A shared pallet data tracking method based on unmanned loading scenarios

Abstract

This invention relates to the field of data tracking technology in unmanned loading scenarios, and in particular to a shared pallet data tracking method based on unmanned loading scenarios. The method includes acquiring real-time and historical status data, classifying data into categories, filtering key feature dimensions, calculating data weight allocation, optimizing the data tracking model, and performing dynamic data tracking operations. This invention can accurately filter key features and dynamically adjust data weights by analyzing pallet flow characteristics and classification biases, significantly improving the model's training efficiency, tracking accuracy, and generalization ability. Simultaneously, combined with a preset strategy library, it achieves efficient and stable dynamic data tracking, enhancing the adaptability and system stability of complex logistics scenarios.

Description

Technical Field

[0001] This invention belongs to the field of logistics automation and intelligent technology, specifically a shared pallet data tracking method based on unmanned loading scenarios. Background Technology

[0002] In the field of modern logistics and supply chain management, the demand for data tracking of shared pallets in unmanned loading scenarios is becoming increasingly prominent. A search revealed a patent, CN113412482B, for "transactional flow of change tracking data." This patent proposes a technical solution based on database transaction change tracking. By generating change tracking entries and storing them in a change tracking stream, it achieves real-time monitoring and recording of database table modifications. Although this solution demonstrates high accuracy in data change tracking, its application scenarios are mainly concentrated at the database level, and it has certain limitations in adapting to the data tracking needs of shared pallets in unmanned loading scenarios.

[0003] On the one hand, the proposed solution does not fully address the linkage mechanism between physical devices and data streams, making it difficult to meet the dynamic data tracking needs of shared pallets across regions and devices during unmanned loading. Unmanned loading scenarios typically require real-time collection and processing of data from multiple devices, while existing technologies are still insufficient in supporting inter-device collaboration and data synchronization. On the other hand, there is still room for improvement in the ability to handle anomalies in complex logistics scenarios; for example, issues such as lost pallets or data conflicts may affect system stability. These problems indicate that existing technologies still need further optimization to enhance the robustness and adaptability of the system when dealing with changing logistics environments.

[0004] Furthermore, existing methods lack scientific classification mechanisms tailored to the characteristics of logistics scenarios in terms of data classification and feature extraction. They fail to effectively segment data based on the changing states of shared pallets at different stages, making it difficult to optimize model training for specific scenarios and impacting the overall system performance. Simultaneously, insufficient attention to key dimensions during feature selection may lead to overemphasis on secondary features, thereby weakening the accuracy of data tracking.

[0005] In terms of resource allocation, traditional methods tend to use fixed weights or empirical allocation strategies, failing to fully consider the discriminative power of different data features. This approach may result in low efficiency in utilizing highly discriminative data during model training, limiting further improvements in system performance. Furthermore, in response optimization, existing techniques lack precision in matching and validating candidate strategies, potentially leading to deviations between the response strategies and actual scenario requirements, thus affecting overall performance. Summary of the Invention

[0006] The purpose of this invention is to provide a shared pallet data tracking method based on unmanned loading scenarios, in order to solve the problems in the existing technology such as insufficient linkage between physical equipment and data flow, lack of dynamic data tracking capabilities across regions and devices, weak ability to handle anomalies in complex logistics scenarios, and imperfect data classification and feature screening mechanisms.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a shared pallet data tracking method based on an unmanned loading scenario, the method comprising:

[0008] The system acquires real-time shared pallet status data from an unmanned loading scenario and a dataset consisting of multiple sets of historical status data recorded during operation. Based on the pallet flow characteristics corresponding to each set of status data in the dataset, the historical status data is divided into multiple categories. A data tracking model is trained based on the dataset, and key feature dimensions are selected by analyzing data classification deviations. Data weights are calculated based on the feature discrimination of different data on the key feature dimensions. The data tracking model is iteratively optimized based on the weight allocation to obtain a trained data tracking model. The real-time shared pallet status data from the unmanned loading scenario is input into the trained data tracking model to obtain data tracking instructions, and dynamic data tracking operations are performed based on these instructions.

