An agricultural product logistics supervision system based on an internet of things

Abstract

The application discloses a kind of based on Internet of Things agricultural product logistics supervision system, specifically related to logistics transport supervision field, including: by data acquisition module records the three-axis acceleration and angular velocity data of tray in loading and unloading and transportation process, reconstructs continuous physical motion trajectory set;Cooperative behavior construction module constructs time sequence tray cooperative behavior graph to trajectory vector;Cooperative fracture identification module calculates trajectory similarity between nodes, and identifies cooperative fracture node set;Mismatch starting point tracing module compares the delivery time of logistics site according to fracture time and filters mismatch starting point;Mismatch end point tracing module filters mismatch end point based on uploaded termination time and compares the receipt time of logistics site;Tracing path construction module determines actual mismatch end point in combination with historical trajectory cumulative displacement and site geographic distance, outputs mismatch path table, to realize the path-level accurate tracing of abnormal behaviors such as wrong box and wrong goods.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics transportation supervision, and more particularly to an agricultural product logistics supervision system based on the Internet of Things. BACKGROUND

[0002] In the process of agricultural product logistics transportation, as the most basic flow unit, pallets are widely used in sorting, loading and unloading, transfer and site handover and other operation links. However, due to the large number of pallets, long transportation chain and frequent operation, problems such as pallet misplacement, site misdelivery, label falling off or allocation path change often occur in actual operation, especially in the scenes of multi-batch mixed loading, manual handling or re-loading on the way. Although temperature monitoring and GPS positioning means have been introduced in the industry for part of the pallets to record the trajectory and collect environmental parameters, due to the consideration of equipment cost and deployment burden, most pallets are not equipped with sensing devices with continuous tracking capability, and the data records between nodes are intermittent and missing. Therefore, when an abnormal situation such as misplacement, misplacement and misplacement of a pallet occurs, it is often difficult to accurately restore its actual flow path and operation link in the transportation process, making it difficult to locate the source of the abnormality and identify the responsible party. Especially under the complex path conditions of multiple sites with overlapping reception, repeated transfer or re-allocation on the way, even if the terminal site finds the difference in goods, it is difficult to determine the time of occurrence, the location of change and the responsibility, resulting in difficulties in tracing the source of misplacement, misplacement and misplacement, unclear path and unclear responsibility. SUMMARY

[0003] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide an agricultural product logistics supervision system based on the Internet of Things to solve the problems raised in the background art.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0005] An agricultural product logistics supervision system based on the Internet of Things, comprising a data acquisition module, a collaborative behavior construction module, a collaborative break identification module, a misplacement starting point tracing module, a misplacement ending point tracing module and a tracing path construction module, wherein:

[0006] The data acquisition module acquires the three-axis acceleration and angular velocity sequence of the transportation pallet during loading and unloading and transportation, and constructs a set of continuous physical motion trajectories of the pallet;

[0007] The collaborative behavior construction module performs time synchronization processing on the trajectory vector set of all pallets, and constructs a time sequence pallet collaborative behavior graph structure;

[0008] The cooperative fracture identification module performs trajectory similarity calculation on each pallet node in the cooperative behavior graph structure and the remaining pallet nodes to identify a set of cooperative fracture nodes in the cooperative behavior graph;

[0009] The mismatch starting point tracing module screens a pallet mismatch starting point by comparing the fracture occurrence time of the cooperative fracture node with the shipment registration time of each logistics site in the transportation cycle;

[0010] The mismatch end point tracing module screens a pallet mismatch end point by comparing the data upload termination time of the fracture node with the receipt registration time of each logistics site.

[0011] The tracing path construction module determines an actual mismatch end point based on the historical trajectory of the fracture node and constructs a pallet mismatch behavior tracing path table.

[0012] In a preferred embodiment, the data acquisition module acquires a three-axis acceleration and angular velocity sequence of the transportation pallet during loading and unloading and transportation, and the construction of the continuous physical motion trajectory set of the pallet specifically includes:

[0013] Based on the inertial sensing node embedded at the bottom of each agricultural product transportation pallet, three-axis acceleration and angular velocity raw data streams are acquired;

[0014] Each sensing node is bound to a unique pallet number, and real-time raw data streams are uploaded to a transportation supervision platform according to the acquisition time stamp;

[0015] The three-axis acceleration and angular velocity data are subjected to gravity component elimination, and the dynamic acceleration and angular velocity change corresponding to the motion change are retained;

[0016] Based on a time series integration method, the acceleration and angular velocity are reconstructed to generate a continuous physical motion trajectory set of the transportation pallet, and each trajectory is labeled with a pallet number.

