Self-adaptive digital intelligence scheduling method and system for lean production

Abstract

The invention provides a lean production-oriented self-adaptive digital intelligent scheduling method and system, and relates to the technical field of data processing, and the method comprises the steps: taking an initial resource allocation scheme as the basis of task sequence optimization, constructing a time detection line on a time axis, detecting the overlapping condition of different tasks in an execution period, and recognizing resource conflict points; generating a preliminary scheduling plan under the condition of meeting the equipment capacity and man-hour constraint; based on multi-source data of a production site, evaluating the preliminary scheduling plan, updating scheduling parameters, and generating an optimized scheduling scheme; and converting the optimized scheduling scheme into an equipment control instruction, and transmitting the equipment control instruction to a control terminal of a production unit to complete closed-loop optimization of production resources. Reasonable adaptation and efficient scheduling of production resources can be achieved, the production efficiency is improved, and resource waste and conflicts are reduced.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an adaptive digital intelligent scheduling method and system for lean manufacturing. Background Technology

[0002] In shop floor scheduling for discrete manufacturing, such as automotive parts assembly lines, existing scheduling methods typically rely on pre-set work order data and basic equipment status information in the Manufacturing Execution System (MES). These methods mostly generate production plans based on fixed scheduling rules or static mathematical models and can handle routine production scheduling to a certain extent. However, with the development of sensing technologies on the production floor, data sources are becoming increasingly abundant. In addition to work order data from MES, these sources include material tracking data from RFID, work-in-process location data from vision systems, and real-time operating parameters from equipment IoT platforms. These multi-source data differ in structure and temporal sequence, making it difficult for existing methods to deeply integrate and uniformly represent them. When constructing scheduling models, the inherent spatial distribution characteristics of these data (such as the aggregation state of materials in buffer areas and the layout correlation of equipment groups) are often ignored or processed only as simplified logical variables. This results in the scheduling system's perception of the actual situation on the production floor being insufficiently refined, and the mapping relationship between the constructed scheduling model and the physical space being biased.

[0003] Therefore, resource allocation schemes generated based on such incomplete feature representations may exhibit insufficient adaptability in dynamic production environments. For example, when materials accumulate at a certain workstation or the load on a specific equipment group is uneven, the static model may struggle to make timely and accurate adjustments, thereby affecting the further improvement of overall production efficiency. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide an adaptive digital intelligent scheduling method and system for lean production, which can realize the rational adaptation and efficient scheduling of production resources, improve production efficiency and reduce resource waste and conflicts.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, an adaptive digital scheduling method for lean manufacturing, the method comprising: Step 1: Collect multi-source data from the production site, preprocess the multi-source data, and generate a structured data set; Step 2: Construct a spatial structure plane based on the structured data set, define structural analysis primitives on the spatial structure plane, compare feature similarity, mark the connected regions with the same feature attributes, and divide them into multiple analysis regions; Step 3: Assign the structured dataset to the corresponding analysis regions according to spatial coordinates, calculate the center position, orientation angle and dispersion characteristics of the data points in each analysis region, extract the characteristic parameters of the spatial distribution of the data, and generate parameter adjustment values; Step 4: Use the parameter adjustment amount as the basis for neural network training, normalize the structured data set and use it as training samples to complete model training and obtain the initial resource allocation scheme. Step 5: Using the initial resource allocation scheme as the basis for task sequence optimization, by constructing a time probe line on the time axis, the overlap of different tasks in the execution period is detected, resource conflict points are identified, and a preliminary scheduling plan is generated under the condition of meeting equipment capacity and working time constraints. Step 6: Based on multi-source data from the production site, evaluate the preliminary scheduling plan, update the scheduling parameters, and generate an optimized scheduling scheme; Step 7: Convert the optimized scheduling scheme into equipment control commands and transmit them to the control terminal of the production unit to complete the closed-loop optimization of production resources.

[0006] Secondly, an adaptive digital scheduling system for lean manufacturing includes: The data acquisition module is used to collect multi-source data from the production site, preprocess the multi-source data, and generate a structured data set. The partitioning module is used to construct a spatial structure plane based on a structured dataset, define structural analysis primitives on the spatial structure plane, compare feature similarities, mark connected regions with the same feature attributes, and divide them into multiple analysis regions. The calculation module is used to allocate the structured data set to the corresponding analysis area according to spatial coordinates, calculate the center position, orientation angle and dispersion characteristics of the data points in each analysis area, extract the characteristic parameters of the spatial distribution of the data, and generate parameter adjustment amounts; The processing module is used to use the parameter adjustment amount as the basis for neural network training, normalize the structured data set and use it as training samples to complete model training and obtain the initial resource allocation scheme. The detection module is used as the basis for task sequence optimization by using the initial resource allocation scheme as the basis. By constructing a time probe line on the time axis, it detects the overlap of different tasks in the execution period, identifies resource conflict points, and generates a preliminary scheduling plan under the condition of meeting equipment capacity and working time constraints. The evaluation module is used to evaluate the preliminary scheduling plan based on multi-source data from the production site, update scheduling parameters, and generate an optimized scheduling scheme. The transmission module is used to convert the optimized scheduling scheme into equipment control commands and transmit them to the control terminal of the production unit to complete the closed-loop optimization of production resources.

[0007] Thirdly, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0008] The above-described solution of the present invention has at least the following beneficial effects: By constructing a time probe line to detect overlaps in task execution periods, this invention can identify potential resource conflict points in advance during the planning stage and optimize the task sequence under various constraints. This proactive conflict resolution method helps reduce waiting, interruptions, and resource contention in the production process, ensures the smooth execution of the scheduling plan, and improves the executability and production efficiency of the plan. It also enhances the adaptive capability of the scheduling system in the face of dynamic changes, effectively reduces local resource overload or idleness, and improves resource utilization efficiency. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the adaptive digital intelligent scheduling method for lean manufacturing provided by an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of an adaptive digital intelligent scheduling system for lean manufacturing provided by an embodiment of the present invention. Detailed Implementation

[0011] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0012] like Figure 1 As shown, embodiments of the present invention propose an adaptive digital scheduling method for lean manufacturing, the method comprising the following steps: Step 1: Collect multi-source data from the production site, preprocess the multi-source data, and generate a structured data set; Step 2: Construct a spatial structure plane based on the structured data set, define structural analysis primitives on the spatial structure plane, compare feature similarity, mark the connected regions with the same feature attributes, and divide them into multiple analysis regions; Step 3: Assign the structured dataset to the corresponding analysis regions according to spatial coordinates, calculate the center position, orientation angle and dispersion characteristics of the data points in each analysis region, extract the characteristic parameters of the spatial distribution of the data, and generate parameter adjustment values; Step 4: Use the parameter adjustment amount as the basis for neural network training, normalize the structured data set and use it as training samples to complete model training and obtain the initial resource allocation scheme. Step 5: Using the initial resource allocation scheme as the basis for task sequence optimization, by constructing a time probe line on the time axis, the overlap of different tasks in the execution period is detected, resource conflict points are identified, and a preliminary scheduling plan is generated under the condition of meeting equipment capacity and working time constraints. Step 6: Based on multi-source data from the production site, evaluate the preliminary scheduling plan, update the scheduling parameters, and generate an optimized scheduling scheme; Step 7: Convert the optimized scheduling scheme into equipment control commands and transmit them to the control terminal of the production unit to complete the closed-loop optimization of production resources.

[0013] In this embodiment of the invention, by constructing a time probe line to detect the overlap of task execution periods, the invention can identify potential resource conflict points in advance during the planning stage and optimize the task sequence under various constraints. This proactive conflict resolution method helps reduce waiting, interruptions, and resource contention in the production process, ensures the smooth execution of the scheduling plan, and improves the executability and production efficiency of the plan. It also enhances the adaptive capability of the scheduling system in the face of dynamic changes, effectively reduces local resource overload or idleness, and improves resource utilization efficiency.

[0014] In a preferred embodiment of the present invention, step 1 involves collecting multi-source data from the production site, preprocessing the multi-source data, and generating a structured data set. Specifically, this includes: in discrete manufacturing workshops such as automotive parts assembly lines, deploying sensors at key equipment locations, material transport lines, and buffer areas; installing RFID devices on material pallets and work-in-process tooling; arranging vision acquisition devices above workstations and on both sides of conveyor belts; and establishing connections between IoT terminals and processing equipment, testing equipment, and logistics equipment. Multi-source data is collected through these devices. The collected data types include manufacturing execution-related work order data, material tracking data from RFID devices, work-in-process location data from the vision system, and real-time operating parameters from the equipment's IoT platform. These real-time operating parameters include a rotational speed of 500 to 3000 rpm, a temperature of 50-100 degrees Celsius, a pressure of 0.3 to 0.8 MPa, and a voltage of 220 to 380 volts. Preprocessing the collected multi-source data involves first removing outliers. The normal range of equipment operating parameters is determined by a 10% fluctuation above and below the factory-set rated parameters, such as a rated rotational speed of 2... For equipment with a rated speed of 000 rpm, the normal range is 1800 to 2200 rpm. For equipment with a rated temperature of 80 degrees Celsius, the normal range is 72 to 88 degrees Celsius. The normal range for material position coordinates is based on the actual length and width boundaries measured in the workshop. For example, if the workshop is 100 meters long and 50 meters wide, the x-axis coordinate is 0 to 100 meters and the y-axis coordinate is 0 to 50 meters. Data exceeding these ranges are considered outliers and are directly discarded. Next, missing values ​​are filled. For missing values ​​of equipment operating parameters, parameter values ​​within 5 minutes of adjacent data collection periods are selected. The average value is obtained by dividing the sum by the number of collections. This average value is used to fill in missing values. For missing values ​​in material tracking data, historical tracking data of the same batch and type of materials in the past 3 months are retrieved, and reasonable values ​​for the missing period are calculated based on the data change trend to fill in the missing values. Finally, the multi-source data with different structure formats are uniformly converted into a standardized data format that includes data identifiers, collection timestamps, physical location x-axis coordinates, y-axis coordinates, material numbers, equipment numbers, operating parameter values, and work-in-process quantities. All the converted standardized data are integrated to generate a structured data set.

