A dynamic material allocation system based on data presetting learning for multiple workstations

Abstract

The present invention relates to the field of resource allocation technology, specifically a material dynamic allocation system based on data pre-set learning for multiple workstations, the system comprising: a topology modeling module, a demand clustering module, a strategy generation module, and a decision optimization module. In the present invention, laser ranging and dynamic time regularization are combined to convert the workstation layout into a moving time parameter, the difference ratio adjusts the spatial correlation coefficient, the topological relationship matrix is ​​updated in real time, DTW quantifies the pattern similarity of the standard deviation of the demand interval and the coefficient of variation of the usage, constructs a multi-dimensional demand feature vector, improves the matching degree between the prediction and actual consumption, establishes a response capability curve based on OEE data, dynamically associates the remaining capacity with the demand characteristics through linear programming, realizes strategy pre-generation and capacity warning, constructs a collaborative unit through the Hungarian algorithm, matches materials and equipment parameters through cosine similarity, triggers strategy reconstruction and synchronously updates the topology, forms a closed-loop optimization system, and enhances the dynamic response capability across workstations.

Description

Technical Field

[0001] The present invention relates to the technical field of resource allocation, and in particular to a material dynamic allocation system for multiple workstations based on data preset learning. Background Art

[0002] Resource allocation technology is a systematic approach to allocating and scheduling limited resources, such as personnel, equipment, and supplies, across time and space through scientific methods. Its system encompasses core modules such as resource demand forecasting, supply planning, scheduling model construction and optimization, and real-time monitoring and adjustment. It is widely used in manufacturing, logistics, project management, and other fields. A typical application of this approach is the dynamic material allocation system, which leverages algorithmic models and data system support. Through technologies such as real-time inventory monitoring, intelligent demand identification, supply-demand matching decisions, and route optimization, it builds a resource priority assessment mechanism based on historical data and production plans, enabling the efficient flow of materials from storage to use. Its core approach is to dynamically optimize resource allocation strategies and transportation routes through parametric models.

[0003] Existing technologies rely on fixed historical data and rules to evaluate resource priorities and are unable to respond to changes in production plans in real time, resulting in delays in supply and demand matching. Static parameter models ignore dynamic changes in equipment efficiency and lack an early warning mechanism when the workstation load approaches the critical value, which can easily lead to imbalanced resource allocation. Path selection is based on fixed topological relationships and does not consider changes in spatial associations after layout adjustments, resulting in deviations in transportation time and decreased efficiency. Demand identification does not quantify the pattern similarity between intervals and usage fluctuations. Single-dimensional predictions are difficult to match complex consumption characteristics, resulting in inventory redundancy. Collaborative scheduling lacks the multi-dimensional adaptability of equipment parameters and material attributes, and is prone to compatibility issues when collaborating across workstations, increasing secondary deployment costs. For example, in flexible manufacturing scenarios, fixed topological relationships cause cross-regional material transportation to take longer than expected, and uneven equipment loads cause production line congestion. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a material dynamic allocation system for multiple workstations based on data preset learning.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: A material dynamic allocation system based on data preset learning for multiple workstations includes:

[0006] A topological modeling module is used to obtain a set of workstation layout coordinates using a laser rangefinder, input the coordinate set into a dynamic time warping algorithm to calculate movement time parameters, adjust the spatial correlation coefficient based on the product of the difference ratio of the time parameters and the equipment movement acceleration, generate a topological relationship matrix, and pass it to the demand clustering module;

[0007] The demand clustering module is used to call the material demand time series records of the production line MES system, input the standard deviation of the demand interval and the coefficient of variation of the single usage into the dynamic time warping algorithm to calculate the pattern similarity, generate the demand feature vector, perform the Cartesian product operation on the topological relationship matrix, and then pass it to the strategy generation module;

[0008] The strategy generation module is used to establish a response capability curve equation based on the equipment OEE data, activate the constraint conditions when the number of material types allocated to the workstation reaches the inflection point of the curve equation, input the demand feature vector and the ratio of the workstation's remaining capacity into the linear programming model to generate a preset strategy set, and pass it to the decision optimization module.

[0009] As a further solution of the present invention, the topological relationship matrix includes spatial correlation coefficient, workstation layout structure, and coordinate space position relationship; the demand feature vector includes demand pattern similarity, demand interval regularity, and material usage characteristics; the preset strategy set specifically includes material configuration strategy, capacity matching plan, and workstation load balancing plan.

[0010] As a further solution of the present invention, the dynamic time warping algorithm achieves dynamic correction of the spatial correlation coefficient by matching the morphological difference of the movement trajectory with the deviation coefficient of the time reference value;

[0011] The dynamic time warping algorithm uses the constraint that the window width is twice the standard deviation of the required interval to align the sequences;

[0012] The response capability curve equation is fitted by cubic polynomial regression to the corresponding relationship between the equipment comprehensive efficiency and the material handling capacity, wherein the regression fitting goodness threshold is set to 0.85.

[0013] As a further solution of the present invention, the topology modeling module includes: a coordinate acquisition submodule that scans the workstation layout using a laser ranging device to collect three-dimensional coordinate data of multiple device nodes, uses a Gaussian filtering algorithm to eliminate environmental noise interference, sets a coordinate offset threshold to 10% of the average spacing between adjacent nodes as a reference value, performs interpolation compensation on abnormal coordinate points, and generates a workstation coordinate set;

[0014] The movement time calculation submodule calls the workstation coordinate set, extracts the Euclidean distance between adjacent coordinate points, uses the ratio of the equipment movement speed to the trajectory length as the time reference value, aligns the time series of the movement trajectory through the dynamic time warping algorithm, and calculates the deviation coefficient of the trajectory morphology difference from the time reference value using the formula morphology difference = trajectory curvature change rate × time reference value squared to obtain the movement time parameter;

[0015] The topological relationship construction submodule calculates the time difference ratio between the differentiated movement paths based on the movement time parameters, uses the inverse of the path length as the weight coefficient, normalizes the row vectors of the spatial correlation matrix, and performs a Hadamard product operation on the normalized result and the time difference ratio to generate a topological relationship matrix;

[0016] The row vector normalization process is achieved by calculating the Euclidean length of each row element and performing a division operation.

[0017] As a further solution of the present invention, the demand clustering module includes: a demand feature extraction submodule that calls the material demand time series records of the production line MES system, extracts the time interval sequence of adjacent demand events, calculates the sequence standard deviation as the time fluctuation amount, collects single material usage data, calculates the coefficient of variation as the usage discrete amount, performs Z-score normalization processing on the time fluctuation amount and the usage discrete amount, and generates a demand feature vector;

[0018] The pattern similarity calculation submodule uses a dynamic time warping algorithm to stretch and compress the time axis of the differentiated demand sequence based on the demand feature vector, calculates the cumulative distance value of the minimum bending path between sequences, maps the distance value to a preset similarity conversion function, and obtains a pattern similarity coefficient;

[0019] The matrix operation submodule calls the pattern similarity coefficient and the topological relationship matrix, constructs the Cartesian product space of the coefficient matrix and the topological matrix, performs the product operation of the corresponding positions of the matrix elements, sums the product results according to the row vector, and generates the demand-topology association matrix;

[0020] The Z-score standardization process is calculated based on the data of the latest 30 production cycles;

[0021] The similarity conversion function is an S-shaped curve function, whose slope parameter is inversely proportional to the material allocation response time threshold;

[0022] The Cartesian product space improves the material distribution efficiency by increasing the weight factor of the movement time between workstations.

