Method and system for automatically generating and processing warehouse outbound information

Abstract

The present invention provides a method and system for automatically generating and processing warehouse outbound information, which relates to the technical field of warehouse management, including obtaining material information and an outbound order of materials to be outbound in a warehouse, extracting first associated information corresponding to a material number from a material basic information database, and extracting outbound logistics information and customer information of historical outbound orders at the same time; intelligently selecting an optimal outbound order template from an outbound order template database according to the outbound logistics information and customer information, generating an initial outbound order, verifying the initial outbound order using an intelligent verification algorithm, and automatically correcting and optimizing erroneous information in the initial outbound order according to the verification result; based on the optimized initial outbound order, obtaining corresponding material unit price information from a dynamic pricing database, and filling it into corresponding fields, generating a complete outbound order, transmitting the complete outbound order to an intelligent printing terminal, outputting a two-dimensional coded paper outbound order, triggering an outbound operation, and automatically generating outbound information.

Description

Technical Field

[0001] The present invention relates to warehouse management technology, and in particular to a method and system for automatically generating and processing warehouse outbound information. Background Art

[0002] The traditional warehouse outbound process usually requires manual filling in of various information on the outbound order, printing of paper outbound orders, and manual binding of the outbound order with the physical object and outbound operations. The whole process is inefficient and prone to errors. With the rapid development of the logistics industry, enterprises have higher and higher requirements for outbound efficiency and accuracy, and an automated outbound information generation and processing method and system are urgently needed.

[0003] In the existing technology, attempts are made to generate outbound orders based on templates, but template selection and information filling still require a lot of manual participation, and lack intelligent verification and optimization functions, so the quality of outbound orders cannot be guaranteed. There are also pricing optimization models introduced, but they are mainly aimed at the overall price strategy and cannot meet the needs of personalized and dynamic pricing. In addition, after the outbound order is generated, it still needs to be manually bound to the physical object, and the degree of automation is not high. In view of the shortcomings of the existing technology, we propose an innovative method and system for automatically generating and processing warehouse outbound information. This method intelligently selects the optimal outbound order template, uses a multi-dimensional intelligent verification algorithm to automatically correct and optimize the initial outbound order, and then realizes dynamic pricing based on the time series prediction model, personalized pricing model and multi-objective optimization algorithm, and finally generates a complete high-quality outbound order. Summary of the invention

[0004] The embodiment of the present invention provides a method and system for automatically generating and processing warehouse outbound information, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention,

[0006] A method for automatically generating and processing warehouse outbound information is provided, comprising:

[0007] Obtain material information and delivery orders of materials to be shipped out of the warehouse, wherein the material information includes the material number, material name and material quantity, and the delivery order includes the delivery order number and customer information; extract the first associated information corresponding to the material number from the material basic information database, and extract the delivery logistics information and customer information of the historical delivery order;

[0008] According to the outbound logistics information and customer information, the optimal outbound order template is intelligently selected from the outbound order template database, and the obtained material information, the first associated information, and the outbound logistics information are filled into the corresponding fields of the optimal outbound order template to generate an initial outbound order, and the initial outbound order is verified by using an intelligent verification algorithm, and the error information in the initial outbound order is automatically corrected and optimized according to the verification result;

[0009] Based on the optimized initial delivery order, the corresponding material unit price information is obtained from the dynamic pricing database and filled into the corresponding fields to generate a complete delivery order. The complete delivery order is transmitted to the intelligent printing terminal, and a QR-coded paper delivery order is output. The QR code of the paper delivery order is automatically identified by the delivery scanning system, and the paper delivery order is bound to the physical material, triggering the delivery operation and automatically generating delivery information.

[0010] In an optional embodiment,

[0011] According to the outbound logistics information and customer information, the optimal outbound order template is intelligently selected from the outbound order template database, including:

[0012] The outbound order template selection problem is modeled as a Markov decision process, which includes a state space, an action space, and a reward function, wherein the state space includes outbound information, and the action space is a set of candidate outbound order templates;

[0013] Construct a deep Q network, the input of which is a state vector, and the output is the Q value of each delivery order template. At the same time, a state transition data set is constructed, and each record in the state transition data set includes the current state, the selected action, the reward obtained, and the next state. A batch of samples are randomly selected from the state transition data set, and the target Q value is calculated based on the time difference algorithm. A loss function is constructed according to the target Q value and the current Q value, and the parameters of the deep Q network are updated based on the gradient descent method to obtain a trained deep Q network;

[0014] The outbound delivery scenario is confirmed based on the outbound logistics information and customer information, and the state vector corresponding to the outbound delivery scenario is input into the trained deep Q network. The Q value of each candidate outbound delivery order template is calculated through forward propagation, and the candidate outbound delivery order template with the largest Q value is selected as the optimal outbound delivery order template.

[0015] In an optional embodiment,

[0016] The state vector corresponding to the outbound delivery scenario is input into the trained deep Q network, and the Q value of each candidate outbound delivery order template is calculated by forward propagation. The method of selecting the candidate outbound delivery order template with the largest Q value as the optimal outbound delivery order template also includes:

[0017] Generate an actionable action mask vector according to the current state, the actionable action mask vector is used to represent the feasibility of each candidate outbound order template, and multiply the actionable action mask vector by the Q value vector output by the corresponding deep Q network element by element to obtain a masked Q value vector;

[0018] The template combination of multiple delivery order templates is used as the action space of the deep Q network for modeling. The complexity of the combination space is controlled by setting the maximum length of the template combination. The reward function is designed according to the overall effect of the template combination, so that the deep Q network can learn the long-term value of the template combination.

[0019] Based on the masked Q value vector and the long-term value of the template combination, the template combination with the largest Q value is selected as the optimal decision. When the optimal decision is not empty, each outbound order template in the optimal decision is applied to the outbound request in turn to generate the final outbound order.

[0020] Real-time monitoring of environmental feedback during the execution of the final delivery order. When the actual effect deviates from the expected result, the current delivery task is stopped, and the deep Q network is used to re-evaluate the status and generate a new decision.

[0021] A manual control trigger mechanism is set up. When the uncertainty of the new decision is higher than the preset threshold, a prompt is issued to the operator and the system suggestion is displayed. The template combination selected by the operator based on professional judgment is used as a manual decision sample, and the new decision generated by the system is used as an exploration sample. It is added to the experience replay pool, and the deep Q network is incrementally trained to finally form an adaptive intelligent outbound order template selection plan.

