Automobile foot mat multi-style mixed flow production scheduling method based on customer order data

Abstract

The invention provides an automobile foot mat multi-style mixed flow production scheduling method based on customer order data, and the method comprises the steps: carrying out the standardization and feature vector generation of order process parameters, combining with the real-time state collection of production equipment, building a static process similarity and dynamic equipment response cost evaluation model, and fusing multi-source data to analyze the dynamic switching cost. The method comprises the following steps: establishing a scheduling sequence, realizing scheduling sequence optimization by using an improved NSGA-II multi-target genetic algorithm, comprehensively considering switching cost, equipment utilization rate and production line rhythm, finally dynamically selecting an optimal scheduling scheme through a TOPSIS decision method and a workshop load rate, and continuously correcting and evaluating model parameters by using an online incremental learning algorithm, thereby realizing scheduling sequence optimization. According to the method, the production scheduling efficiency and the resource utilization rate are improved, the equipment switching loss is remarkably reduced, and flexible production self-adaptive optimization is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing and production scheduling optimization, and particularly relates to a multi-style mixed production scheduling method for automobile foot pads based on customer order data. BACKGROUND

[0002] At present, in the field of multi-variety and small-batch customized production of automobile interior parts, especially automobile foot pads, production workshops generally adopt a multi-variety mixed production scheduling mode to cope with the diversity of customer orders. The mainstream mixed production scheduling system adopts a scheduling method based on fixed line change time estimation to perform sequence optimization and process resource allocation on order tasks. The existing scheduling technology generally regards the process switching of different styles of products as static time consumption, that is, it assumes that the time and resource cost consumed by all line change processes are fixed, and does not consider the real-time running state of the equipment and the dynamic changes of the process parameters. This kind of scheduling strategy is relatively rigid in the actual production process, and when facing complex scenes such as frequent changes of orders and increasing process diversity, it cannot effectively respond to the real-time needs of the actual load and resource switching of the workshop, resulting in difficulty in continuously improving the overall production efficiency. With the increasing customization of automobile interiors, the scheduling system needs to have higher flexibility and intelligent adjustment capability. The typical order dispatching and process resource allocation technology in the industry mainly relies on historical experience parameters or simple heuristic rules to aggregate orders with high similarity, and the resource switching cost in the line change process is often estimated by a unified time window or an average cost, lacking the dynamic quantitative evaluation capability of the actual process differences and the real-time state of the equipment. Some technical solutions introduce simple process feature comparison based on Euclidean distance and cosine similarity, but only give the surface similarity between styles based on static data, and fail to include dynamic elements such as temperature control response, tool wear and restart delay in the actual execution process of the equipment into the evaluation range of the line change cost. SUMMARY

[0003] The present application provides a multi-style mixed production scheduling method for automobile foot pads based on customer order data to solve the above-mentioned technical problems.

[0004] The technical solution of the present application is implemented as follows: a multi-style mixed production scheduling method for automobile foot pads based on customer order data, comprising: S1: obtaining a set of automobile foot pad order tasks to be scheduled, the task set containing the style identifier, process parameter set and production quantity corresponding to each order, and the process parameters including material type, sewing complexity and edge covering method parameters; S2: collecting real-time state data of the production equipment, the real-time state data including the current processing style identifier, the cumulative number of completed work orders, the equipment temperature control parameters, the wear state parameters and the latest line change timestamp; S3: Construct a style-to-style process similarity matrix based on historical process parameter data, calculate the parameter difference degree of different styles in cutting, sewing and pressing processes using the Euclidean distance algorithm, and generate a static process difference degree index; S4: Calculate the dynamic equipment response cost according to the real-time state data of the equipment, including calculating the temperature rise compensation time based on the equipment temperature control parameters, calculating the tool calibration period based on the wear state parameters, and calculating the restart delay time based on the latest line change timestamp; S5: Construct a dynamic resource switching cost evaluation model, weight and integrate the static process difference degree index and the dynamic equipment response cost to generate the dynamic switching cost of style i switching to style j, wherein the weight coefficient is calibrated based on historical line change efficiency data by least squares method; S6: Use the improved NSGA-II multi-objective genetic algorithm to optimize the production sequence, use the process similarity priority aggregation rule to initialize the population, and embed the switching cost guidance strategy in the crossover and mutation operation, so as to minimize the total switching cost, maximize the equipment utilization rate and balance the production line rhythm as multi-optimization objectives; S7: Generate candidate production sequences based on the Pareto front solution set, select the optimal production sequence in combination with the real-time load rate of the workshop, and the selection process uses the fuzzy entropy weight method to sort and decide the TOPSIS of the multi-objective optimization result; S8: Record the error value of the actual switching time and the predicted switching cost after each line change is completed, update the weight coefficient of the dynamic resource switching cost evaluation model using the online incremental learning algorithm, and the update process iteratively corrects the parameters based on the error gradient descent method.

[0005] The automobile foot mat multi-style mixed production scheduling method based on customer order data provided by the application has the following beneficial effects: (1) The application constructs a two-dimensional switching cost evaluation model that integrates static process difference degree and dynamic equipment response cost, significantly improving the fine level of process resource allocation decision-making. By pre-training the style-to-style process similarity matrix to quantify the difference of key process parameters, and combining the real-time running state of the equipment (such as temperature, wear degree, last line change time, etc.) to dynamically calculate the restart delay and calibration period, an individualized instant switching penalty value is formed, making the characterization of resource consumption in the production process more close to the physical reality. This mechanism effectively identifies and suppresses the generation of high-cost switching combinations, reducing unnecessary downtime at the source, thereby significantly improving equipment efficiency and production line continuity; (2) The improved NSGA-II multi-objective genetic algorithm is introduced, and a guiding strategy based on dynamic switching cost is embedded in the population initialization, cross variation and individual evaluation links, realizing effective collaborative optimization of multiple objectives such as minimization of total switching cost, maximization of device utilization and equalization of line beat; the heuristic rule is used to preferentially aggregate orders with high process similarity, thereby improving the quality and convergence speed of the initial solution set; in the evolution process, the switching cost weighted feedback mechanism is used to guide the genetic operation direction, avoiding the generation of inferior sequences with frequent type change and unbalanced load; at the same time, in the fitness evaluation stage of each generation, in addition to the traditional scheduling indicators, the dynamic switching cost between adjacent orders is additionally accumulated, so that the Pareto front solution set has both global performance advantage and execution feasibility. The recommended production sequence output finally not only has good comprehensive scheduling performance, but also fully considers the on-site executability, significantly enhancing the practicality and stability of the production scheduling scheme; (3) The closed-loop online learning mechanism is designed to continuously improve the adaptive ability and long-term prediction accuracy of the production scheduling system. After each actual line change operation is completed, the system automatically collects the real switching time and compares it with the model predicted value, and uses the error signal to correct the weight coefficient in the dynamic cost model in reverse, realizing the progressive learning and compensation of factors such as device aging trend and environmental disturbance. This mechanism breaks through the limitation of static modeling in dealing with working condition drift, so that the switching cost prediction ability is continuously enhanced with the extension of running time, ensuring the effectiveness and robustness of the production scheduling strategy in a long period. BRIEF DESCRIPTION OF DRAWINGS

[0006] Figure 1 A flowchart of the automobile floor mat multi-style mixed production scheduling method based on customer order data of the present application; Figure 2 A sub-flowchart of the automobile floor mat multi-style mixed production scheduling method based on customer order data of the present application; Figure 3 Another sub-flowchart of the automobile floor mat multi-style mixed production scheduling method based on customer order data of the present application. DETAILED DESCRIPTION

[0007] The embodiments of the present application will be described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application.

[0008] The following disclosure provides many different embodiments or examples for implementing various aspects of the present application. Throughout the following description, for the purposes of explanation and non limitation, specific details of certain embodiments of the application are set forth. It will be apparent to one skilled in the art that the application can be practiced in other embodiments that depart from these specific details. Moreover, descriptions of well known devices, methods, procedures, components, and testing equipment are omitted so as to not unnecessarily obscure the disclosure. Embodiments of the present application can employ various objects, apparatus, and processes that are adapted to practice the present application in different embodiments and examples. The particular embodiments presented are meant to be illustrative only and are not meant to be limiting as to the scope of the present application. Furthermore, the aforementioned description is not limiting as to the types of embodiments that work with the present application. Numerous modifications and adaptations will be apparent. Similarly, each flow chart diagram represents one embodiment of the present application. Other steps can be added and other steps can be modified or deleted to suit particular applications. Although process details are given for one embodiment, other embodiments can use different processes or different sequences of processes. Furthermore, examples of the present application can be implemented by hardware, software, firmware, middleware or a combination of hardware and software. Embodiments of the present application can be implemented as computer programs or programs of instructions embodied either fixed on a machine-readable medium, such as a diskette, CD ROM, or a hard disk drive; as a download; or transmitted over a network connection. As will be apparent, the modules can be implemented in a wide variety of configurations; therefore, the foregoing description of the specific configurations is not intended to limit the scope of the application but is merely given as one example of various embodiments of the application. The description of the embodiments of the application is not meant to limit the application to a specific embodiment but is meant to provide a description of the various embodiments of the application.

