A multi-scenario material scheduling method and device

Abstract

The present application provides a multi-scenario material scheduling method and device. The method provided in the present application includes: extracting feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples; obtaining multi-source data of the workspace to be scheduled, the multi-source data including user demand order data and real-time production data of the workspace to be scheduled; inputting the multi-source data into the neural network model, and predicting the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled; generating an initial scheduling strategy based on the traffic change trend and the business rules in the scheduling auxiliary knowledge base; using vehicle load balancing and material delivery time windows as constraints and historical templates as guidance, optimizing the initial scheduling strategy according to the real-time working status of the workspace to be scheduled, and obtaining a real-time scheduling strategy. The multi-scenario material scheduling method and device provided in the present application are used to complete material scheduling efficiently and accurately.

Description

Technical Field

[0001] The present application relates to the technical field of material scheduling, and in particular to a multi-scenario material scheduling method and device. Background Art

[0002] In today's highly information-based manufacturing industry, digital workshops, as an important part of modern manufacturing, are gradually becoming a key link in improving production efficiency, optimizing resource allocation and reducing operating costs. With the widespread application of technologies such as the Internet of Things, big data, and artificial intelligence, the material distribution system in digital workshops has also ushered in significant development.

[0003] At present, the widely used material distribution technology mainly relies on fixed scheduling algorithms and rule engines, and performs static analysis and optimization on material demand, handling equipment status and path planning in the workshop. This method has improved the accuracy and efficiency of material distribution to a certain extent, but with the continuous expansion of production scale and the rapid changes in production demand, there are information islands between various systems and equipment, resulting in the inability to seamlessly transmit and share material distribution information, affecting the efficiency and accuracy of material distribution. At the same time, due to inaccurate material demand forecasts and inadequate material supply assessments, material waste and excessive inventory costs are prone to occur, affecting the economic benefits of the enterprise.

[0004] To sum up, the existing material distribution technology has obvious deficiencies in efficiency and accuracy, and it is difficult to meet the efficient, accurate and flexible material distribution needs of digital workshops. Summary of the invention

[0005] In view of this, the present application provides a multi-scenario material scheduling method and device to complete material scheduling efficiently and accurately.

[0006] Specifically, the present application is implemented through the following technical solutions:

[0007] The first aspect of the present application provides a multi-scenario material scheduling method, the method comprising:

[0008] Extracting feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples, wherein the scheduling auxiliary knowledge base includes at least business rules, historical templates, and a network map of the workspace to be scheduled;

[0009] Acquire multi-source data of the workspace to be scheduled, wherein the multi-source data includes user demand order data and real-time production data of the workspace to be scheduled, wherein the production data includes at least order production data, production machine data and production resource data;

[0010] Inputting the multi-source data into the neural network model, and predicting the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled;

[0011] Generate an initial dispatch strategy according to the traffic change trend and the business rules in the dispatch auxiliary knowledge base, wherein the initial dispatch strategy includes a dispatch scheme and a dispatch priority;

[0012] Taking vehicle load balancing and material delivery time window as constraints and the historical template as a guide, the initial scheduling strategy is optimized according to the real-time working status of the workspace to be scheduled to obtain a real-time scheduling strategy, and the historical template includes a positive template and a negative template.

[0013] The second aspect of the present application provides a multi-scenario material scheduling device, the device comprising an acquisition module, a prediction module, a processing module and an optimization module; wherein:

[0014] The acquisition module is used to extract feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples, wherein the scheduling auxiliary knowledge base includes at least business rules, historical templates, and a network map of a workspace to be scheduled;

[0015] The acquisition module is further used to acquire multi-source data of the workspace to be scheduled, wherein the multi-source data includes user demand order data and real-time production data of the workspace to be scheduled, wherein the production data includes at least order production data, production machine data and production resource data;

[0016] The prediction module is used to input the multi-source data into the neural network model, and predict the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled;

[0017] The processing module is used to generate an initial scheduling strategy according to the traffic change trend and the business rules in the scheduling auxiliary knowledge base, and the initial scheduling strategy includes a scheduling scheme and a scheduling priority;

[0018] The optimization module is used to optimize the initial scheduling strategy based on the real-time working status of the workspace to be scheduled, with vehicle load balancing and material distribution time window as constraints and the historical template as a guide, to obtain a real-time scheduling strategy. The historical template includes a positive template and a negative template.

[0019] The third aspect of the present application provides a multi-scenario material scheduling device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any one of the methods provided in the first aspect of the present application are implemented.

[0020] The fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods provided in the first aspect of the present application.

[0021] The multi-scenario material scheduling method and device provided by the present application firstly extracts feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples, then obtains multi-source data of the workspace to be scheduled, the multi-source data includes user demand order data and real-time production data of the workspace to be scheduled, and then inputs the multi-source data into the neural network model, predicts the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled, and then generates an initial scheduling strategy based on the traffic change trend and the business rules in the scheduling auxiliary knowledge base, and finally optimizes the initial scheduling strategy based on the real-time working status of the workspace to be scheduled with vehicle load balancing and material distribution time windows as constraints and historical templates as guidance, and obtains a real-time scheduling strategy. In this way, through the multi-source data collection of the workspace to be scheduled, combined with the prediction of the neural network model, it is possible to more accurately grasp the user demand order data, vehicle load status and traffic conditions, so as to formulate a more reasonable distribution plan, improve production efficiency and reduce distribution costs. Furthermore, through the multi-source data of the workspace to be scheduled, it is possible to flexibly respond to complex and changeable production environments such as various material handling vehicles, different traffic conditions and various optimization targets, meet the material distribution needs in multiple scenarios, and ensure the reliability and effect of material scheduling. Furthermore, through the accurate prediction of traffic change trends by the neural network model, the initial scheduling plan that best meets the current actual situation can be generated in combination with the business rules in the knowledge base, ensuring the efficiency and reliability of material scheduling. Furthermore, by constraining the initial scheduling strategy through vehicle load balancing and material distribution time windows, the safety of material scheduling can be ensured, accidents can be prevented, and the user experience is improved. Furthermore, through the guidance of historical templates, the accurate material scheduling method can be selected without trial and error, ensuring the accuracy of scheduling. In this way, by combining the neural network model and the knowledge base, the intelligent, real-time and accuracy of multi-scenario material scheduling are realized. In addition, the method provided by the present invention realizes the extraction of feature information in multi-source data through the reuse of some layers of the neural network layer, and at the same time uses the neural network as a whole to predict traffic change trends. The training of domain data once can complete multiple tasks, improving the efficiency of feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of a multi-scenario material scheduling method provided for an exemplary embodiment of the present application;

[0023] Figure 2A schematic diagram of a scheduling auxiliary knowledge base shown as an exemplary embodiment of the present application;

[0024] Figure 3 A schematic diagram of a scheduling auxiliary knowledge base construction process shown in an exemplary embodiment of the present application;

[0025] Figure 4 A structural diagram of a neural network model shown as an exemplary embodiment of the present application;

[0026] Figure 5 A structural diagram of a multi-scenario material scheduling device provided for an exemplary embodiment of the present application;

[0027] Figure 6 A structural diagram of a multi-scenario material scheduling device provided as an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0028] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0029] The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0030] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0031] The present application provides a multi-scenario material scheduling method and device for efficiently and accurately completing material scheduling.