[0022] In an optional embodiment,

[0023] Using an intelligent verification algorithm to verify the initial delivery order, and automatically correcting and optimizing the error information in the initial delivery order according to the verification result includes:

[0024] Convert business rules and common sense constraints into formal representations, perform conflict detection on the formally represented rules, obtain a consistent rule set, design a rule engine, use a rule matching algorithm to perform compliance checks on the initial outbound delivery order based on the consistent rule set, and output the compliance check results of the rule engine;

[0025] At the same time, natural language processing technology is introduced to perform semantic analysis on the keywords in the initial delivery order, and combined with the domain dictionary and knowledge base, typos are identified and corrected, and the results of typos identification and semantic ambiguity resolution are output;

[0026] Construct the logical dependency graph of the initial delivery order, abstract the logical dependency relationship between the delivery order fields into a directed graph, use graph reasoning technology to reason on the logical dependency graph, combine with the association rule mining algorithm to mine the implicit logical relationship between the fields, and output the logical consistency check results between the fields based on the reasoning results and the mined implicit logical relationship;

[0027] Based on the compliance check results of the rule engine, the results of typo recognition and semantic ambiguity resolution, and the logical consistency check results between fields, multi-dimensional verification results are generated. Error information is located and visualized based on the multi-dimensional verification results, and corresponding automatic correction or optimization suggestions are given based on the error type and context information.

[0028] In an optional embodiment,

[0029] Locating and visualizing error information based on multi-dimensional verification results includes:

[0030] Mark errors in historical delivery order data to form an error location and annotation dataset;

[0031] Based on the error location annotation dataset, an end-to-end error location model is built. The error location model uses a bidirectional long short-term memory network as a feature extractor to automatically extract deep semantic features of the outbound order data. At the same time, a conditional random field model is introduced in the decoding layer, an attention mechanism is introduced in the hidden layer, and the embedding layer is initialized using a pre-trained language model;

[0032] The error location annotation dataset is divided into training set, validation set and test set, and the error location model is trained using the gradient descent algorithm. The hyperparameters are automatically tuned using the heuristic search algorithm to obtain a trained target error location model.

[0033] The target error location model is integrated into the intelligent verification process, and the multi-dimensional verification results are used as the input of the target error location model. The local verification results are combined to make a global prediction of the error location and type, and the error information location results are obtained, which are visualized by the front-end visualization tool.

[0034] In an optional embodiment,

[0035] Based on the optimized initial delivery order, the corresponding material unit price information is obtained from the dynamic pricing database and filled into the corresponding fields. The complete delivery order is generated including:

[0036] According to the historical price series of materials, multi-scale time-frequency features reflecting periodic laws are extracted, and the spatiotemporal feature matrix is ​​constructed by combining holidays and emergencies.

[0037] The spatiotemporal feature matrix is ​​input into a pre-built time series prediction model of an encoder-decoder structure, and the semantic information of different time steps is aggregated through an attention mechanism to generate a price prediction sequence in a future time window;

[0038] Obtain the profile features of the customers to be priced, divide the customer groups through clustering algorithms, explore the key pricing influencing factors of different customer groups, and generate a differentiated pricing rule base;

[0039] Input the customer's profile features into the pre-built personalized pricing model to predict the customer's acceptance probability of different prices. Combined with the differentiated pricing rule library, the customer pricing combination strategy is solved through a multi-objective optimization algorithm.

[0040] According to the shipment date in the optimized initial shipment order, the predicted price of the corresponding date is queried from the price prediction sequence, and according to the customer information in the optimized initial shipment order, the corresponding preferential plan is matched in the customer pricing combination strategy;

[0041] The predicted price and preferential scheme are input into the rule engine. After the pricing strategy is executed, the adjusted material unit price is filled into the corresponding field of the outbound delivery note to generate a complete dynamic pricing outbound delivery note.

[0042] In an optional embodiment,

[0043] Input the customer's profile features into the pre-built personalized pricing model to predict the customer's acceptance probability of different prices. Combined with the differentiated pricing rule base, the customer pricing combination strategy is solved through a multi-objective optimization algorithm, including:

[0044] Based on the differentiated pricing rule base, construct a multi-objective optimization problem, with maximizing customer lifetime value, maximizing short-term profits, and minimizing customer churn rate as the optimization objective function; define decision variables based on the differentiated pricing rule base and specific pricing strategies; determine constraints based on business needs and strategy restrictions;

[0045] Based on the determined optimization objective function, decision variables and constraints, a multi-objective optimization model is constructed, and the optimal customer pricing combination strategy is obtained by solving the multi-objective optimization model;

[0046] The steps of solving the multi-objective optimization model include:

[0047] An improved genetic algorithm is used to select, crossover and mutate the population to obtain a new population, and individuals in the new population are updated. The position and speed of the individuals are mapped to the solution space, and the pricing strategy of the individuals is updated accordingly. The optimal solution is selected from the populations of all generations through the elite selection strategy as the output of the customer pricing combination strategy.

[0048] Among them, the calculation formula of the optimization objective function is as follows:

[0049]

[0050] Among them, f(x) represents the optimization objective function, ω1 represents the weight coefficient of customer lifetime value, n represents the number of customers, V i (x i ) represents the lifetime value of the i-th customer, Ci (x i ) represents the short-term cost of the customer, ω2 represents the weight coefficient of the customer's short-term cost, ω3 represents the weight coefficient of the customer's churn rate, P i (x i ) represents the churn rate of the i-th customer.

[0051] According to a second aspect of the embodiments of the present invention,

[0052] A system for automatically generating and processing warehouse outbound information is provided, comprising:

[0053] The first unit is used to obtain material information and delivery orders of materials to be shipped out of the warehouse, wherein the material information includes material number, material name and material quantity, and the delivery order includes delivery order number and customer information; extract the first associated information corresponding to the material number from the material basic information database, and extract the delivery logistics information and customer information of the historical delivery order;

[0054] The second unit is used to intelligently select the optimal outbound order template from the outbound order template database according to the outbound logistics information and the customer information, and fill the obtained material information, the first associated information, and the outbound logistics information into the corresponding fields of the optimal outbound order template to generate an initial outbound order, verify the initial outbound order using an intelligent verification algorithm, and automatically correct and optimize the erroneous information in the initial outbound order according to the verification result;

[0055] The third unit is used to obtain the corresponding material unit price information from the dynamic pricing database based on the optimized initial delivery order, fill it into the corresponding field, generate a complete delivery order, transmit the complete delivery order to the intelligent printing terminal, output a QR-coded paper delivery order, automatically identify the QR code of the paper delivery order through the delivery scanning system, bind the paper delivery order with the physical material, trigger the delivery operation and automatically generate delivery information.