[0017] In a preferred embodiment, the trajectory reconstruction of the acceleration and angular velocity based on the time series integration method specifically includes:

[0018] In the transportation supervision platform, the uploaded time series three-axis acceleration and angular velocity raw data streams are subjected to sliding window filtering processing to eliminate high-frequency vibration and short-time error peaks;

[0019] The acceleration sequence is subjected to first integration calculation to obtain a corresponding velocity sequence;

[0020] The velocity sequence is subjected to second integration, and an initial point calibration value is combined to construct the spatial displacement of the pallet in the entire period;

[0021] Integrating the angular velocity sequence, reconstructing the pallet pose transformation sequence, and combining with the velocity sequence and spatial displacement to form the complete physical motion trajectory.

[0022] In a preferred embodiment, the cooperative behavior modeling module performs time synchronization processing on the trajectory vectors of all pallets, and the timing pallet cooperative behavior graph structure includes:

[0023] Extracting the trajectory data of all pallets from the physical motion trajectory set and performing vectorization conversion on the trajectory;

[0024] Synchronizing and aligning the trajectory vectors of all pallets on the time axis according to the collection timestamps of the sensor nodes;

[0025] Taking the pallet number as the node, establishing a connection edge between the pallet trajectory vectors, and constructing a timing cooperative behavior graph structure.

[0026] In a preferred embodiment, the cooperative fracture identification module performs trajectory similarity calculation on each pallet node and the remaining pallet nodes in the cooperative behavior graph structure, and the cooperative fracture node set in the cooperative behavior graph includes:

[0027] Extracting the timing trajectory vector sequence of each pallet node in the cooperative behavior graph and performing standardization processing;

[0028] Performing time-by-time similarity calculation on the vector sequence of each pallet node and all other nodes within the same time window;

[0029] Statistically calculating the average similarity of each node and constructing a similarity mean mapping table between all nodes, and marking the pallet nodes with an average similarity lower than the set fracture determination threshold as cooperative fracture nodes;

[0030] Summarizing the index numbers of all cooperative fracture nodes and recording the fracture occurrence time, and outputting the cooperative fracture point set.

[0031] In a preferred embodiment, the mismatch starting point tracing module filters the pallet mismatch starting point by comparing the fracture occurrence time of the cooperative fracture node with the shipment registration time of each logistics site within the transportation cycle, specifically including:

[0032] Extracting the fracture occurrence time of each pallet node from the cooperative fracture point set and setting a retrieval time range tolerance;

[0033] Establishing a mapping between the shipment time sequence of the logistics site within the transportation cycle and the site identifier, and comparing the fracture occurrence time with the shipment registration time of the logistics site;

[0034] Filtering the shipment registration time consistent with the fracture occurrence time within the set retrieval time range tolerance, and marking the corresponding site identifier corresponding to the shipment registration time as the pallet mismatch starting point.

[0035] In a preferred embodiment, the mismatch end point screening module screens the mismatch end point of the pallet by comparing the data upload end time of the break node with the receiving registration time of each logistics site, and the screening of the mismatch end point of the pallet specifically includes:

[0036] Extracting the data upload end time of each pallet node in the collaborative break node set;

[0037] Obtaining the receiving registration time and site identifier of each logistics site in the transportation period, performing a one-by-one matching operation on the upload end time and the receiving time, and screening the logistics site with consistent time as a mismatch end point candidate.

[0038] In a preferred embodiment, the trace path construction module determines the actual mismatch end point based on the historical trajectory of the break node corresponding to the cumulative spatial displacement, and constructs a pallet mismatch behavior trace path table, which specifically includes:

[0039] Extracting the trajectory vector sequence of each break node, and calculating the cumulative spatial displacement of the trajectory sequence from the break occurrence time to the upload end time;

[0040] Inputting the trajectory vector sequence and the corresponding cumulative spatial displacement into the trained distance error prediction model, and correcting the cumulative spatial displacement according to the prediction error output by the model;

[0041] Obtaining the geographic coordinates of the mismatch starting point and each mismatch end point site candidate, calculating the distance between all mismatch sites, comparing the calculation result with the corrected cumulative spatial displacement, and selecting the site closest to the cumulative displacement as the actual mismatch end point;

[0042] Connecting the logistics sections corresponding to the pallet mismatch starting point and the actual mismatch end point, and establishing a trace path table.

[0043] In a preferred embodiment, the distance error prediction model is trained in the following way:

[0044] Extracting sample trajectory sections of pallets successfully registered for delivery and receiving in historical transportation periods;

[0045] Performing cumulative spatial displacement calculation on each trajectory section, extracting the trajectory vector sequence of the sample trajectory section, and constructing trajectory statistical features in combination with the cumulative spatial displacement;

[0046] Calculating the geographic path distance based on the coordinates of the start and end sites of the trajectory section, and generating a residual sequence of trajectory cumulative displacement and actual geographic distance;

[0047] Constructing a supervised learning sample set composed of trajectory statistical features and corresponding residual sequences, and inputting it into a multilayer perceptron network to perform residual fitting modeling.

