Data security storage method of intelligent charging robot based on cloud data

Abstract

The invention relates to the technical field of data storage, in particular to a data security storage method of an intelligent charging robot based on cloud data. The method comprises the following steps: acquiring sensing data of a charging robot, and performing multi-source data preprocessing to obtain original acquisition data; performing parking environment tracing according to the original acquisition data to obtain abnormal parking environment data; judging an abnormal mode set according to the abnormal parking environment data identification instruction; reconstructing a robot prompt instruction for the original collected data by using the instruction judgment abnormal mode set, and uploading the robot prompt instruction to a charging robot control platform to add a control instruction; and performing edge encryption and data de-identification on the to-be-stored order data, and performing access permission dynamic behavior adjustment in combination with the instruction judgment abnormal mode set to obtain a charging robot data storage verification chain. According to the invention, the security of data storage of the charging robot, the privacy protection capability and the system stability can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data storage, and in particular to a data security storage method for an intelligent toll collection robot based on cloud data. Background Art

[0002] Existing intelligent toll collection robots usually store data using a centralized cloud server. All robot terminals need to access the remote server through the network for data interaction. This centralized storage structure may lead to large-scale data loss or service interruption in case of server downtime, cyber attacks, or hardware failures, affecting the stability of the toll collection system. In addition, the single-server storage structure is prone to being a target of cyber attacks, such as DDoS attacks and data theft, increasing the risk of data leakage. The data stored in the cloud often involves user privacy information, such as vehicle driving records and payment habits. Traditional methods mainly rely on access control (such as RBAC) or data desensitization (such as hashing) for privacy protection. However, these methods are difficult to resist advanced inference attacks. For example, attackers can infer user behavior characteristics from the stored data through association analysis, pattern matching, etc., thus threatening user privacy. In addition, traditional methods lack traceability and are difficult to detect and prevent malicious tampering or abuse of data. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a data security storage method for an intelligent toll collection robot based on cloud data to solve at least one of the above technical problems.

[0004] To achieve the above object, a data security storage method for an intelligent toll collection robot based on cloud data includes the following steps:

[0005] Step S1: Obtain the sensing data of the toll collection robot, and perform multi-source data preprocessing on the sensing data of the toll collection robot to obtain the original collected data;

[0006] Step S2: Identify the vehicle abnormal parking level according to the original collected data to obtain the vehicle abnormal parking data; trace the parking environment of the vehicle abnormal parking data to obtain the abnormal parking environment data;

[0007] Step S3: Infer the cause and effect of the toll collection robot instruction implementation based on the abnormal parking environment data to obtain an instruction abnormal cause and effect correlation matrix; identify an instruction judgment abnormal mode set according to the instruction abnormal cause and effect correlation matrix;

[0008] Step S4: Use the instruction judgment abnormal mode set to divide the original collected data into abnormal order data to obtain the judgment abnormal order data and the order data to be stored; reconstruct the robot prompt instruction according to the judgment abnormal order data and upload it to the toll collection robot control platform to add a control instruction;

[0009] Step S5: Perform edge encryption and data de-identification on the order data to be stored, and combine the instructions to judge the abnormal pattern set for dynamic behavior adjustment of access rights, so as to obtain the charging robot data storage verification chain.

[0010] Optionally, step S1 is specifically as follows:

[0011] Step S11: Obtain the sensing data of the charging robot, including robot operation logs, lidar data, camera data, and inertial measurement data;

[0012] Step S12: Align the sensing data of the charging robot by time step to obtain the time series aligned sensing data;

[0013] Step S13: Suppress the noise of the sensing data and standardize the data for the time series aligned sensing data to obtain the standardized sensing data;

[0014] Step S14: Perform data anomaly detection and redundant feature dimensionality reduction based on the standardized sensing data to obtain the dimensionality-reduced sensing data;

[0015] Step S15: Perform multi-source data feature fusion and environmental information enhancement based on the dimensionality-reduced sensing data to obtain the original collected data.

[0016] Optionally, step S15 is specifically as follows:

[0017] Step S151: Extract the spatio-temporal features of the sensors based on the dimensionality-reduced sensing data, and perform spatio-temporal correlation analysis on the spatio-temporal features of the sensors to obtain the spatio-temporal correlation features of the sensors;

[0018] Step S152: Perform spatio-temporal fusion of multi-source sensing data on the dimensionality-reduced sensing data based on the spatio-temporal correlation features of the sensors to obtain the fused sensing data of the charging robot;

[0019] Step S153: Obtain the weather data of the parking lot and perform data preprocessing to obtain the parking lot weather data to be analyzed;

[0020] Step S154: Calculate the traffic flow of the parking lot according to the fused sensing data of the charging robot, and combine the traffic flow of the parking lot with the parking lot weather data to be analyzed to perform parking lot environment analysis to obtain the parking lot environment data;

[0021] Step S155: Use the parking lot environment data to enhance the environmental features of the fused sensing data of the charging robot to obtain the original collected data.

[0022] Optionally, the identification of the abnormal parking level of the vehicle in step S2 is specifically as follows:

[0023] Extract the vehicle parking time series features from the original collected data to obtain the vehicle movement trajectory time series data, the vehicle parking time data, and the license plate recognition result time series data;

[0024] Perform vehicle movement behavior association based on the vehicle movement trajectory time series data and the vehicle parking time data to obtain the vehicle movement behavior time data and the vehicle stationary behavior time data;

[0025] Obtain the parking lot structure data, perform road connectivity analysis on the parking lot structure data, and define the roads with connectivity higher than the preset connectivity threshold as no-parking areas to obtain the parking lot no-parking areas;

[0026] Perform spatial superposition on the vehicle stationary behavior time data and the parking lot no-parking areas, and identify the vehicles with a stationary time greater than 10 minutes and a spatial distance from the parking lot no-parking areas less than 5 cm to obtain the no-parking area parking vehicle data;

[0027] Perform license plate repeated recognition vehicle detection based on the license plate recognition result time series data and the vehicle movement behavior time data to obtain the repeated recognition vehicle data;

[0028] Merge the no-parking area parking vehicle data and the repeated recognition vehicle data, and assign the vehicle parking anomaly level and the abnormal parking behavior label to obtain the vehicle abnormal parking data.

[0029] Optionally, the license plate repeated recognition vehicle detection is specifically:

[0030] Calculate the average parking time of the parking lot according to the parking lot structure data and the vehicle parking time data;

[0031] Calculate the robot lifting rod time based on the vehicle recognition result time series data and the robot operation log in the original collected data to obtain the robot instruction execution time;

[0032] Calculate the average vehicle movement time of the scene according to the average parking time of the parking lot and the robot instruction execution time;

[0033] Calculate the license plate recognition frequency of the license plate recognition result time series data to obtain the license plate recognition frequency data;

[0034] Combine the vehicle movement behavior time data and the license plate recognition frequency data to calculate the license plate recognition frequency distribution of the vehicle in different time intervals, and perform exponentially weighted moving average on the license plate recognition frequency distribution to obtain the license plate recognition frequency threshold;

[0035] Time superposition is performed based on the license plate recognition frequency data and the vehicle movement behavior time data to identify vehicles whose license plate recognition frequency is greater than the license plate recognition frequency threshold and whose vehicle movement behavior time is less than the average movement time of scene vehicles, and repeatedly identified vehicle data is obtained.

[0036] Optionally, the parking environment tracing in step S2 is specifically:

[0037] Reconstruct the parking lot spatial point cloud based on the LiDAR data in the original collected data;

[0038] Extract the video spatiotemporal feature data from the original collected data, and establish a parking lot spatiotemporal twin model based on the parking lot space point cloud and the video spatiotemporal feature data;

[0039] The abnormal parking data of vehicles is matched with abnormal time points through the parking lot spatiotemporal twin model to trace back the abnormal parking time and environment state and obtain the abnormal parking time and environment data;

[0040] Calculating the remaining parking space based on the abnormal parking time environment data, and eliminating the abnormal parking data of vehicles whose remaining parking space at the corresponding time is less than 5% according to the remaining parking space, adjusting the abnormal parking level of the vehicle, and obtaining the first abnormal parking correction data of the vehicle;

[0041] Identify parking lot road signs based on abnormal parking time environment data and calculate the completeness of parking lot road signs;

[0042] According to the parking lot road sign integrity, the abnormal parking data of vehicles whose parking lot road sign integrity is less than 20% at the corresponding time are eliminated from the abnormal parking data of vehicles, and the abnormal parking level of the vehicles is adjusted to obtain the second abnormal parking correction data of the vehicles;

[0043] Merging the first vehicle abnormal parking correction data with the second vehicle abnormal parking correction data, and performing vehicle parking abnormality hierarchical clustering to obtain vehicle abnormal parking correction data;

[0044] The environmental data corresponding to the abnormal parking correction data of the vehicle is extracted from the abnormal parking time environmental data to obtain the abnormal parking environmental data.

[0045] Optionally, step S3 specifically includes:

[0046] Step S31: Obtain the prompt instruction set of the charging robot, and combine it with the robot operation log in the original collected data to perform time sequence matching association, set the time sequence matching threshold to 0.5 to filter the robot operation log, and obtain the charging robot instruction operation data;

[0047] Step S32: Perform time series decomposition on the abnormal parking environment data, set the number of reconstruction components to 5 and the modulation factor to [0.1, 1.0], and obtain the multi-scale abnormal parking environment data;

[0048] Step S33: Conduct Bayesian causal inference based on the multi-scale abnormal parking environment data and the toll robot instruction operation data, and construct a parking environment causal inference model in combination with the abnormal parking behavior labels in the vehicle abnormal parking data;

[0049] Step S34: Analyze the causal path of the toll robot instruction judgment according to the parking environment causal inference model, calculate the instruction judgment deviation path coefficient, screen the causal paths with the instruction judgment deviation path coefficient greater than the threshold of 0.1, and generate an instruction anomaly causal association matrix;

[0050] Step S35: Perform hierarchical clustering analysis according to the instruction anomaly causal association matrix, set the maximum number of clusters for hierarchical clustering to 6, extract the abnormal instruction patterns and calculate the instruction anomaly impact factor, and set the impact factor threshold to 0.25, so as to generate an instruction judgment abnormal pattern set.

[0051] Optionally, step S34 is specifically:

[0052] Step S341: Conduct a causal structure analysis on the parking environment causal inference model, analyze the causal path of the toll robot instruction judgment, set the window size for calculating the path coefficient to 50 pieces of data, calculate the causal path coefficient, screen the causal paths with the path coefficient greater than 0.05, and obtain the preliminary instruction judgment causal path data;

[0053] Step S342: Measure the time lag impact based on the preliminary instruction judgment causal path data, set the lag order to 3 to correct the causal relationship weight, and generate the time series corrected instruction judgment causal path data;

[0054] Step S343: According to the time series corrected instruction judgment causal path data, and in combination with the abnormal parking environment variables in the multi-scale abnormal parking environment data, set the abnormal parking environment variable impact weight threshold to 0.1, screen the environment variables with the cumulative contribution rate greater than 90%, and calculate the causal path deviation impact factor;

[0055] Step S344: Conduct an instruction judgment path contribution analysis on the instruction judgment causal path deviation impact factor, set the deviation contribution threshold to 0.2, screen the causal paths with the contribution rate higher than 80%, and obtain the key instruction judgment deviation path data;

[0056] Step S345: Reconstruct the causal influence of the time series corrected instruction judgment causal path data in combination with the key instruction judgment deviation path data, and update the causal path hierarchy to generate an instruction anomaly causal association matrix.