[0009] As Figure 1 shown, the present application provides a multi-style mixed production scheduling method for automobile floor mats based on customer order data, which specifically comprises: S1: Obtain the automobile floor mat order task set to be scheduled, the task set containing the style identifier, process parameter set and production quantity corresponding to each order, and the process parameters including material type, sewing complexity and edge covering method parameters; S2: Collect real-time state data of the production equipment, the real-time state data including the current processing style identifier, the cumulative number of completed work orders, the equipment temperature control parameter, the wear state parameter and the latest line change timestamp; S3: Construct a process similarity matrix between styles based on historical process parameter data, calculate the parameter difference degree of different styles in cutting, sewing and pressing processes using the Euclidean distance algorithm, and generate a static process difference degree index; S4: Calculate the dynamic equipment response cost according to the real-time state data of the equipment, including calculating the temperature rise compensation time based on the equipment temperature control parameter, calculating the tool calibration period based on the wear state parameter, and calculating the restart delay time based on the latest line change timestamp; S5: Construct a dynamic resource switching cost evaluation model, weight and integrate the static process difference degree index and the dynamic equipment response cost to generate the dynamic switching cost of style i switching to style j, wherein the weight coefficient is calibrated based on historical line change efficiency data by the least squares method; S6: Use an improved NSGA-II multi-objective genetic algorithm to optimize the scheduling sequence, use the process similarity priority aggregation rule to initialize the population, and embed the switching cost guidance strategy in the crossover and mutation operation, so as to minimize the total switching cost, maximize the equipment utilization rate and balance the production line beat as the multi-optimization objective; S7: Generate a candidate scheduling sequence based on the Pareto frontier solution set, select the optimal scheduling sequence in combination with the real-time load rate of the workshop, and the selection process uses the fuzzy entropy weight method to sort and decide the TOPSIS of the multi-objective optimization result; S8: Record the error value of the actual switching time and the predicted switching cost after each line change is completed, update the weight coefficient of the dynamic resource switching cost evaluation model using an online incremental learning algorithm, and the update process iteratively corrects the parameters based on the error gradient descent method.

[0010] The step S1: obtaining the automobile foot pad order task set to be arranged, the task set contains the style identifier, the process parameter set and the production quantity corresponding to each order, and the process parameter includes the material type, the sewing complexity and the edge covering method parameter. Specifically, it includes: S1.1: obtaining the automobile foot pad order task set to be arranged from the enterprise resource planning system, the task set includes a plurality of order entries, each order entry contains a corresponding style identifier, a process parameter set and a production quantity, wherein the style identifier is used to uniquely identify different types of automobile foot pad products, the process parameter set includes the material type, the sewing complexity and the edge covering method parameter, and the production quantity represents the total quantity of the order of the style foot pad; Based on the order information database stored in the enterprise resource planning system, a structured data extraction method (parameters: order number, style identifier, process parameter set, production quantity field) is used to realize the batch loading of automobile foot pad order task entries; Further, through the data integrity checking algorithm (parameters: mandatory field set ={style identifier, material type, sewing complexity, edge covering method, production quantity}), the field integrity of the extracted order entries is checked, and an order record validity matrix is obtained, which is used to identify the compliance of each record in the mandatory field; Further, through the data type matching algorithm (parameters: field type mapping table ={style identifier→string, material type→string, sewing complexity→float, edge covering method→string, production quantity→integer}), the matching check of each field value and the preset type is realized, and a field type compliance report is generated, which is used for type constraint in the subsequent data conversion stage; Further, through the business rule filtering logic (parameters: ={production quantity>0, sewing complexity range [0.0,5.0]}), the order that does not conform to the production logic is removed, and the filtered order task set data is generated, which is used to reduce the noise interference in the subsequent similarity calculation process; Through the unique identifier binding method (parameters: primary key field=style identifier), the style identifier in the order entry is associated with the historical production record of the same style to establish an index, realizing the uniqueness and traceability of the style identifier of each entry in the order task set; Exemplary, in the ERP system of a certain automotive interior manufacturing enterprise, there are 100 orders of automotive floor mats to be scheduled in the order information database. Using the structured data extraction method to set the field mapping table {Order_ID, Style_ID, Material_Type, Sewing_Complexity, Edge_Type, Quantity}, all matching entries can be extracted from the database at one time. Through the data integrity checking algorithm, it is detected that 2 records are missing the edge type field and are marked as invalid records. In the field type matching stage, the sewing complexity field is read as a floating point number, and through matching detection, it is found that the fields of 3 records are strings "high" and "low", which do not conform to the preset floating point type constraint and are automatically converted to standard process level values. The business rule filtering logic sets the production quantity threshold to 10, and 1 record with a production quantity of 0 is excluded. After processing, there are 94 valid order task sets, in which the style identifier has been uniquely indexed with the historical production records, and in the subsequent process parameter standardization processing stage of S1.2, the index can be directly called to generate material type coding and process feature vector, avoiding repeated calculation and significantly improving data processing performance; S1.2: Standardizing the process parameter set in the obtained order task set, converting the material type to material category code, converting the sewing complexity to sewing process level index, and converting the edge type to edge process type code to form structured process parameter data in a unified format for subsequent calculation of process similarity between styles; The process parameter set in the order task set output by the enterprise resource planning system is converted to a unified category index using the coding mapping method (parameters: material type, sewing complexity, edge type); Further, through the material category mapping algorithm (parameters: material type data table, preset material category dictionary), the material type field is converted to the material category code, and the material category code data is obtained; Further, through the sewing complexity grading algorithm (parameters: sewing complexity value domain, process level division rule), the sewing complexity field is converted to the sewing process level index, and the sewing process level index data is generated; Further, through the edge type mapping module (parameters: edge type text description, edge process coding table), the edge type is converted to the edge process type code, and the edge process type code data result is formed; Through the field-level standardization processing method (parameters: material category code, sewing process level index, edge process type code), the above three types of coding data are integrated into structured process parameter data in a unified format, realizing unified dimension and comparability; For example, for a set of process parameters of a certain style in the order task set, the material type is "PU leather", the sewing complexity is the input level value 7, and the edge covering method is "hot melt edge covering". The system calls the material category mapping algorithm, retrieves "PU leather" in the material category dictionary and maps it to code M_05; through the sewing complexity grading algorithm, the value 7 is mapped to the level index ; through the edge covering method mapping module, "hot melt edge covering" is corresponded to code B_03. The above codes are processed by the field level standardization method to construct a uniform format vector {M_05, 0.7, B_03}. The vector has consistent dimensions in the key process dimension and can be directly involved in subsequent process similarity calculation and dynamic resource switching cost evaluation. In the test scene, the standardization process ensures the consistency of the data structure of different sources of order data, effectively improving the precision of the construction of the static process difference matrix; S1.3: Based on the standardized structured process parameter data, generate a process feature vector corresponding to each order, which is composed of material category code, sewing process level index and edge covering process type code in a predetermined order, used to represent the feature distribution of the insole style corresponding to the order in the key process dimension; Based on the standardized structured process parameter data output by the previous step, the field position mapping method (parameters: material category code, sewing process level index, edge covering process type code) is used to realize the ordered arrangement of key process parameters in the feature vector, to ensure the structural consistency of different orders in the feature representation; Further, through the feature vector dimension constraint algorithm (parameter: preset feature order matrix), the vectorization conversion of each process parameter occupying a fixed dimension position is realized, and the preliminary process feature array result is obtained; Further, the numerical coding fusion method (parameter: coding weight coefficient set) is used to realize the fusion expression of material category code, sewing process level index and edge covering process type code in a single vector space, and generate a process feature intermediate vector in a unified dimension space; Further, based on the vector normalization processing algorithm (parameter: Z-score standardization), the unified conversion of different parameter dimensions in the vector is realized, and a normalized process feature vector matrix is generated; Through the vector encapsulation processing method, the result of the previous step is converted into a standardized process feature vector that can be directly called by the process similarity calculation module, realizing the input data required for accurate comparison of process features between different orders; For example, for an order containing material category code 3, sewing process level index 2.5, and edge process type code 5, the preset feature order matrix sets the material category code as the first dimension, the sewing process level index as the second dimension, and the edge process type code as the third dimension. First, the field position mapping method is used to arrange (3, 2.5, 5) into the preliminary feature array [3, 2.5, 5]. Then, in the numerical code fusion, the weight coefficient set is set as (0.5, 0.3, 0.2), and the fusion vector 0.5x3+0.3x2.5+0.2x5 is obtained, and the calculation result is 3.15, which is the intermediate vector expression of the single space measurement. In the normalization processing stage, assuming that the sample mean is 3 and the standard deviation is 0.5, the first dimension normalization value is 0, the second dimension normalization value is -1, and the third dimension normalization value is 4, and the normalized process feature vector [0, -1, 4] is finally formed. After encapsulation, the vector is input into the process similarity calculation module, which can significantly improve the accuracy and stability of the process difference calculation between different orders in the subsequent steps. =0, the second dimension normalization value is =-1, and the third dimension normalization value is =4, and the normalized process feature vector [0, -1, 4] is finally formed. After encapsulation, the vector is input into the process similarity calculation module, which can significantly improve the accuracy and stability of the process difference calculation between different orders in the subsequent steps; S1.4: The style identifier, process feature vector, and production quantity of each order are data encapsulated to generate an order task data structure body, which is stored in the form of key-value pairs, with the key being the style identifier and the value being a composite data object containing the process feature vector and the production quantity. This is used for subsequent process similarity comparison and production quantity priority sorting in the production planning process; S1.5: The encapsulated order task data structure body is transmitted to the process similarity analysis unit of the production planning module as one of the input data of the dynamic resource switching cost evaluation model, which is used to support the subsequent process similarity-based production sequence optimization process.