[0032] The multi-scenario material scheduling method and device provided by the present application firstly extracts feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples, then obtains multi-source data of the workspace to be scheduled, the multi-source data includes user demand order data and real-time production data of the workspace to be scheduled, and then inputs the multi-source data into the neural network model, predicts the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled, and then generates an initial scheduling strategy based on the traffic change trend and the business rules in the scheduling auxiliary knowledge base, and finally optimizes the initial scheduling strategy based on the real-time working status of the workspace to be scheduled with vehicle load balancing and material distribution time windows as constraints and historical templates as guidance, and obtains a real-time scheduling strategy. In this way, through the multi-source data collection of the workspace to be scheduled, combined with the prediction of the neural network model, it is possible to more accurately grasp the user demand order data, vehicle load status and traffic conditions, so as to formulate a more reasonable distribution plan, improve production efficiency and reduce distribution costs. Furthermore, through the multi-source data of the workspace to be scheduled, it is possible to flexibly respond to complex and changeable production environments such as various material handling vehicles, different traffic conditions and various optimization objectives, meet the material distribution needs in multiple scenarios, and ensure the reliability and effectiveness of material scheduling. Furthermore, through the accurate prediction of traffic change trends by the neural network model, combined with the business rules in the knowledge base, the initial scheduling plan that best meets the current actual situation can be generated, ensuring the efficiency and reliability of material scheduling. Furthermore, by constraining the initial scheduling strategy through vehicle load balancing and material distribution time windows, the safety of material scheduling can be ensured, accidents can be prevented, and the user experience can be improved. Furthermore, through the guidance of historical templates, the accurate material scheduling method can be selected without trial and error, ensuring the accuracy of scheduling. In this way, by combining the neural network model and the knowledge base, the intelligent, real-time and accurate scheduling of materials in multiple scenarios is realized.

[0033] Specific embodiments are given below to introduce the technical solution of the present application in detail.

[0034] Figure 1 This is a flowchart of a multi-scenario material scheduling method provided by an exemplary embodiment of the present application. Figure 1 , the method provided in this embodiment may include:

[0035] S101. Extract feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples, wherein the scheduling auxiliary knowledge base includes at least business rules, historical templates, and a network map of the workspace to be scheduled.

[0036] Specifically, the historical material data includes all scheduling data of the digital workshop in history. It should be noted that the data type included in the historical material data is set according to actual needs, and is not limited in this embodiment.

[0037] In specific implementation, the historical material data may include historical delivery data, real-time traffic information and the user evaluation.

[0038] Furthermore, the neural network model extracts feature data from the historical material dispatch data. For example, in one embodiment, feature extraction is performed through data such as delivery distance, estimated delivery time, and historical route planning in the historical material data to obtain corresponding feature data.

[0039] Furthermore, real-time traffic data can guide the dispatching process. In specific implementation, real-time traffic data is used to determine the congestion of roads, as well as the travel time on roads and other dispatch-related data. These real-time traffic data are used to ensure the reliability of dispatching and prevent the planned path from being unavailable or the planned path from being unable to complete normal dispatching.

[0040] Furthermore, user evaluation reflects the user's satisfaction with the scheduling effect, and the data in the scheduling process can be marked by user evaluation. In specific implementation, when the user evaluations corresponding to certain specific data are all lower than the preset score, it can be determined that the scheduling effect of these specific data is not ideal, and other data (that is, data with high user evaluations) are given priority when scheduling planning.

[0041] Furthermore, a scheduling auxiliary knowledge base and training samples are generated through the feature data. For example, a scheduling auxiliary knowledge base and training samples can be generated based on all data features.

[0042] It should be noted that the scheduling auxiliary knowledge base contains knowledge related to material scheduling, and the scheduling of materials can be better planned through the scheduling auxiliary knowledge base.

[0043] In specific implementation, the scheduling auxiliary knowledge base adopts a structured knowledge representation method. For example, the scheduling auxiliary knowledge base can be constructed through a rule base, a case base, an ontology, etc., and the knowledge is stored in a computer-understandable form. It should be noted that the rule base is used to store industry rules and expert experience; the case base is used to store historical cases and solutions; and the ontology is used to define concepts, relationships, and attributes in the field.

[0044] Furthermore, business rules are requirements that must be met when scheduling materials; historical templates are cases in which material scheduling is successfully completed in historical material scheduling data; and the network map of the workspace to be scheduled is a map of the place where material scheduling is performed. For example, in one embodiment, the network map of the workspace to be scheduled is a floor plan of a digital workshop.

[0045] Furthermore, when collecting data, the Internet of Things (IoT) technology can be used to collect real-time work order information, material requirements, location, speed, load, power and other information of handling equipment, as well as key data such as warehouse inventory, line inventory, order status, etc. The data is transmitted to the system center in real time through the API interface or message queue for subsequent analysis and processing.

[0046] Furthermore, data preprocessing can include collection, cleaning, conversion, and loading.

[0047] Furthermore, the network map of the workspace to be scheduled may include real-time updated warehouses, lineside warehouses, workstations and transportation network maps.

[0048] Figure 2 This is a schematic diagram of a scheduling auxiliary knowledge base shown in an exemplary embodiment of the present application. Figure 2 The specific knowledge source of the dispatch auxiliary knowledge base is set according to actual needs, and is not limited in this embodiment. For example, in one embodiment, the knowledge entry of the dispatch auxiliary knowledge base can be completed by manual entry and system collection.

[0049] For further information, please refer to Figure 2 The knowledge entered into the dispatch auxiliary knowledge base may include implicit knowledge, explicit knowledge and system knowledge. In specific implementation, for example, in one embodiment, implicit knowledge includes the knowledge of business experts and employee experience; explicit knowledge includes business operation manuals, rules and regulations documents and product specifications; system knowledge includes business rules, historical cases and distribution network maps.

[0050] For further information, please refer to Figure 2 ,The scheduling auxiliary knowledge base can have AI capabilities such as self-evolution, self-transformation, intelligent matching and intelligent association.,It should be noted that the data of the scheduling auxiliary,knowledge base can come from multiple sources.

[0051] For further information, please refer to Figure 2 ,The scheduling auxiliary knowledge base can provide a variety of knowledge ,services and be interconnected through a variety of networks.

[0052] Figure 3 This is a schematic diagram of a scheduling auxiliary knowledge base construction process shown in an exemplary embodiment of the present application. Figure 3,The user or external system inputs data through the interactive interface of the ,scheduling auxiliary knowledge base, and the data is compared and ,queried through the knowledge base.

[0053] For further information, please refer to Figure 3 ,The data in the scheduling auxiliary knowledge base is stored after vectorization.

[0054] A specific embodiment is given below to describe in detail the process of extracting feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples:

[0055] (1) Extracting feature data from historical material scheduling data based on the input layer, wherein the historical material scheduling data at least includes historical delivery data, real-time traffic information, and the user evaluation, and the feature data is valid information in the historical material scheduling data.

[0056] Figure 4 This is a structural diagram of a neural network model shown in an exemplary embodiment of the present application. Figure 4 ,The neural network model includes an input layer, a hidden layer and an output layer.

[0057] Furthermore, the input layer is used to receive multi-source data of the workspace to be scheduled.

[0058] Furthermore, the multi-source data of the workspace to be scheduled is data collected in real time and contains information from many aspects. In specific implementation, the specific data types contained in the multi-source data of the workspace to be scheduled are set according to actual needs, and this is not limited in this embodiment. For example, in one embodiment, the multi-source data of the workspace to be scheduled includes user demand order data and production data produced in real time in the workspace to be scheduled. For another example, in another embodiment, the multi-source data of the workspace to be scheduled includes material requirements, delivery time, route, vehicle status, real-time traffic information, delivery result evaluation, etc.