[0057] Optionally, step S4 is specifically:

[0058] Step S41: using the instruction judgment abnormality pattern set to perform instruction judgment abnormality time association on the original collected data to obtain instruction judgment abnormality sensor data and instruction judgment normal sensor data;

[0059] Step S42: matching the instruction judgment abnormal order prompt instruction with the instruction operation data of the charging robot according to the instruction judgment abnormality sensor data to obtain the judgment abnormality order data;

[0060] Step S43: According to the abnormal sensor data of the instruction judgment, the instruction operation data of the charging robot is matched with the normal order prompt instruction to obtain the order data to be stored;

[0061] Step S44: Calculating the instruction execution deviation based on the abnormal order data, and building an order instruction deviation measurement matrix in combination with the abnormal parking environment data;

[0062] Step S45: Reconstruct the prompt instruction selection path and instruction execution parameters in the charging robot instruction operation data according to the instruction anomaly causal association matrix and the order instruction deviation measurement matrix, obtain the robot control instruction, and upload it to the charging robot control platform to add the control instruction.

[0063] Optionally, step S5 specifically includes:

[0064] Step S51: performing data encryption preprocessing on the order data to be stored, generating an encrypted summary of the order data, and establishing a data integrity verification parameter set;

[0065] Step S52: performing de-identification processing based on the encrypted summary of the order data, removing the user's license plate information and payment ID from the order data to be stored, and generating de-identified order data;

[0066] Step S53: setting a dynamic access control policy based on the de-identified order data and the abnormal pattern set determined in combination with the instruction;

[0067] Step S54: dynamically adjust the access rights of the charging robot control platform to the robot operation log using a dynamic access control strategy to obtain a dynamic adjustment record of the access rights;

[0068] Step S55: dynamically adjust the records based on the access rights, de-identify the order data and the data integrity verification parameter set to build a charging robot data storage verification chain and store it in a preset distributed ledger system.

[0069] Through a decentralized data storage method, combined with edge encryption, data de-identification, and dynamic access permission adjustment, the present invention significantly improves the security, privacy protection ability, and system stability of the data storage of toll robots. First, through multi-source data preprocessing, noise in the sensing data is eliminated, and the data quality is improved, ensuring the accuracy of subsequent anomaly detection. Secondly, in the process of vehicle abnormal parking level recognition and parking environment traceability, through multi-dimensional feature analysis, it is ensured that the toll robot can accurately judge abnormal situations, and an instruction abnormal causal association matrix is constructed in combination with the parking environment information, thereby improving the accuracy and traceability of anomaly event reasoning. Based on this matrix, instruction abnormal pattern recognition is carried out, enabling the toll robot to adopt targeted strategies when detecting abnormal behaviors, effectively reducing the misjudgment rate. In addition, through the division of abnormal order data, preliminary isolation of abnormal data can be completed before order storage, reducing the interference of incorrect data on the system. At the same time, by reconstructing the robot prompt instruction, the response speed of the control platform to abnormal situations is improved. Finally, the present invention performs encryption preprocessing on order data through edge encryption and combines de-identification processing to remove user sensitive information, fundamentally reducing the risk of data leakage. In addition, dynamic access permission adjustment ensures that data access permissions can be flexibly adjusted according to abnormal patterns, significantly reducing the risk of malicious access or unauthorized access. The introduction of the data storage verification chain for toll robots enhances the traceability of data storage, enabling all data operations to be recorded and audited, thereby effectively preventing data tampering or abuse behaviors and improving the data security and compliance of the toll system. In summary, the technical solution of the present invention overcomes the limitations of the traditional centralized storage mode and realizes multiple optimizations of data secure storage, privacy protection, and intelligent anomaly recognition. Brief Description of the Drawings

[0070] Other features, purposes, and advantages of the present invention will become more obvious by reading the detailed description of the non-restrictive embodiments with reference to the following drawings:

[0071] Figure 1 It is a schematic flow chart of the steps of the data secure storage method for the intelligent toll robot based on cloud data of the present invention;

[0072] Figure 2 It is a detailed schematic flow chart of step S1 in the present invention;

[0073] The realization, functional characteristics, and advantages of the object of the present invention will be further described with reference to the embodiments and the drawings. Detailed Embodiments

[0074] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0075] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0076] It should be understood that although terms such as "first" and "second" may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly, the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0077] To achieve the above object, please refer to Figures 1 to 2 , the present invention provides a data security storage method for an intelligent toll collection robot based on cloud data, and the method includes the following steps:

[0078] Step S1: Obtain the sensing data of the toll collection robot and perform multi-source data preprocessing on the sensing data of the toll collection robot to obtain the original collected data;

[0079] In this embodiment, various types of sensing data are collected from the toll robot management platform, including lidar, ultrasonic sensors, cameras, as well as position and speed data. The sensor data acquisition frequency is set to 10 times per second (i.e., the sampling frequency is 10 Hz), and then multi-source data preprocessing is performed. The timestamps of different data sources are aligned, the maximum time deviation threshold is set to 0.1 second, and the interpolation method is used for time synchronization. Through a denoising algorithm, such as a low-pass filter (frequency set to 1 Hz), high-frequency noise is removed to obtain a preprocessed original acquisition data set. The data of each sensor is merged to form the original acquisition data. The original data includes information such as the robot's position coordinates (e.g., x, y, z, unit: meter), speed, obstacle distance, angle, etc. for each sampling. In addition, the format of the collected original data needs to be standardized to adapt to subsequent data processing and analysis. This data set is further used for vehicle abnormal parking recognition and parking environment traceability.

[0080] Step S2: Identify the vehicle abnormal parking level based on the original acquisition data to obtain vehicle abnormal parking data; trace the parking environment of the vehicle abnormal parking data to obtain abnormal parking environment data;

[0081] In this embodiment, the vehicle abnormal parking level is identified by analyzing the original acquisition data. An anomaly detection model based on deep learning (such as a convolutional neural network CNN) can be used. By inputting the robot's position data (such as x, y coordinates) and parking time information, the model is set as a binary classification model, and the output is the label of normal parking and abnormal parking. According to the specific structure of the parking lot (such as the size of the parking lot, the channel structure, etc.), the criteria for abnormal parking behavior are set as follows: the vehicle stays outside the designated area and the stay time exceeds 5 minutes; or the vehicle deviates from the standard parking position by more than 5 meters. The vehicle abnormal parking data will include information such as vehicle ID, parking area number, offset, and parking time. At this time, through the analysis of parking environment traceability, a graph algorithm (such as Dijkstra algorithm) is applied to determine the specific environmental background of the parking position offset, such as whether there are other vehicles blocking, road obstacles, etc., and an abnormal parking environment data set is generated, including information such as parking area, abnormal offset, timestamp, and specific information of surrounding obstacles.

[0082] Step S3: Infer the cause and effect of the toll robot instruction implementation based on the abnormal parking environment data to obtain an instruction abnormal cause and effect correlation matrix; identify the instruction judgment abnormal mode set according to the instruction abnormal cause and effect correlation matrix;

[0083] In this embodiment, the causal relationship of the toll robot instructions is inferred based on the abnormal parking environment data. The Bayesian causal inference model is used to analyze the relationship between the parking environment and the instruction execution. The causal nodes in the model include the parking environment state (such as the position of obstacles, the parking position, etc.), the robot state (such as speed, direction, etc.), and the instruction type (such as parking instruction, path adjustment instruction, etc.). Through setting the maximum number of iterations to 100 times and the inference confidence level to 95%, the instruction abnormal causal association matrix is obtained in the Bayesian inference process. The form of this matrix is an N×M matrix, where N is the number of parking environment states and M is the number of instruction types. The matrix elements represent the causal influence coefficients of each environment state on the execution of various instructions, and the coefficient range is [0,1].

[0084] Step S4: Use the instruction judgment abnormal pattern set to divide the original collected data into abnormal order data, obtaining the judgment abnormal order data and the order data to be stored; reconstruct the robot prompt instruction according to the judgment abnormal order data, and upload it to the toll robot control platform to add a control instruction;

[0085] In this embodiment, according to the instruction abnormal causal association matrix, the abnormal order data is divided. The abnormal order standard is set as: if the deviation of the instruction execution result exceeds the set threshold (for example, the deviation exceeds 10%), it is determined as an abnormal order. Through regression analysis, based on the instruction judgment abnormal pattern set and the causal association matrix, it is identified which specific abnormal instruction selection paths cause the order abnormality. Combining with the parking environment traceability data, the order is reclassified to generate two types of data: "abnormal order data" and "order data to be stored". The abnormal order data will be further reconstructed, including regenerating the robot prompt instruction (such as automatic parking instruction, path adjustment instruction, etc.). The reconstructed prompt instruction will be uploaded to the toll robot control platform in real time through the API interface, and the control platform will update the instruction execution strategy according to the real-time data after receiving the instruction. For example, if it is detected that the obstacle is too close, a new path planning instruction will be generated and sent to the robot through the control platform.

[0086] Step S5: Perform edge encryption and data de-identification on the order data to be stored, and combine the instruction judgment abnormal pattern set to perform dynamic behavior adjustment of access rights, obtaining the toll robot data storage verification chain.

[0087] In this embodiment, the order data to be stored undergoes edge encryption and data de-identification processing. First, a homomorphic encryption algorithm is applied to the order data (for example, AES encryption with a 128-bit encryption key length) to ensure the confidentiality of the data during storage and transmission. At the same time, differential privacy technology is used to de-identify the data, and the noise injection intensity is set to 0.1 to ensure that the identity information of each user cannot be recovered. Combine the instruction to judge the abnormal mode set, and dynamically adjust the data access rights according to the importance and access rights of the instruction. Set an access right rule: if the instruction belongs to a high-risk category (such as an abnormal parking instruction), then restrict its access right, and only users with administrator privileges can view it, and ordinary users can only view non-sensitive data. Finally, all data will be uploaded to the distributed storage system, and a storage verification chain will be generated using blockchain technology. Through the smart contract, it is ensured that the data cannot be tampered with. Each data block in the verification chain contains a timestamp, a data hash value, and an access right record to ensure the integrity and security of the data.

[0088] Optionally, step S1 is specifically as follows:

[0089] Step S11: Obtain the sensing data of the toll collection robot, including the robot operation log, lidar data, camera data, and inertial measurement data;

[0090] In this embodiment, various types of sensor data are obtained from the toll collection robot management platform. These data include the robot operation log, lidar data, camera data, and inertial measurement data. Assume that the robot operation log is sampled once per second, recording the robot action instructions, execution status, and timestamp information. The lidar data is the laser scan point cloud data collected per second, containing 360-degree environmental scan information. Each sampled point cloud contains thousands of data points, reflecting the precise ranging of the surrounding environment. The camera data is the high-definition camera image collected every 30 milliseconds, used to capture visual information to assist in parking environment perception. The inertial measurement unit (IMU) data includes the acceleration, gyroscope, and magnetometer data once per second, used to measure the position, angle, and speed of the robot during movement. The data of each sensor will be aligned on the timestamp to ensure the synchronization of data from different sources, generating the original sensing data set. The units of all data will be unified according to the standardized units. For example, the laser ranging unit is meters, the image size is 1920×1080 pixels, and the units of IMU data are m / s 2 (acceleration) and ° / s (angular velocity).