[0011] The step S2: collecting real-time state data of the production equipment, the real-time state data including current processing style identifier, cumulative completed order number, equipment temperature control parameter, wear state parameter and recent line change timestamp. Specifically, it includes: S2.1: Collecting the current processing style identifier of the production equipment based on the industrial Internet of Things protocol to determine the current processing style of the automobile foot mat and obtain the current processing style parameter for subsequent switching cost calculation; The object of collection is the current processing style identifier data of the production equipment, and the initial condition is that the equipment has accessed the production planning system through the industrial Internet of Things network and has a real-time queryable equipment control state interface; The device data acquisition method based on the industrial Internet of Things (IIoT) protocol (parameters: MQTT topic is the device ID / CurrentStyle, refresh cycle is set to 500 ms) is adopted to realize real-time pulling of the current processing task identification signal of the device; Further, through the style code analysis algorithm output by the device controller (parameters: the mapping table is derived from the order management system, and the fields include the binding relationship between the style identifier and the process feature vector), the original style code collected is decoded and processed to obtain structured style identification data that can be directly matched with the order task set. Further, through the data integrity checking method (parameters: CRC-16 check polynomial is set to 2 16 -17), whether the decoded style identification data has errors during transmission is detected, and the current processing style parameters that pass the check are generated. Further, through the device state synchronization algorithm (parameters: timestamp alignment tolerance ≤ 5 ms), the current processing style parameters are compared with the corresponding structure in the order task set to realize real-time closed-loop synchronization of the process parameter link. Further, through data encapsulation processing, the current processing style identifier, the corresponding process feature vector, and the device ID are stored as a JSON format object for subsequent switching cost calculation module calling, realizing the consistency and expandability of the data interface. Through the real-time acquisition and analysis processing method based on the industrial Internet of Things protocol, the device running information in the previous step is converted into standardized current processing style parameters to realize the preparation of key input data required for switching cost calculation. For example, in a sewing production line for a car mat, the sewing device numbered EQ-Sew-07 transmits the current style code signal through the "EQ-Sew-07 / CurrentStyle" topic of the MQTT protocol, the refresh cycle is configured as 500 ms, and the transmission load is binary style code data length of 16 bits. The original code received by the acquisition module is 0x1A3C, which corresponds to the style identifier "S-Comfort-2024" after style mapping table analysis, and its corresponding process feature vector is [material category code: 05, sewing process level index: 3, edge process type code: B2]. The data integrity check passes the CRC-16 polynomial 2 16-17Calculate the check value and get the consistent check result, confirm the data is valid. The state synchronization module aligns the timestamp error of 3.2ms, which meets the synchronization tolerance of ≤5ms, and binds the current processing style parameters to the equipment ID "EQ-Sew-07" to encapsulate the object in the format of {"DeviceID":"EQ-Sew-07","StyleID":"S-Comfort-2024","CraftVector":[05,3,"B2"]} and push it to the switching cost evaluation module. The output result ensures that the style parameters used in the subsequent dynamic resource switching cost calculation are completely consistent with the real-time processing state of the workshop, significantly improving the reliability of switching cost prediction; S2.2: The cumulative completed work order number of the production equipment is statistically processed to obtain the production load state of the current equipment, and the cumulative completed task quantity index of the equipment is generated as an input parameter for evaluating the running efficiency of the equipment; S2.3: Based on the temperature sensor collected equipment temperature control parameters, the temperature data is filtered to obtain the current thermal state characteristics of the equipment, which are used to calculate the temperature rise compensation time consumption in the line change process; Based on the input of temperature sensor collected equipment temperature control parameters, the object is a high-precision digital temperature sensor installed on the key processing equipment, the signal output unit is Celsius, and the sampling frequency is set to 1Hz by the equipment control system for synchronization, which is used to reflect the current thermal state of the equipment; A low-pass finite impulse response filtering algorithm (parameters: filter order = 16, cutoff frequency = 0.25Hz) is used to realize high-frequency noise suppression of the collected temperature sequence signal, and the real temperature change trend data of the equipment is retained; Further, through the sliding average filtering method (parameters: window size = 5 sampling points), the short-term fluctuation smoothing processing of the temperature signal after low-pass filtering is realized, and the temperature stable value sequence is obtained, which is used for subsequent thermal state characteristic calculation; Further, through the difference calculation method (parameters: order = 1), the instantaneous temperature change rate of the stable value sequence is extracted, and a set of thermal dynamic characteristic values is generated to describe the thermal response speed of the equipment under the current processing task; Further, through the normalization processing method (parameters: minimum-maximum standardization), the unified dimension conversion of the temperature stable value and the change rate is realized, and a temperature feature vector is generated to describe the comprehensive characteristics of the current thermal state of the equipment in the static temperature level and dynamic change trend; Through the temperature rise time prediction formula based on the first-order inertia model of heat conduction, the temperature feature vector is converted into the temperature rise compensation time index, realizing the quantitative prediction of the additional time consumption of temperature rise in the line change process. The temperature rise time prediction formula is as follows:

[0012] wherein T is the temperature rise compensation time consumption, is the difference between the target processing temperature and the current device temperature, k is the device heat conduction coefficient, and R is the heat response rate; For example, in the temperature monitoring scenario of a certain automobile foot mat sewing device, the temperature sensor sampling frequency is configured as 1 Hz, and the collected original temperature sequence is [45.3, 45.5, 46.0, 46.2, 45.9, 46.1] degrees Celsius. The original sequence is processed by using a low-pass finite impulse response filtering algorithm (order 16, cutoff frequency 0.25 Hz) to output smoothed data [45.35, 45.48, 45.93, 46.15, 45.92, 46.08]. Further, a sliding average filter (window size 5) is used to obtain a stable value sequence [45.564, 45.716, 45.888, 46.002]. The differential calculation rate result is [0.152, 0.172, 0.114] degrees Celsius / second. After minimum-maximum normalization, the temperature level and the rate of change are converted into [0.45, 0.48, 0.60, 0.65] and [0.70, 0.80, 0.50] standardized features, respectively. The target processing temperature is set to 60℃, the device heat conduction coefficient k is 0.85, and the heat response rate R is 1.2. The temperature rise compensation time consumption T is calculated by substituting the formula: the current temperature mean is 45.95℃, = 14.05℃, then ≈ 13.78 seconds. It is verified that the predicted value is close to the actual line change temperature rise time consumption, which realizes the accurate quantification of the temperature rise compensation time consumption and effectively supports the calculation of the subsequent dynamic resource switching cost model; S2.4: Collect and normalize the device wear state parameters, extract the device wear degree feature value based on the vibration sensor and current signal, and use it to evaluate the tool calibration period during line change; S2.5: Extract the latest line change timestamp based on the device control system log, and perform format standardization processing on the timestamp data to obtain the latest line change time of the device, which is used to calculate the device restart delay time.

