Automobile manufacturing plan management method and system based on knowledge graph

Abstract

This invention relates to the field of manufacturing planning technology, specifically to a knowledge graph-based method and system for automobile manufacturing planning management. The method includes: collecting planning and scheduling data, resource allocation data, and process execution data from each historical production cycle of automobile manufacturing; identifying the relationships between any two different manufacturing stages involved in the manufacturing process based on the data; wherein the first and second manufacturing stages in a manufacturing stage pair are different manufacturing stages involved in different planning phases within the same production cycle; and establishing a unidirectional dependency edge between the first and second manufacturing stages, pointing from the first manufacturing stage to the second manufacturing stage. This invention, by collecting planning and scheduling, resource allocation, and process execution data from historical production cycles and identifying the relationships between any two different manufacturing stages, can comprehensively reconstruct the true dependency structure of the manufacturing process.

Description

Technical Field

[0001] This invention relates to the field of manufacturing planning technology, specifically to a knowledge graph-based method and system for automobile manufacturing planning management. Background Technology

[0002] Currently, traditional methods rely heavily on manual experience or simple tabular records to determine the sequence of manufacturing processes. Due to the numerous and complex processes involved in automobile manufacturing, manual analysis can easily overlook some implicit dependencies across processes or mistakenly treat random sequences as fixed dependencies. Furthermore, traditional methods rarely use the intervals and co-occurrence frequencies in historical cycles to quantitatively calculate dependency strength, often only distinguishing between the presence and absence of dependencies without differentiating their tightness. This makes it impossible to identify which processes should be arranged as consecutively as possible and which processes can tolerate longer waiting intervals during planning and scheduling, thus reducing the time utilization efficiency of the plan. Furthermore, traditional methods either directly use all observed connections or simply eliminate low-frequency connections based on subjective thresholds; the lack of statistical thresholds and significance detection based on edge frequency distribution results in a large number of noisy connections mixed into the graph; at the same time, traditional methods do not perform statistical significance verification on the retained dependency edges, leading to the reference of many unreliable transition relationships during planning and simulation, increasing interference information in management decisions; and for new models or new processes, traditional methods often require rebuilding the entire graph or simply overlaying the newly added relationships onto the old graph; the lack of a smooth update mechanism based on incremental change matrices and historical weight information causes the knowledge graph to either fail to reflect new changes in a timely manner or to fluctuate drastically due to the high proportion of a single addition, thus damaging the stability of existing dependencies. Summary of the Invention

[0003] To achieve the above objectives, the present invention provides the following technical solution: a knowledge graph-based automobile manufacturing plan management method, comprising: Collect planning and scheduling data, resource allocation data, and process execution data from each historical production cycle of automobile manufacturing, and identify the correlation between any two different manufacturing stages involved in the manufacturing process based on the data; For any manufacturing link pair, the first manufacturing link and the second manufacturing link in the manufacturing link pair are different manufacturing links involved in different planning stages within the same production cycle. The planning execution of the first manufacturing link precedes that of the second manufacturing link. A one-way dependency edge from the first manufacturing link to the second manufacturing link is established between the first manufacturing link and the second manufacturing link. The weight of the one-way dependency edge is determined based on the stage interval and frequency of occurrence of the manufacturing link pair in the historical production cycle. An automotive manufacturing knowledge graph is constructed based on the established relationships between various manufacturing stages, unidirectional dependency edges, and their weights. A stage transition matrix is ​​then constructed based on the automotive manufacturing knowledge graph. The elements in the stage transition matrix are used to represent the plan execution transition probability between two manufacturing stages in the knowledge graph. Obtain the production plan request from the target manufacturing workshop, and determine the target vehicle model, the corresponding process parameter sequence, and the corresponding supply chain data based on the production plan request; The target vehicle model, process parameter sequence, and supply chain data are input into the knowledge graph matching model to obtain the candidate plan set corresponding to the target manufacturing workshop. The process parameter sequence, supply chain data, and candidate plan set are then input into the planning optimization model to obtain the target production plan.

[0004] Preferably, the weights of unidirectional dependent edges are determined based on the stage intervals and frequency of occurrence of manufacturing links in the historical production cycle, including: For manufacturing link pairs with unidirectional dependency edges, determine the frequency of occurrence of the first and second manufacturing links in all planning stages included in the historical production cycle. Based on the number of times manufacturing links appear in all production cycles of historical production cycles, determine the stage interval between the planning stages where the first and second manufacturing links in the manufacturing link pair are located, and determine the frequency of occurrence of manufacturing link pairs under each stage interval. For each stage interval, based on the occurrence frequency of the first manufacturing stage and the second manufacturing stage in the manufacturing stage pair, and the occurrence frequency of the manufacturing stage pair under the stage interval, the dependency strength component between the first manufacturing stage and the second manufacturing stage in the manufacturing stage pair is determined; among them, the stage interval is negatively correlated with the dependency strength component between the first manufacturing stage and the second manufacturing stage associated with the unidirectional dependency edge. Based on the dependency strength components of each manufacturing stage at all corresponding stage intervals, the dependency strength between the first and second manufacturing stages associated with a unidirectional dependency edge is determined and used as the weight of the unidirectional dependency edge.

[0005] Preferably, an automotive manufacturing knowledge graph is constructed based on the established relationships between various manufacturing stages, unidirectional dependency edges, and their weights, including: Retain relationships from which the frequency of edge occurrence exceeds a frequency threshold, and retain unidirectional dependent edges from which the frequency of edge occurrence exceeds a frequency threshold; wherein, the edge occurrence frequency represents the frequency of occurrence of the established edge connection relationship in all planning stages included in the historical production cycle; the frequency threshold is determined based on the mean and standard deviation of the edge occurrence frequency distribution of each established relationship and each unidirectional dependent edge. For each established association and each unidirectional dependency edge, association significance is detected, and associations and unidirectional dependencies with significance greater than or equal to the set significance threshold are retained. Based on the retained relationships and the connection relationships and weights of unidirectional dependent edges, an automotive manufacturing knowledge graph is constructed.

[0006] Preferably, the automotive manufacturing knowledge graph includes all manufacturing processes in a pre-defined manufacturing system as nodes; correspondingly, the method includes: Add new vehicle models or processes to the manufacturing plan, and establish relationships and unidirectional dependency edges based on the manufacturing links involved in each planning stage of the new vehicle models or processes, so as to update the connection relationships between the manufacturing links involved in each planning stage of the new vehicle models or processes in the knowledge graph. When the proportion of newly added models or processes exceeds a preset threshold, the relationships and weights of unidirectional dependencies between each manufacturing process are recalculated based on the manufacturing processes involved in the planning phase of the newly added models or processes, to obtain the current incremental change matrix at the current time point. The current incremental change matrix is ​​used to reflect the weight changes of the pairwise connections between each manufacturing process in the knowledge graph caused by the addition of new models or processes. Based on the current incremental change matrix and the historical weight information generated when the number of newly added models or processes in the automotive manufacturing knowledge graph exceeds the preset percentage threshold, the target weight matrix of the automotive manufacturing knowledge graph at the current time point is determined. The historical weight information includes the incremental change matrix generated at each historical time point when the number of newly added models or processes in the automotive manufacturing knowledge graph exceeds the preset percentage threshold and the target weight matrix at each historical time point. The weights of the pairwise connections between all manufacturing stages in the automotive manufacturing knowledge graph are updated based on the target weight matrix.