[0056] According to a third aspect of the embodiments of the present invention,

[0057] An electronic device is provided, comprising:

[0058] processor;

[0059] a memory for storing processor-executable instructions;

[0060] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0061] A fourth aspect of the embodiments of the present invention is:

[0062] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0063] In this embodiment, a multi-source heterogeneous information fusion extraction method is used to extract multi-dimensional related information such as basic information of materials, historical outbound information, customer preferences, etc., to provide rich contextual semantic information for subsequent intelligent processing. Design a template optimization algorithm for outbound orders based on reinforcement learning. By analyzing massive historical outbound scenes, autonomous learning and optimization of outbound decisions, intelligent selection of the best outbound order template, and improvement of the quality of outbound order generation. Design a template optimization algorithm for outbound orders based on reinforcement learning. By analyzing massive historical outbound scenes, autonomous learning and optimization of outbound decisions, intelligent selection of the best outbound order template, and improvement of the quality of outbound order generation. Integrate the dynamic pricing model with the outbound order generation process, obtain the optimal material price in real time according to influencing factors such as outbound time and customer level, and generate a complete and accurate outbound order in an integrated manner. Integrate QR code and code scanning technology to innovatively realize the automatic matching and binding of outbound orders and physical materials, synchronously trigger outbound operations and information generation, open up information flow and logistics, and realize end-to-end automated and intelligent outbound. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a flowchart of a method for automatically generating and processing warehouse outbound information according to an embodiment of the present invention;

[0065] Figure 2 It is a schematic diagram of the structure of the automatic generation and processing system of warehouse outbound information according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0068] Figure 1 FIG. 1 is a flow chart of a method for automatically generating warehouse outbound information according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0069] S101. Obtain material information and delivery orders of materials to be shipped out of the warehouse, wherein the material information includes material number, material name and material quantity, and the delivery order includes delivery order number and customer information; extract the first associated information corresponding to the material number from the material basic information database, and extract the delivery logistics information and customer information of the historical delivery order.

[0070] Specifically, it provides a standardized outbound task interface to external ERP systems, order systems, etc., and receives outbound task orders automatically created by external systems. When a new outbound task order is received, the intelligent outbound order generation process is immediately triggered. According to the material number provided in the outbound task order, the predefined ETL rules are used to automatically extract relevant information from the material basic information library, outbound history library, and customer information library.

[0071] S102. According to the outbound logistics information and customer information, the optimal outbound order template is intelligently selected from the outbound order template database, and the obtained material information, first related information, and outbound logistics information are filled into the corresponding fields of the optimal outbound order template to generate an initial outbound order. The initial outbound order is verified using an intelligent verification algorithm, and the erroneous information in the initial outbound order is automatically corrected and optimized according to the verification result.

[0072] In an optional embodiment,

[0073] According to the outbound logistics information and customer information, the optimal outbound order template is intelligently selected from the outbound order template database, including:

[0074] The outbound order template selection problem is modeled as a Markov decision process, which includes a state space, an action space, and a reward function, wherein the state space includes outbound information, and the action space is a set of candidate outbound order templates;

[0075] Construct a deep Q network, the input of which is a state vector, and the output is the Q value of each delivery order template. At the same time, a state transition data set is constructed, and each record in the state transition data set includes the current state, the selected action, the reward obtained, and the next state. A batch of samples are randomly selected from the state transition data set, and the target Q value is calculated based on the time difference algorithm. A loss function is constructed according to the target Q value and the current Q value, and the parameters of the deep Q network are updated based on the gradient descent method to obtain a trained deep Q network;

[0076] The outbound delivery scenario is confirmed based on the outbound logistics information and customer information, and the state vector corresponding to the outbound delivery scenario is input into the trained deep Q network. The Q value of each candidate outbound delivery order template is calculated through forward propagation, and the candidate outbound delivery order template with the largest Q value is selected as the optimal outbound delivery order template.

[0077] Exemplarily, first, the outbound order template selection problem is formalized into a Markov decision process, which is defined by a five-tuple, where the state space represents the feature set of the outbound scene, including attributes such as outbound material type, quantity, customer level, and urgency. The action space represents the set of candidate outbound order templates, and each action corresponds to a template. The transition probability represents the probability of transitioning to the next state after selecting an action in a certain state. For the outbound order selection problem, the next state is usually random and can be ignored. The reward function represents the immediate reward value obtained by selecting an action in a certain state. The reward function can be designed according to the matching degree and generation efficiency of the selected template. The discount factor represents the degree of attenuation of future rewards, and its value range is [0, 1]. The goal of the Markov decision process is to find an optimal strategy that maximizes the long-term cumulative reward for selecting actions in each state.

[0078] Using the deep learning method, a deep Q network is constructed as an approximation of the optimal state-action value function. The input of the deep Q network is the state vector in the state space, and the output is the Q value of each outbound order template in the action space. The network structure of the deep Q network can be selected from a multi-layer fully connected network, a convolutional neural network or a recurrent neural network. For example, the historical demand sequence of outbound materials is input into the convolutional neural network to capture demand trends from dynamic changes. Or the unstructured information such as images and text descriptions of materials is input into the recurrent neural network to automatically learn abstract representations.

[0079] The deep Q network is trained using the Experience Replay mechanism. Specifically, a state transition dataset is constructed based on the actual outbound data. Each record includes the current state, the selected action, the reward obtained, and the next state. A batch of samples are randomly selected from the state transition dataset. For each sample, the target Q value is calculated using the temporal difference algorithm (TD algorithm). The target Q value represents the expected long-term cumulative reward obtained after selecting and executing an action in the current state. A loss function is constructed based on the target Q value and the current Q value. The loss function represents the gap between the predicted Q value and the target Q value, which is used to measure the prediction error of the network. The commonly used loss function is the mean square error (MSE) loss function. Furthermore, the parameters of the deep Q network are updated based on the gradient descent method to reduce the loss function, thereby improving the prediction accuracy of the network. The gradient is calculated and the network parameters are updated by the back propagation algorithm. The above steps are repeated until the preset number of training rounds is reached or the convergence condition is reached. In each round of training, the network parameters are continuously updated by randomly sampling different samples to gradually optimize the performance of the network.

[0080] Determine the outbound scenario based on the actual outbound logistics information and customer information. Input the state vector corresponding to the outbound scenario into the trained deep Q network, and calculate the Q value of each candidate outbound order template through forward propagation. Based on the calculated Q value, select the candidate outbound order template with the largest Q value as the optimal outbound order template. Generate the corresponding outbound order based on the optimal outbound order template.

[0081] In this embodiment, by automating and intelligentizing the selection process of the delivery order template, manual intervention and subjective judgment are reduced, and the efficiency and accuracy of the selection are improved. By adopting the reinforcement learning method, by constructing a state transition data set and calculating the target Q value based on the time difference algorithm, the decision-making ability of the model can be continuously optimized to make the selection more accurate. By outputting the Q value of each candidate delivery order template, the decision-making process of the model has a certain degree of transparency, which is easy to analyze and explain. The selection process of the delivery order template is digitized and intelligent, which improves the accuracy, flexibility and scalability of the selection.

[0082] In an optional embodiment,

[0083] The state vector corresponding to the outbound delivery scenario is input into the trained deep Q network, and the Q value of each candidate outbound delivery order template is calculated by forward propagation. The method of selecting the candidate outbound delivery order template with the largest Q value as the optimal outbound delivery order template also includes:

[0084] Generate an actionable action mask vector according to the current state, the actionable action mask vector is used to represent the feasibility of each candidate outbound order template, and multiply the actionable action mask vector by the Q value vector output by the corresponding deep Q network element by element to obtain a masked Q value vector;

[0085] The template combination of multiple delivery order templates is used as the action space of the deep Q network for modeling. The complexity of the combination space is controlled by setting the maximum length of the template combination. The reward function is designed according to the overall effect of the template combination, so that the deep Q network can learn the long-term value of the template combination.

