Intelligent production scheduling method based on NLP (Natural Language Processing) algorithm

Abstract

The invention relates to the technical field of natural language processing, in particular to an intelligent production scheduling method based on a natural language processing (NLP) algorithm, and the method comprises the following steps: obtaining order information to be scheduled, preprocessing the text content in the order information based on natural language processing, and calculating the TF-IDF value of each word group in the text content, extracting keywords in the text content based on the TF-IDF value; obtaining an information index for evaluating the priority of the production tasks, wherein each production task corresponds to one piece of order information; evaluating the priority of the production task corresponding to each piece of order information based on the information index of the production task priority to obtain a priority sequence; obtaining current production resource information; the production scheduling plan information is formulated based on the production resource information and the production task corresponding to the priority sequence, important demands and preferences of customers can be reflected through natural language processing and TF-IDF value calculation, the production scheduling plan is flexibly adjusted, and the intelligence of the production scheduling process is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to an intelligent production scheduling method based on a natural language processing (NLP) algorithm. Background Art

[0002] In modern manufacturing, production scheduling is a key link in enterprise production management. Traditional production scheduling often relies on manual experience and arranges production tasks by considering factors such as delivery date and product complexity. However, with the intensification of market competition and the diversification of customer needs, traditional production scheduling methods have been unable to meet the efficient and flexible production needs of enterprises.

[0003] On the one hand, traditional production scheduling methods often ignore the text content in order information, such as product name, model, etc., which contain important customer needs and preferences. On the other hand, traditional production scheduling methods lack a scientific and objective evaluation system to accurately assess the priority of production tasks, resulting in unreasonable production task arrangement and low production efficiency.

[0004] To this end, we propose an intelligent production scheduling method based on natural language processing (NLP) algorithm to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent production scheduling method based on natural language processing (NLP) algorithm to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: an intelligent production scheduling method based on natural language processing (NLP) algorithm, the method comprising the following steps: S1: Obtain order information to be scheduled, where the order information includes product name, product model, product quantity and delivery date, pre-process the text content in the order information based on natural language processing and calculate the TF-IDF value of each phrase in the text content, and extract keywords in the text content based on the TF-IDF value; S2: Obtain information indicators for evaluating the priority of production tasks, where the information indicators include delivery date, product complexity, and TF-IDF value of keywords. Each production task corresponds to an order information. Based on the information indicators of the production task priority, the priority of the production task corresponding to each order information is evaluated to obtain a priority sequence. S3: Obtain current production resource information, wherein the production resource information includes production equipment information, human resource information and raw material inventory information; formulate production scheduling information based on the production resource information and the production tasks corresponding to the priority sequence, wherein the production scheduling information includes the start time, end time and required resources of each production task.

[0007] Preferably, the step of preprocessing the text content in the order information based on natural language processing and calculating the TF-IDF value of each phrase in the text content, and extracting keywords in the text content based on the TF-IDF value includes: Obtain the text content in the order information, remove meaningless symbols in the text content, divide the parsed text content into phrases, and obtain a phrase set; Formulate stop word screening rules, where the screening rules include screening conditions and trigger points; Remove the stop words in the text content based on the stop word screening rules to obtain a set of valid phrases; The TF-IDF value of each valid phrase is calculated, and the valid phrase corresponding to the TF-IDF value that meets the preset threshold is used as a keyword.

[0008] Preferably, the step of formulating stop word screening rule information includes: Get the standard number threshold corresponding to the historical stop words; Set multiple threshold points, each threshold point corresponds to a standard number threshold; A plurality of threshold points are arranged in a straight line to obtain a threshold line, a common point is set corresponding to the threshold line, a connection channel between the common point and each threshold point is established, and a trigger point and a trigger condition corresponding to the trigger point are set corresponding to the connection channel; Based on the trigger condition, the trigger point corresponding to the common point is started to obtain the target threshold point corresponding to the common point, and the standard number threshold corresponding to the target threshold point is used as the screening condition; The filtering rules are obtained based on the filtering conditions and the trigger points.

[0009] Preferably, the step of obtaining the standard number threshold corresponding to the historical stop words includes: Obtain stop words in the text content of historical order information to obtain historical stop words, obtain the historical occurrence counts of each historical stop word in each text content to obtain multiple historical occurrence counts corresponding to each stop word; The average historical occurrence count of each historical stop word is evaluated based on multiple historical occurrence counts, and the average historical occurrence count is used as the standard count threshold corresponding to the historical stop word.