[0009] Preferably, the historical state data is divided into multiple categories based on the pallet flow characteristics corresponding to each group of state data in the dataset, including: for the k-th group of historical state data: each detection stage from the first misidentification to the correct identification of the k-th group of historical state data is recorded as the analysis stage of the k-th group of historical state data; the flow change values ​​corresponding to the k-th group of historical state data in all analysis stages are arranged in chronological order to obtain the feature change sequence of the k-th group of historical state data; the position index of each value in the feature change sequence is used as the horizontal axis coordinate, and the value corresponding to each position index is used as the vertical axis coordinate. Form at least two coordinate points, and use all coordinate points as input for feature distribution analysis to obtain each two-dimensional coordinate vector and its corresponding mapping value; take the two-dimensional coordinate vector corresponding to the maximum mapping value as the principal pointing vector, and take the arctangent value of the ratio of the vertical component to the horizontal component of the principal pointing vector as the feature distribution pointing value; calculate the classification evaluation value of the k-th group of historical state data based on the detection stage corresponding to the first misidentification of the k-th group of historical state data, the stage interval between the first misidentification and the first correct identification, and the feature distribution pointing value; determine the category to which the k-th group of historical state data belongs based on the classification evaluation value.

[0010] Preferably, the step of calculating the classification evaluation value of the k-th group of historical state data based on the detection stage corresponding to the first misidentification of the k-th group of historical state data, the stage interval between the first misidentification and the first correct identification, and the feature distribution pointing value includes: recording the difference between the feature distribution pointing value and the preset benchmark pointing value as the first evaluation parameter; calculating the inverse normalization result of the detection stage corresponding to the first misidentification of the k-th group of historical state data, and determining the product of the stage interval between the first misidentification and the first correct identification, the inverse normalization result, and the first evaluation parameter as the classification evaluation value of the k-th group of historical state data.

[0011] Preferably, determining the category to which the k-th group of historical state data belongs based on the classification evaluation value includes: if the classification evaluation value is greater than a preset category division threshold, then the k-th group of historical state data is determined to belong to the high-frequency circulation category; otherwise, it is determined to belong to the low-frequency circulation category.

[0012] Preferably, the step of filtering key feature dimensions by analyzing data classification deviation includes: for the j-th category: constructing a first feature matrix based on all correctly identified state data in the j-th category, wherein each row of the first feature matrix is ​​a set of correctly identified state data; processing the first feature matrix using a feature filtering algorithm to obtain a first feature filtering result, wherein each column of data in the first feature filtering result constitutes a first candidate feature dimension; constructing a second feature matrix based on incorrectly identified state data in the j-th category, wherein each row of the second feature matrix is ​​a set of incorrectly identified state data; processing the second feature matrix using a feature filtering algorithm. The second feature selection results are obtained, with each column constituting a second candidate feature dimension. The data in the first (or second) feature matrix are standardized to eliminate dimensional differences between different feature dimensions, ensuring each feature value is on the same order of magnitude. For example, features such as flow change rate and trajectory coordinates are normalized to the [0,1] interval using min-max normalization to avoid excessive influence of features with large numerical ranges on the selection results. The preprocessed feature matrix is ​​then input into a random forest model for training. All feature dimensions are initially ranked using the model's output feature importance score (e.g., Gini impurity reduction). For the first feature matrix, features related to correct pallet identification (e.g., position deviation rate under high-frequency flow conditions) are prioritized. For the second feature matrix, features leading to identification errors (e.g., signal interference strength during low-frequency flow) are analyzed. Based on the initial ranking, the RFE algorithm iteratively eliminates the least important features. After each elimination, 5-fold cross-validation is used to calculate the model's classification accuracy on the remaining features. Iteration stops when the accuracy no longer significantly improves (the improvement is less than 0.5%), and the remaining features constitute the candidate feature set. For example, when processing the first feature matrix, if the accuracy does not decrease significantly after eliminating the "pallet material reflectivity" feature, it is excluded; while the "real-time position coordinate deviation" feature is retained because it has a significant impact on correct identification. Redundancy detection is performed on the candidate feature set, the Pearson correlation coefficient between features is calculated, and highly redundant features with a correlation coefficient greater than 0.8 are removed (retaining the more important features). The final data for each column is the corresponding first candidate feature dimension (or second candidate feature dimension). This process combines the feature importance of the tree model with the recursive elimination strategy, which can accurately retain the features that play a key role in pallet status identification, and avoid overfitting through cross-validation, ensuring that the screening results are adapted to the dynamic tracking requirements of unmanned loading scenarios. The first and second candidate feature dimensions are matched and analyzed, and the key feature dimensions are selected based on the matching results.

