Logistics full-link aging analysis method, apparatus and device, and storage medium

Abstract

The invention relates to the technical field of logistics, and discloses a logistics full-link aging analysis method and device, equipment and a storage medium. The method comprises the following steps: acquiring target dynamic logistics data of a plurality of links and target static logistics data of a plurality of links in full-link logistics; performing dynamic and static feature splicing on the target static logistics data and the target dynamic logistics data to obtain a logistics target feature matrix; extracting and mixing long-term trend features and short-term fluctuation features of the logistics target feature matrix to obtain comprehensive features, and performing feature weighted fusion on the comprehensive features based on target weight distribution to obtain weight matrix features of each link; based on a full connection layer, the weight matrix features are mapped into the predicted transportation time efficiency of each link; obtaining the actual transportation time efficiency of each link; and determining an aging deviation result of the logistics whole link based on the actual transportation aging of the plurality of links and the respective corresponding predicted transportation aging. According to the scheme, the logistics aging stability and the operation efficiency can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics, and in particular to a logistics full-link time limit analysis method, device, equipment and storage medium. BACKGROUND

[0002] With the vigorous development of e-commerce and Internet shopping, users' requirements for the timeliness of express logistics services are increasingly stringent, and the whole transportation time limit has become a key indicator for measuring the quality and core competitiveness of logistics services. In order to ensure the high timeliness of the whole transportation link, modern express logistics enterprises generally adopt a fine control mode for each link, and monitor and evaluate the operation process of goods from collection, sorting, transfer to delivery and other links.

[0003] However, the existing logistics timeliness monitoring and analysis scheme still has many deficiencies. First, the traditional monitoring method relies on a single data source, such as only relying on PDA (portable data terminal) scanning data or vehicle GPS positioning data, resulting in incomplete information chain and easy data blind spots, for example, the vehicle has arrived at the distribution center but has not been unloaded and scanned in time, and the delay in this period is difficult to accurately capture. Second, the existing analysis method is mostly post-statistics, lacking forward-looking prediction ability, and when the delay occurs, manual investigation of the cause is time-consuming and labor-intensive, with low processing efficiency, and it is difficult to actively intervene and optimize. Third, the definition of delay responsibility often relies on manual experience, lacking objective and quantitative attribution analysis tools, which is not conducive to fine management and continuous optimization. Finally, the logistics system is a complex dynamic system, which is affected by multiple factors such as weather, traffic, holidays, and cargo attributes, and the existing model is difficult to comprehensively fuse and analyze these heterogeneous data, resulting in limited prediction accuracy and analysis depth.

[0004] Therefore, how to integrate multi-source heterogeneous data and build a logistics full-link timeliness analysis system that can realize full-process visual monitoring, intelligent abnormal early warning, accurate timeliness prediction and dynamic decision optimization has become a technical problem to be solved in the industry. SUMMARY

[0005] The present application provides a logistics full-link timeliness analysis method, device, equipment and storage medium, which aims to solve the problems of single data source, lack of prediction, lagging abnormal discovery and insufficient decision support in the prior art.

[0006] According to one aspect of the present application, a logistics full-link timeliness analysis method is disclosed, the method comprising: obtaining target dynamic logistics data of a plurality of links in a full-link logistics and target static logistics data of the plurality of links, wherein the target dynamic logistics data comprises target dynamic timeliness data and target dynamic environment data; concatenate the target static logistics data and the target dynamic logistics data to obtain a logistics target feature matrix; extract and mix long-term trend features and short-term fluctuation features of the logistics target feature matrix to obtain a comprehensive feature, the long-term trend features being used to represent cross-period persistence rules, and the short-term fluctuation features being used to represent mutation fluctuations of a target time length; based on target weight distribution, feature-weighted fusion is performed on the comprehensive feature to obtain a weight matrix feature of each link; based on a full connection layer, the weight matrix feature is mapped to a predicted transportation time limit of each link, wherein the number of neurons of the full connection layer is the same as the number of links of the full-link logistics, and each neuron outputs a predicted transportation time limit corresponding to a link; an actual transportation time limit of each link is obtained; based on the actual transportation time limits of multiple links and the respective predicted transportation time limits, a time limit deviation result of the full-link logistics is determined.

[0007] In some embodiments, the obtaining of the dynamic logistics data of multiple links and the static logistics data of multiple links in the full-link logistics includes: initial dynamic logistics data of multiple links and initial static logistics data of multiple links in the full-link logistics are obtained; data processing is performed on the initial dynamic logistics data and the initial static logistics data to obtain corresponding target dynamic logistics data and corresponding target static logistics data.

[0008] In some embodiments, the concatenating of the target static logistics data and the target dynamic logistics data to obtain a logistics target feature matrix includes: the target static logistics data is integrated into a one-dimensional static vector; the target dynamic logistics data is integrated into a two-dimensional time sequence feature matrix, wherein the two-dimensional time sequence feature matrix is determined based on the number of links and the dynamic logistics data corresponding to each link; the one-dimensional static vector is aligned with the time dimension of the two-dimensional time sequence feature matrix to expand the one-dimensional static vector into a two-dimensional static feature expansion matrix; the two-dimensional static feature expansion matrix and the two-dimensional time sequence feature matrix are column-spliced to obtain the logistics target feature matrix.

[0009] In some embodiments, the extracting and mixing of the long-term trend features and the short-term fluctuation features of the logistics target feature matrix to obtain a comprehensive feature includes: The long-term trend features and short-term fluctuation features of the logistics target feature matrix are extracted and then mixed based on a hybrid network. The hybrid network is a combination of a long short-term memory network and a gated recurrent unit. The long short-term memory network is used to extract the long-term trend features, and the gated recurrent unit is used to extract the short-term fluctuation features.

[0010] In some embodiments, the step of performing feature weighting and fusion on the comprehensive features based on the target weight allocation to obtain the weight matrix features of each stage includes: Determine the impact of each link on the entire logistics chain; Target weights are assigned to each stage according to the degree of impact, with higher impact corresponding to higher target weights; Based on the target weight of each stage, the comprehensive features are weighted and fused to obtain the weight matrix features of each stage.

[0011] In some embodiments, determining the timeliness deviation of the entire logistics chain based on the actual transportation timeliness of multiple links and their corresponding predicted transportation timeliness includes: Based on the actual transportation timeliness of each link and its corresponding predicted transportation timeliness, the timeliness deviation of each link is determined; The link timeliness deviation is obtained by summing up the timeliness deviations of multiple links; Obtain the actual timeliness and allowable deviation threshold of the link; Based on the timeliness deviation of the link, the actual timeliness of the link, and the allowable deviation threshold, the timeliness deviation of the entire logistics link is determined.