[0086] Based on the masked Q value vector and the long-term value of the template combination, the template combination with the largest Q value is selected as the optimal decision. When the optimal decision is not empty, each outbound order template in the optimal decision is applied to the outbound request in turn to generate the final outbound order.

[0087] Real-time monitoring of environmental feedback during the execution of the final delivery order. When the actual effect deviates from the expected result, the current delivery task is stopped, and the deep Q network is used to re-evaluate the status and generate a new decision.

[0088] A manual control trigger mechanism is set up. When the uncertainty of the new decision is higher than the preset threshold, a prompt is issued to the operator and the system suggestion is displayed. The template combination selected by the operator based on professional judgment is used as a manual decision sample, and the new decision generated by the system is used as an exploration sample. It is added to the experience replay pool, and the deep Q network is incrementally trained to finally form an adaptive intelligent outbound order template selection plan.

[0089] Exemplarily, based on the current state, a feasible action mask vector of the same size as the action space is generated. This vector is used to represent the feasibility of each candidate delivery order template, where 1 indicates that the template is feasible and 0 indicates that the template is not feasible. The feasible action mask vector is applied to the Q value vector output by the deep Q network. For infeasible delivery order templates, their corresponding Q values ​​are set to a smaller negative infinity to exclude these infeasible templates. The template combination of multiple delivery order templates is modeled as the action space of the deep Q network. By setting the maximum length of the template combination, the complexity of the combination space can be controlled. Each combination consists of the indexes of multiple templates.

[0090] Based on the masked Q value vector and the long-term value of the template combination, the template combination with the largest Q value is selected as the optimal decision. Specifically, the comprehensive score of each template combination is calculated based on the masked Q value vector and the long-term value of the template combination. The comprehensive score can be the weighted sum of the Q value and the long-term value, or other comprehensive indicators. The template combination with the highest comprehensive score is selected from all template combinations as the optimal decision. If the optimal decision is not empty, that is, there is an optimal template combination, then each outbound order template in the optimal decision is applied to the outbound request in turn to generate the final outbound order. Real-time monitoring of environmental feedback during the execution of the final outbound order. If the actual effect deviates from the expected, stop the current outbound task, and use the deep Q network to re-evaluate the state and generate a new decision. At the same time, a manual control trigger mechanism is set. When the uncertainty of the new decision is higher than the preset threshold, a prompt is issued to the operator and the system suggestion is displayed. The template combination selected by the operator based on professional judgment is used as a manual decision sample, and the new decision generated by the system is used as an exploration sample, which is added to the experience replay pool to perform incremental training on the deep Q network.

[0091] In this embodiment, by introducing a feasible action mask vector, the scheme can constrain the feasibility of the selection of the delivery order template, avoid selecting an inappropriate template, and improve the rationality and accuracy of the decision. The combination of multiple delivery order templates is modeled as an action space, and the reward function is designed according to the overall effect of the template combination, so that the deep Q network can learn the long-term value of different template combinations, thereby optimizing the selection of template combinations. An artificial control trigger mechanism is set up to incorporate the operator's professional judgment into the decision-making process when the decision uncertainty is high, forming a human-machine collaborative decision-making model, and improving the interpretability and credibility of the decision. The function of intelligent delivery order template selection is further improved, and the accuracy, adaptability, interpretability and credibility of the decision are improved. At the same time, human-machine collaborative decision-making and continuous learning optimization of the model are realized, providing effective technical support for intelligent decision-making in complex delivery scenarios.

[0092] In an optional embodiment,

[0093] Using an intelligent verification algorithm to verify the initial delivery order, and automatically correcting and optimizing the error information in the initial delivery order according to the verification result includes:

[0094] Convert business rules and common sense constraints into formal representations, perform conflict detection on the formally represented rules, obtain a consistent rule set, design a rule engine, use a rule matching algorithm to perform compliance checks on the initial outbound delivery order based on the consistent rule set, and output the compliance check results of the rule engine;

[0095] At the same time, natural language processing technology is introduced to perform semantic analysis on the keywords in the initial delivery order, and combined with the domain dictionary and knowledge base, typos are identified and corrected, and the results of typos identification and semantic ambiguity resolution are output;

[0096] Construct the logical dependency graph of the initial delivery order, abstract the logical dependency relationship between the delivery order fields into a directed graph, use graph reasoning technology to reason on the logical dependency graph, combine with the association rule mining algorithm to mine the implicit logical relationship between the fields, and output the logical consistency check results between the fields based on the reasoning results and the mined implicit logical relationship;

[0097] Based on the compliance check results of the rule engine, the results of typo recognition and semantic ambiguity resolution, and the logical consistency check results between fields, multi-dimensional verification results are generated. Error information is located and visualized based on the multi-dimensional verification results, and corresponding automatic correction or optimization suggestions are given based on the error type and context information.

[0098] For example, first of all, it is necessary to comprehensively collect and organize the business rules in the field of warehousing and logistics. By communicating with domain experts and analyzing historical outbound order data, key business rules related to outbound order verification, such as required fields, numerical range constraints, format specifications, etc., are extracted. In addition to business rules, some common sense constraints need to be summarized. These constraints reflect the general logical relationship between the fields in the outbound order, such as date sequence, quantity and unit matching, etc. Common sense constraints can be obtained by analyzing domain knowledge and summarizing experience.

[0099] Convert the extracted business rules and common sense constraints into formal representations, such as first-order predicate logic, production rules, etc. Formal representation facilitates the storage, management, and reasoning of rules. For example, "IF the delivery date > current date THEN the delivery date is illegal". When formalizing the rules, it is necessary to check whether there are conflicts or inconsistencies in the rule set. Through rule conflict detection algorithms, such as the Rete algorithm, conflicts such as mutual exclusion, duplication, and loops are identified, and necessary conflict resolution is performed to ensure the validity and consistency of the rule set.

[0100] Furthermore, an efficient and scalable rule engine architecture is designed. The rule engine usually includes core components such as the rule base, fact base, and inference engine. The rule base stores formally expressed rules; the fact base stores the outbound order data to be verified; the inference engine is responsible for reasoning the facts according to the rules to obtain the verification results. Select appropriate rule matching algorithms, such as forward chain reasoning, backward chain reasoning, Rete algorithm, etc. According to the characteristics of outbound order verification, such as the number of rules, the real-time requirements of verification, etc., select an algorithm with better time and space complexity. For example, the Rete algorithm achieves efficient rule matching by constructing a rule network. Use programming languages ​​such as Java and Python to implement the rule engine. According to the designed architecture and the selected algorithm, write the core components of the rule engine, such as the rule compiler and inference engine. The rule engine needs to support dynamic loading, updating, and deletion of rules to cope with business changes.