[0013] Preferably, the step of matching and parsing the first candidate feature dimension and the second candidate feature dimension, and filtering key feature dimensions based on the matching results, includes: using a feature association algorithm to match the first candidate feature dimension and the second candidate feature dimension to obtain multiple feature matching groups; calculating the correlation degree between the two dimensions in each feature matching group, and taking the feature matching group with a correlation degree greater than a preset correlation threshold as the target matching group; and taking the average dimension of the two dimensions in each target matching group as a key feature dimension.

[0014] Preferably, the step of calculating data weight allocation based on the feature discrimination of different data in key feature dimensions includes: for the correctly identified state data of the qth group: the incorrectly identified data in the category of the correctly identified state data of the qth group are recorded as the reference data of the correctly identified data of the qth group; the mean of the feature values ​​of all reference data of the correctly identified data of the qth group in each key feature dimension is calculated respectively, and the absolute difference between the feature value and the corresponding mean of the correctly identified data of the qth group in each key feature dimension is recorded as the discrimination index of the correctly identified data of the qth group in each key feature dimension; the sum of the inverse normalization results of the discrimination index of the correctly identified data of the qth group in all key feature dimensions is used as the weight coefficient of the correctly identified data of the qth group; and the target weight of the correctly identified data of the qth group in each training stage is calculated based on the weight coefficient.

[0015] Preferably, the acquisition of pallet flow characteristics corresponding to each group of state data in the dataset includes: statistically analyzing the pallet flow trajectory corresponding to each group of state data in the dataset to obtain the corresponding trajectory distribution histogram; using a boundary segmentation algorithm to segment the histogram to obtain no less than two segmentation intervals; calculating the average value of the flow change rate of all state data in each segmentation interval, and using it as the pallet flow characteristic of each group of state data in the corresponding segmentation interval.

[0016] Preferably, the step of calculating the target weight of the qth group of correctly identified data in each training phase based on the weight coefficient includes: for any group of correctly identified state data: multiplying the original weight of the group of data in each training phase by the corresponding weight coefficient to obtain the target weight of the group of data in each training phase.

[0017] Preferably, the step of performing dynamic data tracking operation based on the data tracking instruction includes: matching the data tracking instruction with a preset tracking strategy library to obtain the candidate tracking strategy with the highest matching degree; performing consistency verification on the candidate tracking strategy, the verification content including the matching degree between the flow type covered by the candidate tracking strategy and the current identification flow; adjusting the execution priority of the candidate tracking strategy based on the verification result, and prioritizing the execution of the candidate tracking strategy that has passed the verification.

[0018] Compared with existing technologies, the technical advantages of this invention are as follows: The shared pallet data tracking method provided by this invention for unmanned loading scenarios significantly improves the accuracy and adaptability of the data tracking system through in-depth analysis and processing of shared pallet status data. This method divides historical status data into multiple categories based on the pallet flow characteristics corresponding to each group of status data in the dataset. This allows the model to be trained for the characteristics of different categories of data, better capturing the dynamic flow patterns of pallets during unmanned loading, thereby improving the training efficiency and accuracy of the data tracking model. During feature selection, key feature dimensions are selected by analyzing data classification deviations. This accurately identifies features that play a crucial role in data tracking, avoiding interference from irrelevant or secondary features, allowing the model to focus more on learning important features, further improving tracking accuracy. Data weight allocation is calculated based on the feature discrimination of different data on key feature dimensions, and the data tracking model is iteratively optimized based on the weight coefficients. This dynamic weight adjustment mechanism enables the model to adaptively optimize during training, enhancing its adaptability to complex logistics scenarios and improving the model's generalization performance. Real-time shared pallet status data collected in unmanned loading scenarios is input into a trained data tracking model to obtain data tracking instructions, and dynamic data tracking operations are executed based on these instructions. By matching the data tracking instructions with a preset tracking strategy library, the candidate tracking strategy with the highest matching degree is obtained. Then, the candidate tracking strategies are validated for consistency, and the execution priority of the candidate tracking strategies is adjusted based on the validation results, prioritizing the execution of candidate tracking strategies that have passed the validation. This ensures the accuracy and relevance of the tracking strategy, improving data tracking efficiency and system stability. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall process of the shared tray data tracking method in an embodiment of the present invention, showing the complete process from data acquisition to dynamic data tracking operation.