[0012] In some embodiments, the logistics end-to-end timeliness analysis method further includes: Obtain the deviation difference between the predicted timeliness and the actual timeliness of the target stage; Obtain the deviation update threshold; When the deviation difference meets the deviation update threshold, the delay data that caused the delay in the target process is obtained; The initial dynamic logistics data is updated based on the delay data.

[0013] According to another aspect of this application, a logistics end-to-end timeliness analysis device is also disclosed, characterized in that the device comprises: The logistics data acquisition module is used to acquire target dynamic logistics data and target static logistics data for multiple links in the whole-chain logistics. The target dynamic logistics data includes target dynamic timeliness data and target dynamic environmental data. The logistics target feature matrix determination module is used to perform dynamic and static feature concatenation on the target static logistics data and the target dynamic logistics data to obtain the logistics target feature matrix. The comprehensive feature determination module is used to extract and mix the long-term trend features and short-term fluctuation features of the logistics target feature matrix to obtain comprehensive features. The long-term trend features are used to characterize cross-cycle persistence patterns, and the short-term fluctuation features are used to characterize abrupt fluctuations in the target duration. The weight matrix feature determination module is used to perform feature weighting and fusion on the comprehensive features based on the target weight allocation to obtain the weight matrix features of each step; The predicted transportation timeliness determination module is used to map the features of the weight matrix to the predicted transportation timeliness of each link based on the fully connected layer. The number of neurons in the fully connected layer is the same as the number of links in the whole-link logistics division, and each neuron outputs the predicted transportation timeliness corresponding to one link. The actual transportation timeliness acquisition module is used to acquire the actual transportation timeliness of each link; The timeliness deviation determination module is used to determine the timeliness deviation results of the entire logistics chain based on the actual transportation timeliness of multiple links and their corresponding predicted transportation timeliness.

[0014] Optionally, the logistics data acquisition module includes: an initial logistics data acquisition unit, used to acquire initial dynamic logistics data and initial static logistics data of multiple links in the entire logistics chain; and a data processing unit, used to process the initial dynamic logistics data and the initial static logistics data to obtain corresponding target dynamic logistics data and corresponding target static logistics data.

[0015] Optionally, in this embodiment, the logistics target feature matrix determination module includes: a static vector integration unit, used to integrate the target static logistics data into a one-dimensional static vector; a dynamic matrix integration unit, used to integrate the target dynamic logistics data into a two-dimensional time-series feature matrix, wherein the two-dimensional time-series feature matrix is ​​determined based on the number of links and the dynamic logistics data corresponding to the number of links; a static matrix expansion unit, used to align the one-dimensional static vector with the time dimension of the two-dimensional time-series feature matrix to expand the one-dimensional static vector into a two-dimensional static feature expansion matrix; and a matrix concatenation unit, used to concatenate the two-dimensional static feature expansion matrix and the two-dimensional time-series feature matrix column by column to obtain the logistics target feature matrix.

[0016] Optionally, in this embodiment, the comprehensive feature determination module includes a hybrid network processing unit, which is used to extract the long-term trend features and short-term fluctuation features of the logistics target feature matrix based on the hybrid network and then mix them. The hybrid network is a combination of a long short-term memory network and a gated recurrent unit. The long short-term memory network is used to extract the long-term trend features, and the gated recurrent unit is used to extract the short-term fluctuation features.

[0017] Optionally, in this embodiment, the weight matrix feature determination module includes: an influence degree determination unit, used to determine the influence degree of each link on the entire logistics chain; a weight allocation unit, used to allocate target weights to each link according to the degree of influence, wherein a higher degree of influence corresponds to a higher target weight; and a weighted fusion unit, used to perform feature weighted fusion on the comprehensive features based on the target weight of each link to obtain the weight matrix features of each link.

[0018] Optionally, in this embodiment, the timeliness deviation result determination module includes: a link deviation determination unit, used to determine the link timeliness deviation of each link based on the actual transportation timeliness of each link and its corresponding predicted transportation timeliness; a link deviation accumulation unit, used to accumulate the link timeliness deviations of multiple links to obtain the link timeliness deviation; a threshold acquisition unit, used to acquire the actual timeliness of the link and the allowable deviation threshold; and a full-link deviation determination unit, used to determine the timeliness deviation of the entire logistics link based on the link timeliness deviation, the actual timeliness of the link, and the allowable deviation threshold.

[0019] Optionally, in this embodiment, the device further includes: a deviation difference acquisition module, used to acquire the deviation difference between the target predicted timeliness and the target actual timeliness of the target link; an update threshold acquisition module, used to acquire a deviation update threshold; a delay data acquisition module, used to acquire delay data that causes the delay of the target link when the deviation difference meets the deviation update threshold; and a data update module, used to update the initial dynamic logistics data based on the delay data.

[0020] According to another aspect of this application, an electronic device is also disclosed, the electronic device including a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the electronic device to perform various steps of the logistics end-to-end timeliness analysis method as described in any of the preceding claims.

[0021] According to another aspect of this application, a computer-readable storage medium is also disclosed, on which instructions are stored, which, when executed by a processor, implement the various steps of the logistics end-to-end timeliness analysis method as described in any of the preceding claims.

