Self-adaptive intelligent scheduling optimization method for industrial production line

Abstract

The invention provides a self-adaptive intelligent scheduling optimization method for an industrial production line. The method comprises the following steps: constructing a production state model reflecting actual productivity and constraint conditions; establishing a multi-target scheduling optimization model including production efficiency, delivery date satisfaction degree and energy consumption level; based on the real-time production data, generating an optimal scheduling scheme by adopting a dynamic scheduling algorithm; an improved simulated annealing algorithm is introduced into the scheduling optimization process, and the solving efficiency is improved through a self-adaptive annealing strategy; scheduling execution data are collected for performance evaluation, and the scheduling rule base is updated in combination with a reinforcement learning mechanism, so that closed-loop feedback of scheduling optimization is realized. According to the method, collaborative optimization of multi-target scheduling performance can be realized, and the method has relatively high real-time performance and adaptivity and is suitable for intelligent scheduling requirements in a complex dynamic industrial production scene.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of industrial production scheduling, and particularly relates to an adaptive intelligent scheduling optimization method for an industrial production line. BACKGROUND

[0002] Industrial production scheduling is one of the core problems of modern manufacturing. Traditional production scheduling mainly relies on manual experience and static rules, and has problems such as low scheduling efficiency and difficulty in responding to production changes. In recent years, intelligent optimization algorithms have been widely used in production scheduling, such as genetic algorithm, particle swarm optimization, ant colony algorithm, etc., but most of them are aimed at a single scheduling target, and have high algorithm complexity and poor real-time performance. However, the existing production scheduling technology still has the following shortcomings:

[0003] (1) Most methods only consider a single scheduling target, and it is difficult to balance the multi-target requirements of production, inventory, energy consumption, etc.

[0004] (2) Traditional scheduling schemes do not consider the dynamic changes of the production process, and it is difficult to quickly adjust the scheduling strategy when equipment fails or orders are inserted.

[0005] (3) Existing scheduling algorithms lack consideration of solution speed and global optimality, and the scheduling optimization converges slowly.

[0006] Therefore, we need to develop an adaptive intelligent scheduling optimization method for an industrial production line, which can realize multi-target performance coordination optimization of production tasks, rapid adaptive response under sudden state, and efficient global optimization of scheduling schemes. SUMMARY

[0007] The purpose of the present application is to provide an adaptive intelligent scheduling optimization method for an industrial production line to solve the problems mentioned in the background art, such as incomplete consideration of existing production scheduling scheme targets, slow response to sudden changes, and slow algorithm optimization, which makes it difficult to support real-time scheduling requirements.

[0008] To achieve the above purpose, the present application provides an adaptive intelligent scheduling optimization method for an industrial production line, which is as follows:

[0009] Step S1: Through production state modeling, a mathematical model reflecting the actual production capacity and constraint conditions of the production line is constructed as a production state model;

[0010] Step S2: Based on the production state model, a scheduling optimization objective function that balances the production performance from multiple aspects including production efficiency, delivery period satisfaction, and energy consumption level is constructed to form a multi-objective scheduling optimization model;

[0011] The multi-objective scheduling optimization model is constructed by constructing each sub-objective function including a production efficiency objective function, a delivery period satisfaction objective function and an energy consumption level objective function and normalizing, and obtaining by weighted integration, as follows: , wherein are weight coefficients of production efficiency, delivery period satisfaction and energy consumption level, respectively, , , are normalized production efficiency objective function, delivery period satisfaction objective function and energy consumption objective function, respectively.

[0012] Step S3: Based on real-time production state data, an optimal production scheduling scheme is obtained by a dynamic scheduling algorithm and is issued to a production workshop for execution.

[0013] Step S4: The improved simulated annealing algorithm is introduced into the dynamic scheduling algorithm to accelerate the solving speed of the multi-objective scheduling optimization model.

[0014] Step S5: Real-time production execution data of an industrial production line based on the optimal production scheduling scheme are collected, production performance is evaluated, and production performance evaluation results are fed back to optimize scheduling parameters, forming a closed-loop optimization.

[0015] The dynamic scheduling algorithm includes the following key processes: initial scheduling, dynamic scheduling optimization and rule library updating.

[0016] The initial scheduling is to generate an initial feasible scheduling solution of a production task according to the production state model and the multi-objective scheduling optimization model.

[0017] The dynamic scheduling optimization specifically includes: setting a current time as , a scheduling period as , and a next scheduling time as ; within a time window , the system receives real-time production state data of a workshop industrial production line, dynamically updates production state parameters such as production equipment, production materials and workers, and obtains production state data at the time ; in combination with the initial feasible scheduling solution, running performance of the current scheduling solution is evaluated in real time; production state changes and running performance are input into an adaptive scheduling rule library, and a set of targeted scheduling decision rules are dynamically matched: .

[0018] Based on the foregoing scheme, the construction of the production efficiency objective function includes:

[0019] Let the production task set be , n is the total number of production tasks, the planned completion time of the production task is , the actual completion time is , and the total delay time of all production tasks is:

[0020]

[0021] Let the theoretical shortest completion time of all production tasks be , wherein is the shortest processing time of the production task , and the production efficiency objective function can be defined as: .

[0022] Based on the foregoing scheme, the construction of the delivery date satisfaction objective function includes:

[0023] An indication variable is introduced to indicate whether the production task is delayed, and the calculation formula is as follows: , the total number of all production tasks is , and the delivery date satisfaction objective function can be defined as: .