[0020] Figure 2 This is a flowchart of the historical state data classification processing in an embodiment of the present invention, which explains in detail how to classify historical state data based on feature distribution pointing values ​​and classification evaluation values.

[0021] Figure 3 This is a flowchart of the key feature dimension screening process in an embodiment of the present invention, which describes the specific steps of screening key feature dimensions by matching and parsing the first feature matrix and the second feature matrix.

[0022] Figure 4 This is a flowchart illustrating the data weight allocation calculation process in an embodiment of the present invention, demonstrating the implementation process of weight allocation based on the discriminative power of key feature dimensions.

[0023] Figure 5This is a flowchart illustrating the execution of dynamic data tracking operations in an embodiment of the present invention, demonstrating the logical flow of matching and verifying data tracking instructions with a preset tracking strategy library. Detailed Implementation

[0024] This invention provides a shared pallet data tracking method based on unmanned loading scenarios, the specific implementation of which is described in conjunction with the appendix. Figure 1 To be continued Figure 5 Detailed explanation follows. (Attached) Figure 1 This demonstrates the overall workflow of the shared tray data tracking method, including a data acquisition module, a data classification module, a feature filtering module, a weight allocation module, a data tracking module, a tracking strategy library, and a verification module. These modules are connected and work together in a specific logical order to achieve dynamic tracking of shared tray status data.

[0025] First, the data acquisition module is used to acquire real-time shared pallet status data in unmanned loading scenarios, as well as a dataset consisting of multiple sets of status data recorded during historical operations. The data acquisition module integrates real-time and historical status data into a unified data storage area through a sensor network and data transmission equipment.

[0026] The pallet flow characteristics corresponding to each group of state data in the dataset are obtained through statistical analysis. Specifically, the pallet flow trajectories corresponding to each group of state data in the dataset are statistically analyzed to generate a trajectory distribution histogram. A boundary segmentation algorithm is then used to segment the histogram to obtain at least two segmentation intervals. The average flow change rate of all state data within each segmentation interval is calculated and used as the pallet flow characteristic for each group of state data in that interval. These characteristic data are then passed to the data classification module for further processing.

[0027] The data classification module is responsible for categorizing historical status data based on the pallet flow characteristics corresponding to each group of status data in the dataset. (See attached image) Figure 2 As shown, for the k-th group of historical state data, each detection stage from the first misidentification to the correct identification is marked as an analysis stage, and the corresponding flow change values ​​of the k-th group of historical state data in all analysis stages are arranged in chronological order to form a feature change sequence.

[0028] The position index of each value in the feature change sequence is used as the horizontal axis coordinate, and the value corresponding to each position index is used as the vertical axis coordinate, thus forming at least two coordinate points. These coordinate points serve as input for feature distribution analysis, yielding each two-dimensional coordinate vector and its corresponding mapped value. The two-dimensional coordinate vector corresponding to the maximum mapped value is defined as the principal pointing vector, and the arctangent of the ratio of the vertical component to the horizontal component in the principal pointing vector is determined as the feature distribution pointing value.

[0029] Based on the detection stage corresponding to the first misidentification of the k-th historical state data, the stage interval between the first misidentification and the first correct identification, and the feature distribution pointing value, the classification evaluation value of the k-th historical state data is calculated. The calculation process of the classification evaluation value is as follows: the difference between the feature distribution pointing value and the preset benchmark pointing value is recorded as the first evaluation parameter; the inverse normalization result of the detection stage corresponding to the first misidentification is calculated; and the product of the stage interval between the first misidentification and the first correct identification, the inverse normalization result, and the first evaluation parameter is determined as the classification evaluation value of the k-th historical state data.

