A shared bicycle positioning management method and system based on artificial intelligence

Abstract

The application relates to the technical field of shared bicycles, and particularly discloses a shared bicycle positioning management method and system based on artificial intelligence, wherein the method comprises the following steps: acquiring multi-source sensor data uploaded by a vehicle and acquiring environment data about positioning information; estimating data quality information of the multi-source sensor data based on the positioning information and the environment data; acquiring a weight set about the multi-source sensor data according to the data quality information; obtaining fusion data features by weighting and fusing the multi-source sensor data based on the weight set; generating motion state information according to the fusion data features based on a pre-trained state analysis model; and performing operation management based on the motion state information and corresponding positioning information; the method can effectively deal with the problems that sensor data is disturbed, drifts or missing in a city environment, so that an operation platform can more accurately identify the real state of the vehicle, reduce misjudgment, and thus optimize vehicle scheduling, maintenance, anti-theft and other operation management activities.

Description

Technical Field

[0001] This application relates to the field of shared bicycle technology, and more specifically, to a shared bicycle positioning management method and system based on artificial intelligence. Background Technology

[0002] As a vital mode of convenient urban transportation, the operation and management of shared bicycles heavily relies on precise monitoring of vehicle status. Currently, operating platforms are striving to improve the accuracy and reliability of status assessments by integrating multi-source sensor data, such as location information and accelerometer signals, combined with smart lock status and user operation records. The development trend in this field focuses on the application of data fusion technology to support refined operational activities, including vehicle dispatching, billing, and maintenance, thereby optimizing user experience and operational efficiency.

[0003] However, in real-world urban environments, location data often drifts due to signal interference, and accelerometers are susceptible to instantaneous fluctuations caused by external environmental interference, leading to misjudgments of vehicle status. When these multi-source data conflict or are missing, such as in cases of abnormal location information during illegal movement, traditional methods struggle to accurately identify the vehicle's true movement status, impacting operational decisions and the effectiveness of anti-theft mechanisms.

[0004] There is currently no effective technical solution to the above problems. Summary of the Invention

[0005] The purpose of this application is to provide a shared bicycle location management method and system based on artificial intelligence, so as to improve the accuracy of vehicle status judgment and thereby optimize operation and management activities such as vehicle dispatching, maintenance, and anti-theft.

[0006] In a first aspect, this application provides an artificial intelligence-based method for managing the location of shared bicycles, used to locate and analyze the real-time status of bicycles during operation for operational management. The method includes the following steps:

[0007] S1. Obtain multi-source sensor data uploaded by the vehicle, the multi-source sensor data including positioning information, and obtain environmental data about the positioning information stored on the operation platform;

[0008] S2. Estimate the data quality information of the multi-source sensor data based on the positioning information and the environmental data;

[0009] S3. Obtain a weight set for the multi-source sensor data based on the data quality information;

[0010] S4. Obtain fused data features by weighting and fusing the multi-source sensor data based on the weight set;

[0011] S5. Based on the pre-trained state analysis model, generate motion state information according to the features of the fused data;

[0012] S6. Perform operation management based on the motion state information and the corresponding positioning information.

[0013] The method in this application introduces a data quality assessment mechanism based on location information and environmental data, and performs weighted fusion of multi-source sensor data according to the assessment results. This improves the accuracy and reliability of vehicle status judgment for shared bicycles in the case of unreliable or conflicting sensor data, and can effectively address the problems of sensor data interference, drift or missing data in urban environments.

[0014] The aforementioned AI-based shared bicycle location management method, in step S1, includes the step of obtaining environmental data about the location information stored on the operation platform, which comprises:

[0015] S11. Determine the search size based on the accuracy of the positioning information;

[0016] S12. Predict the search base point based on the positioning information;

[0017] S13. Extract environmental data from the environmental data stored on the operation platform, with the search base point as the center and the search size as the diameter, as the environmental data related to the positioning information.

[0018] In summary, the method of this application can adaptively adjust the search range of environmental data according to the accuracy of the positioning information, and determine a more stable search center based on the predicted location. Thus, even when there are errors or drifts in the positioning information, it can more accurately obtain environmental data related to the actual location of the vehicle, improve the accuracy of subsequent data quality assessment and status analysis, and enhance the reliability of the entire shared bicycle positioning management method.

[0019] The aforementioned AI-based shared bicycle positioning management method includes weighted data, where the multi-source sensor data includes positioning information, accelerometer information, and lock status information within a preset time window; and the environmental data includes geographical area information and historical operation information.

[0020] The aforementioned AI-based shared bicycle location management method, weighting, step S2 includes:

[0021] S21. Calculate the stability score of the positioning information based on the position changes of the positioning information at continuous time points;

[0022] S22. Calculate and obtain a communication reliability score based on the environmental data;

[0023] S23. The stability score and the communication reliability score are weighted and calculated to obtain the data quality information.

[0024] The aforementioned AI-based shared bicycle location management method, weighting, step S22 includes:

[0025] S221. Obtain multiple location points within a search size range from the environmental data based on a preset diffusion rule;

[0026] S222. For each location point, calculate the basic communication quality score of that location point based on the geographical area information and historical operation information in the environmental data corresponding to that location point.

[0027] S223. Determine the weighting factors corresponding to the multiple location points based on the distance relationship between the location points and the search base point;

[0028] S224. Based on the weighting factors, the basic communication quality score is weighted and summed to obtain the communication reliability score.

[0029] The aforementioned AI-based shared bicycle location management method, weighting, includes step S3 as follows:

[0030] S31. Based on the data quality information, obtain the weight set from a preset weight configuration table, wherein the weight configuration table includes the weights of different data in the multi-source sensor data under different data quality information.

