A steel cutting production scheduling method and device based on dynamic risk assessment

Abstract

The application provides a steel cutting production scheduling method and device based on dynamic risk assessment, relates to the field of steel processing pre-production management, and realizes data structured expression by constructing a multi-dimensional numerical matrix through real-time acquisition of order, equipment and inventory data; candidate production scheduling sequences are generated by using a Monte Carlo method, combined with dynamic assessment of offcut loss and delay risk, and iterative optimization is performed by combining simulated annealing and an adaptive threshold mechanism; the optimal production scheduling scheme is determined within an error allowable range through probability similarity screening and regression fitting. The method effectively balances production efficiency and resource utilization, reduces cutting loss and delivery risk, and is suitable for intelligent production scheduling decision-making in complex manufacturing scenarios.

Description

Technical Field

[0001] This invention relates to the fields of industrial software and artificial intelligence, and in particular to a method and apparatus for steel cutting and scheduling based on dynamic risk assessment. Background Technology

[0002] In the steel shearing and processing field, the quality of production scheduling plans has a decisive impact on material utilization, production efficiency, and order fulfillment capabilities. However, existing production scheduling technologies have significant limitations and are difficult to adapt to the diverse, small-batch, and highly flexible production needs of modern steel processing enterprises.

[0003] Traditional production scheduling systems have significant deficiencies in data processing, failing to effectively integrate and matrix-process key data such as order parameter sets, equipment status vectors, inventory balance distribution, and material replacement rule tables. Information such as order geometric parameters (length, width, thickness, quantity), priority weights, equipment real-time throughput, task load, maintenance status, wear index, and remaining dimensions, material types, storage locations, and replacement rules of inventory coils are often managed in a scattered manner, resulting in low data utilization and an inability to provide a comprehensive and accurate basis for production scheduling decisions.

[0004] At the algorithmic level, mainstream scheduling methods lack efficient search capabilities within the constrained space, such as the Monte Carlo method, when generating candidate scheduling sequences. Furthermore, their comprehensive assessment of scrap material loss and delay risks is not precise enough. Simultaneously, during sequence optimization, it is difficult to effectively combine simulated annealing algorithms and adaptive loss threshold mechanisms for crossover and mutation operations, easily leading to local optima and failing to achieve global optimization.

[0005] In terms of dynamic response capabilities, existing systems struggle to adjust production scheduling plans in real time based on changes in production scheduling data. When order priorities, equipment status, or inventory levels change, the system cannot update the multidimensional numerical matrix and regenerate the optimal production sequence in a timely manner, leading to a disconnect between the production scheduling plan and the actual production scenario, which affects production efficiency and the timeliness of order delivery.

[0006] In addition, existing technologies have shortcomings in the screening and fitting of production scheduling sequences. They lack an effective mechanism to screen out redundant sequences through probability similarity and to determine the optimal solution within the allowable error range by using regression fitting, which makes it necessary to improve the accuracy and reliability of production scheduling results.

[0007] The combined effect of these technological bottlenecks has led to significant pain points in the steel processing industry, including low average material utilization, insufficient on-time order delivery, and excessively high equipment idle time. Overcoming these technological limitations and building an intelligent scheduling system with multi-source data matrix processing and dynamic global optimization capabilities has become an urgent need to drive the industry's transformation and upgrading. Summary of the Invention

[0008] The main objective of this invention is to overcome the shortcomings of existing steel cutting and scheduling technologies, such as insufficient data integration, single optimization strategy, weak global optimization capability, and poor dynamic convergence. This invention proposes a steel cutting and scheduling method and apparatus based on dynamic risk assessment. By converting scheduling data into a multi-dimensional numerical matrix, integrating the Monte Carlo method to generate candidate sequences, and combining simulated annealing and adaptive loss threshold mechanisms for iterative optimization, the dynamic convergence and optimal determination of the scheduling scheme are achieved through similarity screening and regression fitting.

[0009] The present invention adopts the following technical solution:

[0010] A steel cutting and scheduling method based on dynamic risk assessment includes the following steps:

[0011] S1 acquires production scheduling data in real time, which includes order parameter sets, equipment status vectors, and inventory balances; based on the production scheduling data, it performs numerical processing to convert it into a multi-dimensional numerical matrix, which includes an order feature matrix, an equipment capacity matrix, and a surplus inventory matrix;

[0012] S2, based on the Monte Carlo method, performs calculations on the order feature matrix, the equipment capacity matrix, and the surplus material inventory matrix within the constraint space to generate a candidate production scheduling sequence that satisfies the order priority and equipment constraints. For each order in the candidate production scheduling sequence, the scrap material loss and delay risk are dynamically calculated to obtain a comprehensive loss assessment value for each order.

[0013] S3 combines simulated annealing algorithm and adaptive loss threshold mechanism to perform crossover and mutation operations on the candidate production scheduling sequence; redundant sequences are eliminated by similarity screening, and the potential optimal production scheduling scheme within the error allowable range is determined by regression fitting; the sequence crossover, mutation, screening and fitting process is executed iteratively until the optimal production scheduling scheme that meets the preset conditions is obtained.

[0014] The production scheduling data includes the order parameter set. Device state vector Inventory balance distribution and Material Replacement Rules Table ;in, This represents the total number of orders currently awaiting production. For the first The data object of each order; For the first Status data of the device; This represents the total number of different types of stocked coils. For the first Inventory data for roll-type materials;

[0015] Geometric parameters of the order are extracted based on the order parameter set, including the length, width, thickness, and quantity of steel required in the order. Order priority weights are labeled and dynamically adjusted based on delivery urgency and customer level. The constructed order feature matrix is ​​as follows:

[0016] ;

[0017] in, For order number, For the required length of steel in the order, For the required width of steel in the order, The quantity of steel required for the order. Priority weight;

[0018] Based on the device state vector, the real-time throughput, current task load, and maintenance status of the device are obtained. The wear index of the device is calculated, and the remaining service life is predicted by combining historical data. The constructed device capability matrix is ​​as follows:

[0019] ;

[0020] in, For equipment number, For order length, For the maximum processing length, This represents the current task queue length.