[0048] The technical effect and advantages of the agricultural product logistics supervision system based on the Internet of Things are as follows:

[0049] The multi-tray agricultural product logistics supervision system based on group behavior separation can construct a multi-tray behavior coordination graph only through embedded inertial sensor data without relying on high-cost GPS positioning and temperature control labels, and identify group structure mutation behaviors caused by errors in loading and unloading sequences, position misplacement or cross interference between trays, so as to realize early identification and path tracking of potential wrong box, wrong goods or temperature control chain rupture risks. The abnormal tray path can be backtracked in a closed loop. The system has the advantages of low deployment cost, wide identification range, strong adaptability to dynamic path changes and the like, significantly improves the accuracy of agricultural product tray-level abnormal identification and the operability of responsibility tracing, and is suitable for logistics environments with dense error-prone links such as multi-batch mixed loading, high-frequency loading and unloading and graded cold chain transportation. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 FIG. 1 is a structural schematic diagram of the agricultural product logistics supervision system based on the Internet of Things. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0052] Embodiment 1

[0053] Figure 1 The agricultural product logistics supervision system based on the Internet of Things is given, which comprises a data acquisition module, a cooperative behavior construction module, a cooperative rupture identification module, a mismatch starting point tracing module, a mismatch end point tracing module and a tracing path construction module, wherein:

[0054] The data acquisition module acquires the three-axis acceleration and angular velocity sequence of the transport tray during loading and unloading and transportation, and constructs a continuous physical motion trajectory set of the tray;

[0055] The cooperative behavior construction module performs time synchronization processing on the trajectory vector set of all trays, and constructs a time sequence tray cooperative behavior graph structure;

[0056] The cooperative rupture identification module performs trajectory similarity calculation on each tray node and the remaining tray nodes in the cooperative behavior graph structure, and identifies a cooperative rupture node set in the cooperative behavior graph;

[0057] The mismatch origin tracing module screens the mismatch origins of pallets by comparing the break occurrence time of the collaborative break node with the shipment registration time of each logistics station within the transportation cycle;

[0058] The mismatch endpoint tracing module screens the mismatch endpoints of pallets by comparing the data upload termination time of the broken node with the receiving registration time of each logistics station.

[0059] The traceability path construction module determines the actual mismatch endpoint based on the cumulative spatial displacement corresponding to the historical trajectory of the broken node, and constructs a traceability path table for the pallet mismatch behavior.

[0060] The data acquisition module collects the three-axis acceleration and angular velocity sequences of the transport pallet during loading, unloading and transportation, and constructs a set of continuous physical motion trajectories of the pallet.

[0061] An inertial sensor node with a unique sensor identifier is set for each regulated agricultural product pallet. The sensor node is installed at the central axis position at the bottom of the pallet to prevent local vibration caused by uneven force on the four corners of the pallet from causing deviations in the acceleration data. The inertial sensor node integrates a three-axis accelerometer and a three-axis gyroscope, and has local caching and wireless data transmission capabilities, which can continuously collect physical motion data throughout the entire transportation cycle. Each sensor node is given a unique identification code during the manufacturing stage, and a one-to-one mapping relationship is established with the pallet's physical number through an embedded initialization program. This mapping is uniformly imported into the transportation supervision platform before deployment to form a number binding table. The above binding table is stored in a key-value pair structure with the pallet number and sensor identification code.

[0062] The sensor node automatically collects the original values of three-axis acceleration and three-axis angular velocity at a set sampling period. The sampling period is uniformly issued by the supervision platform during the system initialization phase. In this embodiment, the sampling frequency is set to 1 Hz, covering multiple dynamic characteristics generated in the typical transportation process due to loading and unloading, stacking, jolting and stationary stage, avoiding discontinuity of trajectory or distortion of motion pattern recognition due to insufficient frequency, while controlling the data upload pressure. The data collected by the sensor node is arranged according to the preset data structure, and each piece of data contains sensor number, binding tray number, timestamp, three-axis acceleration value and three-axis angular velocity value. The sensor node uses standard network communication protocol to upload data to the transportation supervision platform in real time. The supervision platform broadcasts a standard time beacon frame to all activated sensor nodes every 30 minutes, and the node receives and calibrates the local clock according to the frame header time field, so that the timestamp error is controlled within ±1 second. If the node is lost in the beacon synchronization period, the system will mark the clock drift identifier for its data, which will be used for error compensation in the subsequent trajectory reconstruction step. After the uploaded original data is received by the platform, it automatically completes the association and classification of tray motion data based on the number binding table and timestamp sequence. Finally, an inertial data record set based on tray number index is formed, providing the basis for the original input data for subsequent trajectory reconstruction, collaborative behavior recognition and abnormal linkage analysis steps. Each record is sequentially classified into the data sequence of the corresponding tray number, and when the record integrity reaches more than 95%, it enters the subsequent processing flow.

