Intelligent identification method, device and equipment for express item missing scanning operation

Abstract

The invention discloses an intelligent identification method, device and equipment for express item missing scanning operation, the method is applied to the technical field of logistics, and the method comprises the steps: obtaining a current operation code, current time and a current website type of a to-be-detected express item, and constructing to-be-detected multi-dimensional data; inputting the to-be-tested multi-dimensional data into the affiliation state prediction model, and outputting at least two current affiliation states of the to-be-tested express item and a current prediction probability corresponding to each current affiliation state; generating a tracking result of the to-be-detected express based on the at least two current attribution states corresponding to the to-be-detected express and the current prediction probability corresponding to each current attribution state; according to the tracking result of the to-be-detected express item, determining a current time sequence of the to-be-detected express item in the plurality of time windows; inputting the current sequential sequence and the real-time operation sequence label of the express item to be detected into the missing scanning identification model, and outputting a current scanning operation result of the express item to be detected; according to the invention, the identification efficiency and the identification accuracy of missed scanning express items are improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an intelligent identification method, apparatus and device for missing items in express mail scanning operations. Background Technology

[0002] In the express delivery and logistics industry, monitoring the status of incoming and outgoing parcels at network points is a core aspect of operations management. However, existing technologies suffer from the following problems: Inconsistent data rules: Different monitoring reports use different data statistical rules, leading to data discrepancies and affecting the accuracy of operational decisions. Difficulty in identifying missed scan data: Traditional methods rely on manual judgment or simple logical judgment, which cannot accurately identify missed scan data such as "arrived but not dispatched," "delivered but not arrived," "signed but not arrived," and "signed but not dispatched," resulting in distorted monitoring data. Static operation code attribution rules: Existing systems use fixed attribution rules, which cannot adapt to business changes and abnormal scenarios. Insufficient prediction accuracy: Existing prediction methods lack learning from implicit patterns in historical data, resulting in low prediction accuracy. Summary of the Invention

[0003] This invention provides an intelligent identification method, device, and equipment for missing package scanning operations, which improves the identification efficiency and accuracy of missing packages.

[0004] On one hand, the present invention provides an intelligent identification method for missing items in express mail scanning operations, the method comprising: Obtain the current operation code, current time, and current branch type of the parcel to be tested, and construct the multidimensional data of the parcel to be tested; The multidimensional data to be tested is input into the home status prediction model, and at least two current home statuses of the parcel to be tested and the current prediction probability corresponding to each current home status are output. Based on at least two current attribution states corresponding to the package under test and the current predicted probability corresponding to each current attribution state, the tracking result of the package under test is generated. Based on the tracking results of the package under test, determine the current time sequence of the package under test in multiple time windows; The current time sequence and real-time operation sequence label of the package under test are input into the missed scan identification model, and the current scan operation result of the package under test is output; the current scan operation result indicates whether there is a missed scan operation in the package under test.

[0005] In one exemplary embodiment, the training method for the attribution state prediction model includes: The sample operation code, time, and branch type of the sample parcel are extracted from historical logistics data to construct multidimensional sample data of the sample parcel; the multidimensional sample data is labeled with the sample ownership status label of the sample parcel and the sample prediction probability corresponding to the sample ownership status label. The sample multidimensional data is input into the XGBoost model to extract the historical attribution features of the sample opcode, and based on the historical attribution features, the sample prediction attribution status and sample prediction probability of the sample opcode are predicted. Based on the difference between the predicted sample ownership status and the sample ownership status label, and the difference between the sample probability label and the sample predicted probability, the XGBoost model is trained to obtain the ownership status prediction model.

[0006] In one exemplary embodiment, the training method of the missed scan identification model includes: Historical operation data of sample parcels in sample outlets is obtained, and the historical operation data is converted into sample time series corresponding to multiple time windows according to a preset duration. The sample time series includes the operation code type, number of operations, and time interval between adjacent operations corresponding to each time window. The sample time series is labeled with the sample operation sequence tag and sample scanning operation tag of the sample parcel in the sample future time period. The sample time series is input into the benchmark sequence prediction model to capture the long-term and short-term dependencies of the sequence and generate a sample prediction benchmark sequence for future time periods; the benchmark sequence prediction model is obtained by training an LSTM model; The sample prediction baseline sequence is compared with the sample operation sequence label window by window to calculate the sample bias data; The sample bias data is input into an isolated forest network, which outputs sample anomaly scores. The isolated forest network is trained based on the sample anomaly score and the sample scanning operation label to obtain a scanning result prediction model; The missed scan identification model is constructed based on the baseline sequence prediction model and the scan result prediction model.

[0007] In one exemplary embodiment, training the isolated forest network based on the sample anomaly score and the sample scan operation label to obtain a scan result prediction model includes: Based on the comparison between the abnormal score of the sample and the preset score threshold, the sample scanning operation result of the sample package is determined; the sample scanning operation result indicates whether there is a missed scan operation on the sample package. Based on the difference between the sample scanning operation results and the sample scanning operation labels, the isolated forest network is trained to obtain the scanning result prediction model.

[0008] In one exemplary embodiment, determining the sample scanning operation result of the sample package based on the comparison result of the sample anomaly score and a preset score threshold includes: When the abnormal score of the sample is greater than the preset score threshold, it is determined that the sample package has been missed during scanning. When the abnormal score of the sample is less than or equal to the preset score threshold, it is determined that the sample package has no missed scanning operation.

[0009] In one exemplary embodiment, the method further includes: The historical parcel volume, historical time period type, and historical weather of the test site corresponding to the parcel under test are obtained as historical multidimensional features. The historical multidimensional features are input into the total number of express shipments prediction model. Based on the self-attention layer of the total number of express shipments prediction model, the first attention weight and first coding feature of historical time period type on the historical number of express shipments, and the second attention weight and second coding feature of historical weather on the historical number of express shipments are extracted. Historical coding features are determined based on the first attention weight, the first coding feature, the second attention weight, and the second coding feature; The decoder based on the total parcel volume prediction model decodes the historical encoding features to obtain the predicted total parcel volume of the test site in the future period.