[0101] Optimize the performance of the rule engine to improve the efficiency of outbound order verification. Optimization technologies such as rule compilation, parallel reasoning, and caching mechanisms can be used. At the same time, introduce natural language processing (NLP) technology to perform semantic analysis on key words such as material names and specification descriptions in outbound orders. NLP technology can help understand the meaning, part of speech, and dependency relationships of words, and provide semantic-level support for error identification. Commonly used NLP technologies include word segmentation, part of speech tagging, named entity recognition, and semantic role tagging. Build professional dictionaries and knowledge bases in the field of warehousing and logistics to provide domain-specific background knowledge for semantic understanding. The dictionary contains key words such as common material names, specifications, and models; the knowledge base stores domain knowledge such as the hierarchical relationship between materials and attribute descriptions. These resources can be constructed through domain literature analysis and expert knowledge acquisition.

[0102] Use NLP technology and domain dictionaries to identify typos in delivery orders. Use methods such as building confusion sets and calculating edit distance to find possible typos. Combined with features such as contextual semantics and glyph similarity, typos are automatically corrected or correction suggestions are given for manual confirmation. By analyzing the context and semantic roles of words, the actual meaning of ambiguous words can be determined. For example, the material name "screwdriver" may refer to tools or parts in different contexts and needs to be distinguished based on the context. Concept hierarchical relationships and attribute constraints in the knowledge base can provide strong support for ambiguity resolution.

[0103] The logical dependency relationship between the fields of the delivery order is abstracted into a directed graph, called a logical dependency graph. The nodes in the graph represent the fields of the delivery order, and the edges represent the dependencies between the fields, such as equal to, greater than, and contained. The logical dependency graph can be constructed by analyzing business rules and summarizing domain knowledge. Using graph reasoning technology, the logical dependency graph of the delivery order is inferred to check the logical consistency between the fields. Commonly used graph reasoning technologies include rule-based reasoning, constraint-based reasoning, and graph neural network-based reasoning. By applying reasoning rules or constraints on the logical dependency graph, logical contradictions and numerical inconsistencies in the delivery order can be discovered. Through association rule mining, decision tree learning and other methods, frequent patterns and dependency rules between the fields of the delivery order are discovered to supplement and improve the logical dependency graph. The novel and implicit logical relationships mined can help discover anomalies and errors in the delivery order.

[0104] Furthermore, a structured representation of the verification results is designed, including information such as error type, error field, error description, and correction suggestions. The verification results are stored and managed in this form to facilitate subsequent analysis and display. Storage solutions such as relational databases and NoSQL databases can be used to support flexible queries and retrieval. According to the verification results, the error information in the outbound order is accurately located. Visualization techniques such as heat maps and tags are used to intuitively display the error fields and content. Visualization helps users quickly understand the distribution and severity of errors and improve the efficiency of error identification and processing. For the identified errors, automatic correction or optimization suggestions are given. Using rule reasoning, machine learning and other technologies, possible correction solutions are generated based on the error type and context information. For errors that cannot be corrected automatically, optimization suggestions are given to guide users to make manual corrections.

[0105] In an optional embodiment,

[0106] Locating and visualizing error information based on multi-dimensional verification results includes:

[0107] Mark errors in historical delivery order data to form an error location and annotation dataset;

[0108] Based on the error location annotation dataset, an end-to-end error location model is built. The error location model uses a bidirectional long short-term memory network as a feature extractor to automatically extract deep semantic features of the outbound order data. At the same time, a conditional random field model is introduced in the decoding layer, an attention mechanism is introduced in the hidden layer, and the embedding layer is initialized using a pre-trained language model;

[0109] The error location annotation dataset is divided into training set, validation set and test set, and the error location model is trained using the gradient descent algorithm. The hyperparameters are automatically tuned using the heuristic search algorithm to obtain a trained target error location model.

[0110] The target error location model is integrated into the intelligent verification process, and the multi-dimensional verification results are used as the input of the target error location model. The local verification results are combined to make a global prediction of the error location and type, and the error information location results are obtained, which are visualized by the front-end visualization tool.

[0111] For example, first, a representative sample subset is selected from the massive historical outbound order data to finely annotate various types of errors. The annotation process can be crowdsourced and use artificial intelligence-assisted tools to improve annotation efficiency and quality. Through repeated iterations and cross-validation, a large-scale error location annotation dataset is finally formed. On the basis of the annotated dataset, a sequence annotation model based on deep learning is introduced, and a bidirectional long short-term memory network (BiLSTM) is used as a feature extractor to automatically learn the deep semantic features of the text sequence. At the same time, a conditional random field model is introduced in the decoding layer, which can consider the transition probability of adjacent error labels and improve the overall coherence of error location. By introducing an attention mechanism in the hidden layer of BiLSTM, the contextual information most relevant to the current error discrimination is dynamically focused, which enhances the model's discriminatory power. At the same time, a pre-trained language model such as BERT can be used to initialize the embedding layer of BiLSTM, introduce prior knowledge of large-scale corpus pre-training, and improve the generalization performance of the model.

[0112] Based on the constructed annotated dataset and the designed deep learning model, an end-to-end trainable error localization model is built. The training set, validation set, and test set are divided proportionally from the annotated dataset, and the model is trained using the gradient descent algorithm. A loss function suitable for the error localization task is selected, such as the cross entropy loss function. The loss function is used to measure the difference between the model's prediction results and the true annotations. The model is optimized through multiple iterations of training. In each iteration, the training set is divided into several small batches, the loss function is calculated on each small batch, and the model parameters are updated using the optimizer. In this way, the prediction ability of the model is gradually improved. By experimenting with different hyperparameter combinations, such as learning rate, batch size, etc., a heuristic search algorithm (such as grid search, random search, or Bayesian optimization) can be used to automatically search for the best hyperparameter combination, and finally a trained target error localization model is obtained.

[0113] Integrate the trained target error location model into the intelligent verification process. Use the multi-dimensional verification results as the input of the target error location model, and make a global prediction of the error location and type by combining the local verification results. Specifically, convert the multi-dimensional verification results into a feature representation acceptable to the model. This can include encoding the verification results into numerical features, text features, or other forms suitable for model input. For example, for numerical features, the original verification results can be directly used as input; for text features, the bag-of-words model or word embedding model can be used to convert the text verification results into vector representations. Use the error location model to predict each local verification result. This can be a classification task, such as predicting the error type; or a regression task, such as predicting the probability distribution of the error location. For each local verification result, the model will output a prediction result. Combine all local prediction results to make a global prediction of the error location and type. There are many ways to achieve global prediction, such as weighted averaging based on the confidence of each local prediction result, or using a voting mechanism to select the final error location and type. Post-process the global prediction results to further improve accuracy and consistency. For example, the conditional random field (CRF) model can be applied to constrain the error location to ensure the rationality and consistency of the global prediction results. Finally, the error information location results are visualized through the front-end visualization tool so that users can intuitively view the error location and type.

