Automatic sorting method and device based on natural language processing and storage medium
Through natural language processing technology, natural language orders are converted into structured data, combined with abnormal detection and clustering technology, picking strategies are generated, which solves the problem of low processing efficiency of natural language orders in the existing technology, realizes automatic picking, and improves warehousing Logistics processing capability.
Patent Information
- Application Number
- CN202411159016.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-22
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
When handling non-standardized and natural language forms of orders, the prior art relies on fixed data formats and rules to directly process order information in natural language forms, resulting in the need of manual intervention and reduce overall efficiency.
Using the automatic picking method based on natural language processing, by obtaining natural language orders input by customers, word segmentation, dependency extraction, named entity recognition and relationship filling are performed, and the selection is generated by combining abnormal detection and clustering technology. Wave-shot strategy and automatic picking is achieved through acoustic, photoelectric guidance technology.
It realizes automatic identification and processing of natural language orders, reduces manual intervention, improves picking efficiency, and can handle diversified, small batch, and high frequency order models.
Smart Images

Figure CN119026599A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of logistics and warehousing, and in particular to an automatic picking method based on natural language processing. Background Art
[0002] As globalization and networking become increasingly prevalent, warehousing logistics, as a key link in the circulation of goods, is directly related to the company's operating costs and customer satisfaction with its efficiency and accuracy. Although traditional warehouse management systems have gradually introduced automation technology in recent years, most of them are still limited by fixed operating procedures and identification technologies. Such systems may be able to meet basic needs when processing standardized, large-volume orders, but they are stretched when faced with diversified, small-batch, high-frequency modern order models.
[0003] In the field of warehousing and logistics management, existing automation technologies are mainly concentrated in the links of cargo identification, handling, storage and outbound delivery. For example, the cargo can be quickly and accurately identified through barcodes, RFID and other technologies; the cargo can be automatically handled by automated handling equipment such as AGV (Automated Guided Vehicle); the cargo can be densely stored and efficiently managed through high-bay warehouses and automated storage systems; and the cargo can be quickly sorted and outbound through automated sorting systems.
[0004] In addition, with the rapid development of technologies such as the Internet of Things, big data, and cloud computing, the level of informatization and intelligence in warehousing and logistics management has been significantly improved. However, these technologies are mainly concentrated in the execution level of logistics operations, while in the processing and management of order information, especially the processing of non-standardized, natural language order instructions, the existing technology still has obvious deficiencies. Although some advanced warehouse management systems have begun to try to introduce data analysis and artificial intelligence technologies to optimize the order processing process, most of these systems still rely on fixed data formats and rules to operate and cannot directly process order information in natural language form; manual intervention is required to convert and interpret information, which reduces overall efficiency. Summary of the invention
[0005] In view of this, the purpose of the present invention is to provide an automatic picking method based on natural language processing to solve the problem in the prior art that the automatic processing of information relies on fixed data formats and rules to operate, cannot directly process order information in natural language form, and requires manual intervention to convert and interpret the information, thereby reducing the overall efficiency.
[0006] According to a first aspect of an embodiment of the present invention, there is provided an automatic picking method based on natural language processing, the method comprising:
[0007] Get natural language orders entered by customers;
[0008] Performing a preprocessing operation on the natural language order, and performing a word segmentation operation on the preprocessed natural language order to segment the natural language order text into separate words; using the Stanford Parser tool to extract the dependency relationship between words, and extracting the phrase structure in the natural language order text through component syntactic analysis;
[0009] Using a pre-trained neural network model to identify named entities in the natural language order text, using a rule-based method or a deep learning-based method, combining dependencies between words and phrase structures to extract relationships between the named entities, and filling the named entities and the relationships between named entities into predefined slots, thereby converting the unstructured natural language order text into a structured data format;
[0010] Extracting preset features related to anomaly detection from structured natural language orders, inputting the extracted features related to anomaly detection into a pre-trained anomaly detection model, wherein the pre-trained anomaly detection model calculates an anomaly score or probability of each order according to the input features, and determines whether the order is abnormal according to a preset anomaly score or probability threshold;
[0011] The detected abnormal orders are classified according to the abnormal reasons, and the abnormal orders are corrected, cancelled or suspended according to the abnormal order classification results, and the current order queue is updated according to the abnormal order processing results;
[0012] Extract the features of all structured natural language orders in the updated order queue, and use the TF-IDF method to assign weights to the extracted features to form a feature vector; cluster all structured natural language orders in the current order queue according to the feature vector, group similar orders together, and obtain multiple order clusters;
[0013] Generate a picking wave strategy for each order cluster, determine the location of the required goods in the automated warehouse according to the picking wave strategy for each order cluster, and plan a picking path in the automated warehouse according to the location of the required goods in the automated warehouse;
[0014] Determine the picking path that needs to be executed, and use acoustic and optical guidance technology to help pickers or equipment quickly and accurately find the required goods in the picking path until the picking wave strategy for each order cluster is completed.
[0015] Preferably, it also includes:
[0016] The picking personnel or equipment inputs abnormal feedback information through voice or text in the picking path, and the abnormal feedback information is parsed through natural language processing technology to obtain the feedback intention of the picking personnel or equipment, and real-time path adjustment or abnormal handling process is executed according to the feedback intention.
[0017] Preferably, it also includes:
[0018] The generated picking path is converted into a natural language description using natural language processing technology.
[0019] include:
[0020] Inputting the original data of the picking path, extracting the key coordinate information in the original data, filling the key information into the selected path natural language template, and replacing the original placeholder in the path natural language template to obtain a structured picking path;
[0021] Add audio-visual guidance information to each coordinate point in the structured picking path.
[0022] Preferably,
[0023] The method of planning a picking path in the automated warehouse according to the location of the required goods in the automated warehouse includes:
[0024] Define data structure and path planning;
[0025] The definition data structure includes:
[0026] Set a starting point and an end point, where the end point is the location of the required goods in the automated warehouse;
[0027] Define each path node between the starting point and the end point, including: the coordinates of the current node, the g value of the current node, the h value of the current node, the f value of the current node, and the parent node pointer; the g value of the current node is the cost from the starting point to the current node, the h value of the current node is the heuristic estimated cost from the current node to the end point, and the f value of the current node is the h value of the current node plus the g value of the current node;
[0028] An open list and a closed list are set, wherein all path nodes are sorted according to the f value in the open list, and the node with a smaller f value has a higher priority for inspection; the closed list is used to store the path nodes that have been inspected;
[0029] Set the heuristic function to calculate the h value of the current node;
[0030] The path planning includes:
[0031] Remove the path node with the smallest f value from the open list as the current check node. For each neighbor node of the current check node, if the neighbor node is already in the closed list, ignore the neighbor node. If the neighbor node is not in the closed list, calculate the g value from the starting point to the neighbor node.
[0032] If the g value of the neighbor node is smaller than the g value of any path node already in the open list, then add the neighbor node to the open list and set the parent node pointer to point to the current check node;
[0033] Add the current check node to the closed list and reorder the open list according to the f value;
[0034] Repeat the above process until the current check node is the end point, and then backtrack through the parent node pointer to build the picking path.
[0035] Preferably,
[0036] The structured natural language orders in the current order queue are clustered according to the feature vectors, similar orders are grouped together, and multiple order clusters are obtained, including:
[0037] Select K initial centroids, where K is the preset number of clusters;
[0038] For each feature vector, calculate its distance to each centroid and assign it to the cluster with the closest centroid;
[0039] For each cluster assigned to a feature vector, calculate the average of all the feature vectors it contains and set the average as the new centroid;
[0040] The above process is repeated until the change in the centroid of each cluster is less than a certain threshold or the maximum number of iterations is reached, and multiple order clusters are obtained.