[0031] The aforementioned AI-based shared bicycle location management method, weighting, step S4 includes:

[0032] S41. Perform feature extraction on the multi-source sensor data to obtain multi-source features;

[0033] S42. The multi-source features are weighted based on the weight set to obtain the weighted multi-source features;

[0034] S43. Based on the positioning information, extract environmental features according to the environmental data;

[0035] S44. Combine the weighted multi-source features and the environmental features to obtain the fused data features.

[0036] The aforementioned AI-based shared bicycle location management method, in step S5, the state analysis model includes an input layer, a feature embedding layer, a bidirectional LSTM layer, a temporal attention layer, a fully connected classification layer, and an output layer connected in sequence. The classification results of the fully connected classification layer include: normal riding, parked, abnormal movement, malfunction, lock not closed, and signal loss.

[0037] The aforementioned AI-based shared bicycle location management method, weighting, step S6 includes:

[0038] S61. Determine whether the vehicle is in a state requiring operational intervention based on the motion status information and positioning information;

[0039] S62. When the vehicle is in the state that requires operational intervention, generate an operation management instruction based on the state and location information required for operational intervention.

[0040] Secondly, this application also provides an artificial intelligence-based shared bicycle positioning management system for locating and analyzing the real-time status of shared bicycles during operation, the system comprising:

[0041] The acquisition module is used to acquire multi-source sensor data uploaded by the vehicle, the multi-source sensor data including positioning information, and to acquire environmental data about the positioning information stored on the operation platform;

[0042] The data quality analysis module is used to estimate the data quality information of the multi-source sensor data based on the positioning information and the environmental data;

[0043] The weight configuration module is used to obtain a weight set for the multi-source sensor data based on the data quality information.

[0044] The feature fusion module is used to obtain fused data features by weighting and fusing the multi-source sensor data based on the weight set;

[0045] The motion analysis module is used to generate motion state information based on the features of the fused data, using a pre-trained state analysis model.

[0046] The management module is used for operation management based on the motion status information and the corresponding positioning information.

[0047] The system in this application introduces a data quality assessment mechanism based on location information and environmental data, and performs weighted fusion of multi-source sensor data according to the assessment results. This improves the accuracy and reliability of vehicle status judgment for shared bicycles in the event of unreliable or conflicting sensor data, and can effectively address the problems of sensor data interference, drift, or missing data in urban environments.

[0048] As can be seen from the above, the shared bicycle positioning management method and system provided in this application based on artificial intelligence, wherein the method of this application introduces a data quality assessment mechanism based on positioning information and environmental data, and performs weighted fusion of multi-source sensor data according to the assessment results, thereby improving the accuracy and reliability of vehicle status judgment of shared bicycles in the case of unreliable or conflicting sensor data. It can effectively deal with the problems of sensor data interference, drift or missing data in urban environments, enabling the operation platform to more accurately identify the real status of vehicles, reduce misjudgments, thereby optimizing vehicle scheduling, maintenance, anti-theft and other operation management activities, and improving operational efficiency and user experience. Attached Figure Description

[0049] Figure 1 A flowchart of an AI-based shared bicycle location management method provided in an embodiment of this application.

[0050] Figure 2 This is a schematic diagram of the network structure of the state analysis model.

[0051] Figure 3 This is a schematic diagram of the structure of an AI-based shared bicycle location management system provided in an embodiment of this application.

[0052] Figure reference numerals: 201, Acquisition module; 202, Data quality analysis module; 203, Weight configuration module; 204, Feature fusion module; 205, Motion analysis module; 206, Management module. Detailed Implementation

[0053] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0054] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0055] Firstly, please refer to Figure 1 and Figure 2 This application provides an artificial intelligence-based method for managing the location of shared bicycles, used to locate and analyze the real-time status of bicycles during operation for operational management. The method includes the following steps:

[0056] S1. Obtain multi-source sensor data uploaded by the vehicle, including positioning information, and obtain environmental data about the positioning information stored on the operation platform;

[0057] S2. Estimate the data quality information of multi-source sensor data based on positioning information and environmental data;

[0058] S3. Obtain a weight set for multi-source sensor data based on data quality information;

[0059] S4. Obtaining fused data features based on weighted fusion of multi-source sensor data using weight sets;

[0060] S5. Based on a pre-trained state analysis model, generate motion state information according to the features of the fused data;

[0061] S6. Conduct operation management based on motion status information and corresponding location information.

[0062] Specifically, multi-source sensor data refers to the collection of information gathered by various sensors mounted on the vehicle (shared bicycle). This data can include, but is not limited to, data collected by positioning modules, accelerometers, gyroscopes, electronic locks, battery status sensors, etc., primarily for sensing the vehicle's real-time status and location. Environmental data refers to auxiliary information related to the vehicle's location stored on the operating platform. This data can include, but is not limited to, geographic information system data, communication signal coverage maps, historical operational data, regional attribute information, etc., primarily for providing contextual information about the vehicle's environment and assisting in judging the reliability of sensor data or the vehicle's status. Data quality information refers to the evaluation results of the reliability or validity of the acquired multi-source sensor data. This can be expressed as a score, confidence level, or probability distribution, primarily for quantifying the usability of different data sources and providing a basis for subsequent data processing. The weight set refers to the set of weights assigned to different data sources in the multi-source sensor data. This weight set can be dynamically adjusted based on data quality information or obtained from preset rules, primarily for reflecting the importance or reliability differences of different data sources during data fusion. Weighted fusion refers to the process of weighting multi-source sensor data according to a weight set and integrating it into a unified data representation. This is mainly to make comprehensive use of multi-source information and reduce the impact of low-quality data.

