Intelligent production scheduling optimization method and system based on large model

Abstract

The invention discloses an intelligent production scheduling optimization method based on a large model. The method comprises the following steps: S1, data acquisition and preprocessing; collecting equipment operation data, material supply data, production progress data and production task information; performing cleaning, normalization and abnormal value processing on the collected data to generate structured data; s2, determining an optimization objective of a scheduling scheme and constraint conditions of a production workshop, constructing a scheduling model, and generating a preliminary scheduling scheme; s3, evaluating the preliminary scheduling scheme by using a scheduling scheme evaluation algorithm; and S4, outputting an optimal production scheduling scheme, and applying the scheme to a production system. According to the method, the entropy weight multi-objective decision-making method is combined with the multi-objective dragonfly optimization algorithm, so that the production scheduling scheme can be effectively evaluated and optimized. Compared with a traditional mathematical programming method and a heuristic algorithm, the method provided by the invention has the advantages that the scheduling efficiency is remarkably improved while the quality of the solution is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field related to production scheduling control, and in particular to an intelligent production scheduling optimization method and system based on a large model. Background Art

[0002] With the advent of the Industrial 4.0 era, intelligent manufacturing has become an important direction for the development of the manufacturing industry. In intelligent manufacturing, production scheduling optimization is a key link to improve production efficiency, reduce costs, and improve product quality. Production scheduling optimization involves multiple goals, such as minimizing production costs, minimizing production time, maximizing equipment utilization, etc. There are often conflicts between these goals, making production scheduling a complex multi-objective optimization problem. At present, production scheduling optimization methods mainly include traditional mathematical programming methods, heuristic algorithms, and intelligent optimization algorithms. Traditional mathematical programming methods have large computational complexity when dealing with large-scale problems and are difficult to meet real-time requirements. Although heuristic algorithms have fast calculation speeds, they often cannot guarantee the optimality of solutions. Intelligent optimization algorithms, such as genetic algorithms and particle swarm optimization algorithms, have solved the above problems to a certain extent, but when dealing with multi-objective optimization problems, there are still problems such as slow convergence, easy to fall into local optimality, and insufficient diversity of solutions.

[0003] In addition, existing technologies often lack effective methods for determining objective weights when dealing with multi-objective optimization problems, which makes it difficult to balance the importance of different objectives in practical applications. At the same time, existing algorithms are also insufficient in maintaining the diversity of solutions, and are prone to produce overly concentrated solution sets, thus affecting the flexibility and adaptability of scheduling solutions. Summary of the invention

[0004] In view of the above problems, the present invention proposes an intelligent production scheduling optimization method based on a large model. The method combines the entropy weight multi-objective decision-making method and the multi-objective dragonfly optimization algorithm, and realizes efficient optimization of the production scheduling problem through steps such as data collection and preprocessing, scheduling model construction, scheduling scheme evaluation and optimal scheme output. The present invention determines the weight of each optimization target through the entropy weight method, so that the algorithm can more reasonably balance the importance of different targets; at the same time, the multi-objective dragonfly optimization algorithm is introduced to improve the convergence speed of the algorithm and the diversity of solutions, thereby obtaining a better production scheduling solution.

[0005] The technical solution adopted by the present invention to solve the technical problem is: first, a large model-based intelligent production scheduling optimization method is provided, comprising the following steps: S1. Data collection and preprocessing: collect equipment operation data, material supply data, production progress data and production task information; clean, normalize and process outliers of the collected data to generate structured data; S2. Determine the optimization goal of the scheduling plan and the constraints of the production workshop, extract features and learn patterns of historical production data through a large model, build a scheduling model, and generate a preliminary scheduling plan; S3, using the scheduling scheme evaluation algorithm to evaluate the preliminary scheduling scheme based on the dynamic parameter adjustment capability of the large model; S4, output the optimal production scheduling plan and apply the plan to the production system; Among them, the S3 scheduling scheme evaluation algorithm uses the entropy weight multi-objective decision-making method combined with the multi-objective dragonfly optimization algorithm to evaluate the scheduling scheme.