[0114] In this embodiment, by formalizing the business rules and common sense constraints, and designing the rule engine and rule matching algorithm, the initial outbound order can be checked for compliance to ensure that the outbound order meets the predefined business rules and constraints. By introducing natural language processing technology, combined with the domain dictionary and knowledge base, the keywords in the initial outbound order can be semantically analyzed, typos and semantic ambiguities can be identified and corrected, and the semantic accuracy of the outbound order can be improved. According to the multi-dimensional verification results, various error information in the outbound order can be accurately located and visualized, which is convenient for manual review and intervention. Through intelligent verification and automatic correction, various errors in the outbound order can be effectively reduced, the overall quality of the outbound order can be improved, thereby improving the outbound efficiency and user experience. By introducing an error location model based on deep learning, full use is made of technologies such as semantic understanding, sequence labeling and attention mechanism, and high-precision positioning and visual display of error information in the outbound order are achieved, further enhancing the ability and effect of intelligent verification, and having good practical value and application prospects.

[0115] S103. Based on the optimized initial delivery order, the corresponding material unit price information is obtained from the dynamic pricing database and filled into the corresponding fields to generate a complete delivery order. The complete delivery order is transmitted to the intelligent printing terminal, and a QR-coded paper delivery order is output. The QR code of the paper delivery order is automatically identified by the delivery scanning system, and the paper delivery order is bound to the physical material, triggering the delivery operation and automatically generating delivery information.

[0116] The dynamic pricing database provides real-time and accurate material unit price information for each outbound task. The real-time unit price information of outbound materials at the current outbound time and for specific customers is obtained from the dynamic pricing database and filled into the material unit price field of the optimized outbound order to form an outbound order with complete information.

[0117] After generating a complete electronic outbound order, the system automatically encodes the outbound order into a QR code, transmits it to the intelligent printing terminal, and generates a QR-coded paper outbound order. During the material picking, packing and boxing process, the outbound operator uses an outbound scanner to scan the QR code on the material box to match and bind the material with the paper outbound order. After the outbound execution is completed, the automatically collected outbound information is attached to the original electronic outbound order to form a detailed and complete outbound information set. Submit the data storage service and store the outbound information set persistently in the outbound history database, material inventory information database and other related databases for subsequent business review and analysis.

[0118] In an optional embodiment,

[0119] Based on the optimized initial delivery order, the corresponding material unit price information is obtained from the dynamic pricing database and filled into the corresponding fields. The complete delivery order is generated including:

[0120] According to the historical price series of materials, multi-scale time-frequency features reflecting periodic laws are extracted, and the spatiotemporal feature matrix is ​​constructed by combining holidays and emergencies.

[0121] The spatiotemporal feature matrix is ​​input into a pre-built time series prediction model of an encoder-decoder structure, and the semantic information of different time steps is aggregated through an attention mechanism to generate a price prediction sequence in a future time window;

[0122] Obtain the profile features of the customers to be priced, divide the customer groups through clustering algorithms, explore the key pricing influencing factors of different customer groups, and generate a differentiated pricing rule base;

[0123] Input the customer's profile features into the pre-built personalized pricing model to predict the customer's acceptance probability of different prices. Combined with the differentiated pricing rule library, the customer pricing combination strategy is solved through a multi-objective optimization algorithm.

[0124] According to the shipment date in the optimized initial shipment order, the predicted price of the corresponding date is queried from the price prediction sequence, and according to the customer information in the optimized initial shipment order, the corresponding preferential plan is matched in the customer pricing combination strategy;

[0125] The predicted price and preferential scheme are input into the rule engine. After the pricing strategy is executed, the adjusted material unit price is filled into the corresponding field of the outbound delivery note to generate a complete dynamic pricing outbound delivery note.

[0126] Exemplarily, the historical price sequence of the target material is obtained from the time series database, the missing values ​​are interpolated, and data preprocessing is performed, such as detrending and seasonal adjustment. Time-frequency analysis methods such as wavelet transform are used to extract the periodic characteristics of the material price sequence at different scales, such as daily, weekly, and monthly periodic components, to obtain a time-frequency feature matrix. Relevant information about holidays and emergencies is obtained from external data sources, such as holiday type, duration, severity of emergencies, and scope of impact, etc. This information is quantified and feature extracted, and the extracted time-frequency features are aligned with holiday and emergency features in time, and spliced ​​into a complete spatiotemporal feature matrix.

[0127] The encoder-decoder structure time series prediction model is trained using historical data. The encoder uses structures such as CNN or RNN to extract high-level semantic representations of spatiotemporal features, and the decoder uses a similar structure to generate future price sequences. An attention mechanism is introduced between the encoder and the decoder. By calculating the correlation between features at different time steps and the current time step, the importance of features is dynamically adjusted to achieve adaptive aggregation of key spatiotemporal information. The obtained spatiotemporal feature matrix is ​​input into the trained time series prediction model, and a price prediction sequence for a period of time in the future is generated through forward propagation.

[0128] Obtain various attribute characteristics of target customers from the company's internal CRM, ERP and other systems, such as historical transaction records, user portraits, credit ratings, etc., and clean and feature engineer these heterogeneous data. Use clustering algorithms such as K-means and DBSCAN to group customers, evaluate clustering effects through indicators such as silhouette coefficient and Calinski-Harabasz index, and adjust clustering parameters until they are optimal. For each customer group, use association rule mining, decision tree and other algorithms to analyze key pricing factors, such as price sensitivity, brand preference, etc., and summarize to form targeted pricing rules. Aggregate the pricing rules of each customer group into a differentiated pricing rule library as the basis for subsequent personalized pricing.

[0129] Use historical transaction data to train the personalized pricing model. The model input is the customer portrait features, and the output is the customer's acceptance probability of different prices. Common models include logistic regression, random forest, neural network, etc. Input the target customer's portrait features into the trained personalized pricing model to obtain the predicted value of their acceptance probability of different prices. Combined with the differentiated pricing rule library of customer groups, a multi-objective optimization model is constructed. The optimization objectives may include maximizing profits, maximizing sales, maximizing customer satisfaction, etc., and multiple objectives are converted into single-objective problems by weighted summation. Heuristic search algorithms such as genetic algorithms and particle swarm optimization can be used to solve the multi-objective optimization model to obtain the optimal pricing combination strategy for target customers, including base prices and various discounts.

[0130] Extract key fields such as the delivery date and customer information from the optimized initial delivery order. According to the delivery date, find the predicted price of the corresponding date in the generated price forecast sequence. According to the customer information, match the customer's preferential plan in the obtained customer pricing combination strategy, including the base price and various preferential discounts. Further, input the obtained predicted price and preferential plan into the rule engine, and the rule engine adjusts the material unit price based on the preset pricing strategy such as "base price × (1-preferential discount)". Fill the adjusted material unit price into the corresponding field of the optimized initial delivery order to complete the dynamic pricing of the delivery order. Output a complete dynamic pricing delivery order, including key information such as delivery date, customer information, material unit price, etc., which can be passed to the downstream system for subsequent processing.