[0010] In one exemplary embodiment, the method further includes: Obtain the network attributes, three-segment code, and salesperson attributes of the network corresponding to the parcel to be tested; A graph structure is constructed based on the branch attributes, three-segment codes, and salesperson attributes; in the graph structure, nodes represent the branch attributes, three-segment codes, and salesperson attributes, and edges represent the association between two connected nodes. The weight of each edge is determined based on the strength of the association between the two nodes connected by each edge in the graph structure. Based on the first convolutional layer of the graph convolutional network and the weight of each edge, the features of each node are weighted and fused with the features of its direct neighbors to obtain local association features. Based on the second convolutional layer, the features of the neighbor nodes corresponding to the direct neighbor nodes are fused into the local association features to capture higher-order association relationships, determine the similarity between each node and the indirect neighbor nodes, and filter out the recommended associated nodes for each node.

[0011] On the other hand, an intelligent identification device for missing item scanning operation of express mail is provided, the device comprising: The test data acquisition module is used to acquire the current operation code, current time, and current branch type of the parcel under test, and construct the test multidimensional data of the parcel under test; The current attribution status prediction module is used to input the multidimensional data to be tested into the attribution status prediction model and output at least two current attribution statuses of the parcel to be tested and the current prediction probability corresponding to each current attribution status. The result generation module is used to generate the tracking result of the package under test based on at least two current attribution states corresponding to the package under test and the current predicted probability corresponding to each current attribution state. The current sequence determination module is used to determine the current time sequence of the package under test in multiple time windows based on the tracking results of the package under test. The scan result prediction module is used to input the current time sequence and real-time operation sequence label of the parcel to be tested into the missed scan identification model, and output the current scan operation result of the parcel to be tested; the current scan operation result indicates whether there is a missed scan operation in the parcel to be tested.

[0012] On the other hand, an electronic device is provided, the device including a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to realize the intelligent identification method for the express mail missing scan operation as described above.

[0013] On the other hand, a computer storage medium is provided, which stores at least one instruction or at least one program, which is loaded and executed by a processor to realize the intelligent identification method for the express mail missing scan operation as described above.

[0014] On the other hand, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform an intelligent identification method for implementing the express mail missing scan operation as described above.

[0015] The intelligent identification method, device, and equipment for missing item scanning operations in express mail provided by this invention have the following technical effects: This invention acquires the current operation code, current time, and current branch type of a parcel to be tested, constructing multi-dimensional data for the parcel. This multi-dimensional data is then input into a home status prediction model, which outputs at least two current home statuses and a current predicted probability for each current home status. Based on these at least two current home statuses and their current predicted probabilities, a tracking result for the parcel is generated. According to the tracking result, the current time sequence of the parcel within multiple time windows is determined. The current time sequence and real-time operation sequence label of the parcel are input into a missed scan identification model, which outputs the current scan operation result for the parcel. This current scan operation result indicates whether a missed scan operation exists for the parcel. This improves the efficiency and accuracy of identifying missed scan parcels. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of an intelligent identification system for missing package scanning provided in the embodiments of this specification; Figure 2 This is a flowchart illustrating an intelligent identification method for missing items in express mail scanning provided in the embodiments of this specification; Figure 3 This is a flowchart illustrating a training method for an attribution state prediction model provided in an embodiment of this specification. Figure 4 This is a flowchart illustrating a training method for a missed scan identification model provided in the embodiments of this specification; Figure 5 This is a flowchart illustrating a method for predicting the total number of parcels at a test site in a future time period, as provided in the embodiments of this specification. Figure 6 This is a schematic diagram of the structure of an intelligent identification device for missing item scanning operation provided in the embodiments of this specification; Figure 7 This is a schematic diagram of the structure of a server provided in the embodiments of this specification. Detailed Implementation

[0018] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0020] Please see Figure 1 , Figure 1 This is a schematic diagram of an intelligent identification system for missing item scanning operations in express mail, as provided in the embodiments of this specification. Figure 1 As shown, the intelligent identification system for the missing package scanning operation can include at least server 01 and client 02.

[0021] Specifically, in the embodiments of this specification, server 01 may include a standalone server, a distributed server, or a server cluster composed of multiple servers. It may also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Server 01 may include a network communication unit, a processor, and a memory, etc. Specifically, server 01 can be used to input the current time sequence and real-time operation sequence tag of the package under test into the missed scan identification model, and output the current scan operation result of the package under test.

[0022] Specifically, in this embodiment, the client 02 may include physical devices such as smartphones, desktop computers, tablets, laptops, digital assistants, smart wearable devices, smart speakers, in-vehicle terminals, and smart TVs. It may also include software running on the physical device, such as web pages provided to users by service providers, or applications provided by those service providers. Specifically, the client 02 can be used to display the current scanning results of the package to be tested.

[0023] The following describes an intelligent identification method for missing items in express mail scanning operations according to the present invention. Figure 2 This is a flowchart illustrating an intelligent identification method for missing package scanning provided in this specification. This specification provides the operational steps described in the embodiments or flowchart, but based on conventional or non-inventive labor, more or fewer operational steps may be included. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only possible execution order. In actual system or server product execution, the methods shown in the embodiments or drawings can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment). Specifically, as shown... Figure 2 As shown, the method may include: S201: Obtain the current operation code, current time, and current branch type of the parcel to be tested, and construct the multidimensional data of the parcel to be tested. In the embodiments of this specification, in logistics operations, the operation code is the core identifier for identifying the express delivery processing stage (such as arrival, sorting, and delivery), and its attribution status (corresponding to the correct processing point / stage) directly affects the accuracy of express delivery tracking. First, key information needs to be extracted from historical logistics data: the operation code covers more than 30 core processing action codes recorded by the system (such as "01 - Arrival Scan" and "05 - Delivery Scan"). The time dimension needs to be accurate to the second to capture the processing time sequence pattern. The point type includes 6 categories such as hub centers, regional points, and community stations. At the same time, auxiliary features such as operation equipment number and operator ID are added to form a raw dataset containing 12 dimensions. The data cleaning phase needs to focus on addressing three types of issues: First, missing values, which are filled with the average of operation codes from the same network point and time period, and missing operation codes are completed using pattern matching; second, outliers, which are removed by using the 3σ principle to eliminate abnormal records that exceed a reasonable time interval (such as a single scan taking more than 10 minutes); and third, duplicate values, which are deduplicated based on the unique key of "operation code + time + network point", and finally a standardized multidimensional feature vector is constructed.