[0015] This embodiment achieves accurate and unified representation of multi-source data in the spatial dimension, fully explores the inherent spatial distribution characteristics of production site data, and ensures that the division of analysis areas is highly consistent with the physical spatial layout of the workshop.

[0016] In a preferred embodiment of the present invention, step 2, constructing a spatial structure plane based on a structured data set, defining structural analysis primitives on the spatial structure plane, performing feature similarity comparisons, marking connection regions with the same feature attributes, and dividing them into multiple analysis regions, may include: Step 201: Project each data point in the structured dataset onto a two-dimensional planar coordinate grid to form a spatial structure plane. Based on the spatial structure plane, establish an equally spaced grid, defining each grid cell as an analysis cell. Specifically, using the physical spatial layout of the discrete manufacturing workshop as a reference, set the lower left corner of the workshop entrance as the origin of the two-dimensional planar coordinate grid. The x-axis is parallel to the length direction of the workshop, and the y-axis is parallel to the width direction of the workshop. Each data point is measured according to its actual physical position within the workshop, obtaining its corresponding x-axis and y-axis coordinate values ​​with centimeter-level accuracy. All data points are then processed according to their respective coordinates. The values ​​are projected one by one onto a two-dimensional plane to form a spatial structure plane that completely corresponds to the actual production area of ​​the workshop. Based on this spatial structure plane, a fixed distance interval is determined by combining the density of the workshop equipment layout and the size of the material buffer area. In areas with small equipment spacing, such as the precision machining area, the equipment spacing is 1 to 2 meters, and the interval is set to 0.5 to 1 meter. The material buffer area interval is set to 2 to 5 meters to ensure that each analysis cell can cover 1 to 2 devices or part of the buffer area and accurately contain the relevant data points in the area. The spatial structure plane is divided into equally spaced grids according to the set intervals. Each rectangular grid cell formed after the division is defined as an analysis cell.

[0017] Step 202: Calculate the distribution feature descriptor for each analysis cell, and starting from the initial analysis cell, sequentially compare the differences in distribution feature descriptors between adjacent analysis cells. Quantify the degree of difference by calculating the weighted distance of the distribution feature descriptors. Specifically, for each analysis cell, calculate its distribution feature descriptor, which includes the number of valid data points in the cell, the average value of the equipment operating parameters corresponding to the data points, the material type percentage, and the work-in-process quantity. The number of valid data points is the total number of data points in the cell after removing outliers. The average value of the equipment operating parameters is calculated by summing the values ​​of each parameter such as speed, temperature, pressure, and voltage, and then dividing each value by the number of valid data points. The material type percentage is the quantity of each material divided by the total quantity of all materials in the cell. The work-in-process quantity is the total quantity of work-in-process recorded by all data points in the cell.

[0018] The starting analysis cell is determined to be the first analysis cell near the initial material buffer area at the workshop entrance. Starting from this cell, the differences in distribution feature descriptors between each analysis cell and its four adjacent analysis cells in the four directions (up, down, left, and right) are compared one by one, from left to right and top to bottom. When calculating the weighted distance, the weights of each distribution feature are first determined according to the production scheduling priority requirements. The weights for the number of data points range from 0.25 to 0.35, the weights for the average value of equipment operating parameters range from 0.25 to 0.35, the weights for the proportion of material types range from 0.15 to 0.25, and the weights for the quantity of work-in-process range from 0.15 to 0.25. The sum of all weights is always kept at 1, and can be fine-tuned within the corresponding range according to actual scheduling needs. Then, the numerical difference of the same distribution feature item between two adjacent cells is calculated. The absolute value of the difference is used to avoid cancellation of positive and negative values. Each absolute value is multiplied by its corresponding weight. Finally, all weighted absolute values ​​are added together, and the sum is the weighted distance between the distribution feature descriptors of the two cells. This weighted distance quantifies the degree of difference between the two.

[0019] Step 203: Based on the degree of difference, a difference threshold is set. When the weighted distance of the distribution feature descriptors of adjacent analysis cells is less than the threshold, they are classified into the same feature region. Starting from the classified feature region, the search continues to look for analysis cells with similar distribution feature descriptors until no adjacent cells that meet the difference condition can be found, thus obtaining the feature region labeling result. Specifically, this includes setting a difference threshold based on the average weighted distance of adjacent cells under different production states in the workshop's historical production data, such as normal production, material shortage, and equipment maintenance, plus one standard deviation. The average value of the weighted distance of adjacent cells in historical production data commonly ranges from 0.5 to 1.2, and the standard deviation commonly ranges from 0.1 to 0.3. The corresponding difference threshold ranges from 0.6 to 1.5. For example, when the historical data average is 0.8 and the standard deviation is 0.2, the threshold is set to 1.0 to ensure that more than 95% of the same type of feature region can be correctly classified. For the cell being analyzed, after calculating its weighted distance to adjacent cells, if the weighted distance is less than the set difference threshold, the adjacent cells are classified into the same feature region as the current cell. Starting from the formed feature region, the search is expanded to all unclassified adjacent cells one by one. Each time, only one adjacent cell is expanded, and the average weighted distance between the cell and all cells in the current feature region is calculated. If the average is less than the threshold, the cell to be searched is included in the current feature region. This search process is repeated until there are no unclassified adjacent cells that meet the conditions. The expansion stops, and the labeling of the feature region is completed. The unlabeled analysis cells are processed in the order from workshop entrance to exit, from left to right, and from top to bottom, until all analysis cells are labeled to the corresponding feature regions, and the complete feature region labeling results are obtained.

[0020] Step 204a: Based on the feature region marking results, identify the boundary analysis cells of each feature region, extract the outer contour points of the boundary cells, and form a boundary point sequence; based on the boundary point sequence, connect adjacent boundary points to construct an initial polygon region, specifically including: according to the feature region marking results, check each of the four adjacent cells of each cell one by one. If there are cells with different feature regions or unmarked cells, then the cell is determined to be a boundary analysis cell of the feature region; for each boundary analysis cell, take the center of the cell as the internal reference point of the feature region, and determine whether each vertex is located outside the feature region. If the line connecting the vertex and the center points to the external region, then the vertex is an outer contour point. Each boundary analysis cell extracts 1 to 4 outer contour points according to its edge position in the feature region to ensure that the outer contour points can accurately reflect the boundary position of the cell; starting from the leftmost outer contour point of the feature region, arrange all the outer contour points corresponding to the feature region in a clockwise direction to ensure that the sequence is continuous without jumps, and the straight-line distance between adjacent points does not exceed the diagonal length of the analysis cell. For example, when the cell side length is 1 meter, the diagonal length is about 1.414 meters, forming a continuous boundary point sequence. Following the sequence of boundary points, adjacent outer contour points are connected sequentially with straight lines to form a closed polygon, which is the initial polygon region of the corresponding feature area.

[0021] Step 204b: Based on the initial polygonal region, detect the overlapping relationship between adjacent polygonal regions and identify the boundary intersections of the overlapping regions. Based on the boundary intersections, perform a union operation on the overlapping polygonal regions to merge the overlapping regions, eliminate duplicate boundaries, and generate a merged polygonal region. Specifically, this includes: checking each initial polygonal region with its adjacent initial polygonal regions one by one, performing pairwise intersection checks on all edges of the two polygons. If there are intersecting edges and the intersection point is within the valid range of the two edges (i.e., the intersection point coordinates are between the start and end coordinates of the two edges), then it is determined that the two polygonal regions have an overlapping relationship. For the two overlapping polygonal regions, traverse all their edges, calculate the intersection points of each edge with each edge of the other polygon, and remove duplicate intersection point coordinates to obtain all boundary intersections, ensuring that the intersection point coordinates are accurate to the centimeter level and that no intersection points of the overlapping parts are missed. Based on the identified boundary intersections, using the intersection points as dividing points, retain the non-overlapping parts of the two polygons and the outer edges between the intersection points, delete the inner edges of the overlapping parts, and integrate the entire spatial range of the two polygonal regions into a whole, forming a complete closed region, which is the merged polygonal region.