[0059] For further information, please refer to Figure 4 The hidden layer is used to capture the complex relationships and long-term dependencies between data through a multi-layer long short-term memory network (LSTM) or self-attention mechanism network (Transformer) structure.

[0060] It should be noted that due to the complexity and uncertainty in the material distribution process, the long short-term memory network can effectively handle the long-term dependencies in time series data and is suitable for predicting traffic conditions, distribution strategies, etc. in the future.

[0061] For further information, please refer to Figure 4The output layer is used to output traffic change trends. In specific implementation, for example, in one embodiment, the output layer outputs traffic change trends such as the best delivery route, the estimated arrival time, and the scheduling priority of the delivery task.

[0062] It should be noted that before extracting feature data, the historical material scheduling data should be cleaned and managed to form structured data to prepare for subsequent sorting.

[0063] When this step is implemented specifically, feature extraction is performed on valid information in historical material scheduling data to obtain feature data.

[0064] (2) Based on the characteristic data, training samples with the same dimension and the scheduling object as the carrier are generated, and the input samples of the training samples are the differences between the material demand and the material supply at different times.

[0065] Specifically, the feature data is processed to obtain information related to material scheduling. For example, the delivery distance, estimated time, route planning, etc. are used to locate the main data content (carrier), and these data are normalized to obtain data of the same dimension, and these data are confirmed as training samples.

[0066] It should be noted that the specific dimension of the training sample is determined according to actual needs and is not limited in this embodiment.

[0067] Furthermore, the input samples for obtaining the training samples may be the difference between the demand quantity of the material and the actual supply quantity of the material at different times.

[0068] (3) Marking the entity objects in the valid information, and analyzing the association relationship between the entity objects based on the historical material scheduling data.

[0069] Specifically, the entity object in the effective information is an object with actual physical meaning. In specific implementation, the entity object in the effective information is annotated, and the entity object is determined according to the annotation.

[0070] Furthermore, the association relationship between various entity objects is determined based on the historical material scheduling data.

[0071] When this step is implemented specifically, it is possible to automatically mark entity objects such as materials, orders, work orders, material requirements, handling equipment, routes, and distribution result evaluations and their attribute information, determine the semantic information of each entity object in combination with historical material scheduling data, and automatically form the corresponding association relationships between entity objects based on the semantics of each entity object.

[0072] (4) Generate a target triple based on the entity object and the association relationship, use the user evaluation as the evaluation value of the target triple, and construct the scheduling auxiliary knowledge base based on the target triple.

[0073] Specifically, the two entity objects corresponding to the association relationship and the association relationship can form a set of target triples. It can be understood that each association relationship uniquely corresponds to a target triple, and the number of target triples is the same as the number of association relationships.

[0074] It should be noted that the evaluation value corresponding to the target triplet can directly adopt the value evaluated by the user.

[0075] Furthermore, a scheduling auxiliary knowledge base is constructed based on the target triples.

[0076] The multi-scenario material scheduling method provided in this embodiment, through data cleaning and governance, the pre-processed historical material scheduling data is converted into structured data, which lays the foundation for subsequent sorting. This process makes the originally scattered and disordered historical material scheduling data orderly and queryable, which is convenient for subsequent knowledge extraction and application. Furthermore, through entity object labeling and relationship extraction, the process of effectively extracting business knowledge from data is realized. This helps to make implicit knowledge explicit, so that the generated scheduling auxiliary knowledge base can provide more basis for scheduling decisions. Furthermore, the scheduling auxiliary knowledge base is formed through the target triples. The scheduling auxiliary knowledge base integrates multi-dimensional information such as business rules, historical templates, and network maps of the workspace to be scheduled, providing strong knowledge support for the formulation of scheduling strategies. The scheduling auxiliary knowledge base can improve the intelligence and accuracy of scheduling decisions.

[0077] Furthermore, as time goes by and the environment changes, the knowledge in the scheduling auxiliary knowledge base will gradually become outdated or invalid. Therefore, the scheduling auxiliary knowledge base needs to be updated regularly to maintain its timeliness and accuracy.

[0078] A specific embodiment is given below to introduce the process of scheduling auxiliary update in detail:

[0079] (1) Capturing update identification data from the multi-source data, wherein the update identification data at least includes traffic policy adjustment data and distribution technology update data.

[0080] Specifically, the update identification data is data that promotes the automatic update of the scheduling auxiliary knowledge base. In specific implementation, when the update identification data is detected, it is determined to update the scheduling auxiliary knowledge base.

[0081] It should be noted that the specific data types included in the update identification data are determined according to actual needs, and are not limited in this embodiment. For example, in one embodiment, the update identification data includes traffic policy adjustment data and distribution technology update data.

[0082] (2) Cleaning and verifying the update identification data to obtain update data, and updating the scheduling auxiliary knowledge base based on the update data, wherein the update processing at least includes data addition and data replacement.

[0083] In specific implementation, after the update identification data is cleaned and verified, the data obtained is the update data. The update identification data can be cleaned and verified based on a traditional data processing method, or based on a neural network data processing method, which is not limited in this embodiment.

[0084] Furthermore, the updated data may be directly added to the scheduling auxiliary knowledge base to complete the updating of the scheduling auxiliary knowledge base; or the updated data may replace the original data of the same type in the scheduling auxiliary knowledge base to complete the updating of the scheduling auxiliary knowledge base.

[0085] It should be noted that thresholds and update rules can be set for the scheduling auxiliary knowledge base. When the value of the data type corresponding to the threshold exceeds the threshold, the scheduling auxiliary knowledge base is updated; when the update rule is met, the scheduling auxiliary knowledge base is updated.

[0086] Furthermore, the scheduling auxiliary knowledge base can also use machine learning technology to automatically extract new rules or patterns from large amounts of data to further enrich the content of the scheduling auxiliary knowledge base.

[0087] Furthermore, key indicators such as response time, scheduling accuracy, and resource utilization are monitored and evaluated. Through A / B testing, simulation and other methods, the neural network model, scheduling algorithm and system integration solution are continuously optimized to improve the overall performance of the system.

[0088] (3) Performing periodic error correction and supplementation processing on the scheduling auxiliary knowledge base based on the expert opinions.

[0089] Specifically, expert opinions may be obtained at preset time intervals. It should be noted that the specific duration of the preset time is determined according to actual needs, and is not limited in this embodiment.

[0090] In the specific implementation, domain experts or system administrators will review and evaluate the current scheduling auxiliary knowledge base, periodically correct errors that may occur during the automatic update process, supplement important missing information, and adjust the structure and content of the scheduling auxiliary knowledge base according to actual needs.

[0091] The multi-scenario material scheduling method provided in this embodiment monitors and evaluates changes in data sources in real time, and can automatically trigger the update process. This real-time update mechanism ensures that the data in the scheduling auxiliary knowledge base is always synchronized with the external environment, avoiding decision-making errors caused by information lags. Furthermore, manual review can correct errors that occur during the automatic update process, supplement important missing information, and adjust the structure and content of the scheduling auxiliary knowledge base according to actual needs, thereby improving the accuracy and completeness of the scheduling auxiliary knowledge base. In this way, by continuously updating the scheduling auxiliary knowledge base, it is possible to cope with various emergencies and uncertain factors, ensure stable operation in various complex and changing scenarios, and improve the robustness of material scheduling.

[0092] A specific embodiment is given below to introduce the training process of the neural network model in detail:

[0093] (1) Determine the network architecture of the benchmark network based on the data type of the training samples of the multi-scenario material scheduling method.

[0094] Specifically, different multi-scenario material scheduling method training samples correspond to different network architectures. It is understandable that in different material scheduling scenarios, due to the different emphases of material scheduling, the required capabilities of the neural network model are also different.