[0006] Furthermore, the S3 entropy weight multi-objective decision-making method combined with the multi-objective dragonfly optimization algorithm includes the following steps: A1. Extract multiple production samples from the preprocessed data, use the big model to generate simulation samples of future production scenarios, and standardize the production samples to unify the various optimization goals in the same direction. Each sample represents the actual production situation in a period of time. A2. Calculate the entropy value of each optimization indicator based on the extracted sample data E j , and calculate the weight of each optimization indicator based on the entropy value ω j ; A3. Initialize the dragonfly population, generate a candidate solution set for the initial scheduling plan through the large model, assign an initial position and speed to each dragonfly, and each dragonfly is a scheduling plan; A4. Calculate the fitness of each dragonfly and optimize the target weight ω j and the values ​​of each optimization objective f j (X i ), calculate the comprehensive fitness; the formula for calculating the comprehensive fitness is: ; in, f j ( X l ) is the i The scheduling scheme is j The value of the optimization target; ω j is the weight of each optimization objective; n is the total number of optimization objectives; A5. Use the Pareto dominance relationship to find the current non-dominated solution set as the current optimal solution; A6. For each dragonfly, calculate the contribution of separation, alignment, cohesion, food attraction, and predator avoidance behaviors based on the information of its neighbors, and update the speed and position of the dragonfly by integrating these behaviors; A7. Re-evaluate the fitness of all dragonflies and update the Pareto optimal solution set; if the fitness of the newly generated scheduling solution is better than the current non-dominated solution set S Some scheduling schemes in , then add them to the non-dominated solution set S , and remove the dominated scheduling scheme; A8. When the predetermined number of iterations G is reached, the algorithm terminates; otherwise, steps A4 to A7 are repeated; A9. Select the optimal production scheduling solution from the Pareto optimal solution set.

[0007] Furthermore, the formula for calculating the entropy value of each optimization target in A2 is:

[0008] in, p jl is the jth optimization objective l is the ratio of samples, and m is the number of samples.

[0009] Furthermore, the formula for calculating the weight of each optimization indicator in A2 is: ; in, ω j is the weight of the jth optimization objective, n is the total number of optimization objectives, E j is the entropy value of each optimization indicator.

[0010] Furthermore, A3 initializes the dragonfly population including: Setting the size N of the dragonfly population, that is, the number of initial scheduling solutions; the size N of the dragonfly population is dynamically adjusted by the large model according to the complexity of the production task and the learning results of the historical scheduling data; Generate a random initial position for each dragonfly X i , X i ∈[ lb , ub ]in, lb and ub are the lower and upper bounds of the decision variables, respectively; Generate a random initial speed for each dragonfly V i , V i ∈[- V max , V max ]; in, Vmax is the maximum speed, which is set according to the size and complexity of the production scheduling problem; Set the maximum number of iterations G of the algorithm; Set the neighborhood radius r to determine the neighbors of the dragonfly.

[0011] Furthermore, A5 uses the Pareto dominance relationship to find the current set of non-dominated solutions, which are the current optimal solutions: A51. Initialize the non-dominated solution set; take all scheduling schemes in the population as the initial candidate solution set; A52, screening non-dominated solutions; Traverse each scheduling solution in the initial candidate solution set i , and compare it with other scheduling solutions in the candidate solution set g Make pairwise comparisons; If the scheduling plan i The fitness of the scheduling scheme is greater than g , and the scheduling scheme i Significantly outperforms the scheduling solution on at least one key optimization objective g , then the scheduling scheme i Control scheduling plan g ; If the scheduling plan i Control scheduling plan g , then the scheduling plan g Remove from the candidate solution set; If the scheduling plan i is not dominated by any other scheduling scheme, then the scheduling scheme i Add non-dominated solution set; A53, maintaining the diversity of solutions; retaining sparsely distributed scheduling solutions according to the distance; A54, termination condition judgment; when the preset number of iterations G is reached, the iteration is stopped; A55. Take the final non-dominated solution set as the Pareto optimal solution set.

[0012] Further, in step A52, key optimization objectives include the following objectives: minimizing production costs, minimizing production time, or maximizing equipment utilization.