[0024] Based on the foregoing scheme, the construction of the energy consumption objective function includes:

[0025] Let the production equipment set be , the unit time energy consumption of the production equipment is , and the on-time length is , and the total energy consumption of all production equipment is: , and the energy consumption objective function is defined as: , wherein is the theoretical minimum energy consumption of the production task under the production scheduling scheme.

[0026] Based on the foregoing scheme, the multi-objective scheduling optimization model is as follows:

[0027]

[0028] Constraint conditions :

[0029] , wherein : the actual completion time of the production task , : the processing time of the th process of the production task ; , Production tasks The , The start time of each process step; Production tasks The The first process is for the first Demand for this type of resource; : No. The total available amount of this type of resource; Total number of production tasks Production tasks The total number of processes; Total number of resource types.

[0030] Based on the aforementioned scheme, and based on the real-time constraints reflected by the production state data Φ(t') at time t', the production state constraints B(t') and the selected scheduling decision rules R(t') are updated, and the multi-objective scheduling optimization model is re-solved within the current feasible scheduling solution set. neighborhood By finding the optimal solution within the set, a new non-dominated solution set is obtained. :

[0031]

[0032] Constraints :

[0033] ;

[0034] Among the constraints Corresponding to The production status at any given moment includes dynamically imposed constraints such as new equipment availability, urgent orders, and delivery date changes. for The scheduling decision rules in the process.

[0035] Based on the aforementioned scheme, the rule base update involves self-learning and updating the adaptive scheduling rule base, following the reinforcement learning model:

[0036]

[0037]

[0038] middle For changes in production status The following scheduling decision rules are adopted. The utility value function, This is the learning rate.

[0039] Based on the foregoing scheme, the simulated annealing algorithm adopts an adaptive annealing strategy, including adaptive temperature decay and adaptive neighborhood radius adjustment, to dynamically adjust annealing parameters;

[0040] The adaptive neighborhood radius adjustment is based on the acceptance rate The neighborhood radius for generating a new scheduling solution is dynamically adjusted :

[0041] (28)

[0042] wherein and are the lower limit and the upper limit of the acceptance rate respectively, and are neighborhood scaling factors, and When the acceptance rate is too low, it indicates that the neighborhood is too large, and the neighborhood radius needs to be appropriately reduced to improve search accuracy; and when the acceptance rate is too high, it indicates that the neighborhood is too small, and the neighborhood radius needs to be expanded to speed up the search.

[0043] Based on the foregoing scheme, when the optimal production scheduling scheme is actually issued to the workshop for execution, the production state data deviating from the planned state data is continuously monitored, the Mahalanobis distance between and is defined as:

[0044]

[0045] wherein, is the covariance matrix of the state variable, measuring the correlation between components; when the Mahalanobis distance breaks through a preset distance threshold, a scheduling deviation risk is timely warned and the root cause is automatically diagnosed.

[0046] The present application has the following advantages and effects relative to the prior art:

[0047] (1) A multi-objective scheduling optimization model is constructed by taking production efficiency, delivery date satisfaction, and energy consumption level as sub-targets, and after normalization processing, the weighted integration is performed to realize the coordination and unification of different performance targets, and the production efficiency, cost control, and delivery reliability are taken into account, so that the overall scheduling performance of the production line is significantly improved;

[0048] (2) An adaptive scheduling rule library based on production state data driving is introduced, the optimal rule is dynamically matched in combination with real-time state changes and scheduling performance, and through a reinforcement learning mechanism, continuous evolution is performed, so that in a complex dynamic environment, sudden changes can be quickly responded to, scheduling decisions can be dynamically adjusted, and the continuity and flexibility of production scheduling can be maintained;

[0049] (3) The simulated annealing algorithm is improved, adaptive temperature attenuation and neighborhood radius dynamic adjustment strategy are introduced, the search range and speed are flexibly controlled according to the acceptance rate, the operation efficiency is significantly improved while the scheduling solution quality is ensured, the optimization time is shortened, and the fast and accurate scheduling requirement of the industrial field is met. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the specific embodiments or prior art of the present application, the drawings needed to be used in the specific embodiments or prior art description will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual proportions.

[0051] Figure 1 is a flowchart of an adaptive intelligent scheduling optimization method for an industrial production line provided by an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to more clearly illustrate the technical solutions in the specific embodiments or prior art of the present application, the drawings needed to be used in the specific embodiments or prior art description will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual proportions.

[0053] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, many specific details are provided to give a sufficient understanding of the embodiments of the present application. However, one skilled in the art will realize that the technical solutions of the present application can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. can be used. In other cases, well-known methods, devices, implementations or operations are not shown or described in detail to avoid obscuring the aspects of the present application.

[0054] The block diagram shown in the drawings is only a functional entity, which does not necessarily correspond to a physically independent entity. That is, the functional entities can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0055] The flow chart shown in the drawing is only an exemplary illustration, not necessarily including all contents and operations / steps, and not necessarily executed in the order described. For example, some operations / steps can be further decomposed, and some operations / steps can be combined or partially combined, so that the actual execution order can be changed according to actual conditions.

[0056] The application will be described in detail below in combination with specific embodiments:

[0057] As shown in the accompanying drawings, Figure 1 Embodiment 1 of the application provides an industrial production line adaptive intelligent scheduling optimization method, and the specific steps of the method are as follows:

[0058] Step S1: Through production state modeling, a mathematical model reflecting the actual production capacity and constraint conditions of the industrial production line is constructed as a production state model to support subsequent scheduling optimization.