[0030] If the classification evaluation value is greater than the preset category division threshold, the k-th group of historical state data is determined to belong to the high-frequency circulation category; otherwise, it is determined to belong to the low-frequency circulation category. Through the above steps, the data classification module completes the category division of the historical state data and transmits the classification results to the feature filtering module.

[0031] The feature selection module's task is to select key feature dimensions by analyzing data classification bias. (See attached image) Figure 3 As shown, for the j-th category, a first feature matrix is ​​constructed based on all correctly identified state data in that category, where each row of the first feature matrix represents a set of correctly identified state data. A feature selection algorithm is then used to process the first feature matrix to obtain a first feature selection result, where each column of data in the first feature selection result constitutes a first candidate feature dimension. Similarly, a second feature matrix is ​​constructed based on the incorrectly identified state data in the j-th category, where each row of the second feature matrix represents a set of incorrectly identified state data.

[0032] A feature selection algorithm is used to process the second feature matrix to obtain the second feature selection result. Each column of data in the second feature selection result constitutes a second candidate feature dimension. The first and second candidate feature dimensions are matched and parsed, and a feature association algorithm is used to match them to obtain multiple feature matching groups.

[0033] The correlation between the two dimensions in each feature matching group is calculated, and feature matching groups with a correlation greater than a preset correlation threshold are designated as target matching groups. The average dimension of the two dimensions in the target matching group is defined as a key feature dimension. Through the above steps, the feature filtering module completes the filtering of key feature dimensions and passes the filtering results to the weight allocation module.

[0034] The weight allocation module calculates data weights based on the feature discrimination of different data points along key feature dimensions. (See attached image.) Figure 4 As shown, for the correctly identified state data in the qth group, the incorrectly identified data in its category are used as reference data.

[0035] Calculate the mean of the eigenvalues ​​of all reference data for the correctly identified data in group q for each key feature dimension, and denote the absolute difference between the eigenvalue and the corresponding mean of the correctly identified data in group q for each key feature dimension as the discrimination index. Use the sum of the inverse normalized results of the discrimination indices of the correctly identified data in group q for all key feature dimensions as the weighting coefficient of the correctly identified data in group q.

[0036] The target weight for the correctly identified data in group q is calculated based on the weight coefficients. Specifically, the original weight of this group of data in each training phase is multiplied by the corresponding weight coefficient to obtain the target weight for that group of data in each training phase. Through these steps, the weight allocation module completes the dynamic allocation of data weights and transmits the allocation results to the data tracking module.

[0037] The data tracking module iteratively optimizes the data tracking model based on the weight allocation results to obtain a fully trained data tracking model. After training, the shared pallet status data collected in real time in the unmanned loading scenario is input into the trained data tracking model to obtain data tracking instructions. The data tracking instructions are then passed to the tracking strategy library for matching operations.

[0038] As attached Figure 5 As shown, the data tracking command is matched with multiple strategies in the preset tracking strategy library to obtain the candidate tracking strategy with the highest matching degree. The candidate tracking strategy is passed to the verification module for consistency verification. The verification content includes the matching degree between the flow type covered by the candidate tracking strategy and the current identification flow.

[0039] Based on the verification results, the execution priority of candidate tracking strategies is adjusted, with priority given to executing candidate tracking strategies that pass verification. Through the above steps, the data tracking module achieves dynamic tracking of shared tray status data.

[0040] The entire system achieves efficient tracking of shared pallet status data in unmanned loading scenarios through the coordinated operation of data acquisition, data classification, feature filtering, weight allocation, data tracking, tracking strategy library, and verification modules. The close interconnections between modules and smooth data flow ensure the system's stability and reliability.

[0041] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principle of this invention will be further explained below in conjunction with a specific application scenario.

[0042] In unmanned loading scenarios, the shared pallet data tracking method operates as follows: First, real-time and historical status data are acquired through a data acquisition module, and this data is integrated into a unified data storage area. Taking a logistics center as an example, the center deploys multiple sets of sensor devices to monitor the pallet's flow trajectory, position changes, and status information during the unmanned loading process.

[0043] The sensor network uploads data to the data acquisition module via a wireless transmission protocol, generating a dataset containing pallet flow characteristics. At this point, the pallet flow characteristics corresponding to each set of state data in the dataset are obtained through statistical analysis. Specifically, the pallet flow trajectory is statistically analyzed to generate a trajectory distribution histogram, and a boundary segmentation algorithm is used to segment the histogram, resulting in at least two segmentation intervals.