[0013] Furthermore, the scheduling scheme of A53 retaining sparse distribution includes: Calculate the distance between each scheduling scheme in the non-dominated solution set; Compute any two scheduling solutions in the solution set X i and X g The Euclidean distance d between Xi , X g ), the formula is: ; in, f k ( X i )and f k ( X g ) are the scheduling schemes X i and X g In the k The fitness value on the optimization target, n is the total number of optimization targets; Calculate each scheduling solution X i Density ρ ( X i ), the formula is: ; in, S is the current non-dominated solution set, ρ ( X i ) represents the scheduling plan X i density; According to the density value ρ ( X i ) for non-dominated solution sets S All scheduling schemes in are sorted, and the first h scheduling schemes with the smallest density values ​​are selected as the reserved solution set, where h is the preset number of reserved solutions.

[0014] Furthermore, the formula for A6 to update the position of the dragonfly is: ; in, represents the new position of the i-th dragonfly at time step t+1; represents the position of the i-th dragonfly at the current time step t; α is the control separation force D i The influence coefficient on the overall position update; D i , A i , C i , F i , R i and E enemyrepresent separation force, alignment force, cohesion force, food attraction force, natural enemy avoidance force and additional displacement to avoid natural enemies, respectively, and Δt is the time step.

[0015] An intelligent production scheduling optimization method based on a large model using the above optimization method is an intelligent production scheduling optimization system based on a large model, comprising: Data collection module, used to collect equipment operation data, material supply data, production progress data and production task information; The preprocessing module is used to clean, normalize and process outliers of the collected data to generate structured data; The model building module integrates a pre-trained large model, analyzes the nonlinear relationship in the production data through its attention mechanism, and automatically generates the optimization objectives of the scheduling model and the constraints of the production workshop; The scheduling scheme evaluation module optimizes the weight distribution and neighborhood radius parameters in the multi-objective dragonfly optimization algorithm in real time based on the dynamic parameter adjustment capability of the large model, and evaluates the preliminary scheduling scheme in combination with the entropy weight multi-objective decision-making method; The output module is used to select the optimal production scheduling solution from the Pareto optimal solution set and apply it to the production system.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention adopts the entropy weight multi-objective decision-making method combined with the multi-objective dragonfly optimization algorithm to effectively evaluate and optimize the production scheduling scheme. Compared with traditional mathematical programming methods and heuristic algorithms, the method of the present invention significantly improves the scheduling efficiency while ensuring the quality of the solution.

[0017] (2) The present invention considers the diversity of solutions in algorithm design and avoids the problem of over-concentration of solutions by retaining sparsely distributed scheduling solutions. This helps to find more comprehensive and more adaptable scheduling solutions and improves the flexibility and robustness of the scheduling system.

[0018] (3) The present invention dynamically determines the weight of each optimization target through the entropy calculation method, so that the algorithm can reasonably balance the importance of different targets according to actual production conditions and business needs. This dynamic adjustment mechanism improves the practicality and adaptability of the scheduling scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required to be used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can also be obtained according to these drawings. Figure 1It is a flow chart of an intelligent production scheduling optimization method based on a large model of the present invention; Figure 2 It is a flow chart of a scheduling scheme evaluation algorithm of a large-model-based intelligent production scheduling optimization method of the present invention; Figure 3 A flowchart of a large-model-based intelligent production scheduling optimization method of the present invention for finding the current non-dominated solution set using the Pareto dominance relationship; Figure 4 This is a structural diagram of an intelligent production scheduling optimization system based on a large model of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0021] This application discloses an intelligent production scheduling optimization method based on a large model, such as Figure 1 As shown, the following steps are included: S1. Data collection and preprocessing: collect equipment operation data, material supply data, production progress data and production task information; clean, normalize and process outliers of the collected data to generate structured data; S2. Determine the optimization goal of the scheduling plan and the constraints of the production workshop, extract features and learn patterns of historical production data through a large model, build a scheduling model, and generate a preliminary scheduling plan; S3, using the scheduling scheme evaluation algorithm to evaluate the preliminary scheduling scheme based on the dynamic parameter adjustment capability of the large model; S4, output the optimal production scheduling plan and apply the plan to the production system; Among them, the S3 scheduling scheme evaluation algorithm uses the entropy weight multi-objective decision-making method combined with the multi-objective dragonfly optimization algorithm to evaluate the scheduling scheme.

