Logistics sending amount prediction method, device and equipment and storage medium

Abstract

The present application relates to the technical field of intelligent logistics, and more particularly to a logistics mail sending quantity prediction method, device, equipment and storage medium, the method first acquires login role information, colloquial query information and a pre-trained time series prediction model, then generates a permission identifier based on the login role information, performs natural language analysis on the colloquial query information to obtain target analysis parameters, then performs hierarchical data association based on the permission identifier and the target analysis parameters to obtain a standardized historical data set, then inputs the standardized historical data set into the time series prediction model to perform offline mail sending quantity prediction to obtain a target prediction result, and finally performs format conversion on the target prediction result based on the permission identifier to obtain a display result; the method aims to solve the technical problems of insufficient data association, high network dependency and extensive permission management in existing logistics mail sending quantity prediction systems, and effectively improves logistics operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics technology, and in particular to a method, apparatus, equipment and storage medium for predicting the volume of logistics shipments. Background Technology

[0002] In logistics operations management, dispatch volume forecasting is a crucial basis for operational decisions such as resource allocation, staff scheduling, and business planning, playing a key role in improving logistics operational efficiency and optimizing business layout. However, with the increasing scale and hierarchy of the logistics industry and the continuous refinement of dispatch organizational structures, existing dispatch volume forecasting systems have gradually revealed numerous shortcomings in practical applications, failing to meet the actual needs of modern logistics operations.

[0003] Existing forecasting systems suffer from significant deficiencies in data correlation capabilities. Most can only provide forecast data for a single branch and cannot automatically correlate and integrate data based on the hierarchical relationship between primary branches and subordinate branches. Operations personnel must manually query data at different levels multiple times, severely reducing business processing efficiency. Furthermore, the system interaction methods are relatively simple and rigid, generally employing fixed parameter input modes. Users must manually fill in information such as dates and branch codes, lacking natural language interaction capabilities and presenting a high operational threshold.

[0004] Furthermore, existing systems largely rely on online query modes, requiring a continuous server connection during operation. This makes them unusable in scenarios with unstable or interrupted networks, compromising business continuity. Regarding forecasting, the system only supports single-period forecasts for the next day, failing to provide continuous forecasts for multiple days and thus hindering adaptability to diverse business planning needs. In terms of access control, the system lacks refined display rules for users at different levels, resulting in a mismatch between displayed data and user permissions. This compromises data security and reduces user experience. These issues collectively limit the practicality of the logistics shipment volume forecasting system, preventing it from meeting the demands of refined and efficient logistics operation management. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, the present invention aims to provide a method, device, equipment and storage medium for predicting the volume of logistics shipments, which aims to solve the technical problems of insufficient data correlation, high dependence on network and crude permission management in the existing logistics shipment volume prediction system, and effectively improve the efficiency of logistics operation.

[0006] The first aspect of this invention provides a method for predicting the volume of logistics shipments, comprising: acquiring login role information, conversational query information, and a pre-trained time-series prediction model; generating an access identifier based on the login role information; performing natural language parsing on the conversational query information to obtain target parsing parameters; performing hierarchical data association based on the access identifier and the target parsing parameters to obtain a standardized historical dataset; inputting the standardized historical dataset into the time-series prediction model for offline shipment volume prediction to obtain a target prediction result; and performing format conversion on the target prediction result based on the access identifier to obtain a display result.

[0007] Optionally, in a first implementation of the first aspect of the present invention, generating a permission identifier based on the login role information includes: obtaining a preset permission rule base; matching the branch-level permission range corresponding to the login role based on the permission rule base; performing hash encryption processing on the branch-level permission range using a hash encryption algorithm to obtain the permission identifier, and caching the permission identifier.

[0008] Optionally, in a second implementation of the first aspect of the present invention, the step of performing natural language parsing on the colloquial query information to obtain target parsing parameters includes: using a parsable text conversion tool to convert the colloquial query information into parsable text information; calling a pre-trained semantic parsing model to perform semantic parsing processing on the parsable text information to obtain initial parsing parameters; obtaining preset verification rules, and performing integrity verification on the initial parsing parameters based on the verification rules to obtain the target parsing parameters.

[0009] Optionally, in a third implementation of the first aspect of the present invention, the semantic parsing model includes a text segmentation and annotation module, a semantic matching module, a core parameter extraction module, and a parameter format conversion module, wherein the text segmentation and annotation module, the semantic matching module, the core parameter extraction module, and the parameter format conversion module are connected sequentially; the step of calling the pre-trained semantic parsing model to perform semantic parsing processing on the parsable text information to obtain initial parsing parameters includes: performing text segmentation and annotation on the parsable text information based on the text segmentation and annotation module to obtain annotation information; performing semantic matching processing on the annotation information based on the semantic matching module to obtain matching information; performing core parameter extraction processing on the matching information based on the core parameter extraction module to obtain time parameters and halftone parameters; and performing format conversion processing on the time parameters and the halftone parameters respectively based on the parameter format conversion module to obtain the initial parsing parameters.

[0010] Optionally, in a fourth implementation of the first aspect of the present invention, the target parsing parameters include target time parsing parameters and target network point parsing parameters; the step of performing hierarchical data association based on the permission identifier and the target parsing parameters to obtain a standardized historical dataset includes: obtaining preset hierarchical affiliation rules; determining the queryable network point level based on the permission identifier and determining the queryable network point node based on the target network point parsing parameters; performing hierarchical data association based on the hierarchical affiliation rules according to the queryable network point level and the queryable network point node to obtain a historical dataset corresponding to the queryable network point level and the queryable network point node; performing abnormal data removal processing on the historical dataset using an anomaly detection algorithm to obtain a preprocessed dataset; selecting the historical contemporaneous dataset corresponding to the target time parsing parameters from the preprocessed dataset, and converting the historical contemporaneous dataset into a standard format using a structured coding algorithm to obtain the standardized historical dataset.