[0013] As Figure 2 shown, the step S3: constructing a style-to-style process similarity matrix based on historical process parameter data, using the Euclidean distance algorithm to calculate the parameter difference degree of different styles in the cutting, sewing and pressing processes, and generating a static process difference degree index. Specifically, it includes: S3.1: Obtain historical automobile foot mat production order data, the historical order data including the cutting parameters, sewing parameters and pressing parameters corresponding to each style, and based on the parameter set, construct a style process feature vector matrix to form a comparable structured data basis between styles; Based on the production management system database interface call historical order data query method (parameters: query time range, product type is automobile foot pad, data field includes cutting parameter, sewing parameter and pressing parameter), the batch extraction of historical automobile foot pad production order data is realized; Further, through the data field mapping method (parameters: field name and process parameter category association mapping table), the accurate positioning of the cutting parameter, sewing parameter and pressing parameter in the extraction record is realized, and the process parameter set corresponding to each style is obtained; Further, the data cleaning algorithm (parameters: missing value filling rule, abnormal value detection threshold) is used to realize the consistency detection and repair processing of the extracted cutting parameter, sewing parameter and pressing parameter, and generate a structured parameter data set after cleaning; Further, based on the multi-dimensional feature coding method (parameters: cutting parameter dimension, sewing parameter dimension, pressing parameter dimension and sequencing rule), the process parameter set of each style is mapped to a numerical feature vector, and a process feature vector set is obtained; Further, the matrix construction method (parameters: row index is style identifier, column index is process feature dimension) is used to realize the combination of the process feature vector set into a style process feature vector matrix, to form a structured data basis comparable between styles; Through the matrix construction method, the process feature vector set of the previous step is converted into a style process feature matrix, which provides a unified multi-dimensional numerical data input for subsequent process similarity calculation; For example, in the order management system, the query time range is set to the past 12 months, the product type is limited to automobile foot pad, and the data field includes process-related parameters such as "cutting tool model", "sewing needle spacing", "edge glue pressing temperature", etc. Call the data interface to obtain 100 historical order entries, and the field mapping table associates "cutting tool model" to cutting parameter dimension, "sewing needle spacing" to sewing parameter dimension, and "edge glue pressing temperature" to pressing parameter dimension. Use the data cleaning algorithm to fill in the missing needle spacing value as the average needle spacing of the style, and adjust the abnormal glue pressing temperature (detection condition is higher than 300℃) to the experienced safe value 180℃. Numerical coding is performed on the cleaned data: the cutting tool model is coded as an integer category ID, the sewing needle spacing is converted to a floating point number in millimeters, and the edge glue pressing temperature is recorded as a floating point number in Celsius. According to the order of [cutting parameter, sewing parameter, pressing parameter], it is combined into a three-dimensional process feature vector, such as the feature vector of a certain style is 12, 3.5, 180. Sort all 100 feature vectors by style identifier to form a 100 row x 3 column style process feature vector matrix, which is stored in a matrix data structure for subsequent S3.2 normalization processing and process similarity calculation; S3.2: Normalizing the style process feature vector matrix, using the minimum-maximum standardization method to map the interval of each dimension parameter, to eliminate the influence of different parameter dimension differences on the subsequent Euclidean distance calculation, and obtaining the standardized process feature vector matrix; S3.3: Based on the standardized process feature vector matrix, using the Euclidean distance algorithm to calculate the distance between any two styles of process parameter vectors, generating the original value of the process difference between the two styles, to quantify the process difference of different styles in the key process; S3.4: Normalizing all the original values of the process difference between the two styles, using the Z-score standardization method to align the distribution of the distance value, generating the standardized process difference matrix, to ensure the comparability and consistency of the process difference index in the subsequent resource switching cost model; The input is the original value matrix of the process difference between the two styles generated in step S3.3, and the matrix elements are the process parameter difference values calculated based on the Euclidean distance algorithm. The Z-score standardization method (parameters: sample mean μ, sample standard deviation σ) is used to perform distribution alignment processing on the difference value matrix, to realize the unified dimension conversion of different difference intervals. Further, the normalized difference matrix element value Z is calculated by the following formula:

[0014] Where X is the original process difference value, μ is the arithmetic mean of all element values, and σ is the sample standard deviation of all element values. Further, by element-by-element calculation, the original difference value of each style pair is replaced by the corresponding Z-score standardized value, realizing the target state of the distribution mean of each element in the matrix being zero and the standard deviation being one. Further, a small positive number ε is introduced to correct the deviation in the case where σ tends to zero in the calculation process, to prevent division by zero error and maintain numerical stability. Through the Z-score standardization processing method, the original process difference matrix in the previous step is converted into a standardized process difference matrix, realizing the elimination of different parameter dimension differences and the comparability and consistency of the process difference index in the subsequent dynamic resource switching cost evaluation. For example, in the production of automobile floor mats, the input process difference original matrix is 3x3, and the element values are 0.85, 0.76, 1.02, 0.91, 0.88, 0.95, 0.79, 1.05, 0.87, respectively. The matrix mean μ is 0.898, and the sample standard deviation σ is 0.091. Based on the formula, the first element Z value is , the result is -0.527, and the normalized matrix is obtained by sequentially calculating all elements. After normalization, the mean of the numerical value distribution of each element is zero, and the standard deviation is one, so that the differences in cutting, sewing and pressing processes can be compared on the same numerical scale. After inputting the matrix into the equipment scheduling optimization algorithm, the similarity aggregation rule and the switching cost guide strategy can significantly improve the accuracy and convergence speed of sequence optimization; S3.5: Based on the normalized process difference matrix, a style process similarity matrix is generated, where the similarity value is 1 minus the process difference value, to form a style similarity quantification basis at the process level, which is used as a static process difference index in the subsequent dynamic resource switching cost evaluation model; Based on the input of the normalized process difference matrix, the matrix element inversion operation method (parameters: normalized difference value range [0, 1]) is used to realize the mapping and conversion of the difference value to the similarity value; Further, the numerical subtraction operation method (parameters: difference value d ij ) is used to realize the calculation of the similarity value and obtain the process similarity value matrix element between style i and style j, and the formula used is as follows:

[0015] Where S ij is the process similarity value between style i and style j, and d ij is the normalized process difference value generated in the previous step; Further, the matrix traversal assignment method (parameters: total number of styles n) is used to realize the calculation of each element of the similarity matrix one by one, and a complete process similarity matrix between styles containing n×n elements is generated; Further, the matrix symmetry detection method (parameters: similarity matrix S) is used to realize the symmetry verification of the matrix, and the average correction processing is performed when there is a numerical deviation, so as to obtain a symmetric similarity matrix with consistent numerical values; Further, the matrix normalization method (parameters: similarity value range [0, 1]) is used to realize the global normalization of the similarity matrix, so as to ensure the comparability and consistency of the similarity values of different styles; Through the matrix inversion and normalization processing method, the normalized process difference matrix result of the previous step is converted into a process similarity matrix between styles, realizing the quantitative representation of the static process difference index and providing accurate input for the dynamic resource switching cost evaluation model; For example, in a historical process difference data set containing 5 styles of automobile floor mats, the difference matrix element range is 0.1 to 0.8, and when calculating the similarity matrix using the above formula, the difference value d 12 =0.25 between style 1 and style 2 is calculated, and the similarity value S 12=0.75. After the similarity matrix is ​​traversed and calculated, the matrix element S is checked. 21 With S 12 The difference was 0.02, which was corrected to a consistent value of 0.76 using the averaging method. After global normalization, the matrix elements remained within the range of [0,1]. After this step, the similarity matrix was used as the input for the static process difference index in the dynamic resource switching cost evaluation model, enabling subsequent production scheduling optimization algorithms to aggregate and schedule high-similarity orders, significantly improving the continuous processing efficiency of equipment and reducing the cost of frequent line changes.

[0016] like Figure 3 As shown, step S4 involves calculating the dynamic equipment response cost based on real-time equipment status data. This includes calculating the temperature rise compensation time based on equipment temperature control parameters, calculating the tool calibration cycle based on wear status parameters, and calculating the restart delay time based on the most recent line changeover timestamp. Specifically, this includes: S4.1: Calculate the temperature compensation model based on the equipment's temperature control parameters to obtain the time required for the equipment to recover from a shutdown state to the target processing temperature. This calculation uses the equipment's historical temperature control curves as training data, employs a first-order inertial system model to model the temperature control response process, and outputs a predicted temperature rise time. The input object for the temperature compensation model calculation based on the equipment temperature control parameters is the current thermal state parameters of the equipment after being collected and filtered from the temperature sensor. The parameters include the real-time temperature values ​​of multiple key temperature measurement points inside the equipment and their corresponding collection timestamps. The historical temperature control curve regression analysis method (parameter source: historical heating process data stored in the equipment process archive) is used to achieve parameterized fitting of the equipment thermal response path, and the temperature change process in the heating stage is modeled as the dynamic equation structure of a first-order inertial system. Furthermore, the parameters of the first-order inertial system model, including the time constant, are fitted using the least squares method. With steady-state gain K, the temperature rise rate characteristics of different equipment types under different operating conditions are identified, and a dataset of temperature rise response parameters specific to the current equipment is obtained. Furthermore, through a first-order inertial system temperature control response model (the equation structure is as follows:) To achieve the prediction and calculation of temperature rise compensation time, where The target processing temperature, The actual temperature of the device currently collected by the sensor, where t is the time variable of the heating process; Furthermore, the above model equations are solved analytically using the formula inversion method to obtain the predicted temperature rise compensation time. Its computational structure is