[0021] Based on the remaining inventory distribution records of the remaining dimensions, material types, and storage locations of rolls of material in each warehouse, and by marking the recommended use cases and replacement rules for the remaining materials, the following remaining material inventory matrix is ​​constructed:

[0022] ;

[0023] in, For the roll material number, For the remaining dimensions, This is the number of the replaceable material.

[0024] Step S2 specifically includes the following:

[0025] S21 sets constraint space parameters, which include order priority sorting rules, equipment operating thresholds, and production resource constraints; based on the Monte Carlo method, random sampling operations are performed on the order feature matrix, equipment capacity matrix, and surplus inventory matrix to generate an initial production scheduling sequence within the constraint space;

[0026] S22 performs compliance checks on the initial production schedule based on order priority sorting rules and equipment operating thresholds, removes sequences that do not meet the constraints, and retains candidate production schedules that meet the requirements.

[0027] S23 calculates the scrap material loss value for each order in the candidate production scheduling sequence by combining the material requirement parameters in the surplus material inventory matrix and the order feature matrix; and calculates the delay risk coefficient for each order based on the equipment load parameters in the equipment capacity matrix and the delivery date parameters in the order feature matrix.

[0028] S24 integrates the scrap material loss value with the delay risk coefficient according to preset weights to obtain the comprehensive loss assessment value for each order.

[0029] In step S21, the Monte Carlo search space is defined as follows:

[0030] ;

[0031] in, This represents the Monte Carlo search space, which is the set of candidate production sequences. Indicates the first One candidate production sequence, Represents the full permutation of the order set. Indicates the index of the currently generated sequence. This indicates the maximum number of sequences that can be generated.

[0032] In step S24, the method for calculating the comprehensive loss assessment value is as follows:

[0033] From the surplus inventory matrix Select rolls that meet the criteria: ;

[0034] in, A set of rolls that meet the conditions, For the first One roll of material, This refers to the remaining length of the coil. The alternative material number for the coil material is [the number of alternative materials for the coil material]. For the required length of steel in the order, For the material type of the order, For material replacement rule functions;

[0035] Selecting the optimal roll material using a value function:

[0036] ;

[0037] in, This is the optimal roll material selection function. To find the roll material object with the lowest overall cost from the candidate list, For size matching, This refers to the transportation distance from the warehouse to the processing equipment. To calculate transportation costs;

[0038] For successfully matched orders Calculate the waste value of scrap materials:

[0039] ;

[0040] in, To select the length of the roll, It is the first The length of the order to be cut in one go. This refers to the number of orders to be cut from this roll of material;

[0041] Processing start time prediction:

[0042] ;

[0043] in, The current system time. For device queue load rate, For equipment Real-time throughput, For equipment Task queue length, As a unit time base;

[0044] Delay risk coefficient calculation:

[0045] ;

[0046] in, For order processing time, This refers to the order delivery deadline;

[0047] Calculate the first Comprehensive loss assessment value of the next iteration :

[0048] ;

[0049] in, For material consumption items, For time consumption items, The delay penalty coefficient;

[0050] ;

[0051] in, Indicates the number of urgent orders. This represents the total number of orders.

[0052] Step S3 specifically includes the following:

[0053] S31 initializes the simulated annealing parameters, including the initial temperature, cooling coefficient, and termination temperature, and sets the initial reference value for the adaptive loss threshold.

[0054] S32 selects the candidate production scheduling sequence with the better comprehensive loss assessment value as the parent generation, fixes the position of high priority orders, performs a region cross operation on the remaining orders to generate the child generation sequence; and randomly swaps the positions of the child generation sequences obtained by the region cross to generate a new candidate production scheduling sequence.

[0055] S33 combines an adaptive loss threshold mechanism with a simulated annealing algorithm to evaluate and accept new candidate production scheduling sequences. It then filters the candidate production scheduling sequences by calculating probability similarity and retains those that meet the requirements. Based on the steel length error range, it performs scatter-point fitting regression calculation on the filtered sequences to determine the potential optimal production scheduling scheme.

[0056] S34. Lower the simulated annealing temperature and repeat S32 to S33 until the termination condition is met, and finally determine the optimal production schedule. If the requirements are not met, return to S31 to readjust the parameters and iterate.

[0057] Step S32 includes:

[0058] Two candidate production scheduling sequences with better overall loss assessment values ​​are used as parent sequences. The positions of orders with priority level ≥ 4 are fixed, and regional crossover and random position swapping are performed on the remaining orders to generate child sequences as new candidate production scheduling sequences.

[0059] Region crossover operation: Randomly select the crossover interval [start, end] from the two parent sequences, where start is the starting position order of the parent sequence and end is the ending position order of the parent sequence. The child sequence directly inherits the interval segment of one of the parent sequences, and the remaining positions are filled with orders with priority ≤ 3 according to the order of the other parent sequence.

[0060] Random position swap: Perform n swaps for each subsequence, and only swap the order positions with priority ≤ 3, n = 1, 2, 3, ... 5;

[0061] Combining simulated annealing with a probabilistic acceptance of suboptimal solutions, specifically:

[0062] Annealing temperature update formula:

[0063] ;

[0064] in, Let be the annealing temperature at the t-th iteration. The initial temperature for the production scheduling process. For decay rate, This represents the number of iterations.

[0065] The probability of accepting a suboptimal solution is determined using the Metropolis criterion:

[0066] ;

[0067] in, The probability of accepting a suboptimal solution, This is the difference between the loss assessment values ​​of the new solution and the current solution.