[0011] Optionally, in a fifth implementation of the first aspect of the present invention, the time-series prediction model includes a scenario-based feature filtering module, an offline prediction module, and a correction module, wherein the scenario-based feature filtering module, the offline prediction module, and the correction module are connected sequentially; the step of inputting the standardized historical dataset into the time-series prediction model for offline shipment volume prediction to obtain a target prediction result includes: performing scenario-based feature filtering on the standardized historical dataset based on the scenario-based feature filtering module to obtain a feature dataset; performing offline prediction on the feature dataset based on the offline prediction module to obtain an initial prediction result; and correcting the initial prediction result based on the correction module to obtain the target prediction result.

[0012] Optionally, in a sixth implementation of the first aspect of the present invention, the step of converting the format of the target prediction result based on the permission identifier to obtain the display result includes: determining the data display boundary based on the permission identifier; obtaining a preset view rendering rule; matching the corresponding view encapsulation format for the data display boundary based on the view rendering rule; and converting the format of the target prediction result based on the view encapsulation format to obtain the display result.

[0013] A second aspect of the present invention provides a logistics shipment volume prediction device, comprising: a data acquisition module for acquiring login role information, conversational query information, and a pre-trained time-series prediction model; an identifier generation module for generating an access identifier based on the login role information; a parsing module for performing natural language parsing on the conversational query information to obtain target parsing parameters; a data association module for performing hierarchical data association based on the access identifier and the target parsing parameters to obtain a standardized historical dataset; a prediction module for inputting the standardized historical dataset into the time-series prediction model for offline shipment volume prediction to obtain a target prediction result; and a display module for converting the target prediction result in a format based on the access identifier to obtain a display result.

[0014] A third aspect of the present invention provides a logistics shipment volume prediction device, the logistics shipment volume prediction device comprising: a memory and at least one processor, the memory storing instructions; the at least one processor calling the instructions in the memory to cause the logistics shipment volume prediction device to perform the various steps of the logistics shipment volume prediction method described in any of the preceding claims.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the logistics shipment volume prediction method described in any of the preceding claims.

[0016] In the technical solution of this invention, login role information, conversational query information, and a pre-trained time-series prediction model are first obtained. An access permission identifier is generated based on the login role information. Natural language parsing is performed on the conversational query information to obtain target parsing parameters. Hierarchical data association is performed based on the access permission identifier and the target parsing parameters to obtain a standardized historical dataset. This standardized historical dataset is then input into the time-series prediction model for offline shipment volume prediction to obtain the target prediction result. Finally, the target prediction result is format-converted based on the access permission identifier to obtain the display result. This approach aims to solve the technical problems of insufficient data association, high network dependence, and crude access permission management in existing logistics shipment volume prediction systems, effectively improving logistics operation efficiency. Attached Figure Description

[0017] Figure 1 This is a first flowchart of a logistics shipment volume prediction method provided in an embodiment of the present invention; Figure 2 This is a second flowchart of the logistics shipment volume prediction method provided in an embodiment of the present invention; Figure 3 This is a third flowchart of the logistics shipment volume prediction method provided in the embodiments of the present invention; Figure 4This is a fourth flowchart of the logistics shipment volume prediction method provided in the embodiments of the present invention; Figure 5 This is a fifth flowchart of the logistics shipment volume prediction method provided in the embodiments of the present invention; Figure 6 This is a sixth flowchart of the logistics shipment volume prediction method provided in the embodiments of the present invention; Figure 7 This is the seventh flowchart of the logistics shipment volume prediction method provided in the embodiments of the present invention; Figure 8 This is a schematic diagram of the structure of the logistics shipment volume prediction device provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of the logistics shipment volume prediction device provided in an embodiment of the present invention. Detailed Implementation

[0018] This invention provides a method, apparatus, device, and storage medium for predicting the volume of logistics shipments. In this invention, the terms "first," "second," "third," "fourth," etc. (if present)," in the specification, claims, and accompanying drawings 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 a sequence other than that 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 device 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.

[0019] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the logistics shipment volume prediction method in this invention includes: 101. Obtain login role information, conversational query information, and pre-trained time series prediction model; In this embodiment, login role information refers to the role-related information bound to the user account obtained by the system after the user completes login verification through the system account and password. This covers the permission association information corresponding to three types of roles: Level 1 branch administrator, branch manager, and service department administrator. Its core function is to provide the foundation for the generation of subsequent permission identifiers, ensuring that all subsequent operations, such as data association, model invocation, and result display, comply with the user's permission boundaries. Conversational query information refers to the conversational expressions entered by the user through the system's interactive interface for querying logistics shipment volume predictions. This can be obtained through text input or voice input followed by text conversion. It covers various conversational inquiries related to shipment volume prediction, and its core content includes key information related to the prediction period and branch information, providing raw input for subsequent natural language parsing to extract target parsing parameters. The pre-trained time series prediction model adopts a time series prediction framework adapted to short-term prediction of logistics shipment volume. It uses historical shipment volume data of logistics network points as the core training data and incorporates key factors affecting shipment volume such as time period, date, weather, holidays, and promotional activities. The parameters are optimized through model training to complete the model pre-training. After training, the model parameters are fixed in advance and stored locally, so that offline calls can be realized without online loading. It can support short-term logistics shipment volume prediction for the next several days.