[0059] Preferably, through production state modeling, the actual running state of the industrial production line is comprehensively described to provide data basis for subsequent scheduling optimization, and the production state modeling includes production capacity evaluation and production constraint modeling.

[0060] Exemplarily, the production capacity evaluation is obtained by analyzing and calculating the equipment processing efficiency, standard capacity, production planning and other production factors of the industrial production line to obtain the overall actual comprehensive production capacity of the industrial production line. Specifically, for the i-th equipment in the production line, let the theoretical processing efficiency be (piece / hour), the equipment availability in the continuous production duration (hours) be , then the actual production capacity (piece) of the equipment can be represented as:

[0061] (1)

[0062] In the formula , is the standard processing efficiency given in the equipment station data, is the equipment availability considering factors such as equipment failure maintenance and quality debugging, is the production time allocated to the equipment in the current production planning cycle, which can be obtained from the production planning system.

[0063] Further, the formula is applied to all key process equipment of the industrial production line to evaluate the comprehensive production capacity of the entire industrial production line, as follows:

[0064] (2)

[0065] The formula​​ In For the comprehensive production capacity of the industrial production line, as a production capacity model, To For the actual production capacity of each key process equipment. Through comprehensive production capacity evaluation, the capacity level of the industrial production line in the current state is quantitatively described.

[0066] Exemplarily, the production constraint modeling is to build a mathematical model reflecting the actual constraint conditions of the industrial production line under comprehensive consideration of process time constraints, equipment resource constraints, process route constraints and other constraint factors, which is denoted as a production constraint model.

[0067] Specifically, the process time constraint refers to the sum of the processing time of the production task in each process cannot exceed the total production time limit, which can be expressed as:

[0068] (3)

[0069] Formula In To is the processing time of the production task through each process, is the total number of processes, is the latest completion period of the production task.

[0070] Specifically, the equipment resource constraint refers to the amount of equipment resources occupied by the same process at the same time cannot exceed the available number of equipment, which can be expressed as:

[0071] (4)

[0072] In formula (4) To is the number of the type of equipment simultaneously occupied by each process, is the number of parallel processes, is the total available amount of the type of equipment.

[0073] Specifically, the process route constraint refers to the sequence relationship of each process in the production task. Let process and process have a prior constraint, the start time and processing time of process are and , and the start time of process is , then:

[0074] (5)

[0075] Formula (5) ensures the continuity and integrity of the process route.

[0076] Further, the above various constraint conditions are integrated to establish a complete production constraint model as follows:

[0077] (6)

[0078] (7)

[0079] In formula (6)~(7) represents the first constraint equation set, is the first specific constraint equation or inequality in the first constraint, is the constraint equation set of the industrial production line.

[0080] Specific example: assuming that a production line produces product A, including three processes of milling, drilling and polishing, the processing time and the type of equipment used are: machine tool 0.5 hours, machine tool 0.3 hours and manual work station 0.2 hours. The process time constraint is that the total processing time of product A does not exceed the planned period of 8 hours, that is, 0.5 + 0.3 + 0.2 ≤ 8 hours; the equipment resource constraint is that the number of machine tools used at the same time does not exceed the total available amount of 5, that is, milling + drilling ≤ 5, the number of manual work stations used at the same time does not exceed the total available amount of 3, that is, polishing ≤ 3; the process route constraint is that the milling process must be completed before drilling, that is, milling start time + 0.5 ≤ drilling start time, and the drilling process must be completed before polishing, that is, sdrilling start time + 0.3 ≤ polishing start time. In this way, the complete production constraint model B integrates the above time constraints, resource constraints and process constraints, forming a comprehensive mathematical description of the production system.

[0081] Further, based on the production capacity model and the production constraint model, a production state model is obtained;

[0082] In summary, through production capacity evaluation and systematic modeling of production constraints, the embodiment of the present application constructs a mathematical model that comprehensively and accurately reflects the actual state of the industrial production line, i.e. the production state model. The production state model quantitatively describes the production state from multiple dimensions such as capacity evaluation, time constraints, resource constraints and process constraints, greatly improving the comprehensiveness and accuracy of production state perception compared to empirical estimation and simplified modeling, thereby laying a solid foundation for subsequent intelligent scheduling optimization.

[0083] Step S2: From multiple aspects including production efficiency, delivery period satisfaction, energy consumption, a target function of comprehensive balanced production performance is constructed to form a multi-objective scheduling optimization model.

[0084] Preferably, based on the production state model, a multi-objective scheduling model considering production performance in multiple aspects is constructed from production efficiency, delivery period satisfaction, energy consumption level, etc.

[0085] Specifically, the establishment of the multi-objective scheduling model requires defining production efficiency target function, delivery period satisfaction target function, and energy consumption level target function, each sub-target function, and performing dimensionless processing, and then integrating each sub-target function into a total scheduling optimization target function through weighting method.

[0086] For example, the production efficiency target function is defined as:

[0087] Let the production task set be , the planned completion period of the production task is , the actual completion time is , and the total delay time of all production tasks is:

[0088] (8)

[0089] Further, let the theoretical shortest completion time of all production tasks be , where is the theoretical shortest processing time of the production task , and the production efficiency target function can be defined as:

[0090] (9)

[0091] Equation (9) represents the ratio of the actual total production time to the theoretical shortest total production time. The larger the value, the higher the production efficiency, where can be calculated from equation (1) and equation (3):

[0092] (10)

[0093] where is the total number of processes of the production task , is the actual processing efficiency of the equipment used in the th process of the production task , which is a basic parameter related to the theoretical processing efficiency of the equipment uᵢ in equation (1), considering the actual running state of the equipment.