[0063] More specifically, the state analysis model refers to an artificial intelligence model used to identify the current motion or stationary state of a vehicle. It can employ machine learning or deep learning models such as support vector machines, random forests, convolutional neural networks, and recurrent neural networks. Its primary purpose is to determine the vehicle's true state based on the fused data features. The fused data features refer to the comprehensive feature representation formed by multi-source sensor data and potential environmental features after weighted fusion processing. This is primarily to reflect the vehicle's true state more comprehensively and accurately, serving as the input to the state analysis model. Motion state information refers to the judgment result of the state analysis model regarding the vehicle's current state. Operational management refers to taking corresponding operational actions based on the vehicle's motion state and location information. This can include vehicle dispatching, maintenance and repair assignment, anomaly alarms, and billing processing, primarily aimed at optimizing operational efficiency and user experience.

[0064] Specifically, the method of this application first acquires multi-source sensor data, including location information, uploaded by the vehicle, and obtains environmental data related to this location information from the operation platform. Then, based on the acquired location information and environmental data, a data quality assessment is performed on the multi-source sensor data to generate data quality information. Based on the assessed data quality information, a weight set is determined, reflecting the reliability of different data sources under the current conditions. Using this weight set, the multi-source sensor data is weighted and fused to form fused data features. These fused data features are input into a pre-trained state analysis model, which outputs the vehicle's current motion state information. Finally, based on the determined motion state information and the corresponding location information, appropriate operation and management operations are executed. The entire process, by introducing data quality assessment and weighted fusion, enables subsequent state analysis to rely on more reliable data representations, thereby improving the accuracy of state judgment and providing a foundation for refined operation and management.

[0065] Through the above-mentioned scheme, the method of this application introduces a data quality assessment mechanism based on location information and environmental data, and performs weighted fusion of multi-source sensor data according to the assessment results. This improves the accuracy and reliability of vehicle status judgment for shared bicycles in the case of unreliable or conflicting sensor data. It can effectively deal with the problems of sensor data interference, drift or missing data in urban environments, enabling the operation platform to more accurately identify the true status of vehicles, reduce misjudgments, and thus optimize vehicle scheduling, maintenance, anti-theft and other operation and management activities, thereby improving operational efficiency and user experience.

[0066] In some preferred embodiments, step S1, the step of obtaining environmental data about location information stored by the operation platform, includes:

[0067] S11. Determine the search size based on the accuracy of the positioning information;

[0068] S12. Predict search base points based on location information;

[0069] S13. Extract environmental data from the environmental data stored on the operation platform, with the search base point as the center and the search size as the diameter, as environmental data related to the location information.

[0070] Specifically, the search size is preferably determined by combining a preset mapping relationship with the accuracy of the positioning information. The preset mapping relationship refers to a pre-established correspondence rule used to convert the input positioning accuracy value into a specific search size value, which can be implemented using a lookup table, mathematical formula, or a model trained based on empirical data. In other embodiments, step S12 can also directly determine the search base point based on the positioning information.

[0071] More specifically, the search size refers to the size parameter used to limit the spatial range from which environmental data is extracted from the operations platform. For example, it can represent the diameter or radius of a circular area, or the side length of a square area. The predicted search base point refers to a reference point estimated based on the current positioning information using a certain algorithm, which is more stable or closer to the vehicle's actual position than the original positioning point. This can be achieved using filtering algorithms (such as Kalman filtering), smoothing algorithms (such as moving average), or by combining historical trajectory information.

[0072] Specifically, the accuracy of sensors used in vehicles from different production batches or models varies, and this accuracy difference affects the relevance of environmental data related to that location. In step S11, the method of this application dynamically calculates and determines a search size using the accuracy information contained in the positioning information and a preset mapping relationship. The higher the positioning accuracy, the smaller the calculated search size; conversely, the lower the accuracy, the larger the calculated search size. This ensures that the search range can be adaptively adjusted according to the reliability of the positioning. Next, in step S12, a search base point is predicted based on the current positioning information. This prediction process can smooth fluctuations in the original positioning point or combine other information (such as vehicle motion state) to estimate a more stable position as the search center. Finally, in step S13, combining the search size calculated in step S11 and the search base point predicted in step S12, all environmental data located within a circular area centered on this base point and with this size as its diameter are extracted from the environmental data stored on the operation platform. This extracted environmental data is used as environmental data related to the current positioning information. In this way, the solution overcomes the problems caused by directly using inaccurate original positioning points and fixed search ranges, and the acquired environmental data is more likely to be correlated with the vehicle's actual location. This optimized environmental data serves as more reliable input for subsequent data quality estimation and data fusion steps, thereby improving the accuracy and reliability of the entire shared bicycle positioning management method. This method, which dynamically adjusts the search range based on positioning accuracy and determines the search center based on predicted location, makes the acquired environmental data more closely reflect the actual situation of the vehicles, providing a more solid foundation for subsequent data processing and status analysis.

[0073] Through the above scheme, the method of this application can adaptively adjust the search range of environmental data according to the accuracy of the positioning information, and determine a more stable search center based on the predicted location. In this way, even if there are errors or drifts in the positioning information, it can more accurately obtain environmental data related to the actual location of the vehicle, improve the accuracy of subsequent data quality assessment and status analysis, and enhance the reliability of the entire shared bicycle positioning management method.