[0020] 102. Generate an access permission identifier based on the login role information; In this embodiment, a preset permission rule base is first obtained. This rule base clarifies the correspondence between various login roles and branch-level permission scopes. Then, based on this rule base, the branch-level permission scope corresponding to the current login role is matched to clarify the boundaries of branch data that different login roles can query and use. Specifically, the permission scope of roles related to first-level branches covers themselves and their subordinate branches, the permission scope of roles related to branches is limited to the branches they manage, and the permission scope of roles related to service departments is limited to their own branches. Subsequently, a hash encryption algorithm is used to encrypt the matched branch-level permission scope to generate a unique and secure permission identifier. Encryption effectively prevents permission information from being tampered with or leaked, ensuring the security of permission data. Finally, the generated permission identifier is cached to provide permission verification basis for subsequent data association, model invocation, and result display throughout the entire process. By matching login roles and permission scopes, combined with encryption and caching, the accuracy, security, and callability of permission identifiers are ensured. The permission boundaries of different login roles are strictly defined, effectively avoiding cross-authority violations and ensuring the compliance and security of data use. At the same time, the cached permission identifiers can be quickly invoked for subsequent full-process permission verification, improving operational efficiency.

[0021] 103. Perform natural language parsing on the colloquial query information to obtain the target parsing parameters; In this embodiment, the colloquial query information refers to the user's inquiry related to mail volume prediction input through the system's interactive interface. This can be obtained through text input or voice recording followed by conversion. The expression is colloquial and includes potential key information related to mail volume prediction, such as the prediction period and network location. During the parsing process, a parsable text conversion tool is first used to transform the colloquial query information into standardized parsable text information, eliminating the arbitrariness and non-standardization of colloquial expressions and providing a unified and standardized input foundation for subsequent semantic parsing. Then, a pre-trained semantic parsing model is invoked to perform deep processing on the parsable text information, including word segmentation, part-of-speech tagging, and semantic matching. Invalid interjections that do not affect the prediction requirements are filtered out, and initial parsing parameters related to mail volume prediction are accurately extracted. These initial parsing parameters mainly cover time parameters and network location parameters. Time parameters correspond to different prediction periods; if not specified, T+1 (one day after the current date) is used by default. Network location parameters are obtained by matching the corresponding codes with network-related information in the parsed text; if not specified, the default network location within the currently logged-in user's permission range is used. Finally, preset verification rules are obtained, and the initial parsing parameters are checked for completeness according to these rules. If key parameters are missing, the system automatically prompts and guides the user to supplement them, ensuring the completeness and validity of the parameters, ultimately yielding the target parsing parameters. This effectively breaks the limitations of traditional fixed parameter input, adapts to users' conversational interaction habits, lowers the operational threshold, and ensures the accuracy of key parameter extraction through standardized conversion and precise semantic parsing. Combined with the completeness verification step, it avoids the impact of missing or incorrect parameters on subsequent prediction processes.

[0022] 104. Based on the permission identifier and the target parsing parameters, perform hierarchical data association to obtain a standardized historical dataset; In this embodiment, a preset hierarchical membership rule is first obtained. This rule stores the hierarchical membership relationships of various network points and keeps them updated in real time, providing a unified logical basis for hierarchical data association. Then, the queryable network point hierarchy is clarified based on the permission identifier, defining the permission boundaries of data association. Simultaneously, the query network point node is determined based on the target network point parsing parameters in the target parsing parameters, clarifying the core object of data association and ensuring that the scope of data association both meets permission requirements and aligns with user query needs. Based on this, hierarchical data association operations are performed according to the hierarchical membership rule, combined with the queryable network point hierarchy and the query network point node. According to the permission differences of different queryable network point hierarchies, corresponding association logic is matched, associating only network point data within the permission range, thereby obtaining the historical dataset corresponding to the queryable network point hierarchy and the query network point node. To ensure data quality, an anomaly detection algorithm is used to perform anomaly data removal processing on the historical dataset, removing invalid data such as missing values ​​and extreme outliers, resulting in a preprocessed dataset. Finally, historical data corresponding to the target time parameter in the target parsing parameters are selected from the preprocessed dataset to ensure the correlation between the data and the prediction period. Then, a structured coding algorithm is used to transform the historical dataset into a unified standard format, eliminating data format differences and ultimately obtaining a standardized historical dataset. This ensures the compliance and accuracy of hierarchical data association and effectively solves the problems of ambiguous permission boundaries and insufficient data correlation in traditional data association.

[0023] 105. Input the standardized historical dataset into the time-series prediction model to perform offline shipment volume prediction and obtain the target prediction result; In this embodiment, the time-series prediction model training process uses historical shipment volume data from logistics network points as the core training sample, incorporating various key factors affecting shipment volume such as time period, date, weather, holidays, and promotional activities. Model parameters are pre-fixed and stored locally, enabling offline invocation without online loading, and can stably support short-term shipment volume prediction needs for the next several days. During the prediction process, after inputting the standardized historical dataset into the time-series prediction model, the model automatically extracts the core feature parameters from the data and independently performs prediction calculations relying on local computing power, without relying on the network throughout the process, ensuring a prediction response speed of seconds and guaranteeing prediction efficiency. Simultaneously, locally cached auxiliary prediction data is used to specifically correct the initial prediction results output by the model, effectively reducing prediction errors and improving the accuracy of the prediction results. After calculation and correction, the time-series prediction model outputs the target prediction result that meets the requirements, synchronously caching prediction-related parameters, data source information, etc., locally and recording logs, providing reliable support for subsequent related processes, eliminating dependence on the network, ensuring that shipment volume prediction can still be carried out normally in scenarios with unstable or interrupted networks, and guaranteeing business continuity.