[0022] As a specific implementation method, the S3 entropy weight multi-objective decision-making method combined with the multi-objective dragonfly optimization algorithm includes the following steps: Figure 2 As shown: A1. Extract multiple production samples from the preprocessed data, use the big model to generate simulation samples of future production scenarios, and standardize the production samples. Each sample represents the actual production situation within a period of time. A2. Calculate the entropy value of each optimization indicator based on the extracted sample data E j , and calculate the weight of each optimization indicator based on the entropy value ω j ; A3. Initialize the dragonfly population, generate a candidate solution set for the initial scheduling plan through the large model, assign an initial position and speed to each dragonfly, and each dragonfly is a scheduling plan; A4. Calculate the fitness of each dragonfly and optimize the target weight ω j and the values ​​of each optimization objective f j (X i ), calculate the comprehensive fitness; the formula for calculating the comprehensive fitness is: ; in, f j ( X l ) is the i The scheduling scheme is j The value of the optimization target; ω j is the weight of each optimization objective; n is the total number of optimization objectives; A5. Use the Pareto dominance relationship to find the current non-dominated solution set as the current optimal solution; A6. For each dragonfly, calculate the contribution of separation, alignment, cohesion, food attraction, and predator avoidance behaviors based on the information of its neighbors, and update the speed and position of the dragonfly by integrating these behaviors; A7. Re-evaluate the fitness of all dragonflies and update the Pareto optimal solution set; if the fitness of the newly generated scheduling solution is better than the current non-dominated solution set S Some scheduling schemes in , then add them to the non-dominated solution set S , and remove the dominated scheduling scheme; A8. When the predetermined number of iterations G is reached, the algorithm terminates; otherwise, steps A4 to A7 are repeated; A9. Select the optimal production scheduling solution from the Pareto optimal solution set.

[0023] As a specific implementation, the formula for calculating the entropy value of each optimization target in A2 is:

[0024] in, p jl is the jth optimization objective lThe ratio of samples is used to measure the relative importance or distribution of each sample on a specific optimization target; m is the number of samples.

[0025] As a specific implementation method, the formula for calculating the weight of each optimization index in A2 is: ; in, ω j is the weight of the jth optimization objective, n is the total number of optimization objectives, E j is the entropy value of each optimization indicator.

[0026] The big model is a pre-trained Transformer architecture model, which interacts with the scheduling optimization process in the following ways: in the data preprocessing stage, the embedding layer of the big model is used to perform semantic analysis on unstructured production logs; in the model building stage, the correlation between key production constraints and goals is identified through the attention mechanism of the big model; in the solution evaluation stage, the reinforcement learning module of the big model is called to adaptively adjust the convergence speed of the dragonfly optimization algorithm. Example

[0027] Assume that we have collected the following sample data, each sample represents the actual production situation within a week, as shown in Table 1:

[0028] Table 1 shows the initial sample data collected.

[0029]