[0074] In some preferred embodiments, the multi-source sensor data includes positioning information, accelerometer information, and lock status information within a preset time window; the environmental data includes geographical area information and historical operation information.

[0075] Specifically, the preset time window refers to a time period set in advance by the system to limit the time range for acquiring sensor data. It can be configured according to actual operational needs so that the acquired positioning information, accelerometer information, and lock status information are presented as sequential information within the time window (i.e., containing multiple sets of data).

[0076] More specifically, location information refers to data reflecting the vehicle's geographical location, which can be obtained using technologies such as the Global Positioning System (GPS), BeiDou Navigation Satellite System, cellular network base station positioning, or Wi-Fi positioning. Accelerometer information refers to data reflecting changes in the vehicle's motion state (such as acceleration, deceleration, stationary state, and vibration), which can be obtained using microelectromechanical systems (MEMS) accelerometer sensors. Lock status information refers to data reflecting the current state of the vehicle's smart lock (such as locked, unlocked, or faulty), which can be obtained using sensors integrated within the smart lock (such as Hall effect sensors or microswitches).

[0077] More specifically, environmental data refers to non-sensor-based real-time data collected related to the vehicle's location or operating area, which can be stored in the operating platform's database. Geographic area information refers to the geographic attribute data of the vehicle's location, which may include, but is not limited to, area type (e.g., indoor, outdoor, underground), terrain features, no-parking zone delineation, etc., and can be obtained from geographic information system databases or map services. Historical operating information refers to historical operational data statistics related to the vehicle or its location, which may include, but is not limited to, historical communication quality, historical fault records, regional vehicle usage patterns, historical positioning accuracy statistics, etc., and can be extracted from the operating platform's historical database.

[0078] More specifically, location information provides the basic position, accelerometer information provides changes in motion status, and lock status information provides the usage status. These sensor data, used in combination, can complement and verify each other, overcoming the limitations of a single data source. For example, when the location signal is poor, accelerometer and lock status information can help determine whether the vehicle is still moving or has been locked and parked. Geographic area information provides the geographic attributes of the vehicle's location, helping to assess the reliability of the location signal or determine whether it is within a specific operating area. Historical operational information provides historical data statistics for the area or vehicle, serving as prior knowledge to help determine the reliability of current data or predict the vehicle's status.

[0079] More specifically, the method of this application provides richer and more targeted information input for subsequent data quality assessment, weighted fusion, and state analysis steps by acquiring these specific types of data. The selection of these data types allows for a more comprehensive reflection of the vehicle's true state and its environment, contributing to improved accuracy in data quality assessment. This enables weighted fusion to more effectively utilize data from different sources and provides more efficient features for the state analysis model, ultimately enhancing the accuracy and reliability of judging the motion state of shared bicycles.

[0080] In some preferred embodiments, step S2 includes:

[0081] S21. Calculate the stability score of the location information based on the location changes at continuous time points;

[0082] S22. Calculate and obtain a communication reliability score based on environmental data;

[0083] S23. Calculate the stability score and communication reliability score by weighting them to obtain data quality information.

[0084] Specifically, step S21 quantifies the reliability of the positioning data by evaluating whether the changes between location points conform to the expected motion pattern or whether abnormal jumps occur. This can be achieved by methods such as calculating the rate of change of distance between consecutive location points, detecting abnormal peaks in velocity or acceleration, or applying smoothing algorithms and evaluating the deviation between the original data and the smoothed results.

[0085] More specifically, step S22, calculating and obtaining a communication reliability score based on environmental data, refers to using data related to the vehicle's current location or historical operations to assess the reliability of data transmission from the vehicle to the operating platform under that environment. This can be achieved by methods such as querying historical communication success rates for a specific geographical area, analyzing current network signal strength, or combining models of the impact of factors such as weather and terrain on communication quality.

[0086] More specifically, data quality information comprehensively reflects the overall availability and reliability of multi-source sensor data. It can be achieved through methods such as simple linear weighted summation, lookup table-based mapping, or using machine learning models trained on historical data to obtain weighting functions.

[0087] Specifically, step S21 directly assesses the stability of the positioning data by analyzing the positional changes of the positioning information at continuous time points, identifying and quantifying drift or fluctuations caused by factors such as signal interference. Step S22 utilizes environmental data to assess the reliability of the communication link for data transmission, considering communication differences under different regions or historical conditions. These two scores evaluate multi-source sensor data from two dimensions: the quality of the data source itself and the quality of data transmission. Step S23 weights and combines the scores from these two dimensions to obtain a comprehensive data quality information. This combination method makes the data quality assessment more comprehensive and can more accurately reflect the actual reliability of multi-source sensor data. More specifically,

[0088] Through the above calculations, the method of this application can more accurately assess the actual reliability of multi-source sensor data of shared bicycles, overcome the problem of inaccurate data quality information that may be caused by relying solely on general estimations, and improve the accuracy of subsequent status analysis and operation management.

[0089] In some preferred embodiments, step S22 includes:

[0090] S221. Obtain multiple location points within the search size range from the environmental data based on preset diffusion rules;

[0091] S222. For each location point, calculate the basic communication quality score of that location point based on the geographical area information and historical operation information in the environmental data corresponding to that location point.

[0092] S223. Determine the weighting factors corresponding to multiple location points based on the distance relationship between the location points and the search base point;

[0093] S224. Based on the weighting factors, the basic communication quality scores are weighted and summed to obtain the communication reliability score.