[0024] 106. Based on the permission identifier, the target prediction result is format-converted to obtain the display result.

[0025] In this embodiment, the data display boundaries are first clearly defined based on permission identifiers. According to the permission levels corresponding to the permission identifiers, the range of target prediction results that different users can view is defined, ensuring that the displayed content strictly conforms to the permission boundaries and avoiding unauthorized data display. Then, preset view rendering rules are obtained. These rules clarify the view presentation standards and encapsulation specifications corresponding to different data display boundaries. Based on these rules, the corresponding view encapsulation format is matched to the determined data display boundaries, achieving a precise correspondence between permissions and display formats, ensuring that users with different permissions receive a display style suitable for their needs. On this basis, the target prediction results are format-converted according to the matched view encapsulation format. Based on the requirements of the view encapsulation format, the level of detail, classification method, and presentation format of the prediction results are adjusted to adapt to the need for rapid switching between multi-day predictions. Simultaneously, results with prediction accuracy below a preset threshold are anomaly-marked and given relevant prompts, ensuring that the displayed content not only meets permission requirements but also has good readability and usability, ultimately yielding the final display result. It achieves deep adaptation between target prediction results and permission identifiers. By clearly defining data display boundaries and matching exclusive view encapsulation formats, it strictly ensures the compliance and security of data use and avoids unauthorized viewing. Differentiated format conversion methods cater to the actual usage needs of users with different permissions, simplifying the interference of unnecessary information while highlighting core content, thus improving user experience and data viewing efficiency. The integration of multi-day prediction switching adaptation and anomaly labeling functions not only facilitates users to quickly obtain prediction results for different time periods but also guides users to reasonably judge the usability of prediction results, improving the credibility of prediction results. The entire format conversion process is automatically completed based on preset rules without manual intervention, improving the generation efficiency and standardization of display results. It provides intuitive and accurate display support for logistics network operation decisions, further improving the closed loop of the entire prediction method and enhancing its overall practicality and adaptability.

[0026] Please see Figure 2 In this embodiment of the invention, generating an access identifier based on the login role information includes: 201. Obtain the preset permission rule base; 202. Based on the permission rule base, match the branch-level permission range corresponding to the logged-in role; 203. The network point level permission range is hash-encrypted using a hash encryption algorithm to obtain the permission identifier, and the permission identifier is cached.

[0027] In this embodiment, a pre-defined permission rule base contains a fixed correspondence between three types of login roles—Level 1 Branch Administrator, Branch Administrator, and Service Department Administrator—and their respective branch-level permission scopes, providing a standardized basis for permission matching. After user authentication, the system reads the login role information bound to the account and performs precise matching based on the permission rule base, clearly defining the branch-level permission scope exclusive to each role. Specifically, Level 1 Branch Administrators correspond to the permission scope of themselves and all their subordinate branches, Branch Administrators correspond to the permission scope of the branches they manage, and Service Department Administrators correspond to the permission scope of their own service departments, ensuring a one-to-one correspondence between permissions and roles without overlapping or exceeding permissions. The encryption process uses the MD5 hash encryption algorithm, which has the advantages of high computational efficiency, strong uniqueness, and outstanding anti-tampering capabilities. It can convert the branch-level permission scope into a fixed-format encrypted identifier, technically eliminating the risk of permission information being tampered with or forged. After generating the encrypted permission identifier, the system caches and stores it, providing a stable basis for permission verification for subsequent data association, model invocation, result display, and other full-process operations. This achieves accurate generation and secure management of permission identifiers, strictly defines the data operation boundaries of different roles, and ensures the compliance and security of system data. The MD5 hash encryption algorithm strengthens the protection capability of permission data, while the caching mechanism enables fast retrieval of permission identifiers, significantly improving the operational efficiency of the entire process of permission verification and providing reliable technical support for the overall permission control of the logistics shipment volume prediction method.

[0028] Please see Figure 3 In this embodiment of the invention, the step of performing natural language parsing on the colloquial query information to obtain target parsing parameters includes: 301. Use a parsable text conversion tool to convert the colloquial query information into parsable text information; 302. Call the pre-trained semantic parsing model to perform semantic parsing processing on the parsable text information to obtain initial parsing parameters; 303. Obtain the preset verification rules, and perform integrity verification on the initial parsing parameters based on the verification rules to obtain the target parsing parameters.

[0029] In this embodiment, the parsable text conversion tool specifically employs a combination of a speech-to-text tool and a text normalization tool. The speech-to-text tool accurately converts spoken query information into text format, while the text normalization tool preprocesses the direct text input or the spoken text after speech conversion, eliminating redundant interjections, disordered word order, and other issues in spoken expressions, unifying the text format, and transforming it into semantically clear and formatted parsable text information, providing standardized input for subsequent semantic parsing. The pre-trained semantic parsing model can accurately segment, tag, and semantically match the parsable text information, effectively filtering out invalid interjections that do not affect prediction requirements, and accurately extracting initial parsing parameters related to the shipment volume prediction. The initial parsing parameters mainly cover time parameters and network parameters. The time parameter corresponds to the prediction period for multiple days in the future, and defaults to T+1 (one day after the current date) when the user does not specify. The network parameter matches the corresponding code through network-related information in the parsed text, and defaults to the network within the currently logged-in user's permission range when the user does not specify. Subsequently, preset verification rules are obtained, which clarify the completeness standards of key parameters required for shipment volume prediction. Based on these rules, the initial parsing parameters are verified one by one. An automatic prompt mechanism is triggered for missing key parameters, guiding the user to supplement and improve them, ensuring the completeness and validity of the parameters, and ultimately obtaining the target parsing parameters. This approach achieves adaptation and standardized conversion to multiple input formats for conversational query information, breaking the limitations of traditional fixed parameter input, lowering the user's operational threshold, improving the accuracy and efficiency of initial parsing parameter extraction, and effectively reducing parameter extraction deviations caused by semantic misunderstandings. The completeness verification step further ensures the validity of the target parsing parameters, preventing the subsequent prediction process from being affected by missing or incorrect parameters.