[0041] Preferably, it also includes:
[0042] After the required goods are picked, feature information in the structured natural language order or product description is extracted, one or more product labels are generated based on the feature information, and vector representations are generated for the one or more product labels respectively;
[0043] Based on one or more product labels, intelligent packaging suggestions are matched in a preset intelligent packaging suggestion library to generate final intelligent packaging suggestions and product labels.
[0044] Preferably, it also includes:
[0045] Obtain customer feedback on picking and delivery services through multiple channels, including text reviews, ratings, and pictures;
[0046] Clean the collected customer feedback, remove irrelevant information and noise, and use natural language processing technology to perform word segmentation, part-of-speech tagging and syntactic analysis on the feedback;
[0047] Through sentiment analysis algorithms, the sentiment tendency of each feedback opinion is judged based on part-of-speech tagging and syntactic analysis. The sentiment tendency includes: positive, negative and neutral;
[0048] Extract keywords from feedback and identify issues contained in customer feedback using a pre-trained LDA topic model;
[0049] Categorize the issues contained in the feedback and prioritize them based on the classification results, frequency of occurrence, and sentiment;
[0050] Optimize the processing of problems according to their priority.
[0051] Preferably, it also includes:
[0052] In the return and exchange process, obtain the return and exchange application information submitted by the customer;
[0053] Clean the return and exchange application information to remove irrelevant information and noise, and use natural language processing technology to perform word segmentation, part-of-speech tagging and syntactic analysis on the return and exchange application information;
[0054] Through the sentiment analysis algorithm, the sentiment tendency of each return application information is judged based on part-of-speech tagging and syntactic analysis, and the sentiment tendency includes: anger, disappointment and indifference;
[0055] The TF-IDF algorithm is used to extract keywords from the return and exchange application information, and the user's return reason is obtained based on the keywords;
[0056] Automatically match preset return and exchange response templates based on return reasons and sentiment tendencies.
[0057] According to a second aspect of an embodiment of the present invention, there is provided an automatic picking device based on natural language processing, the device comprising:
[0058] Order acquisition module: used to obtain natural language orders entered by customers;
[0059] Preprocessing module: used to perform preprocessing operation on the natural language order, perform word segmentation operation on the preprocessed natural language order, and segment the natural language order text into separate words; use Stanford Parser tool to extract dependency relationship between words, and extract phrase structure in the natural language order text through component syntactic analysis;
[0060] Natural language analysis module: used to identify named entities in the natural language order text using a pre-trained neural network model, extract the relationships between the named entities by combining the dependency relationship between words and phrase structure based on a rule-based method or a deep learning method, and fill the named entities and the relationships between named entities into predefined slots, thereby converting the unstructured natural language order text into a structured data format;
[0061] Abnormal order judgment module: used to extract preset features related to abnormality detection from structured natural language orders, input the extracted features related to abnormality detection into a pre-trained abnormality detection model, and the pre-trained abnormality detection model calculates the abnormality score or probability of each order according to the input features, and judges whether the order is abnormal according to the preset abnormality score or probability threshold;
[0062] Abnormal order processing module: used to classify detected abnormal orders according to abnormal reasons, modify, cancel or suspend abnormal orders according to abnormal order classification results, and update the current order queue according to abnormal order processing results;
[0063] Order classification module: used to extract the features of all structured natural language orders in the updated order queue, and use the TF-IDF method to assign weights to the extracted features to form feature vectors; cluster all structured natural language orders in the current order queue according to the feature vectors, group similar orders together, and obtain multiple order clusters;
[0064] Wave path generation module: used to generate a picking wave strategy for each order cluster, determine the location of the required goods in the automated warehouse according to the picking wave strategy for each order cluster, and plan the picking path in the automated warehouse according to the location of the required goods in the automated warehouse;
[0065] Picking execution module: used to determine the picking path that needs to be executed, and use acoustic and optical guidance technology to help pickers or equipment quickly and accurately find the required goods in the picking path until the picking wave strategy for each order cluster is completed.
[0066] According to a third aspect of an embodiment of the present invention, there is provided a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a main controller, each step in the above method is implemented.
[0067] The technical solution provided by the embodiments of the present invention may have the following beneficial effects:
[0068] The present application obtains the natural language order input by the customer, performs word segmentation and dependency extraction on the natural language order, then identifies the named entities in the natural language order text based on a pre-trained neural network model, extracts the relationships between the named entities, and fills the named entities and the relationships between the named entities into predefined slots, thereby converting the unstructured natural language order text into a structured data format, so that the present application can identify the user's natural language order, and at the same time, based on the natural language recognition results, judge and process abnormal orders, group batch orders, and execute wave picking strategies for orders in the same group, thereby saving the picking time of picking personnel and improving picking efficiency.
[0069] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0070] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0071] Figure 1 is a flowchart of an automatic picking method based on natural language processing according to an exemplary embodiment;
[0072] Figure 2 is a system schematic diagram of an automatic picking device based on natural language processing according to another exemplary embodiment;
[0073] In the attached figure: 1-order acquisition module, 2-preprocessing module, 3-natural language analysis module, 4-abnormal order judgment module, 5-abnormal order processing module, 6-order classification module, 7-wave path generation module, 8-picking execution module. DETAILED DESCRIPTION
[0074] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.
[0075] Embodiment 1
[0076] Figure 1 is a flowchart of an automatic picking method based on natural language processing according to an exemplary embodiment. Figure 1 As shown, the method includes:
[0077] S1, obtain the natural language order input by the customer;
[0078] S2, performing a preprocessing operation on the natural language order, performing a word segmentation operation on the preprocessed natural language order, so as to segment the natural language order text into separate words; using a Stanford Parser tool to extract dependency relationships between words, and extracting phrase structures in the natural language order text through component syntactic analysis;
[0079] S3, using a pre-trained neural network model to identify named entities in the natural language order text, extracting the relationships between the named entities by combining the dependency relationship between words and phrase structure based on a rule-based method or a deep learning method, and filling the named entities and the relationships between named entities into predefined slots, thereby converting the unstructured natural language order text into a structured data format;
[0080] S4, extracting preset features related to anomaly detection from the structured natural language order, inputting the extracted features related to anomaly detection into a pre-trained anomaly detection model, wherein the pre-trained anomaly detection model calculates an anomaly score or probability of each order according to the input features, and determines whether the order is abnormal according to a preset anomaly score or probability threshold;
[0081] S5, classifying the detected abnormal orders according to the abnormal reasons, modifying, canceling or suspending the abnormal orders according to the abnormal order classification results, and updating the current order queue according to the abnormal order processing results;