[0091] Step S12: Align the time step of the sensing data of the toll collection robot to obtain the time-series aligned sensing data;

[0092] In this embodiment, for various types of sensor data collected originally, time step alignment is performed. Since the data sampling frequencies of different sensors are inconsistent, it is necessary to perform time step alignment on them. Assume that the time interval between each sampling of lidar data is 100 ms, camera data is sampled every 30 ms, and IMU data is sampled every 100 ms. To ensure the synchronization of different sensor data, first, the minimum time step is selected as 30 ms, and then the data of different sensors are interpolated. For example, for lidar data, the linear interpolation method is used to evenly distribute the data points at a time step of 30 ms. For camera data, the sampling frequency can be adjusted to 30 ms through the interpolation method, and the image frame timestamps are processed to ensure data synchronization. Finally, the time series aligned sensor data set obtained through interpolation and alignment will form a unified time series format, and all sensor data can be further processed according to the unified time step.

[0093] Step S13: Perform sensor data noise suppression and data standardization on the time series aligned sensor data to obtain standardized sensor data;

[0094] In this embodiment, noise suppression and data standardization processing are performed on the time series aligned sensor data. Noise suppression first targets lidar data, and a median filter (window size of 5) is applied to remove abnormal points caused by external interference. For image data, Gaussian blur filtering (standard deviation of 1.5) is used to remove noise points to maintain image quality. High-frequency noise is removed from IMU data through a low-pass filter (cutoff frequency of 2 Hz). During the data standardization process, first, the data of all sensors are normalized, and the Min-Max normalization method is used to adjust all data to the [0, 1] interval for subsequent analysis. The ranging value of each point cloud of the lidar is adjusted by this method, and the color value of each pixel point of the camera image is also normalized. Through standardization, it is ensured that the data of each sensor is within the same dimension, which is beneficial to subsequent data processing and fusion.

[0095] Step S14: Perform data anomaly detection and redundant feature dimensionality reduction based on the standardized sensor data to obtain dimensionality-reduced sensor data;

[0096] In this embodiment, data anomaly detection is first performed. For the standardized sensing data, the Z-score algorithm is used to detect outliers. The threshold is set to 3. If the deviation of a data point from its mean exceeds 3 times the standard deviation, then this data point is determined to be abnormal data and is removed or corrected. Next, redundant feature dimensionality reduction is performed on the sensing data after removing the abnormal data. The principal component analysis (PCA) method is adopted, and the target dimensionality reduction is set to 2 main components. Specifically, the covariance matrix of the data is calculated through PCA, and then the first two principal components are extracted. Finally, the sensing data is reduced from a high dimension to 2 dimensions. The data after dimensionality reduction will reduce redundant features and maintain the main variability of the data, ensuring that the data is more compact and efficient, facilitating subsequent processing and analysis.

[0097] Step S15: Perform multi-source data feature fusion and environmental information enhancement based on the dimensionality-reduced sensing data to obtain the original collected data.

[0098] In this embodiment, multi-source data feature fusion and environmental information enhancement are performed according to the dimensionality-reduced sensing data. First, based on the sensing data and parking environment information (such as obstacles, road conditions, etc.), feature fusion is performed through the weighted average method. The weights of lidar, camera, and IMU data are set to 0.5, 0.3, and 0.2 respectively, and these data are fused with different importance levels. During the fusion process, the distance information provided by lidar data is given a higher weight, while camera data is used for environmental recognition, and IMU data provides motion state information. The fused data generates a unified multi-dimensional feature vector through a weighted algorithm, and this vector contains environmental perception and vehicle state information. In addition, combined with environmental information enhancement, the data features are enhanced by using an environmental model (such as a parking scenario model) to ensure the vehicle's perception ability in a complex environment. Finally, an original collected data set containing environmental information and robot state is obtained, and this data set provides sufficient support for subsequent parking task execution and decision-making analysis.

[0099] Optionally, step S15 is specifically:

[0100] Step S151: Extract sensor spatio-temporal features based on the dimensionality-reduced sensing data, and perform spatio-temporal correlation analysis on the sensor spatio-temporal features to obtain sensor spatio-temporal correlation features;

[0101] In this embodiment, spatio-temporal features are extracted from the dimensionality-reduced sensing data. The dimensionality-reduced sensing data includes lidar, camera data, and IMU data, and these data are organized according to timestamps and spatial positions to form a spatio-temporal data sequence. To extract spatio-temporal features, a sliding window-based method is used. Each time window is 5 seconds, and spatial clustering (such as the DBSCAN algorithm) is performed on the lidar scan data within each window. By detecting changes at different time points and spatial positions, dynamic spatial features are extracted. At the same time, the IMU data provides temporal features of the motion state (such as speed, acceleration, etc.), and frequency domain features can be extracted through Fourier transform to analyze the periodic changes of vehicle motion. Next, spatio-temporal correlation analysis is performed on the sensor spatio-temporal features using a spatio-temporal graph convolutional network (ST-GCN). The spatio-temporal neighborhood range is set to 2 seconds and 10 meters, and the correlation between different sensors in the spatio-temporal dimension is calculated. Finally, the spatio-temporal correlation feature matrix obtained through graph convolution reflects the interaction between different sensors and its change relationship over time, forming sensor spatio-temporal correlation features.

[0102] Step S152: Perform spatio-temporal fusion of multi-source sensing data on the dimensionality-reduced sensing data based on the sensor spatio-temporal correlation features to obtain the fused sensing data of the toll collection robot.

[0103] In this embodiment, mapping of spatio-temporal features of lidar, camera, and IMU data is performed, and the weight coefficients are set as 0.4 for lidar, 0.3 for camera data, and 0.3 for IMU data. These data are fused by the weighted average method in combination with the spatio-temporal correlation features. To improve the fusion effect, a deep learning method can be introduced, and feature-level fusion is performed through a multi-modal neural network. The input of this network is the spatio-temporal features from different sensors, and the output is the fused sensing data. Specifically, an autoencoder structure is used for dimensionality reduction and feature extraction. The decoder part reconstructs the features of each sensor into a shared space, in which the data retains the maximum information and the minimum redundancy. To enhance the accuracy in the fusion process, the feature dimension of the decoder output is set to 128, and the network parameters are tuned through cross-validation to ensure that the fused data set can effectively integrate the spatio-temporal feature information of the sensors, forming a high-dimensional fused data vector.

[0104] Step S153: Obtain the parking lot weather data and perform data preprocessing to obtain the parking lot weather data to be analyzed.

[0105] In this embodiment, the parking lot weather data is obtained through public weather platforms such as regional meteorological observatories. These data include real-time weather information such as temperature, humidity, precipitation, and wind speed, and the sampling frequency is once every 10 minutes. The real-time weather data of the location where the parking lot is located is obtained from a public weather platform (such as Weather Cloud) through an API interface. The format of the weather data is usually a timestamp and the corresponding meteorological value. Then, the obtained weather data is preprocessed, mainly including removing missing values (for example, using linear interpolation to fill in the missing temperature values) and unifying the units to standard units (such as converting the temperature unit to Celsius and the wind speed unit to meters per second). In addition, to ensure data quality, anomaly detection is performed, and a method based on IQR (Interquartile Range) is used to detect and remove extreme values that do not conform to the actual weather conditions. The preprocessed data will be sorted by date and timestamp for subsequent analysis.

[0106] Step S154: Calculate the traffic flow of the parking lot according to the fusion sensing data of the toll collection robot, and combine the traffic flow of the parking lot with the weather data of the parking lot to be analyzed to perform parking lot environment analysis and obtain parking lot environment data;

[0107] In this embodiment, the traffic flow of the parking lot is calculated according to the fusion sensing data of the toll collection robot. The calculation of the traffic flow depends on the vehicle detection information from the lidar and camera in the fusion data. Specifically, the lidar provides the real-time position of the vehicle, the vehicle speed information is processed through a Kalman filter, and the type and parking state of the vehicle are identified by combining the camera to calculate the hourly vehicle in-out flow. Assuming that the capacity of the parking lot is 500 parking spaces, the vehicle flow calculation formula is: Vehicle flow = (Number of vehicles entering and leaving within a unit time) / (Total number of parking spaces). Then, in combination with the weather data of the parking lot, the influence of weather factors on the traffic flow is considered. For high-temperature weather (temperature higher than 30 °C), it is assumed that the traffic flow decreases by 10%; for the case where the precipitation exceeds 5 mm, the traffic flow decreases by 15%. The parking lot environment analysis is based on the above traffic flow and weather data for weighted analysis, and the weight coefficients are set as 0.7 for traffic flow and 0.3 for weather influence to obtain comprehensive parking lot environment data. The environment data will be used as the basic data for subsequent feature enhancement.

[0108] Step S155: Use the parking lot environment data to perform environmental feature enhancement on the fusion sensing data of the toll collection robot to obtain the original collected data.

[0109] In this embodiment, the environmental data of the parking lot is used to enhance the environmental features of the fusion sensing data of the toll collection robot. First, based on the calculated environmental data of the parking lot (including traffic flow and weather data), the dynamic environmental features of the parking lot are added to the fusion data. The influence factor of traffic flow on environmental feature enhancement is set to 0.6, and the influence factor of weather data is 0.4. In specific operations, the environmental data of the parking lot is combined with the robot fusion data through the weighted average method to calculate the environmental weighted feature value at each moment. The enhanced data will increase the context relevance of environmental information. For example, a higher traffic flow will result in more complex movements of the robot in the parking lot, and weather conditions (such as precipitation) affect the accuracy of the robot's sensors. These environmentally enhanced features will be trained through a neural network model (such as a convolutional neural network) to improve the decision-making ability of the robot, and finally the original collected data with enhanced environmental features is obtained.

[0110] Optionally, the vehicle abnormal parking level recognition described in step S2 is specifically:

[0111] Extract the vehicle parking time sequence features from the original collected data to obtain the vehicle movement trajectory time sequence data, vehicle parking time data, and license plate recognition result time sequence data;

[0112] In this embodiment, the vehicle parking time sequence features are extracted from the original collected data. The original collected data includes data from lidar, cameras, and IMU sensors, mainly focusing on the vehicle's trajectory, parking time, and license plate recognition results. First, the vehicle movement trajectory time sequence data is extracted from the lidar and camera data. Assuming a sampling frequency of 5Hz, and the vehicle position is smoothed through the Kalman filter algorithm to obtain the accurate trajectory data of the vehicle in the parking lot. Then, based on the vehicle position and timestamp in the parking lot, the parking duration is calculated, and the start and end times of parking for each vehicle are recorded to form the vehicle parking time data. License plate recognition is performed through a high-definition camera installed at the entrance of the parking lot, and the license plate is recognized in real time by combining a license plate recognition algorithm (such as a CNN model) to generate the license plate recognition time sequence data. This dataset includes information such as license plate number, recognition timestamp, and recognition accuracy to ensure the uniqueness and accuracy of the license plate.