[0030] Please see Figure 4 In this embodiment of the invention, the semantic parsing model includes a text segmentation and annotation module, a semantic matching module, a core parameter extraction module, and a parameter format conversion module, which are sequentially connected. The step of calling the pre-trained semantic parsing model to perform semantic parsing processing on the parsable text information to obtain initial parsing parameters includes: 401. Based on the text segmentation and annotation module, perform text segmentation and annotation on the parsable text information to obtain annotation information; 402. Based on the semantic matching module, perform semantic matching processing on the annotation information to obtain matching information; 403. Based on the core parameter extraction module, the matching information is processed to extract core parameters to obtain time parameters and network point parameters; 404. Based on the parameter format conversion module, the time parameter and the network point parameter are converted into their respective formats to obtain the initial parsing parameters.

[0031] In this embodiment, the semantic parsing model is a dedicated functional model for predicting the volume of shipments in logistics, rather than a purely theoretical single-algorithm model. The overall model uses the BERT pre-trained model as a unified core architecture, and each functional module adopts the optimal algorithm adapted to its own task. These are sub-task processing methods that serve the overall model. Specifically, the text segmentation and annotation module performs text segmentation and annotation on parsable text information. It uses a combination of the jieba segmentation algorithm and the part-of-speech tagging algorithm to accurately segment the parsable text information, breaking down continuous text into independent semantic units. At the same time, it performs part-of-speech tagging on each semantic unit, clarifying the part-of-speech categories such as nouns, verbs, and modal particles. Invalid modal particles that do not affect the prediction requirements are filtered out simultaneously, resulting in annotation information containing segmentation results and part-of-speech tags, which provides a foundation for subsequent semantic matching. The semantic matching module performs semantic matching processing on the labeled information, using a cosine similarity matching algorithm to compare the labeled information with a pre-stored semantic library related to logistics shipment volume prediction in the model. This identifies semantic units relevant to shipment volume prediction, filters out irrelevant semantic information, and obtains matching information highly correlated with the prediction requirements, ensuring the accuracy of the parsing direction. The core parameter extraction module extracts core parameters from the matching information, accurately extracting time and network point parameters based on preset parameter extraction rules. The time parameter corresponds to the prediction period for the next several days; if the user does not specify a time, the default T+1 (one day after the current date) is automatically extracted. The network point parameter is obtained by associating the network point name, level, etc., with the corresponding network point code; if the user does not specify a network point, the default network point within the currently logged-in user's permission range is extracted. The parameter format conversion module performs format conversion processing on both the time and network point parameters, using a standardized coding algorithm to convert the extracted time and network point parameters into a unified format that the model can recognize and call, eliminating differences in parameter expression and ultimately obtaining the initial parsing parameters.

[0032] In this embodiment, the training process of the semantic parsing model relies on a massive amount of colloquial query samples related to logistics shipment volume. First, query samples covering different prediction scenarios and different colloquial expression styles are collected. These samples are then manually labeled to clarify the time parameters, network parameters, and invalid semantic information corresponding to each sample, thus constructing a labeled sample set. Subsequently, the model structure is initialized using BERT (Bidirectional Encoder Representations). Using a pre-trained model from Transformers as the basic framework, labeled sample sets are input into the model for training. The model parameters are continuously optimized through gradient descent algorithm, adjusting word segmentation accuracy, semantic matching threshold, and parameter extraction rules. At the same time, semantic features specific to the logistics industry are incorporated to improve the model's adaptability to colloquial queries in logistics scenarios. During the training process, continuous verification and iteration are carried out until the model's word segmentation accuracy, semantic matching accuracy, and parameter extraction accuracy reach the preset standards, completing the model pre-training. After training, the model parameters are fixed and stored to ensure stability and accuracy when called. The model is adapted to logistics scenarios through training with specific samples, and can accurately identify the core semantics and parameters related to logistics shipment volume prediction, adapting to different colloquial expression styles, breaking the limitations of traditional fixed parameter input, reducing the user's operation threshold, and improving parsing efficiency.

[0033] Please see Figure 5 In this embodiment of the invention, the target parsing parameters include target time parsing parameters and target network point parsing parameters; the step of performing hierarchical data association based on the permission identifier and the target parsing parameters to obtain a standardized historical dataset includes: 501. Obtain the preset hierarchical membership rules; 502. Determine the queryable network level based on the permission identifier, and determine the query network node based on the target network parsing parameters; 503. Based on the hierarchical membership rules, perform hierarchical data association according to the queryable network point level and the query network point node to obtain the historical dataset corresponding to the queryable network point level and the query network point node; 504. An anomaly detection algorithm is used to remove abnormal data from the historical dataset to obtain a preprocessed dataset; 505. Select the historical contemporaneous dataset corresponding to the target time parsing parameter from the preprocessed dataset, and use a structured coding algorithm to convert the historical contemporaneous dataset into a standard format to obtain the standardized historical dataset.