[0023] As a further embodiment of the present invention, the strategy generation module includes: a response capability modeling submodule that collects equipment OEE data, extracts equipment operating time, theoretical cycle time, and number of qualified products, calculates the equipment comprehensive efficiency benchmark value as the product of operating time utilization and performance, uses cubic polynomial regression to fit the corresponding relationship between comprehensive efficiency and material processing volume, and generates a response capability curve equation;

[0024] The constraint condition activation submodule calls the response capacity curve equation, calculates the zero point coordinate of the second-order derivative function, monitors the real-time value of the number of material types currently in the workstation, and triggers the capacity constraint flag when the value reaches the horizontal axis value corresponding to the zero point coordinate, thereby generating a workstation constraint flag set;

[0025] The strategy generation submodule calculates the ratio matrix of the remaining capacity of the station and the demand characteristic vector based on the station constraint identification set, constructs a linear programming model whose objective function is to maximize the sum of the products of the matrix elements, sets the material type constraints as a system of inequality equations, uses the simplex method to solve the model feasible solution, and generates a preset strategy set;

[0026] The goodness-of-fit threshold for the cubic polynomial regression was set at 0.85;

[0027] The zero point coordinate calculation accuracy is retained to two decimal places;

[0028] The inequality equation group includes an upper limit constraint on material types and a non-negative constraint on station capacity.

[0029] As a further embodiment of the present invention, the system further comprises:

[0030] The decision optimization module is used to construct the workstation groups with correlation coefficients higher than the critical value in the topological relationship matrix into collaborative units through the Hungarian algorithm, perform cosine similarity matching between the material types in the preset strategy set and the equipment parameters of the collaborative units, trigger the strategy reconstruction mechanism to generate dynamic deployment instructions and synchronously update them to the topological modeling module.

[0031] As a further solution of the present invention, the dynamic allocation instructions are specifically material distribution paths, workstation collaboration methods, and material-equipment combination solutions.

[0032] As a further solution of the present invention, the critical value is a weighted sum of three times the mean and standard deviation of the topological relationship matrix elements, wherein the weight of the mean is 0.6 and the weight of the standard deviation is 0.4;

[0033] The strategy reconstruction mechanism adjusts the strategy weight coefficient through the gradient descent method, and its learning rate is set to the inverse of the difference between the real-time load of the workstation and the maximum capacity of the collaborative unit.

[0034] As a further solution of the present invention, the decision optimization module includes: a collaborative unit construction submodule calls the topological relationship matrix, calculates the mean and standard deviation of the matrix elements, sets the correlation coefficient critical value to be the weighted sum of three times the mean and the standard deviation, constructs a bipartite graph of the workstation nodes, uses the Hungarian algorithm to traverse the adjacency matrix to capture the maximum weight matching, and generates a collaborative unit set;

[0035] The strategy matching submodule extracts the equipment processing speed and capacity parameters from the collaborative unit set to form a feature vector, calculates the cosine angle value between the material type characteristics and the equipment parameters in the preset strategy set, sets the matching threshold as the inverse of the feature vector modulus ratio, filters the strategy items with angle values ​​lower than the threshold, and generates an optimization strategy candidate set;

[0036] The dynamic allocation submodule calls the optimization strategy candidate set, detects the difference between the real-time load of the workstation and the maximum capacity of the collaborative unit, adjusts the strategy weight coefficient using the gradient descent method, encodes the updated strategy instructions into a JSON format data packet, and generates a dynamic allocation instruction;

[0037] The coefficient of the triple weighted sum is selected based on the correlation distribution corresponding to the peak value of the workstation collaboration efficiency in the historical data;

[0038] The calculation formula of the characteristic vector modulus ratio is the square root of the sum of the equipment processing speed and capacity parameters;

[0039] The iterative termination condition of the gradient descent method is that the change in the weight coefficient is less than 0.01 in three consecutive iterations.

[0040] Compared with the prior art, the advantages and positive effects of the present invention are:

[0041] In the present invention, a laser ranging device is combined with a dynamic time warping algorithm to convert the workstation layout coordinates into movement time parameters, and the spatial correlation coefficient is adjusted by the difference ratio, so that the topological relationship matrix reflects the changes in the physical environment in real time. The dynamic time warping algorithm quantifies the pattern similarity between the standard deviation of the demand interval and the coefficient of variation of the usage, constructs a multi-dimensional demand feature vector, and improves the matching accuracy between the demand forecast and the actual consumption of the workstation. Based on the equipment OEE data, a response capability curve equation is established, and the linear programming model is combined to dynamically associate the remaining capacity of the workstation with the demand characteristics to achieve strategy pre-generation and capacity inflection point warning, thereby optimizing resource allocation efficiency. The Hungarian algorithm constructs a high-correlation coefficient workstation collaboration unit, and the cosine similarity matches the material type and equipment parameters, triggers strategy reconstruction and synchronously updates the topological relationship, forming a closed-loop optimization system to enhance the dynamic response capability of cross-workstation collaboration. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0044] See also Figure 1A material dynamic allocation system for multiple workstations based on data preset learning includes:

[0045] Topology modeling module, demand clustering module, strategy generation module, and decision optimization module.

[0046] The topological modeling module includes: the coordinate acquisition submodule refers to the laser ranging device. First, the operator deploys the 3D laser scanner model Leica ScanStation P50 at the center of the predetermined scanning area of ​​the workstation, sets the scanning density to 10,000 points per square meter, and the scanning angle range covers 360 degrees horizontally and 270 degrees vertically. The scanning program is started, and the laser head emits pulsed lasers and receives reflected signals to obtain the original 3D coordinate data cloud of the surface points of each device in the workpiece processing area, such as robot A, CNC machine tool B, conveyor belt C, etc. For example, for a feature point PA_raw on the base of robot A, the data collected is The original coordinates may be (1005.3, 2010.1, 503.7) mm. At the same time, vibration and temperature changes in the environment may introduce noise, causing the original data points to have random fluctuations of ±5 mm near their true positions. Next, the system calls the Gaussian filtering algorithm to process the collected original coordinate data cloud. Specifically, for each data point, all neighboring points within a radius of 50 mm are selected. The Gaussian function is applied to calculate the weight based on the spatial distance between each neighboring point and the center point. The closer the distance, the greater the weight, and vice versa. Then, the coordinates of these neighboring points are weighted averaged to obtain the filtered coordinates of the center point, for example After Gaussian filtering of PA_raw and its neighboring points, its coordinates are updated to PA_filtered(1004.9,2009.8,503.5) mm. This process effectively reduces the environmental noise interference of ±5 mm, reducing the standard deviation of the coordinate data from the original 3.5 mm to 0.8 mm. Subsequently, in order to identify and process abnormal coordinate points, a coordinate offset threshold is set. The threshold is calculated based on the average distance between adjacent nodes. First, the pairwise Euclidean distances between all identified equipment nodes in the workstation (such as the center of mass of robot A, machine tool B, and conveyor belt C) are calculated. For example, node A(1000,2000 ,500) mm, node B (4000,2000,500) mm, node C (1000,5000,500) mm, then the distance between A and B is 3000 mm, the distance between A and C is 3000 mm, and the distance between B and C is about 4242.6 mm. Assuming that there are mainly these three nodes in the workstation, adjacent nodes are defined as node pairs with direct material flow or collaborative relationships. For example, AB and AC are adjacent, and their average distance is (3000+3000) / 2=3000 mm (the calculation is simplified here, and all adjacent pairs will be considered in actual applications). The coordinate offset threshold is set to 10% of the average distance, that is, The setting of this threshold refers to the maximum deviation range allowed for equipment layout accuracy in industry practice. Usually, an average spacing of 10% can ensure recognition accuracy while avoiding misjudging normal large-span equipment as abnormal. Experimental verification shows that when the threshold is set in the range of 8%-12% of the average spacing, the accuracy of abnormal point recognition is the highest. 10% is selected as the balance point. Finally, the system traverses the filtered coordinate points. If a coordinate point deviates from its theoretical position based on model prediction (or the interpolated position of the surrounding reliable points) by more than 300 mm, it is determined to be an abnormal coordinate point. For example, a certain acquisition point Perror(1200,2300,850) mm, and its surrounding valid points are linearly interpolated (such as using the two valid path points before and after it Pprev(1000,20 00,500) and Pnext(1100,2100,550). If Perror should be located at the midpoint of the line connecting the two, then the theoretical position is (1050,2050,525), and its deviation is much greater than 300 mm. The compensated coordinate obtained is Pcompensated(1050,2050,525) mm. The system uses this compensated coordinate to replace Perror. This process is iterated until all abnormal points are compensated. Finally, a workstation coordinate set that accurately describes the spatial position of each device node in the workstation is output. For example, this set includes robot A_final(1004.9,2009.8,503.5) mm, machine tool B_final(3995.2,2005.1,501.0) mm, etc.