[0082] S6, extracting features of all structured natural language orders in the updated order queue, assigning weights to the extracted features using the TF-IDF method to form a feature vector; clustering all structured natural language orders in the current order queue according to the feature vector, grouping similar orders together, and obtaining multiple order clusters;
[0083] S7, generating a picking wave strategy for each order cluster, determining the location of the required goods in the automated warehouse according to the picking wave strategy for each order cluster, and planning a picking path in the automated warehouse according to the location of the required goods in the automated warehouse;
[0084] S8, determine the picking path to be executed, and use the acoustic and optical guidance technology to help the pickers or equipment quickly and accurately find the required goods in the picking path until the picking wave strategy for each order cluster is completed;
[0085] Understandably,
[0086] S1, the system efficiently receives and processes the diverse order information entered by customers in natural language through the user interface or flexible API interface. The system not only accurately captures core data such as product name and quantity, but also carefully identifies and records key details such as customers' special requirements and delivery addresses, thereby ensuring the integrity and accuracy of each order information; the customer's special requirements refer to the specific requirements that customers may have for the order, such as packaging method, delivery time, invoice information, etc.;
[0087] S2. After receiving the natural language order, this application uses advanced natural language processing technology to deeply analyze the order, accurately extract key information such as product name, quantity, color, size, etc., and provide an accurate data basis for subsequent picking operations. The process of deeply analyzing the order through natural language processing technology is as follows:
[0088] Preprocessing (cleaning, standardization): Text cleaning: remove irrelevant characters, punctuation marks, special symbols, etc.; Text standardization: unify text formats, such as converting to lowercase, removing extra spaces, etc.;
[0089] Tokenization: dividing text into individual words or phrases;
[0090] Dependency syntactic analysis, using tools such as Stanford Parser to deeply analyze the dependency relationship between words, so as to accurately determine the grammatical structure of the sentence;
[0091] Component syntactic analysis: By identifying phrase structures in sentences, such as noun phrases and verb phrases, the system can gain a deeper understanding of the semantic information in the order;
[0092] S3, using machine learning models (such as conditional random fields (CRFs) and bidirectional LSTM-CRF models) to identify named entities in text, such as product names, brands, quantities, etc. These models are usually trained based on a large amount of labeled data to identify various entities;
[0093] Relationship extraction and slot filling:
[0094] Rule-based methods: use regular expressions or predefined rules to match and extract relationships between entities;
[0095] Deep learning-based methods: Use deep learning models (such as Transformer models, BERT, etc.) to automatically extract the relationship between entities, make full use of the contextual information in the text, and improve the accuracy of relationship extraction;
[0096] Slot filling: Fill the extracted entities and relations into predefined slots, such as product name slot, quantity slot, etc., which helps convert unstructured natural language text into structured data format;
[0097] Verify the extracted data to ensure its accuracy and completeness, and format the data to facilitate its use and storage in subsequent steps;
[0098] Application example: A customer inputs through the user interface: "Please order two blue XX brand T-shirts, size L, to be delivered to No. XX, XX Street, Chaoyang District, Beijing." The system receives and processes this natural language order, accurately capturing key information such as order number, product name, quantity, color, size, and delivery address;
[0099] Preprocessing (segmentation results): perform segmentation on the order content and divide it into separate words or phrases, such as: Please / order / two / blue / XX / brand / T-shirts / , / size / L / , / delivery / to / No.XX / , / XX / street / , / Chaoyang / District / , / Beijing / ;
[0100] Dependency syntactic analysis: Dependency syntactic analysis reveals the dependency relationship between words and determines the grammatical structure of the sentence. The results of dependency syntactic analysis are as follows:
[0101] Order depends on please (auxiliary verb), two depends on order (quantity complement), blue depends on T-shirt (color attribute), XX brand depends on T-shirt (brand attribute), T-shirt depends on order (object), for depends on size (preposition), L depends on for (preposition object), delivery is another verb center, which depends on the implicit root node, to depends on delivery (preposition), Beijing, Chaoyang District, XX Street, XX No. constitute the address phrase in sequence, all of which depend on to (address component);
[0102] Constituent Syntactic Analysis: Constituent syntactic analysis identifies the phrase structure in a sentence, such as:
[0103] Main sentence: [Please][order][two pieces of [blue [XX brand T-shirts]]][, [size [L]]][, [delivered [to [No. XX, [XX Street], Chaoyang District, Beijing]]]]]]];
[0104] Among them, [two [blue [XX brand T-shirts]]] is an object phrase, [size [is L]] is a parallel object complement phrase, and [deliver [to [...]]]] is another parallel verb phrase;
[0105] Named Entity Recognition (NER):
[0106] Named entity recognition technology is used to identify named entities in text, such as product names, addresses, etc. For example:
[0107] Product name: XX brand T-shirt; Quantity: two pieces; Color: blue; Size: L; Address: No. XX, XX Street, Chaoyang District, Beijing;
[0108] Relationship extraction and slot filling:
[0109] Relationship extraction is used to identify the relationship between entities and fill this information into predefined slots: product name slot, quantity slot, color slot, size slot, and delivery address slot;
[0110] Slot filling results: Product name slot: XX brand T-shirt; Quantity slot: 2 (or converted to digital form); Color slot: blue; Size slot: L; Delivery address slot: No. XX, XX Street, Chaoyang District, Beijing;
[0111] S4: During order processing, if the order exceeds the preset value range of the original abnormality, NLP technology can be used to detect abnormal information in the order, such as abnormal quantity of goods, unknown delivery address, etc. Once an abnormality is detected, it can automatically remind relevant personnel to handle it, or automatically adjust it according to the preset rules to ensure the accuracy and feasibility of the order; specifically, it includes:
[0112] Abnormal orders include:
[0113] Abnormal order amount: Calculate the total order amount and check whether it exceeds or falls below the set threshold (if it is too high, it may involve fraud; if it is too low, it may involve misoperation);
[0114] Abnormal product quantity characteristics: Count the number of products in the order to check whether there is an abnormally large number of products (such as exceeding the reasonable purchase amount of an individual) or whether there is an abnormal number of individual products (such as the quantity is zero but the price is not zero);
[0115] Abnormal delivery address: Check whether the delivery address is valid (such as whether there is a non-existent area or street), and whether the address changes frequently (which may involve fraud);
[0116] Abnormal user behavior: Combined with the user's historical order data, analyze whether the user's purchasing behavior is abnormal (such as sudden changes in purchasing habits, frequent order cancellations, etc.);
[0117] Before performing anomaly detection, the order data needs to be feature extracted and preprocessed, which includes extracting features related to anomaly detection from the order data (such as order amount, order quantity, payment method, delivery address, order time, etc.), and cleaning, converting and standardizing these features to ensure that the algorithm can effectively process the data;
[0118] Next, we need to use known normal and abnormal order data to train the selected model architecture. Through training, the model can learn the characteristic distribution of normal orders and abnormal orders. During the training process, we also need to tune the model parameters to improve the detection accuracy and efficiency of the model.
[0119] After the model training is completed, the model will calculate the abnormality score or probability of each order based on the input feature data, and determine whether the order is abnormal based on the preset threshold. Finally, the detection results will be output in the form of a report or list for further analysis and processing by relevant personnel;
[0120] The performance of anomaly detection algorithms may be affected by many factors, such as data quality, algorithm selection, parameter settings, etc. Therefore, in practical applications, it is necessary to continuously collect feedback and iteratively optimize the algorithm. By adjusting algorithm parameters, improving feature extraction methods, or introducing new algorithm models, the accuracy and efficiency of anomaly detection can be gradually improved. The performance of anomaly detection algorithms may change with the increase in data volume and changes in business scenarios, so regular evaluation and optimization are required.