[0034] In this embodiment, the target parsing parameters include target time parsing parameters and target network point parsing parameters. The target time parsing parameter corresponds to the time period demand for shipment volume prediction, covering short-term prediction periods for the next several days. When the user does not specify a time, the default value is T+1. The target network point parsing parameter corresponds to the network point object being queried. It matches the corresponding network point code by parsing network point-related content in the colloquial query information. When the user does not specify a network point, the value is the default network point within the currently logged-in user's permission range. Based on the permission identifier and target parsing parameters, hierarchical data association is performed to obtain a standardized historical dataset. First, a preset hierarchical affiliation rule is obtained. This rule stores the hierarchical affiliation relationships of various network points and keeps them updated in real time, clarifying the affiliation relationships between first-level network points and their subordinate branches and service departments, as well as between branches and their subordinate service departments, providing a unified logical basis for hierarchical data association. Then, based on the permission identifier, the queried network point level is determined, defining the permission boundaries of data association. Simultaneously, based on the target network point parsing parameters, the queried network point node is determined, clarifying the objects of data association and ensuring that the scope of data association both meets permission requirements and aligns with user query needs. Based on this, according to the hierarchical affiliation rules, the hierarchical data association operation is performed by combining the queryable branch level and the queryable branch node. According to the permission differences of different queryable branch levels, the corresponding association logic is matched, and only the branch data within the permission range is associated, thereby obtaining the historical dataset corresponding to the queryable branch level and the queryable branch node.

[0035] To ensure data quality, the Isolation Forest algorithm is used to remove outliers from the historical dataset. This algorithm boasts high detection efficiency and sensitivity to outliers, effectively identifying and removing missing values, extreme outliers, and other invalid data to obtain a preprocessed dataset. Then, historical data corresponding to the target time parsing parameters are selected from the preprocessed dataset to ensure the correlation between the data and the prediction period, improving the accuracy of subsequent predictions. Next, a JSON standardization encoding algorithm is used to transform the historical dataset into a unified standard format, eliminating format differences between data from different sources and achieving standardized data integration, ultimately resulting in a standardized historical dataset. Through standardized association, filtering, and standardization, the problems of data redundancy, inconsistent formats, and ambiguous permission boundaries in traditional data association are effectively solved, improving data processing efficiency and providing solid data support for the accuracy of logistics shipment volume prediction, while also adapting to the query needs of users at different levels.

[0036] Please see Figure 6 In this embodiment of the invention, the time-series prediction model includes a scenario-based feature filtering module, an offline prediction module, and a correction module, which are sequentially connected. The step of inputting the standardized historical dataset into the time-series prediction model for offline shipment volume prediction to obtain the target prediction result includes: 601. Based on the scenario-based feature filtering module, the standardized historical dataset is subjected to scenario-based feature filtering to obtain a feature dataset; 602. Perform offline prediction on the feature dataset based on the offline prediction module to obtain initial prediction results; 603. Based on the correction module, the initial prediction result is corrected to obtain the target prediction result.

[0037] In this embodiment, the time-series prediction model is an integrated pre-trained model trained by the system. The training and solidification of all model parameters revolve around the core objective of LSTM time-series prediction. The scenario-based feature selection module, offline prediction module, and correction module are not independent functional units, but rather organic components that are closely connected and work together in the order of data flow within the pre-trained model. The algorithms used in each module are all supporting means for the core LSTM algorithm, not independent and isolated modeling algorithms. The three modules form a progressive and strongly correlated closed loop of pre-feature adaptation, time-series prediction, and error optimization, all centered around the core task of offline shipment volume prediction. Specifically, the scenario-based feature selection module performs scenario-based feature selection on the standardized historical dataset. The ReliefF feature selection algorithm is used, combined with the core influencing factors of the logistics shipment scenario, to accurately select feature parameters that are highly correlated with shipment volume prediction from the standardized historical dataset. These parameters include historical shipment volume for the same period, daily average fluctuation coefficient, seasonal influencing factors, weather influencing parameters, and features related to holidays and promotional activities. Irrelevant and redundant features are removed to obtain a feature dataset with prominent and targeted features, providing a high-quality input foundation for subsequent offline prediction. The feature dataset is predicted offline based on the offline prediction module. The LSTM (Long Short-Term Memory) time series prediction algorithm is adopted. This algorithm has a powerful time series data modeling capability and can accurately capture the time series pattern of historical shipment data. Relying on the model parameters that are fixed in advance and stored locally, there is no need to load resources online. The prediction calculation can be executed quickly using only local computing power, ensuring that the time consumption of a single prediction is controlled within seconds and efficiently outputting the initial prediction results, which fully meets the core requirements of offline prediction. The initial prediction results are corrected based on the correction module. The Kalman filter correction algorithm is used to call the auxiliary prediction data cached locally, including supplementary data synchronized to the local machine in advance, such as recent weather forecast results, holiday arrangements, and branch operation adjustment notices. The algorithm dynamically corrects the initial prediction results, effectively offsetting the prediction deviations caused by various interference factors, improving the accuracy of the prediction results, and finally obtaining the target prediction result.