[0113] Based on the vehicle movement trajectory time sequence data and the vehicle parking time data, the vehicle movement behaviors are associated to obtain the vehicle movement behavior time data and the vehicle stationary behavior time data;

[0114] In this embodiment, vehicle movement behaviors are associated based on the sequential data of vehicle movement trajectories and vehicle parking time data. During this process, by analyzing the dynamic changes in the vehicle movement trajectory data and combining the vehicle parking time data, the movement behaviors of each vehicle are extracted. The stationary behavior of a vehicle is defined as a time period during which the vehicle speed is lower than 1 km / h for more than 5 seconds. According to this criterion, the time window technique (such as the sliding window algorithm) is used to classify vehicle behaviors, obtaining the time data of the stationary and moving behaviors of the vehicle. During a specific time period (such as when the vehicle speed is lower than 1 km / h and exceeds 5 seconds), this period is classified as a stationary behavior, otherwise it is regarded as a moving behavior. All the data of stationary behaviors are classified through clustering analysis (such as the K-means algorithm), and the total stationary duration of the vehicle is calculated. After merging the stationary behavior data and the moving behavior data, the time data of the moving and stationary behaviors of each vehicle are obtained.

[0115] Obtain the parking lot structure data, and perform road connectivity analysis on the parking lot structure data. Define the roads with connectivity higher than the preset connectivity threshold as no-parking areas, obtaining the no-parking areas of the parking lot.

[0116] In this embodiment, the structure data of the toll robot application scenario is obtained through the toll robot management platform. The structure data includes the CAD drawings or maps of the parking lot and is digitally processed to generate a road network diagram. This diagram contains all the lanes, driving routes, and road information related to parking spaces in the parking lot. Then, the connectivity analysis method in graph theory (such as the depth-first search algorithm) is used to perform connectivity analysis on the parking lot road network. The connectivity threshold is set to 0.75, indicating that if the connectivity score between two roads is greater than 0.75, they are considered connected. Based on the analysis, any road that is not related to the main road and has a connectivity score lower than this threshold with other roads will be defined as a no-parking area. The identification of the no-parking area is automatically generated by the algorithm and associated with other information (such as parking space numbers) in the parking lot structure data, obtaining the final no-parking area data.

[0117] Perform spatial overlay on the vehicle stationary behavior time data and the no-parking areas of the parking lot, and identify the vehicles with a stationary time greater than 10 minutes and a spatial distance from the no-parking areas of the parking lot less than 5 cm, obtaining the data of vehicles parked in no-parking areas.

[0118] In this embodiment, spatial superposition is performed on the vehicle stationary behavior time data and the no-parking areas in the parking lot. The no-parking areas in the parking lot are determined based on previous connectivity analysis, and the stationary behavior data (such as parking position and time) of each vehicle has been extracted. The stationary behavior of each vehicle is calculated for spatial overlap with the no-parking areas in the parking lot using GIS spatial analysis methods (such as buffer analysis). A spatial distance threshold of 5 cm is set, that is, if the distance between the vehicle's stationary position and the edge of the no-parking area is less than 5 cm and the stationary duration is greater than 10 minutes, it is determined that the vehicle is parked in the no-parking area. The specific operation is to calculate the distance between the vehicle position and the no-parking area through a spatial superposition analysis tool. If the conditions that the distance between the vehicle's stationary position and the edge of the no-parking area is less than 5 cm and the stationary duration is greater than 10 minutes are met, mark the vehicle as a vehicle parked in the no-parking area, and output corresponding vehicle ID, stationary time, no-parking area position and other information.

[0119] Vehicle detection for duplicate license plate recognition is performed based on the time series data of license plate recognition results and the vehicle movement behavior time data to obtain duplicate recognition vehicle data;

[0120] In this embodiment, vehicle detection for duplicate license plate recognition is performed based on the time series data of license plate recognition results and the vehicle movement behavior time data. First, the license plate numbers of the vehicles in the parking lot and their corresponding timestamps are obtained through the license plate recognition system. Using a duplicate detection algorithm based on a time window, the window size is set to 10 minutes. If the same license plate number is detected more than twice within this time period or the license plate is recognized multiple times in a short period, the vehicle is considered a duplicate recognition vehicle. Then, by associating with the vehicle movement behavior time data, it is checked whether the vehicle moves to different parking spaces in a short period or is recognized repeatedly in different time periods. If the vehicle appears in multiple time periods and has an abnormal parking trajectory (such as the parking time interval is less than 1 minute), it is further marked as a duplicate recognition vehicle, and its repeated parking time, parking space information and license plate number are recorded.

[0121] The vehicle data of the vehicles parked in the no-parking area and the duplicate recognition vehicle data are merged for vehicle data, and the vehicle parking abnormal level and abnormal parking behavior label are assigned to obtain vehicle abnormal parking data.

[0122] In this embodiment, the parking vehicle data in the no-parking area and the repeatedly recognized vehicle data are merged, and a vehicle parking anomaly level and an abnormal parking behavior label are assigned. In this process, first, the parking vehicle data in the no-parking area and the repeatedly recognized vehicle data are merged to construct a comprehensive data set including vehicle ID, parking location, parking duration, license plate number, and other information. Then, according to the set rules (such as the parking location overlapping with the no-parking area, the parking duration exceeding 10 minutes, and the repeatedly recognized vehicle), an abnormal parking level (such as mild, moderate, severe) is assigned to each vehicle. For example, if a vehicle has parked in the no-parking area and has been repeatedly recognized, the vehicle is assigned a moderate abnormal parking level; if the abnormal parking frequency is high, the abnormal parking level is correspondingly increased. Based on this, an abnormal parking behavior label, such as "parking in the no-parking area" or "repeatedly recognized parking", is further assigned to each abnormal parking of each vehicle. These data will be recorded in the database and real-time alarms for abnormal behaviors will be sent through the subsequent monitoring system.

[0123] Optionally, the detection of the vehicle with repeated license plate recognition is specifically as follows:

[0124] Calculate the average parking time of the parking lot according to the parking lot structure data and the vehicle parking time data;

[0125] In this embodiment, the average parking time of the parking lot is calculated according to the parking lot structure data and the vehicle parking time data. The parking lot structure data includes parking space distribution, lane information, lane capacity, etc. The vehicle parking time data records the time of each vehicle from entering the parking lot to completing parking. First, the entry time and exit time of each vehicle are obtained from the monitoring system (camera) of the parking lot, and the parking duration of each vehicle is calculated. Then, the parking durations of all vehicles are counted, and the average parking time of the parking lot is calculated using statistical methods. Suppose there are 100 parking spaces in the parking lot, and the entry and exit times of 1000 vehicles are recorded. Using the formula: average parking time = ∑(vehicle parking duration) / total number of vehicles. Suppose the calculation result is 10 minutes, that is, the average parking time of the vehicles in the parking lot is 10 minutes.

[0126] Calculate the robot lifting rod time based on the vehicle recognition result time series data and the robot operation log in the original collected data to obtain the robot instruction execution time;

[0127] In this embodiment, the robot's time to raise the pole is calculated based on the time-series data of vehicle recognition results and the robot operation logs in the original collected data, and the time for implementing the robot instruction is obtained. The robot operation logs include the timestamps and operation results of each time the robot raises and lowers the pole. The time-series data of license plate recognition results provides the specific time when the vehicle enters the parking lot. By analyzing the relationship between the pole-raising time in the robot operation logs and the vehicle entry time, the time required for the robot to execute the pole-raising instruction each time is calculated. Assume that the average execution time of the robot's pole-raising operation is 5 seconds, and at a certain moment, when the license plate recognition system recognizes that the vehicle enters the parking lot, the robot needs 5 seconds to execute the pole-raising operation. Finally, the time for implementing the robot instruction, that is, the robot's time to raise the pole, is 5 seconds. The toll-collecting robots deployed in the parking lot can be a group of robots, that is, an intelligent entity. The robots in this intelligent entity include the display and pole-raising robots deployed at the entrance and exit, and the prompt robots distributed on both sides of the parking lot roads. Among them, the main function of the display and pole-raising robots is to control the pole-raising operation at the parking lot entrance, calculate and display the amount payable by the vehicle, and display the payment QR code to ensure the entry and exit management of the parking lot. The prompt robots are responsible for providing navigation information and parking guidance to the vehicles inside the parking lot to ensure that the vehicles are smoothly parked in the appropriate positions.

[0128] Calculate the average moving time of the vehicle in the scenario according to the average parking time in the parking lot and the time for implementing the robot instruction;

[0129] In this embodiment, the average moving time of the vehicle refers to the time when the vehicle moves in the parking lot. When calculating, the parking time of the vehicle and the operation time of the robot to raise the pole need to be combined. First, use the obtained average parking time in the parking lot and the obtained robot's time to raise the pole to calculate the average value of the vehicle moving time. For example, assume that the average parking time in the parking lot is 10 minutes and the robot's time to raise the pole is 5 seconds. Then the average moving time of the vehicle in the scenario can be calculated as: average moving time of the vehicle = average parking time - robot's time to raise the pole = 10 minutes - 5 seconds. At the same time, the time for implementing the robot instruction needs to be considered, and the average moving time of the vehicle in the scenario obtained at this time is about 9.92 minutes.

[0130] Calculate the license plate recognition frequency for the time-series data of license plate recognition results to obtain license plate recognition frequency data;

[0131] In this embodiment, the license plate recognition frequency calculation is performed on the time series data of the license plate recognition results to obtain the license plate recognition frequency data. The license plate recognition frequency represents the number of times a license plate is recognized per unit time. The entry and exit records of each vehicle are extracted from the license plate recognition system (camera group) in the parking lot, and the recognition timestamps of each vehicle are calculated. Then, the number of times the license plate is recognized within a specific period (such as per hour, per minute) is calculated. For example, assuming that within one hour, the license plates of 30 vehicles are recognized in total, then the license plate recognition frequency is 30 times / hour. If the number of recognized vehicles within a certain period exceeds the normal range of this frequency, it may indicate the phenomenon of repeated recognition. Finally, the obtained license plate recognition frequency data can help the subsequent steps to identify whether there is a situation of repeated recognition of vehicles.

[0132] Combine the vehicle movement behavior time data and the license plate recognition frequency data to calculate the license plate recognition frequency distribution of the vehicle within different time intervals, and perform an exponentially weighted moving average on the license plate recognition frequency distribution to obtain the license plate recognition frequency threshold.

[0133] In this embodiment, using the license plate recognition frequency data and the vehicle movement behavior time data, the recognition frequencies of the vehicle within different time intervals are calculated. The calculation results of the recognition frequencies of each time interval will form a frequency distribution. Then, an exponentially weighted moving average process is performed on the frequency distribution. Assuming that the weighting coefficient is 0.7, a smoothed license plate recognition frequency distribution is generated. The formula for the exponentially weighted average is: weighted average = α × current value + (1 - α) × previous value, where α is the weighting coefficient. On this basis, by setting a threshold (such as the 90th percentile of the frequency value), the license plate recognition frequency threshold is obtained. This threshold will be used in the subsequent steps to determine whether there is an abnormal recognition situation.

[0134] Perform time superposition based on the license plate recognition frequency data and the vehicle movement behavior time data, and identify the vehicles whose license plate recognition frequency is greater than the license plate recognition frequency threshold and the vehicle movement behavior time is less than the average movement time of the scenario vehicles to obtain the repeated recognition vehicle data.

[0135] In this embodiment, using the calculated license plate recognition frequency threshold, the license plate recognition frequencies of all vehicles within different time intervals are compared with this threshold. If the license plate recognition frequency within a certain period is greater than this threshold, and the vehicle movement behavior time within this period is less than the average movement time of the scenario vehicles, then this vehicle is considered a repeatedly recognized vehicle. Set a condition: when the license plate recognition frequency is greater than the threshold (such as 30 times / hour) and the vehicle movement behavior time within this time period is less than 9.92 minutes, it is determined as a repeatedly recognized vehicle. Through this method, the vehicles that are frequently recognized in a short time can be identified, and further abnormal behavior marking and monitoring can be carried out.