[0022] Step 204c: Based on the merged polygonal regions, detect gap regions between polygonal regions and identify the boundary features of the gap regions; based on the boundary features of the gap regions, perform a difference operation on adjacent polygonal regions to fill the gap regions, generate continuous polygonal boundaries, and process the boundary vertices to generate polygonal outlines. Specifically, this includes: checking all merged polygonal regions, traversing the boundaries of adjacent polygonal regions, calculating the shortest distance between adjacent boundaries; if the shortest distance is greater than 0 and less than the side length of the analyzed cell (e.g., when the cell side length is 1 meter, the shortest distance is 0 to 1 meter), then the space not covered by any polygonal region is determined to be a gap region; analyzing the boundary features of the gap regions, including the coordinates of the four vertices of the gap region, the area size, and the identifiers of adjacent polygonal regions, to clarify the gap region and its boundary characteristics. The positional relationship and production characteristics of the surrounding polygonal regions are correlated. Based on the boundary characteristics of the gap region, the weighted distance between the gap region and the adjacent polygonal regions is calculated. The gap region is assigned to the adjacent polygonal regions with smaller weighted distances and higher similarity in production characteristics. The difference operation is performed between the polygonal region and the gap region, and the boundary vertices of the gap region are incorporated into the boundary of the polygonal region. Adjacent vertices are reconnected, and the spatial range of the gap region is completely incorporated into the polygonal region, filling the gap and making the adjacent polygonal regions seamlessly connected to generate a continuous polygonal boundary. The moving average method of three adjacent vertices is used to smooth the vertices of the connected polygonal boundary, adjust the vertex coordinates so that the included angle of the boundary line segment is not less than 30 degrees, remove sharp edges, make the boundary transition natural, and form a complete and continuous polygonal outline.

[0023] Step 204d: Based on the polygonal contours, verify the topological relationships of the polygonal regions and generate corrected polygonal regions. Specifically, this includes: verifying the topological relationships of the regions corresponding to each polygonal contour; first, checking if the coordinates of the first and last vertices of the polygon are completely consistent; then, traversing all edges formed by connecting adjacent vertices, confirming that the endpoint of the previous edge coincides with the starting point of the next edge, ensuring all edges are continuous without breaks, and the region is a closed polygon; when verifying the relationship between adjacent polygonal regions, first obtain all boundary segments of two adjacent polygons, perform intersection checks on each pair of segments, calculate the overlap length of the intersecting segments using a segment intersection algorithm, and the sum of all overlap lengths is the boundary intersection length of the adjacent regions; if the boundary intersection length is... If the degree is greater than 0 and less than 1% of the length of any intersecting line segment, and the area intersection of the two regions is 0, then the boundary connection of the adjacent regions is confirmed to be correct, with no inclusion or overlap relationship and no logical contradiction. For boundary breaks, overlapping residues, or inaccurate connections found during the verification process, vertices are added at the breaks to ensure that the distance between adjacent vertices after addition does not exceed 1 / 2 of the side length of the analysis cell, thus ensuring boundary continuity. For overlapping residues, the boundary coordinates of the overlapping region are calculated, and duplicate boundary segments are deleted. For inaccurate connections, the deviation distance between the current vertex and the corresponding vertex in the adjacent region is calculated, and the vertex coordinates are adjusted to make the deviation distance less than 1 cm, ensuring that the adjacent boundaries fit perfectly. After comprehensive verification and correction, the corrected polygonal region is generated.

[0024] Step 205: Based on the corrected polygonal regions, define the final analysis region boundaries, dividing the spatial structure plane into multiple analysis regions. Specifically, this includes: using the boundary vertex coordinates of the corrected polygonal regions as a basis, arranging all vertex coordinates in a clockwise order and connecting them sequentially to form a closed boundary. The boundary range is strictly defined according to the vertex connection order of the polygonal outline, with each vertex coordinate accurate to the centimeter level; according to the boundary range of each corrected polygonal region, the entire spatial structure plane is divided into multiple independent regions, each of which is an analysis region. A unique identifier is assigned to each analysis region, and it is associated with the corresponding production features such as material A processing area, equipment group 1 operating area, and buffer area B; the sum of the areas of all analysis regions is calculated and compared with the total area of ​​the spatial structure plane. The total area of ​​the spatial structure plane is the product of the workshop length and width. It is ensured that the difference between the sum of the analysis region areas and the total area of ​​the plane is less than 0.1%, so that all analysis regions completely cover the entire spatial structure plane without overlap or gaps, thus completing the division of the analysis regions.

[0025] This embodiment effectively avoids overlapping or gaps in feature regions, ensuring the accuracy of the analysis region division and laying a reliable spatial foundation for subsequent data distribution feature calculations.

[0026] In a preferred embodiment of the present invention, step 3, based on the feature region marking results, extracts the boundary point sequence of each feature region to construct an initial polygonal region; based on the initial polygonal region, performs boundary processing on adjacent polygonal regions using polygon Boolean operations to generate a corrected polygonal region, may include: Step 301: Based on the feature region marking results, identify the boundary analysis cells of each feature region, extract the outer contour points of the boundary cells, and form a boundary point sequence; based on the boundary point sequence, connect adjacent boundary points to construct an initial polygon region. Specifically, this includes: re-checking the feature region marking results, comparing the first marking results with the actual production layout diagram of the workshop, correcting incorrectly marked cells (such as cells that mistakenly mark the equipment maintenance area as the production area), and ensuring that the range of each feature region is accurate; on this basis, re-examine the features of the four adjacent cells of each cell to accurately identify the boundary analysis cells of each feature region, avoiding omissions or misjudgments; and extracting the outer contour points of each boundary analysis cell a second time to verify the first extraction. Check if the contour point coordinates are within the physical boundary of the workshop. Delete points that exceed the boundary and add any missing boundary points. The outer contour points are still the vertices of the cell edges that are far from the interior of the feature area. Rearrange these outer contour points in clockwise order according to their spatial position to ensure that the arrangement of the points accurately reflects the actual direction of the boundary, forming an ordered sequence of boundary points. Connect adjacent outer contour points with straight lines according to the order of the points in the boundary point sequence to form a closed polygon. Calculate the area of ​​this polygon and the total area of ​​the analysis cells within the feature area: Total area of ​​analysis cells = Area of ​​a single analysis cell × Number of cells within the feature area, where the area of ​​a single analysis cell is the square of its side length. The polygon area is calculated using the shoelace formula, i.e., recording the coordinates sequentially according to the vertex order (…). , ()( , )…( , ),calculate (in Substitute into the formula for the area of ​​the polygon Obtain the area of ​​the polygon; use the formula error The calculation error is ensured to be no more than 5%. This polygon is the initial polygon region that has been confirmed twice and can accurately surround the corresponding feature region.

[0027] Step 302: Based on the initial polygonal region, detect the overlapping relationship between adjacent polygonal regions and identify the boundary intersection points of the overlapping regions; based on the boundary intersection points, perform a union operation on the overlapping polygonal regions to merge the overlapping regions, eliminate duplicate boundaries, and generate a merged polygonal region. Specifically, this includes: checking each of the initial polygonal regions after secondary confirmation one by one, using a magnification method to magnify the polygonal boundaries by 10 times before checking to ensure that no tiny overlapping areas smaller than 1 square meter are missed, focusing on detecting the spatial range between adjacent polygonal regions to accurately determine whether there are overlapping parts; for polygonal regions with overlapping, meticulously search for all boundary intersection points where their boundaries intersect, traverse all edges of the two polygons, calculate the intersection points of each edge with each edge of the other polygon using a line segment intersection algorithm, remove duplicate coordinates of the intersection points, retain unique coordinate values, and record the accurate coordinates of each intersection point and the corresponding numbers of the two intersecting edges to provide a basis for subsequent merging operations. Using the boundary intersection as a reference, a union operation is performed on the overlapping polygonal regions, retaining the non-overlapping parts of the two polygons and the outer edges between the intersection points, while completely deleting the inner edges of the overlapping parts, thus integrating the entire range of the two regions into a single entity. The areas of the two original regions and the overlapping region are calculated using the shoelace formula (by determining the vertices of the overlapping polygons through the intersection points and substituting them into the formula). The theoretical value is calculated according to the formula: Area of ​​merged region = Area of ​​original region 1 + Area of ​​original region 2 - Area of ​​overlapping region. At the same time, the actual area of ​​the merged region is calculated using the shoelace formula to ensure that the error between the actual area and the theoretical value is less than 0.5%, generating a merged polygonal region with clear boundaries and accurate range.