[0007] Preferably, a process transition matrix is ​​constructed based on an automotive manufacturing knowledge graph, including: Obtain the weighted adjacency matrix corresponding to the automotive manufacturing knowledge graph, and for each manufacturing stage in the automotive manufacturing knowledge graph, normalize the weights of all connections between the manufacturing stage and other manufacturing stages based on the temperature coefficient, with the manufacturing stage as the starting node, to obtain the normalized weighted adjacency matrix. Based on the normalized weighted adjacency matrix and the difference in process complexity between two manufacturing stages in the knowledge graph, a stage transition matrix is ​​constructed.

[0008] Preferably, the knowledge graph matching model is a model obtained by training the initial matching model using the first historical data set. The knowledge graph matching model is used to select multiple plan data from the historical plan library where multiple targets match the first preset association condition set. Accordingly, the initial matching model is trained using the first historical dataset, including: Obtain multiple first process parameter sequences for multiple first vehicle models within a first preset time period, as well as a first assembly path and a first material consumption sequence corresponding to each first process parameter sequence. The path information included in the first assembly path is used to indicate the process flow of vehicle assembly, and the consumption information included in the first material consumption sequence is used to indicate the information on the consumption of parts corresponding to the first process parameter sequence. The initial multimodal correlation model is fine-tuned based on multiple first process parameter sequences, multiple first assembly paths, and multiple first material consumption sequences to obtain the target multimodal correlation model. Construct a first historical dataset and train the initial matching model based on the first historical dataset and the target multimodal association model.

[0009] Preferably, a first historical dataset is constructed, including: Obtain multiple second process parameter sequences within a second preset time period, and determine multiple second vehicle models based on the second process parameter sequences, wherein the first historical data set includes multiple second process parameter sequences; Determine the second assembly path and second material consumption sequence for each of the multiple second vehicle models; Based on the second assembly path, determine the assembly text feature set and the assembly timing feature set; based on the second material consumption sequence, determine the consumption text feature set and the consumption quantity feature set; and obtain the equipment information set corresponding to the target manufacturing workshop.

[0010] Preferably, the planning optimization model is a model obtained by training the initial optimization model using a second set of historical data. The planning optimization model is used to select multiple planning data from the candidate planning set whose objectives match the second preset optimization condition set, and to adjust the resource allocation of the selected planning data. Accordingly, the initial optimization model is trained using a second historical dataset, including: The initial optimization model performs deep cross-computation on multiple constraint feature subsets in the second historical data set to obtain multiple weight allocation results. These multiple weight allocation results are then processed by multiple different multilayer perceptrons to obtain constraint perception results. The deep cross-computation is used to calculate the weight allocation of different constraint feature subsets. The cold start constraint sets corresponding to multiple manufacturing workshops are obtained, and the multiple cold start constraint sets are processed by multiple different multilayer perceptrons to obtain the cold start perception results. The cold start constraints in the cold start constraint set are used to indicate new equipment that has not been put into use within a third preset time period. The environmental features corresponding to multiple manufacturing workshops are acquired, and the multiple environmental features are processed by a first multilayer perceptron to obtain the environmental perception results. The first multilayer perceptron is different from multiple different multilayer perceptrons. The constraint perception results, cold start constraint set, and environment perception results are concatenated to obtain the target optimization result. The target optimization result is then processed by the second multilayer perceptron to obtain the output result of the initial optimization model. The second multilayer perceptron is different from multiple different multilayer perceptrons and the first multilayer perceptron. The initial optimization model is iteratively trained based on the output results.

[0011] Preferably, a second historical dataset is constructed, including: Obtain multiple third process parameter sequences for multiple third vehicle models within a third preset time period, and determine a process parameter feature set based on the multiple third process parameter sequences, wherein the second historical data set includes the process parameter feature set; Obtain the production constraint features corresponding to each of the multiple third vehicle models. The second historical data set includes multiple production constraint features. Each production constraint feature includes the third assembly delay identifier sequence, the third material shortage identifier sequence, and the production capacity information of the multiple third vehicle models in multiple manufacturing workshops within the third preset time period. The efficiency statistical feature set for multiple sub-preset time periods within the third preset time period is obtained, as well as the environmental features corresponding to each of the multiple manufacturing workshops. The second historical data set includes the efficiency statistical feature set and multiple environmental features. Each efficiency statistical feature in the efficiency statistical feature set is used to indicate the utilization rate of production equipment in multiple manufacturing workshops within the third preset time period.

[0012] A knowledge graph-based automobile manufacturing planning management system, applicable to the aforementioned knowledge graph-based automobile manufacturing planning management method, including: The data acquisition unit is configured to collect planning and scheduling data, resource allocation data, and process execution data from each historical production cycle of automobile manufacturing, and to identify the relationship between any two different manufacturing stages involved in the manufacturing process based on the data. The weight determination unit is configured to, for any pair of manufacturing links, the first manufacturing link and the second manufacturing link in the pair are different manufacturing links involved in different planning stages within the same production cycle, the planning execution of the first manufacturing link precedes that of the second manufacturing link, a one-way dependency edge is established between the first manufacturing link and the second manufacturing link, and the weight of the one-way dependency edge is determined based on the stage interval and frequency of occurrence of the pair of manufacturing links in the historical production cycle. The process transition unit is configured to construct an automotive manufacturing knowledge graph based on the established relationships between each manufacturing process, unidirectional dependency edges, and their weights, and to construct a process transition matrix based on the automotive manufacturing knowledge graph; the elements in the process transition matrix are used to represent the probability of plan execution transition between two manufacturing processes in the knowledge graph; The request parsing unit is configured to obtain the production plan request of the target manufacturing workshop, and determine the target vehicle model, the process parameter sequence corresponding to the target vehicle model, and the supply chain data corresponding to the target vehicle model based on the production plan request. The planning management unit is configured to input the target vehicle model, process parameter sequence, and supply chain data into the knowledge graph matching model to obtain the candidate plan set corresponding to the target manufacturing workshop, and input the process parameter sequence, supply chain data, and candidate plan set into the planning optimization model to obtain the target production plan.