[0030] Use the min-max normalization method to convert each indicator to the range of 0 to 1: Normalized value = (original value − minimum value) / (maximum value − minimum value) For production costs (hope the cost is as low as possible), production time and equipment utilization (hope the higher the better), we standardize as follows: Production cost (min = 90, max = 130): Sample 1: (100-90) / (130-90)=0.25 Sample 2: (90-90) / (130-90)=0 Sample 3: (120-90) / (130-90)=0.75 Sample 4: (110-90) / (130-90)=0.5 Sample 5: (130-90) / (130-90)=1 Production time (minimum = 8 hours, maximum = 12 hours): Sample 1: 1-(10-8) / (12-8)=0.5 Sample 2: 1-(12-8) / (12-8)=0 Sample 3: 1-(8-8) / (12-8)=1 Sample 4: 1-(9-8) / (12-8)=0.75 Sample 5: 1-(11-8) / (12-8)=0.25 Equipment utilization (minimum = 70%, maximum = 90%): Sample 1: 1-(80-70) / (90-70)=0.5 Sample 2: 1-(75-70) / (90-70)=0.75 Sample 3: 1-(90-70) / (90-70)=0 Sample 4: 1-(85-70) / (90-70)=0.25 Sample 5: 1-(70-70) / (90-70)=1 Step 1: Calculate the proportion p jl For each optimization target, calculate its proportion p in all samples jl : Production cost: p11=0.1818, p21=0.1636, p31=0.2182, p41=0.2, p51=0.2364; Production time: p12=0.2, p22=0.24, p32=0.16, p42=0.18, p52=0.22; Equipment utilization: p13=0.2, p23=0.1875, p33=0.225, p43=0.2125, p53=0.175; Step 2: Calculate entropy E j Using the formula Calculate the entropy value for each optimization objective: =0.9948 =0.9938 =0.9976 Step 3: Calculate weights ω j Using the formula Calculate the weight of each optimization objective: =0.3768 =0.4493 =0.1739 Step 4: Use the formula Calculate the comprehensive fitness: =0.4058 =0.1304 =0.7319 =0.5689 =0.6630 Furthermore, A3 initializes the dragonfly population including: Setting the size N of the dragonfly population, that is, the number of initial scheduling solutions; the size N of the dragonfly population is dynamically adjusted by the large model according to the complexity of the production task and the learning results of the historical scheduling data; Generate a random initial position for each dragonfly X i , X i ∈[ lb , ub ]in, lb and ub are the lower and upper bounds of the decision variables, respectively; Generate a random initial speed for each dragonfly V i , V i ∈[- V max , V max ]; in, V max is the maximum speed, which is set according to the size and complexity of the production scheduling problem; Set the maximum number of iterations G of the algorithm; Set the neighborhood radius r to determine the neighbors of the dragonfly. Example:

[0031] This embodiment describes how to use the dragonfly algorithm in the large model-based intelligent production scheduling optimization method to initialize the dragonfly population and generate an initial solution set for the production scheduling problem.

[0032] Identify decision variables in production scheduling problems, such as machine assignment, process sequence, etc.

[0033] Determine the upper and lower bounds of each decision variable, namely: lb and ub.

[0034] According to the scale and complexity of the problem, the size of the dragonfly population is set: N, which represents the number of initial scheduling solutions.

[0035] For each dragonfly i (i=1,2,...,N), randomly generate an initial position: X i .Location: X i : is a vector, each component of which represents the value of a decision variable, and the value of each component is within the corresponding upper and lower bounds: [lb,ub].

[0036] For example, if the decision variable is the working time of a machine, its upper and lower bounds might be [0,24] (hours).

[0037] For each dragonfly i, a random initial speed is generated V i Each component of the velocity vector represents the change in the value of the decision variable, and the value of each component is between: [− V max , V max ]Inside. V max It is the maximum speed set according to the scale and complexity of the production scheduling problem.

[0038] Set the maximum number of iterations of the algorithm: G, which determines the total number of times the algorithm runs.

[0039] Set the neighborhood radius: r, which is used to determine the neighbors of a dragonfly. The neighborhood radius determines the range of other dragonflies that a dragonfly considers when updating its position.

[0040] The above initialized dragonfly population is applied to the production scheduling optimization problem, where each dragonfly represents a possible scheduling solution. The dragonfly algorithm is used to iteratively optimize these initial solutions to find the optimal or approximately optimal production scheduling solution.

[0041] As a specific implementation method, A5 uses the Pareto dominance relationship to find the current non-dominated solution set, such as Figure 3 As shown, the current optimal solution includes: A51. Initialize the non-dominated solution set; take all scheduling schemes in the population as the initial candidate solution set; A52, screening non-dominated solutions; Traverse each scheduling solution in the initial candidate solution set i , and compare it with other scheduling solutions in the candidate solution set g Make pairwise comparisons; If the scheduling plan i The fitness of the scheduling scheme is greater thang , and the scheduling scheme i Significantly outperforms the scheduling solution on at least one key optimization objective g , then the scheduling plan i Control scheduling plan g ; If the scheduling plan i Control scheduling plan g , then the scheduling plan g Remove from the candidate solution set; If the scheduling plan i is not dominated by any other scheduling scheme, then the scheduling scheme i Add non-dominated solution set S ; A53, maintaining the diversity of solutions; retaining sparsely distributed scheduling solutions according to the distance; A54, termination condition judgment; when the preset number of iterations G is reached, the iteration is stopped; A55. Take the final non-dominated solution set as the Pareto optimal solution set.