[0047] Table 1 Example of workstation equipment node coordinate data

[0048]

[0049] As shown in Table 1, the data of some device nodes after coordinate collection and preliminary processing are listed, showing the correction effect of Gaussian filtering on the original coordinate data.

[0050] The movement time calculation submodule calls the workstation coordinate set generated in the previous step, for example, the coordinate data of robot A_final (1004.9, 2009.8, 503.5) mm and machine tool B_final (3995.2, 2005.1, 501.0) mm. First, the Euclidean distance between adjacent coordinate points is extracted as the theoretical shortest path length for the movement of the equipment (such as the automatic guided vehicle AGV). For the movement from A to B, the Euclidean distance is Calculated as mm,

[0051] Right now mm, which is about 2.99 meters. Next, set the time reference value, which is defined as the ratio of the track length to the average moving speed of the device. Assuming that the average straight-line moving speed of the AGV is is 0.5 m / s, then the time reference value ,This time reference value represents the time consumed in an ideal straight line path without any detour or speed change. Subsequently, the dynamic time warping algorithm (DTW) is used to align the actual or planned moving trajectory (expressed in the form of a time series, for example, the position coordinates of the AGV are recorded every 0.1 seconds) with the time series of the ideal straight line trajectory. DTW constructs a cost matrix in which each element The actual trajectory The time point and the ideal trajectory The position deviation (Euclidean distance) between the time points is calculated, and then a path with the minimum cumulative cost from the lower left corner to the upper right corner of the matrix is ​​found. This path represents the best alignment of the two time series and can handle the time axis deviation from the ideal uniform linear motion caused by acceleration, deceleration or short stops that may exist in the actual trajectory. In the process of calculating the deviation coefficient of the trajectory morphological difference from the time reference value, a specific formula is used to calculate the morphological difference, that is, the morphological difference , where the trajectory curvature change rate is It is a dimensionless parameter that quantifies the curvature of the trajectory. For example, it can be obtained by discretizing the actual trajectory into a series of small line segments, calculating the instantaneous curvature at the endpoint of each line segment (for example, approximating it with the inverse of the radius of the circumcircle of three consecutive points), and then calculating the cumulative change of these curvature values ​​along the path length, and then normalizing it. Assume that for the actual moving trajectory from A to B, After calculation (for example, the actual path is 5% longer than the straight path and contains two major turns, comprehensive evaluation) it is 0.08 (dimensionless), the time base value , then the morphological difference , this morphological difference reflects the non-ideal degree of the trajectory. Finally, based on this morphological difference and the time reference value, the final moving time parameter is obtained. This parameter is the additional time caused by the bending of the trajectory added to the basic time. The calculation formula is , substitute the numerical value, This value is used as the estimated moving time between node A and node B.

[0052] formula The parameters are described as follows: : The final calculated moving time parameter, in seconds, represents the estimated moving time of the device after taking into account non-ideal factors such as path curvature. : Time reference value, in seconds (s), defined as the ratio of the theoretical trajectory length to the average moving speed of the device, i.e. ,in is the Euclidean distance (meters), is the speed of the device (m / s). It is the moving time under ideal conditions. : Curvature change rate of trajectory, a dimensionless parameter that quantifies the degree of curvature or complexity of the actual moving trajectory relative to the ideal straight line path. The larger the value of this parameter, the more the trajectory deviates from the straight line and the more curvature it has. It is usually obtained by analyzing the geometry of the actual trajectory, for example, collecting the coordinates of a series of points on the path. , calculate curvature piecewise , and then calculate the change in curvature, such as In this example, by analyzing the historical movement data or planning path of the AGV from point A to point B, the sensor is set to record the (x, y) coordinates of the AGV every 100 milliseconds to obtain the path point sequence. For every three consecutive points , the curvature (the inverse of the radius) of the circle passing through these three points can be calculated, thus obtaining a series of curvature values. It can be the variance of these curvature values, or the average deviation from a straight path (curvature is 0), normalized. For example, if the average curvature of a path increases by 8% compared to a straight path, then Take 0.08.

[0053] Parameter assignment and calculation process: 1. Extract the coordinates of point A (1004.9, 2009.8, 503.5) mm and point B (3995.2, 2005.1, 501.0) mm from the workstation coordinate set. 2. Calculate the Euclidean distance between A and B. mm Meters. 3. AGV average moving speed Set to 0.5 m / s (obtained from the equipment manual or actual calibration, for example, multiple round trip tests on a 10-meter straight track show a stable average speed between 0.48-0.52 m / s, so 0.5 m / s is used). 4. Calculate the time base value 5. Track curvature change rate Based on actual path analysis, this value is set to 0.08 (this value is obtained by analyzing historical trajectory data and finding that the average turning radius of this path segment is approximately 5 meters, while the turning radius of a straight path is infinite. This difference is quantified using a specific model. For example, after comprehensively quantifying factors such as the average number of turns and turning angles and comparing them with a standard zero-turn path, the deviation is 8%). 6. Substitute this into the formula to calculate the movement time parameter: , , , .

[0054] The topological relationship construction submodule calculates the time difference ratio between differentiated movement paths based on the movement time parameter, uses the inverse of the path length as the weight coefficient, normalizes the row vectors of the spatial correlation matrix, and performs a Hadamard product operation on the normalized result and the time difference ratio to generate a topological relationship matrix;

[0055] The movement time calculation submodule calls the workstation coordinate set generated in the previous step, for example, the coordinate data of robot A_final (1004.9, 2009.8, 503.5) mm and machine tool B_final (3995.2, 2005.1, 501.0) mm. First, the Euclidean distance between adjacent coordinate points is extracted as the theoretical shortest path length for the movement of the equipment (such as the automatic guided vehicle AGV). For the movement from A to B, the Euclidean distance is Calculated as millimeters, that is mm, which is about 2.99 meters. Next, set the time reference value, which is defined as the ratio of the track length to the average moving speed of the device. Assuming that the average straight-line moving speed of the AGV is is 0.5 m / s, then the time reference value ,This time reference value represents the time consumed in an ideal straight line path without any detour or speed change. Subsequently, the dynamic time warping algorithm (DTW) is used to align the actual or planned moving trajectory (expressed in the form of a time series, for example, the position coordinates of the AGV are recorded every 0.1 seconds) with the time series of the ideal straight line trajectory. DTW constructs a cost matrix in which each element The actual trajectory The time point and the ideal trajectory The position deviation (Euclidean distance) between the time points is calculated, and then a path with the minimum cumulative cost from the lower left corner to the upper right corner of the matrix is ​​found. This path represents the best alignment of the two time series and can handle the time axis deviation from the ideal uniform linear motion caused by acceleration, deceleration or short stops that may exist in the actual trajectory. In the process of calculating the deviation coefficient of the trajectory morphological difference from the time reference value, a specific formula is used to calculate the morphological difference, that is, the morphological difference , where the trajectory curvature change rate is It is a dimensionless parameter that quantifies the curvature of the trajectory. For example, it can be obtained by discretizing the actual trajectory into a series of small line segments, calculating the instantaneous curvature at the endpoint of each line segment (for example, approximating it with the inverse of the radius of the circumcircle of three consecutive points), and then calculating the cumulative change of these curvature values ​​along the path length, and then normalizing it. Assume that for the actual moving trajectory from A to B, After calculation (for example, the actual path is 5% longer than the straight path and contains two major turns, comprehensive evaluation) it is 0.08 (dimensionless), the time base value , then the morphological difference , this morphological difference reflects the non-ideal degree of the trajectory. Finally, based on this morphological difference and the time reference value, the final moving time parameter is obtained. This parameter is the additional time caused by the bending of the trajectory added to the basic time. The calculation formula is , substitute the numerical value, This value is used as the estimated moving time between node A and node B.