[0094] It should be noted that formula (9) describes the approximation degree of actual production task performance to optimal performance, which comprehensively considers the production equipment capacity, i.e., actual processing efficiency and operation time, and evaluates the utilization degree of the actual scheduling scheme to the equipment capacity by comparing the ratio of the theoretical shortest total production time to the actual total production time, when The closer the value is to 1, the closer the actual production time is to the theoretical shortest time, i.e., the equipment capacity is fully utilized; in the calculation formula (10) of the theoretical shortest processing time , the actual processing efficiency of the equipment is directly used The higher the processing efficiency of the equipment is, the greater the work amount completed per unit time is, and the stronger the equipment production capacity is, so that the value is smaller.

[0095] Exemplarily, a delivery date satisfaction degree objective function is defined as follows:

[0096] An indication variable of whether the production task is delayed is introduced, and the calculation formula is as follows:

[0097] (11)

[0098] The total delay number of all production tasks is , and the delivery date satisfaction degree objective function can be defined as:

[0099] (12)

[0100] Formula (12) represents the ratio of the number of production tasks completed on time to the total number of production tasks, and the greater the value is, the higher the satisfaction degree to the delivery date is.

[0101] Exemplarily, an energy consumption objective function is defined as follows:

[0102] Suppose the production equipment set is , the unit time energy consumption of the production equipment is , and the on time length is , then the total energy consumption of all production equipment is:

[0103] (13)

[0104] The energy consumption objective function can be defined as:

[0105] (14)

[0106] In formula (14) ​​The minimum energy consumption under the theoretical optimal condition for the production task. The value can be calculated by the benchmark energy consumption of the equipment under the optimal operating state, or extracted from the optimal production scheduling scheme obtained after the last scheduling period optimization. The energy consumption target reflects the closeness between the actual energy consumption level and the ideal minimum energy consumption level;

[0107] Further, the above each sub-target function is normalized to make the value range uniform to the interval [0, 1]:

[0108] (15)

[0109] Further, the above multiple sub-targets are integrated into the total scheduling optimization target function by using the weighting method as follows:

[0110] (16)

[0111] In formula (16), w1, w2 and w3 are weight coefficients of production efficiency, delivery date satisfaction degree and energy consumption level respectively, and w1+w2+w3=1. The value of the scheduling optimization target function is larger, indicating that the comprehensive performance of the scheduling scheme is better. Based on the scheduling optimization target function, the multi-objective scheduling optimization model is as follows:

[0112] (17)

[0113] Constraint condition :

[0114] Among them, : actual completion time of production task , : processing time of the first process of production task ; , : start time of the first process of production task , ; d1, d2, …, dq: demand amount of the first type of resource by the first process of production task , ; : total amount of the first type of resource available; : total number of production tasks : total number of processes of production task ; q: total number of resource types;

[0115] ​​​​​​​It should be noted that formula (17) is a multi-objective scheduling optimization model of the scheduling optimization objective function of the embodiment, wherein the scheduling optimization objective function f(x) integrates three optimization directions of production efficiency , delivery period satisfaction , energy consumption level ; the constraint conditions include but are not limited to task completion time constraint, equipment resource occupation constraint, process order constraint, and other production process constraints.

[0116] In summary, on the basis of accurate modeling of the production state, the embodiment of the application builds a mathematical model covering multiple scheduling objectives such as production efficiency, delivery reliability, and energy consumption level, forming a scheduling problem that takes into account the multi-objective optimization of the production process. The model makes full use of the production state sensing results, takes a series of quantitative indicators measuring production performance as the optimization direction, and establishes a complete closed loop from state sensing, target definition to optimization solving, which has strong systematicness and applicability. In practical application, the weight coefficients of each sub-objective can be flexibly set according to the focus of production management, balancing multiple decision preferences in the production process, improving production efficiency while considering the timeliness of product delivery and energy saving and environmental protection in the production process, and realizing global optimization of manufacturing resource allocation.

[0117] Step S3: Based on real-time production state data, an optimal production scheduling scheme is obtained through a dynamic scheduling algorithm and is issued to the production workshop for execution, thereby improving the production efficiency of key processes in the production task.

[0118] Preferably, based on the production state model in step S1 and the multi-objective scheduling optimization model in step S2, a dynamic scheduling algorithm with self-learning and self-adaptive capability is designed, the core of the algorithm is an adaptive scheduling rule base and an online feedback mechanism, the dynamic evolution of the adaptive scheduling rule base is used to realize the rapid response to the changes of real-time production state data of the workshop production line, and the adaptive update of the parameters of the dynamic scheduling algorithm is realized through the state feedback and performance feedback online feedback mechanisms, so that the scheduling performance is continuously maintained at a high level in the dynamic and complex production environment. The dynamic scheduling algorithm mainly includes three key processes: initial scheduling, dynamic scheduling optimization, and rule base updating.