[0024] S203: Input the multidimensional data to be tested into the home status prediction model, and output at least two current home statuses of the parcel to be tested and the current prediction probability corresponding to each current home status.

[0025] S205: Based on at least two current attribution states corresponding to the package under test and the current predicted probability corresponding to each current attribution state, generate the tracking result of the package under test.

[0026] S207: Based on the tracking results of the package under test, determine the current time sequence of the package under test in multiple time windows.

[0027] S209: Input the current time sequence and real-time operation sequence label of the package under test into the missed scan identification model, and output the current scan operation result of the package under test; the current scan operation result indicates whether there is a missed scan operation in the package under test.

[0028] In the embodiments of this specification, the current attribution status of the current operation code includes "attributed hub center" and "attributed regional outlet". The output of the attribution status prediction model not only includes the classification result, but also generates the probability distribution of each operation code corresponding to each attribution status (e.g., the probability of operation code "03 - sorting scan" attributing to the hub center is 92%, and the probability of attributing to the regional outlet is 8%), providing a probabilistic basis for subsequent decisions. The daily operation sequence of the outlet is a key carrier reflecting the integrity of express mail processing. If there is a missed scan (e.g., no delivery scan is recorded after the express mail arrives at the port), it will lead to the interruption of express mail tracking and affect the customer experience. The missed scan identification model can be trained using an LSTM model. This step first learns the normal pattern of the operation sequence through the LSTM model: First, the daily operation data of the outlet is converted into a time series sequence, with a time window of 1 hour. Each window includes features such as operation code type, number of operations, and time interval between adjacent operations within that time period, forming a daily operation sequence of length 24 (corresponding to 24 hours). The LSTM model captures short- and long-term dependencies in sequences through two hidden layers (64 neurons per layer). For example, under normal circumstances, "sorting scan" will occur 2-4 hours after "arrival scan" and "delivery scan" will occur 4-8 hours after "sorting scan". The model learns from the normal operation sequences of the past 6 months (excluding known missed scan records) to generate a prediction benchmark for the operation sequence of the next 24 hours (including the type of opcode, time interval and frequency that should occur).

[0029] In some embodiments, such as Figure 3 As shown, Figure 3 This is a flowchart illustrating a training method for an attribution state prediction model, which includes: S301: Extract the sample operation code, time, and branch type of the sample parcel from historical logistics data to construct multidimensional sample data of the sample parcel; the multidimensional sample data is labeled with the sample ownership status label of the sample parcel and the sample prediction probability corresponding to the sample ownership status label. S303: Input the multidimensional data of the sample into the XGBoost model, extract the historical attribution features of the sample opcode, and predict the sample prediction attribution status and sample prediction probability of the sample opcode based on the historical attribution features. S305: Based on the difference between the predicted sample ownership status and the sample ownership status label, and the difference between the sample probability label and the sample predicted probability, the XGBoost model is trained to obtain the ownership status prediction model.

[0030] In the embodiments described in this specification, the XGBoost classifier is selected as the model because it has strong nonlinear fitting capabilities and feature importance evaluation functions, and can effectively learn the complex relationship between opcodes and their attribution status. During training, historical data from n months are divided into training and testing sets in a 7:3 ratio. The attribution status of the opcode (e.g., "attributed to hub center" or "attributed to regional outlet") is used as the label. Hyperparameters such as the learning rate (set to 0.1) and tree depth (set to 6) are optimized through 5-fold cross-validation. The model output not only includes the classification results, but also generates the probability distribution of each opcode for each attribution status (e.g., the probability of opcode "03- sorting scan" attributing to a hub center is 92%, and the probability of attributing to a regional outlet is 8%), providing a probabilistic basis for subsequent decisions.

[0031] In some embodiments, to adapt to dynamic business changes (such as the addition of new operation codes or adjustments to branch functions), an automatic update mechanism is established every 7 days: the system incrementally collects logistics data daily, automatically merges new data with historical data every Monday morning, and retrains the model using an online learning strategy (fixing the basic parameters of the historical model and only fine-tuning the weights of leaf nodes with new data), synchronously updating the operation code attribution determination rules in the rule base. This mechanism avoids the cumbersome process of manually adjusting rules periodically in the traditional way, increasing the rule base update frequency from once a month to once a week, while maintaining the model accuracy at over 95%, significantly reducing the cost of manual intervention. For example, updated sample multidimensional data is acquired at preset intervals, and the attribution status prediction model is updated based on the updated sample multidimensional data to obtain an updated attribution status prediction model.

[0032] In some embodiments, such as Figure 4 As shown, Figure 4 This is a flowchart illustrating a training method for a missed scan detection model, which includes: S401: Obtain historical operation data of sample parcels in the sample network points, and convert the historical operation data into sample time sequence corresponding to multiple time windows according to a preset duration; the sample time sequence includes the operation code type, number of operations and adjacent operation time intervals corresponding to each time window; the sample time sequence is labeled with the sample operation sequence label and sample scanning operation label of the sample parcel in the sample future time period. S403: Input the sample time series sequence into the benchmark sequence prediction model to capture the long-term and short-term dependencies of the sequence and generate a sample prediction benchmark sequence for future time periods; the benchmark sequence prediction model is obtained by training an LSTM model; S405: Compare the sample prediction benchmark sequence with the sample operation sequence label window by window, and calculate the sample bias data; S407: Input the sample bias data into the isolated forest network and output the sample anomaly score; S409: Train the isolated forest network based on the sample anomaly score and the sample scanning operation label to obtain a scanning result prediction model; S4011: Construct the missed scan identification model based on the baseline sequence prediction model and the scan result prediction model.

[0033] In the embodiments described in this specification, an LSTM model is used to learn the normal patterns of daily operation sequences at service points, generating a prediction baseline. Combined with the Isolation Forest algorithm, the actual sequences are compared with the prediction results to calculate anomaly scores. Based on these scores, the system can intelligently identify missed scans such as "delivered but not dispatched," "dispatched but not arrived," "signed but not arrived," and "signed but not dispatched," improving the identification accuracy to 98.5%, significantly higher than traditional methods.