[0063] The acceleration signal obtained by collection is a composite acceleration vector containing gravity component. Since there are multiple stationary states of the tray in the stacking, handling and transportation process, the system identifies all time periods with acceleration module length within the set stable threshold and angular velocity below the angular speed stationary threshold as the gravity estimation interval in the preliminary analysis stage. The acceleration module length stable threshold is set to 9.6-10.2 m / s², and the angular velocity stable threshold is set to ±0.05 rad / s, which is determined based on the measured statistical data under static working conditions. In each identified stationary segment, the weighted average value of the three-axis acceleration data in the corresponding time window is calculated as the static gravity estimation value of the segment. The gravity estimation results of multiple stationary segments constitute an acceleration static offset model, which is used to fit and remove the gravity component from the entire acceleration time sequence. In the removal process, in order to enhance continuity and eliminate jumps caused by inconsistent estimates, the system uses a combination of sliding window fitting and linear interpolation to smooth and correct the gravity estimates, with a window length of 30 seconds and an update period of 10 seconds. The interpolation method uses a piecewise linear transition strategy. After the above processing, the acceleration time sequence corresponding to each tray is converted into a dynamic acceleration sequence with the gravity component removed.

[0064] After completing the gravity component rejection and obtaining the pure dynamic response acceleration and angular velocity sequence, the platform needs to further smooth the time series data to suppress the high-frequency disturbance and short-time error peaks caused by non-motion behavior factors (such as mechanical resonance, accidental vibration feedback, and hardware noise) during transportation, and prevent deviation from trajectory integration and behavior graph modeling. The supervision platform reads the dynamic acceleration and angular velocity sequence for each tray, and performs a sliding window filtering operation on each dimension vector (x, y, z). The sliding window size is adjusted according to the stability of different working conditions in the tray transportation scene. In this embodiment, the window length is fixed at 11 sampling points, and the window step interval is 1 sampling point. That is, in each iteration, the minimum and maximum values in the current window are removed, and the intermediate value is retained as the smoothed replacement value for the center sampling point.

[0065] After completing the filtering and smoothing of the three-axis acceleration and angular velocity data, the transportation supervision platform needs to perform integration operations on the smoothed time series data to obtain the physical motion trajectory of each transportation tray in the whole life cycle. First, integrate the three-axis acceleration sequence corresponding to each tray number to reconstruct the dynamic response change process in the online velocity space. Specifically, the first-order integral operation is performed on the acceleration vector smoothed by the sliding window according to the time interval to obtain the instantaneous velocity sequence in three directions. A velocity initial value is introduced at the initial integration time, which is derived from the standard static state (set as a zero vector) configured by the platform when the goods on the tray enter the warehouse. The integration interval is the inverse of the original sampling frequency, assuming a default sampling frequency of 50 Hz (default 1 Hz), the corresponding integration step is 0.02 seconds. After obtaining the velocity sequence, further second-order integration operation is performed on the velocity sequence to calculate the position displacement vector of the tray during transportation (this displacement is a theoretical value, not an accurate value, and the error in the actual scene is generally not less than 50%, error elimination is needed). By accumulating the instantaneous velocity of each time period, the linear velocity cumulative curve is constructed, and the trajectory displacement surface in three-dimensional space is further formed. The initial position of the displacement is input by the loading site calibration system. At the same time, the three-axis angular velocity sequence is integrated to generate the angular displacement sequence, thereby constructing the attitude transformation sequence of the tray at each time. The angular displacement data is developed based on the Euler angle framework, and the platform accumulates the angular velocity components at each time according to the time interval, and outputs the attitude change values in the pitch, roll, and yaw directions. The attitude sequence and the spatial displacement sequence are completely aligned in the time dimension, and are jointly used to identify dynamic behavior characteristics such as tray dumping, unbalanced loading, and sharp turning. After all the trajectory data is generated, the platform binds the tray number to each trajectory structure and uniquely identifies it, and constructs a set of tray physical motion trajectories, each trajectory in the set includes three-dimensional spatial displacement, three-dimensional attitude angle, acceleration change sequence, instantaneous velocity change sequence, timestamp sequence, and number field.

[0066] The cooperative behavior modeling module performs time synchronization processing on the trajectory vectors of all pallets to build a time sequence pallet cooperative behavior graph structure.