[0121] S5, when abnormal orders are detected, NLP technology and other data analysis technologies (such as text classification algorithms) are used to classify the abnormal orders that have been detected and further explain these anomalies, including:
[0122] Abnormal orders can be classified using text classification algorithms (such as Naive Bayes, Logistic Regression, etc.) to determine the specific type and severity of the abnormality: such as quantity abnormality (such as order quantity is too large or too small), description abnormality (such as unclear or contradictory product description), address abnormality (such as invalid or incomplete delivery address), etc.; some abnormalities may be just minor errors, such as spelling errors, while others may cause the order to be unable to be fulfilled, such as wrong delivery address. According to the severity, those abnormalities with greater impact on the business can be processed first. The purpose of classification is to refine the abnormality into specific types, such as quantity abnormality, description abnormality, address abnormality, etc., so as to better understand the nature and cause of the abnormality. This classification not only helps with subsequent processing, but also provides valuable insights for the business team;
[0123] Based on the type and severity of the abnormal order, further determine the best handling method, such as:
[0124] Order correction: If the exception is caused by input errors or inconsistent data, the system may automatically or manually correct the order information, such as correcting the quantity of goods, modifying the delivery address, etc. The corrected order will re-enter the processing flow. The corrected order will be treated as a normal order and will participate in the subsequent grouping and order picking wave planning. If the processing result of an abnormal order is not correctly verified, it may be incorrectly assigned to a picking wave, and the pickers may pick the wrong goods or quantities, which will cause more errors and delays, thereby affecting the picking efficiency and accuracy of the entire wave;
[0125] Order cancellation: For some serious exceptions, such as out-of-stock products, payment failures, or invalid delivery addresses, the system may automatically cancel the order and send a notification to the user. The canceled order will be marked as canceled and removed from the current processing queue and will not be included in subsequent order groupings;
[0126] Order suspension: In some cases, the system may suspend order processing for further investigation or waiting for user confirmation. For example, if the system detects potential fraud or requires more information from the user, the order may be suspended until the problem is resolved. Similarly, suspended orders will not participate in subsequent order grouping;
[0127] Notify users: Whether an order is modified, canceled, or suspended, the system will send notifications to users through appropriate channels (e.g., email, SMS, in-app notifications, etc.). These notifications will inform users of the change in order status and any further actions that may be required;
[0128] S6, this step is to cluster and decompose the N orders in the order queue after abnormal order processing into M task sets: In this step, the order queue contains N orders, which contain the information of the goods purchased by the user. In order to optimize the picking process, a clustering algorithm is used to decompose these orders and divide them into M task sets. The orders in each task set have some common characteristics or similarities when picking, such as similar types of goods, the same storage location or similar delivery time. Through clustering and decomposition, orders with similar picking characteristics can be combined to form a task set, which can reduce the moving distance and time of the pickers or picking robots in the warehouse and improve the picking efficiency.
[0129] Specifically include:
[0130] Natural language processing can parse the text description of goods or orders and extract keywords or features. These features can be used as inputs for clustering algorithms to improve the accuracy of grouping, intelligently group orders and plan order picking waves. This ensures that pickers or equipment can perform picking operations in the optimal order. Based on factors such as the order's product combination, delivery time, and delivery location, algorithms are used to intelligently group orders to form multiple picking batches to improve picking efficiency.
[0131] Among them, "wave" refers to the process of combining multiple orders into one or more picking batches (or waves). Each wave contains a group of orders that need to be picked together. These orders have certain common characteristics or similarities when picking, such as they may need to visit the same picking area, contain similar types or quantities of goods, or have similar delivery times, etc. The purpose of wave picking is to improve picking efficiency and reduce picking costs by optimizing the picking process. By combining multiple orders into one wave, pickers or picking robots can traverse the picking needs of multiple orders at one time, reducing the number and distance of movements in the warehouse, thereby saving time and physical strength. In e-commerce picking systems, waves are usually automatically generated according to certain rules or algorithms. These rules or algorithms may be based on a comprehensive consideration of multiple factors such as the type, quantity, storage location, and delivery time of the order. Once a wave is generated, a specific picking task will be generated for each wave, and the optimal picking path will be planned to guide pickers or picking robots to perform picking operations;
[0132] Before clustering and grouping, the order features need to be converted into feature vectors. The conversion process is as follows:
[0133] Data preprocessing: collect order data, including product information, customer address, purchase time, etc., and perform necessary preprocessing, which includes steps such as data cleaning, standardization and normalization to ensure data quality and consistency;
[0134] Text feature extraction: Use NLP technology to extract key features from product descriptions, such as product type, brand, size, color, etc., and use methods such as TF-IDF to assign weights to features to form feature vectors. TF-IDF (TermFrequency-Inverse Document IDF (Inverse Document Frequency) is a statistical method used to evaluate the importance of a word to a document set or one of the documents in a corpus. It includes: Term Frequency (TF): the frequency of a word appearing in a document; Inverse Document Frequency (IDF): a measure of the general importance of a word. If a word or phrase appears in many documents, its IDF value will be low; conversely, if it only appears in a few documents, its IDF value will be high. By calculating the TF-IDF value of each word, that is, the product of TF and IDF, different weights can be assigned to different words, thereby highlighting those words that are more unique and important in a specific document. In the picking system, the TF-IDF vector can be used to weight the order data to highlight key information and reduce the impact of non-key information. The weighted order data can be used as the basis for the subsequent picking strategy formulation and optimization. For example, if the TF-IDF values of certain products in multiple orders are very high, it may indicate that these products are popular or commonly used products, and the system can give priority to the picking tasks of these products.
[0135] Based on the TF-IDF value or as needed, select the most important words as features, which are the key features;
[0136] Suppose that in a "goods-to-person" picking system of an e-commerce warehouse, the picking process of a specific category of goods (such as clothing) needs to be optimized. There is a set of product descriptions, which contain information such as the brand, color, style, etc. of the goods. The goal of this application is to extract the key features in these descriptions and use TF-IDF weights to evaluate the importance of these features, as shown below:
[0137] Extract product description text from the database, for example:
[0138] Product 1 description: "Nike Air Max 90 red running shoes new for summer";
[0139] Product 2 description: "Adidas Ultra Boost black sneakers with breathable mesh";
[0140] Product 3 description: "Puma RS-X blue casual shoes retro style";
[0141] Product 4 description: "Nike Air Force 1 classic white basketball shoe";
[0142] Product 5 description: "New Balance 574 grey retro running shoes with comfortable cushioning";
[0143] By using word segmentation, removing stop words, and lowercasing, each product description is split into vocabulary units, and the vocabulary list is obtained: {"Nike","Air","Max","90","red","running shoes","summer","new products","Adidas","Ultra","Boost","black","sports shoes","breathable","mesh","Puma","RS-X","blue","casual shoes","retro","style","Air","Force","1","white","basketball shoes","classic","New","Balance","574","grey","comfortable","cushioning"};
[0144] Calculate TF (term frequency)
[0145] For each product description, count the number of occurrences of each word and normalize them;
[0146] Calculate IDF (Inverse Document Frequency)
[0147] Calculate IDF on the entire dataset (all product descriptions);
[0148] Calculating TF-IDF
[0149] Multiply the TF value and IDF value of each word. For example, for "Nike" in product 1, assuming TF = 1 / 6 (because "Nike" appears once in the description and the total number of words is 6), IDF depends on the inverse logarithm of the frequency of "Nike" appearing in all product descriptions. Assuming IDF = 0.5, then TF-IDF("Nike") = 1 / 6*0.5;
[0150] Construct the feature vector:
[0151] Construct a feature vector for each product description. Since the dimension of the feature vector is equal to the number of words in the vocabulary, and each element corresponds to the TF-IDF weight of a word in the vocabulary, each feature vector will be a long vector with most elements being 0 (because most words will not appear in a single product description), such as:
[0152] Feature vector of product 1: [TF-IDF("Nike"),...,TF-IDF("red"),TF-IDF("running shoes"),TF-IDF("summer"),TF-IDF("new products"),0,0,...,0] (Note: the omitted parts are padded with 0 to match the vector length of the vocabulary);
[0153] Feature vector of product 2: [0,0,...,0,TF-IDF("Adidas"),TF-IDF("Ultra"),TF-IDF("Boost"),TF-IDF("Black"),TF-IDF("Sneakers"),TF-IDF("Breathable"),TF-IDF("Mesh"),0,...,0];
[0154] Clustering algorithms (such as K-means or DBSCAN) and features extracted by NLP are used to intelligently group orders. Orders are clustered according to feature vectors, and similar orders are grouped together. The goal of clustering is to make the similarity of orders within the same group high, while the similarity of orders between different groups low. Specifically, it includes:
[0155] Initialization: Select K initial centroids (K is the preset number of clusters);
[0156] Iterative process: assign feature vectors to the nearest centroid; for each feature vector, calculate its distance to each centroid and assign it to the cluster where the nearest centroid is located;
[0157] Recalculate the centroid: For each cluster, calculate the average of all feature vectors and set the average as the new centroid;
[0158] Repeat the above two steps until the change of the centroid is less than a certain threshold or the maximum number of iterations is reached;
[0159] Output: K clusters and their centroids;
[0160] S7, formulate batch processing strategy: formulate batch processing strategy according to clustering results, such as order quantity, commodity type, etc. Orders in the same cluster can be regarded as similar orders, so batch picking and packaging can be performed. In addition, the picking order can be optimized according to the size and importance of the cluster, such as processing clusters with a large number of orders first, or processing clusters with lighter weight and smaller volume first, so as to improve the overall efficiency; specifically including:
[0161] After confirming the order and performing intelligent grouping and wave planning, the system will accurately determine the location of the required goods in the automated warehouse based on the inventory information and order requirements in the WMS. This usually involves specific information such as the shelf number, number of layers and location, and prepare corresponding picking tools and equipment to fully prepare for subsequent picking operations;
[0162] Path planning and optimization: Based on the results of generating picking tasks, this embodiment plans one or more optimal picking paths for picking personnel or equipment by using advanced path planning algorithms (such as shortest path algorithm, heuristic algorithm, A* algorithm, etc.) taking into account factors such as warehouse layout, commodity distribution, and characteristics of picking tools. This can not only reduce walking distance and time, but also ensure efficient picking operations;
[0163] There are two ways to generate a path:
[0164] (1) Algorithm generation method, including:
[0165] Define data structure and path planning:
[0166] The definition of data structure includes:
[0167] Node: contains location information (such as coordinates), g value (cost from the starting point to the current node), h value (heuristic estimated cost from the current node to the end point), f value (g value plus h value), parent node pointer (used for backtracking path). In practical applications, factors such as warehouse layout, shelf arrangement, and aisle width will affect the selection of heuristic functions and the calculation of g value.