[0095] In specific implementation, when the data in the training samples of the multi-scenario material scheduling method need to obtain long-term dependencies in the time series, the network architecture is determined to be a long short-term memory network.

[0096] It should be noted that when obtaining the benchmark model, for material scheduling problems, it is necessary to consider using a recurrent neural network (RNN) or a long short-term memory network (LSTM). This type of network is suitable for processing sequence data and can capture long-term dependencies in time series.

[0097] When implementing it, first determine the number of layers in the network and the number of neurons in each layer. For example, in one embodiment, a neural network with two LSTM layers can be used, each LSTM layer is followed by a dropout layer to prevent overfitting. Then select the activation function: for the hidden layer, the ReLU activation function can be used; for the output layer, the linear activation function can be used.

[0098] Furthermore, for regression problems, the mean square error (MSE) can be used as the loss function. In terms of optimizer, the Adam optimizer can be used as the optimizer based on its adaptive learning rate adjustment feature.

[0099] Furthermore, you can use the Python framework to implement the neural network model. When implementing it, you need to install the required libraries and then implement the code for forward propagation, backpropagation, and weight update.

[0100] (2) Determining structural parameters in the network architecture based on the prediction object, where the structural parameters include at least the number of network layers and the number of neurons.

[0101] Specifically, the structural parameters can be determined based on the corresponding relationship between the predicted object and the structural parameters. In specific implementation, based on the characteristics of the predicted object, a neural network with two long short-term memory network layers is selected, and each long short-term memory network layer is followed by a dropout layer to prevent overfitting.

[0102] (3) Extracting effective knowledge from the training samples and annotating the benchmark model based on the effective knowledge.

[0103] In the specific implementation, the data related to material scheduling in the training samples are determined as effective knowledge, and the data in the benchmark model corresponding to the effective knowledge are labeled.

[0104] (4) Constructing a neural network model based on the training samples.

[0105] Specifically, the trained neural network model is obtained from the training sample data benchmark model.

[0106] Furthermore, for the hidden layer of the neural network model, a ReLU activation function can be used; for the output layer of the neural network model, a linear activation function can be used.

[0107] Furthermore, after completing the model training, the test set can be used to evaluate the performance of the model. If the model performs poorly, the model can be optimized by adjusting the network architecture, increasing the amount of training data, changing the loss function or optimizer, etc. In addition, techniques such as cross-validation can be used to prevent overfitting.

[0108] Furthermore, after determining the optimal model architecture and hyperparameters, it can be deployed to a production environment. Specifically, during the deployment process, the trained model is saved to disk, and code is written to load the model and perform inference on it. At the same time, the performance of the model needs to be checked regularly, and continuous learning and improvement needs to be performed based on new data.

[0109] Furthermore, in order to maintain the effectiveness of the model, the model can be updated regularly to adapt to new data and environmental changes. In addition, the generalization ability of the model can be monitored to ensure that it still performs well on new data.

[0110] The multi-scenario material scheduling method provided in this embodiment can capture the complex relationships and long-term dependencies in time series data through a carefully designed network architecture. This enables the model to accurately predict key information such as material demand, traffic conditions, and delivery time in the future, providing a reliable basis for scheduling decisions. Furthermore, the neural network model is trained with a large amount of data to adapt to complex environmental factors in different scenarios, ensuring the reliability and accuracy of the neural network model prediction.

[0111] S102, obtaining multi-source data of the workspace to be scheduled, wherein the multi-source data includes user demand order data and real-time production data of the workspace to be scheduled, wherein the production data at least includes order production data, production machine data and production resource data.

[0112] Specifically, the user demand order data included in the multi-source data is sent by the user. In specific implementation, when the user sends an order, the user demand order data is determined according to the user's order.

[0113] Furthermore, the production data of the workspace to be scheduled is obtained in real time. For example, in one embodiment, when receiving a user's order, the order production data, production machine data and production resource data of the current workspace to be scheduled are obtained in real time to obtain the production data.

[0114] Furthermore, multi-source data may include work order information, material requirements, location, speed, load, power and other information of handling equipment, as well as warehouse inventory, line inventory, order status and other structured text, semi-structured text, unstructured text, tables, documents, pictures, audio and video and / or web pages.

[0115] Furthermore, the data sources of multi-source data may include enterprise resource planning system (ERP), production execution system (MES), warehouse management system (WMS), warehouse control system (WCS), robot control system (RCS), Internet of Things platform, unit management and control system, etc.

[0116] Furthermore, the data collection method may include data collection, system connection, manual import and / or input.

[0117] Further, after obtaining the multi-source data of the workspace to be scheduled, the multi-source data can be preprocessed. In specific implementation, key structured information is obtained from the unstructured text of the multi-source data, noise and interference information in the text data are cleaned, garbled characters and illegal characters in the structured data are removed, and grammatical errors and / or logically unsmooth statements are repaired. For example, in one embodiment, data collection includes collecting historical distribution data, real-time traffic data, distribution result evaluation, etc. Data cleaning includes removing invalid data such as outliers and missing values ​​to ensure data quality. Feature extraction includes extracting useful features from the original data, such as distribution distance, estimated time, path planning, etc. Data normalization includes scaling the feature data to the same dimension to improve model training efficiency and performance. For another example, in another embodiment, feature engineering can be used to design new feature variables based on the characteristics of the problem and business logic to enhance the expressiveness of the model. In other words, the difference between the demand and supply of each material at each time point can be calculated as a new feature variable. For example, in another embodiment, data normalization may be used to scale all numerical features into a uniform range to help the neural network converge faster, which may be achieved through MinMaxScaler or StandardScaler.

[0118] When this step is implemented, real-time distribution data collection can be carried out. Through the Internet of Things (IoT) technology, work order information, material requirements, location, speed, load, power and other information of handling equipment, as well as key data such as warehouse inventory, line inventory, order status, etc. can be collected in real time. The data is transmitted to the system center in real time through the API interface or message queue for subsequent analysis and processing.

[0119] Furthermore, combined with the above examples, knowledge can be effectively captured from a large amount of labeled and unlabeled data to perform data annotation. It should be noted that the model architecture is designed according to the training requirements of multi-source data, high reuse, semantic analysis and causal analysis, to assist in the extraction of entity, attribute and relationship data in the scheduling auxiliary knowledge base, and expand the generalization ability of the scheduling auxiliary knowledge base. At the same time, a comprehensive performance evaluation system is established to monitor and evaluate key indicators such as the system's response time, scheduling accuracy, and resource utilization. Through A / B testing, simulation and other methods, the neural network model, scheduling algorithm and system integration solution are continuously optimized to improve the overall performance of the system.

[0120] Furthermore, the goal of the material scheduling problem is to optimize the material flow within the workshop to improve production efficiency and reduce inventory costs. For example, in one embodiment, it is necessary to predict which production line each material should be assigned to at different time points to meet production needs and minimize material inventory. In other words, it can be modeled as a regression problem, where the input features include material type, quantity, arrival time, and production line requirements, and the output is the optimal ratio of each material to be assigned to each production line at each time point.

[0121] S103: input the multi-source data into the neural network model, and predict the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled.

[0122] Specifically, the network map of the workspace to be scheduled is used to limit the prediction range of the neural network model. It is understandable that the results predicted by the neural network model cannot exceed the range of the network map of the workspace to be scheduled, and the results predicted by the neural network model should represent traffic according to the existing roads in the network map of the workspace to be scheduled.

[0123] Furthermore, the future traffic changes of the workspace to be scheduled can be determined through the traffic change trend.