[0022] The present invention includes, but is not limited to, the following beneficial effects: (1) By integrating multi-source heterogeneous data such as manual order entry, PDA scanning and GPS positioning, a complete cargo transportation trajectory is constructed, and the transportation process is divided into multiple clear physical links for independent analysis, eliminating the information blind spots in traditional logistics management and realizing end-to-end full-link visual monitoring; (2) By adopting an LSTM-GRU hybrid network combined with an attention mechanism, it can simultaneously capture long-term trends (such as periodic congestion) and short-term sudden fluctuations (such as sudden traffic accidents) in logistics timeliness data, and dynamically identify key nodes affecting timeliness, thereby improving the predictability of transportation timeliness in each link. (3) By measuring accuracy, it can provide early warning of potential delay risks and change passive response to proactive management; (4) By allocating link weights and analyzing deviations, it can locate specific delay links and corresponding responsible entities (such as distribution centers and outlets), quantify the responsibility ratio of each link, provide clear basis for performance evaluation and problem rectification, and promote the refined management of logistics operations; (5) This solution can dynamically adapt to complex scenarios such as sudden weather and traffic congestion, solve the problem of missing scan data through data compensation mechanism, reduce computing costs while ensuring performance, is suitable for deployment on edge devices or cloud services, takes into account practicality and economy, and can effectively improve the stability of logistics timeliness and operational efficiency. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0024] Figure 1 This is a flowchart of a logistics end-to-end timeliness analysis method according to an embodiment of this application; Figure 2 This is another flowchart of the logistics end-to-end timeliness analysis method according to the embodiments of this application; Figure 3 This is another flowchart of the logistics end-to-end timeliness analysis method according to the embodiments of this application; Figure 4 This is another flowchart of the logistics end-to-end timeliness analysis method according to the embodiments of this application; Figure 5 This is another flowchart of the logistics end-to-end timeliness analysis method according to the embodiments of this application; Figure 6 This is another flowchart of the logistics end-to-end timeliness analysis method according to the embodiments of this application; Figure 7 This is a structural block diagram of the logistics end-to-end timeliness analysis device according to an embodiment of this application; Figure 8 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0025] This invention provides a method, apparatus, device, and storage medium for end-to-end logistics timeliness analysis. The method includes acquiring target dynamic logistics data and target static logistics data for multiple links in the end-to-end logistics process. The target dynamic logistics data includes target dynamic timeliness data and target dynamic environment data. The method further involves concatenating dynamic and static features of the target static and dynamic logistics data to obtain a logistics target feature matrix. It also involves extracting and merging long-term trend features and short-term fluctuation features from the logistics target feature matrix to obtain comprehensive features. The long-term trend features characterize cross-cycle persistence, while the short-term fluctuation features characterize abrupt fluctuations in target duration. Based on target weight allocation, the comprehensive features are weighted and fused to obtain a weight matrix feature for each link. A fully connected layer maps the weight matrix features to the predicted transportation timeliness of each link, where the number of neurons in the fully connected layer is the same as the number of links in the end-to-end logistics process, and each neuron outputs the predicted transportation timeliness for one link. The method also involves acquiring the actual transportation timeliness of each link and determining the timeliness deviation result for the entire logistics chain based on the actual transportation timeliness of multiple links and their corresponding predicted transportation timeliness. This solution achieves multi-source data fusion and refined monitoring of the entire process, accurately predicting delivery time, locating delay links and responsible parties, dynamically adapting to complex logistics scenarios, providing a scientific basis for logistics scheduling and resource allocation, and effectively improving the stability and efficiency of logistics delivery time.

[0026] The terms "first," "second," "third," "fourth," etc. (if present) 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 described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus 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 apparatus.

[0027] For ease of understanding, the specific process of the embodiments of the present invention will be described below. Figure 1 A flowchart for a logistics end-to-end timeliness analysis method, such as Figure 1 As shown, it includes the following steps: S100. Obtain target dynamic logistics data and target static logistics data for multiple links in the whole-chain logistics. Among them, the target dynamic logistics data includes target dynamic timeliness data and target dynamic environment data.

[0028] Understandably, end-to-end logistics encompasses numerous stages, including pickup, transportation, sorting, and delivery. Target static logistics data consists of relatively stable information that doesn't easily change over time, such as the weight, volume, category, fixed routes involved in transportation, and the infrastructure configuration of each logistics node. This data provides a basic attribute framework for subsequent analysis of goods and the logistics network. Target dynamic logistics data, on the other hand, focuses on information that changes in real time. Specifically, target dynamic timeliness data includes the actual time spent at each stage, such as the actual transportation time from the pickup point to the regional distribution center, and the actual time spent sorting goods within the distribution center. Target dynamic environmental data involves real-time traffic conditions (such as whether there is congestion on transportation routes or road construction), weather conditions (whether there is rain, snow, or extreme weather such as typhoons), and the real-time load of logistics nodes (such as the number of goods currently awaiting sorting at the distribution center and the backlog of parcels awaiting delivery at sales outlets). By comprehensively acquiring these dynamic and static logistics data, we can provide rich and accurate raw data for subsequent steps such as splicing dynamic and static features and extracting long-term trends and short-term fluctuations. This ensures that the subsequent analysis of the timeliness of the entire logistics chain is based on a solid and comprehensive data foundation, thereby more accurately grasping the timeliness of each link and the timeliness performance of the entire chain.

[0029] Furthermore, Figure 2 This is another flowchart of the logistics end-to-end timeliness analysis method according to an embodiment of this application. This flowchart is an exemplary illustration of step S100, obtaining target dynamic logistics data and target static logistics data for multiple stages in the end-to-end logistics process. Figure 2 As shown, it includes the following steps: S200: Obtain initial dynamic logistics data and initial static logistics data for multiple links in the entire logistics chain.

[0030] Specifically, staff at express delivery outlets input basic cargo information using terminal devices (such as PDAs) or PCs. This basic information includes, but is not limited to, cargo type, weight, originating outlet, destination outlet, and entry time. This provides static baseline information for subsequent timeliness analysis (such as prioritizing different cargo types) and liability attribution (such as associating originating and destination outlets). Furthermore, PDAs are used to scan cargo barcodes / QR codes, collecting dynamic status data at each operational node. These nodes include delivery scanning, unloading scanning, loading scanning, and receipt scanning. GPS positioning devices are then installed on transport vehicles to collect real-time vehicle dynamic data, indirectly linking it to the cargo transportation status. Specifically, each vehicle carrying logistics goods is equipped with a GPS positioning device that supports geofence triggering function. Data is uploaded by timestamp (e.g., updated every 30 seconds), including vehicle entry time (automatically triggered by the system when the vehicle enters the geofence area of ​​the distribution center / outlet, accurate to the second), vehicle exit time (automatically recorded when the vehicle leaves the geofence area), real-time trajectory of the transportation route (latitude and longitude coordinate sequence), driving speed (unit: km / h), etc., to provide data support for route complexity analysis (such as detour coefficient calculation) and transportation timeliness deviation (such as vehicle delay in entering the station).

[0031] S202. Process the initial dynamic logistics data and initial static logistics data to obtain the corresponding target dynamic logistics data and the corresponding target static logistics data.