[0028] Step 303: Based on the merged polygonal regions, detect the gap regions between the polygonal regions and identify the boundary features of the gap regions; based on the boundary features of the gap regions, perform a difference operation on adjacent polygonal regions to fill the gap regions, generate continuous polygonal boundaries, and process the boundary vertices to generate polygonal contours. Specifically, this includes: comprehensively detecting all merged polygonal regions, traversing the boundaries of adjacent polygonal regions, and calculating the shortest distance between adjacent boundaries: for all boundary line segments of two adjacent polygons, use the point-to-line segment distance formula P(… , ) to line segment AB( , )-( , distance Calculate the distance between any point on each line segment and another line segment. Traverse all line segment pairs and take the minimum distance as the shortest distance between adjacent boundaries. Accurately identify all gap regions and record in detail the coordinates of the four vertices, area size (calculated using the shoelace formula), key vertices on the boundary, and adjacent region identifiers for each gap region, clarifying the relationship between the gap region and the surrounding polygonal regions. Based on the boundary features of the gap region and the production characteristic attributes of the surrounding polygonal regions, calculate the weighted distance between the gap region and each adjacent region: first extract the feature items of the gap region (data point density = number of data points in the gap region / area of ​​the gap region, average value of equipment operating parameters, material type proportion, work-in-process quantity density = number of work-in-process items in the gap region / area of ​​the gap region), then extract the corresponding feature items of the adjacent regions, and substitute them into the formula according to the weights set in step 202 (data point density 0.25 to 0.35, average value of equipment operating parameters 0.25 to 0.35, material type proportion 0.15 to 0.25, work-in-process quantity density 0.15 to 0.25). The gap area is assigned to the adjacent polygonal region with the smallest weighted distance and the highest similarity in production features. A difference operation is performed between this adjacent polygonal region and the gap area, incorporating the boundary vertices of the gap area into the boundary of the polygonal region. Adjacent vertices are reconnected, completely incorporating the spatial range of the gap area into the polygonal region. Based on the actual physical obstacles in the workshop, such as columns and conveyor belts, the coordinates of the boundary vertices are adjusted to ensure the distance from the vertex to the obstacle edge is no less than 5 cm, ensuring the boundary conforms to the actual layout and filling the gap area, making all polygonal regions a continuous and connected whole. The boundary vertices of the filled polygons are then finely adjusted to optimize vertex positions and check for self-intersecting edges (using a line segment intersection algorithm to determine whether line segments other than adjacent edges intersect), ensuring the contour has no self-intersecting edges and that all vertex coordinates are within the physical boundary of the workshop, generating a smooth and accurate polygonal contour.

[0029] Step 304: Based on the polygonal outline, verify the topological relationship of the polygonal region and generate the corrected polygonal region. Specifically, this includes: verifying the topological relationship of the refined polygonal outline, checking each polygonal region one by one, confirming that the vertex order is consistent (clockwise) when calculating the area using the shoelace formula, checking that the coordinates of the first and last vertices coincide, and ensuring that all edges are continuous and unbroken to ensure that the region range is accurate; verifying the connection relationship between adjacent polygonal regions, calculating the boundary intersection length of adjacent regions (using the same method as the line segment overlap length summation in step 204d), confirming that the intersection length is greater than 0 and less than 1 cm, and that the area intersection of the two regions is 0, i.e., no overlap, no gap, reasonable topological structure, no logical conflict, and each region is accurately mapped to the actual production functional areas of the workshop, such as processing areas, buffer areas, and logistics channels. For any topological issues discovered during the verification process, the vertex coordinates and boundary orientation of the polygon outline are adjusted in a timely manner to correct the region range. If self-intersecting edges exist, the intersecting vertices are split and the boundaries are replanned. If the boundary connection deviation exceeds 1 cm, the vertex coordinates are adjusted to make the deviation less than 0.5 cm. After repeated verification and adjustment, the topological relationship of each polygon region is ensured to be completely correct and the region range is accurate, and the corrected polygon region is generated.

[0030] This embodiment improves the accuracy of the scheduling scheme in perceiving the actual situation on the production site, reduces the mapping deviation between the scheduling model and the physical space, and enhances the adaptability of the resource allocation scheme to the dynamic production environment.

[0031] In a preferred embodiment of the present invention, step 4, which involves allocating the structured data set to corresponding analysis regions according to spatial coordinates, calculating the center position, orientation angle, and dispersion characteristics of the data points in each analysis region, extracting the characteristic parameters of the data spatial distribution, and generating parameter adjustment amounts, may include: Step 401: Match each data point in the structured dataset to the corresponding analysis region to establish a mapping relationship between data points and analysis regions. Based on the mapping relationship, calculate the coordinates of the data points in each analysis region to obtain the coordinates of the data distribution center of each analysis region. Specifically, each data point contains physical location x-axis coordinates and y-axis coordinates. Compare the x-axis coordinates and y-axis coordinates of each data point with the boundary coordinates of all analysis regions to determine whether the coordinates of the data point fall within the boundary range of a certain analysis region. If it falls within a certain analysis region, establish a corresponding relationship between the data point and the analysis region. After completing the matching of all data points, a mapping relationship between data points and analysis regions is formed. Based on this mapping relationship, all data points within each analysis region are selected, and the x-axis coordinates of these data points are extracted. The sum of all x-axis coordinates is obtained by adding them together, and then the sum of the x-axis coordinates is divided by the number of data points within the analysis region to obtain the average x-axis coordinate value. Using the same method, the y-axis coordinates of all data points are extracted, and the sum of all y-axis coordinates is obtained by adding them together, and then the sum of the y-axis coordinates is divided by the number of data points to obtain the average y-axis coordinate value. The average x-axis coordinate value and the average y-axis coordinate value together constitute the coordinates of the data distribution center position of the analysis region.

[0032] Step 402: Based on the center position coordinates, calculate the covariance matrix of data points relative to the center position within each analysis region, extract the principal eigenvectors of the covariance matrix, and determine the principal direction angles. Specifically, this includes: for each analysis region, subtracting the average x-axis coordinate of the center position from the x-axis coordinate of each data point within that region to obtain the x-axis deviation of each data point relative to the center; and subtracting the average y-axis coordinate of the center position from the y-axis coordinate of each data point to obtain the y-axis deviation of each data point relative to the center. When calculating the elements in the first row and first column of the covariance matrix, multiply each x-axis deviation value by itself to obtain the squared x-axis deviation value, sum all the squared x-axis deviation values ​​to obtain the sum of the squared x-axis deviation values, and then divide the sum of the squared x-axis deviation values ​​by the number of data points in that region minus one. When calculating the elements in the first row and second column and the second row and first column of the covariance matrix, multiply each data point's x-axis deviation value by the corresponding y-axis deviation value to obtain the deviation product value, sum all the deviation products to obtain the sum of the deviation products, and then divide the sum of the deviation products by the number of data points minus one.

[0033] When calculating the second row and second column of the covariance matrix, each y-axis deviation value is multiplied by itself to obtain the squared y-axis deviation value. All squared y-axis deviation values ​​are summed to obtain the total squared y-axis deviation. This total squared y-axis deviation is then divided by the number of data points minus one. These three elements construct the complete covariance matrix. The covariance matrix is ​​then subjected to eigenvalue decomposition, and the eigenvector corresponding to each eigenvalue is calculated. The eigenvector corresponding to the eigenvalue with the largest value is selected as the principal eigenvector. The principal eigenvector contains both x-axis and y-axis components. The tangent value is obtained by dividing the y-axis component of the principal eigenvector by the x-axis component. The arctangent operation is then performed on this tangent value to obtain the angle between the principal eigenvector and the positive x-axis direction. This angle represents the principal direction angle of the data distribution.

[0034] Step 403: Based on the principal direction angle, project the data points within the analysis area onto the principal direction and the direction perpendicular to the principal direction. Calculate the standard deviation of the projected distances in both directions to obtain the dispersion characteristics. Specifically, the angle perpendicular to the principal direction is the principal direction angle plus 90 degrees, ensuring that the two directions are perpendicular to each other and cover all dimensions of the plane. For each data point, calculate the vector of the data point relative to the center position, decompose this vector onto the principal direction, and the length of the decomposed vector is the projected distance of the data point on the principal direction. Record the projected distances of all data points in the principal direction sequentially. Using the same method, project the distance of each data point relative to the center position... The vector is decomposed into the vertical direction to obtain and record the projected distance of each data point in the vertical direction. To calculate the standard deviation of the projected distance in the principal directions, first, all projected distances in the principal directions are summed to obtain the total projected distance in the principal directions. Then, the total projected distance in the principal directions is divided by the number of data points to obtain the average projected distance in the principal directions. Next, the average projected distance in the principal directions is subtracted from each projected distance in the principal directions to obtain the principal direction deviation. Each principal direction deviation is multiplied by itself to obtain the squared principal direction deviation. All squared principal direction deviations are summed to obtain the total squared principal direction deviations. This total squared principal direction deviations are divided by the number of data points minus one, and the square root of the result is taken to obtain the standard deviation of the projected distance in the principal directions. Using the same calculation logic, the average projected distance in the vertical direction, the vertical deviation, the squared vertical deviation, and the total squared vertical deviation are calculated sequentially to obtain the standard deviation of the projected distance in the vertical direction. The standard deviations in both directions together constitute the dispersion characteristic of the data distribution.