[0010] Preferably, the step of removing stop words from the text content based on the stop word screening rule comprises: Based on the standard frequency threshold corresponding to each historical stop word, multiple standard frequency thresholds are obtained, and the multiple standard frequency thresholds are matched one by one with multiple threshold points; Get the phrase in the order information, place it at a common point, use the phrase and the historical stop words as the trigger condition, start the trigger point corresponding to the trigger condition to obtain the target trigger point; The threshold point corresponding to the connection channel where the target trigger point is located is used as the target threshold point, and the standard number threshold corresponding to the target threshold point is used as the screening condition of the phrase; The number of occurrences of each phrase in each text content is counted, and the phrases corresponding to the number of occurrences that do not exceed the corresponding standard number threshold are regarded as stop words, and the stop words are removed.

[0011] Preferably, the step of calculating the TF-IDF value of each valid phrase includes: Set is the set of valid phrases, Represents each valid phrase, set It is the collection of text content of all order information. Represents the text content of each order information; the TF-IDF model is as follows: , ,in, is the number of occurrences of valid phrases, is the inverse document frequency, For valid phrases in text content The number of occurrences in For text content The sum of the number of valid phrases in , is the total number of text contents in the order information. To contain a valid phrase The total number of text contents, the TF-IDF value is and The product value of ,in, is the TF-IDF value of the corresponding valid phrase.

[0012] Preferably, the step of evaluating the priority of the production task corresponding to each order information based on the information indicator of the production task priority to obtain the priority sequence includes: Build a knowledge graph, map the order information to the knowledge graph, and generate information features related to the order; Obtain historical information indicators for evaluating production task priorities, where the information indicators include delivery date, product complexity, and TF-IDF values ​​of keywords; The information features generated by the knowledge graph are integrated with the historical information indicators to form a feature set; Set weight values ​​for each feature in the corresponding feature set; The fused feature set and weight value are used to comprehensively calculate the priority score of each production task, and the production tasks corresponding to the priority scores are sorted in descending order to obtain a priority sequence.

[0013] Preferably, the steps of constructing a knowledge graph, mapping order information to the knowledge graph, and generating information features related to the order include: Define entities and relationships, where entities include order number, product name, quantity, delivery date, and customer name in order information, and relationships include containment and belonging relationships; build a knowledge graph based on entities and relationships; Assign a unique identity to each entity and relationship, and define clear types and attributes for each relationship; set an identifier for each date range or specific product, create nodes and relationships in the graph database, and store the entities, relationships and identifiers together in the graph database to obtain a knowledge graph; Preprocess the order data, extract order information, and map the order information to entities and relationships in the knowledge graph; Generate information features based on entities and relationships in the knowledge graph, wherein the information features include a first feature and a second feature, the first feature is an information feature generated based on entities and relationships, and the second feature is an information feature generated based on relationship paths in the knowledge graph; Preferably, the step of generating information features based on entities and relationships in the knowledge graph includes: Extract basic features from each entity in the knowledge graph; for numerical entities, directly use their values ​​as features; for text entities, use text processing technology to convert them into numerical features; analyze the relationship between entities, define features for each relationship according to the type and attributes of the relationship, use relationship weight and relationship distance as the numerical features of the relationship, combine the extracted entity features and relationship features to form a feature vector, and combine the feature vector by feature splicing or feature interaction to obtain the first feature; perform random walks on the knowledge graph to find paths related to the target relationship, and determine a set of paths as features based on the length of the path, the frequency of occurrence, and the entities and relationships contained in the path; for each relationship path, extract its structural features, and convert the relationship path into a low-dimensional vector representation as a path feature; when there are multiple relationship paths between an entity and multiple other entities, aggregate the features of the multiple relationship paths, and combine the aggregated path features with the original entity features to form the second feature.

[0014] Preferably, the step of performing a random walk on the knowledge graph to find a path related to the target relationship includes: Taking the order number as the starting point and the customer name as the target point, determine that all possible paths from the order number to the customer name need to be found; traverse the knowledge graph from the starting point, record each entity and relationship passed during the traversal process, form a path, and store all identified paths in a suitable data structure; set a path length threshold, filter out paths corresponding to the length that meets the path length threshold to obtain multiple pre-selected paths, filter out paths containing multiple different entities and relationships from the pre-selected paths based on entity and relationship diversity, evaluate the importance of each path based on the random walk path sorting algorithm, and filter out paths with higher scores according to the importance scores of the paths as paths related to the target relationship.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. Preprocess the text content in the order information through natural language processing technology, and calculate the TF-IDF value of each phrase in the text content to extract keywords. Keywords can reflect the important needs and preferences of customers and provide an important basis for the priority evaluation of production tasks. At the same time, this method also comprehensively considers factors such as delivery date and product complexity, and establishes a scientific and objective production task priority evaluation system; 2. Through natural language processing and TF-IDF value calculation, it can accurately extract keywords from order information, reflect important customer needs and preferences, help enterprises understand customer needs more accurately, and flexibly adjust production schedules according to customer preferences, formulate production schedules that better meet customer expectations, and realize intelligent production scheduling; establish a scientific and objective production task priority evaluation system, comprehensively consider delivery date, product complexity, and keyword TF-IDF values ​​and other factors. It helps enterprises arrange production tasks more reasonably, improve production efficiency, and reduce production costs. It helps enterprises improve production scheduling efficiency.