[0044] The average rate of change of all state data in each segmented interval is calculated as the pallet flow feature of each group of state data in the corresponding segmented interval. These feature data are then passed to the data classification module.

[0045] The data classification module categorizes historical state data based on pallet flow characteristics. For example, for the k-th group of historical state data, each detection stage from initial misidentification to correct identification is marked as an analysis stage. The flow change values ​​of the k-th group of historical state data across all analysis stages are arranged chronologically to form a feature change sequence. The position index of each value in the feature change sequence serves as the horizontal axis coordinate, and the corresponding value serves as the vertical axis coordinate, resulting in at least two coordinate points. These coordinate points are used as input for feature distribution analysis, yielding each two-dimensional coordinate vector and its corresponding mapped value.

[0046] The two-dimensional coordinate vector corresponding to the maximum mapped value is defined as the principal pointing vector, and the arctangent of the ratio of the vertical component to the horizontal component in the principal pointing vector is determined as the feature distribution pointing value. Based on the detection stage corresponding to the first misidentification of the k-th group of historical state data, the stage interval between the first misidentification and the first correct identification, and the feature distribution pointing value, the classification evaluation value of the k-th group of historical state data is calculated.

[0047] The calculation process for the classification evaluation value involves recording the difference between the feature distribution pointing value and the preset benchmark pointing value as the first evaluation parameter, and calculating the inverse normalization result of the detection stage corresponding to the first misidentification. The product of the stage interval between the first misidentification and the first correct identification, the inverse normalization result, and the first evaluation parameter is determined as the classification evaluation value of the k-th group of historical state data. If the classification evaluation value is greater than the preset category division threshold, the k-th group of historical state data is determined to belong to the high-frequency circulation category; otherwise, it is determined to belong to the low-frequency circulation category. Through the above steps, the data classification module completes the category division of the historical state data and transmits the classification results to the feature selection module. The task of the feature selection module is to select key feature dimensions by analyzing the data classification deviation.

[0048] Taking the j-th category as an example, a first feature matrix is ​​constructed based on all correctly identified state data in that category, where each row of the first feature matrix represents a set of correctly identified state data. A feature selection algorithm is then used to process the first feature matrix to obtain the first feature selection result, where each column of data in the first feature selection result constitutes a first candidate feature dimension.

[0049] Similarly, a second feature matrix is ​​constructed based on the misidentified state data in the j-th category, where each row of the second feature matrix contains a set of misidentified state data. A feature filtering algorithm is used to process the second feature matrix to obtain the second feature filtering result, where each column of data constitutes a second candidate feature dimension. The first and second candidate feature dimensions are then matched and parsed using a feature association algorithm to obtain multiple feature matching groups.

[0050] The correlation between the two dimensions in each feature matching group is calculated, and feature matching groups with a correlation greater than a preset correlation threshold are designated as target matching groups. The average dimension of the two dimensions in the target matching group is defined as a key feature dimension. Through the above steps, the feature filtering module completes the filtering of key feature dimensions and passes the filtering results to the weight allocation module.

[0051] The weight allocation module calculates data weights based on the feature discrimination of different data across key feature dimensions. Taking the correctly identified state data in group q as an example, the incorrectly identified data in its category are used as reference data.

[0052] Calculate the mean of the eigenvalues ​​of all reference data for the q-th correctly identified data in each key feature dimension, and denote the absolute difference between the eigenvalue and the corresponding mean of the q-th correctly identified data in each key feature dimension as the discrimination index. Use the sum of the inverse normalized results of the discrimination indices for the q-th correctly identified data in all key feature dimensions as the weight coefficients for the q-th correctly identified data. Based on these weight coefficients, calculate the target weights for the q-th correctly identified data in each training phase. Specifically, multiply the original weights of this group of data in each training phase by the corresponding weight coefficients to obtain the target weights for that group of data in each training phase. Through these steps, the weight allocation module completes the dynamic allocation of data weights and transmits the allocation results to the data tracking module.