[0136] Optionally, the parking environment tracing in step S2 is specifically:

[0137] Reconstruct the parking lot spatial point cloud based on the LiDAR data in the original collected data;

[0138] In this embodiment, the laser radar data in the original collected data is the spatial data obtained by real-time collection of the parking lot by a laser radar (LiDAR) device. The data includes the three-dimensional coordinates of the vehicle and all static objects in the parking lot. The scanning frequency of the laser radar is set to 10Hz, and the maximum scanning distance is set to 30 meters. These point cloud data are spatially aligned through a data processing algorithm (such as an ICP algorithm) to eliminate the displacement and distortion between multiple point cloud data collected at different times. After point cloud cleaning and filtering, a high-precision three-dimensional spatial point cloud model of the parking lot is reconstructed to represent the spatial layout of the ground, pillars, obstacles and parking spaces in the parking lot.

[0139] Extract the video spatiotemporal feature data from the original collected data, and establish a parking lot spatiotemporal twin model based on the parking lot space point cloud and the video spatiotemporal feature data;

[0140] In this embodiment, target detection is performed on each frame of video in the original collected data to extract vehicles and their movement trajectories. Feature data such as speed, direction, and dwell time of each vehicle are extracted in the time series dimension, and the spatiotemporal features in the video are fused with the parking lot space point cloud data extracted from the lidar data. Multimodal data fusion technology is used to align the point cloud data with the spatiotemporal features of the video, and a spatiotemporal twin model of the parking lot is established through a nested convolutional network to form a three-dimensional model that can reflect the dynamic state of the parking lot in real time. The model has a matching accuracy of about 95%, which can ensure low computing latency when the traffic volume is large.

[0141] The abnormal parking data of vehicles is matched with abnormal time points through the parking lot spatiotemporal twin model to trace back the abnormal parking time and environment state and obtain the abnormal parking time and environment data;

[0142] In this embodiment, the established spatiotemporal twin model is used in combination with the abnormal parking data of the vehicle, and time point matching is performed by matching the time series data of the abnormal parking of the vehicle with the timestamp in the spatiotemporal twin model. Through retrospective analysis, the environmental status of the parking lot at that time point is identified, such as the number of remaining parking spaces, visibility of road signs, etc. The time point range in the abnormal parking data can be set to ±5 seconds to ensure matching accuracy. According to the matching results, the parking lot environmental data at that time point is obtained, including the remaining parking spaces in the parking lot, the integrity and clarity of the road signs, etc.

[0143] Calculate the remaining parking lot space based on the abnormal parking time environmental data, and remove the vehicle abnormal parking data with the remaining parking lot space less than 5% at the corresponding time in the vehicle abnormal parking data according to the remaining parking lot space, adjust the vehicle parking abnormal level, and obtain the first vehicle abnormal parking correction data;

[0144] In this embodiment, in the spatio-temporal twin model of the parking lot, based on the remaining parking space data of the parking lot, calculate the remaining parking spaces in the parking lot at the abnormal parking time point. First, subtract the occupied parking spaces from the overall available parking spaces in the parking lot to obtain the remaining parking space ratio. If the remaining parking spaces in a certain period are less than 5%, it is considered that the parking lot space is tense, and the vehicle data repeatedly identified in the abnormal parking during this period is removed. This step is screened by setting a threshold, and the abnormal parking level of the vehicle is adjusted according to the corrected abnormal parking records, and is assigned to three levels: low, medium, and high.

[0145] Identify the road surface markings in the parking lot based on the abnormal parking time environmental data, and calculate the integrity of the road surface markings in the parking lot;

[0146] In this embodiment, through the video data in the spatio-temporal twin model of the parking lot, use image processing algorithms (such as Canny edge detection algorithm and Hough transform) to identify the road surface markings in the parking lot, such as parking lines, signs, etc. Calculate the ratio of the area of the identified road surface markings to the total area of the actual markings to obtain the integrity of the parking lot markings. If the coverage rate of the markings is less than 20%, it is considered that the integrity of the markings is poor, which affects the normal behavior of vehicle parking. In this example, through the algorithm recognition result, when the integrity of the road surface markings in the parking lot is detected to be 12%, this data is marked as abnormal.

[0147] Remove the vehicle abnormal parking data with the road surface marking integrity less than 20% at the corresponding time in the vehicle abnormal parking data according to the road surface marking integrity in the parking lot, adjust the vehicle parking abnormal level, and obtain the second vehicle abnormal parking correction data;

[0148] In this embodiment, combined with the integrity of the parking lot road surface markings, vehicle data with abnormal parking occurring when the marking integrity is less than 20% is screened out. At this time, the markings in the parking lot affect the parking behavior of the vehicle, and the resulting abnormal parking may be an incorrect operation caused by environmental factors. Therefore, adjust the abnormal parking levels of these vehicles, and based on the marking integrity data and the distribution of the prompt robots, eliminate and adjust the levels of parking in the no-parking areas in the abnormal parking data for the corresponding time period to generate corrected abnormal parking data. For example, during a certain period, the marking integrity is low and there is a risk of incorrect parking, but there are prompt robots distributed around the low-integrity markings and the indication is correct. If the vehicle still parks in the no-parking area, it is defined as intentional illegal parking, and accordingly, the parking record of the vehicle in the no-parking area during this period in the abnormal parking data is retained, and the abnormal level is maintained. During a certain period, the marking integrity is low, but there are no prompt robots distributed around the low-integrity markings or the prompt robots give incorrect indications. If the vehicle still parks in the no-parking area, it is defined as incorrect parking, and accordingly, the parking record of the vehicle in the no-parking area during this period in the abnormal parking data is eliminated, and the abnormal level is adjusted.

[0149] Merge the first vehicle abnormal parking correction data and the second vehicle abnormal parking correction data, and perform hierarchical clustering on the vehicle parking abnormal levels to obtain vehicle abnormal parking correction data;

[0150] In this embodiment, after completing the two corrections, merge the correction data of both. Using a hierarchical clustering algorithm (such as the k-means or DBSCAN algorithm), group all the abnormal parking data, and aggregate the vehicles with similar abnormal levels together. By calculating the dispersion of each cluster, identify the main abnormal patterns, and determine the clustering results according to the abnormal levels of the vehicles. Finally, form a corrected abnormal parking data set, and each cluster represents a different type of abnormal parking, such as "incorrect parking" or "intentional illegal parking".

[0151] Extract the environmental data corresponding to the vehicle abnormal parking correction data from the abnormal parking time environmental data to obtain abnormal parking environmental data.

[0152] In this embodiment, according to the merged vehicle abnormal parking correction data, trace back the abnormal parking time environmental data associated with each vehicle. Through spatio-temporal matching, associate the environmental status of the parking lot (such as the remaining parking spaces, the clarity of the road surface markings, etc.) with the time period of abnormal parking. Through data extraction methods, obtain the environmental data during these time periods, and finally obtain the environmental data of the abnormal parking period. These data will be used to further analyze the environmental factors behind the abnormal parking behavior and provide a basis for optimizing the parking lot management in the future.

[0153] Optionally, step S3 is specifically as follows:

[0154] Step S31: Obtain the prompt instruction set of the toll collection robot, and perform temporal matching degree correlation in combination with the robot operation logs in the original collected data. Set the temporal matching threshold to 0.5 to filter the robot operation logs, and obtain the toll collection robot instruction operation data;

[0155] In this embodiment, the prompt instruction set is obtained through the toll collection robot management platform, including all toll collection instructions and related operation commands, such as raising the pole, opening the door, vehicle guidance, etc. These instructions are issued through the robot management platform and stored as a series of log records of timestamps and instruction types. At the same time, the robot operation logs in the original collected data provide the specific operation time series data of the robot in the parking lot, such as time points, executed actions (raising the pole, instruction confirmation, etc.) and their execution results. Next, a temporal matching algorithm, such as dynamic time warping (DTW), is used to perform temporal matching between the prompt instruction set and the robot operation logs. Set the temporal matching threshold to 0.5, that is, when the matching degree between two time points is greater than 0.5, it is considered that their correlation is strong, and these operation records are retained. Finally, the filtered instruction operation data contains all toll collection robot operation records with high matching degrees.

[0156] Step S32: Perform time series decomposition on the abnormal parking environment data, set the number of reconstructed components to 5 and the modulation factor [0.1, 1.0], and obtain the multi-scale abnormal parking environment data;

[0157] In this embodiment, the abnormal parking environment data usually includes the dynamic environment state of the parking lot, such as the occupancy of parking spaces, the integrity of parking signs, etc. In order to analyze the time series change trend of the data, first use the seasonal decomposition method (such as STL decomposition) to perform time series decomposition on the abnormal parking environment data. The number of reconstructed components is set to 5, indicating that the data is decomposed into 5 different components (trend, seasonality, residual, etc.). In addition, set the modulation factor range to [0.1, 1.0] to adjust the influence degree of each component, especially the influence in the seasonal or trend components. In this way, the obtained multi-scale abnormal parking environment data can not only reflect the short-term fluctuations, but also extract the long-term trend information, providing a multi-dimensional perspective for subsequent analysis.

[0158] Step S33: Perform Bayesian causal inference based on the multi-scale abnormal parking environment data and the toll collection robot instruction operation data, and construct a parking environment causal inference model in combination with the abnormal parking behavior labels in the vehicle abnormal parking data;

[0159] In this embodiment, Bayesian causal inference is used to reveal the causal relationship between different environmental factors (such as parking lot space occupancy rate, road surface sign integrity, etc.) and vehicle parking behavior (such as abnormal parking). First, multi-scale abnormal parking environment data and toll robot instruction operation data are fused to form a time-series data set. Each data point includes the state of the parking lot environment, the operation instructions of the robot and its results. On this basis, combined with the abnormal parking behavior labels of the vehicle, a conventional Bayesian network model is used for causal inference to deduce the causal relationship between different environmental factors and machine instructions. For example, if there is a strong causal relationship between the instruction "raise the pole" and "vehicle passing through", it indicates the influence of the robot operation on the abnormal parking behavior. Finally, a causal inference model of the parking environment is constructed, which can reveal the relationship between environmental factors and abnormal vehicle parking. Let X = {x_1, x_2,..., x_n} be the multi-scale abnormal parking environment data, Y = {y_1, y_2,..., y_m} be the toll robot instruction operation data, and Z = {z_1, z_2,..., z_p} be the vehicle abnormal parking data. A Bayesian network model is constructed, where X, Y, and Z are nodes representing different variables. The causal relationship between each node is represented by a directed edge. The specific structure is as follows: X1→Y1→Z1. The causal inference here is carried out through a Bayesian network.