[0119] Specifically, the initial scheduling is to generate an initial feasible scheduling solution of the production task according to the production state model and the multi-objective scheduling optimization model, taking the production state model of formula (6)-(7) as the constraint condition and the multi-objective scheduling optimization model of formula (17) as the optimization objective, and solving the following optimization problem by using an improved non-dominated sorting genetic algorithm:

[0120] (18)

[0121] Constraints :

[0122] in For scheduling solution The corresponding subset of constraint equations contains various strong constraints that the scheduling process must satisfy; the non-dominated solution set obtained by solving equation (18) is the initial feasible scheduling solution set. Each solution Sᵢ represents all production tasks {τ1, τ2, ..., τ} on the industrial production line. n A feasible scheduling scheme for production tasks includes the start time, processing sequence, and resource allocation scheme for each production task.

[0123] Furthermore, we proceed to the dynamic scheduling optimization. Let the current time be... The scheduling cycle is The next scheduling time is ; within the time window Internally, the system receives real-time production status data from the industrial production line in the workshop, dynamically updates production status parameters such as production equipment, production materials, and personnel, and obtains... Real-time production status data Simultaneously, the current scheduling solution is evaluated in real time, taking into account the aforementioned optimization objective. operational performance Changes in production status and operational performance The input is fed into the adaptive scheduling rule base, and a set of targeted scheduling decision rules is dynamically matched:

[0124] (19)

[0125] Specifically, the adaptive scheduling rule base is organized in the form of a three-dimensional matrix. The three dimensions of the matrix are: production status characteristics (such as machine failure, material shortage, etc.), scheduling performance characteristics (such as reduced production efficiency, increased delay risk, etc.), and decision rule dimensions (such as order insertion, line switching, resource borrowing, etc.). The matrix elements represent the applicability of the corresponding scheduling rule. This is achieved through changes in production status. By matching the analysis with the operational performance kpi(t), the set of scheduling decision rules with the highest applicability is selected: ;

[0126] Furthermore, based on the real-time constraints reflected by the production state data Φ(t') at time t', the production state constraints B(t') and the selected scheduling decision rules R(t') are updated, and the multi-objective scheduling optimization model is solved again in the current feasible scheduling solution set. neighborhood By finding the optimal solution within the set, a new non-dominated solution set is obtained. :

[0127] (20)

[0128] Constraints :

[0129] ;

[0130] Among the constraints Corresponding to The production status at any given moment includes dynamically imposed constraints such as new equipment availability, urgent orders, and delivery date changes. for The scheduling decision rules in the equation (18) are updated in terms of both the feasible solution space and constraints to quickly respond to real-time changes in production status. The optimal non-dominated solution obtained by optimization is the non-dominated solution set. The optimal scheduling scheme S(t') with the best overall performance is selected as the new scheduling decision, i.e., the optimal production scheduling scheme. Detailed scheduling instructions are generated and sent to the workshop for execution to complete this round of dynamic scheduling.

[0131] Furthermore, the rule base update involves self-learning and updating the adaptive scheduling rule base. This includes: firstly, retaining scheduling decision rules that significantly improve operational performance in the rule base through mechanisms such as roulette wheel selection, thereby enhancing their applicability; secondly, using reinforcement learning algorithms to establish a mapping relationship between scheduling performance feedback and changes in production status, uncovering new scheduling rules to supplement the rule base. Through multiple scheduling cycles, the adaptive scheduling rule base will continuously improve, and scheduling knowledge will accumulate, enabling it to make fast and accurate scheduling decisions for unknown production states. It possesses self-learning and self-optimization capabilities. The update of the adaptive scheduling rule base follows the following reinforcement learning model:

[0132]

[0133] (twenty one)

[0134] In equation (21) For changes in production status The following decision rules are adopted The utility value function, The learning rate is used. Based on the improvement in scheduling performance, the utility value function is updated using the time difference learning formula, and the production state-scheduling decision pair with the highest utility value is added to the adaptive scheduling rule base, realizing the self-evolution and updating of the adaptive scheduling rule base.

[0135] In summary, the dynamic scheduling algorithm in this embodiment takes an adaptive scheduling rule base as its core. Through the real-time production data collection and performance evaluation mechanism described in step S3, it adjusts and optimizes the model and algorithm online, and continuously learns and accumulates scheduling knowledge. Finally, it forms an intelligent scheduling system with self-learning, self-optimization, and self-adaptive capabilities, providing an effective technical means for real-time optimization scheduling of dynamic workshop production processes.

[0136] Step S4: Introduce the improved simulated annealing algorithm into the dynamic scheduling algorithm to accelerate the solution speed of the multi-objective scheduling optimization model.

[0137] Preferably, this embodiment introduces the simulated annealing algorithm into the dynamic scheduling optimization process and optimizes parameters such as the initial annealing temperature, cooling coefficient, and perturbation function of the simulated annealing algorithm to accelerate the search speed for the global optimum, reduce the risk of getting trapped in local optima, and ensure the rapid generation of high-quality scheduling schemes in dynamic production environments. The simulated annealing algorithm simulates the physical annealing process, probabilistically switching between local search and global exploration, effectively escaping local optima and ultimately converging to the global optimum with a high probability. This embodiment makes targeted improvements to the classic simulated annealing algorithm, combining a multi-objective scheduling optimization model and dynamic scheduling requirements to obtain an intelligent scheduling optimization strategy.