[0124] In the specific implementation, in this step, multi-source data are input into the neural network model, and the traffic change trend is output through the neural network model.

[0125] The constraint prediction of the traffic change trend of the workspace to be scheduled based on the neural network model and the network map of the workspace to be scheduled includes:

[0126] Determine the path constraints of the occupied paths and their occupied time periods according to the network map of the workspace to be scheduled;

[0127] Determine a real-time spatial network map at each time point according to the path constraint, wherein the real-time spatial network map does not include the path corresponding to the path constraint;

[0128] The neural network model predicts the optimal path within the scope of the real-time spatial network map based on the usage of the dispatched vehicles and the resources to be dispatched;

[0129] The step of optimizing the initial scheduling strategy according to the real-time working state of the workspace to be scheduled to obtain a real-time scheduling strategy includes:

[0130] The real-time working status in the workspace to be scheduled is collected in real time, and the real-time working status represents the status of resources participating in the scheduling work in the workspace to be scheduled; the uncalled vehicles are determined according to the real-time working status, and the material delivery time window is determined based on the material delivery order demand; the load balancing of the uncalled vehicles and the minimum material delivery time are taken as the goals, and the material delivery time window is taken as the constraint, the initial scheduling strategy is optimized to obtain the real-time scheduling strategy; wherein, the optimization is optimized in the direction of the positive template and away from the negative template.

[0131] S104: Generate an initial scheduling strategy according to the traffic change trend and the business rules in the scheduling auxiliary knowledge base, wherein the initial scheduling strategy includes a scheduling scheme and a scheduling priority.

[0132] Specifically, business rules are used to restrict the scheduling of materials. When implementing, all requirements specified in the business rules should be met by the initial scheduling strategy.

[0133] Furthermore, through business rule restrictions, the reliability and scientificity of material scheduling can be guaranteed, and accidents in material scheduling can be prevented.

[0134] Furthermore, by combining traffic change trends and business rules, an initial scheduling strategy suitable for future traffic conditions and meeting business rule requirements can be generated. Material scheduling can be completed through multiple scheduling schemes in the initial scheduling strategy.

[0135] It should be noted that the multiple scheduling schemes in the initial scheduling strategy correspond to different scheduling priorities. In specific implementation, the scheduling priority corresponding to the scheduling scheme with high accuracy and high efficiency can be set to a larger value.

[0136] A specific embodiment is given below to introduce the generation process of the initial scheduling strategy in detail:

[0137] (1) Retrieving the historical template based on the traffic change trend to obtain a target template that matches the traffic change trend.

[0138] In specific implementation, the historical model whose traffic conditions in the historical template are closest to the traffic change trend can be determined as the target template.

[0139] It can be understood that the traffic conditions corresponding to the target model are closest to the current traffic change trend. Therefore, the material scheduling calculated using the target template is most consistent with the future traffic conditions, and the logistics scheduling reliability and efficiency are the highest.

[0140] (2) Generating the initial scheduling strategy based on the business rule constraints on the target template; wherein the initial scheduling strategy is not generated in real time.

[0141] Optionally, the business rules include at least traffic rules, network map rules and vehicle rules.

[0142] During specific implementation, the part of the target template corresponding to the business rules needs to meet the requirements of the business rules.

[0143] It should be noted that the initial scheduling strategy is continuously generated. For example, when there is a traffic change in the workspace to be scheduled, the initial scheduling strategy is generated. In other words, the initial scheduling strategy is continuously generated and has nothing to do with the material scheduling order issued by the user.

[0144] The multi-scenario material scheduling method provided in this embodiment can quickly generate a preliminary scheduling strategy based on the prediction results of the neural network model and the rules in the scheduling auxiliary knowledge base, ensuring that a feasible initial scheduling strategy can be immediately provided when facing real-time dynamic traffic conditions. Furthermore, the process of generating the initial scheduling strategy combines the prediction ability of the neural network model and the business rules in the scheduling auxiliary knowledge base, reflecting the characteristics of intelligent decision-making. This helps to improve the accuracy and scientificity of scheduling decisions.

[0145] Optionally, the data types of the business rules and the historical templates contained in the dispatch auxiliary knowledge base are logical process data, and the generating of the initial dispatch strategy according to the traffic change trend and the business rules in the dispatch auxiliary knowledge base includes:

[0146] (1) Generate current logical conditions based on the traffic change trend and current environmental conditions.

[0147] In specific implementation, the dispatch auxiliary knowledge base can be inferred and deduced based on the pre-trained reasoning engine, and the logical conditions at this time can be obtained by combining the traffic change trend and the current environment.

[0148] It should be noted that the logical conditions reflect the implicit connections of the data in the current traffic and scheduling auxiliary knowledge base.

[0149] (2) The current logical condition is input into the scheduling auxiliary knowledge base for matching to obtain the business rule and the historical template, and the logical process corresponding to the business rule and the historical template is applicable to the generation of the current logical condition.

[0150] Specifically, the data corresponding to the logical conditions in the scheduling auxiliary knowledge base are retrieved and captured to obtain business rules and historical templates.

[0151] The multi-scenario material scheduling method provided in this embodiment uses business rules, historical cases and other information in the scheduling auxiliary knowledge base, combined with the traffic change trend of the neural network model to perform logical reasoning and deduction, complete logical matching, and obtain accurate business rules and historical templates. In this way, intelligent decision support can be provided, and the accuracy and scientificity of material scheduling can be improved.

[0152] S105, taking vehicle load balancing and material delivery time window as constraints and the historical template as a guide, optimizing the initial scheduling strategy according to the real-time working status of the workspace to be scheduled to obtain a real-time scheduling strategy, wherein the historical template includes a positive template and a negative template.

[0153] Specifically, the historical template includes a positive template and a negative template. It should be noted that when the user's evaluation is good, the corresponding historical template is a positive template; when the user's evaluation is bad, the corresponding historical template is a negative template.

[0154] In a specific implementation, when generating a real-time scheduling strategy, the data in the positive template can be referenced. For example, in one embodiment, the data in the positive template can be directly added to the initial scheduling strategy. For another example, in another embodiment, the data in the positive template can be replaced with the corresponding data in the initial scheduling strategy. Further, the real-time scheduling strategy is modified based on the data in the negative template. In a specific implementation, the data in the negative template can be matched with the initial scheduling strategy, and the values ​​with a matching degree higher than a preset matching degree can be deleted.

[0155] Furthermore, vehicle load balancing and material delivery time windows are the minimum requirements that the real-time scheduling strategy must comply with. In other words, when the real-time scheduling strategy is scheduling, the vehicle load cannot exceed the vehicle load balancing, and the material scheduling time cannot exceed the material delivery time window.

[0156] Furthermore, the real-time working status is the operating data of each device in the workspace to be scheduled that is collected in fact.

[0157] When this step is implemented, on the premise of meeting the vehicle load balance and material delivery time window, follow the scheduling plan in the historical template, optimize the initial scheduling strategy according to the real-time working status, and form a real-time scheduling strategy.

[0158] A specific embodiment is given below to introduce the acquisition process of the real-time scheduling strategy in detail:

[0159] (1) Real-time collection of real-time traffic information and vehicle status within the workspace to be scheduled.

[0160] Specifically, the real-time traffic information represents the traffic situation in the workspace to be dispatched, and the vehicle status represents the operating status of each transport vehicle in the workspace to be dispatched.

[0161] In specific implementation, the actual operation data of the workspace to be scheduled can be determined based on the Internet of Things system, and the data can be analyzed to obtain real-time traffic information and vehicle status.

[0162] (2) Determine the material delivery time window based on user demand and the real-time traffic information.