[0160] Step S34: Analyze the causal path of the toll robot instruction judgment according to the causal inference model of the parking environment, calculate the instruction judgment deviation path coefficient, screen the causal paths with the instruction judgment deviation path coefficient greater than the threshold of 0.1, and generate an instruction abnormal causal association matrix;

[0161] In this embodiment, through the constructed causal inference model of the parking environment, different causal paths are analyzed, such as "parking sign integrity → robot instruction execution → abnormal vehicle parking behavior". In this step, the strength of each causal path is parsed by the model, that is, the instruction judgment deviation path coefficient is calculated. This coefficient measures the possibility that the robot instruction affects the abnormal parking behavior in a specific environment. For example, the deviation path coefficient of a certain path is 0.3, indicating that the robot instruction plays a greater role in this path. Set the threshold of the deviation path coefficient to 0.1. When the path coefficient is greater than this threshold, it is considered that this path has a significant impact. Filter out the paths that meet the conditions and generate an instruction anomaly causal association matrix for further analysis of the potential connection between the robot instruction and abnormal parking. Let P = {p_1, p_2,..., p_q} be all path coefficients, and P(i) represents the deviation path coefficient of path i. The deviation path coefficient P(i) measures the difference between the Observed Path and the Expected Path, reflecting the actual influence of the path. Calculate the deviation path coefficient of each path, and the formula is: P(i) = |Observed Path - Expected Path|; Observed Path (the observed path) is specifically: in the actual environment, the data collected according to the execution of the robot instruction and the parking behavior of the vehicle. That is, whether the robot instruction is executed according to the expected path, and whether the resulting parking behavior is abnormal. For example, if the robot still executes a certain instruction when the parking sign integrity is low and causes abnormal parking behavior of the vehicle, then this path is the observed path. Expected Path (expected path): The expected path obtained based on historical data and logical derivation in the causal model. It is based on a theoretical or statistical model to predict how the robot should instruct and the parking behavior of the vehicle should be normal under certain parking sign integrity conditions. For example, if the parking sign integrity is low, theoretically, the possibility of the robot's instruction execution should be reduced to avoid abnormal parking behavior. Set the threshold of the deviation path coefficient to 0.1, and filter out the paths greater than this threshold. Construct a matrix C. Suppose there are q paths, denoted as P_1, P_2,..., P_q. Each path represents the causal relationship from a certain environmental variable (such as the integrity of the parking sign) to a certain behavior (such as abnormal parking behavior). In the matrix C, the element C(i, j) represents the strength of the causal relationship between path i and path j. Specifically, C(i, j) represents the degree of interaction or mutual influence between path i and path j, which is usually quantified in the following way: If path i and path j are independent, then C(i, j) = 0. If path i and path j are correlated, the value of C(i, j) is calculated based on the conditional probability in the causal inference model.For example, if path i and path j simultaneously involve similar environmental features or robot instruction execution patterns, the value of C(i,j) is relatively high, indicating a relatively strong causal association between them. The form of the instruction anomaly causal association matrix is as follows: Among them, C(i,j) represents the strength of the causal relationship between path P_i and path P_j, and the non-zero elements in the matrix indicate a significant interactive effect between these two paths.

[0162] Step S35: Perform hierarchical clustering analysis based on the instruction anomaly causal association matrix, set the maximum number of clusters for hierarchical clustering to 6, extract the abnormal instruction patterns and calculate the instruction anomaly influence factor, and set the influence factor threshold to 0.25, so as to generate an instruction judgment abnormal pattern set.

[0163] In this embodiment, based on the generated instruction anomaly causal association matrix, the hierarchical clustering algorithm is applied to cluster and analyze the instruction anomaly causal paths. Set the maximum number of clusters to 6, that is, divide the causal paths into at most 6 categories. In this way, various possible abnormal instruction patterns can be identified, such as "the lifting rod instruction causes abnormal parking" or "the incomplete parking sign causes misexecution of the robot instruction". Next, calculate the anomaly influence factor of each instruction pattern to measure the degree of influence of the instruction on the abnormal parking behavior. The threshold of the influence factor is set to 0.25. When the influence factor is greater than this threshold, it is considered that the instruction pattern has a significant abnormal influence. Finally, screen out the instruction judgment abnormal pattern set according to the influence factor, providing a basis for the prediction and control of subsequent abnormal parking behaviors.

[0164] Optionally, step S34 is specifically as follows:

[0165] Step S341: Perform causal structure analysis on the parking environment causal inference model, analyze the causal paths for the toll robot instruction judgment, set the window size for calculating the path coefficient to 50 pieces of data, calculate the causal path coefficient, and screen out the causal paths with a path coefficient greater than 0.05 to obtain the preliminary instruction judgment causal path data;

[0166] In this embodiment, a structural analysis is performed on the causal inference model for the parking environment, aiming to analyze the causal path for the toll robot instruction judgment. By using the causal inference model, multiple possible causal paths are identified, such as "Parking sign integrity → Robot instruction execution → Abnormal parking behavior", and the causal coefficient of each path is calculated. The calculation window size is set to 50 data items, that is, the past 50 parking environment data items are analyzed to obtain the path coefficient of each causal path. The path coefficient reflects the influence degree of each causal path. The calculation formula for the path coefficient is as follows: the deviation path coefficient of the path, and the formula is: P(i) = |Observed Path - Expected Path|; the deviation path coefficient of the path, and the formula is: P(i) = |Observed Path - Expected Path|; Assuming that in the analysis of the past 50 data items, the path coefficient of the path "Parking sign integrity → Robot instruction execution → Abnormal parking behavior" is 0.07, then this path meets the screening criteria and is regarded as one of the preliminary instruction judgment causal paths.

[0167] Step S342: Based on the preliminary instruction judgment causal path data, measure the time lag effect, and set the lag order to 3 to correct the causal relationship weight, generating the time series corrected instruction judgment causal path data;

[0168] In this embodiment, the time lag effect is measured through the preliminary instruction judgment causal path data. The time lag effect means that some causal relationships may show their effects after a certain time delay. To more accurately understand the long-term effect of these causal paths on the parking behavior, the lag order is set to 3, that is, it is considered that the influence of the causal path will gradually appear within the next 3 time points. Therefore, it is necessary to perform lag correction on the preliminary instruction judgment causal path data to update the weight of the causal relationship. For the path "Parking sign integrity → Robot instruction execution → Abnormal parking behavior", assuming that it is found that the influence of this path has a delay effect, when the lag order is 3, the weights of the path are corrected from the initial 0.07 to 0.05, 0.03, and 0.02. These corrected weight values reflect the gradual weakening of the lag effect on the path strength at different time points.

[0169] Step S343: According to the time series corrected instruction judgment causal path data, and in combination with the abnormal parking environment variables in the multi-scale abnormal parking environment data, set the influence weight threshold of the abnormal parking environment variables to 0.1, and screen the environment variables with a cumulative contribution rate greater than 90% to calculate the causal path deviation influence factor;

[0170] In this embodiment, by combining the causal path data with the abnormal parking environment variables in the multi-scale abnormal parking environment data for the timing correction instruction, the influence of each environment variable on the causal path is evaluated. Taking "parking sign integrity" as an example, the time step is set to 1 second, and the data at each moment is standardized. Through time series modeling (for example, using the autoregressive moving average model ARMA or the LSTM model), the temporal dependence relationship between each environment variable and the parking path can be extracted. To evaluate the influence of the environment variable on the path, it is necessary to calculate the contribution degree of each environment variable. For this purpose, temporal regression analysis is used to establish a regression model of the environment variable on the instruction judgment causal path. The form of the regression model is: where P ath (t) is the temporal data of the instruction judgment causal path, X i (t) is the value of the environment variable i at time t, β i is the regression coefficient, indicating the influence degree of the environment variable on the path, α is the constant term, and ∈(t) is the error term. The temporal data of the instruction judgment causal path refers to the data sequence arranged in time order of the instructions executed by the toll robot and its influence on environmental changes, system responses, etc. within a given time range. Through the regression model, the regression coefficients β i of each environment variable can be obtained, and these regression coefficients reflect the contribution degree of the environment variable to the path. The larger the regression coefficient, the higher the contribution degree of the environment variable to the path. According to the results of the regression analysis, a contribution degree threshold (for example, 0.1) is set, and the environment variables with a greater influence on the path are screened out. If the regression coefficient β i of a certain environment variable is greater than this threshold, it means that this environment variable has a significant influence on the instruction judgment path. Further, the environment variables with a cumulative contribution rate greater than 90% are screened out, that is, these variables have a higher contribution degree to the causal path and are worthy of further attention. The influence weight threshold of the abnormal parking environment variable is set to 0.1, which means that only when the contribution degree of the environment variable to the path exceeds this value will it be included in the analysis. In addition, the environment variables with a cumulative contribution rate greater than 90% are screened out, and these environment variables have a greater influence on the causal path and are key influencing factors. In the multi-scale abnormal parking environment data, there are multiple environment variables (such as parking sign integrity, road surface condition, weather, etc.). Through analysis, it is found that the influence weight of the "parking sign integrity" variable is 0.12, which meets the threshold requirements, while the influence weight of the "weather condition" variable is 0.08, which is lower than the threshold and is therefore excluded. The two environment variables of "parking sign integrity" and "road surface condition" are screened out as key influencing factors, and the contribution rate exceeds 90%. The formula for setting the causal path deviation influence factor is: ΔP = P ew - P riginal; where P_new is the path coefficient after correction, and P_original is the original path coefficient without correction. The deviation influence factor represents the difference between the path coefficient after correction and the original path coefficient.

[0171] Step S344: Analyze the contribution degree of the instruction judgment path to the causal path deviation influence factor of the instruction judgment, set the deviation contribution threshold to 0.2, and screen the causal paths with a contribution degree higher than 80% to obtain the key instruction judgment deviation path data;

[0172] In this embodiment, the contribution degree of the instruction judgment causal path deviation influence factor is analyzed to identify the most critical causal path among multiple paths. By analyzing the contribution degree of each causal path, the deviation contribution threshold is set to 0.2, indicating that only when the contribution degree of the path exceeds 0.2, it is considered to play an important role in the abnormal parking behavior. By screening the paths with a contribution degree higher than 80%, the final key instruction judgment deviation path data is obtained. Suppose the contribution degrees of different paths are calculated, such as the contribution degree of the path "Parking sign integrity → Robot instruction execution → Abnormal parking behavior" is 0.22, while the contribution degree of the path "Road surface condition → Robot instruction execution → Abnormal parking behavior" is 0.18. Since the contribution degree of the latter is lower than 0.2, it is not included in the screening range. Finally, only the paths with a contribution degree higher than 0.2 are retained, such as "Parking sign integrity → Robot instruction execution → Abnormal parking behavior".

[0173] Step S345: Reconstruct the causal influence of the timing correction instruction judgment causal path data in combination with the key instruction judgment deviation path data, and update the causal path level to generate an instruction abnormal causal association matrix.

[0174] In this embodiment, the causal influence of the timing correction instruction judgment causal path data is reconstructed in combination with the key instruction judgment deviation path data. By performing weighted reconstruction on the influence of the key path, the hierarchical structure of the causal path is updated, and finally an instruction abnormal causal association matrix is generated. This matrix reflects the strength of the causal relationship between each instruction path and provides the potential influence of the robot instruction on the abnormal parking behavior. Suppose the timing correction instruction judgment causal path data includes the paths "Parking sign integrity → Robot instruction execution → Abnormal parking behavior" and "Road surface condition → Robot instruction execution → Abnormal parking behavior". According to the screened key instruction judgment paths, the reconstructed causal influence matrix C is as follows: Among them, the matrix elements represent the degree of causal influence between paths. The update of matrix C reflects the contribution degree and lag effect of the path, further supporting the abnormal causal association analysis of the robot instruction.