[0094] Specifically, the preset diffusion rule refers to the strategy used to determine which specific location points to select for environmental data sampling within the search size range. It can be implemented using rules based on grid partitioning, rules based on random sampling, or rules based on specific pattern distribution.

[0095] More specifically, the basic communication quality score is a quantitative assessment of the communication environment quality at a specific location. It can be calculated by comprehensively considering information such as the signal coverage level, historical data transmission success rate, and network latency of the geographical area to which the location belongs.

[0096] More specifically, the weighting factor is a numerical value used to measure the contribution of the basic communication quality score of each location point to the final communication reliability score. It can be calculated using a function related to the distance from the location point to the search base point (such as the reciprocal of the distance, a Gaussian decay function, etc.).

[0097] Specifically, this solution addresses the problem of accurately assessing the communication reliability of shared bicycles in complex environments by employing a spatially weighted fusion approach based on environmental data. First, multiple location points are acquired from the environmental data stored on the operations platform, based on preset diffusion rules within a search size. This process expands the sampling range of environmental data, avoiding potential errors from relying on a single location point. Next, for each acquired location point, a basic communication quality score is calculated using the geographical area information and historical operational information from the corresponding environmental data, quantifying the communication environment characteristics of that point. Subsequently, a corresponding weighting factor is determined based on the distance relationship between each location point and the search baseline, ensuring that location points closer to the search baseline have a greater impact on the assessment results. Finally, the basic communication quality scores of each location point are weighted and summed with their corresponding weighting factors to obtain the final communication reliability score.

[0098] Through the above-mentioned solution, this application can overcome the problems of inaccuracy and lack of robustness caused by relying solely on environmental data from a single location point to assess communication reliability. Especially when the positioning accuracy is low or the environmental data is sparse, it can more accurately reflect the impact of the overall communication environment of the area where the vehicle is located on the communication quality and avoid distortion of the communication reliability score.

[0099] In some preferred embodiments, in step S222, the geographic area information corresponding to each location point includes environmental simplicity, and the historical operational information corresponding to each location point includes historical communication success rate. The process of calculating the basic communication quality score for that location point includes:

[0100] The product of the calculation environment simplicity and the historical communication success rate is used as the basic communication quality score.

[0101] Specifically, environmental simplicity refers to a quantitative indicator reflecting the degree to which the objective environment of a location affects communication signals. It can be obtained through analysis of Geographic Information System (GIS) data, such as by comprehensively calculating a value based on factors like building density, terrain undulation, vegetation cover, and the distribution of electromagnetic interference sources. A higher value indicates a more favorable environment for communication. Historical communication success rate refers to the percentage of successful communication attempts made at that location throughout history.

[0102] More specifically, the method of this application calculates the basic communication quality score of a location by multiplying the environmental simplicity of the location point by the historical communication success rate. This score comprehensively reflects the objective environmental conditions and historical actual communication performance of the location point, thereby enabling a more accurate and effective assessment of the communication capability of the location point. This provides reliable basic data for subsequent calculation of communication reliability score, and thus improves the accuracy of the entire method in judging the vehicle status.

[0103] In some preferred embodiments, step S3 includes:

[0104] S31. Based on the data quality information, obtain the weight set from the preset weight configuration table. The weight configuration table includes the weights of different data in the multi-source sensor data under different data quality information.

[0105] Specifically, a pre-defined weight configuration table refers to a pre-established data structure or model used to store or define the relationship between data quality information and the corresponding weight set. A weight set is a set of values, each corresponding to a specific data source type in multi-source sensor data, used to indicate the relative importance of that data source in the data fusion process; it can be represented by a vector or a list.

[0106] Specifically, the core of step S31 lies in utilizing a pre-defined weight configuration table to find and obtain the corresponding weight set based on the currently evaluated data quality information. The key to this approach is that the weight configuration table is pre-designed and configured, containing the weights that should be assigned to different data source types in multi-source sensor data under different data quality levels. Through this mapping relationship, this scheme can dynamically adjust the importance of each data source in the subsequent fusion process based on the actual quality of the current data, especially considering the impact of different communication qualities on the reliability of different sensor data. For example, when poor communication quality leads to a decrease in the reliability of positioning data, the weight configuration table will provide a weight set that reduces the weight of positioning data while potentially increasing the weight of other relatively unaffected data (such as accelerometer data) based on the evaluated low data quality information. In this way, in the subsequent weighted fusion steps, the importance of more reliable data sources under the current conditions can be effectively highlighted, and the interference from unreliable data sources can be reduced, thereby generating more accurate fused data characteristics that better reflect the true state of the vehicle, providing a more solid data foundation for subsequent state analysis and operation management. This step incorporates the assessment results of the data quality information and provides key weight parameters for subsequent data weighting and fusion, enabling the entire data processing flow to adaptively adjust according to the actual reliability of the data and improving the accuracy of the final state judgment.

[0107] Through the above scheme, this application can dynamically adjust the fusion weights of different data sources in multi-source sensor data based on the data quality information obtained from the evaluation, especially the data quality information reflecting changes in communication quality. Therefore, when data quality is poor or the communication environment is harsh, the weight of more affected data can be reduced, while the weight of relatively reliable data can be increased, thereby reducing the negative impact of low-quality data on the fusion results.

[0108] In some preferred embodiments, step S4 includes:

[0109] S41. Extract features from multi-source sensor data to obtain multi-source features;

[0110] S42. Weight the multi-source features based on the weight set to obtain the weighted multi-source features;

[0111] S43. Based on location information, extract environmental features from environmental data;

[0112] S44. Combine the weighted multi-source features and environmental features to obtain fused data features.