[0067] The motion trajectory data of all pallets is extracted from the generated set of physical motion trajectories. The trajectory data is converted into a uniform format trajectory vector structure. During the vectorization conversion process, the following five types of core elements are extracted for each trajectory data: first, the three-dimensional space displacement sequence obtained by double integrating the acceleration sequence is extracted; second, the three-dimensional attitude angle sequence obtained by integrating the angular velocity sequence is extracted, corresponding to the rotation angle changes on the three axes; third, the original three-axis acceleration sequence is retained and normalized by unit time to form a continuous acceleration change sequence; fourth, the instantaneous speed change sequence is formed by combining the differential results in the displacement sequence; and finally, the timestamp sequence in the original data is retained to ensure the correspondence of the dimensional data on the time axis. The above five types of elements are normalized in the same time sequence to form a multi-dimensional trajectory vector. The trajectory vector on each time slice is a data block with uniform dimensions. The pallet number is used as the node identifier, and the trajectory vector set is mapped into the graph structure space. In this graph structure, each node of the graph corresponds to a pallet individual, and the attribute field of the node is the trajectory vector sequence of the pallet in the entire transportation period.

[0068] The cooperative fracture identification module performs trajectory similarity calculation between each pallet node and the remaining pallet nodes in the cooperative behavior graph structure to identify the set of cooperative fracture nodes in the cooperative behavior graph.

[0069] The trajectory vector sequence carried by all pallet nodes in the constructed time sequence cooperative behavior graph of the transportation pallets is extracted. Each trajectory vector sequence contains multi-dimensional data of the pallet at each time slice in the entire transportation period. The synchronization degree of the pallets in the time sequence behavior is judged, and a fixed length time window is selected for behavior similarity comparison between node pairs. The window length is set to 300 seconds in actual engineering, corresponding to an average loading and unloading paragraph in general transportation operation. The window start time is advanced in a sliding manner from the beginning to the end according to the timestamp sequence, and the step length can be set to 30 seconds to support covering continuous time periods. In each time window, the trajectory vector sequence of each pallet node in the graph structure is extracted in turn. The trajectory vector sequence of the node in the current window is compared with the vector sequence of all other pallet nodes in the same window. The mean square error of the multi-dimensional vector difference on each component dimension is calculated at each time slice, then the average error is taken within the time window, and the error value is normalized to obtain the behavior similarity score. The score range is limited to 0-1, and the higher the value, the more consistent the cooperative behavior. After the calculation is completed, the similarity sequence of each node and all other nodes in the window is obtained.

[0070] After completing the hourly similarity calculation between all node pairs, the average similarity value of each node is calculated. That is, the similarity of a certain node with all other nodes is averaged to form a single indicator reflecting its consistency in cooperative behavior in the current time window. Based on this indicator, a complete similarity average mapping table is constructed, with the key being the tray number and the value being the similarity average of the node. In the similarity average mapping table, the average similarity of each node is compared with the preset fracture judgment threshold. The threshold is set based on the distribution of cooperative behavior and non-cooperative behavior in the historical transportation samples, and the default value is 0.5. This value indicates that if the behavior similarity of a certain tray with most trays in the current time window is less than 0.5 on average, the tray is considered not to participate in the overall cooperative behavior, i.e., the cooperative behavior fracture occurs. The tray number of the node marked as a fracture node and its corresponding time window start time are recorded together and are included in the cooperative fracture point set. This point set is finally used as the structural mutation set in the cooperative behavior network. It reflects the overall disconnection of a certain tray with other trays in a specific period, and is completely based on physical trajectory data, without relying on additional labels, temperature or manual annotation, thus having strong universality and endogeneity. In complex transportation scenarios, it can effectively capture local behavior variations caused by position replacement, operation interference or manual misplacement.

[0071] The mismatch starting point tracing module screens the tray mismatch starting point by comparing the fracture occurrence time of the cooperative fracture node with the delivery registration time of each logistics site in the transportation cycle.

[0072] After identifying the fracture nodes in the cooperative behavior graph, the system obtains the fracture occurrence time of each tray in the fracture node set. This fracture time identifies the time point at which the corresponding tray trajectory behavior significantly deviates from the cooperative behavior of other trays in the transportation process, usually with the start time stamp of the node average similarity being lower than the fracture threshold in the sliding time window as its occurrence time. Each tray node only retains the time stamp of the first time it is judged as a fracture node, which is used to trace the starting position of its behavior anomaly. In actual deployment, the system binds this time stamp information with the tray number to generate a fracture event time index table, with each row in the table recording a tray number and its corresponding fracture occurrence time. The delivery registration information of all logistics receiving sites in the transportation cycle is collected, and the formal delivery registration time of each site is extracted. Each logistics site needs to complete identity registration and time identification standardization setting in the system in advance to ensure that the delivery time is recorded in the time stamp format accurate to seconds. On this basis, the system constructs a complete logistics site delivery time mapping table with the site number as the key and the delivery registration time as the value. This mapping structure supports efficient reverse query of the corresponding site from the time index to form the key matching logic in mismatch tracing.