[0034] The daily operation sequence of the branch is a key indicator of the completeness of parcel processing. Missed scans (such as parcels arriving at the port without being recorded for dispatch scanning) will lead to interruptions in parcel tracking and negatively impact customer experience. This step first uses an LSTM model to learn the normal patterns of the operation sequence: The daily operation data of the branch is first converted into a time series, with a 1-hour time window. Each window includes features such as the operation code type, number of operations, and time intervals between adjacent operations within that time period, forming a 24-hour daily operation sequence. The LSTM model captures the short-term and long-term dependencies of the sequence through two hidden layers (64 neurons per layer)—for example, under normal circumstances, "sorting scan" will occur 2-4 hours after "arrival scan," and "dispatch scan" will occur 4-8 hours after "sorting scan." By learning from the normal operation sequences of the past 6 months (excluding known missed scan records), the model generates a prediction benchmark for the operation sequence of the next 24 hours (including the expected operation code type, time interval, and frequency).

[0035] To accurately identify anomalies, an Isolation Forest algorithm is introduced in conjunction with LSTM prediction results: Isolation Forest isolates anomalous sequences (those with large deviations from the normal pattern) more quickly by randomly partitioning the feature space, making it suitable for processing high-dimensional time-series data. The system compares the actual operation sequence with the LSTM prediction benchmark window by window, calculating deviation indices (such as the number of missing opcodes, operation time deviation duration, and operation frequency deviation rate). Isolation Forest generates anomaly scores based on these indices—higher scores indicate a more severe deviation from the normal pattern. When the score exceeds a preset threshold (calibrated based on historical missed scan cases, with the threshold set at 0.85), it is determined to be a suspected missed scan.

[0036] To address four common types of missed scans in logistics scenarios, the system establishes specific identification rules: "Arrived but not dispatched" refers to a "port arrival scan" occurring in the actual sequence but without a subsequent "delivery scan," with an interval exceeding 12 hours; "Delivered but not arrived" refers to a "delivery scan" occurring but without a preceding "port arrival scan"; "Signed but not arrived" refers to a "signed for but not received" scan occurring but without a "port arrival scan"; and "Signed but not dispatched" refers to a "signed for but not dispatched" scan occurring. By using LSTM to ensure the accuracy of normal pattern learning and combining it with the high sensitivity of isolated forests to anomalies, this fusion model improves the accuracy of missed scan identification to 98.5%, a 16.5 percentage point improvement compared to traditional rule engines based on fixed thresholds (accuracy of approximately 82%). This significantly reduces parcel tracking complaints caused by missed scans and greatly improves operational stability.

[0037] In some embodiments, training the isolated forest network based on the sample anomaly score and the sample scanning operation label to obtain a scanning result prediction model includes: Based on the comparison between the abnormal score of the sample and the preset score threshold, the sample scanning operation result of the sample package is determined; the sample scanning operation result indicates whether there is a missed scan operation on the sample package. Based on the difference between the sample scanning operation results and the sample scanning operation labels, the isolated forest network is trained to obtain the scanning result prediction model.

[0038] In the embodiments of this specification, the sample scanning operation result of the sample package is first determined by comparing the sample anomaly score with a preset score threshold. Then, the target loss data is determined based on the difference between the sample scanning operation result and the sample scanning operation label. The network parameters of the isolated forest network are then adjusted based on the target loss data until the training termination condition is met, and the isolated forest network at the end of training is used as the scanning result prediction model.

[0039] In some embodiments, determining the sample scanning operation result of the sample package based on the comparison result of the sample anomaly score and a preset score threshold includes: When the abnormal score of the sample is greater than the preset score threshold, it is determined that the sample package has been missed during scanning. When the abnormal score of the sample is less than or equal to the preset score threshold, it is determined that the sample package has no missed scanning operation.

[0040] In the embodiments of this specification, a preset score threshold corresponding to abnormal situations can be set. When the abnormal score of the sample is greater than the preset score threshold, it is determined that the sample package has a missed scan operation; when the abnormal score of the sample is less than or equal to the preset score threshold, it is determined that the sample package has no missed scan operation.

[0041] In some embodiments, such as Figure 5 As shown, Figure 5 This is a flowchart illustrating a method for predicting the total number of parcels received at a test site in a future time period. The method includes: S501: Obtain the historical parcel volume, historical time period type, and historical weather of the test site corresponding to the parcel under test in different historical periods, as historical multidimensional features; S503: Input the historical multidimensional features into the total shipment prediction model, and extract the first attention weight and first coding feature of historical time period type on the historical shipment volume, and the second attention weight and second coding feature of historical weather on the historical shipment volume based on the self-attention layer of the total shipment prediction model. S505: Determine historical coding features based on the first attention weight, the first coding feature, the second attention weight, and the second coding feature; S507: The decoder based on the total parcel volume prediction model decodes the historical encoding features to obtain the predicted total parcel volume of the test site in the future period.

[0042] In the embodiments of this specification, parcel volume prediction is the core basis for logistics resource allocation (such as personnel scheduling, vehicle dispatching, and warehouse inventory preparation), and requires comprehensive consideration of multiple influencing factors. First, three core features are extracted: historical parcel volume features include daily parcel volume, weekly average parcel volume, monthly peak volume for the past 12 months, and the parcel volume trend during the same period (e.g., the same period last year, the same period last month); holiday features cover statutory holidays (Spring Festival, National Day, etc.) and e-commerce promotional festivals (Double Eleven, 618, etc.), which are converted into numerical features through one-hot encoding, marking the special period of 7 days before and after the holidays; weather features include daily rainfall, wind force level, and temperature range in the area where the branch is located, and the impact coefficient of severe weather (e.g., heavy rain, typhoon) on parcel volume is marked based on historical data (e.g., parcel volume decreases by 15% during heavy rain). These three types of features constitute a 28-dimensional input feature vector, which is then standardized and used for model training. In the application of the model, the historical parcel volume, historical time period type, and historical weather of the test site corresponding to the parcel under test are obtained in different historical time periods as historical multidimensional features. The historical multidimensional features are input into the parcel volume prediction model. Based on the self-attention layer of the parcel volume prediction model, the first attention weight and first encoding feature of the historical time period type on the historical parcel volume, and the second attention weight and second encoding feature of the historical weather on the historical parcel volume are extracted. Based on the first attention weight, the first encoding feature, the second attention weight, and the second encoding feature, the historical encoding features are determined. Based on the decoder of the parcel volume prediction model, the historical encoding features are decoded to obtain the predicted parcel volume of the test site in the future time period.