[0131] In an optional embodiment,

[0132] Input the customer's profile features into the pre-built personalized pricing model to predict the customer's acceptance probability of different prices. Combined with the differentiated pricing rule base, the customer pricing combination strategy is solved through a multi-objective optimization algorithm, including:

[0133] Based on the differentiated pricing rule base, construct a multi-objective optimization problem, with maximizing customer lifetime value, maximizing short-term profits, and minimizing customer churn rate as the optimization objective function; define decision variables based on the differentiated pricing rule base and specific pricing strategies; determine constraints based on business needs and strategy restrictions;

[0134] Based on the determined optimization objective function, decision variables and constraints, a multi-objective optimization model is constructed, and the optimal customer pricing combination strategy is obtained by solving the multi-objective optimization model;

[0135] The steps of solving the multi-objective optimization model include:

[0136] An improved genetic algorithm is used to select, crossover and mutate the population to obtain a new population, and individuals in the new population are updated. The position and speed of the individuals are mapped to the solution space, and the pricing strategy of the individuals is updated accordingly. The optimal solution is selected from the populations of all generations through the elite selection strategy as the output of the customer pricing combination strategy.

[0137] Among them, the calculation formula of the optimization objective function is as follows:

[0138]

[0139] Among them, f(x) represents the optimization objective function, ω1 represents the weight coefficient of customer lifetime value, n represents the number of customers, V i (x i ) represents the lifetime value of the i-th customer, C i (x i ) represents the short-term cost of the customer, ω2 represents the weight coefficient of the customer's short-term cost, ω3 represents the weight coefficient of the customer's churn rate, P i (x i ) represents the churn rate of the i-th customer.

[0140] In this embodiment, by constructing a spatiotemporal feature matrix and inputting it into a time series prediction model of an encoder-decoder structure, the material price sequence in the future time window can be predicted, providing data support for dynamic pricing. The multi-scale time-frequency features reflecting the periodicity of material prices are extracted, and combined with holidays and emergency factors, the accuracy and stability of price prediction are improved. A personalized pricing model is constructed to predict the probability of customers accepting different prices, and combined with a differentiated pricing rule base, the customer pricing combination strategy is solved through a multi-objective optimization algorithm, and the personalization and optimization of pricing are achieved. The predicted price and preferential scheme are input into the rule engine, and after the pricing strategy is executed, the material unit price field in the outbound order can be automatically filled, which improves the automation and efficiency of the pricing process. By comprehensively considering multiple goals such as customer lifetime value, short-term profit and customer churn rate, more reasonable and preferential pricing schemes can be provided to customers, improving customer satisfaction and loyalty.

[0141] Figure 2 FIG. 1 is a schematic diagram of the structure of a system for automatically generating and processing warehouse outbound information according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0142] The first unit is used to obtain material information and delivery orders of materials to be shipped out of the warehouse, wherein the material information includes material number, material name and material quantity, and the delivery order includes delivery order number and customer information; extract the first associated information corresponding to the material number from the material basic information database, and extract the delivery logistics information and customer information of the historical delivery order;

[0143] The second unit is used to intelligently select the optimal outbound order template from the outbound order template database according to the outbound logistics information and the customer information, and fill the obtained material information, the first associated information, and the outbound logistics information into the corresponding fields of the optimal outbound order template to generate an initial outbound order, verify the initial outbound order using an intelligent verification algorithm, and automatically correct and optimize the erroneous information in the initial outbound order according to the verification result;

[0144] The third unit is used to obtain the corresponding material unit price information from the dynamic pricing database based on the optimized initial delivery order, fill it into the corresponding field, generate a complete delivery order, transmit the complete delivery order to the intelligent printing terminal, output a QR-coded paper delivery order, automatically identify the QR code of the paper delivery order through the delivery scanning system, bind the paper delivery order with the physical material, trigger the delivery operation and automatically generate delivery information.

[0145] According to a third aspect of the embodiments of the present invention,

[0146] An electronic device is provided, comprising:

[0147] processor;

[0148] a memory for storing processor-executable instructions;

[0149] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0150] A fourth aspect of the embodiments of the present invention is:

[0151] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0152] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically generating and processing warehouse outbound information, characterized in that: include: Obtain material information and delivery orders of materials to be shipped out of the warehouse, wherein the material information includes the material number, material name and material quantity, and the delivery order includes the delivery order number and customer information; extract the first associated information corresponding to the material number from the material basic information database, and extract the delivery logistics information and customer information of the historical delivery order; According to the outbound logistics information and customer information, the optimal outbound order template is intelligently selected from the outbound order template database, and the obtained material information, the first associated information, and the outbound logistics information are filled into the corresponding fields of the optimal outbound order template to generate an initial outbound order, and the initial outbound order is verified by using an intelligent verification algorithm, and the error information in the initial outbound order is automatically corrected and optimized according to the verification result; Based on the optimized initial delivery order, the corresponding material unit price information is obtained from the dynamic pricing database and filled into the corresponding fields to generate a complete delivery order. The complete delivery order is transmitted to the intelligent printing terminal, and a QR-coded paper delivery order is output. The QR code of the paper delivery order is automatically identified by the delivery scanning system, and the paper delivery order is bound to the physical material, triggering the delivery operation and automatically generating delivery information. Using an intelligent verification algorithm to verify the initial delivery order, and automatically correcting and optimizing the error information in the initial delivery order according to the verification result includes: Convert business rules and common sense constraints into formal representations, perform conflict detection on the formally represented rules, obtain a consistent rule set, design a rule engine, use a rule matching algorithm to perform compliance checks on the initial outbound delivery order based on the consistent rule set, and output the compliance check results of the rule engine; At the same time, natural language processing technology is introduced to perform semantic analysis on the keywords in the initial delivery order, and combined with the domain dictionary and knowledge base, typos are identified and corrected, and the results of typos identification and semantic ambiguity resolution are output; Construct the logical dependency graph of the initial delivery order, abstract the logical dependency relationship between the delivery order fields into a directed graph, use graph reasoning technology to reason on the logical dependency graph, combine with the association rule mining algorithm to mine the implicit logical relationship between the fields, and output the logical consistency check results between the fields based on the reasoning results and the mined implicit logical relationship; Generate multi-dimensional verification results based on the compliance check results of the rule engine, the results of typo recognition and semantic ambiguity resolution, and the results of logical consistency check between fields. Locate and visualize error information based on the multi-dimensional verification results, and provide corresponding automatic correction or optimization suggestions based on the error type and context information. Based on the optimized initial delivery order, the corresponding material unit price information is obtained from the dynamic pricing database and filled into the corresponding fields. The complete delivery order is generated including: Based on the historical price series of materials, multi-scale time-frequency features reflecting periodic laws are extracted, and the spatiotemporal feature matrix is ​​constructed by combining holidays and emergencies. The spatiotemporal feature matrix is ​​input into a pre-built time series prediction model of an encoder-decoder structure, and the semantic information of different time steps is aggregated through an attention mechanism to generate a price prediction sequence in a future time window; Obtain the profile features of the customers to be priced, divide the customer groups through clustering algorithms, explore the key pricing influencing factors of different customer groups, and generate a differentiated pricing rule base; Input the customer's profile features into the pre-built personalized pricing model to predict the customer's acceptance probability of different prices. Combined with the differentiated pricing rule library, the customer pricing combination strategy is solved through a multi-objective optimization algorithm. According to the shipment date in the optimized initial shipment order, the predicted price of the corresponding date is queried from the price prediction sequence, and according to the customer information in the optimized initial shipment order, the corresponding preferential plan is matched in the customer pricing combination strategy; The predicted price and preferential scheme are input into the rule engine. After the pricing strategy is executed, the adjusted material unit price is filled into the corresponding field of the outbound delivery note to generate a complete dynamic pricing outbound delivery note.