[0013] Compared with the prior art, the beneficial effects of the present invention are: (1) By collecting planning, scheduling, resource allocation and process execution data in historical production cycles and identifying the relationship between any two different manufacturing links, this invention can fully restore the real dependency structure of the manufacturing process and avoid the omission or incorrect connection of links caused by human experience. By establishing a one-way dependency edge from the first execution link to the next execution link for each manufacturing link and determining the weight based on the stage interval and the frequency of occurrence, the weight of the dependency edge not only reflects the frequency of co-occurrence of links, but also reflects the length of the time interval. The shorter the interval, the greater the weight, thereby distinguishing between tightly coupled processes and loosely related processes and improving the authenticity of the transfer probability during planning and simulation. (2) By using a dual screening method of edge occurrence frequency threshold and association significance detection, the present invention filters out low-frequency noise and statistically insignificant connections, so that the final graph retains only reliable and stable dependencies, reducing interference information in planning management; by introducing a temperature coefficient to normalize the weighted adjacency matrix, the smoothness of the transition probability can be flexibly adjusted, and the differences in process complexity levels are integrated, so that the transition matrix not only reflects historical statistical patterns, but also incorporates the engineering constraint of process difficulty, thereby guiding the plan generation to avoid overly complex cross-level transitions and improving the plan's executability. Attached Figure Description

[0014] Figure 1 This is a schematic flowchart of the overall method in one embodiment of the present invention; Figure 2 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0015] Figure descriptions: 1. Data acquisition unit; 2. Weight determination unit; 3. Link transfer unit; 4. Request parsing unit; 5. Plan management unit. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1, please refer to Figure 1 This invention provides a technical solution: a knowledge graph-based automobile manufacturing plan management method, comprising: S1. Collect planning and scheduling data, resource allocation data and process execution data from each historical production cycle of automobile manufacturing, and identify the relationship between any two different manufacturing links involved in the manufacturing process based on the data. S2. For any manufacturing link pair, the first manufacturing link and the second manufacturing link in the manufacturing link pair are different manufacturing links involved in different planning stages within the same production cycle. The planning execution of the first manufacturing link precedes that of the second manufacturing link. A one-way dependency edge from the first manufacturing link to the second manufacturing link is established between the first manufacturing link and the second manufacturing link. The weight of the one-way dependency edge is determined based on the stage interval and frequency of occurrence of the manufacturing link pair in the historical production cycle. S3. Construct an automotive manufacturing knowledge graph based on the established relationships between each manufacturing stage, unidirectional dependency edges, and their weights, and construct a stage transition matrix based on the automotive manufacturing knowledge graph; the elements in the stage transition matrix are used to represent the plan execution transition probability between two manufacturing stages in the knowledge graph; S4. Obtain the production plan request from the target manufacturing workshop, and determine the target vehicle model, the process parameter sequence corresponding to the target vehicle model, and the supply chain data corresponding to the target vehicle model based on the production plan request. S5. Input the target vehicle model, process parameter sequence, and supply chain data into the knowledge graph matching model to obtain the candidate plan set corresponding to the target manufacturing workshop output by the knowledge graph matching model. Input the process parameter sequence, supply chain data, and candidate plan set into the planning optimization model to obtain the target production plan.

[0018] It should be noted that the data collected from various production cycles in automobile manufacturing consists of three types: planning and scheduling data, resource allocation data, and process execution data. Planning and scheduling data records the sequence and timing of production tasks within each cycle; resource allocation data records how resources such as materials, equipment, and manpower are allocated to various manufacturing stages; and process execution data records the actual production parameters and operational results of each stage. Based on this historical data, the relationships between any two different manufacturing stages in the manufacturing process are identified. These relationships may manifest as temporal dependencies, resource sharing constraints, or process sequence requirements, etc. For each pair of manufacturing stages, referred to as a manufacturing stage pair, the first and second manufacturing stages are different stages involved in different planning phases within the same production cycle, and the planned execution time of the first manufacturing stage is earlier than that of the second manufacturing stage. A one-way dependency edge is established between these two stages, pointing from the first manufacturing stage to the second manufacturing stage, indicating that the latter can only proceed after the former is completed. At the same time, the weight of this one-way dependency edge is calculated based on the stage interval of this stage pair in the historical production cycle, i.e., the average difference between the actual execution times of the two stages, and the frequency of this stage pair in the entire historical cycle. The larger the weight, the tighter or more frequent the dependency relationship is. Using the relationships between all the manufacturing stages established above, along with each unidirectional dependency edge and its corresponding weight, an automotive manufacturing knowledge graph is constructed. This graph uses manufacturing stages as nodes and weighted unidirectional dependency edges as directed edges, comprehensively describing the sequence and transition strength between each stage in the manufacturing process. Furthermore, the dependencies of the graph are transformed into a stage transition matrix. Each element in this matrix represents the probability of a planned execution transition from one manufacturing stage to another in the graph. The transition probability can be understood as the likelihood of executing another stage after completing one stage in historical data. For a target manufacturing workshop, the first step is to receive its production plan request. Based on this request, the target vehicle model to be produced is determined, and the corresponding process parameter sequence for that vehicle model is obtained. This means that a series of process steps and parameter requirements must be followed to complete the production of that vehicle model, as well as the relevant supply chain data for that vehicle model, including parts supply status, logistics capabilities, and inventory levels. The target vehicle model, process parameter sequence, and supply chain data are input into a pre-trained knowledge graph matching model. Based on the input information, the model searches for the most similar or matching historical production path or planning pattern in the pre-built automotive manufacturing knowledge graph, and finally outputs a candidate plan set that matches the current target manufacturing workshop conditions. The candidate plan set contains multiple potentially feasible production plan schemes. The process parameter sequence, supply chain data, and the candidate plan set obtained in the previous step are then input into a planning optimization model to screen and optimize the candidate plans, ultimately outputting an optimal target production plan. This plan can be directly used to guide the actual production activities in the target manufacturing workshop. In a specific embodiment, the formula for calculating the weight of a one-way dependent edge is as follows: ;in, This indicates the manufacturing process, where... Prior to There are unidirectional dependent edges. , This indicates the frequency of this step in the historical production cycle. This indicates the stage interval between two phases. Indicates the interval penalty coefficient; The formula for calculating the transition probability of the plan is as follows: ; in, Indicates from the manufacturing stage Transfer (plan execution flow) to the manufacturing stage The probability of.

[0019] In an optional embodiment, the weight of a unidirectional dependency edge is determined based on the stage interval and frequency of occurrence of manufacturing process pairs in a historical production cycle, including: For manufacturing link pairs with unidirectional dependency edges, determine the frequency of occurrence of the first and second manufacturing links in all planning stages included in the historical production cycle. Based on the number of times manufacturing links appear in all production cycles of historical production cycles, determine the stage interval between the planning stages where the first and second manufacturing links in the manufacturing link pair are located, and determine the frequency of occurrence of manufacturing link pairs under each stage interval. For each stage interval, based on the occurrence frequency of the first manufacturing stage and the second manufacturing stage in the manufacturing stage pair, and the occurrence frequency of the manufacturing stage pair under the stage interval, the dependency strength component between the first manufacturing stage and the second manufacturing stage in the manufacturing stage pair is determined; among them, the stage interval is negatively correlated with the dependency strength component between the first manufacturing stage and the second manufacturing stage associated with the unidirectional dependency edge. Based on the dependency strength components of each manufacturing stage at all corresponding stage intervals, the dependency strength between the first and second manufacturing stages associated with a unidirectional dependency edge is determined and used as the weight of the unidirectional dependency edge.