[0043] In some embodiments, the training method for the total shipment prediction model includes: The sample outlets are obtained with sample parcel volume, sample time period type, and sample weather data at different sample time periods as multidimensional features of the samples; the multidimensional features of the samples are labeled with the total sample parcel volume of the sample outlets in future sample time periods. The multidimensional features of the sample are input into the Transformer model, and the first sample weight and first sample coding feature of the sample time period type on the sample express quantity, and the second sample weight and second sample coding feature of the sample weather on the sample express quantity are extracted based on the self-attention layer of the Transformer model. The comprehensive coding features of the samples are determined based on the first sample weight, the first sample coding feature, the second sample weight, and the second sample coding feature. The decoder based on the Transformer model decodes the comprehensive coding features of the sample to obtain the total number of predicted express parcels of the sample outlet in the future time period of the sample. Based on the difference between the predicted total number of parcels from the sample and the total number of parcels labeled in the sample, the Transformer model is trained to obtain the total number of parcels prediction model.

[0044] In the embodiments of this specification, the Transformer architecture is used to process time series prediction tasks. Its self-attention mechanism can effectively capture the correlation between different time steps and different features. For example, it can identify the pattern that "the volume of express deliveries begins to increase 3 days before Double Eleven, reaches its peak on the day of the festival, and gradually declines 5 days after the festival," and can also correlate the superimposed impact of "heavy rain + weekend" on the volume of express deliveries. The Transformer model is divided into two parts: an encoder and a decoder. The encoder consists of 6 layers of encoding blocks, each layer containing a multi-head self-attention layer (8 heads) and a feedforward neural network. Through self-attention weight calculation, it extracts key patterns from the multi-dimensional features of the past 30 days (such as the correlation between holidays and express delivery volume, and the correlation between weather and delivery efficiency), and encodes the historical data into a fixed-dimensional feature vector. The decoder also consists of 6 layers of decoding blocks. Based on the feature vector output by the encoder, combined with the known features of the target time step (T+1 / 2 / 3) (such as the weather forecast for T+1 day, whether it is a holiday), it generates the predicted values ​​of express delivery volume for the next 1, 2, and 3 days.

[0045] To further improve prediction accuracy, a multi-model weighted average fusion strategy is adopted: three basic prediction models are constructed (one based on a Transformer model with different feature combinations, and the other a traditional ARIMA time series model). After each model outputs its prediction results, the system dynamically adjusts its weights based on its historical prediction accuracy. For example, if the MAE of a Transformer model over the past 30 days is 5.2% and the ARIMA model's error is 8.5%, then the former's weight is set to 0.65 and the latter's to 0.35. The weights are updated weekly based on the latest prediction results. Through this fusion strategy, the model's prediction accuracy for parcel volume on T+1 / 2 / 3 reaches 93.2%, 91.5%, and 90.8% respectively, with the overall accuracy improved to 92%, a 5 percentage point improvement compared to the single Transformer model (87% accuracy). High-precision prediction allows branches to adjust resources 3 days in advance: for example, if a 20% increase in parcel volume is predicted for T+2, 5 additional sorting personnel and 2 additional transport vehicles can be deployed in advance, reducing resource waste and avoiding insufficient transport capacity.

[0046] In some embodiments, the method further includes: Obtain the network attributes, three-segment code, and salesperson attributes of the network corresponding to the parcel to be tested; A graph structure is constructed based on the branch attributes, three-segment codes, and salesperson attributes; in the graph structure, nodes represent the branch attributes, three-segment codes, and salesperson attributes, and edges represent the association between two connected nodes. The weight of each edge is determined based on the strength of the association between the two nodes connected by each edge in the graph structure. Based on the first convolutional layer of the graph convolutional network and the weight of each edge, the features of each node are weighted and fused with the features of its direct neighbors to obtain local association features. Based on the second convolutional layer, the features of the neighbor nodes corresponding to the direct neighbor nodes are fused into the local association features to capture higher-order association relationships, determine the similarity between each node and the indirect neighbor nodes, and filter out the recommended associated nodes for each node.

[0047] In the embodiments of this specification, dimensions such as outlets, three-segment codes, and salespersons are treated as graph nodes, and relational edges are constructed to represent the strength of association between dimensions. A GCN (Graph Convolutional Network) is applied to learn features from the relational graph, capturing high-order associations between nodes. Based on the learned feature representations, the system can achieve intelligent dimension switching and drill-down analysis. When the user selects a dimension, the system automatically recommends the most relevant drill-down dimension, improving dimension drill-down efficiency by 40% and providing accurate support for operational decisions.

[0048] Logistics operation analysis needs to be approached from multiple dimensions (network points, three-segment codes, and salespersons). Traditional manual dimension switching suffers from low efficiency and inaccurate identification of related dimensions. Intelligent analysis needs to be achieved through graph structure modeling and GCN learning. First, a logistics relationship graph is constructed: three types of core objects are set as graph nodes—network point nodes include 10 attributes such as network point number, average daily parcel volume, and location; three-segment code nodes (representing parcel sorting areas, such as "xxxxxx" corresponding to xx city xx district) include 8 attributes such as area, population density, and average daily parcel volume; salesperson nodes include 6 attributes such as employee ID, average daily delivery volume, and on-time delivery rate. Relationship edges are used to represent the strength of association between nodes, and the calculation logic is as follows: Association strength between a branch and a three-segment code = Number of three-segment code parcels processed by the branch / Total parcel volume of the branch; Association strength between a branch and a salesperson = Delivery volume of the salesperson assigned to that branch / Total delivery volume of the salesperson; Association strength between a salesperson and a three-segment code = Delivery volume of the three-segment code handled by the salesperson / Total delivery volume of the salesperson. These indicators quantify the association relationships in each dimension into edge weights in the graph structure.