[0113] Specifically, feature extraction refers to the process of converting raw multi-source sensor data into a structural representation, which can be achieved using signal processing or data conversion techniques; environmental features refer to the representation extracted from environmental data that is associated with the current location.

[0114] Specifically, the above processing first extracts features from multi-source sensor data to obtain multi-source features. These multi-source features are then weighted based on a weight set, which reflects the data quality of different data sources. Weighting adjusts the influence of features from different data sources. Simultaneously, based on the current location information, location-related environmental features are extracted from the environmental data stored on the operation platform. Finally, the weighted multi-source features are concatenated with the extracted environmental features to form fused data features. This process combines vehicle state information reflected by sensor data (considering data quality) with contextual information provided by environmental data, generating a representation containing both aspects. In this way, the comprehensiveness of the fused data features is improved, providing representative input for subsequent state analysis and overcoming the limitations of relying solely on sensor data or indirectly utilizing environmental information. Therefore, the obtained fused data feature information is comprehensive, accurately reflecting the vehicle's true state and improving the accuracy of subsequent state analysis.

[0115] In some preferred embodiments, step S41 includes:

[0116] S411. Extract the corresponding features from the multi-source sensor data for positioning information, accelerometer information and lock status information respectively;

[0117] S412. Combine the extracted features to obtain multi-source features.

[0118] Specifically, step S411 performs feature extraction independently for each type of data. For positioning information, features reflecting the vehicle's motion state can be extracted, such as kinematic features like speed, acceleration, or direction obtained by calculating position changes at consecutive time points. For accelerometer information, features reflecting the vehicle's vibration or impact state can be extracted, such as calculating the mean, variance, standard deviation, and root mean square value of acceleration data, or performing frequency domain analysis to extract energy features at specific frequencies. For lock status information, since it is usually a discrete value (e.g., locked, unlocked), its status value or state sequence can be directly used as features. This method of extraction allows for the use of the most suitable algorithms and methods to capture key information based on the characteristics of each data type.

[0119] More specifically, multi-source features contain high-quality information extracted specifically from different data sources, providing comprehensive input for subsequent data processing.

[0120] In some preferred embodiments, in step S5, such as Figure 2 As shown, the state analysis model consists of an input layer, a feature embedding layer, a bidirectional LSTM layer, a temporal attention layer, a fully connected classification layer, and an output layer connected in sequence. The classification results of the fully connected classification layer include: normal riding, parked, abnormal movement, malfunction, lock not closed, and signal loss.

[0121] Specifically, a feature embedding layer is a network layer that maps the original input feature vector to a higher-dimensional, more expressive feature space. It can be implemented using linear transformation layers, non-linear transformation layers, or fully connected layers incorporating activation functions. A bidirectional LSTM layer is a recurrent neural network layer capable of simultaneously handling the temporal and reverse-temporal dependencies of the input sequence. It can be implemented by concatenating or summing the outputs of the forward and backward LSTM layers. A temporal attention layer is a network layer that assigns different weights based on the importance of different time steps in the input sequence and then weights and integrates the sequence information accordingly. It can be implemented using additive attention, multiplicative attention, or self-attention mechanisms.

[0122] Specifically, the state analysis model receives weighted fusion data features as input. These fusion data features are obtained by weighting multi-source sensor data and environmental data of the vehicle, combined with data quality information, and include the vehicle's temporal information over a period of time, as well as location-related environmental information. The input layer receives these fusion data feature sequences. The feature embedding layer transforms the original fusion features into a more abstract feature representation that is easier for the model to learn. The bidirectional LSTM layer processes the embedded feature sequence. By considering the context of the sequence, this layer can capture long-term dependencies and bidirectional correlations in the vehicle's motion patterns, such as understanding whether a stationary state is a parking after riding or preparation for abnormal movement. The temporal attention layer works on the output of the bidirectional LSTM layer. It evaluates the importance of each time step in the sequence to the final state judgment and generates a set of attention weights. With these weights, the model can automatically focus on the key moments or data patterns that best indicate the vehicle's current state, such as sudden changes in acceleration, changes in lock status, or long periods of position invariance. This layer integrates the temporal information with attention weights into a single vector representation. The fully connected classification layer receives the weighted vector obtained through an attention mechanism and inputs it into one or more fully connected layers, ultimately mapping it to a predefined vehicle state category. The output layer outputs the vehicle state judgment result obtained by the classification layer. This model, combined with the preceding steps, forms a complete state analysis process. The preceding steps acquire multi-source data and environmental data, and perform weighted fusion based on data quality, providing the state analysis model with high-quality, information-rich input. This data preprocessing effectively mitigates potential noise, drift, or conflict issues in the original data, enabling the state analysis model to learn and make judgments on a more reliable data foundation. The state analysis model then utilizes its deep learning structure, particularly bidirectional temporal processing and temporal attention mechanisms, to deeply mine the complex motion patterns and state transition laws of vehicles from these fused features. This approach, combining intelligent data fusion with advanced deep learning models, can more accurately identify the true state of vehicles, especially in complex or abnormal scenarios that are difficult to handle with traditional methods, thus effectively solving the problem of state misjudgment in existing technologies.

[0123] Through the above design, the method of this application can effectively process the temporal motion data and environmental characteristics of shared bicycles. As a result, the model can more accurately distinguish between different states such as normal riding, parking, and abnormal movement, reducing the probability of state misjudgment in complex or abnormal scenarios.