2. The method according to claim 1, characterized in that: According to the outbound logistics information and customer information, the optimal outbound order template is intelligently selected from the outbound order template database, including: The outbound order template selection problem is modeled as a Markov decision process, which includes a state space, an action space, and a reward function, wherein the state space includes outbound information, and the action space is a set of candidate outbound order templates; Construct a deep Q network, the input of which is a state vector, and the output is the Q value of each delivery order template. At the same time, a state transition data set is constructed, and each record in the state transition data set includes the current state, the selected action, the reward obtained, and the next state. A batch of samples are randomly selected from the state transition data set, and the target Q value is calculated based on the time difference algorithm. A loss function is constructed according to the target Q value and the current Q value, and the parameters of the deep Q network are updated based on the gradient descent method to obtain a trained deep Q network; The outbound delivery scenario is confirmed based on the outbound logistics information and customer information, and the state vector corresponding to the outbound delivery scenario is input into the trained deep Q network. The Q value of each candidate outbound delivery order template is calculated through forward propagation, and the candidate outbound delivery order template with the largest Q value is selected as the optimal outbound delivery order template.

3. The method according to claim 2, characterized in that The state vector corresponding to the outbound delivery scenario is input into the trained deep Q network, and the Q value of each candidate outbound delivery order template is calculated by forward propagation. The method of selecting the candidate outbound delivery order template with the largest Q value as the optimal outbound delivery order template also includes: Generate an actionable action mask vector according to the current state, the actionable action mask vector is used to represent the feasibility of each candidate outbound order template, and multiply the actionable action mask vector by the Q value vector output by the corresponding deep Q network element by element to obtain a masked Q value vector; The template combination of multiple delivery order templates is used as the action space of the deep Q network for modeling. The complexity of the combination space is controlled by setting the maximum length of the template combination. The reward function is designed according to the overall effect of the template combination, so that the deep Q network can learn the long-term value of the template combination. Based on the masked Q value vector and the long-term value of the template combination, the template combination with the largest Q value is selected as the optimal decision. When the optimal decision is not empty, each outbound order template in the optimal decision is applied to the outbound request in turn to generate the final outbound order. Real-time monitoring of environmental feedback during the execution of the final delivery order. When the actual effect deviates from the expected result, the current delivery task is stopped, and the deep Q network is used to re-evaluate the status and generate a new decision. A manual control trigger mechanism is set up. When the uncertainty of the new decision is higher than the preset threshold, a prompt is issued to the operator and the system suggestion is displayed. The template combination selected by the operator based on professional judgment is used as a manual decision sample, and the new decision generated by the system is used as an exploration sample. It is added to the experience replay pool, and the deep Q network is incrementally trained to finally form an adaptive intelligent outbound order template selection plan.

4. The method according to claim 1, characterized in that Locating and visualizing error information based on multi-dimensional verification results includes: Mark errors in historical delivery order data to form an error location and annotation dataset; Based on the error location annotation dataset, an end-to-end error location model is built. The error location model uses a bidirectional long short-term memory network as a feature extractor to automatically extract deep semantic features of the outbound order data. At the same time, a conditional random field model is introduced in the decoding layer, an attention mechanism is introduced in the hidden layer, and the embedding layer is initialized using a pre-trained language model; The error location annotation dataset is divided into training set, validation set and test set, and the error location model is trained using the gradient descent algorithm. The hyperparameters are automatically tuned using the heuristic search algorithm to obtain a trained target error location model. The target error location model is integrated into the intelligent verification process, and the multi-dimensional verification results are used as the input of the target error location model. The local verification results are combined to make a global prediction of the error location and type, and the error information location results are obtained, which are visualized by the front-end visualization tool.

5. The method according to claim 1, characterized in that Input the customer's profile features into the pre-built personalized pricing model to predict the customer's acceptance probability of different prices. Combined with the differentiated pricing rule base, the customer pricing combination strategy is solved through a multi-objective optimization algorithm, including: Based on the differentiated pricing rule base, construct a multi-objective optimization problem, and optimize the objective function with short-term profit and customer churn rate; define decision variables based on the differentiated pricing rule base and specific pricing strategies; determine constraints based on business needs and strategy restrictions; Based on the determined optimization objective function, decision variables and constraints, a multi-objective optimization model is constructed, and the optimal customer pricing combination strategy is obtained by solving the multi-objective optimization model; The steps of solving the multi-objective optimization model include: An improved genetic algorithm is used to select, crossover and mutate the population to obtain a new population, and individuals in the new population are updated. The position and speed of the individuals are mapped to the solution space, and the pricing strategy of the individuals is updated accordingly. The optimal solution is selected from the populations of all generations through the elite selection strategy as the output of the customer pricing combination strategy. Among them, the calculation formula of the optimization objective function is as follows: Among them, f(x) represents the optimization objective function, ω1 represents the weight coefficient of customer lifetime value, n represents the number of customers, V i (x i ) represents the lifetime value of the i-th customer, C i (x i ) represents the short-term cost of the customer, ω2 represents the weight coefficient of the customer's short-term cost, ω3 represents the weight coefficient of the customer's churn rate, P i (x i ) represents the churn rate of the i-th customer.

6. A system for automatically generating and processing warehouse outbound information, used to implement the method described in any one of claims 1 to 5, characterized in that: include: The first unit is used to obtain material information and delivery orders of materials to be shipped out of the warehouse, wherein the material information includes material number, material name and material quantity, and the delivery order includes delivery order number and customer information; extract the first associated information corresponding to the material number from the material basic information database, and extract the delivery logistics information and customer information of the historical delivery order; The second unit is used to intelligently select the optimal outbound order template from the outbound order template database according to the outbound logistics information and the customer information, and fill the obtained material information, the first associated information, and the outbound logistics information into the corresponding fields of the optimal outbound order template to generate an initial outbound order, verify the initial outbound order using an intelligent verification algorithm, and automatically correct and optimize the erroneous information in the initial outbound order according to the verification result; The third unit is used to obtain the corresponding material unit price information from the dynamic pricing database based on the optimized initial delivery order, fill it into the corresponding field, generate a complete delivery order, transmit the complete delivery order to the intelligent printing terminal, output a QR-coded paper delivery order, automatically identify the QR code of the paper delivery order through the delivery scanning system, bind the paper delivery order with the physical material, trigger the delivery operation and automatically generate delivery information.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images