[0020] It should be noted that for a pair of manufacturing links associated with a one-way dependent edge, it is necessary to first determine the basic occurrence of these two links in the historical production cycle. Specifically, the total frequency of the first and second manufacturing links occurring independently in all planning stages included in the entire historical production cycle is counted, which reflects the activity level of each link in history. Examine the records of this manufacturing process that occurs in all production cycles; whenever the second manufacturing process occurs within a certain time difference after the first manufacturing process occurs in a production cycle, record the time difference between the planning stages of these two processes, which is called the stage interval; at the same time, count the number of times this process occurs under each stage interval, for example, how many times it occurs when the interval is one day, and how many times it occurs when the interval is two days. For each specific stage interval, three values ​​are used to calculate the dependency strength component under that interval: the total frequency of occurrence of the first manufacturing stage, the total frequency of occurrence of the second manufacturing stage, and the frequency of co-occurrence of the two stages under this specific stage interval. The calculated dependency strength component has a key property: the longer the stage interval, the smaller the dependency strength component, that is, the two are negatively correlated. This means that if two stages have historically appeared one after the other with a long time interval, the direct dependence between them is relatively weak; conversely, if the interval is very short or even appears almost consecutively, the dependency strength component is large. By combining the dependency strength components of all different stage intervals corresponding to this manufacturing link, such as by summing or aggregating, the total dependency strength of the unidirectional dependency edge is obtained; this total dependency strength is the weight of the edge; the higher the weight, the stronger the overall tendency to switch to the second manufacturing link after various possible time lengths in history after the completion of the first manufacturing link, and the more important the dependency relationship is in the knowledge graph.

[0021] In an optional embodiment, an automotive manufacturing knowledge graph is constructed based on the established relationships between various manufacturing stages, unidirectional dependency edges, and their weights, including: Retain relationships from which the frequency of edge occurrence exceeds a frequency threshold, and retain unidirectional dependent edges from which the frequency of edge occurrence exceeds a frequency threshold; wherein, the edge occurrence frequency represents the frequency of occurrence of the established edge connection relationship in all planning stages included in the historical production cycle; the frequency threshold is determined based on the mean and standard deviation of the edge occurrence frequency distribution of each established relationship and each unidirectional dependent edge. For each established association and each unidirectional dependency edge, association significance is detected, and associations and unidirectional dependencies with significance greater than or equal to the set significance threshold are retained. Based on the retained relationships and the connection relationships and weights of unidirectional dependent edges, an automotive manufacturing knowledge graph is constructed.

[0022] It should be noted that all previously established relationships and unidirectional dependency edges underwent quality screening. The first filtering criterion is edge frequency. Every edge, whether it's an association or a one-way dependency, has an edge frequency, representing the total number of times the connection occurs across all planning phases in the entire historical production cycle. The distribution of edge frequencies is calculated, specifically using the mean and standard deviation of these frequencies to determine a frequency threshold. Only associations and one-way dependencies with an edge frequency higher than this threshold are retained. This filters out weak connections that occur occasionally and lack representativeness. The second screening criterion is the correlation significance test; for each edge retained after frequency screening, a statistical significance test is further performed to determine whether the dependency between the two manufacturing links is statistically significant, rather than random noise; the correlation significance index of each edge is calculated and compared with a pre-set significance threshold; only edges with a correlation significance greater than or equal to the threshold are retained. A knowledge graph for automobile manufacturing is constructed by utilizing all the relationships and unidirectional dependencies retained after dual screening through frequency filtering and saliency detection, along with the weights of each edge. In this graph, nodes represent various manufacturing processes, and directed edges represent the retained dependencies that are saliency and have sufficient frequency. Each edge is accompanied by its weight. The knowledge graph constructed in this way is more compact and reliable, eliminating low-frequency noise and statistically insignificant connections.

[0023] In an optional embodiment, the automotive manufacturing knowledge graph includes all manufacturing processes within a predefined manufacturing system as nodes; correspondingly, the method includes: Add new vehicle models or processes to the manufacturing plan, and establish relationships and unidirectional dependency edges based on the manufacturing links involved in each planning stage of the new vehicle models or processes, so as to update the connection relationships between the manufacturing links involved in each planning stage of the new vehicle models or processes in the knowledge graph. When the proportion of newly added models or processes exceeds a preset threshold, the relationships and weights of unidirectional dependencies between each manufacturing process are recalculated based on the manufacturing processes involved in the planning phase of the newly added models or processes, to obtain the current incremental change matrix at the current time point. The current incremental change matrix is ​​used to reflect the weight changes of the pairwise connections between each manufacturing process in the knowledge graph caused by the addition of new models or processes. Based on the current incremental change matrix and the historical weight information generated when the number of newly added models or processes in the automotive manufacturing knowledge graph exceeds the preset percentage threshold, the target weight matrix of the automotive manufacturing knowledge graph at the current time point is determined. The historical weight information includes the incremental change matrix generated at each historical time point when the number of newly added models or processes in the automotive manufacturing knowledge graph exceeds the preset percentage threshold and the target weight matrix at each historical time point. The weights of the pairwise connections between all manufacturing stages in the automotive manufacturing knowledge graph are updated based on the target weight matrix.

[0024] It should be noted that the automotive manufacturing knowledge graph initially included all manufacturing processes in the pre-defined manufacturing system, with these processes serving as nodes in the graph; based on this, the knowledge graph needs to be able to dynamically update as production plans change. When a company adds a new model or a new process to its manufacturing plan, it establishes the relationships and one-way dependencies between these processes based on the specific manufacturing steps involved in each stage of the plan. These newly established relationships are then used to update the existing connections between these manufacturing steps in the knowledge graph. In other words, the new content will supplement or change the connectivity between some nodes in the graph. Determine the proportion of the new model or process in the overall manufacturing plan. If this proportion exceeds a pre-set threshold, it indicates that the scale of the new content is significant enough that it cannot be simply superimposed on the existing graph and requires further updates. At this point, based on all the manufacturing stages involved in the new model or process, the relationships between these manufacturing stages and the weights of unidirectional dependencies are recalculated. The result is a current incremental change matrix. This matrix is ​​specifically used to represent how the original connection relationship between any two manufacturing stages in the knowledge graph has changed in terms of weight due to the introduction of the new model or process. The current incremental change matrix is ​​combined with historical weight information to determine the final target weight matrix at the current time point. Historical weight information includes the N most recent records of when the proportion of newly added models or processes exceeded a threshold. Each record includes two parts: the incremental change matrix generated at that historical time point and the final target weight matrix obtained at that historical time point. Using the current incremental change matrix and these N historical records, the target weight matrix for the current moment is calculated. The purpose of this is to maintain the smoothness and stability of the graph weights while absorbing the latest changes, avoiding drastic fluctuations. The value of N is usually determined based on the company's production rhythm and knowledge graph update strategy. In the automotive manufacturing scenario, a common value for N is 3 to 10 times. In the automotive manufacturing scenario, the proportion threshold may range from 15% to 30%, meaning that the weight recalculation of the dependencies between manufacturing stages in the knowledge graph is triggered only when the proportion of a newly added model or process in all active models / processes exceeds the proportion threshold. Historical weight information is not a single value but a set containing two parts: the incremental change matrix generated each time the weight recalculation is triggered and the target weight matrix corresponding to that historical time point. Based on the calculated target weight matrix, the weights of the pairwise connections between all manufacturing links in the automotive manufacturing knowledge graph are globally updated; the updated knowledge graph reflects all production dependencies, including the latest models or processes, for subsequent planning and management.