[0032] Understandably, manual order entry may result in inconsistent formats, necessitating standardization of the initial static logistics data to obtain the target static logistics data. First, the order entry data is standardized according to preset rules. Second, the system automatically verifies the matching relationship between the "sender point and destination point" (e.g., excluding invalid data where "sender point = destination point") and the completeness of the order entry time (e.g., supplementing missing entry times with the current operation time). Finally, the target static logistics data is output, which can be directly linked to the unique identification code (barcode / QR code) of the goods, flowing throughout the entire supply chain. Additionally, the initial dynamic logistics data suffers from issues such as "multi-source heterogeneity (inconsistent timestamps between PDA scan data and GPS data), redundancy and duplication (multiple scans at the same stage), and data gaps (missed scans leading to blank data at each stage)." These issues require processing to obtain the target dynamic logistics data. Specifically, the first step involves using ETL (Extract, Transform, Load) tools to extract, transform, and load the initial dynamic data, reconstructing the entire supply chain according to time sequence. First, all dynamic data for the same goods (associated through a unique identifier) ​​are extracted from the PDA scanning database and GPS positioning database. Second, the scattered data is mapped into 12 standardized steps according to transportation logic, sorted in ascending order by timestamp to form a complete time-series chain. The standardized step sequence is: Order entry → Outbound scanning → Originating distribution center entry → Originating distribution center unloading → Originating distribution center loading → Transit distribution center entry → Transit distribution center unloading → Transit distribution center loading → Destination distribution center entry → Destination distribution center unloading → Destination distribution center loading → Final receipt. Finally, the sorted time-series data is loaded into the full-link timeliness analysis database, with each step corresponding to a structured format of "step name - timestamp - associated data (e.g., distribution center code, vehicle number)". A time window clustering algorithm is used to merge duplicate scan records at the same node (only the first valid data is retained from multiple unloading scans). The second step: For multiple scans at the same operation node (e.g., the same goods are scanned 3 times during distribution center unloading), a "time window clustering algorithm" is used to remove duplicates. Specifically, firstly, for the same operational step of the same goods (such as unloading scanning), a time window with a certain time threshold is set (multiple scans within 30 minutes of the same step are considered redundant). Secondly, duplicate scan records within the time window are clustered, and only the first valid scan data is retained (e.g., if three unloading scans are performed at 10:05, 10:08, and 10:12, only the record at 10:05 is retained). Finally, the merged redundant data is marked with a "redundancy flag" to facilitate subsequent data traceability and avoid accidental deletion of valid data. Thirdly, for data loss caused by "missed scans" in the initial dynamic data (e.g., the vehicle has entered the station but there is no unloading scan record), GPS data is used for compensation.Specifically, the system first automatically detects gaps in the time-series data. For example, if there is a GPS record for "entering the distribution center" (vehicle enters at 10:00), but no "unloading" scan data is detected within 2 hours, it is determined that the unloading scan data is missing. Secondly, based on the GPS-recorded "vehicle entry time," the system adds the "preset operation time" of the corresponding link in the distribution center (e.g., the preset unloading time for the originating distribution center is 1.5 hours) to estimate the unloading time (10:00 + 1.5 hours = 11:30). Finally, the estimated unloading time and compensation flag ("GPS compensation") are entered into the time-series data to fill in the missing links and ensure the integrity of the target dynamic logistics data chain. After the data cleaning and processing of the above steps, the corresponding target dynamic logistics data and the corresponding target static logistics data are obtained.

[0033] S102. Perform dynamic and static feature splicing on the target static logistics data and the target dynamic logistics data to obtain the logistics target feature matrix.

[0034] Furthermore, Figure 3 This is another flowchart of the logistics end-to-end timeliness analysis method according to an embodiment of this application. This flowchart is an exemplary illustration of step S102, which involves splicing static and dynamic features of the target static logistics data and the target dynamic logistics data to obtain a logistics target feature matrix. Figure 3 As shown, it includes the following steps: S300: Integrate the target static logistics data into a one-dimensional static vector.

[0035] Specifically, target static logistics data is selected, including core static attributes such as cargo type, cargo weight, cargo volume, originating point code, and destination point code. For categorical attributes (cargo type, originating point code, and destination point code), One-Hot encoding is used. For example, if there are three cargo types: "fresh produce," "fragile goods," and "general goods," and a certain cargo is fresh produce, its One-Hot code is [1,0,0]. For numerical attributes (cargo weight and cargo volume), standardization is performed to eliminate the influence of dimensions. All the coded and standardized static attribute data are then concatenated in sequence to form a one-dimensional static vector. For example, if the cargo type code is [1,0,0], the standardized weight is 0.5, the standardized volume is 0.3, the originating point code is [1,0], and the destination point code is [0,1], then the one-dimensional static vector is [1,0,0,0.5,0.3,1,0,0,1].

[0036] S302. Integrate the target dynamic logistics data into a two-dimensional time-series feature matrix, wherein the two-dimensional time-series feature matrix is ​​determined based on the number of links and the dynamic logistics data corresponding to the number of links.

[0037] Specifically, first, determine the number of links in the entire logistics chain, such as from delivery at the point of sale, unloading at the distribution center, loading at the distribution center, to final delivery and signature. Assuming there are n links, for each link, extract the corresponding dynamic data from the target dynamic logistics data, such as the operation time (accurate to the second), the equipment number, and the operator number for each link. Then, using the number of links n as the number of rows in the matrix, for each link, arrange all its corresponding dynamic data (such as operation time, equipment number, and operator number, after appropriate encoding, such as numerical mapping or One-Hot encoding for equipment number and operator number) by column to determine the number of columns in the matrix. For example, if each link has m dynamic features, the matrix dimension is n×m, forming a two-dimensional time-series feature matrix.

[0038] S304. Align the one-dimensional static vector with the time dimension of the two-dimensional time-series feature matrix to expand the one-dimensional static vector into a two-dimensional static feature extension matrix.

[0039] Specifically, the process involves obtaining the timestamp information corresponding to each stage in the two-dimensional time-series feature matrix to clarify the distribution of the time dimension, such as the intervals and order of the stages. It's understandable that since the one-dimensional static vector itself lacks a time dimension (static data doesn't change over time), it needs to be expanded according to the time dimension of the two-dimensional time-series feature matrix to align with it. For example, if the two-dimensional time-series feature matrix has n time-related stages, the one-dimensional static vector is copied n times to form a matrix with n rows (the same as the number of stages) and the same number of columns as the one-dimensional static vector. This resulting matrix is ​​the two-dimensional static feature expansion matrix, where each row contains complete static features, and the number of rows is the same as the number of rows (stages) in the two-dimensional time-series feature matrix, achieving alignment of the time dimension (represented by the number of stages).

[0040] S306. Concatenate the columns of the two-dimensional static feature extension matrix and the two-dimensional time-series feature matrix to obtain the logistics target feature matrix.