[0138] For example, the non-dominated solution set obtained by the dynamic scheduling optimization in step S3. as the initial solution space Then, set the initial annealing temperature. Termination temperature attenuation coefficient Parameters such as: initial annealing temperature Calculate using the following formula:

[0139] (twenty two)

[0140] In the formula The average change in the objective function value for scheduling optimization can be estimated based on problem experience; The initial acceptance probability is typically set to 0.8 to 0.9. This method for setting the initial annealing temperature takes into account both the difficulty of the problem exploration and the quality of the initial solution, providing a reasonable starting point for subsequent temperature decay.

[0141] Furthermore, a simulated annealing iterative optimization loop is executed. Based on the simulated annealing algorithm, at the current temperature T, a new scheduling solution S' is randomly selected from the neighborhood of the current scheduling solution S, and its change in the scheduling optimization objective function is calculated:

[0142] Δf(S',S) = F(S') - F(S) (23)

[0143] like Then accept As the new current scheduling solution; otherwise, with probability accept As the new current scheduling solution:

[0144] (twenty four)

[0145] Equation (24) Let S' represent the probability of accepting the new scheduling solution S'. After the new scheduling solution S' is accepted, use... Replace the current solution Otherwise, keep If unchanged, continue iteratively searching.

[0146] It should be noted that, to accommodate the needs of multi-objective optimization, the boost vector of the scheduling optimization objective function is adjusted. The calculation needs to be performed separately on each sub-objective function to form the target boost vector. :

[0147] (25)

[0148] Furthermore, in determining the acceptance of a new solution, it is necessary to... Each component is processed according to equation (24). The criterion is that a new scheduling solution is considered acceptable only if all objectives meet the acceptance criteria. Accepted. This strict acceptance criterion guarantees the Pareto optimality of the solution. At the end of each iteration, the annealing temperature is updated in an exponentially decaying manner:

[0149] (26)

[0150] In equation (26) This is the temperature decay coefficient, typically taken as 0.80~0.99. The temperature decreases proportionally, causing the simulated annealing algorithm to accept inferior solutions with a higher probability in the early stages, thus escaping local optima; while in the later stages, it accepts inferior solutions with a lower probability, accelerating the local search speed.

[0151] Furthermore, this embodiment introduces an adaptive annealing strategy, which mainly includes adaptive temperature decay and adaptive neighborhood radius adjustment. The annealing parameters are dynamically adjusted according to the improvement effect of the scheduling solution during the optimization process.

[0152] Specifically, the adaptive temperature decay dynamically adjusts the temperature decay coefficient based on the scheduling solution improvement effect. :

[0153] (27)

[0154] in, The mean of the scheduling optimization objective function at the current temperature is to optimize the scheduling. The number of iterations at the current temperature. This is an adjustment factor. If the objective function value of scheduling optimization improves significantly at the same temperature, the temperature decay rate should be appropriately slowed down to increase the search intensity; conversely, the temperature decrease should be accelerated to reduce ineffective searches.

[0155] The adaptive neighborhood radius adjustment is based on the acceptance rate. Dynamically adjust the neighborhood radius generated by the new scheduling solution :

[0156] (28)

[0157] in and These are the lower and upper limits of the acceptance rate, respectively. and Let be the neighborhood scaling factor, and When the acceptance rate If the acceptance rate is too low, it indicates that the neighborhood is too large, and the neighborhood radius needs to be appropriately reduced to improve search accuracy; while if the acceptance rate is too high, it indicates that the neighborhood is too small, and the neighborhood radius needs to be expanded to speed up the search.

[0158] The termination condition for the simulated annealing algorithm can be set as: the current temperature. Below the termination temperature The optimal solution does not improve after several consecutive temperature decreases. Upon reaching the termination condition, the globally optimal scheduling solution is output. This serves as the optimal production scheduling scheme.

[0159] For example, the scheduling optimization process based on the simulated annealing algorithm can be summarized as follows:

[0160] Input initial solution space Set annealing parameters , , wait;

[0161] Randomly select the initial scheduling solution Set the current scheduling solution Current temperature ;

[0162] Repeat the following steps until the termination condition is met:

[0163] a) in A new scheduling solution is randomly selected from the neighborhood. ;

[0164] b) Calculate the objective function lifting vector ,according to criteria to determine whether to accept the new scheduling solution ;

[0165] c) if the new scheduling solution is accepted , ; otherwise, keep the same ;

[0166] d) update the historical optimal solution, i.e., the global optimal scheduling solution ;

[0167] e) adaptively adjust the annealing temperature and the neighborhood radius ; ;

[0168] output the global optimal scheduling solution as the optimal production scheduling scheme.

[0169] Further, embedding the above simulated annealing optimization process into the dynamic scheduling algorithm of step S3 can realize the fusion and promotion of global optimization capability and dynamic adaptive capability. Each scheduling cycle forms an annealing optimization sub-cycle, and global optimization is performed under the dynamically updated production state constraints. At the same time, the scheduling decision, i.e., the optimal production optimization scheme, can also feedback to guide the annealing optimization of the next cycle, so that the annealing parameters can be adjusted online to adapt to the dynamically changing scheduling environment. The synergistic enhancement of scheduling optimization and online learning is the core of the embodiments of the present application.

[0170] Step S5: Real-time collection of production execution data of the industrial production line based on the optimal production scheduling scheme, evaluation of production performance, and feedback of production performance evaluation results to optimize scheduling parameters, forming a closed-loop optimization.