[0017] in to reach the temperature value corresponding to the steady state determination threshold value; The temperature rise time prediction value is calculated by a temperature rise compensation model, the result of the previous step is converted into time data required for the quantitative equipment to recover from the shutdown state to the target processing temperature, and the real-time update of the temperature control dimension parameter in the dynamic equipment response cost model is realized; For example, in the automobile foot mat sewing station, the temperature sensor measures the initial temperature of the current equipment as 80°C, the target processing temperature as 35°C, and the time constant obtained by fitting the historical temperature control curve as 120 seconds, and the steady state gain K as 1.0. Using the above model equation, the steady state determination temperature is set as 78°C, the parameters are substituted into the formula =120×3.258=390.96 seconds. The predicted time consumption is used for the temperature control compensation term of the subsequent dynamic response cost model, and the field verification shows that the deviation between the model prediction value and the actual measured time consumption is less than 5 seconds, realizing the high-precision application effect of the temperature control parameter in the production scheduling optimization; S4.2: Perform wear compensation analysis on the equipment wear state parameter to obtain the tool calibration period required by the current equipment after line change. The analysis is based on the cumulative running time of the equipment and the wear model of the key components, and a mapping relationship between the wear degree and the calibration time consumption is established by using the linear regression method, and a tool calibration time prediction value is output; S4.3: Perform restart delay calculation based on the time difference between the latest line change timestamp and the current timestamp to obtain the restart delay time required by the equipment from switching to the current task from the last processing task. The calculation combines the equipment control system response time distribution model, and dynamically estimates the restart delay by using the sliding window average method, and outputs the restart delay time value; S4.4: Weighted sum of the temperature rise compensation time consumption, tool calibration period and restart delay time to generate the dynamic equipment response cost when the equipment switches from the current processing style to the next target style. The weight coefficient is calibrated based on the equipment type and the historical line change efficiency data by the least square method, and a dynamic response cost index is output; The input temperature rise compensation time consumption, tool calibration period and restart delay time are all from the calculation results of steps S4.1 to S4.3, and the data form is numerical time parameter, the unit is second, and the structured vector form is transmitted to the weighted fusion processing unit; The weighted sum method (parameters: temperature rise compensation time , tool calibration period , restart delay time ) is used to realize the unified quantitative processing of the three response times, and convert them into a single dynamic equipment response cost index C. The formula is as follows:

[0018] ​​​Furthermore, using the least squares method (parameter: historical line-changing efficiency dataset) Response time sample matrix The system calibrates the weight coefficients w1, w2, and w3, and performs grouped regression based on different equipment types to obtain a set of weighted parameters suitable for each equipment category. Furthermore, the calibration weight coefficients are scaled by normalization and the original weight values ​​are transformed using the Z-score method to ensure that they maintain consistent dimensions in the fusion operation and avoid the bias of the results by a single response time index. The dynamic response cost index output by weighted summation is the quantified value of time loss of the equipment in the case of processing style switching, which is used for the generation of the equipment state-cost mapping table in S4.5. Through the above weighted fusion and weight calibration process, the multi-source response time parameters calculated in the previous step are transformed into a unified dynamic response cost index, so as to realize the reasonable contribution allocation of different response items to the switching cost and improve the accuracy and adaptability of cost assessment. For example, in a mixed-flow production line for car floor mats, the time required for temperature rise compensation of a certain cutting machine... The tool calibration cycle is 85 seconds. The restart delay time is 45 seconds. The response time is 38 seconds, the equipment type is a high-speed cutting machine, and the corresponding historical sample matrix X contains response time records for 300 work orders. The actual total switching time vector Y contains the corresponding total time data. The weight coefficients obtained by least squares regression are w1=0.5, w2=0.3, w3=0.2, and the normalized coefficients are w1'=0.52, w2'=0.31, w3'=0.17. Substituting these values ​​into the weighted summation formula, the following calculation is performed: The dynamic response cost was calculated to be 66.54 seconds. This metric was used in subsequent production scheduling optimization to evaluate the time loss of switching the high-speed cutting machine from the current processing style to another style, which helps to reduce the switching frequency in the optimization constraints, thereby significantly improving the overall operating efficiency of the production line. S4.5: Bind the dynamic equipment response cost index to the current processing style identifier of the equipment to generate an equipment status-cost mapping table, which can be called by the subsequent dynamic resource switching cost evaluation model, thereby realizing personalized modeling of switching costs between different styles.

[0019] Step S5: Construct a dynamic resource switching cost assessment model, which weights and fuses static process difference indicators with dynamic equipment response costs to generate the dynamic switching cost of switching from style i to style j. The weighting coefficients are calibrated using the least squares method based on historical line changeover efficiency data. Specifically, this includes: S5.1: Construct a process-equipment response cost mapping table based on historical line change efficiency data, use least squares method to regress model the correlation between historical line change time and process difference degree, equipment state parameters, to obtain an initial weight coefficient set, providing model parameter basis for subsequent dynamic fusion; S5.2: Normalize the static process difference degree index and dynamic equipment response cost, use Z-score standardization method to unify the dimension conversion of different dimension input data, to eliminate the influence of different dimensions on the model fusion result; S5.3: Based on the normalized static process difference degree index and dynamic equipment response cost, perform weighted fusion operation, where the weight coefficient comes from the least squares regression model output constructed in S5.1, to generate a preliminary dynamic switching cost matrix; Based on the normalized static process difference degree index and dynamic equipment response cost, a linear weighted fusion algorithm (parameters: weight coefficients w1, w2) is used to realize the unified cost quantification of different dimension resource switching factors; Further, through matrix weighted operation (parameters: static process difference degree matrix S ij , dynamic response cost matrix D ij , weight vector w), the comprehensive switching cost initial value of each pair of styles i and j is calculated, and a preliminary dynamic switching cost matrix C is obtained, each element of which reflects the total loss of style i switching to style j under the current state; Further, the matrix element is calculated by the following weighted fusion formula:

[0020] Where w1 is the static process difference degree weight, w2 is the dynamic equipment response cost weight, S ij is the process difference degree matrix element, D ij is the equipment response cost matrix element; Further, through matrix traversal algorithm (parameters: order style list, weight vector), the element generation process of all style combinations is realized, and is stored in the form of two-dimensional array for subsequent calling; Through this weighted fusion processing method, the normalized static and dynamic indicators in the previous step are converted into switching cost matrix with direct comparability, realizing the comprehensive measurement of resource switching cost; For example, in the mixed flow production scene of automobile floor mats, the value range of static process difference degree matrix S ij is 0~0.8, the value range of dynamic equipment response cost matrix D ij is 0~5.0, and the weights w1 and w2 are calibrated to 0.35 and 0.65 respectively from historical line change efficiency data, and the normalized Sij and D ij The weighted calculation is performed to obtain = 0.35x0.6 + 0.65x3.2, which is calculated by the following formula:

[0021] The value is 2.29, indicating that the expected cost of switching from style 1 to style 2 is 2.29 time units or equivalent loss units. In production testing, this matrix is applied for production scheduling optimization, so that high-cost switching sequences are effectively avoided, the idle time of the production line is significantly reduced, the equipment utilization rate is greatly improved, and the production cycle stability is obviously improved; S5.4: Nonlinear correction is performed on the preliminary dynamic switching cost matrix, and a Sigmoid function is introduced to amplify the switching situation with high difference or high device response cost, so as to enhance the sensitivity of the model to high-cost switching, and generate a corrected dynamic switching cost matrix; The preliminary dynamic switching cost matrix obtained by weighted fusion is processed by using a S-type nonlinear amplification function (parameters: curve slope coefficient k, translation coefficient x0), so as to realize the value enhancement of the high process difference or high dynamic device response cost situation; Further, the Sigmoid transformation formula is used to map the matrix elements, and the formula is as follows:

[0022] wherein, is the element value in the preliminary dynamic switching cost matrix, i and j are style indexes, k is the gain coefficient of the S-type curve slope, and x0 is the position parameter of the turning point of the cost curve; Further, the Sigmoid function output is multiplied with the original cost value element by element, so as to realize the nonlinear amplification of the high-cost situation, while keeping close to linear for the low-cost situation, so as to avoid unnecessary adjustment of the low-impact switching sequence in production scheduling optimization; Further, by setting the calibration rules of k value and x0 value (according to the mean and variance of historical cost distribution), the sensitivity of the Sigmoid function in different production scenarios is optimized, so that the amplification multiple rapidly increases for the cost difference value exceeding the turning point, thereby highlighting the influence weight of high-cost switching; Through the above nonlinear correction algorithm, the preliminary dynamic switching cost matrix generated in the previous step is converted into a corrected matrix containing process difference and device response cost sensitivity, so as to realize the key inhibition of high-cost switching situation, and improve the precision and robustness of the production scheduling optimization model in multi-variety mixed production; For example, in the application scenario of a certain automobile foot mat production line, the preliminary dynamic switching cost matrix The element range of the matrix is 0.8 to 5.2, the historical cost distribution mean is 2.6, and the variance is 0.9. Set x0 to 2.6 and k to 1.15. For the matrix element =4.8, calculate by the above nonlinear correction formula: =0.907. Multiply the amplification coefficient by the original value to obtain the corrected switching cost 4.354, which is higher than the mean interval, effectively increasing the punishment for high-cost switching in production scheduling optimization. In actual production scheduling, sequences containing such high correction value switching are quickly eliminated in genetic evolution iteration, significantly reducing the fluctuation of overall equipment switching time in the workshop, and stabilizing the production rhythm; S5.5: Based on the real-time input of the current equipment state and order process parameters, the dynamic switching cost matrix is updated online, and the latest change record is integrated using a sliding window mechanism to generate the final dynamic resource switching cost evaluation model output C_ij as a key input indicator for production sequence optimization; Based on the real-time input data of the equipment state-cost mapping table and the order process feature vector, a sliding window update mechanism (window length parameter: take the last N change records, N is determined by the scheduling system load fluctuation period) is used to realize the online iterative update of the dynamic switching cost matrix ; Further, by the window data weighted average method (weight is set according to the record time decreasing function), the latest static process difference degree change and dynamic equipment response cost change are integrated to correct the cost unit value of the corresponding style switching path; Further, an exponential smoothing algorithm (smoothing coefficient α is controlled by the inverse ratio of the model prediction error mean square deviation) is used to trend filter the weighted and corrected switching cost unit, to eliminate short-term fluctuations caused by temporary equipment abnormalities or extreme order values, and to obtain a smoothed cost matrix; Further, through online normalization processing (Z-score method, mean and variance are dynamically calculated in the sliding window), the unit values of the smoothed cost matrix are converted into standardized indicators to ensure the consistency of the input of the subsequent optimization algorithm; Further, based on the standardized matrix, difference sensitive adjustment is performed, a nonlinear amplification operator is introduced to additionally punish style switching with a cost increase exceeding a certain threshold, and the response speed of the model to new high-cost switching situations is improved; Through the combination of online updating and trend smoothing, the dynamic in the previous step is converted to C ij , realizing real-time adaptation to the latest production state and order structure as a key input technical indicator for production sequence optimization; Exemplary, in a certain car mat manufacturing workshop, the sliding window length is set to 10 times the line change record, the decreasing weight parameter is taken as the power sequence of 0.85, and the smoothing coefficient a is set to 0.3. In the last workshop state collection, the order process characteristic vector change caused the cutting process difference to rise by 0.12, and the device temperature rise compensation time to decrease by 5 seconds. The window fusion calculation obtains the update cost value of the cutting process from style A to style B as 42.6 seconds, which is reduced to 40.1 seconds after exponential smoothing, and the corresponding unit value in the normalized matrix is 0.71. This value significantly improves the proportion of low-cost switching sequences in the production line scheduling optimization, and the production scheduling result shows that the comprehensive utilization rate of the device is improved and the number of high-cost switching is reduced.

[0023] The step S6: using the improved NSGA-II multi-objective genetic algorithm for scheduling sequence optimization, using the process similarity priority aggregation rule for initialization population, embedding the switching cost guiding strategy in the crossover and mutation operation, taking the minimum total switching cost, the maximum device utilization rate and the balanced production line beat as the multi-optimization target. Specifically, it includes: S6.1: Based on the switching cost matrix output by the dynamic resource switching cost evaluation model, the initialization population generation rule is used to cluster the order task set using the process similarity priority aggregation strategy to reduce the proportion of high switching cost sequences in the initial population, thereby improving the convergence efficiency of the algorithm; S6.2: Encode each individual in the initialization population, use an integer coding method based on process priority relationship, map the scheduling sequence to a chromosome gene sequence to facilitate the execution of subsequent genetic operations, while ensuring that the chromosome decoding process is consistent with the actual scheduling logic; S6.3: Improve the crossover operation based on the switching cost guiding strategy, design a cost-sensitive local matching crossover operator (COST-LMX), which preferentially retains the low switching cost scheduling order combination in the gene fragments of the two parent individuals, to enhance the genetic stability of high-quality genes in the population; Based on the improved crossover operation input conditions are the parent chromosome gene sequence coded from S6.2 and the dynamic switching cost matrix C ij generated by S5.5; Using the cost-sensitive local matching crossover operator design method (parameters: switching cost matrix C ij , local matching window length L), the crossover region of the offspring chromosome is located and initially paired; Further, by the window cost calculation method (parameters: L, C ij ), the switching cost of each gene fragment in the matching window of the two parents is quantitatively evaluated, and a local switching cost vector K is obtained; Furthermore, a cost-first selection strategy algorithm (parameter: K) is used to select and judge the segments at the same position of the two parents, prioritizing the retention of sequence segments with lower switching costs, and generating cost-optimized crossover segment combinations; Furthermore, through a cost-constrained repair mechanism (parameter: target switching cost threshold T), the global switching cost is calculated after the offspring chromosome is generated. When the threshold T is exceeded, a local permutation function is called to replace the high-cost fragment with a candidate low-cost fragment, and the corrected offspring gene sequence is generated. By using a cost-sensitive local matching crossover operator, the parent chromosome sequence from the previous step is transformed into a progeny sequence that balances low switching costs and structural integrity, thereby achieving stable transmission of high-quality genes in the population and satisfying the constraints of production optimization objectives. For example, in a mixed-production scenario for car floor mats, a local matching crossover with L=3 is performed on parent sequence A [1,3,5,2,4,6] and parent sequence B [3,1,2,5,6,4], switching cost matrix C. ij C 13 =5, C 35 =4, C 52 =7, C 24 =3, etc. Using an in-window cost calculation method, the switching costs of each window segment of the parent generation are summed. For example, the switching cost of the first window of parent generation A is 5+4=9, and the corresponding window switching cost of parent generation B is 6+2=8. A cost-first selection strategy is executed, retaining the first window segment of parent generation B [3,1,2]. After repeated processing, an initial child generation [3,1,2,2,4,6] is generated, and the global switching cost ΣC is calculated. ij =28, which is higher than the set threshold T=25. The cost constraint repair mechanism is invoked to replace the high-cost segment [2,4] with [5,4] and the global handover cost ΣC is recalculated. ij =23, satisfying the constraints. The final offspring sequence achieves significantly reduced switching costs and a valid structure, making it suitable for subsequent multi-objective fitness evaluation; S6.4: Based on the goals of equipment utilization and production line cycle balance, design a multi-objective fitness evaluation function to perform multi-dimensional evaluation on each chromosome individual, including total switching cost, equipment idle time, and cycle fluctuation coefficient, in order to generate the multi-objective fitness vector of the individual; S6.5: A non-dominated sorting mechanism is used to perform hierarchical sorting of individuals in the population, and a crowding calculation mechanism is introduced to maintain population diversity, forming a Pareto front candidate solution set, which provides an optimized result set for subsequent multi-objective decision-making. S6.6: After each generation of genetic iterations, an elite retention strategy is executed based on the Pareto front distribution of the current population to retain the best non-dominated individuals in the next generation of the population, so as to prevent the loss of high-quality solutions and improve the stability of the algorithm. After receiving the Pareto front distribution data of the current population output by the non-dominated sorting mechanism, the elite retention strategy determination module (parameters: non-dominated level, crowding distance vector) is used to identify the set of individuals in the population that are located in the optimal non-dominated layer. Furthermore, a genetic stability assessment algorithm (parameters: fitness vector, historical retention set) is used to detect the fitness change trend of the best non-dominant individual and obtain the stability judgment result, which is then used to select retention targets. Furthermore, through a multi-objective coverage analysis method (parameters: total switching cost target value, equipment utilization target value, production line cycle time balance target value), the coverage range of candidate retained individuals in the objective function space is calculated, and a coverage index is generated; Furthermore, by performing capacity allocation calculations using the retention set capacity constraint optimization formula, the number of elite individuals retained in each non-dominated layer is dynamically adjusted. The formula is as follows:

[0024] Where N is the number of individuals retained in the current layer, C is the total allowed capacity, d is the average crowding distance of individuals in this layer, and T is the total crowding distance of the entire layer; By using elite replication and index mapping, the sorted best non-dominated individuals are inserted into the initial data structure of the next generation population, thus achieving the intergenerational inheritance of high-quality solutions. The above strategy transforms the Pareto front distribution results from the previous step into a stable and diverse next-generation initial population structure, thereby improving the global stability and convergence performance of the production scheduling optimization process. For example, in a mixed-flow production scheduling optimization scenario, assume the current population contains 50 individuals, with the first layer of the Pareto front containing 12 non-dominated solution individuals. The capacity constraint is set to C=20, the average crowding distance of individuals in this layer is d=0.08, and the total crowding distance of the entire layer is T=0.96. The capacity allocation formula is used to calculate: We obtained N=2, meaning that this layer retains two individuals with a large crowding distance, and the remaining individuals are allocated to other non-dominated layers according to the coverage index. During the elite replication process, the integer-coded chromosome sequences, fitness vectors, and historical line-switching data of the two selected individuals are copied to the next-generation initialization matrix. Verification results show that the total switching cost is significantly reduced in subsequent iterations, equipment utilization remains high, and the fluctuation range of production line cycle time is significantly reduced.