[0163] Specifically, user demand is the time point when the user needs the material.

[0164] During specific implementation, user requirements can be determined based on the data filled in by the user.

[0165] Furthermore, real-time traffic information can determine the transportation time of different materials, and combined with the user's needs, the time required to complete the user's needs can be determined, and this time is determined as the material delivery time window. In other words, the material delivery time window is the total scheduling time of the material.

[0166] (3) Locating a scheduling strategy to be adjusted based on the material delivery time window, where the material delivery time of the scheduling strategy to be adjusted exceeds the material delivery time window.

[0167] In specific implementation, the scheduling policy corresponding to the material delivery time exceeding the material delivery time window is determined as the scheduling policy to be adjusted.

[0168] It should be noted that, since the initial adjustment strategy may not meet actual needs (manifested as the material delivery time exceeding the material delivery time window), it is determined as the scheduling strategy to be adjusted and adjusted.

[0169] (4) Determining a vehicle load distribution method for vehicle load balancing according to the vehicle status collected in real time.

[0170] Specifically, the load of the vehicle is re-determined according to the vehicle state to meet the vehicle load balancing. In specific implementation, the load of the vehicle with too little load is increased, and the load of the vehicle with too much load is reduced to achieve vehicle load balancing.

[0171] (5) Adding delivery vehicles to the scheduling strategy to be adjusted based on the vehicle load distribution method.

[0172] In specific implementation, whether or not additional delivery vehicles are needed and the number of additional vehicles are determined according to the new vehicle load distribution method. For example, in one embodiment, it is determined according to the vehicle load distribution method that three delivery vehicles need to be added.

[0173] (6) Planning the delivery routes of the additional vehicles and generating the real-time scheduling strategy.

[0174] In specific implementation, it is determined whether to add a delivery path according to the new vehicle load distribution method, and the delivery path is added. For example, in one embodiment, the original 5 delivery paths are increased to 8 delivery paths.

[0175] The multi-scenario material scheduling method provided in this embodiment should be able to quickly respond to changes in real-time data and adjust the distribution plan in a timely manner. At the same time, through the material distribution time window, the material scheduling strategy can pursue the minimization of distribution costs and the maximization of efficiency while considering customer satisfaction. Furthermore, by changing the vehicle load distribution method and distribution path, it can be flexibly adjusted according to different scenarios and needs to adapt to the complex and changing distribution environment.

[0176] A specific embodiment is given below to introduce in detail the process of determining the real-time scheduling strategy when an emergency occurs:

[0177] (1) Acquire emergency situation data, and update the multi-source data based on the emergency situation data.

[0178] Specifically, emergency data may include a variety of situations. For example, when a user actively initiates an order modification, the modified order is determined as emergency data. For another example, when a traffic accident occurs, the accident situation is determined as emergency data.

[0179] Furthermore, the emergency data is updated into the multi-source data. In specific implementation, the emergency data can be added into the multi-source data, or the emergency data can be replaced with the data in the multi-source data.

[0180] (2) Reacquiring the real-time scheduling strategy based on the updated multi-source data.

[0181] Specifically, the new multi-source data is re-input into the neural network model, and the acquisition of the real-time scheduling strategy is repeated to obtain a new real-time scheduling strategy.

[0182] The multi-scenario material scheduling method provided in this embodiment can quickly identify emergencies through real-time monitoring and prediction, and immediately start the update process, ensuring that when faced with emergencies, the scheduling strategy can be adjusted in time to reduce losses caused by delays. Furthermore, by timely updating the real-time scheduling strategy, the risks of delivery delays, material shortages or surpluses caused by emergencies can be reduced. At the same time, reasonable adjustments can also help reduce scheduling costs, inventory costs, etc., and improve overall economic benefits. In this way, in emergencies, the real-time scheduling strategy can be quickly updated to ensure the reliability and efficiency of material scheduling, better meet customer needs, and improve customer satisfaction.

[0183] Furthermore, real-time scheduling strategies can be sent down to the execution layer, while the execution status is monitored in real time and dynamically adjusted based on feedback.

[0184] It should be noted that the multi-scenario material scheduling process is integrated with multiple external systems to achieve data sharing and process collaboration. At the same time, in order to ensure data exchange and process collaboration between systems, unified interface standards and protocols need to be designed.

[0185] In specific implementation, external systems may include: Enterprise Resource Planning System (ERP): obtains warehouse inventory, order information and other data; Manufacturing Execution System (MES): obtains work orders, work order changes, material requirements and other data; Warehouse Management System (WMS): obtains basic information such as warehouses, storage locations, materials, handling equipment, inbound and outbound tasks and other data; Warehouse Control System (WCS): data such as handling equipment status, parameters, operating conditions and control instructions, equipment failure conditions, storage locations, pallets, cargo box barcodes, equipment tasks, etc.; Robot Control System (RCS): robot motion data, sensor data, program data, status data, etc.; Internet of Things Platform: handling equipment status, spatial location, path traffic conditions, etc.; Unit Management and Control System: status of docking locations related to logistics distribution in the workshop, action instructions and execution feedback data, etc.

[0186] The multi-scenario material scheduling method provided in this embodiment first uses a partial layer of a neural network model to extract feature data from historical material scheduling data, generates a scheduling auxiliary knowledge base and training samples, and then obtains multi-source data of the workspace to be scheduled, the multi-source data includes user demand order data and real-time production data of the workspace to be scheduled, and then inputs the multi-source data into the neural network model, predicts the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled, and then generates an initial scheduling strategy based on the traffic change trend and the business rules in the scheduling auxiliary knowledge base, and finally optimizes the initial scheduling strategy based on the real-time working status of the workspace to be scheduled with vehicle load balancing and material distribution time windows as constraints and historical templates as guidance, and obtains a real-time scheduling strategy. In this way, through the multi-source data collection of the workspace to be scheduled, combined with the prediction of the neural network model, it is possible to more accurately grasp the user demand order data, vehicle load status and traffic conditions, so as to formulate a more reasonable distribution plan, improve production efficiency and reduce distribution costs. Furthermore, through the multi-source data of the workspace to be scheduled, it is possible to flexibly respond to complex and changeable production environments such as various material handling vehicles, different traffic conditions and various optimization objectives, meet the material distribution needs in multiple scenarios, and ensure the reliability and effect of material scheduling. Furthermore, through the accurate prediction of traffic change trends by the neural network model, the initial scheduling plan that best meets the current actual situation can be generated in combination with the business rules in the knowledge base, ensuring the efficiency and reliability of material scheduling. Furthermore, by constraining the initial scheduling strategy through vehicle load balancing and material distribution time windows, the safety of material scheduling can be ensured, accidents can be prevented, and the user experience can be improved. Furthermore, through continuous learning and optimization, useful features can be automatically extracted, new knowledge rules can be generated, and applied to scheduling decisions, realizing intelligent decision support and ensuring the reliability and accuracy of material scheduling. Furthermore, through real-time monitoring and prediction, it is possible to respond to various emergencies and uncertain factors, re-plan distribution routes and task allocation, and ensure the smooth completion of distribution tasks. In this way, through the combination of data such as neural network models and scheduling auxiliary knowledge bases, the material scheduling process in the digital workshop is efficient, flexible and intelligent.

[0187] Optionally, a series of experiments are designed to test the effectiveness and performance of this application. The experimental contents include:

[0188] Neural network model prediction accuracy test: Use historical data to train the neural network model and test its prediction accuracy in different scenarios.

[0189] Knowledge base query efficiency test: simulate different query scenarios to test the response time and accuracy of knowledge base queries.