[0171] Specifically, the closed-loop optimization is realized based on the execution monitoring and performance feedback of the scheduling scheme. Based on the optimal production scheduling scheme obtained in steps S3 and S4, the embodiments of the present application construct a complete scheduling execution monitoring system and performance evaluation feedback mechanism to implement active control of the optimal production scheduling scheme execution process from three dimensions of real-time sensing, online evaluation, and dynamic optimization.

[0172] Illustratively, the scheduling execution monitoring system is used to collect production execution data in real time, specifically including:

[0173] A real-time sensing network of workshop production running state is established. A large amount of heterogeneous data is collected in real time through deployment of sensors, intelligent gateways and other Internet of Things devices at key process procedures, such as equipment processing, material transfer, and personnel operation. After preprocessing of the collected raw data, such as cleaning, fusion, and semantic annotation, structured production state data is formed to depict the workshop in Real-time production running state:

[0174] (29)

[0175] wherein is the process of the production task is the state feature vector at the moment, containing multi-source heterogeneous information , respectively representing the number of idle, processing, and fault devices; respectively, the number of waiting, in-process, and completed materials in the first process, the first position; respectively, the production cycle, quality pass rate, and unit energy consumption of the process , and other statistical quantities. Through the association and fusion of heterogeneous data, the state feature vector comprehensively depicts the real-time production running state of the process , and makes the state data stream quantitative to present the real-time production picture of each process and even the whole plant.

[0176] Further, the performance evaluation feedback mechanism is used to evaluate the production performance and feedback the production performance evaluation results to optimize the scheduling parameters, specifically including:

[0177] Based on the real-time sensing network, a virtual simulation system conforming to the physical workshop is built through emerging technologies such as digital twinning and virtual assembly. Through real-time state mapping, the digital twinning model of the workshop is reconstructed online. The optimal production scheduling scheme is input into the digital twinning model to realize virtual execution and online estimate various scheduling performance indicators of the optimal production scheduling scheme:

[0178] (30)

[0179] wherein is the global optimal scheduling solution, are the online estimated values of production efficiency, delivery date satisfaction, and energy consumption level. The digital twinning model acts as a "preview" role for scheduling execution, which can pre-evaluate the performance of the optimal production scheduling scheme before it is issued to the workshop for execution, discover potential problems in time, and feedback optimization.

[0180] Preferably, when the global optimal scheduling solution , i.e., the optimal production scheduling scheme is actually issued to the workshop for execution, the deviation of the workshop production state data from the planned state data is continuously monitored, and the Mahalanobis distance between and is defined as: ​

[0181] (31)

[0182] wherein, is the covariance matrix of state variables, measuring the correlation between each component. Mahalanobis distance considers the statistical correlation between state variables, and can more sensitively capture the multi-dimensional changes of the state between vehicles. When the Mahalanobis distance breaks through the preset distance threshold, the deviation risk of scheduling is timely warned and the root cause is automatically diagnosed.

[0183] Further, according to the actual running state , the actual achievement level of each scheduling performance indicator is evaluated as the actual performance:

[0184] (32)

[0185] Further, the actual performance of formula (32) is compared with the online estimated value of formula (30), and the performance deviation is calculated:

[0186] (33)

[0187] Further, combined with static threshold alarm and dynamic trend early warning, a stereoscopic performance evaluation of the optimal production scheduling scheme is formed as the production performance evaluation result; the static threshold alarm is a threshold detection on the components of the performance deviation , identifying the key performance indicators that do not meet the expectations; the dynamic trend early warning is a trend prediction on the time series of the performance deviation , judging the dynamic trend of improvement or deterioration of each indicator.

[0188] Further, the production performance evaluation result is fed back to the dynamic scheduling algorithm in step S3 in a timely manner, driving the online optimization of scheduling parameters and the optimal production scheduling scheme, specifically including the following two feedback optimization paths:

[0189] For example, the state feedback driven scheduling trigger optimization. When the Mahalanobis distance or the performance indicator deterioration trend occurs, a re-scheduling optimization event is triggered in time to trigger the dynamic scheduling optimization process in step S3, generate a new optimal production scheduling scheme, and timely stop loss to reduce further deviation:

[0190] (34)

[0191] wherein This is a preset Mahalanobis distance threshold, indicating when an alert or adjustment will be triggered if the distance exceeds this value; This is the warning threshold for the i-th performance indicator; a negative sign indicates a decline in the performance indicator.

[0192] An example is performance feedback-driven parameter self-optimization. This involves optimizing parameters for each scheduling cycle. Performance deviation The accumulated time series data is input into the adaptive learning module of the scheduling system to uncover the intrinsic relationships between scheduling parameters, workshop status, and scheduling performance, and to incrementally optimize the scheduling rule base and algorithm parameters.

[0193] (35)

[0194] In equation (35) Let be the parameter vector of the algorithm to be optimized, and let be the optimization objective to minimize the performance deviation. norm By introducing performance feedback, the scheduling optimization process forms a closed loop, enabling the system to continuously learn from historical scheduling experience. The algorithm parameters can continuously adapt in a dynamic environment, improving the robustness and foresight of the scheduling system.

[0195] In summary, step S5 established a monitoring and feedback mechanism covering the entire scheduling execution process, achieving proactive closed-loop control from status perception and performance evaluation to optimization decision-making. This mechanism maximizes the use of real-time collected workshop operation data, accurately diagnoses scheduling execution deviations, and achieves bidirectional collaborative optimization of scheduling decisions and online learning through status feedback and performance feedback, forming a data-driven dynamic scheduling optimization closed loop. While improving the accuracy and sensitivity of scheduling execution, it also enhances the adaptability and anti-interference capability of scheduling decisions to unknown and complex working conditions, making it a necessary guarantee for intelligent scheduling. The execution monitoring and feedback mechanism in this embodiment breaks through the limitations of traditional static scheduling and open-loop control, providing a new and effective dynamic closed-loop scheduling and control model for new CPS workshops.