[0041] Specifically, check if the number of rows in the two-dimensional static feature extension matrix and the two-dimensional time-series feature matrix are the same (both are the number of links n). Then, concatenate the two-dimensional static feature extension matrix and the two-dimensional time-series feature matrix in the column direction. For example, if the two-dimensional static feature extension matrix has a dimension of n×p (p is the number of static features) and the two-dimensional time-series feature matrix has a dimension of n×m (m is the number of dynamic features), the concatenated logistics target feature matrix will have a dimension of n×(p+m). This matrix integrates static and dynamic logistics features and can be used for subsequent analysis and calculation.

[0042] S104. Extract and combine the long-term trend features and short-term fluctuation features of the mixed flow target feature matrix to obtain comprehensive features. The long-term trend features are used to characterize the cross-cycle persistence pattern, and the short-term fluctuation features are used to characterize the sudden fluctuations of the target duration.

[0043] Understandably, in this embodiment, the long-term trend features and short-term fluctuation features of the logistics target feature matrix are extracted separately from the hybrid network and then combined. The hybrid network is a combination of a Long Short-Term Memory (LSTM) network and a gated recurrent unit (GRU). The LSTM network is used to extract long-term trend features, and the GRU is used to extract short-term fluctuation features. Specifically, when using the LSTM network to extract long-term trend features, an LSTM network layer is first constructed, with appropriate hidden layer neurons and time steps set. Then, the logistics target feature matrix is ​​input into the LSTM network in a time sequence (the order of each logistics link). Through its unique cell state and gating mechanism, the LSTM network can effectively capture continuous patterns across cycles. For example, for the end-to-end timeliness data of different batches of the same type of goods from dispatch to receipt, the LSTM network can learn the long-term timeliness trend of this type of goods. After forward propagation calculation by the LSTM network, the extracted long-term trend feature vector is output. When using the GRU to extract short-term fluctuation features, a GRU network layer is first constructed, with appropriate hidden layer neurons, time steps, and other parameters set. Then, the logistics target feature matrix is ​​input into the GRU network in the same time sequence. Compared to LSTM networks, GRU networks have a simpler structure, higher computational efficiency, and can effectively capture short-term abrupt changes in fluctuations. For example, a sudden extension in the unloading process of a batch of goods at a distribution center due to equipment failure, or a sudden change in the delivery time of a certain transportation link due to weather conditions, can be quickly captured by the GRU network. The GRU network extracts short-term change information from the input sequence through update and reset gates, outputting a short-term fluctuation feature vector. Further, the dimensions of the long-term trend feature vector output by LSTM and the short-term fluctuation feature vector output by GRU are adjusted to ensure consistency. Then, feature concatenation is used to concatenate the long-term trend feature vector and the short-term fluctuation feature vector in the column direction to obtain a longer comprehensive feature vector. Alternatively, a weighted summation method can be used, assigning different weights to the long-term trend features and short-term fluctuation features according to business needs (e.g., prioritizing long-term trends or short-term fluctuations) (e.g., long-term trend weight 0.6, short-term fluctuation weight 0.4), and then performing a weighted summation to obtain the comprehensive feature. Understandably, through the above steps, the long-term trend features and short-term fluctuation features of the logistics target feature matrix are extracted and mixed. The resulting comprehensive features include both cross-cycle continuous patterns and short-term sudden fluctuation information, providing strong feature support for subsequent analysis such as timeliness monitoring and anomaly detection in the entire logistics chain.

[0044] S106. Based on the target weight allocation, perform feature weighting and fusion on the comprehensive features to obtain the weight matrix features of each link.

[0045] Furthermore, Figure 4 This is another flowchart of the logistics end-to-end timeliness analysis method according to an embodiment of this application. This flowchart is an exemplary illustration of step S106, which involves weighted fusion of comprehensive features based on target weight allocation to obtain the weight matrix features of each link. Figure 4 As shown, it includes the following steps: S400: Determine the impact of each link on the entire logistics chain.

[0046] Specifically, historical data is collected for each stage of the entire logistics chain (such as arrival at the distribution center, unloading at the distribution center, loading at the distribution center, and last-mile delivery). This includes the operation time of each stage, the number of anomalies (such as delays and cargo damage), and the corresponding overall logistics timeliness (total time from dispatch to receipt). Statistical analysis or machine learning methods are used, with overall logistics timeliness as the dependent variable and the operation-related data of each stage as independent variables, to construct regression models (such as multiple linear regression) or use feature importance assessment algorithms (such as feature importance calculation using random forests). For example, by training a random forest model on historical data, the model will output an importance score for each stage's independent variable to the overall logistics timeliness dependent variable. This score reflects the degree of influence of each stage on the overall logistics chain.

[0047] S402. Assign target weights to each link according to the degree of influence, where a higher degree of influence corresponds to a higher target weight.

[0048] Specifically, the comprehensive features obtained after hybrid network processing are input into the Attention mechanism module. For the time step features corresponding to each logistics link, the core calculation logic of the Attention mechanism is used to determine their impact on the overall timeliness. Then, target weights are assigned to each link according to the degree of impact; that is, links with a high degree of impact are assigned high target weights, and links with a low degree of impact are assigned low target weights.

[0049] S404. Based on the target weight of each stage, perform feature weighting and fusion on the comprehensive features to obtain the weight matrix features of each stage.

[0050] It is understandable that the comprehensive features here refer to the feature set obtained after preliminary steps (such as S104 extracting and mixing long-term trend features and short-term fluctuation features), which contains information on the long-term trend and short-term fluctuation of the logistics target feature matrix. Specifically, the comprehensive features are broken down according to the logistics links, resulting in feature sub-matrices corresponding to each link. For example, if there are n links in the entire chain, the comprehensive feature matrix can be broken down into n feature sub-matrices, each corresponding to the features of one link. For each link's feature sub-matrix, it is multiplied by the target weight assigned to that link in S402, and then the weighted feature sub-matrices of all links are merged (e.g., by matrix addition or concatenation, the specific fusion method can be determined according to business needs and data structure). Finally, the weight matrix features of each link are obtained, where the features of high-impact links have a higher proportion in the weight matrix features, better reflecting their impact on the entire logistics chain, and facilitating the subsequent location of key links through methods such as Attention weight visualization.

[0051] S108. Based on the fully connected layer, the weight matrix features are mapped to the predicted transportation time of each link. The number of neurons in the fully connected layer is the same as the number of links in the whole-link logistics division. Each neuron outputs the predicted transportation time of a link.