[0168] OpenList: A priority queue that sorts nodes according to their f-values. Nodes with smaller f-values have higher priorities.
[0169] ClosedList: used to store nodes that have been checked to avoid repeated processing;
[0170] Set the starting and ending points: determine the starting and ending locations for picking (i.e. the shelf locations of the required goods);
[0171] Set the heuristic function: Choose a suitable heuristic function (such as Manhattan distance or Euclidean distance) to calculate the h value;
[0172] Initialize the starting point: add the starting point to OpenList, set its g value to 0, h value to the result of the heuristic function calculation, and f value to the sum of g value and h value;
[0173] Path planning includes:
[0174] Loop condition: When OpenList is not empty, continue execution;
[0175] Select the current node: remove the node with the smallest f value from OpenList as the current node;
[0176] Expand neighbor nodes: For each neighbor node of the current node, if the neighbor node is in the ClosedList, ignore it. If not, calculate the g value from the starting point to the neighbor node (calculated by the g value of the current node and the actual distance from the current node to the neighbor node). If the neighbor node is not in the OpenList, or the new g value is smaller than the existing g value in the OpenList (indicating that a shorter path has been found), update the g value and f value (g value plus h value) of the neighbor node, and set the parent node pointer to point to the current node; if the neighbor node is in the OpenList but the g value is not updated, no further operation is required; add (or update) the neighbor node to the OpenList and reorder it according to the f value;
[0177] Add the current node to ClosedList: This means that the node has been checked.
[0178] Check whether the end point has been reached: If the current node is the end point, backtrack to build the path through the parent node pointer and exit the algorithm;
[0179] Algorithm termination: If OpenList is empty and no path is found, it will display that no path can be found;
[0180] Path backtracking: trace back to the starting point through the parent node pointer of the end point, build and return the optimal path;
[0181] (2) A pseudo code generation method, comprising:
[0182] Variable definition:
[0183] start: starting point node;
[0184] goal: end point node;
[0185] openList: a list containing the nodes to be checked, each with the values of f(n), g(n) and h(n);
[0186] closedList: a list containing the checked nodes;
[0187] current: the node currently being processed;
[0188] node: current's neighbor node;
[0189] g(n): actual cost from the starting point to node n;
[0190] h(n): heuristic estimated cost from node n to the end point;
[0191] f(n): the total estimated cost from the starting point through node n to the end point, f(n) = g(n) + h(n);
[0192] dist(a,b): calculates the actual distance function between node a and node b;
[0193] initialization:
[0194] Define the starting point (start), end point (goal) and heuristic function h(n);
[0195] Create an open list (openList) to store the nodes to be checked, and a closed list (closedList) to store the checked nodes;
[0196] Add the starting point start to openList, and set its g(start) = 0, f(start) = h(start);
[0197] Loop processing:
[0198] If openList is empty then
[0199] Report that path was not found and exit.
[0200] otherwise
[0201] Select the node current (called the current node) with the smallest f(n) from openList.
[0202] Move current from openList to closedList.
[0203] If current is goal then
[0204] Backtrack from current to start, build a path and return it.
[0205] otherwise
[0206] For each neighbor node of current, execute
[0207] Ignored if node is in closedList.
[0208] Calculate g(node)=g(current)+dist(current,node) from start to node (dist is the actual distance between the two points).
[0209] If node is not in openList or the new g(node) is smaller than the existing g(node) in openList, then
[0210] Add node to openList.
[0211] Set node’s parent to current.
[0212] Calculate f(node)=g(node)+h(node).
[0213] Termination conditions:
[0214] When the path is found (i.e., the goal is on the path), the algorithm terminates and returns the path;
[0215] If openList is empty and no path is found, the algorithm terminates and reports that no path could be found;
[0216] The formulas involved are:
[0217] f(n)=g(n)+h(n): Total cost function used to evaluate the priority of node n.
[0218] g(node)=g(current)+dist(current,node): Calculates the actual cost from the starting point to the neighbor node.
[0219] S8, through a visual interface, based on the results of path planning and optimization, displays the recommended picking paths to the pickers in an intuitive and easy-to-understand manner. If the path planning and optimization output multiple results, multiple picking paths can be provided for the pickers to choose from, which enables the pickers to quickly and accurately find the location of the goods, further improving the picking efficiency;
[0220] After completing the path planning, multiple optional paths may be generated, similar to the map navigation system providing multiple route options. In order to improve the user experience and flexibility, this embodiment uses natural language processing technology (NLP) to convert these path planning results into intuitive and easy-to-understand natural language descriptions, specifically:
[0221] Natural language description generation: By combining NLP technology with machine learning algorithms (such as deep learning models), the sequence-to-sequence (Seq2Seq) model is used to convert path planning data into natural language descriptions. The system can convert complex path planning results into instructions or suggestions described in natural language. For example, the system can generate natural language instructions such as "Please go along channel A to intersection B, then turn right to channel C", so that workers can understand and execute path planning more intuitively.
[0222] The conversion process of path planning data into natural language description includes:
[0223] Input route planning data: Receive raw data containing specific route information (such as lane name, distance, turning point, etc.);
[0224] Parsing path planning data: Parsing input data to extract key information, such as lane name, distance, turn instructions, etc.
[0225] Select / build templates: Select or dynamically build appropriate natural language templates based on the characteristics of the path planning data (such as length, complexity, etc.). These templates should contain placeholders for subsequent filling of specific information.
[0226] Fill template: Fill the key information extracted from the path data into the selected template and replace the placeholders in the template to ensure the correctness and fluency of the information;
[0227] Add additional information: Based on the template filling, add additional audio-visual guidance information as needed, such as prompts for LED guide lights, suggestions for using the indoor positioning system, etc.