[0025] In an optional embodiment, constructing a process transition matrix based on an automotive manufacturing knowledge graph includes: Obtain the weighted adjacency matrix corresponding to the automotive manufacturing knowledge graph, and for each manufacturing stage in the automotive manufacturing knowledge graph, normalize the weights of all connections between the manufacturing stage and other manufacturing stages based on the temperature coefficient, with the manufacturing stage as the starting node, to obtain the normalized weighted adjacency matrix. Based on the normalized weighted adjacency matrix and the difference in process complexity between two manufacturing stages in the knowledge graph, a stage transition matrix is ​​constructed.

[0026] It should be noted that the first step in constructing the process transition matrix is ​​to obtain the corresponding weighted adjacency matrix from the automotive manufacturing knowledge graph. The weighted adjacency matrix is ​​a two-dimensional structure, with rows and columns corresponding to various manufacturing processes in the graph. Each element in the matrix represents the weight of the connection between the manufacturing process corresponding to the row and the manufacturing process corresponding to the column. If there is no direct directed edge between two processes, the corresponding element is zero. For each manufacturing stage in the knowledge graph, this stage is taken as the starting node, and the directed edges pointing to all other manufacturing stages and their weights are examined. These weights are then normalized so that the sum of the weights of all edges originating from this stage is a fixed value. A temperature coefficient is introduced during the normalization process. The temperature coefficient controls the smoothness of the weight distribution: when the temperature coefficient is high, the originally larger weights will be relatively reduced, and the originally smaller or zero weights will be relatively increased, making the transition probability more even; when the temperature coefficient is low, the differences between the original weights are emphasized more, and the larger weights become larger after normalization. After this temperature coefficient-based normalization process, the original weighted adjacency matrix is ​​transformed into a normalized weighted adjacency matrix, in which the sum of the non-zero elements in each row is equal, and it reflects the transition tendency after the temperature coefficient adjustment. The final process transition matrix is ​​constructed by combining the normalized weighted adjacency matrix with the difference in process complexity levels between two manufacturing stages in the knowledge graph. The difference in process complexity level refers to the degree of difference in process difficulty or technical requirements involved when transitioning from the current manufacturing stage to the target manufacturing stage. For example, the difficulty difference between transitioning from a simple process to a complex process and transitioning to a process of equal complexity is different. The transition probability is adjusted by comprehensively considering both the normalized transition probability and this difference in process complexity level. If the difference in process complexity level from stage A to stage B is large, even if the normalized transition probability is high, the actual transition may be limited; conversely, a smaller difference makes it easier to achieve. By integrating these two factors, each element in the final process transition matrix represents the actual planned execution transition probability from the current manufacturing stage to the next manufacturing stage, considering process complexity constraints. This process transition matrix can be used for subsequent production planning path derivation and optimization.

[0027] In an optional embodiment, the knowledge graph matching model is a model obtained by training an initial matching model using a first historical data set. The knowledge graph matching model is used to filter out multiple plan data from the historical plan library, all of which match the first preset set of association conditions. Accordingly, the initial matching model is trained using the first historical dataset, including: Obtain multiple first process parameter sequences for multiple first vehicle models within a first preset time period, as well as a first assembly path and a first material consumption sequence corresponding to each first process parameter sequence. The path information included in the first assembly path is used to indicate the process flow of vehicle assembly, and the consumption information included in the first material consumption sequence is used to indicate the information on the consumption of parts corresponding to the first process parameter sequence. The initial multimodal correlation model is fine-tuned based on multiple first process parameter sequences, multiple first assembly paths, and multiple first material consumption sequences to obtain the target multimodal correlation model. Construct a first historical dataset and train the initial matching model based on the first historical dataset and the target multimodal association model.

[0028] It should be noted that the knowledge graph matching model is a trained model. Its function is to filter out multiple plan data from the historical plan database. These plan data must match the first preset association condition set. The first preset association condition set is a set of predefined rules used to determine whether a certain historical plan is suitable as a reference. To train this model, the initial matching model needs to be trained using a first historical dataset; multiple first vehicle models involved within a first preset time period need to be acquired; for each first vehicle model, multiple first process parameter sequences need to be collected, and each first process parameter sequence corresponds to two important data: a first assembly path and a first material consumption sequence; the first assembly path contains path information, which indicates the process flow that the vehicle should follow during assembly, i.e., what to do first and what to do later; the first material consumption sequence contains consumption information, which indicates which parts need to be consumed and the quantity consumed when producing according to this process parameter sequence; Using the multiple first process parameter sequences, multiple first assembly paths, and multiple first material consumption sequences obtained in the previous step, an initial multimodal correlation model is fine-tuned. Fine-tuning refers to further optimizing the model parameters using specific data based on the existing model. After fine-tuning, the target multimodal correlation model is obtained. This model can understand the intrinsic relationship between the three different modal data: process parameters, assembly paths, and material consumption. A first historical dataset is constructed, containing historical samples used to train the matching model. Then, the initial matching model is trained using both the first historical dataset and the target multimodal association model obtained in the previous step. The target multimodal association model may play a role in feature extraction or auxiliary encoding, helping the matching model better understand the matching relationship between the input data and historical plans. After this series of training steps, the initial matching model becomes a working knowledge graph matching model. Specifically, the first process parameter sequence is used for training the knowledge graph matching model, corresponding to historical data within the "first preset time period"; the second process parameter sequence is used to construct the first historical dataset, corresponding to data within the "second preset time period"; and the third process parameter sequence is used to construct the second historical dataset, corresponding to data within the "third preset time period," for training the plan optimization model. The process parameter sequence refers to a set of process parameters arranged according to production processes, covering stamping, etc. The key control parameters in manufacturing processes such as welding, painting, and final assembly are consistent with the general definition of "process parameters" in the automotive manufacturing industry. These are key control indicators that directly affect product quality, production efficiency, and energy consumption, such as the temperature setting of the injection molding machine, welding current and voltage, painting drying time, and stamping die stroke. The path information included in the first assembly path is used to indicate the process flow of vehicle assembly. This definition clearly states that the assembly path is the process flow information of vehicle assembly. In automotive manufacturing, the assembly path usually describes the complete assembly sequence from the body-in-white to the complete vehicle, including the process flow sequence of each line such as the interior assembly line, chassis sub-assembly line, door sub-assembly line, dashboard sub-assembly line, body assembly line, and final assembly line. The consumption information included in the first material consumption sequence is used to indicate the consumption information of the parts corresponding to the first process parameter sequence. That is, the material consumption sequence is a time-series record of the part consumption information, which corresponds one-to-one with the process parameter sequence, recording the material (part) consumption of each process.

[0029] In an optional embodiment, constructing a first historical data set includes: Obtain multiple second process parameter sequences within a second preset time period, and determine multiple second vehicle models based on the second process parameter sequences, wherein the first historical data set includes multiple second process parameter sequences; Determine the second assembly path and second material consumption sequence for each of the multiple second vehicle models; Based on the second assembly path, determine the assembly text feature set and the assembly timing feature set; based on the second material consumption sequence, determine the consumption text feature set and the consumption quantity feature set; and obtain the equipment information set corresponding to the target manufacturing workshop.