[0052] Specifically, the weight matrix features of each stage are used as input data, and a fully connected layer is constructed. The number of neurons in the fully connected layer is set to be the same as the number of stages in the entire logistics chain. For example, if the entire logistics chain is divided into n stages (such as delivery at the point of sale, unloading at the distribution center, loading at the distribution center, and final receipt, etc.), then the fully connected layer has n neurons. The weight matrix features are input into the constructed fully connected layer. The fully connected layer processes the input weight matrix features through linear transformations (each neuron is fully connected to the input weight matrix features, i.e., each neuron receives all the input information of the weight matrix features) and activation functions (such as ReLU, Sigmoid, etc., which can be selected according to business needs to introduce nonlinearity and enhance the model's expressive power). Each neuron outputs the predicted transportation time of one stage. Since the number of neurons in the fully connected layer is the same as the number of stages, and each neuron focuses on processing the feature information of its corresponding stage, the fully connected layer can output the predicted transportation time of each stage after calculation, realizing the mapping from the weight matrix features to the predicted transportation time of each stage.

[0053] Understandably, by leveraging the fully connected layer, the weighted features of the multi-dimensional feature matrix are transformed into specific predicted transportation timeliness for each stage, providing quantitative prediction results for timeliness monitoring and analysis of the entire logistics chain.

[0054] S110. Obtain the actual transportation time for each stage.

[0055] S112. Based on the actual transportation timeliness of multiple links and their corresponding predicted transportation timeliness, determine the timeliness deviation results of the entire logistics chain.

[0056] Furthermore, Figure 5 This is another flowchart of the logistics end-to-end timeliness analysis method according to an embodiment of this application. This flowchart is an exemplary illustration of step S112, which involves determining the timeliness deviation results of the entire logistics chain based on the actual transportation timeliness of multiple links and their corresponding predicted transportation timeliness. Figure 5 As shown, it includes the following steps: S500. Based on the actual transportation timeliness of each link and its corresponding predicted transportation timeliness, determine the link timeliness deviation of each link.

[0057] Specifically, from the actual operational data of the entire logistics chain, the actual completion time of each link (such as delivery to the point of sale, unloading at the distribution center, loading at the distribution center, and final receipt) is extracted, and the actual transportation timeliness of each link (i.e., the time spent from the start to the end of the link) is calculated. The predicted transportation timeliness of each link is obtained based on the output of the fully connected layer. For each link, the actual transportation timeliness is subtracted from its corresponding predicted transportation timeliness to obtain the link's timeliness deviation. If the result is positive, it indicates that the actual timeliness is longer than the predicted timeliness, indicating a delay; if the result is negative, it indicates that the actual timeliness is shorter than the predicted timeliness, indicating early completion.

[0058] S502. The timeliness deviations of multiple links are accumulated to obtain the link timeliness deviation.

[0059] S504. Obtain the actual timeliness and allowable deviation threshold of the link.

[0060] Specifically, the total actual time from dispatch to receipt is calculated from the actual operational data of the entire logistics chain, i.e., the actual timeliness of the chain. Based on the service standards of the logistics business, historical experience, or customer requirements, the allowable timeliness deviation threshold for the entire chain is preset. For example, the allowable timeliness deviation for the entire chain is within ±T (T is the set time value, such as 2 hours).

[0061] S506. Based on the timeliness deviation of the link, the actual timeliness of the link, and the allowable deviation threshold, determine the timeliness deviation of the entire logistics link.

[0062] Specifically, the link timeliness deviation and actual link timeliness are combined and compared with the allowable deviation threshold to determine whether the link timeliness deviation is within the allowable deviation threshold range. If the absolute value of the link timeliness deviation is less than or equal to the allowable deviation threshold, it indicates that the overall link timeliness deviation is within an acceptable range; if the absolute value of the link timeliness deviation is greater than the allowable deviation threshold, it indicates that the overall link timeliness deviation exceeds the acceptable range, and further anomaly analysis and processing are required.

[0063] Furthermore, Figure 6 This is another flowchart of the logistics end-to-end timeliness analysis method according to the embodiments of this application, such as... Figure 6 As shown, it includes the following steps: S600, Obtain the deviation difference between the predicted timeliness and the actual timeliness of the target in the target stage.

[0064] Specifically, based on the monitoring needs of the entire logistics chain, key links requiring focused attention are selected as target links, such as unloading at the distribution center and last-mile delivery. Then, based on the prediction results output by the fully connected layer, the predicted transportation time corresponding to the target link is extracted. From the actual operational data of the entire logistics chain, the actual completion time of the target link is extracted, and the target actual time (i.e., the actual time spent from start to finish of the link) is calculated. The target predicted time is subtracted from the target actual time to obtain the deviation difference. If the result is positive, it indicates that the actual time of the target link is longer than the predicted time, resulting in a delay; if it is negative, it indicates that the actual time is shorter than the predicted time, resulting in early completion.

[0065] S602, Obtain the deviation update threshold.

[0066] Specifically, based on factors such as the accuracy requirements of logistics operations and historical delay data, a deviation update threshold is pre-set. For example, it is set that when the deviation difference is greater than or equal to T (T is a set time value, such as 1 hour), subsequent delay data acquisition and data update operations are required.

[0067] S604. When the deviation difference meets the deviation update threshold, obtain the delay data that caused the delay in the target process.

[0068] Specifically, the calculated deviation difference is compared with the obtained deviation update threshold. If the deviation difference is greater than or equal to the deviation update threshold, it indicates that the delay in the target process has reached the standard requiring data updates. When the threshold condition is met, relevant data causing the delay in the target process is collected through channels such as the logistics system's monitoring logs, equipment records, and personnel operation records. For example, if the target process is the unloading process at the distribution center, the delay data may include the unloading equipment downtime, the amount of backlogged goods, and the allocation of operational personnel.

[0069] S606. Update initial dynamic logistics data based on delay data.

[0070] Specifically, based on the stages and content involved in the delay data, the corresponding relevant data segments are located in the acquired initial dynamic logistics data. For example, if the delay data indicates a malfunction in the unloading equipment at a distribution center, then the equipment operation data and operation time data for that unloading stage in the initial dynamic logistics data are identified. The collected delay data is then integrated into the corresponding initial dynamic logistics data, correcting or supplementing the original data. For instance, the equipment failure time is recorded in the equipment operation time series of the initial dynamic logistics data, and the backlog of goods is added to the goods status data for that stage. The updated data then enters the subsequent data processing flow, providing more accurate foundational data for subsequent end-to-end logistics tracking and analysis.