[0016] 3. Build a knowledge graph, map the order information to the knowledge graph, and build a knowledge graph based on entities and relationships for the order information, delivery date, and customer name; assign a unique identity to each entity and relationship, and define clear types and attributes for each relationship; set identifiers for each date range or specific product, create nodes and relationships in the graph database, store the entities, relationships, and identifiers together in the graph database to obtain a knowledge graph, and preprocess the order data to extract order information, and map the order information to the entities and relationships in the knowledge graph; generate information features based on the entities and relationships in the knowledge graph; improve the objective evaluation system to accurately assess the priority of production tasks, improve the rationality of production task arrangements, and improve production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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.

[0020] Example See also Figure 1 The present invention provides a technical solution of an intelligent production scheduling method based on a natural language processing (NLP) algorithm: an intelligent production scheduling method based on a natural language processing (NLP) algorithm, comprising the following steps: S1: Obtain order information to be scheduled, where the order information includes product name, product model, product quantity and delivery date, pre-process the text content in the order information based on natural language processing and calculate the TF-IDF value of each phrase in the text content, and extract keywords in the text content based on the TF-IDF value; Preprocessing the text content in the order information based on natural language processing and calculating the TF-IDF value of each phrase in the text content, and extracting keywords in the text content based on the TF-IDF value includes: obtaining the text content in the order information, removing meaningless symbols in the text content, dividing the parsed text content into phrases, and obtaining a phrase set; formulating stop word screening rules, wherein the screening rules include screening conditions and trigger points; removing stop words in the text content based on the stop word screening rules, and obtaining a valid phrase set; calculating the TF-IDF value of each valid phrase, and taking the valid phrase corresponding to the TF-IDF value that meets the preset threshold as a keyword; The steps of formulating stop word screening rule information include: obtaining a standard number threshold corresponding to historical stop words; setting multiple threshold points, each threshold point corresponding to a standard number threshold; arranging multiple threshold points in a straight line to obtain a threshold line, setting a common point corresponding to the threshold line, establishing a connection channel between the common point and each threshold point, setting a trigger point and a trigger condition of the corresponding trigger point corresponding to the connection channel; starting the trigger point corresponding to the common point based on the trigger condition to obtain a target threshold point corresponding to the common point, and using the standard number threshold corresponding to the target threshold point as a screening condition; obtaining a screening rule based on the screening condition and the trigger point; The step of obtaining the standard number threshold corresponding to the historical stop words includes: obtaining the stop words in the text content of the historical order information to obtain the historical stop words, obtaining the historical number of occurrences of each historical stop word in each text content to obtain multiple historical number of occurrences corresponding to each stop word, evaluating the average historical number of occurrences of each historical stop word based on the multiple historical number of occurrences, and using the average historical number of occurrences as the standard number threshold corresponding to the historical stop words; The step of removing stop words from the text content based on the stop word screening rule includes: obtaining multiple standard frequency thresholds based on the standard frequency thresholds corresponding to each historical stop word, and making one-to-one correspondence between the multiple standard frequency thresholds and multiple threshold points; obtaining a phrase in the order information, placing it at a common point, using the consistency of the phrase and the historical stop word as a trigger condition, starting the trigger point corresponding to the trigger condition to obtain a target trigger point, using the threshold point corresponding to the connection channel where the target trigger point is located as the target threshold point, and using the standard frequency threshold corresponding to the target threshold point as the screening condition for the phrase; counting the number of occurrences of each phrase in each text content, using the phrase corresponding to the number of occurrences that does not exceed the corresponding standard frequency threshold as a stop word, and removing the stop words; Specifically, firstly, a plurality of standard frequency thresholds are respectively matched with historical stop words, then the phrases in the text content are obtained, the phrases identical to the historical stop words are extracted, and the number of occurrences of the phrases is counted. Prior to this, a corresponding standard frequency threshold is set for the historical stop words, so as to determine whether the number of occurrences of the phrase exceeds the standard frequency threshold. If it is lower than or equal to the standard frequency threshold, it indicates that the phrase corresponding to the number of occurrences is a stop word. When it exceeds the standard frequency threshold, it is determined to be a valid phrase. Different standard frequency thresholds are automatically selected for different stop words. For each stop word, the corresponding standard frequency threshold is set according to the historical number of occurrences, so that the stop words can be effectively screened out and the recognition effect of the stop words can be improved. The steps for calculating the TF-IDF value of each valid phrase include: is the set of valid phrases, Represents each valid phrase, set It is the collection of text content of all order information. Represents the text content of each order information; the TF-IDF model is as follows: , ,in, is the number of occurrences of valid phrases, is the inverse document frequency, For valid phrases in text content The number of occurrences in For text content The sum of the number of valid phrases in , is the total number of text contents in the order information. To contain a valid phrase The total number of text contents, the TF-IDF value is and The product value of ,in, is the TF-IDF value of the corresponding valid phrase; Specifically, by removing stop words from the relevant text content in the order information, the accuracy of text content recognition can be improved, so that effective phrases can be screened out more accurately. The TF-IDF method combines the two indicators of term frequency (TF) and inverse document frequency (IDF). Term frequency (TF) indicates the number of times a word appears in a document, which reflects the local importance of the word in the document. The inverse document frequency (IDF) is a measure of the universal importance of a word, which reflects the global importance of the word in a document collection. By multiplying TF and IDF, the TF-IDF value of a word in a document can be obtained. The larger the value, the higher the importance of the word in the document. By calculating the TF-IDF value of each effective phrase and comparing it with the preset threshold, the keywords of the text content can be screened out for subsequent production scheduling of each order. The preset threshold refers to the threshold that can affect the content of the document after reaching the preset threshold, resulting in further impact on the production plan, thereby ensuring the reasonable effectiveness of subsequent production scheduling; S2: Obtain information indicators for evaluating the priority of production tasks, where the information indicators include delivery date, product complexity, and TF-IDF value of keywords. Each production task corresponds to an order information. Based on the information indicators of the production task priority, the priority of the production task corresponding to each order information is evaluated to obtain a priority sequence. The step of evaluating the priority of the production task corresponding to each order information based on the information indicator of the production task priority to obtain the priority sequence includes: Build a knowledge graph, map the order information to the knowledge graph, and generate information features related to the order; Obtain historical information indicators for evaluating production task priorities, where the information indicators include delivery date, product complexity, and TF-IDF values ​​of keywords; The information features generated by the knowledge graph are integrated with the historical information indicators to form a feature set; Set weight values ​​for each feature in the corresponding feature set; The fused feature set and weight value are used to comprehensively calculate the priority score of each production task, and the production tasks corresponding to the priority scores are sorted in descending order to obtain a priority sequence.