[0025] Step S7: Generate candidate production scheduling sequences based on the Pareto front solution set, and select the optimal production scheduling sequence by combining the real-time workshop load rate. The selection process uses the fuzzy entropy weighting method to perform TOPSIS ranking decision on the multi-objective optimization results. Specifically, it includes: S7.1: Based on the Pareto front solution set output by the production planning module, extract the multi-objective optimization index vector corresponding to each candidate production sequence, including total switching cost, equipment utilization rate and production line beat balance degree, to construct a multi-attribute decision matrix and provide input for subsequent multi-criteria decision-making; In this sub-step, the Pareto front solution set output by the improved NSGA-II algorithm of the production planning module is used as the input condition, each solution in the Pareto front solution set corresponds to a candidate production sequence optimized by multi-objective optimization, and contains three core optimization indicators: total switching cost, equipment utilization rate and production line beat balance degree; The solution set traversal method (parameters: solution index range of the Pareto front solution set) is used to realize the retrieval of the optimization index data structure of each candidate production sequence from the current solution set; Further, through the index analysis algorithm (parameters: single solution index field identifier), the total switching cost, equipment utilization rate and beat balance degree in the multi-objective optimization result are extracted as scalar values, and are associated with the unique sequence ID of the candidate sequence to form a sequence-index mapping relationship data; Further, the matrix construction method (parameters: number of rows is the number of candidate sequences, number of columns is the dimension number of optimization indicators) is used to realize the combination of three-dimensional indicators of all candidate sequences into a multi-attribute decision matrix M, and to ensure that each row in the matrix corresponds to a candidate sequence and each column corresponds to an optimization indicator type; Further, the index sorting algorithm (parameters: sorted in ascending order according to sequence ID) is used to realize the sorting of the row order of the decision matrix according to the natural ID of the candidate sequence, so as to ensure the consistency of sequence positioning in the subsequent weight allocation and normalization processing steps; Through the above matrix construction processing method, the multi-objective optimization result in the Pareto front solution set is converted into a structured multi-attribute decision matrix, which provides a unified data input structure for subsequent fuzzy entropy weight method and TOPSIS sorting operation; Exemplarily, in a certain mixed flow production scene, the Pareto frontier solution set contains 5 candidate scheduling sequences, and the total switching cost (unit: minute) of each sequence is 12.4, 15.1, 10.7, 13.5, and 9.9 respectively, and the equipment utilization rate (unit: %) corresponds to 0.82, 0.79, 0.85, 0.81, and 0.88 respectively, and the line beat balance degree (unit: reciprocal of beat variance) is 0.56, 0.52, 0.59, 0.54, and 0.61 respectively. The solution set traversal method is used to extract the above three indicators one by one, and a sequence-index mapping relationship is generated, for example, sequence ID=3 corresponds to index vector [10.7, 0.85, 0.59]. The index vectors of the 5 sequences are assembled into a 5x3 decision matrix M: the row index is [1…5], and the column index represents the total switching cost, the equipment utilization rate, and the beat balance degree respectively. In this embodiment, the first row of the matrix M is [12.4, 0.82, 0.56], and the last row is [9.9, 0.88, 0.61]. This matrix will be normalized in S7.2, and then weighted by the fuzzy entropy weight method, and finally support the TOPSIS comprehensive evaluation calculation to realize the full-link support of candidate scheduling sequences from the index set to the ranking decision; S7.2: Perform normalization processing on the multi-attribute decision matrix, and use the maximum-minimum normalization method to perform dimensionless conversion on each optimization index to eliminate the influence of the dimension difference of different indexes on the subsequent weight distribution and distance calculation, and obtain a standardized decision matrix; S7.3: Weight each optimization index in the standardized decision matrix based on the fuzzy entropy weight method, first calculate the information entropy of each index, then determine the objective weight of each index according to the information entropy to reflect the relative importance of each optimization target under the current workshop state, and obtain a fuzzy entropy weight vector; S7.4: Weight the standardized decision matrix based on the fuzzy entropy weight vector, calculate the positive ideal solution and the negative ideal solution after weighting, which represent the comprehensive attribute values of the optimal and worst scheduling schemes respectively, and provide a distance benchmark for subsequent TOPSIS ranking; S7.5: Calculate the Euclidean distance of each candidate scheduling sequence from the positive ideal solution and the negative ideal solution, and construct a relative closeness index based on the distance ratio, which is used to measure the similarity of each candidate sequence to the optimal scheduling scheme, and generate a TOPSIS comprehensive evaluation value of the candidate sequence; S7.6: Weight and correct the TOPSIS comprehensive evaluation value of the candidate scheduling sequence in combination with the real-time load rate of the workshop, which is calculated by the equipment occupancy rate, work-in-process quantity, and planned completion rate collected by the execution control system, and is used to dynamically adjust the priority of the scheduling strategy; S7.7: Sort the candidate scheduling sequences according to the weighted corrected TOPSIS comprehensive evaluation values, select the sequence with the highest evaluation value as the final recommended scheduling scheme, and output the scheme to the execution control system to drive the production equipment for scheduling execution; Under the input condition of obtaining the weighted corrected TOPSIS comprehensive evaluation value set and the corresponding candidate scheduling sequence mapping table, a sorting algorithm (parameters: evaluation value set and sequence identifier) is used to realize the sorting and arrangement of multiple candidate sequences according to the comprehensive index level; Further, through a stable sorting strategy (parameters: evaluation value, sequence generation timestamp), the generation order of sequences with the same evaluation value is maintained in order to ensure the operation traceability of the scheduling process and obtain a stable candidate sequence sorting list; Further, through a sequence selection method (parameters: sorting list, sequence priority threshold), the sequence with the highest evaluation value in the sorting list is selected, and the final recommended scheduling scheme is generated; Further, a data encapsulation interface (parameters: scheduling scheme data, scheduling instruction format) is used to convert the recommended scheme into a scheduling instruction structure executable by the equipment, and a complete scheduling instruction set is generated; Through an industrial control protocol sending module (parameters: scheduling instruction set, execution control system communication address), the result of the previous step is converted into scheduling task data that can be received by the execution control system, realizing the expected technical effect of automatic driving and scheduling execution of the production equipment; For example, in a certain automobile floor mat mixed flow production workshop, the candidate scheduling sequence set is 10, and after fuzzy entropy weight method and TOPSIS sorting calculation, the weighted corrected evaluation value of each sequence is in the range of 0.42 to 0.87. When input into the sorting module, the sequence number and evaluation value are arranged in descending order, and the sequence number with the highest evaluation value is S6, corresponding to the evaluation value of 0.87. The sorting module uses stable sorting, and in the case of the same evaluation value of S3 and S8, the original generation timestamp order is maintained. The selection module sets the priority threshold to 0.85, so that S6 is selected as the final scheme. The data encapsulation interface converts the work order sequence of S6 into a structured scheduling instruction set containing process number, device instruction code and execution time window, for example, the device instruction code of the first work order is M01, and the execution time window is 08:00-08:15, the device instruction code of the second work order is M03, and the execution time window is 08:15-08:30. The industrial control protocol sending module sends the instruction set to the execution control system address 192.168.1.105 through the TCP / IP interface, and the device receives and schedules according to the S6 scheme in turn, the production process equipment switching cost is greatly reduced, the production line beat fluctuation is significantly reduced, and the equipment utilization rate is improved.