[0190] Evaluation of the effectiveness of dynamic scheduling strategies: In a simulated or actual environment, compare the performance of traditional scheduling methods and the dynamic scheduling strategies of this system in terms of distribution cost, efficiency, customer satisfaction, etc.

[0191] System performance testing: Test the stability and performance of the system under high concurrency, large data volume, etc.

[0192] The experimental results of the present invention are analyzed as follows:

[0193] The neural network model has high prediction accuracy: The neural network model shows high accuracy in traffic condition prediction, delivery time prediction, etc., providing strong support for the formulation of dynamic scheduling strategies.

[0194] High efficiency of knowledge base query: The knowledge base adopts structured storage and indexing technology, with fast query response speed, which can meet the needs of real-time scheduling.

[0195] Corresponding to the aforementioned embodiment of a multi-scenario material scheduling method, the present application also provides an embodiment of a multi-scenario material scheduling device.

[0196] The embodiment of a multi-scenario material scheduling device of the present application can be applied to a multi-scenario material scheduling device. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of the multi-scenario material scheduling device in which it is located reading the corresponding computer program instructions in the non-volatile memory into the memory and running them. From the hardware level, if Figure 5 As shown, it is a structural diagram of a multi-scenario material scheduling device provided by an exemplary embodiment of the present application, except Figure 5 In addition to the processor, memory, network interface, and non-volatile memory shown, the multi-scenario material scheduling device in which the device in the embodiment is located may also include other hardware generally based on the actual functions of the multi-scenario material scheduling device, which will not be elaborated herein.

[0197] Figure 6 This is a structural diagram of a multi-scenario material scheduling device provided by an exemplary embodiment of the present application. Figure 6 The device provided in this embodiment includes an acquisition module 610, a prediction module 620, a processing module 630 and an optimization module 640; wherein,

[0198] The acquisition module 610 is used to extract feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples, wherein the scheduling auxiliary knowledge base includes at least business rules, historical templates, and a network map of the workspace to be scheduled;

[0199] The acquisition module 610 is further used to acquire multi-source data of the workspace to be scheduled, wherein the multi-source data includes user demand order data and real-time production data of the workspace to be scheduled, wherein the production data includes at least order production data, production machine data and production resource data;

[0200] The prediction module 620 is used to input the multi-source data into the neural network model, and predict the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled;

[0201] The processing module 630 is used to generate an initial scheduling strategy according to the traffic change trend and the business rules in the scheduling auxiliary knowledge base, and the initial scheduling strategy includes a scheduling scheme and a scheduling priority;

[0202] The optimization module 640 is used to optimize the initial scheduling strategy based on the real-time working status of the workspace to be scheduled, with vehicle load balancing and material distribution time window as constraints and the historical template as a guide, to obtain a real-time scheduling strategy. The historical template includes a positive template and a negative template.

[0203] The device of this embodiment can be used to perform Figure 1 The steps, specific implementation principles and implementation processes of the method embodiment shown are similar and will not be repeated here.

[0204] Optionally, the acquisition module 610 is specifically used to extract feature data from historical material scheduling data based on the input layer, the historical material scheduling data at least includes historical distribution data, real-time traffic information and the user evaluation, and the feature data is valid information in the historical material scheduling data;

[0205] The acquisition module 610 is further specifically used to generate training samples with the same dimension and with the scheduling object as the carrier based on the characteristic data, and the input samples of the training samples are the differences between the material demand and the material supply at different times;

[0206] The acquisition module 610 is further specifically used to mark the entity objects in the valid information and analyze the association relationship between the entity objects based on the historical material scheduling data;

[0207] The acquisition module 610 is further specifically configured to generate a target triple based on the entity object and the association relationship, use the user evaluation as the evaluation value of the target triple, and construct the scheduling auxiliary knowledge base based on the target triple.

[0208] Optionally, the processing module 630 is specifically configured to retrieve the historical template based on the traffic change trend to obtain a target template matching the traffic change trend;

[0209] The processing module 630 is further specifically configured to generate the initial scheduling strategy based on the business rule constraints on the target template; wherein the initial scheduling strategy is not generated in real time.

[0210] Optionally, the optimization module 640 is specifically used to collect real-time traffic information and vehicle status in the workspace to be scheduled;

[0211] The optimization module 640 is further specifically used to determine the material delivery time window according to user demand and the real-time traffic information;

[0212] The optimization module 640 is further specifically used to locate the scheduling strategy to be adjusted based on the material delivery time window, and the material delivery time of the scheduling strategy to be adjusted exceeds the material delivery time window;

[0213] The optimization module 640 is further specifically used to determine the vehicle load distribution mode of the vehicle load balancing according to the vehicle status collected in real time;

[0214] The optimization module 640 is further specifically configured to add a delivery vehicle to the scheduling strategy to be adjusted based on the vehicle load distribution method;

[0215] The optimization module 640 is also specifically used to plan the delivery routes of the added vehicles and generate the real-time scheduling strategy.

[0216] Optionally, the processing module 630 is further specifically configured to determine a network architecture of a benchmark network based on a data type of a training sample of the multi-scenario material scheduling method;

[0217] The processing module 630 is further specifically used to determine the structural parameters in the network architecture based on the prediction object, and the structural parameters at least include the number of network layers and the number of neurons;

[0218] The processing module 630 is further specifically configured to extract effective knowledge from the training sample and annotate the benchmark model based on the effective knowledge;

[0219] The processing module 630 is further specifically configured to construct a neural network model based on the training samples.

[0220] Optionally, the processing module 630 is further specifically used to capture update identification data in the multi-source data, and the update identification data at least includes traffic policy adjustment data and distribution technology update data;

[0221] The processing module 630 is further specifically used to clean and verify the update identification data to obtain update data, and update the scheduling auxiliary knowledge base based on the update data, wherein the update processing at least includes data addition and data replacement;

[0222] The processing module 630 is further specifically configured to perform periodic error correction and supplementation processing on the scheduling auxiliary knowledge base based on the expert opinions.

[0223] Optionally, the acquisition module 610 is further specifically configured to acquire emergency situation data, and update the multi-source data based on the emergency situation data;

[0224] The acquisition module 610 is further specifically configured to reacquire the real-time scheduling strategy based on the updated multi-source data.

[0225] Optionally, the processing module 630 is further specifically configured to generate a current logical condition based on the traffic change trend and the current environmental condition;

[0226] The processing module 630 is further specifically used to input the current logical condition into the scheduling auxiliary knowledge base for matching, to obtain the business rules and the historical template, and the logical processes corresponding to the business rules and the historical template are applicable to the generation of the current logical condition.

[0227] Please continue to refer to Figure 5 The present application also provides a multi-scenario material scheduling device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any one of the methods provided in the first aspect of the present application are implemented.

[0228] The present application also provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of any one of the methods provided in the present application are implemented.

