inputting it into a pre-trained autoencoder for feature dimensionality reduction, and obtaining low-dimensional dense features.

4. The IoT-based urban traffic cooperative intelligent parking management system according to claim 3, characterized in that, The process of determining the abnormal deviation of parking behavior in the current area by comparing it with a historical normal behavior feature database includes: calling a historical normal parking behavior feature template stored in a distributed database, the template containing a range of baseline features for different time periods and different parking space types; calculating the Mahalanobis distance between the currently extracted low-dimensional dense features and the corresponding template baseline features, and mapping the distance value to a deviation score from zero to one hundred; when the deviation score exceeds a dynamic threshold, marking the parking space node as having an abnormal deviation, and recording the abnormal time window.

5. The IoT-based urban traffic cooperative intelligent parking management system according to claim 4, characterized in that, The construction of the node congestion correlation model includes: acquiring real-time traffic flow data provided by the municipal traffic management platform, establishing a road network grid with abnormal parking nodes as the center and a radiation radius of 500 meters; analyzing the average vehicle speed change rate and traffic flow density increment of each road segment within the grid during the abnormal time window; using a graph convolutional network to model the nonlinear relationship between the abnormality degree of parking nodes and the vehicle speed change rate of surrounding road segments, and outputting the influence weight of each node on the road network traffic resistance.

6. The IoT-based urban traffic cooperative intelligent parking management system according to claim 5, characterized in that, The calculation of the data migration urgency score for different parking space nodes includes: determining the set of parking space nodes that need to be adjusted based on the influence weights; analyzing the amount of parking order data stored in the nodes within the set and the level of user sensitivity information; assessing the strength of data dependencies between nodes, which is quantified by the weighted sum of the cross-node order association ratio and the synchronous access frequency; and using a multilayer perceptron model to calculate the migration urgency score for each node, combining the influence weights, data volume, and dependency strength, with higher scores indicating priority processing.

7. The IoT-based urban traffic cooperative intelligent parking management system according to claim 6, characterized in that, The process of generating an isolation configuration scheme that includes path planning and permission adjustment instructions includes: sorting nodes according to migration urgency scores and selecting the node with the highest score as the source node to be migrated; selecting a target storage node cluster in a low-load area, calculating the optimal data transmission path from the source node to the target cluster based on the Dijkstra algorithm, wherein the path must avoid the current network congestion links; and generating an isolation configuration scheme that includes the source node identifier, target cluster address, migration data volume, transmission path sequence, and execution time window.

8. The IoT-based urban traffic cooperative intelligent parking management system according to claim 7, characterized in that, The continuous monitoring of parking space usage pattern changes after implementing the isolation configuration scheme includes: during the monitoring period after the data migration is completed, collecting usage logs of the affected parking spaces, extracting access time distribution, single access duration and concurrent access count; dynamically time-aligning the access pattern characteristics after migration with the baseline pattern of the same period before migration, and calculating the cumulative difference between the two at key pattern points.

9. The IoT-based urban traffic cooperative intelligent parking management system according to claim 8, characterized in that, The method of locating the source of abnormal behavior by comparing the differences in access sequences before and after the configuration takes effect includes: when the cumulative difference value exceeds the adaptive threshold, tracing back the access event sequences with significant differences, parsing the operation source device number and user identifier; associating the access records of other adjacent nodes within the same time window to identify whether there is a collaborative abnormal pattern; and performing pattern matching between the abnormal access sequence and the system vulnerability feature library to determine the type of behavior source that caused the abnormality.

10. The IoT-based urban traffic cooperative intelligent parking management system according to claim 9, characterized in that, The system also includes an adaptive response mechanism: automatically triggering a predefined response script based on the determined behavior source type. The script includes, but is not limited to, temporarily locking the parking space reservation permissions associated with the abnormal source, adjusting the parking space pricing coefficient in the abnormal area, sending a detailed diagnostic report to the operation and maintenance terminal, and prompting manual review of key decision points.

Images