[0035] Step 404: Based on the center position coordinates, principal direction angle, and dispersion characteristics, extract feature parameters, calculate the weight coefficients of each feature parameter, and generate parameter adjustment amounts. Specifically, the extracted feature parameters include the center position x-axis coordinate, center position y-axis coordinate, principal direction angle, principal direction projection standard deviation, and vertical direction projection standard deviation, totaling five feature parameters. The weight coefficients of each feature parameter are determined according to the production scheduling requirements of the discrete manufacturing workshop. The weight coefficients for the center position x-axis coordinate and y-axis coordinate both range from 0.15 to 0.25, and this range is mainly determined by considering the material transport distance. The impact of scheduling efficiency is set, and distance optimization is directly related to production flow costs. The weight should be guaranteed to have a certain proportion but not be excessively tilted. The weight coefficient of the main direction angle ranges from 0.15 to 0.25, and is set according to the matching degree between the equipment layout direction and the task flow direction. The higher the matching degree, the more time the equipment adjustment will be reduced. The weight should be as important as the position coordinate. The weight coefficients of the main direction projection standard deviation and the vertical direction projection standard deviation both range from 0.15 to 0.25. They are set based on the efficiency of centralized resource utilization. The degree of dispersion directly affects the equipment load balance and should be given reasonable weights to ensure the resource optimization effect.

[0036] The weight coefficients of the five feature parameters are all positive numbers and their sum is fixed at one. For example, the weight coefficients for the center position x-axis coordinate, center position y-axis coordinate, principal direction angle, principal direction projection standard deviation, and vertical direction projection standard deviation are all 0.2. This example is completely within the weight range of each parameter. When generating the parameter adjustment amount, the actual value of each feature parameter is first converted into a standardized value. The conversion method is to subtract the historical minimum value of the feature parameter from the actual value of the feature parameter, and then divide by the historical maximum value minus the historical minimum value of the parameter. Then, each standardized feature parameter value is multiplied by the corresponding weight coefficient to obtain five weighted values. The sum of these five weighted values ​​is the parameter adjustment amount.

[0037] This embodiment accurately captures the spatial distribution characteristics and temporal correlation patterns of production data, providing reliable data support for resource allocation.

[0038] In a preferred embodiment of the present invention, step 5, using the parameter adjustment amount as the basis for neural network training, and using the normalized structured data set as training samples to complete model training and obtain an initial resource allocation scheme, may include: Step 501: Based on the parameter adjustment, determine the configuration of the number of neurons and the selection of activation functions for each hidden layer of the neural network, and construct the deep neural network structure. Based on the deep neural network structure, set the hyperparameters for the learning rate, batch size, and number of iterations during the neural network training process. Specifically, the deep neural network includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer is consistent with the number of feature parameters, i.e., five neurons, each corresponding to one of the five feature parameters. Three hidden layers are set: the first layer has twice the number of neurons in the input layer (ten neurons), the second layer has eight neurons, and the third layer has five neurons. The configuration of the number of neurons refers to the numerical range of the parameter adjustment to ensure that the network can fully learn the features and resources. Regarding the mapping relationship; in terms of activation function selection, all three hidden layers use the ReLU activation function, which can effectively alleviate the gradient vanishing problem. The output layer uses a linear activation function to adapt to the continuous numerical output requirements in the resource allocation scheme; in the hyperparameter settings, the learning rate is determined according to the magnitude of parameter adjustment. When the parameter adjustment is large, the learning rate is 0.01, and when the parameter adjustment is small, the learning rate is 0.001, with an overall range between 0.001 and 0.01; the batch size is determined according to the amount of training data. When the amount of training data is large, the batch size is 128, and when the amount of data is small, it is 32, with a range between 32 and 128; the number of iterations is set to 1000 to 5000 times to ensure that the network has enough training times to achieve convergence.

[0039] Step 502: Based on hyperparameters, the normalized structured data set is input into the deep neural network, and forward propagation is performed to obtain the output results of each layer of the neural network. Specifically, this includes: normalizing all feature parameters in the structured data set by subtracting the minimum value of each feature parameter from the minimum value across all data points, and then dividing by the maximum value minus the minimum value, so that the values ​​of all feature parameters are between 0 and 1, avoiding the influence of parameters of different magnitudes on the training results. The normalized structured data is divided into a training set and a validation set in a 7:3 ratio. The training set is used for network training, and the validation set is used to monitor the training effect. During forward propagation, the input layer neurons receive the feature parameter values ​​from the training set and pass each input value to all neurons in the first hidden layer. Each neuron in the first hidden layer receives the output values ​​of all neurons in the input layer, multiplies each input value by the corresponding weight parameter, sums all the weighted values, adds the bias parameter of the neuron, and substitutes the sum into the ReLU activation function to calculate the output value of each neuron in the first hidden layer.

[0040] The output value of the first hidden layer is used as the input value of the second hidden layer. The same calculation method is used, namely, weighted summation plus bias and substitution into the activation function to obtain the output value of the second hidden layer. The output value of the second hidden layer is passed to the third hidden layer, and the above calculation process is repeated to obtain the output value of the third hidden layer. The output value of the third hidden layer is used as the input value of the output layer. Each neuron in the output layer sums the input values ​​with weights and adds the bias parameter, and outputs directly without going through the activation function, thus obtaining the output results of each layer of the neural network. The output result of the output layer corresponds to the predicted value of the initial resource allocation.

[0041] Step 503: Based on the neural network output, calculate the difference between the predicted output and the expected output to construct a loss function and evaluate the training effect of the neural network; based on the loss function, calculate the gradient values ​​of the parameters of each layer of the neural network, specifically including: the expected output value comes from the validated optimal resource allocation scheme in the historical production data of the workshop, and each training sample corresponds to a value of the optimal resource allocation scheme as the expected output; the loss function is constructed using the mean squared error function, calculate the difference between the predicted output value and the expected output value of each output neuron, multiply each difference by itself to obtain the square of the difference, add the squares of the differences of all output neurons, and divide by the number of output neurons to obtain the loss value of a single training sample, and add the loss values ​​of all training samples and divide by the total number of training samples to obtain the average loss function value.

[0042] The training effect is evaluated using the average loss function value. A smaller average loss function value indicates that the predicted output is closer to the expected output, and the better the training effect. The gradient value is calculated using the backpropagation algorithm, starting from the output layer. First, the partial derivative of the output layer's loss function with respect to the output value is calculated. Then, combined with the output layer's weight and bias parameters, the partial derivative of the loss function with respect to these parameters is calculated. The gradient information from the output layer is then passed to the third hidden layer. Using the derivative of the ReLU activation function, the partial derivative of the loss function with respect to the output value of the third hidden layer is calculated. This partial derivative with respect to the weight and bias parameters of this layer is then further calculated. Following the same logic, the gradient information is passed to the second and first hidden layers, and the partial derivatives of the weight and bias parameters of each layer are calculated. These partial derivatives are the gradient values ​​of the parameters in each layer.

[0043] Step 504: Based on the gradient value and the learning rate determined by the parameter adjustment, update the weight parameters and bias parameters of each layer of the neural network to obtain the updated neural network parameters. Repeat the forward propagation calculation, loss function construction, gradient calculation, and parameter update process until the loss function value converges to a preset threshold, completing the neural network training and obtaining the initial resource allocation scheme. Specifically, during parameter update, for each weight parameter of each layer, subtract the learning rate multiplied by the gradient value corresponding to the current weight parameter to obtain the updated weight parameter; for each bias parameter of each layer, subtract the learning rate multiplied by the current bias parameter to obtain the updated weight parameter. The habit rate is multiplied by the gradient value corresponding to the bias parameter to obtain the updated bias parameter. After the update, the validation set data is input into the neural network to perform forward propagation and calculate the average loss function value of the validation set. The process of forward propagation of the training set, calculating the average loss function value, calculating the gradient value, updating the parameters, and validating the validation set is repeated. The average loss function values ​​of the training set and the validation set are recorded once for each iteration. The preset threshold is set according to the accuracy requirements of production scheduling, usually 0.001. When the average loss function value of the validation set is less than the preset threshold for ten consecutive iterations and no longer decreases significantly, the loss function value is considered to have converged, and training is stopped. The neural network parameters at this time are the optimal parameters. The complete structured data set is input into the trained neural network, and the output result obtained by forward propagation is the initial resource allocation scheme. This scheme specifies the task allocation quantity for each device, the processing time of each task, the material allocation ratio, and the flow path of work-in-process.

[0044] In this embodiment, the neural network construction and training process is rigorous, and the initial resource allocation scheme can fully adapt to the production needs of discrete manufacturing workshops.