[0175] Optionally, step S4 is specifically as follows:

[0176] Step S41: Use the instruction judgment abnormal mode set to perform instruction judgment abnormal time association on the original collected data, and obtain instruction judgment abnormal sensing data and instruction judgment normal sensing data;

[0177] In this embodiment, according to the instruction judgment abnormal mode set, the abnormal time association of the original collected data is performed through time series data analysis, and the instruction judgment abnormal sensing data and the instruction judgment normal sensing data are obtained. Taking the real-time data of the charging robot operation in the parking lot as an example, the original collected data includes multiple sensor data (such as on-vehicle cameras, radars, sensor data, etc.), and the output data of each sensor has a timestamp. The instruction judgment abnormal mode set is an abnormal instruction mode set trained from historical data, which is used to identify whether an abnormality occurs at the current moment. By comparing these data with the predetermined abnormal mode, the abnormal time points are identified. For example, assuming that it is detected that the parking robot gives a wrong parking instruction due to misidentification during a certain period, the sensing data during this period will be classified as "instruction judgment abnormal sensing data". Through the threshold judgment method (for example, setting the abnormal judgment threshold to 95%), these abnormal period data can be screened out. For the normal sensing data, the sensor data of the robot in the normal working state will be classified as "instruction judgment normal sensing data".

[0178] Step S42: According to the instruction judgment abnormal sensing data, perform instruction judgment abnormal order prompt instruction matching on the charging robot instruction operation data, and obtain judgment abnormal order data;

[0179] In this embodiment, the charging robot instruction operation data is matched according to the instruction judgment abnormal sensing data, and judgment abnormal order data is obtained. By comparing the robot instruction operation data with the sensing data, it is found which instruction operations are associated with the abnormal sensing data, and then the abnormal order data is generated. For example, when the parking robot executes the "parking completed" instruction, if an abnormal parking environment (such as inaccurate obstacle detection or blurred parking space identification) is detected during this period, then this order (parking record) will be marked as "judgment abnormal order data". The abnormal order data includes a timestamp, an instruction type (such as a parking instruction), and related environmental abnormality marks (such as abnormal parking space occupancy, unclear parking identification, etc.).

[0180] Step S43: According to the instruction judgment abnormal sensing data, perform instruction judgment normal order prompt instruction matching on the charging robot instruction operation data, and obtain order data to be stored;

[0181] In this embodiment, by matching the normal sensing data with the instructions, the order data to be stored is screened out. Similar to step S42, this step analyzes the normal sensing data to match the corresponding time period of the instruction operation data of the toll collection robot, so as to obtain the order data that conforms to the normal operation. For example, when the robot completes the parking task without an abnormal environment, a normal order data will be recorded. The normal order data includes information such as the start time of parking, the end time of parking, the parking space identifier, and the operation instructions. By matching these normal sensing data and the instruction operation data, a set of valid order data is screened out for subsequent storage and management.

[0182] Step S44: Calculate the instruction execution deviation based on the judged abnormal order data, and construct an order instruction deviation metric matrix in combination with the abnormal parking environment data;

[0183] In this embodiment, based on the judged abnormal order data, the instruction execution deviation is calculated, and an order instruction deviation metric matrix is constructed in combination with the abnormal parking environment data. Suppose in a certain order, the parking robot fails to complete the task according to the predetermined route, resulting in the parking position deviating from the target position. At this time, by comparing the difference between the actual executed instruction and the expected instruction (such as the position of the target parking space), the instruction execution deviation is calculated. The deviation metric calculation formula is: Through this calculation method, the deviation of the instruction execution can be quantified. On this basis, in combination with the abnormal parking environment data, such as factors like "the parking space identifier is not clear", a deviation metric matrix is constructed, where each matrix element represents the degree of a certain instruction deviation and the contribution of the environment to the deviation. Matrix C can be represented in the following form: Among them, δ i,j represents the instruction execution deviation between the i-th order and the j-th environmental variable.

[0184] Step S45: Reconstruct the prompt instruction selection path and the instruction execution parameters in the instruction operation data of the toll collection robot according to the instruction abnormal causal association matrix and the order instruction deviation metric matrix, obtain the robot control instruction, and upload it to the toll collection robot control platform to add the control instruction.

[0185] In this embodiment, according to the order instruction deviation measurement matrix and the instruction anomaly causal association matrix, the prompt instruction selection path and the instruction execution parameters in the instruction operation data of the toll collection robot are reconstructed. First, the instruction anomaly causal association matrix determines which instructions are more likely to cause abnormal behavior in a specific environment by analyzing the relationship between abnormal instructions and environmental variables in historical data. Through this matrix, the instruction selection path can be adjusted, and the instruction execution parameters can be updated when necessary (such as adjusting the parking position target or executing a new task). Suppose that when the robot completes the parking task, some instructions (such as parking steering) show a high abnormal frequency in some environments, and the execution parameters of these instructions will be adjusted accordingly to optimize subsequent operations. Finally, the adjusted control instructions are uploaded to the toll collection robot through the control platform to achieve the update and execution of dynamic instructions. Path reconstruction method: Calculate the abnormal contribution degree of each existing prompt instruction path: S i = ∑ j A ij ·C ij ; where S i represents the degree of influence of the prompt instruction path i on the abnormal environment. If S i exceeds the set threshold (such as 0.5), then this path needs to be adjusted. In the historical data, retrieve the low-deviation path P k matching the current environmental variables. If the alternative path P k has a lower abnormal influence weight (for example, S k <0.2), then use it to replace the original path; if there is no suitable alternative path, enter the instruction execution parameter adjustment stage. Suppose the instruction path P A for the toll collection robot to "guide the vehicle into the parking space" is as follows: "Voice prompt: Please drive forward → Radar identifies the parking space boundary → Issue a reverse prompt: Please reverse slowly → Parking space detection passed, parking completed". However, in a low-light environment, the error of the radar identifying the parking space boundary is relatively high, and the influence weight A2 = 0.7 calculated by the causal association matrix exceeds the set threshold (0.5). The system automatically adjusts the path, such as: "Voice prompt: Please drive forward → Call the infrared camera to detect the parking space boundary (replace the radar) → Issue a reverse prompt: Please reverse slowly → Parking space detection passed, parking completed". If no low-deviation alternative path is found during the path optimization process, the key instruction parameters need to be dynamically adjusted to improve the adaptability of the instructions. Through historical data analysis, calculate the correction coefficient of the parameter to the execution error: Δp = ∑ j C ij ·W j; Wherein, Δp represents the change in the instruction execution parameter that needs to be adjusted. If Δp>0, increase the current instruction parameter (such as the parking position correction value); if Δp<0, reduce the current instruction parameter (such as the speed threshold). Assume that in the "parking position correction" instruction, the original parameter settings of the toll collection robot are: default parking offset correction value: ±5cm; parking speed upper limit: 0.5m / s. However, due to the slippery ground, the parking robot has a large inertia when executing the parking instruction, resulting in parking overshoot. The deviation measurement matrix is calculated to be C=0.6, and the abnormal environment variable weight W=0.4, then: Δp=0.6×0.4=0.24; exceeding the threshold (0.2), so the parameters are dynamically adjusted: the parking offset correction value is adjusted to ±3cm; the parking speed upper limit is reduced to 0.3m / s.

[0186] Optionally, step S5 specifically includes:

[0187] Step S51: performing data encryption preprocessing on the order data to be stored, generating an encrypted summary of the order data, and establishing a data integrity verification parameter set;

[0188] In this embodiment, the order data to be stored is encrypted and pre-processed to protect data security, and a data integrity check parameter set is generated to ensure the integrity and verifiability of the stored data. Order data hash summary generation: SHA-256 hash calculation is performed on the order data to be stored (including order number, payment amount, vehicle information, etc.) to obtain the encrypted summary of the order data: H = SHA-256 (order number | | payment amount | | timestamp); the hash calculation window size is set to 1024 bytes to ensure data block consistency. The Merkle tree structure is used to store the hash summary and construct the integrity check parameter: M 订单 =H 左子节点 ||H 右子节点 ; Record the root hash value H 根 As the integrity check benchmark, the integrity check parameters are obtained.

[0189] Step S52: performing de-identification processing based on the encrypted summary of the order data, removing the user's license plate information and payment ID from the order data to be stored, and generating de-identified order data;

[0190] In this embodiment, the encrypted summary of the order data is de-identified to ensure user privacy and remove sensitive information of the user, such as the license plate number and payment ID, while ensuring data traceability. Identify the license plate number field and payment ID field in the order data and perform desensitization processing on them: License plate number → hashing, H 车牌 = SHA-256 (license plate number); payment ID → hashing, H 付款ID= SHA-256(Payment ID). Using data masking, only the hash value is stored to prevent direct leakage of user information. Generate the de-identified order dataset O 去标识 , where sensitive information fields have been replaced with irreversible hash values to ensure compliant data storage.

[0191] Step S53: Based on the de-identified order data and combined with the instruction, determine the abnormal pattern set and set the dynamic access control policy;

[0192] In this embodiment, based on the de-identified order data and combined with the instruction, determine the abnormal pattern set and set the dynamic access control policy to ensure that different access subjects (such as toll collection robots, control platforms, and audit systems) have appropriate access rights. The specific setting of the dynamic access control policy is as follows: The low-privilege access subject (ordinary robot) can only access the order status intelligently and cannot modify it; The medium-privilege access subject (control platform) can query and adjust the order status but cannot delete it; The high-privilege access subject (audit system) can perform data integrity verification and access all data in case of anomalies. Access restriction based on abnormal patterns: Set the abnormal order recognition rule as R 异常 = {O 去标识 , H 订单 , A 指令异常}; If the order data meets R 异常 , then restrict or record its access rights to prevent abnormal order data from being tampered with.

[0193] Step S54: Use the dynamic access control policy to dynamically adjust the access rights of the toll collection robot control platform to the robot operation log, and obtain the dynamic adjustment record of the access rights;

[0194] In this embodiment, use the dynamic access control policy to adjust the access rights of the robot operation log on the toll collection robot control platform in real time to ensure that the rights of different access subjects are adjusted according to the data status. According to the order data status, set the access right matrix: The corresponding privilege values for different access subjects (1 represents accessible, 0 represents inaccessible). When the data status changes, adjust this matrix. If an access subject attempts to access beyond its privileges (such as a robot attempting to modify the order status), the system records the violation operation and triggers an audit system warning. Record the access right change log L 访问 and store it for subsequent auditing and tracing

[0195] Step S55: Construct a toll collection robot data storage verification chain based on the dynamic adjustment record of the access rights, the de-identified order data, and the data integrity verification parameter set, and store it in the preset distributed ledger system.

[0196] In this embodiment, the access rights are used to dynamically adjust the records, de-identified order data, and data integrity verification parameter set, and a data storage verification chain for the charging robot is constructed and stored in the distributed ledger system to ensure data security, traceability, and immutability. Construct the block data structure: Each order data block contains the de-identified order data O 去标识 , the data integrity verification parameter set M 订单 , and the access rights adjustment record L 访问 . The Merkle Patricia Tree (MPT) structure is used to construct the data storage verification chain: B 订单 = H 根 || H 权限调整 || H 存储时间戳 . Hyperledger Fabric is used as the distributed ledger system, and each order data block is stored in the blockchain network and consensus verification is performed. Any access subject can verify the integrity of the order data by querying the block hash value. H 校验 = SHA-256(B 订单 ). If the calculated H 校验 does not match the hash value stored in the blockchain, it indicates that the data has been tampered with and access is refused.

[0197] Therefore, in all respects, the embodiments should be considered exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application document are intended to be embraced by the present invention.