[0056] formula The parameters are described as follows: : The final calculated moving time parameter, in seconds, represents the estimated moving time of the device after taking into account non-ideal factors such as path curvature. : Time reference value, in seconds (s), defined as the ratio of the theoretical trajectory length to the average moving speed of the device, i.e. ,in is the Euclidean distance (meters), is the speed of the device (m / s). It is the moving time under ideal conditions. : Curvature change rate of trajectory, a dimensionless parameter that quantifies the degree of curvature or complexity of the actual moving trajectory relative to the ideal straight line path. The larger the value of this parameter, the more the trajectory deviates from the straight line and the more curvature it has. It is usually obtained by analyzing the geometry of the actual trajectory, for example, collecting the coordinates of a series of points on the path. , calculate curvature piecewise , and then calculate the change in curvature, such as In this example, by analyzing the historical movement data or planning path of the AGV from point A to point B, the sensor is set to record the (x, y) coordinates of the AGV every 100 milliseconds to obtain the path point sequence. For every three consecutive points , the curvature (the inverse of the radius) of the circle passing through these three points can be calculated, thus obtaining a series of curvature values. It can be the variance of these curvature values, or the average deviation from a straight path (curvature is 0), normalized. For example, if the average curvature of a path increases by 8% compared to a straight path, then Take 0.08.

[0057] Parameter assignment and calculation process: 1. Extract the coordinates of point A (1004.9, 2009.8, 503.5) mm and point B (3995.2, 2005.1, 501.0) mm from the workstation coordinate set. 2. Calculate the Euclidean distance between A and B. mm Meters. 3. AGV average moving speed Set to 0.5 m / s (obtained from the equipment manual or actual calibration, for example, multiple round trip tests on a 10-meter straight track show a stable average speed between 0.48-0.52 m / s, so 0.5 m / s is used). 4. Calculate the time base value 5. Track curvature change rate Based on actual path analysis, this value is set to 0.08 (this value is obtained by analyzing historical trajectory data and finding that the average turning radius of this path segment is approximately 5 meters, while the turning radius of a straight path is infinite. This difference is quantified using a specific model. For example, after comprehensively quantifying factors such as the average number of turns and turning angles and comparing them with a standard zero-turn path, the deviation is 8%). 6. Substitute this into the formula to calculate the movement time parameter: , , , ;

[0058] Row vector normalization is achieved by calculating the Euclidean length of each row element and performing a division operation.

[0059] The demand clustering module includes: the demand feature extraction submodule calls the demand time series data recorded in the production line manufacturing execution system (MES) about the cutting material (for example, steel plate SPCC, thickness 2mm) at station A. This data records in detail the timestamp and quantity of each demand initiation in the past few production cycles. First, the system extracts the time interval between two adjacent demand events from these time series records to form a time interval sequence. For example, in the most recent shift (8 hours), the timestamps of the steel plate demanded by station A are 08:05, 08:32, 09:10, 09:40, and 10:25, and the corresponding time intervals are 27 minutes, 38 minutes, 30 minutes, and 45 minutes, forming a sequence Minutes, next, calculate the standard deviation of this time interval series as the "time fluctuation" to measure the uncertainty of the demand time point , for the sequence , its mean Minutes, standard deviation Then, the system collects the specific usage data of the material at each time it is required at station A during the same time period to form a usage sequence. For example, the corresponding four demand amounts are 10, 12, 8, and 15 blocks respectively, that is, The block then calculates the coefficient of variation (CV) of this usage series as a measure of the "usage dispersion" of the demand uncertainty. , the coefficient of variation is defined as the standard deviation divided by the mean, for the series , its mean Block, its standard deviation Block, therefore, the amount of discrete (dimensionless), and finally, the calculated “time fluctuation amount” Minutes and "Discrete Usage" Perform Z-score standardization. The standardization calculation basis is derived from the daily calculations within the last 30 production cycles (for example, the last 30 days). and The mean and standard deviation of the values ​​are obtained by statistics. Assuming that the mean of the time fluctuation is obtained by statistics of the data of the past 30 cycles Minutes, standard deviation Minutes; mean of discrete amounts of usage , standard deviation The setting of these benchmark values ​​is based on long-term tracking and statistical analysis of historical data to ensure their representativeness and stability. The selection of 30 production cycles takes into account seasonal fluctuations in production and monthly plan adjustments, which can better reflect the recent normal level. The standardized time fluctuation of the current cycle is , standardized discrete quantity , these two standardized values ​​constitute a two-dimensional demand feature vector , this vector concisely characterizes the volatility and discrete characteristics of current material demand.

[0060] Table 2 Example data for demand characteristics calculation

[0061]

[0062] As shown in Table 2, the calculation process and intermediate data of the material demand characteristics of station A are displayed, and the standardized demand feature vector is finally obtained.

[0063] The pattern similarity calculation submodule is based on a specific demand generated in the previous step (for example, the steel plate demand of station A, whose feature vector is ) or its original demand sequence, and call other historical or different types of demand patterns (such as the aluminum demand sequence of station B, or the high-fluctuation demand pattern sequence of station A in history) for comparison. First, the dynamic time warping (DTW) algorithm is used to process the two differentiated demand sequences that need to be compared, for example, the current daily demand sequence of steel plates at station A. (in is the demand for the t-th time unit, such as hourly demand), and a historical typical demand pattern sequence ,DTW algorithm stretches or compresses the time axis of the two sequences nonlinearly to find the best alignment path between them. When it is implemented, it constructs a The distance matrix, where the elements The value of (usually p=1 or 2), and then use dynamic programming to find a way from arrive The path with the smallest sum of elements on this path is the minimum curved path. Next, calculate the cumulative distance value on this minimum curved path, which is recorded as , this value quantifies the degree of difference in the morphology of the two demand sequences, even if they are misaligned in time or have different rates. Assume that the cumulative distance value between the current demand sequence of station A and the historical high fluctuation pattern P1 is obtained by DTW calculation (units are related to the demand units, or have been normalized to dimensionless), and then the calculated cumulative distance value is Mapping is performed through a preset similarity conversion function to obtain the final pattern similarity coefficient , the conversion function is set as an S-shaped curve function, the specific form can be ,in is the input DTW distance value, is the slope parameter, is the distance offset (or center point), which determines the conversion center position of the S-shaped curve. According to the description, the slope parameter Material allocation response time threshold Inversely proportional, that is ,in is a positive proportional constant, the material allocation response time threshold It refers to the maximum time allowed from the generation of demand to the delivery of materials. For example, it is set to 30 minutes. The setting of this threshold is based on the comprehensive consideration of production rhythm and material waiting cost. After experimental analysis, when the response time exceeds 30 minutes, the average waiting time loss of the workstation increases significantly. Therefore, it is set to 30 minutes. The proportional constant The value of can be obtained by calibration of historical data, so that the similarity coefficient can effectively distinguish different patterns. For example, ,but (If the unit of D is minute-related), if D is the dimensionless DTW distance, then the unit of k is also adjusted accordingly, or D is first normalized, assuming (dimensionless, obtained by calibration so that the DTW distance in the range of 20-80 can produce good similarity distinction), (the center point of the S-shaped curve, representing the DTW distance of the general difference) is set to 50. This value is selected based on the average of the DTW distances of a large number of historical demand sequence pairs. For example, if the average DTW distance is 52, then It can be taken as 50, then for , pattern similarity coefficient , this 0.6035 is the similarity coefficient between the current demand pattern of workstation A and the historical high fluctuation pattern P1.