[0228] For ease of understanding, according to the above process, the generated natural language description of the path is as follows: Please go straight along the main lane A for about 30 meters. Note that there is a T-shaped intersection B ahead, where there are obvious yellow warning lines and signs. After reaching intersection B, please confirm that the right side is the entrance to the auxiliary lane C, above which there is a signboard of "C zone entrance". Please turn right into the auxiliary lane C and continue forward for about 25 meters. You will see that the target cargo location is on the second floor of the third row of shelves on the left. There is a corresponding electronic display screen above the cargo location to display product information and picking instructions. To ensure accuracy, please pay attention to the LED guide lights on the ground when approaching intersection B. They will light up in advance to guide you to turn correctly. At the same time, please use your handheld device or the indoor positioning system you wear to confirm your current location to avoid entering other lanes by mistake.
[0229] The interactive visual interface can use natural language processing technology to parse user input, such as querying the path of a specific product, making special requests or feedback questions, and give corresponding responses. At the same time, the system can display the floor plan of the warehouse and the recommended picking path. Pickers can intuitively see the recommended path and the specific operations at each path point (such as picking up and putting away goods) on the interface. These paths are displayed through an interactive visual interface. The graphical interface is used to intuitively display the floor plan of the warehouse, the shelf layout and the recommended picking path. Pickers can view detailed information of different paths on the interface, including path length, estimated time, possible obstacles, etc., so as to make the best choice;
[0230] This embodiment provides two modes: automatic execution and manual selection execution. If the picker does not make a selection within the preset time, the default picking path and operation will be automatically executed, which not only ensures efficiency but also gives the picker a certain degree of flexibility and control.
[0231] During the picking process, the execution of picking personnel or equipment is monitored in real time. By recording the picking progress and abnormal situations, the picking operation can be ensured to proceed smoothly and potential problems can be discovered and handled in a timely manner. The system not only monitors the progress and accuracy of picking in real time through tools such as handheld devices and warehouse management systems, but also uses NLP technology to analyze and understand the real-time feedback and problems of picking personnel.
[0232] Smooth execution: During the picking process, the system uses acoustic and optical guidance technologies (such as LED guide lights, electronic display screens, voice prompts, etc.) to help pickers find goods quickly and accurately. These guidance technologies complement NLP technologies and jointly improve picking efficiency;
[0233] Exception handling: If any exception or error is found, such as missing goods or wrong location, for example, if the picker reports that "the specified goods cannot be found", the system will combine NLP sentiment analysis and keyword extraction technology to promptly identify and remind the picker, and take corresponding corrective measures to automatically trigger the exception handling process, such as querying inventory, providing alternative goods or re-planning the picking route, while ensuring that the picking process complies with the preset personalized picking strategy;
[0234] It is worth emphasizing that the above-mentioned audio-visual guidance information includes:
[0235] Barcode / QR code scanning: Confirm whether the picked product is correct by scanning the product’s barcode or QR code;
[0236] Electronic display screen guidance: Electronic display screens are used in warehouses to display picking instructions, and workers perform picking tasks based on the information on the screen;
[0237] Light guidance system: Use light guidance on the shelf to indicate the location where the staff needs to pick, or use LED guide lights or ground projection to light up or display arrows at key path nodes in advance to guide personnel or AGVs to move correctly;
[0238] Fixed-format electronic instructions: The system can send fixed-format electronic instructions to the staff's handheld devices, and the staff will pick according to these instructions;
[0239] Location-based services (LBS): Tracking workers’ locations through indoor positioning systems to ensure they pick items according to the established routes;
[0240] Electronic tags: Install electronic tags at shelves, lane entrances, intersections, etc., and automatically identify and confirm the location through handheld devices or AGV's RFID reader;
[0241] Voice prompts: Prompt information is given by voice at key nodes (such as intersections and turns), such as "Turn right ahead and enter Lane C";
[0242] Real-time path intervention and adjustment: During the picking operation, based on real-time monitoring data, the system can make real-time path intervention and adjustments according to the actual situation, which can respond to emergencies or further improve picking efficiency, ensuring the flexibility and efficiency of picking operations, including:
[0243] Real-time information acquisition and analysis: During the picking process, workers may input real-time information through voice or text, such as excessive accumulation of goods in a certain area, congestion on a certain picking route, etc. NLP technology is used to analyze these natural language information and extract key content, such as congested areas, routes that need to be adjusted, etc.
[0244] Intent recognition and problem location: Through NLP intent recognition technology, the system can understand the intention of the instructions or feedback input by the staff, such as whether it is a request to adjust the path or report a problem. At the same time, combined with technologies such as named entity recognition (NER), it can accurately locate the specific location where the problem occurs or the path that needs to be adjusted;
[0245] Real-time path planning and adjustment: Based on the information analyzed by NLP, the path planning system can adjust the path in real time. For example, if a picking area is congested, the system can re-plan the path to avoid the congested area to ensure picking efficiency, update the path guidance in the handheld device or AGV control system, and provide real-time feedback of new action instructions through acoustic and optical auxiliary means;
[0246] Natural language driven exception handling: Through NLP technology, the system can monitor and analyze the natural language descriptions of pickers in real time to identify potential exceptions or problems. For example, if a picker says "this item cannot be found", the system can automatically trigger the exception handling process, such as checking inventory, providing alternative items, or replanning the picking route;
[0247] Intelligent packaging and label generation: After picking is completed, the system automatically generates packaging requirements and product labels based on order information, which not only ensures the correctness and traceability of the products, but also improves the efficiency of subsequent delivery and after-sales service. The specific process is as follows:
[0248] Input: Receive order or product description text;
[0249] Text parsing and extraction: Use NLP technology to parse and extract key information from order or product descriptions. For example, extract key attributes such as product name, brand, quantity, color, size, etc. from a description.
[0250] Intelligent label generation: In addition to the conventional set values, product labels are automatically generated based on the extracted information. These labels contain not only basic information, but also other attributes inferred from the text description, such as "waterproof", "wear-resistant", "fragile", etc. In addition, word vector technology (such as Word2Vec or BERT) can be used to generate a vector representation for each label, which facilitates subsequent search and recommendation.
[0251] Smart packaging suggestions: Based on product labels, smart packaging suggestions are provided. For example, for fragile or valuable items, the system can automatically suggest strengthening the packaging or using special protective materials.
[0252] Output: Generate final smart packaging and labels;
[0253] Loading operation and delivery preparation: After picking and packing, the system enters loading operation and prepares for corresponding delivery work, which includes printing delivery orders, arranging transportation and other steps to ensure that the goods can be delivered to customers accurately and in a timely manner;
[0254] Customer feedback collection and analysis: The system collects and analyzes customer feedback on delivered goods. By understanding customer satisfaction and needs, the system can continuously improve service quality and enhance customer satisfaction, including:
[0255] Collect customer feedback on picking and delivery services through multiple channels (such as official websites, apps, social media, etc.), including text reviews, ratings, pictures and other multimedia content;
[0256] Clean the collected customer feedback, remove irrelevant information and noise, use NLP technology to perform word segmentation, part-of-speech tagging and syntactic analysis, and use sentiment analysis algorithms to judge the sentiment tendency (positive, negative, neutral) of each feedback;
[0257] Use algorithms such as TF-IDF and TextRank to extract keywords from feedback, and apply topic models such as LDA (Latent Dirichlet Allocation) to identify the main issues and concerns in customer feedback;
[0258] Automatically categorize issues based on keywords and topics (such as packaging issues, delivery delays, damaged goods, etc.), and prioritize issues using multi-dimensional information such as issue frequency and severity of sentiment.
[0259] For each type of problem, formulate specific picking optimization strategies based on the actual warehouse conditions (such as shelf layout, staffing, etc.). The optimization strategies may include adjusting the picking path, improving the packaging method, and improving staff training.