[0049] Feature learning is performed using GCN (Graph Convolutional Network). GCN fuses the features of a node itself and its neighbors through two layers of graph convolutional operations: The first convolutional layer weights and fuses the features of each node with the features of its direct neighbors (nodes with an association strength ≥ 0.3) (the weights being the association strength), learning the local association features of the node; the second convolutional layer, based on the first layer, further fuses the features of the neighbors' neighbors to capture higher-order associations (such as the indirect association between branch A → three-segment code B → salesperson C). For example, the features of a branch not only include its average daily parcel volume, but also incorporate the population density of the three-segment code area it processes frequently, as well as the on-time delivery rate of the salespersons belonging to that branch, forming a more comprehensive node feature representation (the dimensionality is uniformly increased from the original 6-10 dimensions of node attributes to 64 dimensions).

[0050] Based on the node association features learned by GCN, the system enables intelligent dimension switching and drill-down: When a user selects an analysis dimension (such as "Beijing xx district outlet"), the system calculates the feature similarity between that node and nodes in other dimensions (calculating cosine similarity based on the feature vectors output by GCN), and automatically recommends the drill-down dimension with the highest similarity. For example, after selecting "xx district outlet", the system recommends drilling down to "the three-segment code frequently processed by this outlet", and then drilling down to "the salesperson responsible for this three-segment code". This intelligent recommendation avoids the blind attempts of "outlet → region → salesperson" in traditional analysis, reducing the average number of steps for dimension drill-down from 5 to 3, improving efficiency by 40%. Meanwhile, the system provides decision support based on high-order correlation features: for example, by drilling down, it is found that in the correlation chain of "a certain branch in xx district → three-segment code xxxxx → salesperson Zhang", Zhang's on-time delivery rate is only 82% (lower than the branch average of 90%), and the express delivery delay rate in the three-segment code area is 12%. The system automatically prompts "adjust the three-segment code area under Zhang's responsibility or strengthen training", helping operations personnel to accurately locate problems and promote the overall on-time delivery rate of the branch to increase by 3-5 percentage points per month, providing data-driven and precise support for logistics operation decisions.

[0051] As can be seen from the technical solutions provided in the embodiments of this specification above, the embodiments of this specification obtain the current operation code, current time, and current branch type of the parcel to be tested, and construct the multi-dimensional data of the parcel to be tested; input the multi-dimensional data to be tested into the attribution status prediction model, and output at least two current attribution statuses of the parcel to be tested and the current prediction probability corresponding to each current attribution status; generate the tracking result of the parcel to be tested based on the at least two current attribution statuses of the parcel to be tested and the current prediction probability corresponding to each current attribution status; determine the current time sequence of the parcel to be tested in multiple time windows based on the tracking result of the parcel to be tested; input the current time sequence and real-time operation sequence label of the parcel to be tested into the missed scan identification model, and output the current scan operation result of the parcel to be tested; the current scan operation result indicates whether there is a missed scan operation of the parcel to be tested. This improves the identification efficiency and accuracy of missed scan parcels.

[0052] This embodiment provides a deep learning-based intelligent identification method for missed scan data of express mail inbound and outbound shipments. It extracts and cleans information such as operation codes, time, and branch type from historical logistics data to construct a multi-dimensional feature vector. The XGBoost model is used to classify these features, learn the historical attribution status of operation codes, and output the attribution probability distribution of each operation code. An LSTM model is used to learn the normal pattern of daily operation sequences at branches, generating a prediction benchmark. Combined with the Isolation Forest algorithm, the actual sequence is compared with the prediction results to calculate anomaly scores. Based on these scores, the system can intelligently identify missed scan situations such as "delivered but not dispatched," "delivered but not arrived," "signed but not arrived," and "signed but not dispatched." It extracts multi-dimensional features including historical express mail volume, holidays, and weather, and uses a Transformer architecture for time series prediction. The Transformer encoder processes historical data, and the decoder generates express mail volume predictions for the next T+1 / 2 / 3 days. By fusing the results of multiple prediction models using a weighted averaging strategy and dynamically adjusting the weights based on historical prediction accuracy, it provides high-precision express mail volume predictions, offering more accurate decision support for branch resource allocation.

[0053] This specification also provides an intelligent identification device for missing items during express mail scanning, such as... Figure 6 As shown, the device includes: The test data acquisition module 610 is used to acquire the current operation code, current time and current branch type of the parcel to be tested, and construct the test multidimensional data of the parcel to be tested; The current attribution status prediction module 620 is used to input the multidimensional data to be tested into the attribution status prediction model and output at least two current attribution statuses of the parcel to be tested and the current prediction probability corresponding to each current attribution status. The result generation module 630 is used to generate the tracking result of the package under test based on at least two current ownership states corresponding to the package under test and the current predicted probability corresponding to each current ownership state. The current sequence determination module 640 is used to determine the current time sequence of the package under test in multiple time windows based on the tracking results of the package under test. The scanning result prediction module 650 is used to input the current time sequence and real-time operation sequence label of the parcel to be tested into the missed scan identification model, and output the current scanning operation result of the parcel to be tested; the current scanning operation result indicates whether there is a missed scan operation in the parcel to be tested.

[0054] In one exemplary embodiment, the apparatus further includes: The sample multidimensional data construction module is used to extract the sample operation code, time, and branch type of the sample express from historical logistics data to construct the sample multidimensional data of the sample express; the sample multidimensional data is labeled with the sample ownership status label of the sample express and the sample prediction probability corresponding to the sample ownership status label. The sample state prediction module is used to input the multidimensional data of the sample into the XGBoost model, extract the historical attribution features of the sample opcode, and predict the sample prediction attribution state and sample prediction probability of the sample opcode based on the historical attribution features. The attribution state model training module is used to train the XGBoost model based on the difference between the predicted attribution state of the sample and the attribution state label of the sample, and the difference between the probability label of the sample and the predicted probability of the sample, so as to obtain the attribution state prediction model.

[0055] In one exemplary embodiment, the apparatus further includes: The sample sequence acquisition module is used to acquire historical operation data of sample parcels in the sample network, and convert the historical operation data into sample time sequence corresponding to multiple time windows according to a preset duration; the sample time sequence includes the operation code type, operation number and adjacent operation time interval corresponding to each time window; the sample time sequence is labeled with the sample operation sequence label and sample scanning operation label of the sample parcel in the sample future time period. The sample prediction module is used to input the sample time series into the benchmark sequence prediction model, capture the long-term and short-term dependencies of the sequence, and generate the sample prediction benchmark sequence for future time periods; the benchmark sequence prediction model is obtained by training an LSTM model; The sample bias calculation module is used to compare the sample prediction benchmark sequence with the sample operation sequence label window by window to calculate the sample bias data. The sample score prediction module is used to input the sample bias data into the isolated forest network and output the sample anomaly score; The scanning result model training module is used to train the isolated forest network based on the sample anomaly score and the sample scanning operation label to obtain a scanning result prediction model. The model building module is used to build the missed scan identification model based on the baseline sequence prediction model and the scan result prediction model.