[0071] Furthermore, Figure 7 This is a structural block diagram of a logistics end-to-end timeliness analysis device, such as... Figure 7 As shown, the device includes: The logistics data acquisition module is used to acquire target dynamic logistics data and target static logistics data for multiple links in the entire logistics chain. Among them, the target dynamic logistics data includes target dynamic timeliness data and target dynamic environmental data. The logistics target feature matrix determination module is used to perform dynamic and static feature concatenation on the target static logistics data and the target dynamic logistics data to obtain the logistics target feature matrix. The comprehensive feature determination module is used to extract and combine the long-term trend features and short-term fluctuation features of the mixed flow target feature matrix to obtain comprehensive features. The long-term trend features are used to characterize the cross-cycle persistence pattern, and the short-term fluctuation features are used to characterize the sudden fluctuations of the target duration. The weight matrix feature determination module is used to perform feature weighting and fusion on the comprehensive features based on the target weight allocation to obtain the weight matrix features of each stage; The predicted transportation timeliness determination module is used to map the features of the weight matrix to the predicted transportation timeliness of each link based on the fully connected layer. The number of neurons in the fully connected layer is the same as the number of links in the whole-link logistics. Each neuron outputs the predicted transportation timeliness corresponding to one link. The actual transportation timeliness acquisition module is used to acquire the actual transportation timeliness of each link; The timeliness deviation determination module is used to determine the timeliness deviation results of the entire logistics chain based on the actual transportation timeliness of multiple links and their corresponding predicted transportation timeliness.

[0072] In this embodiment, the logistics data acquisition module includes an initial logistics data acquisition unit, used to acquire initial dynamic logistics data and initial static logistics data of multiple links in the whole-chain logistics; and a data processing unit, used to process the initial dynamic logistics data and initial static logistics data to obtain the corresponding target dynamic logistics data and the corresponding target static logistics data.

[0073] In this embodiment, the logistics target feature matrix determination module includes a static vector integration unit for integrating target static logistics data into a one-dimensional static vector; a dynamic matrix integration unit for integrating target dynamic logistics data into a two-dimensional time-series feature matrix, wherein the two-dimensional time-series feature matrix is ​​determined based on the number of links and the dynamic logistics data corresponding to the number of links; a static matrix expansion unit for aligning the one-dimensional static vector with the time dimension of the two-dimensional time-series feature matrix to expand the one-dimensional static vector into a two-dimensional static feature expansion matrix; and a matrix concatenation unit for concatenating the two-dimensional static feature expansion matrix and the two-dimensional time-series feature matrix column by column to obtain the logistics target feature matrix.

[0074] In this embodiment, the comprehensive feature determination module includes a hybrid network processing unit, which is used to extract the long-term trend features and short-term fluctuation features of the target feature matrix of the stream separately based on the hybrid network and then mix them. The hybrid network is a combination of a long short-term memory network and a gated recurrent unit. The long short-term memory network is used to extract the long-term trend features, and the gated recurrent unit is used to extract the short-term fluctuation features.

[0075] In this embodiment, the weight matrix feature determination module includes an influence degree determination unit, used to determine the influence degree of each link on the entire logistics chain; a weight allocation unit, used to allocate target weights to each link according to the degree of influence, wherein a higher degree of influence corresponds to a higher target weight; and a weighted fusion unit, used to perform feature weighted fusion on the comprehensive features based on the target weight of each link to obtain the weight matrix features of each link.

[0076] In this embodiment, the timeliness deviation determination module includes a link deviation determination unit, used to determine the link timeliness deviation of each link based on the actual transportation timeliness of each link and its corresponding predicted transportation timeliness; a link deviation accumulation unit, used to accumulate the link timeliness deviations of multiple links to obtain the link timeliness deviation; a threshold acquisition unit, used to acquire the actual timeliness of the link and the allowable deviation threshold; and a full-link deviation determination unit, used to determine the timeliness deviation of the entire logistics link based on the link timeliness deviation, the actual timeliness of the link, and the allowable deviation threshold.

[0077] In this embodiment, the logistics end-to-end timeliness analysis device further includes a deviation difference acquisition module, used to acquire the deviation difference between the target predicted timeliness and the target actual timeliness of the target link; an update threshold acquisition module, used to acquire a deviation update threshold; a delay data acquisition module, used to acquire delay data that causes delays in the target link when the deviation difference meets the deviation update threshold; and a data update module, used to update the initial dynamic logistics data based on the delay data.

[0078] In this embodiment, by integrating multi-source heterogeneous data such as manual order entry, PDA scanning and GPS positioning, a complete cargo transportation trajectory is constructed. The transportation process is divided into multiple clear physical links for independent analysis, eliminating the information blind spots in traditional logistics management and realizing end-to-end full-link visual monitoring; (2) By adopting an LSTM-GRU hybrid network combined with the Attention mechanism, it can simultaneously capture long-term trends (such as periodic congestion) and short-term sudden fluctuations (such as sudden traffic accidents) in logistics timeliness data, and dynamically identify key nodes affecting timeliness, thereby improving the prediction accuracy of transportation timeliness in each link and improving the efficiency of transportation. (3) By pre-warning potential delay risks, the passive response can be transformed into proactive management; (4) Through the allocation of link weights and deviation analysis, specific delay links and corresponding responsible entities (such as distribution centers and outlets) can be located, and the responsibility ratio of each link can be quantified, providing a clear basis for performance evaluation and problem rectification, and promoting the refined management of logistics operations; (5) This solution can dynamically adapt to complex scenarios such as sudden weather and traffic congestion, and solve the problem of missing scan data through the data compensation mechanism. While ensuring performance, it reduces computing costs and is suitable for deployment on edge devices or cloud services. It takes into account both practicality and economy, and can effectively improve the stability of logistics timeliness and operational efficiency.

[0079] The application of the relevant modules of the device in this example can be referred to the relevant introduction of the method principle above, and will not be repeated here.

[0080] above Figure 7 The logistics end-to-end timeliness analysis device in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The electronic equipment in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0081] Figure 8This is a schematic diagram of the structure of an electronic device 800 provided in an embodiment of the present invention. The electronic device 800 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 810 (e.g., one or more processors) and a memory 820, and one or more storage media 830 (e.g., one or more mass storage devices) for storing application programs 833 or data 832. The memory 820 and storage media 830 can be temporary or persistent storage. The program stored in the storage media 830 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the electronic device 800. Furthermore, the processor 810 may be configured to communicate with the storage media 830 and execute the series of instruction operations in the storage media 830 on the electronic device 800.