[0124] In some preferred embodiments, the single temporal vector of the output of the temporal attention layer satisfies:

[0125] (1)

[0126] Where L1 is a single time-series vector, T1 is the time-series length of the multi-source sensor data, and h t Let α be the output vector of the bidirectional LSTM layer at time t. t for h t The normalized weights satisfy:

[0127] (2)

[0128] Among them, score(h t ) represents the importance score at time t, score(h) i Let ) be the importance score at time i, tanh(⋅) be the hyperbolic tangent activation function, W be the attention weight matrix obtained during training, b be the bias vector obtained during training, v be the weight vector obtained during training, and T be the transpose.

[0129] Specifically, temporal attention is used to calculate and assign different weights based on the characteristics of different time steps in the input sequence, thereby highlighting the role of important time steps and weakening the influence of unimportant time steps when integrating sequence information. score(h t () refers to the measure of the output vector h of the bidirectional LSTM layer at time t. t Importance score, which is calculated by applying h t A linear transformation is performed, a nonlinear activation function (tanh) is applied, and then the result is calculated by multiplying it with the trained weight vector v. This calculation method allows the model to learn h. t Which feature combinations are relevant to the importance of the current time point? The attention weight matrix W, bias vector b, and weight vector v obtained during training are parameters that the model automatically adjusts using learning data during training. For example, a state analysis model is trained through supervised learning, using a labeled vehicle state dataset, optimizing the loss function (such as cross-entropy) to update parameters W, b, and v, which together determine how to calculate the importance score for each time point. The hyperbolic tangent activation function tanh(⋅) is a commonly used nonlinear function used to introduce nonlinear transformation capabilities, enhancing the model's expressive power.

[0130] Specifically, this scheme receives a series of time vectors {h1, h2, ..., h} output from a bidirectional LSTM layer. T}, where T is the time series length. For each time point t in the sequence, calculate the importance score (score(ht)) of its corresponding vector ht. t The calculation process for this score involves taking h... tThe process involves affine transformation, inputting the result into a hyperbolic tangent function, and finally multiplying the output of the activation function by the vector v. After calculating the scores at all time points, the softmax function is used to normalize these scores, yielding the normalized weight α for each time point t. t These weights reflect the relative importance of each time point in the entire sequence. Finally, the output vector h at each time point t is... t Its corresponding normalized weight α t Weighted summation yields a single temporal vector L1. This vector L1 is a compressed representation of the entire temporal sequence, where information from important time points is amplified while information from less important time points is de-emphasized. This single vector L1 can then be input into subsequent fully connected classification layers for the final vehicle state determination. In this way, the state analysis model can automatically focus on information at the most critical time points for determining the vehicle state, thereby improving the accuracy of state analysis. This temporal attention mechanism, as part of the state analysis model, works in conjunction with the bidirectional LSTM layer, enabling the model to more effectively extract discriminative information from the temporal features of multi-source sensor data, especially under conditions of fluctuating data quality or sparse key information, enhancing the model's ability to determine the true motion state of the vehicle.

[0131] In some preferred embodiments, step S6 includes:

[0132] S61. Determine whether the vehicle is in a state requiring operational intervention based on motion status information and location information;

[0133] S62. When a vehicle is in a state that requires operational intervention, generate an operational management instruction based on the required operational intervention status and location information.

[0134] Specifically, the state requiring operational intervention refers to a pre-defined category of abnormal or illegal vehicle status that requires intervention from the operations platform. This category can be represented by an enumeration value or a status code, and can include illegal stationary parking, malfunction, abnormal movement, and unlocked vehicle.

[0135] More specifically, motion state information can be implemented using discrete state labels (e.g., riding, parked, abnormal movement, etc.).

[0136] More specifically, operation management instructions refer to instructions or information generated by the operation platform based on vehicle status and location, used to guide operators or systems to perform specific operations. These instructions can be implemented using structured data messages, task assignment notifications, or system control commands.

[0137] Specifically, by combining motion status and location information, the system can identify pre-defined specific states requiring intervention from the operations platform. These states are explicitly defined as illegal parking, malfunction, abnormal movement, and unlocked vehicles. This judgment mechanism based on specific states and locations allows operations management to focus on vehicles that truly require attention and intervention, improving management efficiency and accuracy. Secondly, when a vehicle is determined to be in a state requiring operational intervention, operations management instructions are generated based on the required intervention state and location information. This step is a direct application of the previous judgment result. Once a specific state requiring intervention is identified, the system automatically generates corresponding operations management instructions based on the vehicle's current location information. By generating instructions based on the specific intervention state and vehicle location, the timeliness and relevance of operational responses are ensured, enabling the operations platform to quickly and effectively handle various abnormal situations. This solution is based on the fusion processing of multi-source sensor data and the generation of motion state information based on a trained model. By utilizing this more accurate and reliable state information and corresponding positioning information, it is possible to more accurately determine whether a vehicle is in a state requiring intervention. This makes the subsequent generated operation and management instructions more targeted and effective, improves the intelligence level and responsiveness of the entire operation and management process, helps reduce vehicle asset losses, enhances user experience, and optimizes operating costs.

[0138] Secondly, please refer to Figure 3 Some embodiments of this application also provide an artificial intelligence-based shared bicycle location management system for locating and analyzing the real-time status of shared bicycles during operation, the system comprising:

[0139] The acquisition module 201 is used to acquire multi-source sensor data uploaded by the vehicle, including positioning information, and to acquire environmental data about the positioning information stored on the operation platform.

[0140] Data quality analysis module 202 is used to estimate data quality information of multi-source sensor data based on positioning information and environmental data;

[0141] The weight configuration module 203 is used to obtain a weight set for multi-source sensor data based on data quality information.