[0064] The matrix operator module calls the pattern similarity coefficient calculated in the previous step (for example, the similarity coefficient vectors calculated for the demand pattern of the current workstation X and the K historical typical demand patterns P1, P2, ..., PK respectively). Or, if the similarity of demand patterns between workstations is calculated, a pattern similarity matrix between workstations is obtained , where the elements is the similarity coefficient of the demand pattern between workstation i and workstation j) and the topological relationship matrix generated in the topological modeling module (Its elements Reflects the topological connection strength between station i and station j). First, the system needs to build a "coefficient matrix" and topological matrix The "coefficient matrix" here is closely related to the "pattern similarity coefficient". Assuming that we are considering the characteristics of the demand pattern of each workstation relative to a global benchmark, or the demand synergy between workstations, if the "pattern similarity coefficient" is a single value (such as the similarity between the overall demand pattern of the current production batch and a certain standard pattern), then this coefficient may be used to scale the entire topological relationship matrix or some of its elements, but more commonly, we need a similar coefficient to "Coefficient matrix" of the same dimension ,For example, It can represent the synchronization or complementarity score of the demand patterns of workstations i and j, which is determined by their respective "pattern similarity coefficients" (similarity relative to a certain benchmark pattern) and the interaction between them. In the description of "constructing the Cartesian product space of the coefficient matrix and the topological matrix, and performing the product operation of the corresponding positions of the matrix elements", "Cartesian product space" may refer to a conceptual combination space, while the actual operation of "performing the product operation of the corresponding positions of the matrix elements" clearly refers to the Hadamard product, provided that and Have the same dimensions (N is the number of workstations), suppose is the interaction coefficient calculated based on the pattern similarity coefficients of workstations i and j. For example, if the demand pattern of workstation i is highly similar to the high-volatility pattern, while the demand pattern of workstation j is highly similar to the stable pattern, their interaction coefficients may be low. On the contrary, if both are similar to the high-cooperation demand pattern, the interaction coefficient is high. Let's take a specific example and assume that Directly the similarity of demand patterns between workstation i and workstation j ,and is the topological connection strength between them, then the corresponding position elements are multiplied to obtain the product matrix , whose elements For example, if the demand pattern similarity between workstation 1 and workstation 2 is , the topological relationship value between them (This value comes from the calculation result of paragraph 3, for example After some scaling or selection of representative values), then , this product More comprehensively reflects the comprehensive correlation between workstations i and j in terms of demand patterns and physical topology. Regarding "the Cartesian product space improves material distribution efficiency by increasing the weight factor of the time consumption of movement between workstations", this usually refers to It already contains the weight factor related to the moving time (as described in paragraph 3, its construction process takes the moving time into consideration), so Naturally, this is also inherited. Then, this product matrix The result of "summing by row vectors" means that Each line , calculate the sum of all its elements , which will be for each station Generates a scalar value that represents the overall strength of the relationship between the workstation and all other workstations at the demand-topology level, for example, if The first behavior , then the row sum of station 1 is ,Finally, according to the description of "generating demand-topology association matrix", if this "demand-topology association matrix" is a matrix, then it may be itself, while the result of "summing by row vectors" is based on If further analysis or aggregation steps are refers to ,but , which directly quantifies the coupling effect of demand pattern similarity and topological connectivity between each pair of workstations.

[0065] The Z-score standardization process is calculated based on the data of the latest 30 production cycles;

[0066] The similarity conversion function is an S-shaped curve function, whose slope parameter is inversely proportional to the material allocation response time threshold;

[0067] Cartesian product space improves material distribution efficiency by increasing the weight factor of the time consumption of movement between workstations.

[0068] The strategy generation module includes: the response capability modeling submodule collects the historical equipment overall efficiency (OEE) data of the designated workstation (for example, welding station W1). This data is usually automatically recorded by the MES system or equipment monitoring system and includes the equipment's operating performance over the past period of time. First, key parameters are extracted from these OEE data records, including the actual operating time of the equipment within the planned operating time. (For example, in an 8-hour or 480-minute shift, the actual running time of the equipment is 400 minutes due to waiting for materials, changing models, etc.), the theoretical cycle time of a unit product is (For example, the theoretical fastest time to weld a standard part is 0.8 minutes per part, which is provided by the process designer or equipment manufacturer), and the number of qualified parts produced during the actual running time (For example, 450 qualified products were produced in 400 minutes). Next, calculate an "overall equipment efficiency benchmark value" , which is defined as the product of "run time utilization" and "performance", where "run time utilization" Defined as actual running time Total planned production time (e.g. shift time), i.e. , and "performance" Here it is defined as (the number of qualified products Theoretical cycle time) / actual running time, i.e. Therefore, the benchmark value of equipment overall efficiency is This benchmark value is a snapshot of efficiency performance for a specific period and a specific material handling volume. Subsequently, in order to more comprehensively describe the relationship between efficiency and output (material handling volume), the system uses a cubic polynomial regression method to collect multiple sets of historical data points (each set of data contains a period of "material handling volume"). and the corresponding “comprehensive efficiency” , the overall efficiency here is This is calculated according to the above method ) is fitted, the goal is to establish a The mathematical model of Represents the material throughput (for example, the number of workpieces processed per hour), Represents the overall efficiency (a value between 0 and 1), the coefficient It is estimated by the least square method. For example, by analyzing the data of the past 20 shifts, we can get 20 The fitted equation may be ,in The unit is piece / hour, is the efficiency value, and the goodness of fit is evaluated by the coefficient of determination R2, and the threshold of R2 is set to 0.85, which requires The setting of this threshold is based on industry experience and multiple simulation experiments. The experiments show that when R2 is lower than 0.85, the prediction error of the model is large and it cannot accurately guide production. When it is higher than 0.85, the model can better reflect the actual trend and ensure the reliability of the model. For example, the R2 obtained in this fitting is 0.89, which meets the requirements. Finally, this cubic polynomial equation This is the generated response capability curve equation for this workstation.

[0069] formula (Simplified to ) are described as follows: : The benchmark value of equipment overall efficiency, dimensionless, usually between 0 and 1, represents the overall performance of the equipment under specific conditions. : Actual startup time, in minutes or hours, refers to the time the equipment is actually used for production within the planned production time. : Total planned production time, unit and The same, for example, the total duration of a shift. : The number of qualified products, in pieces or other counting units, refers to the number of qualified products in The number of products that meet quality requirements produced within a certain period of time. : The theoretical cycle time of a unit product, expressed in minutes / piece or hours / piece, refers to the shortest time required to produce a qualified product under ideal conditions.

[0070] Parameter assignment and calculation process: 1. Collect data from the MES system: planned total production time Minutes. 2. Actual startup and running time Minutes. 3. Qualified product quantity 4. Theoretical cycle time of unit product Minutes / piece (This data is provided by the equipment supplier and verified through testing under ideal conditions in small batches. For example, for continuous production of 10 pieces, the shortest total time is 7.9 minutes, with an average of 0.79 minutes per piece, rounded to 0.8 minutes). 5. Substitute this into the formula to calculate the baseline value of the equipment's overall efficiency: , , , .