[0260] After implementing the optimization strategy, collect customer feedback again, evaluate the optimization effect, and quantify the optimization results by comparing customer satisfaction indicators before and after optimization (such as improved scores, reduced negative feedback, etc.). Based on the evaluation results, continuously adjust and optimize the picking strategy to form a closed-loop feedback system;
[0261] Return and exchange (reverse logistics) request processing: In the return and exchange process, it is first necessary to collect the return and exchange application information submitted by the customer, which usually includes the reason for return and exchange, product description, order information, etc. Most of this information exists in the form of natural language text and needs to be processed by NLP technology, which can ensure the accuracy and efficiency of return and exchange requests and improve customer service experience. The specific process is as follows:
[0262] Collect return and exchange application information submitted by customers;
[0263] Data preprocessing: text cleaning, word segmentation and part-of-speech tagging;
[0264] NLP analysis: Using NLP technology, you can conduct an in-depth analysis of the reasons for returns and exchanges provided by customers. This helps merchants understand the problems with the products and the needs of customers, so as to improve products and services. Mainly includes: sentiment analysis and keyword extraction. Through sentiment analysis algorithms, judge the customer's emotions and attitudes towards returns and exchanges, such as anger, disappointment, indifference, etc. Sentiment analysis can help merchants understand the emotional state of customers so as to better handle return and exchange requests and provide compensation measures. Keyword extraction: Use algorithms such as TF-IDF and TextRank to extract keywords in the reasons for returns and exchanges. Keywords can help merchants quickly understand the main reasons for returns and exchanges, such as "inappropriate size", "quality issues", etc.;
[0265] Automated processing: Return and exchange requests are automatically classified based on the analysis results. The system can automatically or semi-automatically process return and exchange requests and give preset responses. For example, based on the reason for return and exchange, the request is automatically classified into different processing flows, such as "return due to quality issues" and "return due to size incompatibility". A preset response template is provided for each type of return and exchange request to quickly respond to customers;
[0266] Manual review and processing: Service personnel conduct further review and processing based on the automated processing results;
[0267] Data analysis and improvement: Display NLP analysis results in the form of charts or reports, and perform data analysis on the analysis results to identify problems in products and services and make improvements.
[0268] Intelligent replenishment reminders and suggestions: Based on inventory status and sales data, the system intelligently generates replenishment reminders and suggestions, which can help warehouse managers replenish goods in a timely manner, ensure that inventory is sufficient and meets sales demand, maintain the optimal goods supply status, and further improve the intelligent level of warehouse management, including:
[0269] On the backend replenishment page, operators can input replenishment information through voice or text, and NLP technology is used to perform semantic analysis and information extraction. This information may include key information such as product name, quantity, replenishment time, etc.
[0270] Use NLP technology to extract key information such as sales trends, customer needs and preferences from text data such as sales reports and customer feedback. Perform sentiment analysis on customer comments and feedback to understand their satisfaction and demand for products;
[0271] Through data mining and predictive analysis and other technical systems, operators can also provide specific replenishment suggestions, including replenishment quantity, replenishment time, etc., based on sales trends and inventory demand forecasts. For example, by analyzing historical sales data and customer feedback, combined with sentiment tendencies extracted by NLP, time series prediction models (such as ARIMA, LSTM, etc.) are used to predict future demand;
[0272] Process real-time inventory reports through NLP to extract key inventory indicators (such as inventory turnover, inventory days, etc.), inventory optimization algorithm: combine demand forecasts and inventory indicators, and use algorithms (such as the economic order quantity model EOQ) to determine the optimal replenishment quantity and replenishment time;
[0273] Based on the above analysis, the system can generate intelligent replenishment reminders and suggestions. Specifically, when the inventory of a certain product falls below a preset threshold, the system will automatically issue a replenishment reminder. This threshold can be dynamically adjusted based on historical sales data and forecasting models to ensure inventory adequacy and avoid out-of-stocks.
[0274] The system can also provide operators with specific replenishment suggestions based on sales trends and inventory demand forecasts, including replenishment quantity, replenishment time, and recommended suppliers. These suggestions can help operators perform replenishment operations more scientifically and efficiently;
[0275] Based on NLP order processing, the system can automatically or semi-automatically generate replenishment orders and use NLP to ensure the accuracy and completeness of order information. Use NLP technology to handle communications with suppliers, including order confirmation, shipping notifications, etc.
[0276] Embodiment 2
[0277] Figure 2is a system schematic diagram of an automatic picking device based on natural language processing according to another exemplary embodiment, comprising:
[0278] Order acquisition module 1: used to obtain natural language orders entered by customers;
[0279] Preprocessing module 2: used to perform preprocessing operations on the natural language order, perform word segmentation operations on the preprocessed natural language order, and segment the natural language order text into separate words; use the Stanford Parser tool to extract the dependency relationship between words, and extract the phrase structure in the natural language order text through component syntactic analysis;
[0280] Natural language analysis module 3: used to identify named entities in the natural language order text using a pre-trained neural network model, extract the relationship between the named entities by combining the dependency relationship between words and phrase structure based on a rule-based method or a deep learning method, and fill the named entities and the relationship between named entities into predefined slots, thereby converting the unstructured natural language order text into a structured data format;
[0281] Abnormal order judgment module 4: used to extract preset features related to abnormality detection from structured natural language orders, input the extracted features related to abnormality detection into a pre-trained abnormality detection model, and the pre-trained abnormality detection model calculates the abnormality score or probability of each order according to the input features, and judges whether the order is abnormal according to the preset abnormality score or probability threshold;
[0282] Abnormal order processing module 5: used to classify the detected abnormal orders according to the abnormal reasons, modify, cancel or suspend the abnormal orders according to the abnormal order classification results, and update the current order queue according to the abnormal order processing results;
[0283] Order classification module 6: used to extract the features of all structured natural language orders in the updated order queue, and use the TF-IDF method to assign weights to the extracted features to form feature vectors; cluster all structured natural language orders in the current order queue according to the feature vectors, group similar orders together, and obtain multiple order clusters;
[0284] Wave path generation module 7: used to generate a picking wave strategy for each order cluster, determine the location of the required goods in the automated warehouse according to the picking wave strategy for each order cluster, and plan the picking path in the automated warehouse according to the location of the required goods in the automated warehouse;
[0285] Picking execution module 8: used to determine the picking path that needs to be executed, and use sound and light guidance technology to help pickers or equipment quickly and accurately find the required goods in the picking path until the picking wave strategy for each order cluster is completed.
[0286] Embodiment three:
[0287] This embodiment provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a main controller, each step in the above method is implemented;
[0288] It is understandable that the storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.
[0289] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.
[0290] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" refers to at least two.
[0291] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.
[0292] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0293] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.
[0294] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.
[0295] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.
[0296] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0297] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.
Claims
1. An automatic picking method based on natural language processing, characterized in that: The method comprises: Get natural language orders entered by customers; Performing a preprocessing operation on the natural language order, and performing a word segmentation operation on the preprocessed natural language order to segment the natural language order text into separate words; using the Stanford Parser tool to extract the dependency relationship between words, and extracting the phrase structure in the natural language order text through component syntactic analysis; Using a pre-trained neural network model to identify named entities in the natural language order text, using a rule-based method or a deep learning-based method, combining dependencies between words and phrase structures to extract relationships between the named entities, and filling the named entities and the relationships between named entities into predefined slots, thereby converting the unstructured natural language order text into a structured data format; Extracting preset features related to anomaly detection from structured natural language orders, inputting the extracted features related to anomaly detection into a pre-trained anomaly detection model, wherein the pre-trained anomaly detection model calculates an anomaly score or probability of each order according to the input features, and determines whether the order is abnormal according to a preset anomaly score or probability threshold; The detected abnormal orders are classified according to the abnormal reasons, and the abnormal orders are corrected, cancelled or suspended according to the abnormal order classification results, and the current order queue is updated according to the abnormal order processing results; Extract the features of all structured natural language orders in the updated order queue, and use the TF-IDF method to assign weights to the extracted features to form a feature vector; cluster all structured natural language orders in the current order queue according to the feature vector, group similar orders together, and obtain multiple order clusters; Generate a picking wave strategy for each order cluster, determine the location of the required goods in the automated warehouse according to the picking wave strategy for each order cluster, and plan a picking path in the automated warehouse according to the location of the required goods in the automated warehouse; Determine the picking path that needs to be executed, and use acoustic and optical guidance technology to help pickers or equipment quickly and accurately find the required goods in the picking path until the picking wave strategy for each order cluster is completed.