[0068] Set an adaptive loss threshold and gradually tighten the standard during iterations, specifically as follows:

[0069] Initial loss threshold ,in This represents the historical average loss rate.

[0070] Loss threshold varies with iteration number Adaptive tightening:

[0071] , ;

[0072] Set a dynamic acceptance probability, allowing the algorithm to accept inferior solutions with a certain probability. > To avoid getting trapped in local optima, the probability of accepting inferior solutions is gradually reduced with each iteration, eventually stabilizing at the optimal solution:

[0073] ;

[0074] in, This represents the total loss value for the current iteration. The historical best loss value, temperature parameter With the number of iterations attenuation: , This is the initial temperature.

[0075] In step S33, candidate production scheduling sequences are screened based on probability similarity, specifically including:

[0076] Collect order information from candidate production schedules, including material width, length, and quantity;

[0077] Generate all possible production scheduling combinations based on the material length, quantity, and width;

[0078] For each candidate production combination, normalization is performed, and the Euclidean norm of the actual demand is calculated:

[0079] If two candidate production scheduling combinations With candidate production combinations Production scheduling is considered successful if the following conditions are met:

[0080] ;

[0081] in, Candidate production combinations The nth cutting group in Candidate production combinations The nth cutting group in For the order demand feature vector, the symbol is... To normalize the three-dimensional feature vectors of candidate production combinations, the symbols... Calculate the Euclidean norm. This represents the normalized order demand feature vector and the normalized candidate production scheduling combination scheme. The difference vector between the feature vectors;

[0082] Set similarity threshold :

[0083] ;

[0084] in, It is a dynamic adjustment factor. ;

[0085] If similarity threshold Greater than probability similarity If the two candidate production combinations are not found, they will be eliminated; otherwise, they will be retained.

[0086] The optimal production scheduling scheme is obtained by scatter-point fitting regression calculation based on the steel length error range:

[0087] Scatter fitting is performed on the set of candidate production scheduling sequences after screening, taking into account a 10% length error range of the steel after trimming, to obtain the scatter fitting set;

[0088]

[0089] in, =Theoretical length of steel in the order , =Actual cutting length deviation rate ;

[0090] Length error range constraint:

[0091] ;

[0092] The least squares method is used to determine the optimal production scheduling plan, which minimizes the error between the actual production scheduling result and the ideal production scheduling plan. Specifically, this includes:

[0093] a) Construct the input matrix D, which contains the theoretical length of steel in the order. Deviation rate from actual cutting length ;

[0094] b) Solve the weighted least squares equation to obtain the bias prediction model;

[0095] c) Select the candidate production sequence that minimizes the sum of the absolute values ​​of the prediction deviations as the optimal production scheduling scheme.

[0096] The formula for calculating the length deviation rate is as follows:

[0097] Length deviation rate ;

[0098] in, For the required length of steel in the order, This is the actual cutting length.

[0099] A steel shearing and scheduling device based on dynamic convergence, comprising:

[0100] The data acquisition and processing module acquires production scheduling data in real time, including order parameter sets, equipment status vectors, and inventory balances. Based on the production scheduling data, it performs numerical processing to convert it into a multi-dimensional numerical matrix, including an order feature matrix, an equipment capacity matrix, and a surplus inventory matrix.

[0101] The dynamic risk assessment module, based on the Monte Carlo method, performs calculations on the order feature matrix, the equipment capacity matrix, and the surplus material inventory matrix within the constraint space to generate candidate production schedules that meet the order priority and equipment constraints. For each order in the candidate production schedule, the module dynamically calculates the scrap material loss and delay risk, and comprehensively obtains the overall loss assessment value for each order.

[0102] The production scheduling module combines simulated annealing algorithm and adaptive loss threshold mechanism to perform crossover and mutation operations on the candidate production scheduling sequences; it eliminates redundant sequences by similarity screening and uses regression fitting to determine the potential optimal production scheduling scheme within the allowable error range; it iteratively executes the sequence crossover, mutation, screening and fitting process until the optimal production scheduling scheme that meets the preset conditions is obtained.

[0103] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects:

[0104] (1) By constructing an order feature matrix, equipment capacity matrix and surplus inventory matrix, the data is structured and integrated. Combined with the Monte Carlo method, candidate production scheduling sequences are generated and the comprehensive loss assessment value is dynamically calculated, which significantly improves the material utilization rate. The adaptive loss threshold tightens with iteration, effectively avoiding the premature convergence problem caused by the fixed threshold.

[0105] (2) Optimize equipment allocation based on parameters such as real-time load and wear index in the equipment capability matrix to achieve load balancing and avoid overloading of single equipment. Combine the equipment capability to dynamically adjust the processing scheme and extend the service life of the equipment.

[0106] (3) By adopting a mutation strategy of regional crossover and random position exchange, combined with the probability acceptance mechanism of simulated annealing algorithm, local optima are effectively avoided and the success rate of global optimization is improved.

[0107] (4) Redundant sequences are eliminated by probability similarity screening, and the optimal solution is determined within the allowable error range by regression fitting, thereby improving the accuracy and feasibility of the production scheduling solution; the iterative mechanism ensures continuous optimization of the solution and further reduces the risk of loss and delay. Attached Figure Description

[0108] Figure 1 This is the main flowchart of the method of the present invention;

[0109] Figure 2 This is a schematic diagram of the basic framework of the present invention;

[0110] Figure 3 This is a schematic diagram of the data acquisition feature matrix according to an embodiment of the present invention;

[0111] Figure 4 This is a schematic diagram illustrating the annealing mechanism and sequence crossover sorting principle of an embodiment of the present invention.

[0112] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Detailed Implementation

[0113] The present invention will be further described below through specific embodiments.