[0053] The data tracking module iteratively optimizes the data tracking model based on the weight allocation results to obtain a fully trained data tracking model. After training, the shared pallet status data collected in real time during the unmanned loading scenario is input into the trained data tracking model to obtain data tracking instructions. The data tracking instructions are then passed to the tracking strategy library for matching operations. For example, in a certain unmanned loading process, the system collects a set of pallet status data in real time and inputs it into the trained data tracking model to generate data tracking instructions.

[0054] As attached Figure 5 As shown, the data tracking command is matched against multiple strategies in the preset tracking strategy library to obtain the candidate tracking strategy with the highest matching degree. The candidate tracking strategy is then passed to the verification module for consistency verification. The verification content includes the degree of matching between the flow type covered by the candidate tracking strategy and the current identified flow. Based on the verification results, the execution priority of the candidate tracking strategies is adjusted, and the candidate tracking strategies that pass the verification are executed first. Through the above steps, the data tracking module realizes the dynamic tracking operation of shared tray status data.

[0055] The entire system achieves efficient tracking of shared pallet status data in unmanned loading scenarios through the coordinated operation of data acquisition, data classification, feature filtering, weight allocation, data tracking, tracking strategy library, and verification modules.

[0056] The close connections between the modules and the smooth data flow ensure the stability and reliability of the system. For example, in a practical application, when the system detects abnormal movement of pallets during unmanned loading, it can quickly generate corresponding tracking instructions through the data tracking module, match the optimal handling strategy through the tracking strategy library, and finally have the verification module confirm and execute the strategy, thereby effectively dealing with emergencies in complex logistics environments.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for tracking shared pallet data in an unmanned loading scenario, characterized in that, The method includes the following steps: Acquire real-time shared pallet status data and a dataset consisting of multiple sets of status data recorded during historical operations in unmanned loading scenarios; Based on the pallet flow characteristics corresponding to each group of status data in the dataset, the historical status data is divided into multiple categories; The data tracking model is trained based on the dataset, and key feature dimensions are selected by analyzing the data classification bias. Calculate data weight allocation based on the feature discrimination of different data on key feature dimensions; Based on the weight allocation, the data tracking model is iteratively optimized to obtain the trained data tracking model. The shared pallet status data collected in real time in the unmanned loading scenario is input into the trained data tracking model to obtain data tracking instructions; Dynamic data tracking operations are performed based on the data tracking instructions.

2. The method for tracking shared pallet data in an unmanned loading scenario according to claim 1, characterized in that, The historical status data is divided into multiple categories based on the pallet flow characteristics corresponding to each group of status data in the dataset, including: For the k-th set of historical state data: Each detection stage from the first misidentification to the correct identification of the k-th group of historical state data is denoted as the analysis stage of the k-th group of historical state data. Arrange the corresponding flow change values ​​of the k-th group of historical state data in all analysis stages according to time order to obtain the characteristic change sequence of the k-th group of historical state data; Using the position index of each value in the feature change sequence as the horizontal axis coordinate and the value corresponding to each position index as the vertical axis coordinate, at least two coordinate points are formed. All coordinate points are used as input for feature distribution analysis to obtain each two-dimensional coordinate vector and its corresponding mapping value. The two-dimensional coordinate vector corresponding to the maximum mapping value is taken as the main pointing vector, and the arctangent value of the ratio of the vertical component to the horizontal component in the main pointing vector is taken as the feature distribution pointing value. Based on the detection stage corresponding to the first misidentification of the k-th group of historical state data, the stage interval between the first misidentification and the first correct identification, and the feature distribution pointing value, calculate the classification evaluation value of the k-th group of historical state data; The category to which the k-th group of historical state data belongs is determined based on the classification evaluation value.

3. The method for tracking shared pallet data in an unmanned loading scenario according to claim 2, characterized in that, The classification evaluation value of the k-th group of historical state data is calculated based on the detection stage corresponding to the first misidentification of the k-th group of historical state data, the stage interval between the first misidentification and the first correct identification, and the feature distribution pointing value, including: The difference between the feature distribution pointing value and the preset benchmark pointing value is recorded as the first evaluation parameter; Calculate the inverse normalization result of the detection stage corresponding to the first misidentification of the k-th group of historical state data, and determine the classification evaluation value of the k-th group of historical state data by multiplying the stage interval between the first misidentification and the first correct identification, the inverse normalization result, and the first evaluation parameter.