[0030] It should be noted that multiple second process parameter sequences within a second preset time period are obtained; these sequences are process parameter data recorded during actual production; based on these second process parameter sequences, their respective corresponding second vehicle models are identified; the first historical data set will contain all these second process parameter sequences. For each identified second vehicle model, determine the corresponding second assembly path and second material consumption sequence. The second assembly path describes the sequence of assembly processes that the vehicle model needs to go through during production, and the second material consumption sequence records the consumption of various parts required when producing the vehicle model. Two sets of features are extracted from the second assembly path: the first is the assembly text feature set, which converts textual descriptions in the assembly path, such as process names and operating instructions, into usable features; the second is the assembly sequence feature set, which extracts time-series related features from the assembly path, such as the sequence of processes and the interval duration. Simultaneously, two other sets of features are extracted from the second material consumption sequence: the first is the consumption text feature set, which converts textual descriptions in material consumption, such as component names and specifications, into features; the second is the consumption quantity feature set, which extracts quantity information from material consumption, such as the usage of each component. Obtain the set of equipment information corresponding to the target manufacturing workshop; this set contains basic information about all available equipment in the target workshop, such as equipment type, capacity parameters, and operating status; this equipment information will be used together with other features for subsequent model training or matching processes.

[0031] In an optional embodiment, the planning optimization model is a model obtained by training an initial optimization model using a second set of historical data. The planning optimization model is used to select multiple planning data from the candidate planning set whose objectives match a second set of preset optimization conditions, and to adjust the resource allocation of the selected planning data. Accordingly, the initial optimization model is trained using a second historical dataset, including: The initial optimization model performs deep cross-computation on multiple constraint feature subsets in the second historical data set to obtain multiple weight allocation results. These multiple weight allocation results are then processed by multiple different multilayer perceptrons to obtain constraint perception results. The deep cross-computation is used to calculate the weight allocation of different constraint feature subsets. The cold start constraint sets corresponding to multiple manufacturing workshops are obtained, and the multiple cold start constraint sets are processed by multiple different multilayer perceptrons to obtain the cold start perception results. The cold start constraints in the cold start constraint set are used to indicate new equipment that has not been put into use within a third preset time period. The environmental features corresponding to multiple manufacturing workshops are acquired, and the multiple environmental features are processed by a first multilayer perceptron to obtain the environmental perception results. The first multilayer perceptron is different from multiple different multilayer perceptrons. The constraint perception results, cold start constraint set, and environment perception results are concatenated to obtain the target optimization result. The target optimization result is then processed by the second multilayer perceptron to obtain the output result of the initial optimization model. The second multilayer perceptron is different from multiple different multilayer perceptrons and the first multilayer perceptron. The initial optimization model is iteratively trained based on the output results.

[0032] It should be noted that the planning optimization model is obtained by training the initial optimization model with the second set of historical data. The model has two main functions: first, to select all planning data that meet the second preset optimization conditions from the candidate planning set; and second, to adjust the resource allocation of these selected plans. The initial optimization model receives a second set of historical data, which contains various subsets of constraint features. Each subset represents a type of production constraint, such as time, material, or equipment constraints. The model performs deep cross-calculation on these subsets to calculate the weight distribution among different constraint feature subsets, i.e., to determine which constraints are more important in planning. After deep cross-calculation, the model obtains multiple weight distribution results. These weight distribution results are then fed into multiple different multilayer perceptrons for processing. Each multilayer perceptron independently processes one weight distribution result, ultimately outputting the constraint perception result. The constraint perception result reflects the model's comprehensive understanding and trade-offs regarding various production constraints. Obtain the cold start constraint sets corresponding to each of the multiple manufacturing workshops; the cold start constraints contained in the cold start constraint sets are used to indicate new equipment that has not yet been put into use within a third preset time period; these new equipment have no historical operating data, so they need to be specially processed; these cold start constraint sets are also sent to multiple different multilayer perceptrons, but whether the multilayer perceptrons here are the same group or a different group used when processing the constraint feature subsets depends on the description, it is a different group of multiple different multilayer perceptrons for processing cold starts; after processing, the cold start perception results are obtained; The system acquires environmental characteristics corresponding to each of the multiple manufacturing workshops, such as workshop temperature, humidity, and energy supply status. These environmental characteristics are then fed into a first multilayer perceptron for processing. This first multilayer perceptron is different from the various multilayer perceptrons mentioned earlier; it is a dedicated network. The processed output is the environmental perception result. The three vectors obtained earlier—constraint perception result, cold start perception result, and environment perception result—are concatenated to form a longer combined vector, called the target optimization result. This target optimization result is then fed into the second multilayer perceptron for processing. The second multilayer perceptron is different from all the multilayer perceptrons used before; it is an independent output network. After processing by the second multilayer perceptron, the final output result of the initial optimization model is obtained. Based on the difference between this output and the actual optimal plan in the second historical data set, the initial optimization model is iteratively trained. By repeatedly adjusting the parameters of all multilayer perceptrons and deep cross-computation parts in the model, the output of the model gradually approaches the ideal result, and finally the trained plan optimization model is obtained. Among them, the first multilayer perceptron is specifically used to process the environmental characteristics corresponding to multiple manufacturing workshops; the second multilayer perceptron is used to process the spliced ​​target optimization results.

[0033] In an optional embodiment, constructing a second historical data set includes: Obtain multiple third process parameter sequences for multiple third vehicle models within a third preset time period, and determine a process parameter feature set based on the multiple third process parameter sequences, wherein the second historical data set includes the process parameter feature set; Obtain the production constraint features corresponding to each of the multiple third vehicle models. The second historical data set includes multiple production constraint features. Each production constraint feature includes the third assembly delay identifier sequence, the third material shortage identifier sequence, and the production capacity information of the multiple third vehicle models in multiple manufacturing workshops within the third preset time period. The efficiency statistical feature set for multiple sub-preset time periods within the third preset time period is obtained, as well as the environmental features corresponding to each of the multiple manufacturing workshops. The second historical data set includes the efficiency statistical feature set and multiple environmental features. Each efficiency statistical feature in the efficiency statistical feature set is used to indicate the utilization rate of production equipment in multiple manufacturing workshops within the third preset time period.