[0021] The steps of building a knowledge graph, mapping order information to the knowledge graph, and generating information features related to the order include: Define entities and relationships, where entities include order number, product name, quantity, delivery date, and customer name in order information, and relationships include containment and belonging relationships; build a knowledge graph based on entities and relationships; Assign a unique identity to each entity and relationship, and define clear types and attributes for each relationship; set an identifier for each date range or specific product, create nodes and relationships in the graph database, and store the entities, relationships and identifiers together in the graph database to obtain a knowledge graph; Preprocess the order data, extract order information, and map the order information to entities and relationships in the knowledge graph; Generate information features based on entities and relationships in the knowledge graph, wherein the information features include a first feature and a second feature, the first feature is an information feature generated based on entities and relationships, and the second feature is an information feature generated based on relationship paths in the knowledge graph; Specifically, in the order scenario, entities can include orders, products, customers, suppliers, etc. Relationships can be defined as associations between entities, such as the "contains" relationship between orders and products, the "belongs to" relationship between orders and customers, etc. Use a graph database or a relational database to store entities and relationships. Assign a unique ID to each entity and define clear types and attributes for each relationship. Clean and normalize order data to ensure data accuracy and consistency. Extract key information from the order, such as order number, product name, quantity, delivery date, customer name, etc. Map the key information in the order to the entities and relationships in the knowledge graph. For example, map the order number to the ID of the order entity, map the product name to the name attribute of the product entity, map the delivery date to the time relationship between the order and the product, etc. Generate new features using the entities and relationships in the knowledge graph. For example, you can calculate statistical features such as the average price of products in an order, the total amount of an order, and the frequency of purchases by customers. It is also possible to generate more complex features based on the relationship paths in the knowledge graph, such as the cooperation history between customers and suppliers, the supply chain path of products, etc. If the supply chain path of a product is long or involves multiple complex links, it may be more difficult to obtain raw materials, thereby increasing production costs and delivery time. In this case, in order to ensure the stability of the supply chain, companies may give priority to producing products with shorter supply chain paths and relatively easy raw material acquisition. Optimization of supply chain paths can improve production efficiency and reduce delays and waste in the production process. Therefore, for products with clear supply chain paths and smooth processes, companies may give higher production priority to reduce production costs and improve market competitiveness. Fusion of features generated based on knowledge graphs with original numerical features. Features can be fused using methods such as feature concatenation, feature weighting, and feature selection. The fused feature set will contain richer information and help improve the performance of the model. Use the fused feature set to train the machine learning model. Select a suitable model according to the specific task, such as a classification model, a regression model, etc. Use a test data set to evaluate the performance of the model. The performance of the model can be evaluated using indicators such as accuracy, recall, and F1 score. Adjust and optimize the model based on the evaluation results. As order data increases and changes, the knowledge graph needs to be updated and maintained in a timely manner to ensure its accuracy and completeness. In the process of feature generation and fusion, feature selection and optimization are required to remove redundant and irrelevant features and improve the performance and generalization ability of the model. Using knowledge graphs for feature fusion is an effective method that can map order information to entities and relationships in the knowledge graph and generate new features to enhance the performance of the model. Through reasonable feature generation and fusion methods, the information in the knowledge graph can be fully utilized to improve the accuracy and generalization ability of the model.