2. The method according to claim 1, characterized in that Also includes: The picking personnel or equipment inputs abnormal feedback information through voice or text in the picking path, and the abnormal feedback information is parsed through natural language processing technology to obtain the feedback intention of the picking personnel or equipment, and real-time path adjustment or abnormal handling process is executed according to the feedback intention.
3. The method according to claim 2, characterized in that Also includes: The generated picking path is converted into a natural language description using natural language processing technology. include: Inputting the original data of the picking path, extracting the key coordinate information in the original data, filling the key information into the selected path natural language template, and replacing the original placeholder in the path natural language template to obtain a structured picking path; Add audio-visual guidance information to each coordinate point in the structured picking path.
4. The method according to claim 1, characterized in that: The method of planning a picking path in the automated warehouse according to the location of the required goods in the automated warehouse includes: Define data structure and path planning; The definition data structure includes: Set a starting point and an end point, where the end point is the location of the required goods in the automated warehouse; Define each path node between the starting point and the end point, including: the coordinates of the current node, the g value of the current node, the h value of the current node, the f value of the current node, and the parent node pointer; the g value of the current node is the cost from the starting point to the current node, the h value of the current node is the heuristic estimated cost from the current node to the end point, and the f value of the current node is the h value of the current node plus the g value of the current node; An open list and a closed list are set, wherein all path nodes are sorted according to the f value in the open list, and the node with a smaller f value has a higher priority for inspection; the closed list is used to store the path nodes that have been inspected; Set the heuristic function to calculate the h value of the current node; The path planning includes: Remove the path node with the smallest f value from the open list as the current check node. For each neighbor node of the current check node, if the neighbor node is already in the closed list, ignore the neighbor node. If the neighbor node is not in the closed list, calculate the g value from the starting point to the neighbor node. If the g value of the neighbor node is smaller than the g value of any path node already in the open list, then add the neighbor node to the open list and set the parent node pointer to point to the current check node; Add the current check node to the closed list and reorder the open list according to the f value; Repeat the above process until the current check node is the end point, and then backtrack through the parent node pointer to build the picking path.
5. The method according to claim 1, characterized in that The structured natural language orders in the current order queue are clustered according to the feature vectors, similar orders are grouped together, and multiple order clusters are obtained, including: Select K initial centroids, where K is the preset number of clusters; For each feature vector, calculate its distance to each centroid and assign it to the cluster with the closest centroid; For each cluster assigned to a feature vector, calculate the average of all the feature vectors it contains and set the average as the new centroid; The above process is repeated until the change in the centroid of each cluster is less than a certain threshold or the maximum number of iterations is reached, and multiple order clusters are obtained.
6. The method according to claim 2, characterized in that Also includes: After the required goods are picked, feature information in the structured natural language order or product description is extracted, one or more product labels are generated based on the feature information, and vector representations are generated for the one or more product labels respectively; Based on one or more product labels, intelligent packaging suggestions are matched in a preset intelligent packaging suggestion library to generate final intelligent packaging suggestions and product labels.
7. The method according to claim 6, characterized in that Also includes: Obtain customer feedback on picking and delivery services through multiple channels, including text reviews, ratings, and pictures; Clean the collected customer feedback, remove irrelevant information and noise, and use natural language processing technology to perform word segmentation, part-of-speech tagging and syntactic analysis on the feedback; Through sentiment analysis algorithms, the sentiment tendency of each feedback opinion is judged based on part-of-speech tagging and syntactic analysis. The sentiment tendency includes: positive, negative and neutral; Extract keywords from feedback and identify issues contained in customer feedback using a pre-trained LDA topic model; Categorize the issues contained in the feedback and prioritize them based on the classification results, frequency of occurrence, and sentiment; Optimize the processing of problems according to their priority.
8. The method according to claim 7, characterized in that Also includes: In the return and exchange process, obtain the return and exchange application information submitted by the customer; Clean the return and exchange application information to remove irrelevant information and noise, and use natural language processing technology to perform word segmentation, part-of-speech tagging and syntactic analysis on the return and exchange application information; Through the sentiment analysis algorithm, the sentiment tendency of each return application information is judged based on part-of-speech tagging and syntactic analysis, and the sentiment tendency includes: anger, disappointment and indifference; The TF-IDF algorithm is used to extract keywords from the return and exchange application information, and the user's return reason is obtained based on the keywords; Automatically match preset return and exchange response templates based on return reasons and sentiment tendencies.
9. An automatic picking device based on natural language processing, characterized in that: The device comprises: Order acquisition module: used to obtain natural language orders entered by customers; Preprocessing module: used to perform preprocessing operation on the natural language order, perform word segmentation operation on the preprocessed natural language order, and segment the natural language order text into separate words; use Stanford Parser tool to extract dependency relationship between words, and extract phrase structure in the natural language order text through component syntactic analysis; Natural language analysis module: used to identify named entities in the natural language order text using a pre-trained neural network model, extract the relationships between the named entities by combining the dependency relationship between words and phrase structure based on a rule-based method or a deep learning method, and fill the named entities and the relationships between named entities into predefined slots, thereby converting the unstructured natural language order text into a structured data format; Abnormal order judgment module: used to extract preset features related to abnormality detection from structured natural language orders, input the extracted features related to abnormality detection into a pre-trained abnormality detection model, and the pre-trained abnormality detection model calculates the abnormality score or probability of each order according to the input features, and judges whether the order is abnormal according to the preset abnormality score or probability threshold; Abnormal order processing module: used to classify detected abnormal orders according to abnormal reasons, modify, cancel or suspend abnormal orders according to abnormal order classification results, and update the current order queue according to abnormal order processing results; Order classification module: used to extract the features of all structured natural language orders in the updated order queue, and use the TF-IDF method to assign weights to the extracted features to form feature vectors; cluster all structured natural language orders in the current order queue according to the feature vectors, group similar orders together, and obtain multiple order clusters; Wave path generation module: used to generate a picking wave strategy for each order cluster, determine the location of the required goods in the automated warehouse according to the picking wave strategy for each order cluster, and plan the picking path in the automated warehouse according to the location of the required goods in the automated warehouse; Picking execution module: used to determine the picking path that needs to be executed, and use acoustic and optical guidance technology to help pickers or equipment quickly and accurately find the required goods in the picking path until the picking wave strategy for each order cluster is completed.
10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the main controller, each step of the drone control and positioning method based on digital twins as described in any one of claims 1 to 8 is implemented.
Citation Information
Patent Citations
Sorting list generation method and device
CN110533350A
Designated driver telephone ordering method and system
CN113794808A
Order generation method and device based on network chat information
CN116012098A
Path planning method for improving bidirectional jump point search region constraint
CN117949002A
Express whole-course aging prediction method, device and equipment and storage medium
CN118246584A
Cited By
Abnormality processing method, device and equipment and readable storage medium
CN119527767A
Transaction sink matching method and device based on natural language processing, and electronic equipment
CN119741107A
Unmanned aerial vehicle cargo distribution management method and system based on strategy optimization
CN120125126A
Cross-border logistics information management method and system based on logistics management platform
CN120893924A
Cross-border logistics information management method and system based on logistics management platform
CN120893924B