[0056] In one exemplary embodiment, the scan result model training module includes: The sample result determination unit is used to determine the sample scanning operation result of the sample package based on the comparison result of the sample anomaly score and the preset score threshold; the sample scanning operation result indicates whether there is a missed scan operation in the sample package; The scanning model training unit is used to train the isolated forest network based on the difference between the sample scanning operation results and the sample scanning operation labels to obtain the scanning result prediction model.

[0057] In one exemplary embodiment, the sample result determination unit includes: The first determining subunit is used to determine that the sample package has a missed scan operation when the abnormal score of the sample is greater than the preset score threshold. The second determining subunit is used to determine that there is no missed scanning operation in the sample package when the abnormal score of the sample is less than or equal to the preset score threshold.

[0058] In one exemplary embodiment, the apparatus further includes: The historical feature acquisition module is used to acquire the historical parcel volume, historical time period type, and historical weather of the test site corresponding to the parcel under test in different historical periods, as historical multidimensional features. The weight determination module is used to input the historical multidimensional features into the total number of express shipments prediction model, and extract the first attention weight and first coding feature of historical time period type on the historical number of express shipments, and the second attention weight and second coding feature of historical weather on the historical number of express shipments based on the self-attention layer of the total number of express shipments prediction model. The feature encoding module is used to determine historical encoding features based on the first attention weight, the first encoding feature, the second attention weight, and the second encoding feature; The total volume prediction module is used to decode the historical encoding features based on the decoder of the total volume prediction model of the express mail, so as to obtain the predicted total volume of express mail for the test outlet in the future period.

[0059] In one exemplary embodiment, the apparatus further includes: The attribute acquisition module is used to acquire the network attributes, three-segment code, and salesperson attributes of the network point corresponding to the parcel under test. The graph structure construction module is used to construct a graph structure based on the branch attributes, three-segment codes, and salesperson attributes; in the graph structure, nodes represent the branch attributes, three-segment codes, and salesperson attributes, and edges represent the association relationship between two connected nodes. The weight determination module is used to determine the weight of each edge based on the strength of the association between the two nodes connected by each edge in the graph structure. The fusion module is used to perform weighted fusion of the features of each node with the features of its direct neighbors based on the first convolutional layer of the graph convolutional network and the weight of each edge, so as to obtain local association features. The filtering module is used to fuse the neighbor node features corresponding to the direct neighbor node in the local association features based on the second convolutional layer, capture higher-order association relationships, determine the similarity between each node and the indirect neighbor node, and filter out the recommended associated nodes for each node.

[0060] The apparatus and method embodiments described herein are based on the same inventive concept.

[0061] This specification provides an electronic device including a processor and a memory. The memory stores at least one instruction or at least one program, which is loaded and executed by the processor to implement the intelligent identification method for express mail missing scan operation as provided in the above method embodiments.

[0062] Embodiments of the present invention also provide a computer storage medium, which can be disposed in a terminal to store at least one instruction or at least one program related to the intelligent identification method for implementing a package missing scan operation in the method embodiment. The at least one instruction or at least one program is loaded and executed by the processor to implement the intelligent identification method for package missing scan operation provided in the above method embodiment.

[0063] Embodiments of the present invention also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform an intelligent identification method for implementing the express mail missing scan operation provided in the above-described method embodiments.

[0064] Optionally, in the embodiments of this specification, the storage medium may be located at at least one of the multiple network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0065] The memory described in the embodiments of this specification can be used to store software programs and modules. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for the functions, etc.; the data storage area may store data created according to the use of the device, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory may also include a memory controller to provide the processor with access to the memory.

[0066] The intelligent identification method for missing package scanning provided in this specification can be executed on a mobile terminal, computer terminal, server, or similar computing device. Taking a server as an example... Figure 7 This is a hardware structure block diagram of a server for an intelligent identification method for missing package scanning operations provided in the embodiments of this specification. Figure 7As shown, the server 700 can vary significantly due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 710 (CPUs 710 may include, but are not limited to, microprocessors (MCUs) or programmable logic devices (FPGAs), a memory 730 for storing data, and one or more storage media 720 (e.g., one or more mass storage devices) for storing application programs 723 or data 722. The memory 730 and storage media 720 may be temporary or persistent storage. The program stored in the storage media 720 may include one or more modules, each module may include a series of instruction operations on the server. Furthermore, the CPU 710 may be configured to communicate with the storage media 720 and execute the series of instruction operations stored in the storage media 720 on the server 700. Server 700 may also include one or more power supplies 760, one or more wired or wireless network interfaces 750, one or more input / output interfaces 740, and / or one or more operating systems 721, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0067] The input / output interface 740 can be used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of server 700. In one example, the input / output interface 740 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the input / output interface 740 may be a radio frequency (RF) module used for wireless communication with the Internet.

[0068] Those skilled in the art will understand that Figure 7 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, server 700 may also include... Figure 7 The more or fewer components shown, or having the same Figure 7 The different configurations shown.

[0069] As can be seen from the embodiments of the intelligent identification method, device, electronic device, or storage medium for missed scan operations of express parcels provided by the present invention, the present invention acquires the current operation code, current time, and current branch type of the parcel to be tested, and constructs multi-dimensional data of the parcel to be tested; inputs the multi-dimensional data to be tested into the attribution status prediction model, and outputs at least two current attribution statuses of the parcel to be tested and the current prediction probability corresponding to each current attribution status; generates the tracking result of the parcel to be tested based on the at least two current attribution statuses of the parcel to be tested and the current prediction probability corresponding to each current attribution status; determines the current time sequence of the parcel to be tested in multiple time windows based on the tracking result of the parcel to be tested; inputs the current time sequence and real-time operation sequence label of the parcel to be tested into the missed scan identification model, and outputs the current scan operation result of the parcel to be tested; the current scan operation result indicates whether the parcel to be tested has a missed scan operation. This improves the identification efficiency and accuracy of missed scan parcels.