[0073] Due to the physical transmission delay, edge node clock difference and human delivery registration delay of the sampling of the sensor node, the system needs to perform error tolerance matching operation on the fracture time and the site delivery time. Therefore, a retrieval time range tolerance is set to determine whether two time points are considered to be consistent in time. The tolerance can be set to ± 30 seconds to ± 300 seconds during system deployment, and the default value is ± 60 seconds. The setting is based on the on-site registration error statistics to ensure accuracy and fault tolerance in matching. The fracture time index of each fracture node is retrieved in sequence, and the delivery registration time within the retrieval tolerance range is searched in the delivery time mapping table. If the delivery time of a site is within the tolerance interval before and after the fracture time, the site identification is marked as the mismatch starting point candidate of the corresponding pallet node. If there are multiple site registration times within the tolerance interval during the matching process, the system will retain all of them and enter the subsequent discrimination process for final mismatch starting point confirmation. Finally, the system outputs a result index table composed of each fracture node pallet number and its corresponding mismatch starting site identification as the mapping result between behavior fracture and site delivery time.

[0074] The mismatch end point tracing module screens the pallet mismatch end point by comparing the data upload termination time of the fracture node with the receipt registration time of each logistics site.

[0075] For each sensor number corresponding to a fracture node, the upload log index structure is called, and the last valid data upload timestamp is obtained. To exclude non-real termination factors such as communication interruption and sensor failure, the system sets the confirmation condition for data upload termination time as follows: no new data packet is received for more than a set time (e.g. 10 minutes) at the current time, and the upload frequency in the previous time window is still within the normal range. For example, if the upload interval is set to 1 second, the latest sampling time before termination should be a valid record within 10 seconds before termination. If it exceeds this window, it is considered to be terminated abnormally and is excluded. The confirmed upload termination time is bound to the pallet number to generate a fracture node upload termination time table. At the same time, a set of receipt registration data of all logistics receiving sites within the transportation cycle is established. This set contains the pallet receipt time registration records of each site in the transportation scheduling system, and each record is bound to a unique site identification, a pallet number and a standardized receipt timestamp. Standardized preprocessing is performed on this data set to exclude abnormal records such as registration delay and batch conflict, and only valid receipt registration information that matches the transportation path after timestamp correction is retained. All receipt time information is stored in a unified time format (accurate to seconds), and a bidirectional mapping structure of site identification and receipt time is constructed to facilitate fast indexing and reverse matching.

[0076] After the above processing is completed, the system performs a matching operation of the upload termination time and the pickup time. The matching strategy adopts a one-by-one comparison manner. For each broken node pallet, according to its upload termination time, a window search is performed in the pickup time sequence to determine whether there is a time consistent record. In order to avoid the matching failure caused by the microsecond level difference, a matching tolerance range is set (the default is consistent with the time tolerance range in the mismatch starting point tracing module, and can be flexible according to the actual scene). For example, if the upload termination time of a certain pallet is 15:32:20, the system will search for records in the pickup time sequence whose time falls between 15:31:50 and 15:32:50. If the matching is successful, the corresponding logistics site identifier will be recorded as the mismatch end candidate of the pallet.

[0077] The tracing path construction module determines the actual mismatch end point based on the historical trajectory of the broken node and accumulates the space displacement to construct a pallet mismatch behavior tracing path table.

[0078] The broken occurrence time and data upload termination time of each pallet node are read from the cooperative broken point set, and the complete trajectory vector sequence is indexed by the pallet number. The trajectory vector sequence is a multi-dimensional vector flow data set, which at least includes three-dimensional space displacement, three-dimensional attitude angle, acceleration change sequence (acceleration value of each axis), instantaneous speed change sequence (estimated based on sliding integration) and time stamp sequence. All data is stored with uniform time resolution and is subjected to sliding filtering and gravity component elimination processing to ensure the continuity and physical consistency of the data. After extraction is completed, the system extracts the trajectory segment between the broken occurrence time and the upload termination time according to the time index, and performs cumulative space displacement calculation on the trajectory segment. Here, the time sequence integration method is adopted, that is, the filtered three-axis velocity vector is integrated by time, and the space position difference of the midpoint of each trajectory segment is added to obtain the total space displacement corresponding to the trajectory segment. The calculation result needs to be normalized by combining the calibration reference of the starting point position. The final output displacement value is in meters, reflecting the overall movement amplitude of the pallet from breaking to termination.

[0079] To correct for the cumulative displacement bias caused by the integration error of the inertial sensor, the system introduces a distance error prediction model trained on historical trajectory samples. The training process uses supervised learning to construct a residual fitting relationship between sample trajectory segments and actual geographic distances. The training data comes from a sample of shipping and receiving pallets that are fully registered during the historical transportation cycle. Each sample is required to have clear start and end station identifiers and a complete trajectory record on the supervision platform. The system calculates the cumulative spatial displacement for each sample trajectory segment and extracts trajectory statistical features based on the time span of the trajectory segment. The statistical feature dimensions include indicators such as the number of trajectory points, average acceleration, velocity change rate, attitude change range, and maximum displacement direction angle offset. The difference between the cumulative spatial displacement in these trajectory statistical features and the geographic distance between its start and end stations (calculated based on high-precision geographic coordinates) is the displacement error or residual.