[0034] It should be noted that the process of constructing the second historical dataset consists of three main parts, each of which collects different types of data features, and all these features together form the second historical dataset. Obtain multiple third process parameter sequences for multiple third vehicle models within a third preset time period; each third vehicle model corresponds to multiple different process parameter sequences, which record the specific combinations of process parameters used in actual production; extract and form a process parameter feature set from these sequences, which will be incorporated into the second historical data set; The production constraint features corresponding to each third vehicle model are obtained. Each production constraint feature specifically includes the following three items: the first item is the third assembly delay identifier sequence, which records the delay in each assembly stage of the vehicle model within the third preset time period; the second item is the third material shortage identifier sequence, which records the interruption caused by insufficient materials during the production process of the vehicle model; the third item is the capacity information of multiple third vehicle models across multiple manufacturing workshops, that is, the maximum production capacity that each workshop can allocate to each vehicle model. All these production constraint features are added to the second historical data set. Two additional types of data are acquired. The first type is a set of efficiency statistics features, which is derived from multiple sub-preset time periods divided within the third preset time period. Each efficiency statistics feature is used to indicate the utilization rate of production equipment in multiple manufacturing workshops within these sub-time periods, that is, the ratio of actual equipment running time to available time. The second type is the environmental characteristics corresponding to each of the multiple manufacturing workshops, such as external conditions such as temperature, humidity, and cleanliness within the workshop. The set of efficiency statistics features and multiple environmental features are also added to the second historical data set. The second historical data set includes a set of process parameter features, multiple production constraint features, a set of efficiency statistical features, and multiple environmental features; these data provide diverse historical samples and constraint information for training the planning optimization model.

[0035] Example 2, please refer to Figure 2 This invention provides a technical solution: a knowledge graph-based automobile manufacturing planning management system, applicable to the aforementioned knowledge graph-based automobile manufacturing planning management method, comprising: Data acquisition unit 1 is configured to collect planning and scheduling data, resource allocation data and process execution data in each historical production cycle of automobile manufacturing, and to identify the relationship between any two different manufacturing links involved in the manufacturing process based on the data. Weight determination unit 2 is configured to, for any pair of manufacturing links, the first manufacturing link and the second manufacturing link in the pair are different manufacturing links involved in different planning stages within the same production cycle, the planning execution of the first manufacturing link precedes the second manufacturing link, a one-way dependency edge from the first manufacturing link to the second manufacturing link is established between the first manufacturing link and the second manufacturing link, and the weight of the one-way dependency edge is determined based on the stage interval and occurrence frequency of the pair of manufacturing links in the historical production cycle. The process transfer unit 3 is configured to construct an automotive manufacturing knowledge graph based on the established relationships between each manufacturing process, unidirectional dependency edges and their weights, and to construct a process transfer matrix based on the automotive manufacturing knowledge graph; the elements in the process transfer matrix are used to represent the plan execution transfer probability between two manufacturing processes in the knowledge graph; Request parsing unit 4 is configured to obtain the production plan request of the target manufacturing workshop, and determine the target vehicle model, the process parameter sequence corresponding to the target vehicle model, and the supply chain data corresponding to the target vehicle model based on the production plan request; The planning management unit 5 is configured to input the target vehicle model, process parameter sequence, and supply chain data into the knowledge graph matching model to obtain the candidate plan set corresponding to the target manufacturing workshop output by the knowledge graph matching model. The process parameter sequence, supply chain data, and candidate plan set are then input into the planning optimization model to obtain the target production plan.

[0036] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A method for managing a manufacturing plan for an automobile based on a knowledge graph, characterized by, include: Collect planning and scheduling data, resource allocation data, and process execution data from each historical production cycle of automobile manufacturing, and identify the correlation between any two different manufacturing stages involved in the manufacturing process based on the data; For any manufacturing link pair, the first manufacturing link and the second manufacturing link in the manufacturing link pair are different manufacturing links involved in different planning stages within the same production cycle. The planning execution of the first manufacturing link precedes that of the second manufacturing link. A one-way dependency edge from the first manufacturing link to the second manufacturing link is established between the first manufacturing link and the second manufacturing link. The weight of the one-way dependency edge is determined based on the stage interval and frequency of occurrence of the manufacturing link pair in the historical production cycle. An automotive manufacturing knowledge graph is constructed based on the established relationships between various manufacturing stages, unidirectional dependency edges, and their weights. A stage transition matrix is ​​then constructed based on the automotive manufacturing knowledge graph. The elements in the stage transition matrix are used to represent the plan execution transition probability between two manufacturing stages in the knowledge graph. Obtain the production plan request from the target manufacturing workshop, and determine the target vehicle model, the corresponding process parameter sequence, and the corresponding supply chain data based on the production plan request; The target vehicle model, process parameter sequence, and supply chain data are input into the knowledge graph matching model to obtain the candidate plan set corresponding to the target manufacturing workshop. The process parameter sequence, supply chain data, and candidate plan set are then input into the planning optimization model to obtain the target production plan.

2. The knowledge graph-based automobile manufacturing plan management method according to claim 1, characterized by, The weights of unidirectional dependent edges are determined based on the stage intervals and frequency of occurrence of manufacturing links in historical production cycles, including: For manufacturing link pairs with unidirectional dependency edges, determine the frequency of occurrence of the first and second manufacturing links in all planning stages included in the historical production cycle. Based on the number of times manufacturing links appear in all production cycles of historical production cycles, determine the stage interval between the planning stages where the first and second manufacturing links in the manufacturing link pair are located, and determine the frequency of occurrence of manufacturing link pairs under each stage interval. For each stage interval, based on the occurrence frequency of the first manufacturing stage and the second manufacturing stage in the manufacturing stage pair, and the occurrence frequency of the manufacturing stage pair under the stage interval, the dependency strength component between the first manufacturing stage and the second manufacturing stage in the manufacturing stage pair is determined; among them, the stage interval is negatively correlated with the dependency strength component between the first manufacturing stage and the second manufacturing stage associated with the unidirectional dependency edge. Based on the dependency strength components of each manufacturing stage at all corresponding stage intervals, the dependency strength between the first and second manufacturing stages associated with a unidirectional dependency edge is determined and used as the weight of the unidirectional dependency edge.

3. The automobile manufacturing planning management method based on knowledge graphs according to claim 2, characterized in that, Based on the established relationships between various manufacturing stages, unidirectional dependency edges, and their weights, an automotive manufacturing knowledge graph is constructed, including: Retain relationships from which the frequency of edge occurrence exceeds a frequency threshold, and retain unidirectional dependent edges from which the frequency of edge occurrence exceeds a frequency threshold; wherein, the edge occurrence frequency represents the frequency of occurrence of the established edge connection relationship in all planning stages included in the historical production cycle; the frequency threshold is determined based on the mean and standard deviation of the edge occurrence frequency distribution of each established relationship and each unidirectional dependent edge. For each established association and each unidirectional dependency edge, association significance is detected, and associations and unidirectional dependencies with significance greater than or equal to the set significance threshold are retained. Based on the retained relationships and the connection relationships and weights of unidirectional dependent edges, an automotive manufacturing knowledge graph is constructed.

4. The automobile manufacturing planning management method based on knowledge graphs according to claim 3, characterized in that, The automotive manufacturing knowledge graph includes all manufacturing processes within a pre-defined manufacturing system as nodes; correspondingly, the method includes: Add new vehicle models or processes to the manufacturing plan, and establish relationships and unidirectional dependency edges based on the manufacturing links involved in each planning stage of the new vehicle models or processes, so as to update the connection relationships between the manufacturing links involved in each planning stage of the new vehicle models or processes in the knowledge graph. When the proportion of newly added models or processes exceeds a preset threshold, the relationships and weights of unidirectional dependencies between each manufacturing process are recalculated based on the manufacturing processes involved in the planning phase of the newly added models or processes, to obtain the current incremental change matrix at the current time point. The current incremental change matrix is ​​used to reflect the weight changes of the pairwise connections between each manufacturing process in the knowledge graph caused by the addition of new models or processes. Based on the current incremental change matrix and the historical weight information generated when the number of newly added models or processes in the automotive manufacturing knowledge graph exceeds the preset percentage threshold, the target weight matrix of the automotive manufacturing knowledge graph at the current time point is determined. The historical weight information includes the incremental change matrix generated at each historical time point when the number of newly added models or processes in the automotive manufacturing knowledge graph exceeds the preset percentage threshold and the target weight matrix at each historical time point. The weights of the pairwise connections between all manufacturing stages in the automotive manufacturing knowledge graph are updated based on the target weight matrix.