[0071] The constraint activation submodule calls the response capability curve equation generated in the previous step. For example, the response capability curve of welding station W1 is ,in is the overall efficiency, is the material handling capacity (pieces / hour). First, the system calculates the second-order derivative function of the response capability curve equation and solves the point where the second-order derivative is zero (i.e., the inflection point of the curve). This usually corresponds to the turning point where the efficiency growth rate changes significantly. Mathematically, , then the second-order derivative ,make ,Right now , the solution is Pieces / hour, the calculation result is rounded to two decimal places as required, that is Pieces / hour, this The value is interpreted in the current context as a critical point in the station's ability to process material types or a boundary point in the optimal efficiency range (although the direct association between the inflection point and the optimal efficiency point usually requires more context, here it is carried out according to the logic described in the patent). Next, the system monitors the number of material types currently being processed by the station W1 in real time. For example, by reading the production order or bill of materials information associated with the workstation, it is known that the current W1 workstation needs to process three parts with different specifications but similar processes, namely Then, the system will monitor the number of material types in real time. The horizontal axis value corresponding to the zero point of the second derivative calculated previously (The patent description here will be Material handling volume compared to the number of material types may mean It does not represent the inflection point of the processing rate, but a certain quantity related to the processing capacity, or the "number of material types" here and the "material processing volume" in the response capacity curve have a certain correspondence or conversion relationship in a specific scenario. Assume that is a threshold value related to the “carrying capacity of material types”), and when the real-time material type number is Reach or exceed this horizontal axis value When , the system triggers a capacity constraint indicator, indicating that the workstation has reached or is close to a critical point in its design or experience in terms of handling material diversity, which may lead to a decrease in efficiency or an increase in management complexity. In this case, ,and ,because , so the capacity constraint indicator is not triggered. If we assume Refers to a threshold related to the upper limit of the number of material types that a workstation can efficiently handle, such as After calibration, it represents an "equivalent maximum material type processing index", and the current is also converted to a comparable index, or simply, if Directly refers to the maximum number of material types allowed, for example , and currently , then because , the capacity constraint flag of the W1 workstation is set to "activated" or "True". Finally, the system summarizes the constraint flag status of all workstations and generates a workstation constraint flag set. For example, for a workstation containing three workstations W1, W2, and W3, its constraint flag set may be {W1: True, W2: False, W3: True}, indicating that the W1 and W3 workstations have triggered constraints in terms of material type processing.

[0072] The strategy generation submodule is based on the station constraint identification set obtained in the previous step (for example, {W1: True, W2: False, W3: True}, where the material types of W1 and W3 have reached the constraints), as well as the specific operating parameters of each station. It first calculates the ratio matrix between the "station remaining capacity" and the "demand feature vector". Here, the "station remaining capacity" For workstations For example, if workstation W2 does not trigger any constraints, its theoretical maximum capacity is 100 units / hour, and the assigned tasks occupy 60 units / hour, then the remaining capacity is Unit / hour, and for the workstation W1 that has triggered the constraint, its remaining capacity calculation may need to consider the efficiency loss caused by the constraint or directly set it to a lower value, such as Unit / hour, "demand characteristic vector" From the demand clustering module (as in paragraph 4) ), which describes the demand to be allocated Characteristics of demand, such as volatility and discreteness of demand, suppose we have 3 workstations W1, W2, W3 and 2 pending demands D1, D2, and their demand feature vectors are and , we need to convert the demand feature vector into a "demand quantity" or "load index" that can be compared with the capacity. For example, we can take its modulus or weighted sum. Let For demand The equivalent load, then the ratio matrix Elements (or some form of matching score, such as ), for example, if unit, Unit, then , , this ratio matrix reflects the relative sufficiency or suitability of each workstation to handle various demands. Next, the system constructs a linear programming model, whose objective function is set to maximize the sum of the products of these ratios (or matching scores) and the decision variable (indicating whether the demand is assigned to the workstation), that is, ,in is a binary decision variable. If the demand Assigned to workstation ,but , otherwise it is 0. Then, set the constraints of the model. These constraints are given in the form of inequality equations, mainly including "upper limit constraint of material type" and "non-negative constraint of workstation capacity". The upper limit constraint of material type refers to the amount of material allocated to each workstation. The total number of different material types (or demand types) cannot exceed the upper limit of the type preset for the station or determined by the constraint activation submodule ,Right now (here Determine whether requirement j is a new type added to workstation i), or if represents the distribution amount, then ,in is an indicator function, and the non-negative constraint on the station capacity ensures that the total load assigned to each station is Does not exceed its remaining effective capacity , and all allocations (If it represents a specific quantity) must be a non-negative number. Finally, the simplex method is used to solve the constructed linear programming model. The simplex method finds the optimal solution by iteratively moving on the vertices of the feasible solution domain. Each step is made in the direction of improving the objective function value until a solution that satisfies all constraints and maximizes the objective function is found. Combination, for example, after solving may get , indicating that allocating both demands D1 and D2 to workstation W2 is the current optimal (or feasible) strategy. Finally, all calculated feasible solutions (which may include the optimal solution and some suboptimal solutions, or the Pareto solution set under multiple objectives) are sorted and output to form a preset strategy set, for example, strategy A: {D1->W2,D2->W2}, strategy B: {D1->W2,D2->W3} (if W3 also has a certain capacity).

[0073] The decision optimization module includes: the collaborative unit construction submodule calls the topological relationship matrix generated by the previous step (For example, a Matrix, element It represents the topological association strength from station i to station j. The larger the value, the stronger the association. Assume that the matrix is ), first calculate the mean of all off-diagonal elements in the matrix and standard deviation , for the above , the off-diagonal elements are 0.8, 0.3, 0.8, 0.6, 0.3, 0.6, and their mean , standard deviation ; Next, set a "correlation coefficient critical value" , which is defined as the triple weighted sum of the mean and the standard deviation, i.e. , where the weight coefficient , this coefficient The selection is based on the analysis of historical data. It is found that when the collaboration efficiency between workstations reaches its peak, the corresponding topological correlation values ​​are mostly distributed in For example, collect tasks involving two workstations working together in the past year, calculate their completion efficiency, and compare them with the efficiency between the two workstations. The correlation analysis of the values ​​showed that When the collaborative efficiency is improved by more than 15%, the probability of significantly increasing As a conservative and effective threshold, ,because The maximum value is 0.8. The value is too high, indicating that the current sample data may not have a strong correlation. In order to continue the example, we adjust the hypothesis, for example, assuming that the ,but , then, based on this critical value, a workstation node graph is constructed (not a specific bipartite graph, but a general graph where nodes represent workstations and edges represent the strength of association between workstations). If , then an edge is formed between workstations i and j, and the weight of the edge can be itself, for example, and , there is an edge with a weight of 0.8 between workstation 1 and workstation 2, and other direct connections do not meet this condition (0.3 and 0.6 are both less than 0.7), so a graph containing only 1-2 connections is obtained. Then, the system uses the Hungarian algorithm to "traverse the adjacency matrix to capture the maximum weight matching". The Hungarian algorithm is usually used to solve the maximum weight matching problem of weighted bipartite graphs. If the goal here is to form paired collaborative units (for example, to pair N workstations into N / 2 pairs), the workstation set can be copied to construct a bipartite graph, in which one part of the nodes is the original workstation and the other part is also the original workstation, and the edge The weight is (like and ), and then use the Hungarian algorithm to find the matching that maximizes the total weight. Alternatively, if the "collaborative unit" is not limited to pairing, but a small-scale collaborative cluster, other community discovery algorithms may be used. If we strictly follow the "Hungarian algorithm" and "maximum weight matching", we assume that the purpose here is to find the optimal workstation pairing combination. Even if the number of workstations is odd, it can be handled by introducing virtual nodes. For example, for workstations 1, 2, and 3, we want to form an optimal pairing and a single workstation. Suppose we construct a weighted adjacency matrix ,in if else0, and apply the Hungarian algorithm (or its variants such as the auction algorithm for matching non-bipartite graphs) to find several disjoint edge sets that maximize the sum of the weights of the matching edges. In this case, only the 1-2 connection meets the condition, with a weight of 0.8, then a collaborative unit is identified as {W1, W2}, and W3 is alone, and finally the collaborative unit set is output, for example: {{W1, W2}, {W3}}.

[0074] formula The parameters are described as follows: : Correlation coefficient critical value, which is a threshold used to judge whether the element value in the topological relationship matrix is ​​high enough to indicate that there is significant synergy potential between the two workstations. : Topological relationship matrix The mean of the (usually off-diagonal) elements in . : Topological relationship matrix The standard deviation of the (usually off-diagonal) elements in . : Weight coefficient, here it is set to 3. It determines the magnification of the standard deviation when calculating the critical value. It usually corresponds to the "three sigma" criterion in statistics and can cover most data.