[0080] The statistical features of all sample trajectories and the corresponding displacement residual sequences are constructed into a training sample set, which is input into a multi-layer perceptron network with a preset structure to perform residual fitting modeling. The model sets a three-layer perceptron structure, and the number of nodes in each layer is set to 128-64-32, which can be adjusted according to the sample size. ReLU is selected as the activation function, mean square error is selected as the loss function, and the training batch size is set to 32 to 64. After the training is completed, the system saves the model structure and parameters, and inputs the statistical features of the newly collected pallet trajectory segments into the model to obtain the corresponding prediction error value. The system superimposes or subtracts the predicted residual from the actual cumulative spatial displacement (processed as positive or negative values) to generate a corrected spatial displacement estimate. Subsequently, for the candidate mismatched terminal sites that have been screened in the previous step, their geographic coordinates are extracted and the geographic path distance from the mismatched starting site to each candidate site is calculated.

[0081] The absolute difference between the path distances of all candidate sites and the corrected spatial displacement is compared, and the site closest to the corrected displacement is selected as the actual mismatch endpoint. This site is the actual location where the pallet finally arrived but experienced a behavioral break. Based on this, the system connects the logistics station segment corresponding to the mismatch starting point of the pallet and the actual mismatch endpoint, marking it as a suspected mismatch path and outputting a traceability path index table. This index table uses the pallet number as the index field and records key fields such as the starting site, ending site, break time, trajectory offset, prediction error, corrected displacement, and actual path distance. This serves as the data foundation for identifying differential responsibility chains, reconstructing mismatches, and locating responsible parties.

[0082] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0083] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center through a wired (for example, infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0084] Those of ordinary skill in the art can realize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0085] Those of ordinary skill in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device, and module can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0086] In several embodiments provided in the present application, it should be understood that the disclosed system, device, and method can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed ones can be indirect coupling or communication connection through some interfaces, devices, or modules, which can be electrical, mechanical, or other forms.

[0087] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0088] In addition, the functional modules in each embodiment of the present application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0089] The functions, if realized in the form of software function modules and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0090] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0091] Finally: the above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application, should be included in the protection scope of the present application.

Claims

1. An agricultural product logistics supervision system based on the Internet of Things, characterized by: It includes data acquisition module, collaborative behavior construction module, collaborative fracture identification module, mismatch starting point tracing module, mismatch end point tracing module and tracing path construction module, among which: The data acquisition module collects the three-axis acceleration and angular velocity sequences of the transport pallet during loading, unloading and transportation, and constructs a set of continuous physical motion trajectories of the pallet; The collaborative behavior building module performs time synchronization processing on the trajectory vector sets of all pallets and constructs the sequential pallet collaborative behavior graph structure; The collaborative break identification module calculates the trajectory similarity between each pallet node and the remaining pallet nodes in the collaborative behavior graph structure, and identifies the collaborative break node set in the collaborative behavior graph; The mismatch origin tracing module screens the mismatch origins of pallets by comparing the break occurrence time of the collaborative break node with the shipment registration time of each logistics station within the transportation cycle; The mismatch endpoint tracing module screens the mismatch endpoints of pallets by comparing the data upload termination time of the broken node with the receiving registration time of each logistics station; The traceability path construction module determines the actual mismatch endpoint based on the cumulative spatial displacement corresponding to the historical trajectory of the broken node, and constructs a traceability path table for the pallet mismatch behavior.

2. The agricultural product logistics supervision system based on the Internet of Things according to claim 1 is characterized in that: The data acquisition module collects the three-axis acceleration and angular velocity sequence of the transport pallet during loading, unloading and transportation, and constructs a continuous physical motion trajectory set of the pallet, specifically including: Based on the inertial sensor nodes embedded at the bottom of each agricultural product transport pallet, the original data stream of three-axis acceleration and angular velocity is collected; Bind each sensor node with a unique pallet number and upload the real-time raw data stream to the transportation supervision platform according to the collection timestamp; Remove the gravity component from the three-axis acceleration and angular velocity data, and retain the dynamic acceleration and angular velocity changes corresponding to the motion changes; The acceleration and angular velocity trajectories are reconstructed based on the time series integration method to generate a set of continuous physical motion trajectories of the transport pallet, and each trajectory is labeled with the pallet number.