[0045] In a preferred embodiment of the present invention, step 6, using the initial resource allocation scheme as the basis for task sequence optimization, involves constructing a time probe line on the time axis to detect overlaps in the execution periods of different tasks, identifying resource conflict points, and generating a preliminary scheduling plan under the conditions of satisfying equipment capacity and working time constraints. This may include: Step 601: Based on the initial resource allocation scheme, extract the execution time parameters of each task; based on the execution time parameters of each task, establish equally spaced time probe points on the time axis to construct a time probe line. Specifically, this includes: extracting the earliest planned start time, latest planned start time, planned execution duration, and latest planned end time of each task from the initial resource allocation scheme. This information together constitutes the task's execution time parameters; taking the production start time of the day as the starting point of the time axis and the production end time of the day as the ending point of the time axis, determine the interval of the time probe points based on the average execution duration of all tasks. For example, if the average execution duration of all tasks is 30 minutes, then the interval of the time probe points is set to 10 minutes; starting from the starting point of the time axis, mark the time points sequentially according to the set intervals until the end point of the time axis. Each marked time point is a time probe point, and the time interval between two adjacent time probe points is a time interval. All time probe points and time intervals together constitute a time probe line, which is used to monitor the execution status of tasks in different time periods.

[0046] Step 602: Based on the time probe line, sequentially detect the combinations of tasks executed simultaneously within each time interval to identify task pairs with conflicting resource requirements; based on the conflicting task pairs, analyze the degree of conflict in equipment usage and time allocation for each task pair to determine the conflict level. Specifically, this includes: traversing each time interval on the time probe line, viewing all tasks currently being executed within that interval, and counting the equipment type, equipment number, and required time length for each task; for any two tasks within the same time interval, if they require the same equipment and their respective execution time periods overlap within that time interval, or if their total required time length exceeds the maximum available time of that equipment within that time interval, then these two tasks are determined to have conflicting resource requirements. Conflicting task pairs; when analyzing the degree of equipment usage conflict, calculate the overlap duration of the execution time of the two tasks on the equipment, divide the overlap duration by the total available working hours of the equipment within that time interval, and obtain the equipment usage conflict ratio; when analyzing the degree of working hour allocation conflict, calculate the overlap portion of the working hours required by the two tasks, divide the length of the overlap portion by the total available working hours within that time interval, and obtain the working hour allocation conflict ratio; add the equipment usage conflict ratio and the working hour allocation conflict ratio and divide by two to obtain the overall conflict degree; classify the conflict level according to the overall conflict degree: a comprehensive conflict degree between 0 and 0.3 is Level 1 conflict, between 0.3 and 0.6 is Level 2 conflict, and between 0.6 and 1 is Level 3 conflict. The higher the conflict level, the more severe the conflict.

[0047] Step 603: Based on the conflict level, prioritize conflicting task pairs according to their severity and adjust their execution sequence. Specifically, sort all conflicting task pairs from highest to lowest conflict level, prioritizing level 3 conflicts, followed by level 2 conflicts, and finally level 1 conflicts. When processing each conflicting task pair, first query the order priorities corresponding to the two tasks. The closer the order delivery date, the higher the priority. Tasks with higher priority are considered critical tasks, and their original planned execution time and resource allocation are retained. For non-critical tasks, postpone their start time by a period equal to the conflict overlap duration. If the non-critical task still conflicts with other tasks after postponement, continue postponing by the overlap duration until the task has no conflicts with any other task in terms of equipment usage and time allocation. Record the adjusted start and end times of the non-critical tasks and re-organize the execution sequence of all tasks according to the adjusted times to ensure a logically sound temporal sequence between tasks.

[0048] Step 604: Based on the adjusted task execution order, verify the satisfaction of equipment capacity constraints and working time constraints, and generate a preliminary scheduling plan. Specifically, this includes: sorting all conflicting task pairs in descending order of conflict level, prioritizing level 3 conflicting task pairs, followed by level 2 conflicts, and finally level 1 conflicts; when processing each conflicting task pair, first query the order priority corresponding to the two tasks, with higher priority orders having closer delivery dates; tasks with higher priority are critical tasks, and their original planned execution time and resource allocation are retained first; for non-critical tasks, postpone their start time by a period equal to the conflict overlap period. If the non-critical task still conflicts with other tasks after postponement, continue to postpone it step by step according to the overlap period until the task has no conflict with all other tasks in terms of equipment usage and working time allocation; record the adjusted start and end times of non-critical tasks, and re-organize the execution order of all tasks according to the adjusted times to ensure that the temporal logic between tasks is reasonable.

[0049] In this embodiment, the handling of task conflicts is highly targeted, and the initial scheduling plan strictly complies with equipment capacity and working time constraints.

[0050] In a preferred embodiment of the present invention, step 7, evaluating the preliminary scheduling plan based on multi-source data from the production site, updating scheduling parameters, and generating an optimized scheduling scheme, may include: Step 701: Based on the preliminary scheduling plan, extract the task execution sequence and resource allocation scheme from the plan to generate a plan execution dataset; based on multi-source data from the production site, obtain the actual operating status of each production unit and generate an actual execution dataset. Specifically, this includes: extracting the planned start time, planned end time, planned equipment number, planned material consumption quantity, and planned working hours for each task from the preliminary scheduling plan, classifying and organizing this information by task number, and generating a plan execution dataset; multi-source data from the production site is collected in real time by sensors, radio frequency identification equipment, visual acquisition devices, and IoT terminals deployed in the workshop, including actual equipment running time, actual number of processed tasks, actual material consumption quantity, actual work-in-process turnover speed, and actual equipment downtime. This data reflects the actual operating status of each production unit. Integrate these real-time collected data by task number and time dimension to generate an actual execution dataset.

[0051] Step 702: Based on the planned execution dataset and the actual execution dataset, compare and analyze the time execution deviation and resource usage deviation of each task, and calculate the deviation index; based on the deviation index, evaluate the matching degree between the preliminary scheduling plan and the actual production operation, and generate the matching degree evaluation result, specifically including: comparing the planned execution data and actual execution data of each task; when calculating the time execution deviation, subtract the planned end time from the actual end time of the task to obtain the time deviation value; if the actual end time is earlier than the planned end time, the deviation value is negative, otherwise it is positive; when calculating the resource usage deviation, subtract the planned resource consumption from the actual resource consumption of the task to obtain the resource deviation value, similarly distinguishing between positive and negative deviations; when calculating the deviation index, first calculate the time... The time deviation rate is calculated by dividing the absolute value of the time deviation of each task by the planned execution time of that task. The resource deviation rate is calculated by dividing the absolute value of the resource deviation of each task by the planned resource consumption of that task. The time deviation rates of all tasks are summed and divided by the total number of tasks to obtain the average time deviation rate. The resource deviation rates of all tasks are summed and divided by the total number of tasks to obtain the average resource deviation rate. The average time deviation rate and the average resource deviation rate are summed and divided by two to obtain the comprehensive deviation index. The degree of matching is evaluated based on the comprehensive deviation index. The smaller the comprehensive deviation index, the higher the degree of matching between the preliminary scheduling plan and the actual production operation. A matching evaluation result containing the comprehensive deviation index, average time deviation rate, and average resource deviation rate is generated.

[0052] Step 703: Based on the matching degree evaluation results, identify the types and directions of scheduling parameters that need adjustment, and determine the adjustment range for each scheduling parameter. Specifically, this includes: analyzing the matching degree evaluation results; if the average time deviation rate is high, it indicates that the time-related scheduling parameters are set unreasonably, and the types of parameters that need adjustment include task start time interval, equipment switchover time reserve, and order delivery period buffer time. The adjustment direction is determined based on the sign of the time deviation. If the actual completion time of most tasks is later than planned, the adjustment direction is to extend the equipment switchover time reserve or the order delivery period buffer time; if the actual completion time of most tasks is earlier than planned... The adjustment direction is to shorten the relevant parameters; if the average resource deviation rate is high, it indicates that the scheduling parameters related to resource allocation are set unreasonably. The types of parameters that need to be adjusted include equipment resource quotas, material allocation ratios, and work-in-process buffer capacity. The adjustment direction is also determined according to the sign of the resource deviation. If the actual resource consumption is more than the plan, the allocation ratio of the relevant resources should be increased, and vice versa. When determining the adjustment range, it is determined according to the magnitude of the comprehensive deviation index. The larger the comprehensive deviation index, the larger the adjustment range. The adjustment range should be controlled between 5% and 15% of the original parameter value to ensure that the adjusted parameters do not exceed the range allowed by the production process.

[0053] Step 704: Based on the adjustment range, update the scheduling parameters, re-optimize the task execution sequence and resource allocation scheme, and generate an optimized scheduling scheme. Specifically, this includes: updating the corresponding scheduling parameters according to the determined adjustment range and direction, such as extending the equipment switchover time reserve from 5 minutes to 6 minutes, adjusting the resource quota for a certain equipment from processing 3 tasks per hour to processing 4 tasks per hour, and adjusting the material allocation ratio from 6:4 to 7:3; based on the updated scheduling parameters, re-examine the execution logic of all tasks, and reorder the task execution sequence in conjunction with the actual equipment capacity, real-time material supply, and work-in-process flow status to avoid new resource and timing conflicts between tasks; simultaneously, according to the updated resource allocation parameters, reallocate the equipment resources, material resources, and working hours resources for each task to ensure that resource allocation better matches task requirements; and integrating the optimized task execution sequence resource allocation scheme and time nodes to generate an optimized scheduling scheme that better fits the actual operating status of the production site, improving the accuracy and adaptability of scheduling.