[0075] Parameter assignment and calculation process (using adjusted hypothetical data to make the process meaningful): 1. Topological relationship matrix 2. Calculate the mean of the off-diagonal elements 3. Calculate the standard deviation of the off-diagonal elements 4. Set the weight coefficient The selection of this coefficient is based on historical data showing that when the topological correlation coefficient between workstations exceeds When the average efficiency of the collaborative unit they form is 20%-25% higher than that of the unit below this threshold when performing complex tasks, and the failure rate is reduced by 10%. For example, 100 sets of historical collaborative task data were selected, corresponding to the Value, found in The top 5% of the value (roughly corresponding to The synergy success rate and efficiency index are significantly improved when the pairing is performed within the range above 5. 5. Substitute the formula to calculate the critical value of the correlation coefficient: , , , , (As mentioned before, this result is too high for the example matrix. To continue, assume Adjusted to 0.7).

[0076] The strategy matching submodule is based on the collaborative unit set generated in the previous step, such as {{W1, W2}, {W3}}, and performs matching evaluation on each unit (or device within the unit) and each strategy in the "preset strategy set". First, the key "device processing speed" and "capacity parameters" are extracted from the collaborative unit or individual device to form the device feature vector For example, for the collaborative unit {W1, W2}, its comprehensive processing speed may be the sum of the two or some weighted average, such as 15 pieces / minute, and the comprehensive capacity is 80 pieces (batch capacity or cache capacity), then its feature vector , for a single workstation W3, its eigenvector may be Next, the system analyzes each strategy in the "preset strategy set". A strategy usually involves allocating a certain material (or task) to a certain device or collaborative unit. It is necessary to extract a "material type feature" vector for the material (or task). , this vector should reflect the material's requirements for the equipment. For example, a certain material M1 requires a high processing speed and medium batch processing capacity, and its characteristics can be quantified as , then, calculate the material feature vector and device feature vector The cosine angle value between (i.e. cosine similarity), , for example, for material M1( ) and collaborative unit {W1,W2}( ): , , , , this value is very close to 1, indicating that the directions of the two feature vectors are highly consistent. Then, a matching threshold is set. , the threshold is described as the "inverse of the eigenvector modulus ratio". According to "the calculation formula of the eigenvector modulus ratio is the square root of the sum of the device processing speed and capacity parameters", this refers to the calculation of a single modulus, that is, , for the “inverse of the modulus ratio”, we interpret it as To ensure that the threshold is in the interval [0,1], it can be compared with the cosine similarity, for example, , ,but The logic of setting this threshold is that when the modulus lengths (representing the overall "scale" or "capacity") of the two vectors differ too much, even if the directions are consistent, the matching degree should be affected. The closer the modulus length ratio is to 1, the higher the threshold, requiring more consistent directions. Conversely, when the modulus length difference is large, the threshold is lowered, and the requirement for directional consistency is slightly relaxed. Finally, filter those "strategy items with angle values ​​lower than the threshold" in the preset strategies. If the "angle value" refers to itself, while "threshold" refers to an angle threshold (For example, by Inverse cosine is obtained), then filter strategy, which is equivalent to , if the "angle value" refers to , and "below the threshold" is literal, i.e. screening This strategy will select items with poor matching, which may make sense in some scenarios (such as finding differentiated supplementary strategies), but usually the optimization is to find items with high matching. Here, we follow the literal "angle value ( ) is below the threshold ( )", then for the above example, , ,because , so this policy item will not be filtered out (if you follow this literal logic), if you understand it as filtering If the quality match , the strategy will be selected, considering the goal of "optimizing the strategy candidate set", the latter understanding is more common, but according to the title strictly, assuming that the screening , then, only strategies whose cosine similarity is not high enough (below the threshold determined by the modulus ratio) will be retained. For example, if another strategy calculates And its corresponding , then because , the strategy item will be selected into the optimization strategy candidate set, which contains those strategies that have been initially screened according to the current matching criteria (which may be looking for specific differences or satisfying specific mismatch conditions).

[0077] The dynamic allocation submodule calls the optimization strategy candidate set generated by the previous step of screening (for example, including strategies S1', S2', S3', which may be selected because the cosine similarity of their equipment and material characteristics is "lower" than a dynamic threshold determined by the inverse of the module length ratio, suggesting that they may represent some options that require special attention, are not highly matched but still valuable, or, if the screening logic is to take "higher" than the threshold, they are high-quality matches). First, the system detects the real-time load of the current workstation (or collaborative unit). The maximum capacity determined in the previous collaborative unit construction phase (For example, the maximum collaborative capacity of collaborative units {W1, W2} is 100 units / hour) , this difference reflects the current idle capacity or overload situation of the collaborative unit. For example, if the current load of unit {W1, W2} is 90 units / hour, then Units / hour (with idle time), if the load is 110, then (Overload), Next, the system uses the gradient descent method to adjust the weight coefficients of each strategy in the optimization strategy candidate set (Each strategy have an initial weight, such as a uniform distribution or an initial value based on a previous evaluation, such as ), the goal of gradient descent is to optimize a performance indicator function , this function may be related to improving overall output, reducing waiting time or balancing the load of each unit, and will be affected by The impact of (for example, giving priority to reducing the policy weight of overloaded units, or increasing the policy weight of idle units), specifically, calculating the performance index About the weight of each strategy The partial derivative of , and then update the weights along the negative gradient direction: ,in is the learning rate (e.g., , this value is adjusted through experiments and tested in the range of 0.01 to 0.1. 0.05 can converge well within 100 iterations and is not easy to oscillate). The iterative process of gradient descent will continue until the termination condition is met. The termination condition is set as follows: in three consecutive iterations, the weight coefficients of all strategies The change (absolute value) is less than 0.01, that is, for any strategy , , ,and The threshold of 0.01 is set based on the sensitivity analysis of the impact of weight changes on the final scheduling instructions. When the change is less than this value, it is considered that the actual benefit improvement brought by further iteration is no longer obvious. For example, after several iterations, the weight update is , and if the change in each weight is less than 0.01 in the subsequent three iterations, the iteration stops. Finally, these strategies optimized and updated by the gradient descent method (now with a new weight distribution, some strategies may have significantly increased weights and become dominant strategies) and their related execution parameters (such as which workstation to assign to, specific quantity, priority, etc.) are encoded as a data packet in JSON (JavaScript Object Notation) format. JSON format is chosen for its lightweight and easy machine parsing and human-readable features. For example, a deployment instruction may be encoded as: {"command_id":"CMD001","timestamp":"2025-05-16T14:35:00Z","target_unit": "Unit_W1_W2","strategy_selected":"S1'_optimal","tasks":[{"material_id":"M005","quantity":50,"priority":1},{"material_id":"M008","quantity":30,"priority":2}],"estimated_completion_time":"2025-05-16T15:05:00Z"}. This JSON data packet is the final generated dynamic allocation instruction, which will be sent to the execution system (such as the AGV scheduling system and workstation control system) to guide actual material allocation and production activities.

[0078] The coefficient of the triple weighted sum is selected based on the correlation distribution corresponding to the peak of the workstation collaboration efficiency in the historical data;

[0079] The formula for calculating the eigenvector modulus ratio is the square root of the sum of the device processing speed and capacity parameters;

[0080] The iterative termination condition of the gradient descent method is that the change in the weight coefficient is less than 0.01 in three consecutive iterations.

[0081] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A material dynamic allocation system based on data preset learning for multiple workstations, characterized in that: The system comprises: A topological modeling module is used to obtain a set of workstation layout coordinates using a laser rangefinder, input the coordinate set into a dynamic time warping algorithm to calculate movement time parameters, adjust the spatial correlation coefficient based on the product of the difference ratio of the time parameters and the equipment movement acceleration, generate a topological relationship matrix, and pass it to the demand clustering module; The demand clustering module is used to call the material demand time series records of the production line MES system, input the standard deviation of the demand interval and the coefficient of variation of the single usage into the dynamic time warping algorithm to calculate the pattern similarity, generate the demand feature vector, perform the Cartesian product operation on the topological relationship matrix, and then pass it to the strategy generation module; The strategy generation module is used to establish a response capability curve equation based on the equipment OEE data, activate the constraint conditions when the number of material types allocated to the workstation reaches the inflection point of the curve equation, input the demand feature vector and the ratio of the workstation's remaining capacity into the linear programming model to generate a preset strategy set, and pass it to the decision optimization module.