4. A shared pallet data tracking method based on an unmanned loading scenario according to claim 2, characterized in that, The step of determining the category of the k-th group of historical state data based on the classification evaluation value includes: If the classification evaluation value is greater than the preset category division threshold, then the k-th group of historical state data is determined to belong to the high-frequency circulation category; otherwise, it is determined to belong to the low-frequency circulation category.

5. A shared pallet data tracking method based on an unmanned loading scenario according to claim 1, characterized in that, The process of selecting key feature dimensions by analyzing data classification bias includes: For the j-th category: A first feature matrix is ​​constructed based on all correctly identified state data in the j-th category, where each row of the first feature matrix is ​​a set of correctly identified state data. The first feature matrix is ​​processed using a feature selection algorithm to obtain a first feature selection result, wherein each column of data in the first feature selection result constitutes a first candidate feature dimension. A second feature matrix is ​​constructed based on the state data of misidentification in the j-th category, where each row of the second feature matrix contains a set of state data of misidentification. The second feature matrix is ​​processed using a feature selection algorithm to obtain the second feature selection result. Each column of data in the second feature selection result constitutes a second candidate feature dimension. The first and second candidate feature dimensions are matched and analyzed, and the key feature dimensions are selected based on the matching results.

6. A shared pallet data tracking method based on an unmanned loading scenario according to claim 5, characterized in that, The process of matching and parsing the first and second candidate feature dimensions, and filtering key feature dimensions based on the matching results, includes: A feature association algorithm is used to match the first candidate feature dimension with the second candidate feature dimension to obtain multiple feature matching groups; Calculate the correlation between the two dimensions in each feature matching group, and take the feature matching group with a correlation greater than the preset correlation threshold as the target matching group; The average dimension of the two dimensions in each target matching group is used as a key feature dimension.

7. A shared pallet data tracking method based on an unmanned loading scenario according to claim 1, characterized in that, The calculation of data weight allocation based on the feature discrimination of different data on key feature dimensions includes: For the qth group of correctly identified state data: The incorrectly identified data in the category of the correctly identified state data in group q is recorded as the reference data of the correctly identified data in group q. Calculate the mean of the feature values ​​of all reference data for the qth group of correctly identified data in each key feature dimension. The absolute difference between the feature value and the corresponding mean of the qth group of correctly identified data in each key feature dimension is denoted as the discrimination index of the qth group of correctly identified data in each key feature dimension. The sum of the inverse normalization results of the discrimination index of the qth group of correctly identified data on all key feature dimensions is used as the weight coefficient of the qth group of correctly identified data. The target weight of the correctly identified data in the qth group is calculated based on the weight coefficients in each training phase.

8. A shared pallet data tracking method based on an unmanned loading scenario according to claim 1, characterized in that, The acquisition of pallet flow characteristics corresponding to each group of state data in the dataset includes: The pallet flow trajectory corresponding to each group of status data in the dataset is statistically analyzed to obtain the corresponding trajectory distribution histogram. The histogram is then segmented using a boundary segmentation algorithm to obtain at least two segmentation intervals. Calculate the average rate of change of all state data in each segmented interval, and use it as the pallet flow characteristic of each group of state data in the corresponding segmented interval.

9. A shared pallet data tracking method based on an unmanned loading scenario according to claim 7, characterized in that, The calculation of the target weight of the qth group of correctly identified data in each training phase based on the weight coefficients includes: For any set of correctly identified state data: multiply the original weights of the data set in each training phase by the corresponding weight coefficients to obtain the target weights of the data set in each training phase.

10. A shared pallet data tracking method based on an unmanned loading scenario according to claim 1, characterized in that, The execution of dynamic data tracking operations based on the data tracking instructions includes: The data tracking instructions are matched with the preset tracking strategy library to obtain the candidate tracking strategy with the highest matching degree; The candidate tracking strategies are subjected to consistency verification, and the verification content includes the degree of matching between the flow types covered by the candidate tracking strategies and the current identified flow. The execution priority of candidate tracking strategies is adjusted based on the verification results, with priority given to candidate tracking strategies that pass the verification.

Images