[0198] The above description is only a specific implementation manner of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A data security storage method for an intelligent toll collection robot based on cloud data, characterized in that, It includes the following steps: Step S1: Obtain the sensing data of the toll collection robot, and perform multi-source data preprocessing on the sensing data of the toll collection robot to obtain the original collected data; Step S2: Identify the vehicle abnormal parking level according to the original collected data to obtain the vehicle abnormal parking data; Trace the parking environment of the vehicle abnormal parking data to obtain the abnormal parking environment data; Step S3: Infer the cause and effect of the implementation of the toll collection robot instruction based on the abnormal parking environment data to obtain the instruction abnormal cause and effect correlation matrix; Identify the instruction judgment abnormal mode set according to the instruction abnormal cause and effect correlation matrix; Step S4: Use the instruction judgment abnormal mode set to divide the original collected data into abnormal order data to obtain the judgment abnormal order data and the order data to be stored; Reconstruct the robot prompt instruction according to the judgment abnormal order data and upload it to the toll collection robot control platform to add a control instruction; Step S5: Perform edge encryption and data de-identification on the order data to be stored, and combine the instruction judgment abnormal mode set to perform dynamic behavior adjustment of access rights to obtain the toll collection robot data storage verification chain.

2. The data security storage method of the intelligent toll collection robot based on cloud data according to claim 1, characterized in that, Step S1 is specifically as follows: Step S11: Obtain the sensing data of the toll collection robot, including the robot operation log, lidar data, camera data, and inertial measurement data; Step S12: Align the time steps of the sensing data of the toll collection robot to obtain the time series aligned sensing data; Step S13: Suppress the noise of the sensing data and standardize the data for the time series aligned sensing data to obtain the standardized sensing data; Step S14: Perform data anomaly detection and redundant feature dimensionality reduction on the standardized sensing data to obtain the dimensionality-reduced sensing data; Step S15: Perform multi-source data feature fusion and environmental information enhancement based on the dimensionality-reduced sensing data to obtain the original collected data.

3. The data security storage method of the intelligent toll collection robot based on cloud data according to claim 2, wherein, Step S15 is specifically as follows: Step S151: Extract the sensor spatio-temporal features based on the dimensionality-reduced sensing data, and perform spatio-temporal correlation analysis on the sensor spatio-temporal features to obtain the sensor spatio-temporal correlation features; Step S152: Perform multi-source sensor data spatio-temporal fusion on the dimensionality-reduced sensing data based on the sensor spatio-temporal correlation features to obtain the toll collection robot fused sensing data; Step S153: Obtain the parking lot weather data and perform data preprocessing to obtain the parking lot weather data to be analyzed; Step S154: Calculate the traffic flow of the parking lot according to the toll collection robot fused sensing data, and perform parking lot environment analysis by combining the traffic flow of the parking lot and the parking lot weather data to be analyzed to obtain the parking lot environment data; Step S155: Use the parking lot environment data to enhance the environmental features of the toll collection robot fused sensing data to obtain the original collected data.

4. The data security storage method of the intelligent toll collection robot based on cloud data according to claim 1, wherein, The vehicle abnormal parking level identification described in Step S2 is specifically as follows: Extract the vehicle parking time series features from the original collected data to obtain the vehicle moving trajectory time series data, vehicle parking time data, and license plate recognition result time series data; Perform vehicle moving behavior association based on the vehicle moving trajectory time series data and the vehicle parking time data to obtain the vehicle moving behavior time data and the vehicle stationary behavior time data; Acquire parking lot structure data, perform road connectivity analysis on the parking lot structure data, define roads with connectivity higher than a preset connectivity threshold as no-parking areas, and obtain no-parking areas of the parking lot; Spatially superimpose the vehicle stationary behavior time data and the parking lot's prohibited parking area, identify vehicles whose stationary time is greater than 10 minutes and whose stationary position is less than 5 cm away from the parking lot's prohibited parking area, and obtain the vehicle data parked in the prohibited parking area; Perform repeated license plate recognition vehicle detection based on the license plate recognition result time series data and vehicle movement behavior time data to obtain repeated recognition vehicle data; The vehicle data of vehicles parked in prohibited parking areas and the vehicle data of repeated identification are merged, and the vehicle parking abnormality level and abnormal parking behavior label are assigned to obtain the vehicle abnormal parking data.

5. The data security storage method of the intelligent toll collection robot based on cloud data according to claim 4, characterized in that, The vehicle detection for repeated license plate recognition is specifically as follows: Calculate the average parking time of the parking lot based on the parking lot structure data and vehicle parking time data; The robot lifting time is calculated based on the time series data of the vehicle identification result and the robot operation log in the original collected data to obtain the robot instruction implementation time; The average moving time of the scene vehicles is calculated based on the average parking time in the parking lot and the robot command execution time; Calculate the license plate recognition frequency on the license plate recognition result time series data to obtain the license plate recognition frequency data; The license plate recognition frequency distribution of vehicles in different time intervals is calculated by combining the vehicle movement behavior time data and the license plate recognition frequency data, and the license plate recognition frequency distribution is subjected to exponential weighted moving average to obtain the license plate recognition frequency threshold; Time superposition is performed based on the license plate recognition frequency data and the vehicle movement behavior time data to identify vehicles whose license plate recognition frequency is greater than the license plate recognition frequency threshold and whose vehicle movement behavior time is less than the average movement time of scene vehicles, and repeatedly identified vehicle data is obtained.

6. The data security storage method of the intelligent toll collection robot based on cloud data according to claim 1, characterized in that, The parking environment tracing described in step S2 is specifically as follows: Reconstruct the parking lot spatial point cloud based on the LiDAR data in the original collected data; Extract the video spatiotemporal feature data from the original collected data, and establish a parking lot spatiotemporal twin model based on the parking lot space point cloud and the video spatiotemporal feature data; The abnormal parking data of vehicles is matched with abnormal time points through the parking lot spatiotemporal twin model to trace back the abnormal parking time and environment state and obtain the abnormal parking time and environment data; Calculating the remaining parking space based on the abnormal parking time environment data, and eliminating the abnormal parking data of vehicles whose remaining parking space at the corresponding time is less than 5% according to the remaining parking space, adjusting the abnormal parking level of the vehicle, and obtaining the first abnormal parking correction data of the vehicle; Identify parking lot road signs based on abnormal parking time environment data and calculate the completeness of parking lot road signs; According to the parking lot road sign integrity, the abnormal parking data of vehicles whose parking lot road sign integrity is less than 20% at the corresponding time are eliminated from the abnormal parking data of vehicles, and the abnormal parking level of the vehicles is adjusted to obtain the second abnormal parking correction data of the vehicles; Merge the first vehicle abnormal parking correction data and the second vehicle abnormal parking correction data, and perform hierarchical clustering on the vehicle parking abnormal level to obtain the vehicle abnormal parking correction data; Extract the environmental data corresponding to the vehicle abnormal parking correction data from the abnormal parking time environmental data to obtain the abnormal parking environmental data.

7. The data security storage method of the intelligent toll collection robot based on cloud data according to claim 1, characterized in that, Step S3 is specifically as follows: Step S31: Obtain the toll robot prompt instruction set, and perform temporal matching degree association in combination with the robot operation log in the original collected data. Set the temporal matching threshold to 0.5 to filter the robot operation log to obtain the toll robot instruction operation data; Step S32: Perform time series decomposition on the abnormal parking environmental data, and set the number of reconstructed components to 5 and the modulation factor [0.1, 1.0] to obtain the multi-scale abnormal parking environmental data; Step S33: Perform Bayesian causal inference based on the multi-scale abnormal parking environmental data and the toll robot instruction operation data, and construct a parking environment causal inference model in combination with the abnormal parking behavior labels in the vehicle abnormal parking data; Step S34: Analyze the causal path of the toll robot instruction judgment according to the parking environment causal inference model, calculate the instruction judgment deviation path coefficient, filter the causal path with the instruction judgment deviation path coefficient greater than the threshold of 0.1, and generate an instruction abnormal causal association matrix; Step S35: Perform hierarchical clustering analysis according to the instruction abnormal causal association matrix, set the maximum number of clusters for hierarchical clustering to 6, extract the abnormal instruction patterns and calculate the instruction abnormal influence factor, and set the influence factor threshold to 0.25 to generate an instruction judgment abnormal pattern set.

8. The data security storage method of the intelligent toll collection robot based on cloud data according to claim 7, characterized in that, Step S34 is specifically as follows: Step S341: Analyze the causal structure of the parking environment causal inference model, parse the causal path of the toll robot instruction judgment, set the window size of the path coefficient calculation to 50 pieces of data, calculate the causal path coefficient, and filter the causal path with the path coefficient greater than 0.05 to obtain the preliminary instruction judgment causal path data; Step S342: Measure the time lag effect based on the preliminary instruction judgment causal path data, and set the lag order to 3 to correct the causal relationship weight to generate the time series corrected instruction judgment causal path data; Step S343: According to the time series corrected instruction judgment causal path data, and in combination with the abnormal parking environment variables in the multi-scale abnormal parking environmental data, set the abnormal parking environment variable influence weight threshold to 0.1, filter the environmental variables with the cumulative contribution rate greater than 90% to calculate the causal path deviation influence factor; Step S344: Analyze the contribution degree of the instruction judgment causal path deviation influence factor to the instruction judgment path, and set the deviation contribution threshold to 0.2, filter the causal path with the contribution degree higher than 80% to obtain the key instruction judgment deviation path data; Step S345: Reconstruct the causal influence of the time series corrected instruction judgment causal path data in combination with the key instruction judgment deviation path data, and update the causal path hierarchy to generate an instruction abnormal causal association matrix.

9. The data security storage method of the intelligent toll collection robot based on cloud data according to claim 1, characterized in that, Step S4 is specifically as follows: Step S41: Use the instruction to judge the abnormal mode set to perform instruction judgment and abnormal time association on the original collected data, and obtain the instruction judgment abnormal sensing data and the instruction judgment normal sensing data; Step S42: According to the instruction judgment abnormal sensing data, perform instruction judgment abnormal order prompt instruction matching on the toll collection robot instruction operation data to obtain the judgment abnormal order data; Step S43: According to the instruction judgment abnormal sensing data, perform instruction judgment normal order prompt instruction matching on the toll collection robot instruction operation data to obtain the order data to be stored; Step S44: Based on the judgment abnormal order data, calculate the instruction execution deviation, and combine the abnormal parking environment data to construct an order instruction deviation metric matrix; Step S45: Reconstruct the prompt instruction selection path and instruction execution parameters in the toll collection robot instruction operation data according to the instruction abnormal causality association matrix and the order instruction deviation metric matrix to obtain the robot control instruction, and upload it to the toll collection robot control platform to add the control instruction.

10. The data security storage method of the intelligent toll collection robot based on cloud data according to claim 1, characterized in that Step S5 is specifically as follows: Step S51: Perform data encryption preprocessing on the order data to be stored, generate an order data encryption digest, and establish a data integrity verification parameter set; Step S52: Based on the order data encryption digest, perform de-identification processing, remove the user license plate information and payment ID in the order data to be stored, and generate de-identified order data; Step S53: According to the de-identified order data, and combined with the instruction judgment abnormal mode set, set a dynamic access control policy; Step S54: Use the dynamic access control policy to dynamically adjust the access rights of the toll collection robot control platform for the robot operation log to obtain the dynamic access right adjustment record; Step S55: Construct a toll collection robot data storage verification chain according to the dynamic access right adjustment record, de-identified order data and data integrity verification parameter set, and store it in the preset distributed ledger system.

Images