[0042] As a specific implementation, in step A52, key optimization objectives include but are not limited to the following objectives: minimizing production costs, minimizing production time, or maximizing equipment utilization.

[0043] As a specific implementation, the A53 reserved sparse distribution scheduling scheme includes: Calculate the distance between each scheduling scheme in the non-dominated solution set; Compute any two scheduling solutions in the solution set X i and X g The Euclidean distance d between X i , X g ), the formula is: ; in, f k ( X i )and f k ( X g ) are the scheduling schemes X i and X g The fitness value at the kth optimization target, n is the total number of optimization targets; Calculate each scheduling solution X iDensity ρ ( X i ), the formula is: ; in, S is the current non-dominated solution set, ρ ( X i ) represents the scheduling plan X i density; According to the density value ρ ( X i ) for non-dominated solution sets S All scheduling schemes in are sorted, and the first h scheduling schemes with the smallest density values ​​are selected as the reserved solution set, where h is the preset number of reserved solutions. Example:

[0044] 1. Initialize the non-dominated solution set: Assume that we have the following five scheduling schemes (each with three key optimization indicators: equipment utilization, production cost, and production time), as shown in Table 2:

[0045] Table 2 shows the collected initial scheduling plans.

[0046]

[0047] 2. Screening non-dominated solution set: First, we need to determine which solutions are non-dominated. A solution is non-dominated if it is not inferior to any other solution in all optimization metrics and is superior to another solution in at least one metric.

[0048] By comparing the above solutions, it can be concluded that the non-dominated solution set includes Solution 2, Solution 3 and Solution 4.

[0049] 3. Calculate the distance between each two scheduling plans d(2,3)≈1.458; d(2,4)≈1.031; d(3,4)≈0.433.

[0050] Calculate the density of each scheduling solution ρ ( X i ).

[0051] ≈1.6558; ≈2.9953; ≈3.2794; Sort the scheduling solutions according to the density value, and select the first h scheduling solutions with the smallest density value as the retained solution set. Set h to 2.

[0052] result: Assume that after calculation and sorting, the first two scheduling solutions with the smallest density value are selected as the retained solution set. Retain solution 3 and solution 4. Through this example, we demonstrated how to use the A45 algorithm to filter out non-dominated solutions from a set of scheduling solutions with three key optimization indicators (production cost, production time, and equipment utilization), and retain the sparsely distributed scheduling solutions based on the distance between these solutions. This approach helps to find the best solution that balances different requirements in multi-objective optimization problems. Please note that in practical applications, it may be necessary to standardize the data and the number of retained solutions can be adjusted according to the specific situation.

[0053] As a specific implementation, the formula for A6 to update the position of the dragonfly is: ; in, represents the new position of the i-th dragonfly at time step t+1; represents the position of the i-th dragonfly at the current time step t; α is the control separation force D i The influence coefficient on the overall position update; D i , A i , C i , F i and R i They represent separation force, alignment force, cohesion force, food attraction force and natural enemy avoidance force respectively, and Δt is the time step. The new position of each dragonfly is calculated according to the above formula. The positions of all dragonflies are updated to complete an iteration.

[0054] Separation force D i : Calculate the distance between each dragonfly and the surrounding individuals based on the information of its neighbors to ensure that they are not too close.

[0055] Alignment force A i : Adjust speed to match the speed of neighbors, maintaining group consistency.

[0056] Cohesion C i : Move toward the center of the group to enhance group cohesion.

[0057] Food Attraction F i : Move in the direction of a food source, simulating foraging behavior.

[0058] Natural enemy avoidance Ri : Stay away from potential threats and simulate the behavior of avoiding natural enemies.

[0059] By implementing a large model-based intelligent production scheduling optimization method, enterprises can find the best scheduling solution in terms of minimizing production costs, minimizing production time, and maximizing equipment utilization. This method ensures the diversity of knowledge by retaining sparsely distributed scheduling solutions, thereby improving the flexibility and adaptability of scheduling solutions. By applying the optimal scheduling solution, enterprises have improved production efficiency, reduced costs, and increased equipment utilization.