[0114] This invention is an intelligent steel production scheduling method that integrates multi-source data matrix processing, Monte Carlo global optimization, and dynamic feedback mechanisms. It is mainly applied to the intelligent production management system of steel processing enterprises to achieve synergistic optimization of material utilization and order delivery capability.

[0115] See Figures 1 to 4 This invention proposes a steel cutting scheduling method based on dynamic risk assessment, comprising:

[0116] S1 acquires production scheduling data in real time, which includes order parameter sets, equipment status vectors, and inventory balances; based on the production scheduling data, it performs numerical processing to convert it into a multi-dimensional numerical matrix, which includes an order feature matrix, an equipment capacity matrix, and a surplus inventory matrix.

[0117] In this step, the basic structure for data acquisition and matrix construction can be found in [link to relevant documentation]. Figure 3The system obtains order parameter sets (including order number, length, width, quantity, priority weight), equipment status vectors (including equipment number, maximum processing length, current task queue length, wear index), and surplus material inventory distribution (including coil number, remaining length, material type, and replaceable material number) and material replacement rule table in real time through the data interface.

[0118] At the data layer, a multi-source data matrix processing module is used to transform production scheduling data into a multi-dimensional numerical matrix. Specifically, this involves obtaining the order parameter set through the data interface. Device state vector Inventory balance distribution and Material Replacement Rules Table ;in, This represents the total number of orders currently awaiting production. For the first The data object of each order; This represents the total number of available production equipment. For the first Status data of the device; This represents the total number of different types of stocked coils. For the first Inventory data for roll-type materials.

[0119] Geometric parameters of the order are extracted based on the order parameter set, including the length, width, thickness, and quantity of steel required in the order. Order priority weights are labeled and dynamically adjusted based on delivery urgency and customer level. The constructed order feature matrix is ​​as follows:

[0120] ;

[0121] in, For order number, For the required length of steel in the order, For the required width of steel in the order, The quantity of steel required for the order. Priority weights (levels 1-5);

[0122] Based on the device state vector, the real-time throughput, current task load, and maintenance status of the device are obtained. The wear index of the device is calculated, and the remaining service life is predicted by combining historical data. The constructed device capability matrix is ​​as follows:

[0123] ;

[0124] in, For equipment number, For order length, For the maximum processing length, This represents the current task queue length.

[0125] Based on the remaining inventory distribution records of the remaining dimensions, material types, and storage locations of rolls of material in each warehouse, and by marking the recommended use cases and replacement rules for the remaining materials, the following remaining material inventory matrix is ​​constructed:

[0126] ;

[0127] in, For the roll material number, The length is in ascending order. This is the number of the replaceable material.

[0128] S2, based on the Monte Carlo method, performs calculations on the order feature matrix, the equipment capacity matrix, and the surplus material inventory matrix within a constraint space to generate candidate production scheduling sequences that satisfy order priority and equipment constraints. For each order in the candidate production scheduling sequence, it dynamically calculates scrap material loss and delay risk, and comprehensively obtains the overall loss assessment value for each order. Specifically, this includes the following:

[0129] S21 sets constraint space parameters, which include order priority sorting rules, equipment operating thresholds, and production resource constraints; based on the Monte Carlo method, random sampling operations are performed on the order feature matrix, equipment capacity matrix, and surplus inventory matrix to generate an initial production scheduling sequence within the constraint space.

[0130] In this step, the feature matrix is ​​received. , , Generate candidate production schedules Prioritize satisfying the preconditions of high-priority orders; define the Monte Carlo search space:

[0131] ;

[0132] in, This represents the Monte Carlo search space (the set of all possible production sequences). Indicates the first One candidate production sequence, Represents the full permutation of the order set. Indicates the index of the currently generated sequence. This indicates the maximum number of sequences that can be generated.

[0133] S22 performs compliance checks on the initial production schedule based on order priority sorting rules and equipment operating thresholds, removes sequences that do not meet the constraints, and retains candidate production schedules that meet the requirements.

[0134] S23 calculates the scrap material loss value for each order in the candidate production scheduling sequence by combining the material requirement parameters in the surplus material inventory matrix and the order feature matrix; and calculates the delay risk coefficient for each order based on the equipment load parameters in the equipment capacity matrix and the delivery date parameters in the order feature matrix.

[0135] S24 integrates the scrap material loss value with the delay risk coefficient according to preset weights to obtain the comprehensive loss assessment value for each order. The comprehensive loss assessment value is calculated as follows:

[0136] Perform the following operations for each order in the candidate production schedule:

[0137] From the surplus inventory matrix Select rolls that meet the criteria:

[0138] in, A set of rolls that meet the conditions, For the first One roll of material, This refers to the remaining length of the coil. This is the replacement material number for the coil material. For the required length of steel in the order, For the material type of the order, For material replacement rule functions;

[0139] Selecting the optimal roll material using a value function:

[0140] ;

[0141] in, This is the optimal roll material selection function. To find the roll material object with the lowest overall cost from the candidate list, For size matching, This refers to the transportation distance from the warehouse to the processing equipment. To calculate transportation costs;

[0142] For successfully matched orders Calculate the waste value of scrap materials:

[0143] ;

[0144] in, To select the length of the roll, It is the first The length of the order to be cut in one go. This refers to the number of orders to be cut from this roll of material;

[0145] Processing start time prediction:

[0146] ;

[0147] in, The current system time. For device queue load rate, For equipment Real-time throughput, For equipment Task queue length, As a unit time base;

[0148] Delay risk coefficient calculation:

[0149] ;

[0150] in, For order processing time, This refers to the order delivery deadline;

[0151] Calculate the first Comprehensive loss assessment value of the next iteration :

[0152] ;

[0153] in, For material consumption items, For time consumption items, The delay penalty coefficient;

[0154] ;

[0155] in, Indicates the number of urgent orders. This represents the total number of orders.