[0082] Electronic device 800 may also include one or more power supplies 840, one or more wired or wireless network interfaces 850, one or more input / output interfaces 860, and / or one or more operating systems 831, such as Windows Server, MacOSX, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 8 The illustrated electronic device structure does not constitute a limitation on electronic devices and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0083] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the logistics end-to-end timeliness analysis method.

[0084] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0085] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0086] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing the timeliness of the entire logistics chain, characterized in that, The method includes: Acquire target dynamic logistics data and target static logistics data for multiple links in the end-to-end logistics process, wherein the target dynamic logistics data includes target dynamic timeliness data and target dynamic environmental data; The static and dynamic logistics data of the target are concatenated to obtain a logistics target feature matrix. Extract and mix the long-term trend features and short-term fluctuation features of the logistics target feature matrix to obtain comprehensive features. The long-term trend features are used to characterize cross-cycle persistence patterns, and the short-term fluctuation features are used to characterize abrupt fluctuations in the target duration. Based on the target weight allocation, the comprehensive features are subjected to feature weighting and fusion to obtain the weight matrix features of each link; The weight matrix features are mapped to the predicted transportation time of each link based on the fully connected layer. The number of neurons in the fully connected layer is the same as the number of links in the whole-link logistics. Each neuron outputs the predicted transportation time of one link. Obtain the actual transportation time for each stage; Based on the actual transportation timeliness of multiple links and their corresponding predicted transportation timeliness, the timeliness deviation results of the entire logistics chain are determined.

2. The logistics end-to-end timeliness analysis method according to claim 1, characterized in that, The acquisition of dynamic logistics data and static logistics data from multiple stages in the entire logistics chain includes: Acquire initial dynamic logistics data and initial static logistics data for multiple stages in the entire logistics chain; The initial dynamic logistics data and the initial static logistics data are processed to obtain the corresponding target dynamic logistics data and the corresponding target static logistics data.

3. The logistics end-to-end timeliness analysis method according to claim 1, characterized in that, The step of concatenating the static and dynamic features of the target static logistics data and the target dynamic logistics data to obtain the logistics target feature matrix includes: The target static logistics data is integrated into a one-dimensional static vector; The target dynamic logistics data is integrated into a two-dimensional time-series feature matrix, wherein the two-dimensional time-series feature matrix is ​​determined based on the number of links and the dynamic logistics data corresponding to the number of each link. The one-dimensional static vector is aligned with the time dimension of the two-dimensional temporal feature matrix to expand the one-dimensional static vector into a two-dimensional static feature expansion matrix. The two-dimensional static feature extension matrix and the two-dimensional time-series feature matrix are concatenated column-wise to obtain the logistics target feature matrix.

4. The logistics end-to-end timeliness analysis method according to claim 1, characterized in that, The extraction and mixing of the long-term trend features and short-term fluctuation features of the logistics target feature matrix yields comprehensive features including: The long-term trend features and short-term fluctuation features of the logistics target feature matrix are extracted and then mixed based on a hybrid network. The hybrid network is a combination of a long short-term memory network and a gated recurrent unit. The long short-term memory network is used to extract the long-term trend features, and the gated recurrent unit is used to extract the short-term fluctuation features.

5. The logistics end-to-end timeliness analysis method according to claim 1, characterized in that, The weighted fusion of the comprehensive features based on the target weight allocation to obtain the weight matrix features of each stage includes: Determine the impact of each link on the entire logistics chain; Target weights are assigned to each stage according to the degree of impact, with higher impact corresponding to higher target weights; Based on the target weight of each stage, the comprehensive features are weighted and fused to obtain the weight matrix features of each stage.

6. The logistics end-to-end timeliness analysis method according to claim 1, characterized in that, The timeliness deviation results for the entire logistics chain, determined based on the actual transportation timeliness and their corresponding predicted transportation timeliness at multiple stages, include: Based on the actual transportation timeliness of each link and its corresponding predicted transportation timeliness, the timeliness deviation of each link is determined; The link timeliness deviation is obtained by summing up the timeliness deviations of multiple links; Obtain the actual timeliness and allowable deviation threshold of the link; Based on the timeliness deviation of the link, the actual timeliness of the link, and the allowable deviation threshold, the timeliness deviation of the entire logistics link is determined.

7. The logistics end-to-end timeliness analysis method according to claim 1, characterized in that, The method further includes: Obtain the deviation difference between the predicted timeliness and the actual timeliness of the target stage; Obtain the deviation update threshold; When the deviation difference meets the deviation update threshold, the delay data that caused the delay in the target process is obtained; The initial dynamic logistics data is updated based on the delay data.

8. A logistics end-to-end timeliness analysis device, characterized in that, The device includes: The logistics data acquisition module is used to acquire target dynamic logistics data and target static logistics data for multiple links in the whole-chain logistics. The target dynamic logistics data includes target dynamic timeliness data and target dynamic environmental data. The logistics target feature matrix determination module is used to perform dynamic and static feature concatenation on the target static logistics data and the target dynamic logistics data to obtain the logistics target feature matrix. The comprehensive feature determination module is used to extract and mix the long-term trend features and short-term fluctuation features of the logistics target feature matrix to obtain comprehensive features. The long-term trend features are used to characterize cross-cycle persistence patterns, and the short-term fluctuation features are used to characterize abrupt fluctuations in the target duration. The weight matrix feature determination module is used to perform feature weighting and fusion on the comprehensive features based on the target weight allocation to obtain the weight matrix features of each step; The predicted transportation timeliness determination module is used to map the features of the weight matrix to the predicted transportation timeliness of each link based on the fully connected layer. The number of neurons in the fully connected layer is the same as the number of links in the whole-link logistics division, and each neuron outputs the predicted transportation timeliness corresponding to one link. The actual transportation timeliness acquisition module is used to acquire the actual transportation timeliness of each link; The timeliness deviation determination module is used to determine the timeliness deviation results of the entire logistics chain based on the actual transportation timeliness of multiple links and their corresponding predicted transportation timeliness.

9. An electronic device, characterized in that, The electronic device includes a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the electronic device to perform the various steps of the logistics end-to-end timeliness analysis method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instruction is executed by the processor, it implements each step of the logistics end-to-end timeliness analysis as described in any one of claims 1-7.

Images