[0022] The step of generating information features based on entities and relationships in the knowledge graph includes: Extract basic features from each entity in the knowledge graph; for numerical entities, directly use their values ​​as features; for text entities, use text processing technology to convert them into numerical features; analyze the relationship between entities, define features for each relationship based on the type and attributes of the relationship, use relationship weight and relationship distance as the numerical features of the relationship, combine the extracted entity features and relationship features to form a feature vector, and combine the feature vectors using feature concatenation or feature interaction to obtain the first feature; perform random walks on the knowledge graph to find paths related to the target relationship, and determine a set of paths as features based on the length, frequency of occurrence, and entities and relationships contained in the path; for each relationship path, extract its structural features and convert the relationship path into a low-dimensional vector representation as a path feature; when there are multiple relationship paths between an entity and multiple other entities, aggregate the features of the multiple relationship paths, and combine the aggregated path features with the original entity features to form the second feature; Specifically, a random walk is performed on the knowledge graph to find the path related to the target relationship, where the path is a series of relationship connections from one entity to another; a set of paths is determined as features according to the selection conditions, and these paths should be able to reflect the existence or non-existence of the target relationship; each selected path is converted into a feature vector, where the feature vector represents the weight and type of each relationship in the path; an eigenvalue is calculated for each eigenvector, and this value can represent the relevance or importance of the path to the target relationship. It usually involves the calculation of path probabilities or some measure of entities and relationships on the path; the selected eigenvectors and eigenvalues ​​are used to train a regression model, and the trained regression model is applied to new knowledge graph data to predict new relationships or perform other analysis tasks. The goal of this model is to predict the existence or strength of the target relationship based on the input feature vector, and the performance of the model is evaluated by cross-validation or other methods, and adjusted and optimized as needed. The trained model is applied to new knowledge graph data to predict new relationships or perform other analysis tasks.

[0023] The step of performing a random walk on the knowledge graph to find a path related to the target relationship includes: Take the order number as the starting point and the customer name as the target point, and determine that all possible paths from the order number to the customer name need to be found; start traversing the knowledge graph from the starting point, record each entity and relationship passed during the traversal process, form a path, and store all identified paths in a suitable data structure; set a path length threshold, filter out paths corresponding to the length that meets the path length threshold to obtain multiple pre-selected paths, filter out paths containing multiple different entities and relationships from the pre-selected paths based on entity and relationship diversity, evaluate the importance of each path based on the random walk path sorting algorithm, and filter out paths with higher scores based on the importance scores of the paths as paths related to the target relationship; Specifically, take the order number as the starting point and the customer name as the target point. Clearly, we need to find all possible paths from the order number to the customer name. Use graph traversal algorithms such as depth-first traversal (DFS) or breadth-first traversal (BFS) to traverse the graph from the starting point. Record each entity and relationship passed during the traversal to form a path. Store all identified paths in a suitable data structure, such as a list or graph structure, for subsequent analysis. Set a reasonable path length threshold, such as no more than 5 relationships. Filter out paths whose length does not exceed the threshold, because paths that are too long may contain too much noise and redundant information. Analyze the diversity of entities and relationships contained in the path. Filter out paths that contain multiple different entities and relationships, because these paths may better reflect the complex relationships between entities. Use a path evaluation algorithm, such as the path ranking algorithm (PRA) based on random walks, to evaluate the importance of each path. Filter out paths with higher scores based on the importance score of the path. For example, take the path from the order number to the customer name through the product name as an example: Identify the path: order number-→ product name-→ order details-→ customer name; order number-→ order details-→ customer name (skip the product name, but directly associate), assuming that the path length threshold is 3, then both paths meet the requirements. Analyze the entity and relationship diversity and find that both paths contain key entities such as order number and customer name, and the relationship is clear. Use PRA or other algorithms to evaluate the importance of the path, assuming that both paths have high scores. By identifying and screening all possible and important relationship paths, we can deeply understand the entity relationships in the knowledge graph and provide strong support for subsequent analysis and application.