[0060] An intelligent production scheduling optimization method based on a large model using the above optimization method is an intelligent production scheduling optimization system based on a large model, such as Figure 4 As shown, including: Data collection module, used to collect equipment operation data, material supply data, production progress data and production task information; The preprocessing module is used to clean, normalize and process outliers of the collected data to generate structured data; The model building module integrates a pre-trained large model, analyzes the nonlinear relationship in the production data through its attention mechanism, and automatically generates the optimization objectives of the scheduling model and the constraints of the production workshop; The scheduling scheme evaluation module optimizes the weight distribution and neighborhood radius parameters in the multi-objective dragonfly optimization algorithm in real time based on the dynamic parameter adjustment capability of the large model, and evaluates the preliminary scheduling scheme in combination with the entropy weight multi-objective decision-making method; The output module is used to select the optimal production scheduling solution from the Pareto optimal solution set and apply it to the production system.

[0061] Obviously, the above embodiments are merely examples for clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from them are still within the protection scope of the invention.

Claims

1. An intelligent production scheduling optimization method based on a large model, characterized in that: The following steps are involved: S1, data collection and preprocessing; Collect equipment operation data, material supply data, production progress data and production task information; Clean, normalize and process outliers of the collected data to generate structured data; S2. Determine the optimization goal of the scheduling plan and the constraints of the production workshop, extract features and learn patterns of historical production data through a large model, build a scheduling model, and generate a preliminary scheduling plan; S3, using the scheduling scheme evaluation algorithm to evaluate the preliminary scheduling scheme based on the dynamic parameter adjustment capability of the large model; S4, output the optimal production scheduling plan and apply the plan to the production system; Among them, the S3 scheduling scheme evaluation algorithm uses the entropy weight multi-objective decision-making method combined with the multi-objective dragonfly optimization algorithm to evaluate the scheduling scheme.

2. According to the large model-based intelligent production scheduling optimization method of claim 1, it is characterized in that: The S3 entropy weight multi-objective decision-making method combined with the multi-objective dragonfly optimization algorithm includes the following steps: A1. Extract multiple production samples from the preprocessed data, use the big model to generate simulation samples of future production scenarios, and standardize the production samples. Each sample represents the actual production situation within a period of time. A2. Calculate the entropy value of each optimization indicator based on the extracted sample data E j , and calculate the weight of each optimization indicator based on the entropy value ω j ; A3. Initialize the dragonfly population, generate a candidate solution set for the initial scheduling plan through the large model, and assign an initial position and speed to each dragonfly; A4. Calculate the fitness of each dragonfly and optimize the target weight ω j and the values ​​of each optimization objective f j (X i ), calculate the comprehensive fitness; the formula for calculating the comprehensive fitness is: ； in, f j ( X i ) is the i The scheduling scheme is j The value of the optimization target; ω j is the weight of each optimization objective; n is the total number of optimization objectives; A5. Use the Pareto dominance relationship to find the current non-dominated solution set as the current optimal solution; A6. For each dragonfly, calculate the contribution of separation, alignment, cohesion, food attraction, and predator avoidance behaviors based on the information of its neighbors, and update the speed and position of the dragonfly by integrating these behaviors; A7. Re-evaluate the fitness of all dragonflies and update the Pareto optimal solution set; if the fitness of the newly generated scheduling solution is better than the current non-dominated solution set S Some scheduling schemes in , then add them to the non-dominated solution set S , and remove the dominated scheduling scheme; A8. When the predetermined number of iterations G is reached, the algorithm terminates; otherwise, steps A4 to A7 are repeated; A9. Select the optimal production scheduling solution from the Pareto optimal solution set.

3. The intelligent production scheduling optimization method based on a large model according to claim 2 is characterized in that: A2 calculates the entropy value of each optimization target as follows: ； in, p jl is the jth optimization objective l is the ratio of samples, and m is the number of samples.

4. The intelligent production scheduling optimization method based on a large model according to claim 3 is characterized in that: A2 calculates the weight of each optimization indicator using the following formula: ； in, ω j is the weight of the jth optimization objective, n is the total number of optimization objectives, E j is the entropy value of each optimization indicator.