[0026] The step S8: record the error value of the actual switching time consumption and the predicted switching cost after each line change is completed, update the weighting coefficients of the dynamic resource switching cost evaluation model using the online incremental learning algorithm, and the updating process is based on the error gradient descent method to perform parameter iterative correction. Specifically, it includes: S8.1: collect the actual switching time consumption after each line change is completed, calculate the actual switching time consumption based on the device line change start timestamp and line change end timestamp to obtain real-time performance feedback data of the line change operation; S8.2: calculate the error between the predicted value and the actual time consumption according to the predicted switching cost output by the scheduling planning module before the line change, generate the error residual, and use it to quantify the prediction deviation of the dynamic resource switching cost model; S8.3: execute the online incremental learning algorithm based on the error residual, use the weighted least squares method to update the process difference weight and device response cost weight in the dynamic resource switching cost evaluation model, and obtain the updated weighting coefficient vector; S8.4: correct the gradient descent direction of the updated weighting coefficient vector, perform parameter iterative correction based on the error gradient, adjust the convergence using the learning rate, and generate the optimized dynamic switching cost model parameter set; S8.5: write the optimized dynamic switching cost model parameter set into the resource switching cost evaluation model, replace the original parameter set, and use the updated model to participate in the next round of scheduling sequence optimization calculation to realize the continuous adaptive optimization of the scheduling strategy.

[0027] So far, the technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the drawings, but those skilled in the art will readily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to related technical features without departing from the principles of the present application, and the technical solutions after these changes or replacements will fall within the protection scope of the present application.

[0028] The above description is only the preferred embodiments of the present application and is not intended to limit the present application; for those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A production scheduling method for mixed-flow production of car floor mats based on customer order data, characterized in that: Includes the following steps: S1: Obtain the task set of car floor mat orders to be scheduled for production. The task set includes the style identifier, process parameter set and production quantity corresponding to each order. S2: Collect real-time status data of production equipment, including the current processing style identifier, the cumulative number of completed work orders, equipment temperature control parameters, wear status parameters, and the most recent line change timestamp; S3: Construct a process similarity matrix between styles based on historical process parameter data, and use the Euclidean distance algorithm to calculate the parameter differences of different styles in the cutting, sewing and pressing processes to generate static process difference index; S4: Calculate the dynamic device response cost based on the real-time status data, including calculating the temperature rise compensation time based on the device temperature control parameters, calculating the tool calibration cycle based on the wear status parameters, and calculating the restart delay time based on the most recent line change timestamp. S5: Construct a dynamic resource switching cost evaluation model, and weight and fuse the state process difference index with the dynamic equipment response cost to generate dynamic switching cost; S6: An improved NSGA-II multi-objective genetic algorithm is used to optimize the production scheduling sequence. When initializing the population, the process similarity priority aggregation rule is adopted. The switching cost guidance strategy is embedded in the crossover and mutation operation. The multiple optimization objectives are to minimize the total switching cost, maximize the equipment utilization rate and balance the production line cycle time. S7: Generate candidate production scheduling sequences based on the Pareto front solution set, and select the optimal production scheduling sequence by combining the real-time load rate of the workshop.

2. The method for mixed-flow production scheduling of car floor mats based on customer order data according to claim 1, characterized in that, The process following step S7 also includes: S8: After each line switch is completed, record the error between the actual switch time and the predicted switch cost, and use an online incremental learning algorithm to update the weighting coefficients of the dynamic resource switch cost evaluation model.

3. The method for mixed-flow production scheduling of car floor mats based on customer order data according to claim 1, characterized in that, Step S1 specifically includes: Obtain a set of car floor mat orders to be scheduled from the enterprise resource planning system. The set of orders includes multiple order entries, and each order entry contains a corresponding style identifier, a set of process parameters, and a production quantity. The process parameter set is standardized to form structured process parameter data in a unified format; Based on the structured process parameter data, a process feature vector corresponding to each order is generated; The style identifier, the process feature vector, and the production quantity are encapsulated to generate an order task data structure. The data structure is stored in key-value pair format, where the key is the style identifier and the value is a composite data object containing the process feature vector and the production quantity. The order task data structure is transmitted to the process similarity analysis unit of the production scheduling module.

4. The method for mixed-flow production scheduling of car floor mats based on customer order data according to claim 3, characterized in that, The style identifier is used to uniquely identify different types of car floor mat products. The set of process parameters includes material type, sewing complexity, and edging method parameters. The production quantity represents the total number of car floor mat styles corresponding to each order.

5. The method for mixed-flow production scheduling of car floor mats based on customer order data according to claim 1, characterized in that, Step S2 specifically includes: Based on the Industrial Internet of Things (IIoT) protocol, the current processing style identifier of the production equipment is collected to determine the current processing style of the car floor mat and obtain the current processing style parameters. The cumulative number of work orders completed by the production equipment is statistically processed to obtain the current production load status of the equipment and generate the cumulative task completion index of the equipment. Based on the temperature control parameters of the equipment collected by the temperature sensor, the temperature data is filtered to obtain the current thermal state characteristics of the equipment. The wear status parameters of the equipment are collected and normalized, and the wear degree feature values ​​of the equipment are extracted based on vibration sensors and current signals. The most recent line change timestamp is extracted from the equipment control system logs, and the timestamp data is processed to standardize the format to obtain the time of the most recent line change of the equipment.

6. The method for mixed-flow production scheduling of car floor mats based on customer order data according to claim 1, characterized in that, Step S3 specifically includes: Obtain a set of historical order data for car floor mat production, including cutting parameters, sewing parameters, and pressing parameters for each style; and construct a style process feature vector matrix based on the set of historical order data. The process feature vector matrix of the style is normalized, and the minimum-maximum normalization method is used to perform interval mapping on the parameters of each dimension to obtain the normalized process feature vector matrix. Based on the standardized process feature vector matrix, the Euclidean distance algorithm is used to calculate the pairwise distance between the process parameter vectors of any two styles, generating the original values ​​of the process difference between each pair of styles. The original values ​​of process differences between all pairs of styles are normalized, and the distance values ​​are aligned using a standardization method to generate a standardized process difference matrix. Based on the standardized process difference matrix, a process similarity matrix between styles is generated.

7. The method for mixed-flow production scheduling of car floor mats based on customer order data according to claim 6, characterized in that, Step S3 further includes establishing a process feature matrix based on historical process parameters according to cutting, sewing, and pressing parameters, using minimum-maximum standardization and Z-score standardization to standardize and quantify the process differences of multiple varieties, and generating a process similarity matrix by subtracting the difference value from 1.

8. The method for mixed-flow production scheduling of car floor mats based on customer order data according to claim 1, characterized in that, Step S4 specifically includes: Based on the equipment temperature control parameters, a temperature compensation model is executed to calculate the time required for the equipment to recover from a shutdown state to the target processing temperature. Perform wear compensation analysis on the equipment wear condition parameters to obtain the tool calibration cycle required for the current equipment after line changeover; The restart delay is calculated based on the time difference between the most recent line change timestamp and the current timestamp to obtain the restart delay time required for the device to switch from the previous processing task to the current task. The temperature rise compensation time, the tool calibration cycle and the restart delay time are weighted and summed to generate the dynamic equipment response cost when the equipment switches from the current processing style to the next target style. The dynamic equipment response cost index is bound to the equipment's current processing style identifier to generate an equipment status-cost mapping table.

9. The method for mixed-flow production scheduling of car floor mats based on customer order data according to claim 1, characterized in that, Step S5 specifically includes: Based on historical changeover efficiency data, a process-equipment response cost mapping table is constructed. The least squares method is used to perform regression modeling on the correlation between historical changeover time and process variability and equipment status parameters to obtain an initial set of weight coefficients. The static process difference index and dynamic equipment response cost are normalized. The input data of different dimensions are uniformly transformed using a standardization method to obtain the normalized static process difference index and dynamic equipment response cost. Based on the normalized static process difference index and dynamic equipment response cost, a weighted fusion operation is performed to generate a preliminary dynamic switching cost matrix. The preliminary dynamic switching cost matrix is ​​nonlinearly corrected by introducing a Sigmoid function to amplify the switching cases with high differences or high equipment response costs, thereby generating a corrected dynamic switching cost matrix. Based on the real-time input of current equipment status and order process parameters, the corrected dynamic switching cost matrix is ​​updated online. The latest line switching records are fused using a sliding window mechanism to generate the final dynamic resource switching cost evaluation model output.

10. The method for scheduling multi-style mixed-flow production of car floor mats based on customer order data according to claim 9, characterized in that, Step S5 further includes, based on the weighted fusion of static process difference and dynamic response cost, using the Sigmoid function to nonlinearly amplify the switching cost that is higher than the mean, and ensuring the model's adaptive adjustment to periodic fluctuations and high-cost switching through sliding weighting, exponential smoothing, and anomaly penalty.

Images