[0024] The step of obtaining the historical information indicator for evaluating the production task priority includes: obtaining the number of days from the current time point to the delivery date to obtain the available time, and evaluating the urgency score of the delivery date based on the available time; obtaining the hierarchical structure and the number of parts of the order information to comprehensively evaluate the complexity of the product; obtaining the TF-IDF value of the keyword, and using the delivery date, the complexity of the product and the TF-IDF value of the keyword as the historical information indicator; It should be noted that the earlier the delivery date, the higher the priority of the production task is usually. The more complex the product, the more time and resources may be required for production, so the priority is usually higher. Products with high complexity and tight delivery dates should be given higher priority. The higher the TF-IDF value, the more important the information contained in the order is to the overall production task, so the priority is higher.

[0025] Specifically, the indicators are quantified and a weight is assigned to each indicator. The weight allocation can be determined based on factors such as the actual situation of the enterprise, production strategy, and market demand. For example, the delivery date may be considered the most important indicator and therefore assigned a higher weight; while the product complexity and the TF-IDF value of the keyword may be assigned a relatively low weight based on the specific situation. For each production task, the corresponding order information is collected, including the delivery date, product description (used to assess the product complexity), and any related keywords. Delivery date score: The score is determined based on the early or late delivery date. For example, a benchmark date can be set, and orders earlier than the benchmark date receive a higher score, while orders later than the benchmark date receive a lower score. Product complexity score: The complexity is assessed based on the product description and a score is assigned to it. This may require relying on professional product knowledge or experience to judge. Keyword TF-IDF score (hypothetical calculation): Since TF-IDF values ​​are often used in the field of text analysis, certain adjustments or assumptions may need to be made in the actual production task priority assessment. For example, it can be assumed that certain keywords are associated with important orders and assigned a higher TF-IDF value. However, please note that this assumption may not be accurate or reliable, so it needs to be used with caution in actual applications. According to the score and weight of each indicator, calculate the comprehensive score of each production task. The calculation formula of the comprehensive score can be: Comprehensive score = delivery date score × delivery date weight + product complexity score × product complexity weight + keyword TF-IDF value score × keyword TF-IDF value weight. Finally, sort the production tasks according to the comprehensive score to get the priority sequence. The production tasks with higher scores have higher priorities and should be arranged for production first; in addition to considering traditional factors such as delivery date and production cycle, the priority of the order is evaluated in combination with the TF-IDF value. Keywords with high TF-IDF values ​​often represent important information in the document. In the production scheduling plan, if the keywords of an order have a high TF-IDF value, it may mean that the order contains important information or requirements and needs to be given priority. For example, if the keywords of an order are highly correlated with current market demand, hot-selling products, or customer priorities, then the TF-IDF value of the order may be higher, and thus be given a higher priority; S3: Obtain current production resource information, wherein the production resource information includes production equipment information, human resource information, and raw material inventory information; formulate production scheduling information based on the production resource information and the production tasks corresponding to the priority sequence, wherein the production scheduling information includes the start time, end time, and required resources of each production task; The production scheduling information is formulated based on the production tasks and production resource information corresponding to the priority sequence, and the steps include: obtaining all the production tasks to be executed and evaluating the demand time of each production task; determining the production resource information corresponding to each production task, obtaining the working time of the production resource information, allocating a start time for each production task based on the working time, and calculating the end time according to the demand time and start time of the production task; Specifically, production equipment information: Equipment name: List the names of all production equipment. Equipment status: Whether each equipment is currently idle, in use or under maintenance. Capacity: The maximum production capacity and / or efficiency of each equipment. Available time period: The time period in which each equipment is available (such as daily working hours, rest days, etc.); Human resources information: Employee name: List the names of all employees involved in production. Skills: The skill areas in which each employee is good at. Schedule: The attendance time, rest time and holiday arrangements of each employee. Work efficiency: The work efficiency or production capacity of each employee. Raw material inventory information: Material name: List the names of all required raw materials. Inventory: The current inventory of each raw material. Supplier information: The supplier and delivery cycle of each raw material. Safety stock: The minimum inventory threshold of each raw material; Determine the priority of each task based on factors such as customer requirements, delivery date, order value, etc. For each production task, analyze the required equipment, employees and raw materials. Determine the time required for each task, including preparation time, production time and cleanup time. Start time: Assign a start time to each task based on the availability of equipment, employees and raw materials. End time: Calculate the end time based on the time required and the start time of the task. Required resources: List the required equipment, employees, and raw materials for each task in detail. Consider the time for equipment maintenance, employee rest, and raw material replenishment to ensure the feasibility of the plan. Use scheduling software or algorithms to optimize the plan, reduce idle time, and improve production efficiency. Evaluate potential bottlenecks in the plan, such as equipment failure, employee absence, or raw material shortages, and develop countermeasures.