[0038] The training process of the time-series prediction model needs to be carried out around the short-term prediction needs of logistics shipment volume. First, a massive amount of historical shipment volume data from logistics network points is collected, and relevant data on various factors affecting shipment volume, such as time period, date, weather, holidays, and promotional activities, are integrated to construct a comprehensive training sample set. The training sample set is preprocessed, including data cleaning, outlier removal, and standardization transformation, to ensure the integrity and standardization of the sample data. Then, the model structure is initialized, based on the LSTM network framework, integrating the core logic of scenario-based feature selection, offline prediction, and correction. The preprocessed training sample set is input into the model, and the Adam (Adaptive Moment Estimation) optimization algorithm is used to continuously adjust the model parameters, optimize the feature selection threshold, prediction calculation weights, and correction coefficients, and incorporate logistics industry-specific scenario features to improve the model's adaptability to the fluctuation patterns of logistics shipment volume. During the training process, the training set and validation set are continuously divided, and the model prediction accuracy is verified in real time through the validation set. Iterative optimization is carried out repeatedly until the model's prediction accuracy and response speed reach the preset standards, completing the model pre-training. After training, the model parameters are fixed and stored locally to ensure stability and efficiency when called offline.

[0039] Please see Figure 7 In this embodiment of the invention, the step of converting the format of the target prediction result based on the permission identifier to obtain the display result includes: 701. Determine the data display boundaries based on the permission identifier; 702. Obtain the preset view rendering rules, and match the corresponding view encapsulation format for the data display boundary based on the view rendering rules; 703. Based on the view encapsulation format, the target prediction result is converted to obtain the display result.

[0040] In this embodiment, data display boundaries are determined based on permission identifiers. According to the user permission levels corresponding to the permission identifiers, the range of target prediction results that different users can view is defined, distinguishing the display boundaries for primary network points, branches, and service departments. This ensures that the displayed content strictly conforms to permission requirements, clearly defining that primary network points can view their own and subordinate branches' prediction data, branches can only view their own prediction data, and service departments can only view their own simplified prediction data. This establishes a clear scope benchmark for subsequent format conversion. Preset view rendering rules are obtained. These rules include built-in display specifications, classification standards, detail requirements, and interaction logic corresponding to different data display boundaries. They cover prediction data classification methods, display dimensions, filtering functions, export permissions, and anomaly prompt rules. Based on these rules, corresponding view encapsulation formats are matched to the determined data display boundaries, achieving precise correspondence between permissions and display formats. A multi-level linked display format containing data from themselves and subordinate branches is matched for primary network points; a single-node independent display format is matched for branches; and a simplified core data display format is matched for service departments. Simultaneously, it adapts to the display requirements of multi-day prediction switching and anomaly prompts. Based on the view encapsulation format, the target prediction results are formatted and converted. Following the matched view encapsulation format, the display structure, level of detail, and presentation of the target prediction results are adjusted. For first-level branches, prediction results are displayed separately for themselves and their subordinates, including predicted shipment volume, fluctuation range, and prediction accuracy for each time point. Filtering by branch name and time point, as well as report export, are supported. For branches, prediction results are displayed separately for each time point, retaining core prediction information and supporting report export, while masking data from superiors and subordinates. Prediction results for service departments are simplified, highlighting core prediction values. A fast switching function for multi-day predictions is also implemented, allowing switching between prediction results for different time periods without re-invoking the model. Results with prediction accuracy below a preset threshold are anomaly flagged and accompanied by relevant prompts, ensuring that the displayed results comply with permission specifications while maintaining good readability, usability, and interactivity. The final display result is as follows.

[0041] By defining data display boundaries and matching exclusive view encapsulation formats, the compliance and security of data use are strictly guaranteed, effectively avoiding the risk of unauthorized data access and maintaining data management order. Differentiated view encapsulation formats and format conversion methods cater to the actual usage needs of users at different levels, simplifying unnecessary information interference, highlighting core content, and improving user viewing efficiency and user experience. The integration of multi-day forecast switching and anomaly alert functions not only facilitates users to quickly obtain forecast results for different time periods but also guides users to reasonably judge the usability of forecast results, improving the credibility of forecast results. The entire format conversion process is automatically completed based on preset rules without manual intervention, improving the generation efficiency and standardization of display results, achieving deep adaptation between target forecast results and user permissions, perfecting the closed loop of the entire forecasting method, and further enhancing its practicality and adaptability in the refined operation of the logistics industry.

[0042] The above describes the logistics shipment volume prediction method in the embodiments of the present invention. The following describes the logistics shipment volume prediction device in the embodiments of the present invention. Please refer to [link / reference]. Figure 8 One embodiment of the logistics shipment volume prediction device in this invention includes: Data acquisition module 801: used to acquire login role information, conversational query information, and pre-trained time series prediction models; Identifier generation module 802: Used to generate permission identifiers based on the login role information; Parsing module 803: used to perform natural language parsing on the colloquial query information to obtain target parsing parameters; Data association module 804: used to perform hierarchical data association based on the permission identifier and the target parsing parameters to obtain a standardized historical dataset; Prediction module 805: used to input the standardized historical dataset into the time series prediction model to perform offline shipment volume prediction and obtain the target prediction result; Display module 806: Used to convert the format of the target prediction result based on the permission identifier to obtain the display result.

[0043] Based on the same ideas as the methods in the above embodiments, the apparatus provided in this application can implement the methods in the above embodiments.

[0044] above Figure 8 The logistics shipment volume prediction device in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The logistics shipment volume prediction device in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0045] Figure 9This is a schematic diagram of the structure of a logistics shipment volume prediction device 900 provided in an embodiment of the present invention. The logistics shipment volume prediction device 900 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 910 (e.g., one or more processors) and a memory 920, and one or more storage media 930 (e.g., one or more mass storage devices) storing application programs 933 or data 932. The memory 920 and storage media 930 can be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the logistics shipment volume prediction device 900. Furthermore, the processor 910 may be configured to communicate with the storage media 930 and execute the series of instruction operations in the storage media 930 on the logistics shipment volume prediction device 900 to implement the steps of the logistics shipment volume prediction method provided in the above-described method embodiments.