[0070] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0071] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0072] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer storage medium, such as a read-only memory, a disk, or an optical disk.

[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart identification method for missing items in express mail scanning operations, characterized in that, The method includes: Obtain the current operation code, current time, and current branch type of the parcel to be tested, and construct the multidimensional data of the parcel to be tested; The multidimensional data to be tested is input into the home status prediction model, and at least two current home statuses of the parcel to be tested and the current prediction probability corresponding to each current home status are output. Based on at least two current attribution states corresponding to the package under test and the current predicted probability corresponding to each current attribution state, the tracking result of the package under test is generated. Based on the tracking results of the package under test, determine the current time sequence of the package under test in multiple time windows; The current time sequence and real-time operation sequence label of the package under test are input into the missed scan identification model, and the current scan operation result of the package under test is output; the current scan operation result indicates whether there is a missed scan operation in the package under test.

2. The method according to claim 1, characterized in that, The training method for the attribution status prediction model includes: The sample operation code, time, and branch type of the sample parcel are extracted from historical logistics data to construct multidimensional sample data of the sample parcel; the multidimensional sample data is labeled with the sample ownership status label of the sample parcel and the sample prediction probability corresponding to the sample ownership status label. The sample multidimensional data is input into the XGBoost model to extract the historical attribution features of the sample opcode, and based on the historical attribution features, the sample prediction attribution status and sample prediction probability of the sample opcode are predicted. Based on the difference between the predicted sample ownership status and the sample ownership status label, and the difference between the sample probability label and the sample predicted probability, the XGBoost model is trained to obtain the ownership status prediction model.

3. The method according to claim 1, characterized in that, The training method for the missed scan identification model includes: Historical operation data of sample parcels in sample outlets is obtained, and the historical operation data is converted into sample time series corresponding to multiple time windows according to a preset duration. The sample time series includes the operation code type, number of operations, and time interval between adjacent operations corresponding to each time window. The sample time series is labeled with the sample operation sequence tag and sample scanning operation tag of the sample parcel in the sample future time period. The sample time series is input into the benchmark sequence prediction model to capture the long-term and short-term dependencies of the sequence and generate a sample prediction benchmark sequence for future time periods; the benchmark sequence prediction model is obtained by training an LSTM model; The sample prediction baseline sequence is compared with the sample operation sequence label window by window to calculate the sample bias data; The sample bias data is input into an isolated forest network, which outputs sample anomaly scores. The isolated forest network is trained based on the sample anomaly score and the sample scanning operation label to obtain a scanning result prediction model; The missed scan identification model is constructed based on the baseline sequence prediction model and the scan result prediction model.

4. The method according to claim 3, characterized in that, The step of training the isolated forest network based on the sample anomaly score and the sample scanning operation label to obtain a scanning result prediction model includes: Based on the comparison between the abnormal score of the sample and the preset score threshold, the sample scanning operation result of the sample package is determined; the sample scanning operation result indicates whether there is a missed scan operation on the sample package. Based on the difference between the sample scanning operation results and the sample scanning operation labels, the isolated forest network is trained to obtain the scanning result prediction model.

5. The method according to claim 4, characterized in that, The step of determining the sample scanning operation result of the sample package based on the comparison result of the sample anomaly score and the preset score threshold includes: When the abnormal score of the sample is greater than the preset score threshold, it is determined that the sample package has been missed during scanning. When the abnormal score of the sample is less than or equal to the preset score threshold, it is determined that the sample package has no missed scanning operation.

6. The method according to claim 1, characterized in that, The method further includes: The historical parcel volume, historical time period type, and historical weather of the test site corresponding to the parcel under test are obtained as historical multidimensional features. The historical multidimensional features are input into the total number of express shipments prediction model. Based on the self-attention layer of the total number of express shipments prediction model, the first attention weight and first coding feature of historical time period type on the historical number of express shipments, and the second attention weight and second coding feature of historical weather on the historical number of express shipments are extracted. Historical coding features are determined based on the first attention weight, the first coding feature, the second attention weight, and the second coding feature; The decoder based on the total parcel volume prediction model decodes the historical encoding features to obtain the predicted total parcel volume of the test site in the future period.

7. The method according to any one of claims 1-6, characterized in that, The method further includes: Obtain the network attributes, three-segment code, and salesperson attributes of the network corresponding to the parcel to be tested; A graph structure is constructed based on the branch attributes, three-segment codes, and salesperson attributes; in the graph structure, nodes represent the branch attributes, three-segment codes, and salesperson attributes, and edges represent the association between two connected nodes. The weight of each edge is determined based on the strength of the association between the two nodes connected by each edge in the graph structure. Based on the first convolutional layer of the graph convolutional network and the weight of each edge, the features of each node are weighted and fused with the features of its direct neighbors to obtain local association features. Based on the second convolutional layer, the features of the neighbor nodes corresponding to the direct neighbor nodes are fused into the local association features to capture higher-order association relationships, determine the similarity between each node and the indirect neighbor nodes, and filter out the recommended associated nodes for each node.

8. An intelligent identification device for missing item scanning operation in express mail, characterized in that, The device includes: The test data acquisition module is used to acquire the current operation code, current time, and current branch type of the parcel under test, and construct the test multidimensional data of the parcel under test; The current attribution status prediction module is used to input the multidimensional data to be tested into the attribution status prediction model and output at least two current attribution statuses of the parcel to be tested and the current prediction probability corresponding to each current attribution status. The result generation module is used to generate the tracking result of the package under test based on at least two current attribution states corresponding to the package under test and the current predicted probability corresponding to each current attribution state. The current sequence determination module is used to determine the current time sequence of the package under test in multiple time windows based on the tracking results of the package under test. The scan result prediction module is used to input the current time sequence and real-time operation sequence label of the parcel to be tested into the missed scan identification model, and output the current scan operation result of the parcel to be tested; the current scan operation result indicates whether there is a missed scan operation in the parcel to be tested.

9. An electronic device, characterized in that, The device includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the intelligent identification method for express mail missing scan operation as described in any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer storage medium stores at least one instruction or at least one program, which is loaded and executed by a processor to implement the intelligent identification method for express mail missing scan operation as described in any one of claims 1-7.

Images