[0054] This embodiment conducts deviation analysis based on actual production data, optimizes the scheduling scheme to better fit the actual production site, and improves the flexibility and practicality of scheduling.

[0055] In a preferred embodiment of the present invention, step 8, converting the optimized scheduling scheme into equipment control commands and transmitting them to the control terminal of the production unit to complete the closed-loop optimization of production resources, may include: Step 801: Based on the optimized scheduling scheme, extract the task execution sequence and resource allocation parameters to generate an equipment control instruction set; transmit the equipment control instruction set to the control terminal of the production unit. Specifically, this includes: extracting the complete task execution sequence and resource allocation parameters from the optimized scheduling scheme. The task execution sequence requires extracting the unique number of each task, execution order, planned start time, planned end time, and corresponding production unit identifier; the resource allocation parameters require extracting the equipment number corresponding to each production unit, equipment operating parameters (such as the rotation speed, temperature, and pressure of processing equipment, and the operating speed of logistics equipment), material supply, work-in-process allocation duration, and work-in-process flow nodes; classifying the extracted information according to the production unit type and associating it with the equipment number. Based on the corresponding task information and resource parameters, generate equipment control instruction sets. Each instruction must include the equipment number, task number, execution start time, execution end time, specific operating parameter values, material supply batch and quantity, and process connection requirements, ensuring complete and accurate instruction information. Establish an industrial Ethernet communication link, group the generated equipment control instruction sets by production unit, and transmit them to the corresponding production unit's control terminal via the Modbus communication protocol. Control terminals include programmable logic controllers, CNC systems, and logistics scheduling controllers. After transmission, receive instruction reception confirmation signals from each control terminal. If any terminal fails to acknowledge receipt, retransmit the corresponding instruction set until all control terminals have completed reception confirmation, ensuring that the instruction transmission is successful.

[0056] Step 802: Based on the instructions received by the control terminal, execute production tasks and collect multi-source data from the production site. Based on the collected multi-source data from the production site, monitor the progress of production task execution and resource usage status, and evaluate the actual execution effect of the optimized scheduling scheme. Specifically, after receiving the instructions, the control terminal parses the operating parameters and time requirements in the instructions, drives the corresponding production equipment to start operation, and executes production tasks in the order of the instructions. For example, processing equipment processes materials at a set speed and temperature, and logistics equipment transfers materials according to a specified time and route. At the same time, multi-source data acquisition devices deployed on the production site collect data in real time. Sensors collect parameters such as the actual speed, temperature, pressure, and operating current of the equipment; RFID devices collect the quantity of materials used, the quantity consumed, and the location of work-in-process; visual acquisition devices collect the quantity of work-in-process and the processing quality status of the workstations; and IoT terminals collect the actual running time and downtime of the equipment. The data acquisition interval matches the equipment operating cycle. Processing equipment collects operating parameters every 10 seconds, and logistics equipment collects location data every 30 seconds.

[0057] Based on the collected data, the task execution progress is monitored, and the actual start time and actual completion time of each task are statistically analyzed. The number of completed tasks is calculated, and the number of completed tasks is divided by the total number of tasks in the production unit and then multiplied by 100% to obtain the task execution progress rate. At the same time, the process connection of each task is checked to confirm whether the process flow is completed in the order of instructions. When monitoring the resource usage status, the actual equipment running time is calculated, divided by the planned equipment running time, and then multiplied by 100% to obtain the equipment utilization rate. The actual material consumption is calculated, divided by the planned material supply, and then multiplied by 100% to obtain the material utilization rate. The deviation values ​​between the actual equipment operating parameters and the instruction settings are statistically analyzed, such as the difference between the actual speed and the set speed, and the difference between the actual temperature and the set temperature. When evaluating the actual execution effect, the task execution progress rate, equipment utilization rate, and material utilization rate are compared with the target values ​​set by the optimized scheduling plan. The impact of equipment operating parameter deviations on production quality is analyzed, and it is determined whether there are problems such as task delays, equipment overload, material waste, and process connection blockages. These factors are combined to form the execution effect evaluation result.

[0058] Step 803: Based on the actual execution results, determine whether the expected optimization goal has been achieved. If the expected optimization goal has not been achieved, return to the execution scheduling parameter update step and regenerate the optimized scheduling scheme. Based on the regenerated optimized scheduling scheme, update the equipment control instructions to achieve dynamic closed-loop optimization of production resources. Specifically, this includes: setting expected optimization goals, including a task execution progress rate of no less than 95%, equipment utilization rate of no less than 85%, material utilization rate of no less than 90%, absolute value of equipment operating parameter deviation not exceeding 5% of the set value, and no task delays exceeding 10 minutes. Compare the actual execution results data obtained in step 802 with the preset goals. If all indicators meet the preset requirements, it is determined that the expected optimization goal has been achieved, the current equipment control instructions are maintained, and the production task continues to be executed. If any indicators are not met, such as a task execution progress rate of only 88%, equipment utilization rate of 75%, or equipment overload, the process is reversed. Issues such as downtime and material shortages leading to task delays indicate that the expected goals have not been achieved, prompting a return to step 703, the scheduling parameter update step. The reasons for non-compliance are analyzed based on the performance evaluation results; for example, delays may be due to insufficient equipment changeover time, or low equipment utilization may be due to uneven resource allocation. The scheduling parameters are adjusted accordingly, such as extending the equipment changeover time or reallocating task quotas for equipment. The optimized scheduling scheme is then regenerated following step 704. Based on the regenerated optimized scheduling scheme, step 801 is repeated to extract the updated task execution sequence and resource allocation parameters, generating a new set of equipment control instructions. This instruction is transmitted to the control terminals of each production unit, overriding the original control instructions and driving the equipment to execute production tasks according to the new instructions. Steps 802 to 803 are then repeated, continuously monitoring the performance and dynamically adjusting to form a dynamic closed-loop optimization of production resources.

[0059] like Figure 2 As shown, embodiments of the present invention also provide an adaptive digital scheduling system for lean manufacturing, comprising: The data acquisition module is used to collect multi-source data from the production site, preprocess the multi-source data, and generate a structured data set. The partitioning module is used to construct a spatial structure plane based on a structured dataset, define structural analysis primitives on the spatial structure plane, compare feature similarities, mark connected regions with the same feature attributes, and divide them into multiple analysis regions. The calculation module is used to allocate the structured data set to the corresponding analysis area according to spatial coordinates, calculate the center position, orientation angle and dispersion characteristics of the data points in each analysis area, extract the characteristic parameters of the spatial distribution of the data, and generate parameter adjustment amounts; The processing module is used to use the parameter adjustment amount as the basis for neural network training, normalize the structured data set and use it as training samples to complete model training and obtain the initial resource allocation scheme. The detection module is used as the basis for task sequence optimization by using the initial resource allocation scheme as the basis. By constructing a time probe line on the time axis, it detects the overlap of different tasks in the execution period, identifies resource conflict points, and generates a preliminary scheduling plan under the condition of meeting equipment capacity and working time constraints. The evaluation module is used to evaluate the preliminary scheduling plan based on multi-source data from the production site, update scheduling parameters, and generate an optimized scheduling scheme. The transmission module is used to convert the optimized scheduling scheme into equipment control commands and transmit them to the control terminal of the production unit to complete the closed-loop optimization of production resources.

[0060] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0061] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0062] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An adaptive digital scheduling method for lean manufacturing, characterized in that, The method includes: Step 1: Collect multi-source data from the production site, preprocess the multi-source data, and generate a structured data set; Step 2: Construct a spatial structure plane based on the structured data set, define structural analysis primitives on the spatial structure plane, compare feature similarity, mark the connected regions with the same feature attributes, and divide them into multiple analysis regions; Step 3: Assign the structured dataset to the corresponding analysis regions according to spatial coordinates, calculate the center position, orientation angle and dispersion characteristics of the data points in each analysis region, extract the characteristic parameters of the spatial distribution of the data, and generate parameter adjustment values; Step 4: Use the parameter adjustment amount as the basis for neural network training, normalize the structured data set and use it as training samples to complete model training and obtain the initial resource allocation scheme. Step 5: Using the initial resource allocation scheme as the basis for task sequence optimization, by constructing a time probe line on the time axis, the overlap of different tasks in the execution period is detected, resource conflict points are identified, and a preliminary scheduling plan is generated under the condition of meeting equipment capacity and working time constraints. Step 6: Based on multi-source data from the production site, evaluate the preliminary scheduling plan, update the scheduling parameters, and generate an optimized scheduling scheme; Step 7: Convert the optimized scheduling scheme into equipment control commands and transmit them to the control terminal of the production unit to complete the closed-loop optimization of production resources.