[0142] Feature fusion module 204 is used to obtain fused data features based on weighted fusion of multi-source sensor data using a weight set;

[0143] Motion analysis module 205 is used to generate motion state information based on the features of fused data, using a pre-trained state analysis model.

[0144] The management module 206 is used for operation management based on motion status information and corresponding location information.

[0145] The system in this application introduces a data quality assessment mechanism based on location information and environmental data, and performs weighted fusion of multi-source sensor data according to the assessment results. This improves the accuracy and reliability of vehicle status judgment for shared bicycles in the event of unreliable or conflicting sensor data. It can effectively address the problems of sensor data interference, drift, or missing data in urban environments, enabling the operation platform to more accurately identify the true status of vehicles, reduce misjudgments, and thus optimize operation and management activities such as vehicle scheduling, maintenance, and anti-theft, thereby improving operational efficiency and user experience.

[0146] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0147] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0148] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0149] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1.A shared bicycle positioning management method based on artificial intelligence, for positioning and analyzing the real state of a vehicle in order to perform operation management in a shared bicycle operation, characterized by, The method comprises the following steps: S1, acquiring multi-source sensor data uploaded by a vehicle, the multi-source sensor data comprising positioning information, and acquiring environment data about the positioning information stored by an operation platform; S2, estimating data quality information of the multi-source sensor data based on the positioning information and the environment data; S3, acquiring a weight set about the multi-source sensor data according to the data quality information; S4, acquiring fusion data features by weighted fusion of the multi-source sensor data based on the weight set; S5, generating motion state information according to the fusion data features based on a pre-trained state analysis model; S6, performing operation management based on the motion state information and corresponding positioning information; The step of acquiring environment data about the positioning information stored by the operation platform comprises: S11, determining a search size based on the accuracy of the positioning information; S12, predicting a search base point based on the positioning information; S13, extracting environment data in a range centered on the search base point and having a diameter of the search size from environment data stored by the operation platform as the environment data about the positioning information; The multi-source sensor data comprises positioning information, accelerometer information and lock state information in a preset time window; and the environment data comprises geographic area information and historical operation information. 2.The shared bicycle positioning management method based on artificial intelligence according to claim 1, wherein, Step S2 comprises: S21, calculating a stability score about the positioning information based on the position change of the positioning information at consecutive time points; S22, calculating a communication reliability score according to the environment data; S23, calculating the data quality information by weighted calculation of the stability score and the communication reliability score. 3.The AI-based shared bicycle positioning management method of claim 2, wherein, Step S22 comprises: S221, acquiring a plurality of position points from the environment data based on a preset diffusion rule; S222, for each position point, calculating a basic communication quality score of the position point according to the geographic area information and the historical operation information of the environment data corresponding to the position point; S223, determining a weighting factor corresponding to the plurality of position points according to the distance relationship between the position points and the search base point; S224, obtaining the communication reliability score by weighted summation of the basic communication quality scores according to the weighting factors. 4.The AI-based shared bicycle positioning management method of claim 1, wherein, Step S3 comprises: S31, acquiring the weight set from a preset weight configuration table according to the data quality information, the weight configuration table comprising weights of different data in the multi-source sensor data under different data quality information. 5.The shared bicycle positioning management method based on artificial intelligence according to claim 1, wherein, Step S4 comprises: S41, performing feature extraction on the multi-source sensor data to obtain multi-source features; S42, performing weighted processing on the multi-source features based on the weight set to obtain weighted multi-source features; S43, extracting environment features according to the environment data based on the positioning information; S44, splicing the weighted multi-source features and the environment features to obtain the fusion data features. 6.The shared bicycle positioning management method based on artificial intelligence according to claim 1, wherein, In step S5, the state analysis model comprises an input layer, a feature embedding layer, a bidirectional LSTM layer, a time attention layer, a full connection classification layer, and an output layer connected in sequence, and the classification result of the full connection classification layer comprises normal riding, parking, abnormal movement, fault, lock not closed, and signal loss. 7.The shared bicycle positioning management method based on artificial intelligence according to claim 1, wherein, Step S6 comprises: S61, determining whether the vehicle is in a state requiring operation intervention according to the motion state information and the positioning information; S62, generating an operation management instruction according to the state requiring operation intervention and the positioning information when the vehicle is in the state requiring operation intervention. 8.A shared bicycle positioning management system based on artificial intelligence for positioning and analyzing a real state of a vehicle in order to perform operation management in a shared bicycle operation, characterized by, The system comprises: An acquisition module configured to acquire multi-source sensor data uploaded by a vehicle, the multi-source sensor data comprising positioning information, and acquire environment data about the positioning information stored by an operation platform; A data quality analysis module configured to estimate data quality information of the multi-source sensor data based on the positioning information and the environment data; A weight configuration module configured to acquire a weight set about the multi-source sensor data according to the data quality information; A feature fusion module configured to acquire fused data features by weighting and fusing the multi-source sensor data based on the weight set; A motion analysis module configured to generate motion state information according to the fused data features based on a pre-trained state analysis model; An operation management module configured to perform operation management based on the motion state information and corresponding positioning information; The step of acquiring environment data about the positioning information stored by the operation platform comprises: S11, determining a search size based on the accuracy of the positioning information; S12, predicting a search base point based on the positioning information; S13, extracting environment data within a range centered on the search base point and having a diameter of the search size from environment data stored by the operation platform as the environment data about the positioning information; The multi-source sensor data comprises positioning information, accelerometer information, and lock state information within a preset time window; and the environment data comprises geographic area information and historical operation information.

Images