[0156] S3 combines simulated annealing algorithm and adaptive loss threshold mechanism to perform crossover and mutation operations on the candidate production scheduling sequences; redundant sequences are eliminated through similarity screening, and the potential optimal production scheduling scheme within the allowable error range is determined by regression fitting; the sequence crossover, mutation, screening and fitting process is executed iteratively until the optimal production scheduling scheme that meets the preset conditions is obtained. Specifically, it includes the following:

[0157] S31 initializes the simulated annealing parameters, including the initial temperature, cooling coefficient, and termination temperature, and sets the initial reference value for the adaptive loss threshold.

[0158] S32 selects the candidate production scheduling sequence with the better comprehensive loss assessment value as the parent sequence. After fixing the positions of high-priority orders, it performs a region crossover operation on the remaining orders to generate the child sequence. The child sequences obtained from the region crossover are then randomly swapped to generate new candidate production scheduling sequences. This step specifically includes the following:

[0159] Two candidate production scheduling sequences with better overall loss assessment values ​​are used as parent sequences. The positions of orders with priority level ≥ 4 are fixed, and regional crossover and random position swapping are performed on the remaining orders to generate child sequences as new candidate production scheduling sequences.

[0160] Region crossover operation: Randomly select the crossover interval [start, end] from the two parent sequences, where start is the starting position order of the parent sequence and end is the ending position order of the parent sequence. The child sequence directly inherits the interval segment of one of the parent sequences, and the remaining positions are filled with orders with priority ≤ 3 according to the order of the other parent sequence.

[0161] Random position swap: Perform n swaps for each subsequence, and only swap the order positions with priority ≤ 3, n = 1, 2, 3, ... 5;

[0162] Combining simulated annealing with a probabilistic acceptance of suboptimal solutions, specifically:

[0163] Annealing temperature update formula:

[0164] ;

[0165] in, Let be the annealing temperature at the t-th iteration. The initial temperature for the production scheduling process. For decay rate, This represents the number of iterations.

[0166] The probability of accepting a suboptimal solution is determined using the Metropolis criterion:

[0167] ;

[0168] in, The probability of accepting a suboptimal solution, This is the difference between the loss assessment values ​​of the new solution and the current solution.

[0169] Set an adaptive loss threshold and gradually tighten the standard during iterations, specifically as follows:

[0170] Initial loss threshold ,in This represents the historical average loss rate.

[0171] Loss threshold varies with iteration number Adaptive tightening:

[0172] , ;

[0173] Set a dynamic acceptance probability, allowing the algorithm to accept inferior solutions with a certain probability. > To avoid getting trapped in local optima, the probability of accepting inferior solutions is gradually reduced with each iteration, eventually stabilizing at the optimal solution:

[0174] ;

[0175] in, This represents the total loss value for the current iteration. The historical best loss value, temperature parameter With the number of iterations attenuation: , This is the initial temperature.

[0176] S33 combines an adaptive loss threshold mechanism with a simulated annealing algorithm to evaluate and accept new candidate production scheduling sequences. It then filters the candidate production scheduling sequences by calculating probability similarity and retains those that meet the requirements. Based on the steel length error range, it performs scatter-point fitting regression calculation on the filtered sequences to determine the potential optimal production scheduling scheme.

[0177] This step involves filtering candidate production sequences based on probability similarity, specifically including:

[0178] Collect order information from candidate production sequences, including material width, length, and quantity; generate all possible production combinations based on material length, quantity, and width;

[0179] For each candidate production combination, normalization is performed, and the Euclidean norm of the actual demand is calculated:

[0180] If two candidate production scheduling combinations With candidate production combinations Production scheduling is considered successful if the following conditions are met:

[0181] ;

[0182] in, Candidate production combinations The nth cutting group in Candidate production combinations The nth cutting group in For the order demand feature vector, the symbol is... To normalize the three-dimensional feature vectors of candidate production combinations, the symbols... Calculate the Euclidean norm. This represents the normalized order demand feature vector and the normalized candidate production scheduling combination. The difference vector between the feature vectors;

[0183] Set similarity threshold :

[0184] ;

[0185] in, It is a dynamic adjustment factor. ;

[0186] If similarity threshold Greater than probability similarity If the two candidate production combinations are not found, they will be eliminated; otherwise, they will be retained.

[0187] The optimal production scheduling scheme is obtained by scatter-point fitting regression calculation based on the steel length error range:

[0188] Scatter fitting is performed on the set of candidate production scheduling sequences after screening, taking into account a 10% length error range of the steel after trimming, to obtain the scatter fitting set;

[0189]

[0190] in, =Theoretical length of steel in the order , =Actual cutting length deviation rate ;

[0191] Length error range constraint:

[0192] ;

[0193] The least squares method is used to determine the optimal production scheduling plan, which minimizes the error between the actual production scheduling result and the ideal production scheduling plan. Specifically, this includes:

[0194] a) Construct the input matrix D, which contains the theoretical length of steel in the order. Deviation rate from actual cutting length ;

[0195] b) Solve the weighted least squares equation to obtain the bias prediction model;

[0196] c) Select the candidate production sequence that minimizes the sum of the absolute values ​​of the prediction deviations as the optimal production scheduling scheme.

[0197] The formula for calculating the length deviation rate is as follows:

[0198] Length deviation rate ;

[0199] in, For the required length of steel in the order, This is the actual cutting length.

[0200] S34. Lower the simulated annealing temperature and repeat S32 to S33 until the termination condition is met, and finally determine the optimal production schedule. If the requirements are not met, return to S31 to readjust the parameters and iterate.