5. The automobile manufacturing planning management method based on knowledge graphs according to claim 4, characterized in that, Constructing a process transition matrix based on an automotive manufacturing knowledge graph, including: Obtain the weighted adjacency matrix corresponding to the automotive manufacturing knowledge graph, and for each manufacturing stage in the automotive manufacturing knowledge graph, normalize the weights of all connections between the manufacturing stage and other manufacturing stages based on the temperature coefficient, with the manufacturing stage as the starting node, to obtain the normalized weighted adjacency matrix. Based on the normalized weighted adjacency matrix and the difference in process complexity between two manufacturing stages in the knowledge graph, a stage transition matrix is ​​constructed.

6. The automobile manufacturing planning management method based on knowledge graphs according to claim 5, characterized in that, The knowledge graph matching model is a model obtained by training the initial matching model using the first historical data set. The knowledge graph matching model is used to filter out plan data from the historical plan library where multiple targets match the first preset set of association conditions. Accordingly, the initial matching model is trained using the first historical dataset, including: Obtain multiple first process parameter sequences for multiple first vehicle models within a first preset time period, as well as a first assembly path and a first material consumption sequence corresponding to each first process parameter sequence. The path information included in the first assembly path is used to indicate the process flow of vehicle assembly, and the consumption information included in the first material consumption sequence is used to indicate the information on the consumption of parts corresponding to the first process parameter sequence. The initial multimodal correlation model is fine-tuned based on multiple first process parameter sequences, multiple first assembly paths, and multiple first material consumption sequences to obtain the target multimodal correlation model. Construct a first historical dataset and train the initial matching model based on the first historical dataset and the target multimodal association model.

7. The automobile manufacturing planning management method based on knowledge graphs according to claim 6, characterized in that, Construct the first historical dataset, including: Obtain multiple second process parameter sequences within a second preset time period, and determine multiple second vehicle models based on the second process parameter sequences, wherein the first historical data set includes multiple second process parameter sequences; Determine the second assembly path and second material consumption sequence for each of the multiple second vehicle models; Based on the second assembly path, determine the assembly text feature set and the assembly timing feature set; based on the second material consumption sequence, determine the consumption text feature set and the consumption quantity feature set; and obtain the equipment information set corresponding to the target manufacturing workshop.

8. The automobile manufacturing planning management method based on knowledge graphs according to claim 7, characterized in that, The planning optimization model is a model trained on the initial optimization model using the second historical data set. The planning optimization model is used to select multiple planning data from the candidate planning set whose objectives match the second preset optimization condition set, and to adjust the resource allocation of the selected planning data. Accordingly, the initial optimization model is trained using a second historical dataset, including: The initial optimization model performs deep cross-computation on multiple constraint feature subsets in the second historical data set to obtain multiple weight allocation results. These multiple weight allocation results are then processed by multiple different multilayer perceptrons to obtain constraint perception results. The deep cross-computation is used to calculate the weight allocation of different constraint feature subsets. The cold start constraint sets corresponding to multiple manufacturing workshops are obtained, and the multiple cold start constraint sets are processed by multiple different multilayer perceptrons to obtain the cold start perception results. The cold start constraints in the cold start constraint set are used to indicate new equipment that has not been put into use within a third preset time period. The environmental features corresponding to multiple manufacturing workshops are acquired, and the multiple environmental features are processed by a first multilayer perceptron to obtain the environmental perception results. The first multilayer perceptron is different from multiple different multilayer perceptrons. The constraint perception results, cold start constraint set, and environment perception results are concatenated to obtain the target optimization result. The target optimization result is then processed by the second multilayer perceptron to obtain the output result of the initial optimization model. The second multilayer perceptron is different from multiple different multilayer perceptrons and the first multilayer perceptron. The initial optimization model is iteratively trained based on the output results.

9. The automobile manufacturing planning management method based on knowledge graphs according to claim 8, characterized in that, Construct a second historical dataset, including: Obtain multiple third process parameter sequences for multiple third vehicle models within a third preset time period, and determine a process parameter feature set based on the multiple third process parameter sequences, wherein the second historical data set includes the process parameter feature set; Obtain the production constraint features corresponding to each of the multiple third vehicle models. The second historical data set includes multiple production constraint features. Each production constraint feature includes the third assembly delay identifier sequence, the third material shortage identifier sequence, and the production capacity information of the multiple third vehicle models in multiple manufacturing workshops within the third preset time period. The efficiency statistical feature set for multiple sub-preset time periods within the third preset time period is obtained, as well as the environmental features corresponding to each of the multiple manufacturing workshops. The second historical data set includes the efficiency statistical feature set and multiple environmental features. Each efficiency statistical feature in the efficiency statistical feature set is used to indicate the utilization rate of production equipment in multiple manufacturing workshops within the third preset time period.

10. A knowledge graph-based automobile manufacturing planning management system, applicable to the knowledge graph-based automobile manufacturing planning management method described in any one of claims 1-9, characterized in that, include: The data acquisition unit is configured to collect planning and scheduling data, resource allocation data, and process execution data from each historical production cycle of automobile manufacturing, and to identify the relationship between any two different manufacturing stages involved in the manufacturing process based on the data. The weight determination unit is configured to, for any pair of manufacturing links, the first manufacturing link and the second manufacturing link in the pair are different manufacturing links involved in different planning stages within the same production cycle, the planning execution of the first manufacturing link precedes that of the second manufacturing link, a one-way dependency edge is established between the first manufacturing link and the second manufacturing link, and the weight of the one-way dependency edge is determined based on the stage interval and frequency of occurrence of the pair of manufacturing links in the historical production cycle. The process transition unit is configured to construct an automotive manufacturing knowledge graph based on the established relationships between each manufacturing process, unidirectional dependency edges, and their weights, and to construct a process transition matrix based on the automotive manufacturing knowledge graph; the elements in the process transition matrix are used to represent the probability of plan execution transition between two manufacturing processes in the knowledge graph; The request parsing unit is configured to obtain the production plan request of the target manufacturing workshop, and determine the target vehicle model, the process parameter sequence corresponding to the target vehicle model, and the supply chain data corresponding to the target vehicle model based on the production plan request. The planning management unit is configured to input the target vehicle model, process parameter sequence, and supply chain data into the knowledge graph matching model to obtain the candidate plan set corresponding to the target manufacturing workshop, and input the process parameter sequence, supply chain data, and candidate plan set into the planning optimization model to obtain the target production plan.

Images