[0026] The present invention pre-processes the text content in the order information through natural language processing technology, and calculates the TF-IDF value of each phrase in the text content, thereby extracting keywords. Keywords can reflect the important needs and preferences of customers, and provide an important basis for the priority evaluation of production tasks. At the same time, the method also comprehensively considers factors such as delivery date and product complexity, and establishes a scientific and objective production task priority evaluation system; through natural language processing and TF-IDF value calculation, keywords in order information can be accurately extracted to reflect the important needs and preferences of customers, which helps enterprises to understand customer needs more accurately, and can flexibly adjust production scheduling plans according to customer preferences, formulate production scheduling plans that are more in line with customer expectations, and realize the intelligentization of the production scheduling process; a scientific and objective production task priority evaluation system is established, which comprehensively considers multiple factors such as delivery date, product complexity and TF-IDF value of keywords. It helps enterprises to arrange production tasks more reasonably, improve production efficiency and reduce production costs. It helps enterprises to improve production scheduling efficiency.

[0027] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. 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 can be combined in any one or more embodiments or examples in a suitable manner.

[0028] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent production scheduling method based on natural language processing (NLP) algorithm, characterized in that: The following steps are involved: Obtain order information to be scheduled, where the order information includes product name, product model, product quantity and delivery date, pre-process the text content in the order information based on natural language processing and calculate the TF-IDF value of each phrase in the text content, and extract keywords in the text content based on the TF-IDF value; Obtain information indicators for evaluating the priority of production tasks, where the information indicators include delivery date, product complexity, and TF-IDF value of keywords, and each production task corresponds to an order information; evaluate the priority of the production task corresponding to each order information based on the information indicators of the production task priority to obtain a priority sequence; Obtain current production resource information, wherein the production resource information includes production equipment information, human resource information, and raw material inventory information; formulate production scheduling information based on the production resource information and production tasks corresponding to the priority sequence, wherein the production scheduling information includes the start time, end time, and required resources of each production task.

2. The intelligent production scheduling method based on natural language processing (NLP) algorithm according to claim 1 is characterized by: The steps of preprocessing the text content in the order information based on natural language processing and calculating the TF-IDF value of each phrase in the text content, and extracting keywords in the text content based on the TF-IDF value include: Obtain the text content in the order information, remove meaningless symbols in the text content, divide the parsed text content into phrases, and obtain a phrase set; Formulate stop word screening rules, where the screening rules include screening conditions and trigger points; Remove the stop words in the text content based on the stop word screening rules to obtain a set of valid phrases; The TF-IDF value of each valid phrase is calculated, and the valid phrase corresponding to the TF-IDF value that meets the preset threshold is used as a keyword.

3. The intelligent production scheduling method based on natural language processing (NLP) algorithm according to claim 2 is characterized by: The step of formulating stop word screening rule information comprises: Get the standard number threshold corresponding to the historical stop words; Set multiple threshold points, each threshold point corresponds to a standard number threshold; A plurality of threshold points are arranged in a straight line to obtain a threshold line, a common point is set corresponding to the threshold line, a connection channel between the common point and each threshold point is established, and a trigger point and a trigger condition corresponding to the trigger point are set corresponding to the connection channel; Based on the trigger condition, the trigger point corresponding to the common point is started to obtain the target threshold point corresponding to the common point, and the standard number threshold corresponding to the target threshold point is used as the screening condition; The filtering rules are obtained based on the filtering conditions and the trigger points.

4. The intelligent production scheduling method based on natural language processing (NLP) algorithm according to claim 3 is characterized by: The step of obtaining the standard number threshold corresponding to the historical stop words includes: Obtain stop words in the text content of historical order information to obtain historical stop words, obtain the historical occurrence counts of each historical stop word in each text content to obtain multiple historical occurrence counts corresponding to each stop word; The average historical occurrence count of each historical stop word is evaluated based on multiple historical occurrence counts, and the average historical occurrence count is used as the standard count threshold corresponding to the historical stop word.