[0201] Application Examples

[0202] This embodiment takes a heavy equipment manufacturing enterprise (annual production capacity of 500,000 tons) as the implementation object. The original production scheduling system of this enterprise has problems such as low material utilization rate (72%), high delay rate of urgent orders (18%), and uneven equipment load (Gini coefficient 0.48). The implementation process after adopting the solution of this invention is as follows:

[0203] S1 Data Matrix Construction

[0204] There are currently 50 pending orders (including 10 urgent orders), involving 5 materials such as Q235B and Q355C;

[0205] Parameters of the 3 cutting machines: maximum processing length 12m, real-time task queue lengths are 8 / 15 / 5 respectively;

[0206] 20 rolls of stock (length 1.8-12.3m);

[0207] The matrix is ​​constructed as follows:

[0208] Order Feature Matrix

[0209]

[0210] Equipment Capability Matrix

[0211]

[0212] surplus inventory matrix

[0213]

[0214] Dynamic calculation of penalty coefficient for emergency orders:

[0215] =0.5+0.1×(Number of urgent orders / Total number of orders)=0.5+0.1×(10 / 50)=0.7.

[0216] S2 Dynamic Risk Assessment

[0217] Monte Carlo sequence generation: in constrained space Generate 200 initial sequences ( After verifying the device's capabilities (e.g., DevB maximum queue length 12), illegal sequences are removed, leaving 120 candidate sequences.

[0218] Example of scrap material optimization:

[0219] Traditional solution: Order Ord01 (requires 6200mm) uses Coil01 (8500mm) separately.

[0220] Loss rate = (8500-6200) / 8500×100% = 27.1%.

[0221] This solution combines cutting: Ord01 (6200mm) + Ord12 (1900mm) using Coil01.

[0222] Loss rate = (8500 - (6200 + 1,900)) / 8500 × 100% = 9.8% (a reduction of 64%).

[0223] Delay risk prediction:

[0224] DevB load rate = current queue / maximum capacity = 15 / 12 = 125%

[0225] Expected start time =Delay 37.5min

[0226] Order Ord01 delay risk =22.5min

[0227] Comprehensive loss value 9.8% + 0.7 × log(1 + 22.5 / 60) ≈ 15.3%.

[0228] S3 Iterative Optimization

[0229] Simulated annealing parameters:

[0230] initial temperature =1000, attenuation coefficient α=0.95, 50 iterations

[0231] Adaptive loss threshold =0.9 × 28% = 25.2%, tightening by 5% per iteration;

[0232] Cross-regional operations: Orders with a fixed priority of ≥4 (e.g., Ord01)

[0233] Perform crossover on orders with priority ≤ 3: randomly select and swap orders within the range of [3, 15].

[0234] Regression fit verification:

[0235] Length error dataset

[0236]

[0237] Least squares fitting results: absolute error reduced by 72%.

[0238] Comparison of implementation results (6-month average)

[0239]

[0240] The method of this invention acquires order parameters, equipment status, and inventory balance through a real-time data interface and converts them into a multi-dimensional numerical matrix (including an order feature matrix, an equipment capacity matrix, and a surplus material inventory matrix) to ensure structured data processing and efficient utilization. It employs a Monte Carlo method to integrate multiple strategies to optimize the production scheduling process, generating candidate sequences based on order priority and equipment constraints. It dynamically assesses scrap material loss and delay risks, and combines simulated annealing and adaptive threshold mechanisms to achieve sequence crossover and mutation. Finally, it determines the optimal production scheduling scheme within the allowable error range through similarity screening and regression fitting, iterating until the conditions are met, thereby balancing global optimization and local efficiency, and improving production tolerance and resource utilization.

[0241] Based on this, the present invention also proposes a steel shearing scheduling device based on dynamic risk assessment, used to execute the above-mentioned steel cutting scheduling method based on dynamic risk assessment, comprising:

[0242] The data acquisition and processing module acquires production scheduling data in real time, including order parameter sets, equipment status vectors, and inventory balances. Based on this data, it performs numerical processing, converting it into a multi-dimensional numerical matrix. This multi-dimensional numerical matrix includes an order feature matrix, an equipment capacity matrix, and a surplus inventory matrix. This data acquisition and processing is used to execute step S1 of the aforementioned steel cutting production scheduling method based on dynamic risk assessment.

[0243] The dynamic risk assessment module, based on the Monte Carlo method, performs calculations on the order feature matrix, the equipment capacity matrix, and the surplus material inventory matrix within a constraint space to generate candidate production scheduling sequences that satisfy order priority and equipment constraints. For each order in the candidate production scheduling sequence, it dynamically calculates the scrap material loss and delay risk, and comprehensively obtains the overall loss assessment value for each order. The dynamic risk assessment module is used to execute step S2 of the above-described steel cutting production scheduling method based on dynamic risk assessment.

[0244] The production scheduling module, combining simulated annealing algorithm and adaptive loss threshold mechanism, performs crossover and mutation operations on the candidate production scheduling sequences; it eliminates redundant sequences through similarity screening and uses regression fitting to determine the potential optimal production scheduling scheme within the allowable error range; it iteratively executes the sequence crossover, mutation, screening, and fitting process until the optimal production scheduling scheme that meets preset conditions is obtained. The production scheduling module is used to execute step S3 of the above-mentioned steel cutting production scheduling method based on dynamic risk assessment.

[0245] This invention achieves highly adaptable production scheduling scheme generation by constructing a multi-source data matrix processing model and combining Monte Carlo random sampling with a dynamic loss rate feedback mechanism. The algorithm innovatively employs a probabilistic component similarity matching method to generate candidate sequences based on order length and width parameters. Through real-time loss rate calculation and a dynamic threshold optimization mechanism, it performs global optimization in the solution space, thereby generating an optimal production scheduling scheme that balances material utilization and production feasibility.

[0246] The present invention also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist alone and not assembled into the electronic device.

[0247] The aforementioned computer-readable medium carries one or more programs that, when executed by an electronic device, cause the electronic device to implement the methods described in the above embodiments.