2. The material dynamic allocation system based on data preset learning for multiple workstations according to claim 1 is characterized in that: The topological relationship matrix includes the spatial correlation coefficient, the workstation layout structure, and the coordinate space position relationship. The demand feature vector includes the demand pattern similarity, the demand interval regularity, and the material usage characteristics. The preset strategy set specifically includes the material configuration strategy, the capacity matching plan, and the workstation load balancing plan.

3. The material dynamic allocation system based on data preset learning for multiple workstations according to claim 2 is characterized in that: The dynamic time warping algorithm achieves dynamic correction of the spatial correlation coefficient by matching the morphological difference of the movement trajectory with the deviation coefficient of the time reference value; The dynamic time warping algorithm uses the constraint that the window width is twice the standard deviation of the required interval to align the sequences; The response capability curve equation is fitted by cubic polynomial regression to the corresponding relationship between the equipment comprehensive efficiency and the material handling capacity, wherein the regression fitting goodness threshold is set to 0.

85.

4. The material dynamic allocation system based on data preset learning for multiple workstations according to claim 3 is characterized in that: The topology modeling module includes: a coordinate acquisition submodule that scans the workstation layout using a laser rangefinder to collect three-dimensional coordinate data of multiple device nodes, uses a Gaussian filter algorithm to eliminate environmental noise interference, sets a coordinate offset threshold of 10% of the average spacing between adjacent nodes as a reference value, performs interpolation compensation on abnormal coordinate points, and generates a workstation coordinate set; The movement time calculation submodule calls the workstation coordinate set, extracts the Euclidean distance between adjacent coordinate points, uses the ratio of the equipment movement speed to the trajectory length as the time reference value, aligns the time series of the movement trajectory through the dynamic time warping algorithm, and calculates the deviation coefficient of the trajectory morphology difference from the time reference value using the formula morphology difference = trajectory curvature change rate × time reference value squared to obtain the movement time parameter; The topological relationship construction submodule calculates the time difference ratio between the differentiated movement paths based on the movement time parameters, uses the inverse of the path length as the weight coefficient, normalizes the row vectors of the spatial correlation matrix, and performs a Hadamard product operation on the normalized result and the time difference ratio to generate a topological relationship matrix; The row vector normalization process is achieved by calculating the Euclidean length of each row element and performing a division operation.

5. The material dynamic allocation system based on data preset learning for multiple workstations according to claim 4 is characterized in that: The demand clustering module includes: a demand feature extraction submodule that calls the material demand time series records of the production line MES system, extracts the time interval sequence of adjacent demand events, calculates the sequence standard deviation as the time fluctuation amount, collects single material usage data, calculates the coefficient of variation as the usage discrete amount, performs Z-score normalization processing on the time fluctuation amount and the usage discrete amount, and generates a demand feature vector; The pattern similarity calculation submodule uses a dynamic time warping algorithm to stretch and compress the time axis of the differentiated demand sequence based on the demand feature vector, calculates the cumulative distance value of the minimum bending path between sequences, maps the distance value to a preset similarity conversion function, and obtains a pattern similarity coefficient; The matrix operation submodule calls the pattern similarity coefficient and the topological relationship matrix, constructs the Cartesian product space of the coefficient matrix and the topological matrix, performs the product operation of the corresponding positions of the matrix elements, sums the product results according to the row vector, and generates the demand-topology association matrix; The Z-score standardization process is calculated based on the data of the latest 30 production cycles; The similarity conversion function is an S-shaped curve function, whose slope parameter is inversely proportional to the material allocation response time threshold; The Cartesian product space improves the material distribution efficiency by increasing the weight factor of the movement time between workstations.

6. The material dynamic allocation system based on data preset learning for multiple workstations according to claim 5 is characterized in that: The strategy generation module includes: a response capability modeling submodule that collects equipment OEE data, extracts equipment operating time, theoretical cycle time, and number of qualified products, calculates the equipment comprehensive efficiency benchmark value as the product of operating time utilization and performance, uses cubic polynomial regression to fit the corresponding relationship between comprehensive efficiency and material processing volume, and generates a response capability curve equation; The constraint condition activation submodule calls the response capacity curve equation, calculates the zero point coordinate of the second-order derivative function, monitors the real-time value of the number of material types currently in the workstation, and triggers the capacity constraint flag when the value reaches the horizontal axis value corresponding to the zero point coordinate, thereby generating a workstation constraint flag set; The strategy generation submodule calculates the ratio matrix of the remaining capacity of the station and the demand characteristic vector based on the station constraint identification set, constructs a linear programming model whose objective function is to maximize the sum of the products of the matrix elements, sets the material type constraints as a system of inequality equations, uses the simplex method to solve the model feasible solution, and generates a preset strategy set; The goodness-of-fit threshold for the cubic polynomial regression was set at 0.85; The zero point coordinate calculation accuracy is retained to two decimal places; The inequality equation group includes an upper limit constraint on material types and a non-negative constraint on station capacity.

7. The material dynamic allocation system based on data preset learning for multiple workstations according to claim 6 is characterized in that: The system further comprises: The decision optimization module is used to construct the workstation groups with correlation coefficients higher than the critical value in the topological relationship matrix into collaborative units through the Hungarian algorithm, perform cosine similarity matching between the material types in the preset strategy set and the equipment parameters of the collaborative units, trigger the strategy reconstruction mechanism to generate dynamic deployment instructions and synchronously update them to the topological modeling module.

8. The material dynamic allocation system based on data preset learning for multiple workstations according to claim 7 is characterized in that: The dynamic allocation instructions specifically include material distribution paths, workstation collaboration methods, and material-equipment combination plans.

9. The material dynamic allocation system based on data preset learning for multiple workstations according to claim 8, characterized in that: The critical value is the weighted sum of three times the mean and standard deviation of the topological relationship matrix elements, where the mean weight is 0.6 and the standard deviation weight is 0.4; The strategy reconstruction mechanism adjusts the strategy weight coefficient through the gradient descent method, and its learning rate is set to the inverse of the difference between the real-time load of the workstation and the maximum capacity of the collaborative unit.

10. The material dynamic allocation system based on data preset learning for multiple workstations according to claim 9, characterized in that: The decision optimization module includes: a collaborative unit construction submodule calling the topological relationship matrix, calculating the mean and standard deviation of the matrix elements, setting the correlation coefficient critical value to be the weighted sum of three times the mean and the standard deviation, constructing a bipartite graph of the workstation nodes, using the Hungarian algorithm to traverse the adjacency matrix to capture the maximum weight matching, and generating a collaborative unit set; The strategy matching submodule extracts the equipment processing speed and capacity parameters from the collaborative unit set to form a feature vector, calculates the cosine angle value between the material type characteristics and the equipment parameters in the preset strategy set, sets the matching threshold as the inverse of the feature vector modulus ratio, filters the strategy items with angle values ​​lower than the threshold, and generates an optimization strategy candidate set; The dynamic allocation submodule calls the optimization strategy candidate set, detects the difference between the real-time load of the workstation and the maximum capacity of the collaborative unit, adjusts the strategy weight coefficient using the gradient descent method, encodes the updated strategy instructions into a JSON format data packet, and generates a dynamic allocation instruction; The coefficient of the triple weighted sum is selected based on the correlation distribution corresponding to the peak value of the workstation collaboration efficiency in the historical data; The calculation formula of the characteristic vector modulus ratio is the square root of the sum of the equipment processing speed and capacity parameters; The iterative termination condition of the gradient descent method is that the change in the weight coefficient is less than 0.01 in three consecutive iterations.

Images