5. The intelligent production scheduling method based on natural language processing (NLP) algorithm according to claim 2 is characterized by: The step of removing stop words from the text content based on the stop word screening rule comprises: Based on the standard frequency threshold corresponding to each historical stop word, multiple standard frequency thresholds are obtained, and the multiple standard frequency thresholds are matched one by one with multiple threshold points; Get the phrase in the order information, place it at a common point, use the phrase and the historical stop words as the trigger condition, start the trigger point corresponding to the trigger condition to obtain the target trigger point; The threshold point corresponding to the connection channel where the target trigger point is located is used as the target threshold point, and the standard number threshold corresponding to the target threshold point is used as the screening condition of the phrase; The number of occurrences of each phrase in each text content is counted, and the phrases corresponding to the number of occurrences that do not exceed the corresponding standard number threshold are regarded as stop words, and the stop words are removed.

6. The intelligent production scheduling method based on natural language processing (NLP) algorithm according to claim 2 is characterized by: The step of calculating the TF-IDF value of each valid phrase includes: Set is the set of valid phrases, Represents each valid phrase, set It is the collection of text content of all order information. Represents the text content of each order information; the TF-IDF model is as follows: , ,in, is the number of occurrences of valid phrases, is the inverse document frequency, For valid phrases in text content The number of occurrences in For text content The sum of the number of valid phrases in , is the total number of text contents in the order information. To contain a valid phrase The total number of text contents, the TF-IDF value is and The product value of ,in, is the TF-IDF value of the corresponding valid phrase.

7. The intelligent production scheduling method based on natural language processing (NLP) algorithm according to claim 1, characterized in that: The step of evaluating the priority of the production task corresponding to each order information based on the information indicator of the production task priority to obtain the priority sequence includes: Build a knowledge graph, map the order information to the knowledge graph, and generate information features related to the order; Obtain historical information indicators for evaluating production task priorities, where the information indicators include delivery date, product complexity, and TF-IDF values ​​of keywords; The information features generated by the knowledge graph are integrated with the historical information indicators to form a feature set; Set weight values ​​for each feature in the corresponding feature set; The fused feature set and weight value are used to comprehensively calculate the priority score of each production task, and the production tasks corresponding to the priority scores are sorted in descending order to obtain a priority sequence.

8. The intelligent production scheduling method based on natural language processing (NLP) algorithm according to claim 7 is characterized by: The steps of building a knowledge graph, mapping order information to the knowledge graph, and generating information features related to the order include: Define entities and relationships, where entities include order number, product name, quantity, delivery date, and customer name in order information, and relationships include containment and belonging relationships; build a knowledge graph based on entities and relationships; Assign a unique identity to each entity and relationship, and define clear types and attributes for each relationship; set an identifier for each date range or specific product, create nodes and relationships in the graph database, and store the entities, relationships and identifiers together in the graph database to obtain a knowledge graph; Preprocess the order data, extract order information, and map the order information to entities and relationships in the knowledge graph; Information features are generated based on entities and relationships in the knowledge graph, wherein the information features include a first feature and a second feature, the first feature is an information feature generated based on entities and relationships, and the second feature is an information feature generated based on relationship paths in the knowledge graph.

9. The intelligent production scheduling method based on natural language processing (NLP) algorithm according to claim 8, characterized in that: The step of generating information features based on entities and relationships in the knowledge graph includes: Extract basic features from each entity in the knowledge graph; for numeric entities, use their values ​​directly as features; For text entities, use text processing technology to convert them into numerical features; Analyze the relationship between entities, define features for each relationship according to the type and attributes of the relationship, use the relationship weight and relationship distance as the numerical features of the relationship, combine the extracted entity features and relationship features to form a feature vector, and combine the feature vectors by feature concatenation or feature interaction to obtain the first feature; Perform random walks on the knowledge graph to find paths related to the target relationship, and determine a set of paths as features based on their length, frequency, and the entities and relationships they contain; For each relationship path, extract its structural features and convert the relationship path into a low-dimensional vector representation as the path feature; When there are multiple relationship paths between an entity and multiple other entities, the features of the multiple relationship paths are aggregated, and the aggregated path features are combined with the original entity features to form a second feature.

10. The intelligent production scheduling method based on natural language processing (NLP) algorithm according to claim 9, characterized in that: The step of performing a random walk on the knowledge graph to find a path related to the target relationship includes: Take the order number as the starting point and the customer name as the target point, and determine that all possible paths from the order number to the customer name need to be found; Traverse the knowledge graph from the starting point, record each entity and relationship passed through during the traversal, form a path, and store all identified paths in a suitable data structure; A path length threshold is set, and paths corresponding to the length that meets the path length threshold are screened out to obtain multiple pre-selected paths. Paths containing multiple different entities and relationships are screened out from the pre-selected paths based on entity and relationship diversity. The random walk-based path sorting algorithm evaluates the importance of each path, and according to the importance score of the path, selects the path with a higher score as the path related to the target relationship.

Images