[0248] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0249] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of this disclosure.

[0250] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein.

[0251] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A steel cutting production scheduling method based on dynamic risk assessment, characterized in that, Comprising the following steps: S1 real-time acquisition of scheduling data, the scheduling data including order parameter set, equipment state vector and inventory balance; numerical processing based on the scheduling data, converted into a multi-dimensional numerical matrix, the multi-dimensional numerical matrix including order feature matrix, equipment capacity matrix and excess material inventory matrix; S2 based on the Monte Carlo method, operating the order feature matrix, the equipment capacity matrix and the excess material inventory matrix in the constraint space to generate a candidate scheduling sequence that meets the order priority and the equipment constraint condition; dynamically calculating the offcut loss and delay risk for each order in the candidate scheduling sequence, and comprehensively obtaining the comprehensive loss evaluation value of each order; S3 combining the simulated annealing algorithm and the adaptive loss threshold mechanism, performing cross and mutation operations on the candidate scheduling sequence; removing redundant sequences through similarity screening, and determining the potential optimal scheduling scheme within the error allowed range by using regression fitting; performing sequence cross, mutation, screening and fitting processes in a loop until the optimal scheduling scheme that meets the preset condition is obtained; in step S3, the following steps are specifically included: S31 initializing simulated annealing parameters, including initial temperature, cooling coefficient and termination temperature, and setting the initial reference value of the adaptive loss threshold value; S32 selecting a candidate scheduling sequence with better comprehensive loss evaluation value as the parent, fixing the position of high-priority orders, and performing regional cross operation on the remaining orders to generate a child sequence; performing random position exchange on the child sequence obtained by regional cross to generate a new candidate scheduling sequence; specifically including: Taking two candidate scheduling sequences with better comprehensive loss evaluation values as parent sequences, fixing the positions of orders with priority ≥ 4, and performing regional cross and random position exchange on the remaining orders to generate a child sequence as a new candidate scheduling sequence: Regional cross operation: randomly selecting a cross interval [start, end] from the two parent sequences, start being the starting position order of the parent sequence, and end being the termination position order of the parent sequence, the child sequence directly inheriting the interval segment of one of the parent sequences, and the remaining positions being filled with orders with priority ≤ 3 in the order of the other parent sequence; Random position exchange: performing n exchanges for each child sequence, and only exchanging the positions of orders with priority ≤ 3, n = 1, 2, 3, … 5; Combining the simulated annealing mechanism to accept suboptimal solutions, specifically: Annealing temperature update formula: ； wherein, Ttis the annealing temperature at the tth iteration, T0is the initial temperature of the scheduling process, is the decay rate, is the number of iterations; The probability of accepting suboptimal solutions adopts Metropolis criterion: ； wherein, is the probability of accepting a suboptimal solution, is the difference between the loss evaluation value of the new solution and the current solution, Setting the adaptive loss threshold value, gradually tightening the standard in iteration, specifically: initial wear threshold wherein is the historical average wear rate; Loss threshold as a function of iteration number Adaptive tightening: ， ； Setting dynamic acceptance probability, allowing the algorithm to accept inferior solution with a certain probability , avoid falling into local optimum, gradually reduce the probability of accepting inferior solution with iteration, and finally stabilize at the optimal solution:​ ； wherein, is the first iteration, is the synthetic loss evaluation value of the n-th iteration, is the historical optimal loss value, the temperature parameter decreases with the iteration number n, decreases exponentially: , is the initial temperature; S33 combining the adaptive loss threshold mechanism and the simulated annealing algorithm, evaluating and accepting the new candidate scheduling sequence, screening the candidate scheduling sequences through probability similarity calculation, and retaining the sequences that meet the requirements; based on the steel length error range, performing scatter fitting regression calculation on the screened sequences to determine the potential optimal scheduling scheme; S34 reducing the simulated annealing temperature, repeating the above S32 to S33 until the termination condition is met, and finally determining the optimal scheduling scheme; if the requirement is not met, returning to S31 to adjust the parameters and iterate.

2. The steel cutting production scheduling method based on dynamic risk assessment according to claim 1, wherein, The production scheduling data includes the order parameter set , equipment state vector , inventory reserve distribution and material replacement rule table ; wherein, is the total number of orders to be scheduled currently, is the data object of the th order; is the state data of the th equipment; is the total number of inventory roll types, is the inventory data of the th roll type; Based on the order parameter set, the geometric parameters of the order are extracted, including the length, width, thickness and quantity of the steel required by the order, the order priority weight is labeled, and the order feature matrix is constructed as follows based on the delivery urgency and customer level dynamic adjustment: ； wherein, is an order number, is a required steel length for the order, is a required steel width for the order, is a required steel quantity for the order, is a priority weight; Based on the equipment state vector, the real-time throughput, current task load and maintenance state of the equipment are obtained, the wear index of the equipment is calculated, the remaining service life is predicted combined with historical data, and the equipment capability matrix is constructed as follows: ； wherein, is the equipment number, is the order length, is the maximum processing length, is the current task queue length; According to the inventory distribution, the remaining size, material type and storage location of the coil stock in each warehouse are recorded, the recommended use scenario and replaceable rule of the remaining material are labeled, and the remaining material inventory matrix is constructed as follows: ； wherein, is the roll number, is the remaining size, is the replaceable material number.