[0046] The logistics shipment volume prediction device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating systems 931, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 9 The illustrated structure of the logistics shipment volume prediction device does not constitute a limitation on the logistics shipment volume prediction device. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0047] 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 shipment volume prediction method.

[0048] 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.

[0049] 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 the present 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 described in the various embodiments of the present 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.

[0050] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. 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 method for predicting the volume of shipments in logistics, characterized in that, include: Obtain login role information, conversational query information, and a pre-trained time-series prediction model; Generate an access identifier based on the login role information; Natural language parsing is performed on the colloquial query information to obtain the target parsing parameters; Based on the permission identifier and the target parsing parameters, hierarchical data association is performed to obtain a standardized historical dataset; The standardized historical dataset is input into the time-series prediction model to perform offline shipment volume prediction and obtain the target prediction result. The target prediction result is formatted based on the permission identifier to obtain the display result.

2. The logistics shipment volume prediction method according to claim 1, characterized in that, The generation of permission identifiers based on the login role information includes: Retrieve the preset permission rule library; Based on the permission rule base, the network level permission range corresponding to the login role is matched; The network point level permission scope is hash-encrypted using a hash encryption algorithm to obtain the permission identifier, and the permission identifier is cached.

3. The logistics shipment volume prediction method according to claim 1, characterized in that, The step of performing natural language parsing on the colloquial query information to obtain target parsing parameters includes: The colloquial query information was converted into parsable text information using a parsable text conversion tool. The pre-trained semantic parsing model is invoked to perform semantic parsing on the parsable text information to obtain initial parsing parameters; Obtain preset verification rules, and perform integrity verification on the initial parsing parameters based on the verification rules to obtain the target parsing parameters.

4. The logistics shipment volume prediction method according to claim 3, characterized in that, The semantic parsing model includes a text segmentation and annotation module, a semantic matching module, a core parameter extraction module, and a parameter format conversion module, which are sequentially connected. The pre-trained semantic parsing model is invoked to perform semantic parsing processing on the parsable text information to obtain initial parsing parameters, including: Based on the text segmentation and annotation module, the parsable text information is segmented and annotated to obtain annotation information; Based on the semantic matching module, semantic matching processing is performed on the labeled information to obtain matching information; Based on the core parameter extraction module, the matching information is processed to extract core parameters, resulting in time parameters and network point parameters. The parameter format conversion module performs format conversion processing on the time parameter and the network point parameter respectively to obtain the initial parsing parameters.

5. The logistics shipment volume prediction method according to claim 1, characterized in that, The target parsing parameters include target time parsing parameters and target network point parsing parameters; The process of performing hierarchical data association based on the permission identifier and the target parsing parameters to obtain a standardized historical dataset includes: Retrieve the preset hierarchical membership rules; The queryable network level is determined based on the permission identifier, and the query network node is determined based on the target network parsing parameters; Based on the hierarchical membership rules, hierarchical data association is performed according to the queryable network point hierarchy and the query network point node to obtain the historical dataset corresponding to the queryable network point hierarchy and the query network point node; An anomaly detection algorithm is used to remove abnormal data from the historical dataset to obtain a preprocessed dataset. Historical contemporaneous datasets corresponding to the target time parsing parameters are selected from the preprocessed dataset, and the historical contemporaneous datasets are converted into a standard format using a structured coding algorithm to obtain the standardized historical dataset.

6. The logistics shipment volume prediction method according to claim 1, characterized in that, The time-series prediction model includes a scenario-based feature selection module, an offline prediction module, and a correction module, which are sequentially connected. The step of inputting the standardized historical dataset into the time-series prediction model for offline shipment volume prediction to obtain the target prediction result includes: The standardized historical dataset is subjected to scenario-based feature filtering based on the scenario-based feature filtering module to obtain a feature dataset; The feature dataset is predicted offline based on the offline prediction module to obtain initial prediction results; The initial prediction result is corrected based on the correction module to obtain the target prediction result.

7. The logistics shipment volume prediction method according to claim 1, characterized in that, The process of converting the format of the target prediction result based on the permission identifier to obtain the display result includes: The data display boundaries are determined based on the permission identifier; Obtain preset view rendering rules, and match the corresponding view encapsulation format to the data display boundary based on the view rendering rules; The target prediction result is converted based on the view encapsulation format to obtain the display result.

8. A logistics shipment volume prediction device, characterized in that, include: Data acquisition module: used to acquire login role information, conversational query information, and pre-trained time series prediction models; Identifier generation module: used to generate permission identifiers based on the login role information; Parsing module: used to perform natural language parsing on the colloquial query information to obtain target parsing parameters; Data association module: used to perform hierarchical data association based on the permission identifier and the target parsing parameters to obtain a standardized historical dataset; Prediction module: used to input the standardized historical dataset into the time series prediction model to perform offline shipment volume prediction and obtain the target prediction result; Display module: Used to convert the format of the target prediction result based on the permission identifier to obtain the display result.

9. A logistics shipment volume prediction device, characterized in that, The logistics shipment volume prediction device includes: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the logistics shipment volume prediction device to perform the various steps of the logistics shipment volume prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the logistics shipment volume prediction method as described in any one of claims 1-7.

Images