5. The intelligent production scheduling optimization method based on a large model according to claim 2 is characterized in that: A3 initialization of the dragonfly population includes: Setting the size N of the dragonfly population, that is, the number of initial scheduling solutions; the size N of the dragonfly population is dynamically adjusted by the large model according to the complexity of the production task and the learning results of the historical scheduling data; Generate a random initial position for each dragonfly X i , X i ∈[ lb , ub ]in, lb and ub are the lower and upper bounds of the decision variables, respectively; Generate a random initial speed for each dragonfly V i , V i ∈[- V max , V max ]; in, V max is the maximum speed, which is set according to the size and complexity of the production scheduling problem; Set the maximum number of iterations G of the algorithm; Set the neighborhood radius r to determine the neighbors of the dragonfly.

6. The intelligent production scheduling optimization method based on a large model according to claim 5 is characterized in that: A5 uses the Pareto dominance relationship to find the current set of non-dominated solutions, which are the current optimal solutions: A51. Initialize the non-dominated solution set; take all scheduling schemes in the population as the initial candidate solution set; A52, screening non-dominated solutions; Traverse each scheduling solution in the initial candidate solution set i , and compare it with other scheduling solutions in the candidate solution set g Make pairwise comparisons; If the scheduling plan i The fitness of the scheduling scheme is greater than g , and the scheduling scheme i Outperforms the scheduling solution on at least one key optimization objective g , then the scheduling scheme i Control scheduling plan g ; If the scheduling plan i Control scheduling plan g , then the scheduling plan g Remove from the candidate solution set; If the scheduling plan i is not dominated by any other scheduling scheme, then the scheduling scheme i Add non-dominated solution set S ; A53, maintaining the diversity of solutions; retaining sparsely distributed scheduling solutions according to the distance; A54, termination condition judgment; when the preset number of iterations G is reached, the iteration is stopped; A55. Take the final non-dominated solution set as the Pareto optimal solution set.

7. The intelligent production scheduling optimization method based on a large model according to claim 6 is characterized in that: In A52, key optimization goals include the following objectives: minimizing production costs, minimizing production time, or maximizing equipment utilization.

8. The intelligent production scheduling optimization method based on a large model according to claim 6 is characterized in that: A53 retains sparse distribution scheduling schemes including: Calculate the distance between each scheduling scheme in the non-dominated solution set; Compute any two scheduling solutions in the solution set X i and X g The Euclidean distance d between X i , X g ), the formula is: ； in, f k ( X i )and f k ( X g ) are the scheduling schemes X i and X g The fitness value at the kth optimization target, n is the total number of optimization targets; Calculate each scheduling solution X i Density ρ ( X i ), the formula is: ； in, S is the current non-dominated solution set, ρ ( X i ) represents the scheduling plan X i density; According to the density value ρ ( X i ) for non-dominated solution sets S All scheduling schemes in are sorted, and the first h scheduling schemes with the smallest density values ​​are selected as the reserved solution set, where h is the preset number of reserved solutions.

9. The intelligent production scheduling optimization method based on a large model according to claim 6 is characterized in that: A6 updates the dragonfly's position using the formula: ； in, represents the new position of the i-th dragonfly at time step t+1; represents the position of the i-th dragonfly at the current time step t; α is the control separation force D i The influence coefficient on the overall position update; D i , A i , C i , F i , R i and E enemy represent separation force, alignment force, cohesion force, food attraction force, natural enemy avoidance force and additional displacement to avoid natural enemies, respectively, and Δt is the time step.

10. An intelligent production scheduling optimization system based on a large model using an intelligent production scheduling optimization method based on a large model according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, used to collect equipment operation data, material supply data, production progress data and production task information; The preprocessing module is used to clean, normalize and process outliers of the collected data to generate structured data; The model building module integrates the pre-trained large model, analyzes the nonlinear relationship in the production data through its attention mechanism, and generates the optimization objectives of the scheduling model and the constraints of the production workshop; The scheduling scheme evaluation module optimizes the weight distribution and neighborhood radius parameters in the multi-objective dragonfly optimization algorithm in real time based on the dynamic parameter adjustment capability of the large model, and evaluates the preliminary scheduling scheme in combination with the entropy weight multi-objective decision-making method; The output module is used to select the optimal production scheduling solution from the Pareto optimal solution set and apply it to the production system.

Images