[0229] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0230] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0231] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A multi-scenario material scheduling method, characterized in that: The method comprises: Extracting feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples, wherein the scheduling auxiliary knowledge base includes at least business rules, historical templates, and a network map of the workspace to be scheduled; Acquire multi-source data of the workspace to be scheduled, wherein the multi-source data includes user demand order data and real-time production data of the workspace to be scheduled, wherein the production data includes at least order production data, production machine data and production resource data; Inputting the multi-source data into the neural network model, and predicting the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled; Generate an initial dispatch strategy according to the traffic change trend and the business rules in the dispatch auxiliary knowledge base, wherein the initial dispatch strategy includes a dispatch scheme and a dispatch priority; The business rules at least include traffic rules, network map rules and vehicle rules. The initial scheduling strategy is generated according to the traffic change trend and the business rules in the scheduling auxiliary knowledge base, including: Retrieving the historical template based on the traffic change trend to obtain a target template matching the traffic change trend; Based on the business rules constraining the target template, generating the initial scheduling strategy; wherein the initial scheduling strategy is not generated in real time; Taking vehicle load balancing and material distribution time window as constraints and the historical template as a guide, the initial scheduling strategy is optimized according to the real-time working state of the workspace to be scheduled to obtain a real-time scheduling strategy, and the historical template includes a positive template and a negative template; The method of optimizing the initial scheduling strategy based on the real-time working state of the workspace to be scheduled and taking the vehicle load balancing and material distribution time window as constraints and the historical template as a guide to obtain the real-time scheduling strategy includes: Collecting real-time traffic information and vehicle status in the workspace to be scheduled; Determine the material delivery time window according to user demand and the real-time traffic information; Locating the scheduling strategy to be adjusted based on the material delivery time window, wherein the material delivery time of the scheduling strategy to be adjusted exceeds the material delivery time window; Determining a vehicle load distribution method for vehicle load balancing according to the vehicle status collected in real time; Adding a delivery vehicle to the scheduling strategy to be adjusted based on the vehicle load distribution method; Plan the delivery routes of the additional vehicles and generate the real-time dispatch strategy.

2. The method according to claim 1, characterized in that The neural network model includes an input layer, a hidden layer and an output layer. The method of extracting feature data from historical material scheduling data using a partial layer of the neural network model to generate a scheduling auxiliary knowledge base and training samples includes: Extracting feature data from historical material dispatching data based on the input layer, the historical material dispatching data at least including historical distribution data, real-time traffic information and user evaluation, the feature data being valid information in the historical material dispatching data; Generate training samples with the same dimension and with the scheduling object as the carrier based on the characteristic data, wherein the input samples of the training samples are the difference between the material demand and the material supply at different times; Marking the entity objects in the valid information, and analyzing the association relationship between the entity objects based on the historical material scheduling data; A target triplet is generated based on the entity object and the association relationship, the user evaluation is used as the evaluation value of the target triplet, and the scheduling auxiliary knowledge base is constructed based on the target triplet.

3. The method according to claim 1, characterized in that The training process of the neural network model includes: Determining a network architecture of a benchmark network based on the data types of the training samples of the multi-scenario material scheduling method; Determine structural parameters in the network architecture based on the prediction object, the structural parameters at least including the number of network layers and the number of neurons; Extracting effective knowledge from the training samples, and labeling a benchmark model based on the effective knowledge; A neural network model is constructed based on the training samples.

4. The method according to claim 1, characterized in that: The generating of the scheduling auxiliary knowledge base and training samples includes: Capturing update identification data in the multi-source data, the update identification data at least including traffic policy adjustment data and distribution technology update data; Cleaning and verifying the update identification data to obtain update data, and updating the scheduling auxiliary knowledge base based on the update data, wherein the update processing at least includes data addition and data replacement; The scheduling auxiliary knowledge base is periodically corrected and supplemented based on expert opinions.

5. The method according to claim 1, characterized in that The method further comprises: optimizing the initial scheduling strategy based on the real-time working state of the workspace to be scheduled and taking the vehicle load balancing and the material distribution time window as constraints and the historical template as a guide, and obtaining the real-time scheduling strategy according to the real-time working state of the workspace to be scheduled; Acquire emergency situation data, and update the multi-source data based on the emergency situation data; The real-time scheduling strategy is reacquired based on the updated multi-source data.

6. The method according to claim 1, characterized in that The data types of the business rules and the historical templates contained in the dispatch auxiliary knowledge base are logical process data, and the initial dispatch strategy is generated according to the traffic change trend and the business rules in the dispatch auxiliary knowledge base, including: generating a current logic condition based on the traffic change trend and the current environmental conditions; The current logical condition is input into the scheduling auxiliary knowledge base for matching to obtain the business rule and the historical template. The logical process corresponding to the business rule and the historical template is applicable to the generation of the current logical condition.

7. The method according to claim 1, characterized in that The constraint prediction of the traffic change trend of the workspace to be scheduled based on the neural network model and the network map of the workspace to be scheduled includes: Determine the path constraints of the occupied paths and their occupied time periods according to the network map of the workspace to be scheduled; Determine a real-time spatial network map at each time point according to the path constraint, wherein the real-time spatial network map does not include the path corresponding to the path constraint; The neural network model predicts the optimal path within the scope of the real-time spatial network map based on the usage of the dispatched vehicles and the resources to be dispatched; The step of optimizing the initial scheduling strategy according to the real-time working state of the workspace to be scheduled to obtain a real-time scheduling strategy includes: Collecting the real-time working status in the workspace to be scheduled in real time, wherein the real-time working status represents the resource status involved in the scheduling work in the workspace to be scheduled; Determine the unused vehicles according to the real-time working status, and determine the material delivery time window based on the material delivery order demand; Taking the uncalled vehicle load balance and the minimum material delivery time as the goal and the material delivery time window as the constraint, the initial scheduling strategy is optimized to obtain a real-time scheduling strategy; wherein the optimization is optimized toward the positive template and away from the negative template.

8. A multi-scenario material scheduling device, characterized in that: The device includes an acquisition module, a prediction module, a processing module and an optimization module; wherein, The acquisition module is used to extract feature data from historical material scheduling data using a partial layer of a neural network model to generate a scheduling auxiliary knowledge base and training samples, wherein the scheduling auxiliary knowledge base includes at least business rules, historical templates, and a network map of the workspace to be scheduled; The acquisition module is further used to acquire multi-source data of the workspace to be scheduled, wherein the multi-source data includes user demand order data and real-time production data of the workspace to be scheduled, wherein the production data includes at least order production data, production machine data and production resource data; The prediction module is used to input the multi-source data into the neural network model, and predict the traffic change trend of the workspace to be scheduled based on the constraints of the neural network model and the network map of the workspace to be scheduled; The processing module is used to generate an initial scheduling strategy according to the traffic change trend and the business rules in the scheduling auxiliary knowledge base, and the initial scheduling strategy includes a scheduling scheme and a scheduling priority; The business rules at least include traffic rules, network map rules and vehicle rules. The initial scheduling strategy is generated according to the traffic change trend and the business rules in the scheduling auxiliary knowledge base, including: Retrieving the historical template based on the traffic change trend to obtain a target template matching the traffic change trend; Based on the business rules constraining the target template, generating the initial scheduling strategy; wherein the initial scheduling strategy is not generated in real time; The optimization module is used to optimize the initial scheduling strategy according to the real-time working state of the workspace to be scheduled, with vehicle load balancing and material distribution time window as constraints and the historical template as a guide, to obtain a real-time scheduling strategy, wherein the historical template includes a positive template and a negative template; The method of optimizing the initial scheduling strategy based on the real-time working state of the workspace to be scheduled and taking the vehicle load balancing and material distribution time window as constraints and the historical template as a guide to obtain the real-time scheduling strategy includes: Collecting real-time traffic information and vehicle status in the workspace to be scheduled; Determine the material delivery time window according to user demand and the real-time traffic information; Locating the scheduling strategy to be adjusted based on the material delivery time window, wherein the material delivery time of the scheduling strategy to be adjusted exceeds the material delivery time window; Determining a vehicle load distribution method for vehicle load balancing according to the vehicle status collected in real time; Adding a delivery vehicle to the scheduling strategy to be adjusted based on the vehicle load distribution method; Plan the delivery routes of the additional vehicles and generate the real-time dispatch strategy.

Images