2. The adaptive digital scheduling method for lean manufacturing according to claim 1, characterized in that, A spatial structure plane is constructed based on a structured dataset. Structural analysis primitives are defined on this plane, and feature similarity comparisons are performed. Connecting regions with the same characteristic attributes are marked and divided into multiple analysis regions, including: Each data point in the structured dataset is projected onto a two-dimensional planar coordinate grid to form a spatial structure plane; based on the spatial structure plane, an equally spaced grid is established, and each grid cell is defined as an analysis cell; Calculate the distribution feature descriptor for each analysis cell, and starting from the starting analysis cell, sequentially compare the differences in distribution feature descriptors between adjacent analysis cells. Quantify the degree of difference by calculating the weighted distance of the distribution feature descriptors. Based on the degree of difference, a difference threshold is set. When the weighted distance of the distribution feature descriptors of adjacent analysis cells is less than the threshold, they are classified into the same feature region. Starting from the classified feature region, the analysis cells with similar distribution feature descriptors are continuously searched for in the surrounding area until no adjacent cells that meet the difference condition can be found, and the feature region labeling result is obtained. Based on the feature region labeling results, the boundary point sequence of each feature region is extracted to construct an initial polygon region; based on the initial polygon region, polygon Boolean operations are used to process the boundaries of adjacent polygon regions to generate a corrected polygon region. Based on the corrected polygonal region, the final analysis region boundary is defined, dividing the spatial structure plane into multiple analysis regions.

3. The adaptive digital scheduling method for lean manufacturing according to claim 2, characterized in that, Based on the feature region labeling results, the boundary point sequence of each feature region is extracted to construct an initial polygon region; Based on the initial polygonal region, polygon Boolean operations are used to process the boundaries of adjacent polygonal regions, generating a corrected polygonal region, including: Based on the feature region labeling results, the boundary analysis cells of each feature region are identified, the outer contour points of the boundary cells are extracted, and a boundary point sequence is formed; based on the boundary point sequence, adjacent boundary points are connected to construct an initial polygon region; Based on the initial polygonal region, the overlapping relationship between adjacent polygonal regions is detected, and the boundary intersection points of the overlapping regions are identified. Based on the boundary intersection points, a union operation is performed on the overlapping polygonal regions to merge the overlapping regions, eliminate duplicate boundaries, and generate a merged polygonal region. Based on the merged polygon regions, gap regions between polygon regions are detected, and the boundary features of the gap regions are identified. Based on the boundary features of the gap regions, a difference operation is performed on adjacent polygon regions to fill the gap regions, generate continuous polygon boundaries, and process the boundary vertices to generate polygon contours. Based on the polygonal outline, the topological relationship of the polygonal region is verified, and a corrected polygonal region is generated.

4. The adaptive digital scheduling method for lean manufacturing according to claim 3, characterized in that, The structured dataset is assigned to corresponding analysis regions according to spatial coordinates. The center position, orientation angle, and dispersion characteristics of the data points in each analysis region are calculated. Feature parameters of the spatial distribution of the data are extracted, and parameter adjustment variables are generated, including: Each data point in the structured dataset is matched to its corresponding analysis region to establish a mapping relationship between the data points and the analysis regions. Based on the mapping relationship, the coordinates of the data points in each analysis region are calculated to obtain the coordinates of the data distribution center of each analysis region. Based on the center position coordinates, calculate the covariance matrix of data points in each analysis area relative to the center position, extract the principal eigenvector of the covariance matrix, and determine the principal direction angle; Based on the principal direction angle, the data points in the analysis area are projected onto the principal direction and the direction perpendicular to the principal direction, and the standard deviation of the projection distance in the two directions is calculated to obtain the dispersion characteristics. Based on the center position coordinates, main direction angle, and dispersion characteristics, feature parameters are extracted, the weight coefficients of each feature parameter are calculated, and parameter adjustment amounts are generated.

5. The adaptive digital scheduling method for lean manufacturing according to claim 4, characterized in that, The parameter adjustment amounts are used as the basis for neural network training. The structured dataset is normalized and used as training samples to complete model training, resulting in an initial resource allocation scheme, including: Based on parameter adjustment, the number of neurons and activation function selection for each hidden layer of the neural network are determined, and the deep neural network structure is constructed. Based on the deep neural network structure, hyperparameters such as learning rate, batch size, and number of iterations are set during the neural network training process. Based on hyperparameters, the normalized structured data set is input into a deep neural network, and forward propagation calculation is performed to obtain the output results of each layer of the neural network. Based on the output of the neural network, calculate the difference between the predicted output and the expected output to construct a loss function and evaluate the training effect of the neural network; based on the loss function, calculate the gradient values ​​of the parameters of each layer of the neural network. Based on the gradient value and the learning rate determined by the parameter adjustment, the weight parameters and bias parameters of each layer of the neural network are updated to obtain the updated neural network parameters. The forward propagation calculation, loss function construction, gradient calculation and parameter update process are repeated until the loss function value converges to the preset threshold, the neural network training is completed and the initial resource allocation scheme is obtained.

6. The adaptive digital scheduling method for lean manufacturing according to claim 5, characterized in that, Using the initial resource allocation scheme as the basis for task sequence optimization, a time probe line is constructed on the time axis to detect overlaps in the execution periods of different tasks, identify resource conflict points, and generate a preliminary scheduling plan under the conditions of meeting equipment capacity and time constraints, including: Based on the initial resource allocation scheme, the execution time parameters of each task are extracted; based on the execution time parameters of each task, time probe points at equal intervals are established on the time axis to construct a time probe line; Based on time probe lines, the combination of tasks executed simultaneously within each time interval is detected sequentially to identify task pairs with conflicting resource requirements; based on the conflicting task pairs, the degree of conflict in equipment usage and time allocation of each task pair is analyzed to determine the conflict level. Based on the conflict level, conflicting task pairs are prioritized according to the severity of the conflict, and the execution sequence of conflicting task pairs is adjusted to rearrange the task execution order. Based on the adjusted task execution order, verify the satisfaction of equipment capacity constraints and working time constraints, and generate a preliminary scheduling plan.

7. The adaptive digital scheduling method for lean manufacturing according to claim 6, characterized in that, Based on multi-source data from the production site, the initial scheduling plan is evaluated, scheduling parameters are updated, and an optimized scheduling scheme is generated, including: Based on the preliminary scheduling plan, the task execution sequence and resource allocation scheme in the plan are extracted to generate the plan execution dataset; based on multi-source data from the production site, the actual operating status of each production unit is obtained to generate the actual execution dataset. Based on the planned execution dataset and the actual execution dataset, the time execution deviation and resource utilization deviation of each task are compared and analyzed, and the deviation index is calculated. Based on the deviation index, the degree of matching between the preliminary scheduling plan and the actual production operation is evaluated, and the matching evaluation result is generated. Based on the matching degree evaluation results, identify the types and directions of scheduling parameters that need to be adjusted, and determine the adjustment range of each scheduling parameter; Based on the adjustment range, update the scheduling parameters, re-optimize the task execution sequence and resource allocation scheme, and generate an optimized scheduling scheme.

8. The adaptive digital scheduling method for lean manufacturing according to claim 7, characterized in that, The optimized scheduling scheme is transformed into equipment control commands and transmitted to the control terminal of the production unit to complete the closed-loop optimization of production resources, including: Based on the optimized scheduling scheme, the task execution sequence and resource allocation parameters are extracted to generate a set of equipment control instructions; the set of equipment control instructions is then transmitted to the control terminal of the production unit. Based on the instructions received from the control terminal, production tasks are executed and multi-source data from the production site is collected; based on the collected multi-source data from the production site, the progress of production task execution and resource usage status are monitored, and the actual execution effect of the optimized scheduling scheme is evaluated. Based on the actual execution results, it is determined whether the expected optimization goal has been achieved. If the expected optimization goal has not been achieved, the process returns to the execution of the scheduling parameter update step and regenerates the optimized scheduling scheme. Based on the regenerated optimized scheduling scheme, the equipment control instructions are updated to achieve dynamic closed-loop optimization of production resources.

9. An adaptive digital intelligent scheduling system for lean manufacturing, wherein the system implements the method as described in any one of claims 1 to 8, characterized in that, include: The data acquisition module is used to collect multi-source data from the production site, preprocess the multi-source data, and generate a structured data set. The partitioning module is used to construct a spatial structure plane based on a structured dataset, define structural analysis primitives on the spatial structure plane, compare feature similarities, mark connected regions with the same feature attributes, and divide them into multiple analysis regions. The calculation module is used to allocate the structured data set to the corresponding analysis area according to spatial coordinates, calculate the center position, orientation angle and dispersion characteristics of the data points in each analysis area, extract the characteristic parameters of the spatial distribution of the data, and generate parameter adjustment amounts; The processing module is used to use the parameter adjustment amount as the basis for neural network training, normalize the structured data set and use it as training samples to complete model training and obtain the initial resource allocation scheme. The detection module is used as the basis for task sequence optimization by using the initial resource allocation scheme as the basis. By constructing a time probe line on the time axis, it detects the overlap of different tasks in the execution period, identifies resource conflict points, and generates a preliminary scheduling plan under the condition of meeting equipment capacity and working time constraints. The evaluation module is used to evaluate the preliminary scheduling plan based on multi-source data from the production site, update scheduling parameters, and generate an optimized scheduling scheme. The transmission module is used to convert the optimized scheduling scheme into equipment control commands and transmit them to the control terminal of the production unit to complete the closed-loop optimization of production resources.

10. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 8.

Images