3. The agricultural product logistics supervision system based on the Internet of Things according to claim 2 is characterized in that: The trajectory reconstruction of acceleration and angular velocity based on the time series integration method specifically includes: In the transportation supervision platform, a sliding window filter is performed on the uploaded time series of three-axis acceleration and angular velocity raw data streams to remove high-frequency vibrations and short-term error peaks. Perform an integral calculation on the acceleration sequence to obtain the corresponding velocity sequence; The velocity sequence is integrated twice and combined with the initial point calibration value to construct the spatial displacement of the pallet in the entire time period; An integral operation is performed on the angular velocity sequence to reconstruct the pallet posture transformation sequence, which is then combined with the velocity sequence and spatial displacement to form a complete physical motion trajectory.

4. The agricultural product logistics supervision system based on the Internet of Things according to claim 1 is characterized in that: The collaborative behavior building module performs time synchronization processing on the trajectory vectors of all pallets to build a sequential pallet collaborative behavior graph structure, specifically including: Extract the trajectory data of all pallets from the physical motion trajectory set and perform vectorization conversion on the trajectory; The trajectory vectors of all pallets are time-synchronized and aligned according to the acquisition timestamps of the sensor nodes; Taking pallet numbers as nodes, connecting edges are established between pallet trajectory vectors to construct a temporal collaborative behavior graph structure.

5. The agricultural product logistics supervision system based on the Internet of Things according to claim 1 is characterized in that: The collaborative break identification module performs trajectory similarity calculation on each pallet node and the remaining pallet nodes in the collaborative behavior graph structure, and identifies the collaborative break node set in the collaborative behavior graph specifically including: Extract the time-series trajectory vector sequence of each pallet node in the collaborative behavior graph and perform normalization processing; Perform time-by-time similarity calculation on the vector sequences of each pallet node and all other nodes in the same time window; Count the average similarity of each node, and build a similarity mean mapping table between all nodes. Mark the pallet nodes with average similarity lower than the set fracture judgment threshold as collaborative fracture nodes. Summarize the index numbers of all collaborative break nodes and record the time when the break occurs, and output the collaborative break point set.

6. The agricultural product logistics supervision system based on the Internet of Things according to claim 1 is characterized in that: The mismatch starting point tracing module compares the fracture occurrence time of the coordinated fracture node with the shipment registration time of each logistics station within the transportation cycle to screen the pallet mismatch starting point, specifically including: Extract the fracture occurrence time of each pallet node from the collaborative fracture point set and set the retrieval time range tolerance; Establish a mapping between the shipping time series and the site identification of the logistics site within the transportation cycle, and compare the time when the break occurred with the shipping registration time of the logistics site; The shipment registration time that is consistent with the fracture occurrence time is selected within the set retrieval time range tolerance, and the site identification corresponding to the shipment registration time is marked as the starting point of the pallet mismatch.

7. The agricultural product logistics supervision system based on the Internet of Things according to claim 1 is characterized in that: The mismatch endpoint tracing module compares the data upload termination time of the broken node with the receiving registration time of each logistics station to screen the mismatch endpoints of the pallet, specifically including: Extract the data upload termination time of each pallet node in the collaborative breakpoint set; Obtain the receiving registration time and site identification of each logistics site within the transportation cycle, perform item-by-item matching operations on the upload end time and the receiving time, and select logistics sites with consistent time as mismatch endpoint candidates.

8. The agricultural product logistics supervision system based on the Internet of Things according to claim 1 is characterized in that: The traceability path construction module determines the actual mismatch endpoint based on the cumulative spatial displacement corresponding to the historical trajectory of the broken node, and constructs a pallet mismatch behavior traceability path table, specifically including: Extract the trajectory vector sequence of each fracture node and calculate the cumulative spatial displacement of the trajectory sequence from the fracture occurrence time to the upload termination time; The trajectory vector sequence and the corresponding cumulative spatial displacement are input into the trained distance error prediction model, and the cumulative spatial displacement is corrected according to the prediction error output by the model; Obtain the geographic coordinates of the mismatch starting point and each candidate mismatch end point, calculate the distance between all mismatched sites, compare the calculated results with the corrected cumulative spatial displacement, and select the site closest to the cumulative displacement as the actual mismatch end point; Connect the logistics segments corresponding to the pallet mismatch starting point and the actual mismatch end point to establish a traceability path table.

9. The agricultural product logistics supervision system based on the Internet of Things according to claim 8 is characterized in that: The distance error prediction model training method is: Extract sample trajectory segments where pallet shipment and receipt registrations are successful during the historical transportation cycle; Perform cumulative spatial displacement calculation on each trajectory segment, extract the trajectory vector sequence of the sample trajectory segment, and construct trajectory statistical features based on the cumulative spatial displacement; The geographical path distance is calculated based on the coordinates of the starting and ending stations of the trajectory segment, and a residual sequence of the cumulative displacement of the trajectory and the actual geographical distance is generated; A supervised learning sample set consisting of trajectory statistical features and corresponding residual sequences is constructed and input into a multi-layer perceptron network to perform residual fitting modeling.

Images