3. The steel cutting production scheduling method based on dynamic risk assessment according to claim 1, wherein, In step S2, the following steps are specifically included: S21, set the constraint space parameters, including order priority sorting rules, equipment running threshold and production resource limitation conditions; based on the Monte Carlo method, random sampling operation is performed on the order feature matrix, equipment capability matrix and remaining material inventory matrix, and an initial production scheduling sequence is generated in the constraint space; S22, according to the order priority sorting rules and equipment running threshold, the initial production scheduling sequence is checked for compliance, and the sequence that does not meet the constraint condition is removed, and the candidate production scheduling sequence that meets the requirement is retained; S23, for each order in the candidate production scheduling sequence, the edge and corner loss value is calculated combined with the material demand parameters in the remaining material inventory matrix and the order feature matrix; Based on the equipment load parameters in the equipment capability matrix and the delivery period parameters in the order feature matrix, the delay risk coefficient of each order is calculated; S24, the edge and corner loss value and the delay risk coefficient are fused according to the preset weight to obtain the comprehensive loss evaluation value of each order.

4. The steel cutting production scheduling method based on dynamic risk assessment according to claim 3, wherein, In step S21, the Monte Carlo search space is defined as follows: ； wherein, denotes a Monte Carlo search space, i.e. a set of candidate scheduling sequences, denotes the i-th candidate scheduling sequence, denotes the i-th candidate scheduling sequence, denotes a full permutation of the order set, denotes the current generated sequence index, denotes the maximum number of generated sequences.

5. The steel cutting production scheduling method based on dynamic risk assessment according to claim 3, wherein, In step S24, the comprehensive loss evaluation value calculation method is as follows: from the scrap inventory matrix selecting a roll that meets the conditions from the scrap inventory matrix: ; wherein, a set of coils satisfying the condition, a first coil, a remaining length of the coil, a replaceable material number of the coil, a required steel length for the order, a material type of the order, a material replacement rule function; Select the optimal coil stock through the value function: ； wherein, is a function of selecting the optimal roll, is a function of finding the roll object with the minimum overall cost in the candidate list, is a function of size matching degree, is a function of transportation distance from the warehouse to the processing equipment, is a function of calculating transportation cost; For the orders matched successfully, the edge and corner loss value is calculated: ； wherein, is the length of the selected roll, is the first cut order length, is the number of orders cut from the roll. Processing start time prediction: ； in, The current system time. For device queue load rate, For equipment Real-time throughput, For equipment Task queue length, As a unit time base; Delay risk coefficient calculation: ； wherein, is the order processing duration, is the order delivery deadline; The calculation of the synthetic loss evaluation value of the second iteration :​ ； wherein, is a material consumption item, is a time consumption item, is a delay penalty coefficient; ； wherein, represents the number of emergency orders, is the total number of orders.

6. The steel cutting production scheduling method based on dynamic risk assessment according to claim 1, wherein, In step S33, the candidate production scheduling sequence is screened through the probability similarity, specifically including: Collect the order information in the candidate production scheduling sequence, including material width, length and quantity; Generate all possible production scheduling combinations according to the material length, quantity and width; Normalize each candidate production scheduling combination and calculate the Euclidean norm with the actual demand: If two candidate scheduling combinations and the candidate scheduling combination If the following conditions are met, it is considered that the scheduling is successful, specifically: ； wherein, is the nth cut set in the candidate production scheduling combination is the nth cut set in the candidate production scheduling combination is the nth cut set in the candidate production scheduling combination is the nth cut set in the candidate production scheduling combination is the order demand feature vector, symbol is the normalization of the three-dimensional feature vector of the candidate production scheduling combination, symbol is the Euclidean norm calculation, represents the difference vector between the normalized order demand feature vector and the normalized feature vector of the candidate production scheduling combination . Setting a similarity threshold : ； wherein is a dynamic adjustment factor, ; if the similarity threshold is greater than the probability similarity then the two candidate production scheduling combinations are eliminated, otherwise the two candidate production scheduling combinations are retained; Based on the steel length error range, the scatter point fitting regression calculation is performed to obtain the optimal production scheduling scheme: Scatter point fitting is performed on the set of screened candidate production scheduling sequences, considering the 10% length error range of the cut steel, to obtain a scatter point fitting set; ； wherein, = theoretical length of steel material of order , = actual cutting length deviation rate ; Length error range constraint: ； Using the least square method, the optimal production scheduling scheme is determined to minimize the error between the actual production scheduling result and the ideal production scheduling scheme, specifically including: a) constructing an input matrix D, containing the theoretical length of steel material of orders and the deviation rate of actual cutting length ; b) Solve the weighted least square equation to obtain the bias prediction model; c) Select the candidate production scheduling sequence with the minimum absolute value of the predicted bias as the optimal production scheduling scheme.

7. The steel cutting production scheduling method based on dynamic risk assessment according to claim 6, wherein, The calculation formula of the actual cutting length deviation rate is as follows: Actual cutting length deviation rate ; wherein, is the required steel length for the order, is the actual cut length.

8. A steel cutting production scheduling device based on dynamic risk assessment, characterized in that, A steel cutting production scheduling method based on dynamic risk assessment for performing any one of claims 1-7, comprising: A data acquisition and processing module acquires scheduling data in real time, the scheduling data including an order parameter set, a device state vector, and an inventory amount; numerical processing is performed based on the scheduling data to convert into a multi-dimensional numerical matrix, the multi-dimensional numerical matrix including an order feature matrix, a device capability matrix, and a leftover inventory matrix; A dynamic risk assessment module performs operations on the order feature matrix, the device capability matrix, and the leftover inventory matrix within a constraint space based on a Monte Carlo method to generate a candidate scheduling sequence that satisfies order priority and device constraint conditions; for each order in the candidate scheduling sequence, dynamic calculation is performed on offcut loss and delay risk to comprehensively obtain a comprehensive loss evaluation value of each order; A scheduling scheme generation module performs cross and mutation operations on the candidate scheduling sequence in combination with a simulated annealing algorithm and an adaptive loss threshold mechanism; redundant sequences are removed through similarity screening, and a potential optimal scheduling scheme within an error allowable range is determined using regression fitting; the sequence cross, mutation, screening, and fitting processes are iteratively executed until an optimal scheduling scheme that satisfies a preset condition is obtained.

Images