[0196] In the embodiment, firstly, the production state model is constructed by capacity evaluation and resource constraint modeling; secondly, a multi-objective scheduling optimization model is established, taking production efficiency, delivery period satisfaction and energy consumption level as core indexes, and the unified expression of the objective function is realized through normalization and weighted fusion. In the scheduling optimization process, a dynamic scheduling algorithm based on real-time state data driving is adopted, and the scheduling scheme is dynamically generated by combining real-time matching of the scheduling rule base; further, the scheduling rule base is updated online by using the reinforcement learning method, so as to improve its adaptability and effectiveness. In addition, the improved simulated annealing algorithm is fused, and the adaptive temperature attenuation and neighborhood adjustment mechanism are adopted to improve the global search efficiency. The closed-loop feedback part is also included, the scheduling execution result is monitored and evaluated, when the scheduling deviation occurs, the scheduling optimization process is automatically triggered, and the online learning and continuous optimization of the scheduling system are realized. The method described in the embodiment can improve the real-time response capability, performance achievement rate and stability of the scheduling process.

[0197] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the specification and examples be considered as exemplary only, with the true scope and spirit of the application being indicated by the following claims. It is to be understood that the application is not to be limited to the precise details of apparatus and operation described herein, and that various modifications and changes can be made to the embodiments described without departing from the scope and spirit of the application. The scope of the application is to be indicated by the appended claims rather than the description and figures.

Claims

1. An adaptive intelligent scheduling and optimization method for industrial production lines, characterized in that, include: Step S1: By modeling the production status, construct a mathematical model that reflects the actual production capacity and constraints of the production line, as the production status model; Step S2: Based on the production state model, construct a scheduling optimization objective function that comprehensively balances production performance from multiple aspects, including production efficiency, delivery time satisfaction, and energy consumption level, to form a multi-objective scheduling optimization model; The multi-objective scheduling optimization model is obtained by first constructing and normalizing sub-objective functions, including production efficiency objective function, delivery time satisfaction objective function, and energy consumption level objective function, and then integrating them through weighted summation, as shown in the following formula: ,in These are the weighting coefficients for production efficiency, on-time delivery satisfaction, and energy consumption level, respectively. , , These are the normalized production efficiency objective function, delivery time satisfaction objective function, and energy consumption objective function, respectively. Step S3: Based on real-time production status data, obtain the optimal production scheduling plan through a dynamic scheduling algorithm and send it to the production workshop for execution; Step S4: Introduce the improved simulated annealing algorithm into the dynamic scheduling algorithm to accelerate the solution speed of the multi-objective scheduling optimization model; Step S5: Collect production execution data of the industrial production line based on the optimal production scheduling scheme in real time, evaluate production performance, and feed back the production performance evaluation results to optimize scheduling parameters and form a closed-loop optimization. The dynamic scheduling algorithm includes the following key processes: initial scheduling, dynamic scheduling optimization, and rule base update. The initial scheduling is based on the production state model and the multi-objective scheduling optimization model to generate an initial feasible scheduling solution for the production tasks. The dynamic scheduling optimization specifically includes: assuming the current time is... The scheduling cycle is The next scheduling time is ; within the time window Internally, the system receives real-time production status data from the industrial production line in the workshop, dynamically updates production status parameters such as production equipment, production materials, and personnel, and obtains... Real-time production status data Based on the initial feasible scheduling solution, the current scheduling solution is evaluated in real time. operational performance Changes in production status and operational performance The input is fed into the adaptive scheduling rule base, and a set of targeted scheduling decision rules is dynamically matched: .

2. The adaptive intelligent scheduling and optimization method for industrial production lines according to claim 1, characterized in that, The construction of the production efficiency objective function includes: Let the set of production tasks be... n is the total number of production tasks. The planned completion date is The actual completion time is Therefore, the total delay time for all production tasks is: ； Let the theoretical shortest completion time for all production tasks be... ,in For production tasks If the minimum processing time is given, then the production efficiency objective function can be defined as: .

3. The adaptive intelligent scheduling and optimization method for industrial production lines according to claim 2, characterized in that, The construction of the objective function for delivery time satisfaction includes: Introducing production tasks Indicator variable for whether to postpone The calculation formula is as follows: Then the total delay for all production tasks is Then the objective function for delivery date satisfaction can be defined as: .

4. The adaptive intelligent scheduling and optimization method for industrial production lines according to claim 2, characterized in that, The construction of the energy consumption objective function includes: Let the set of production equipment be... Production equipment Energy consumption per unit time is The boot time is Then the total energy consumption of all production equipment is: The energy consumption objective function is defined as: ,in This represents the theoretical minimum energy consumption of a production task under a production scheduling scheme.

5. The adaptive intelligent scheduling and optimization method for industrial production lines according to claim 4, characterized in that, The multi-objective scheduling optimization model is as follows: Constraints : ,in Production tasks The actual completion time, Production tasks The The processing time for each step; , Production tasks The , The start time of each process step; Production tasks The The first process is for the first Demand for this type of resource; : No. The total available amount of this type of resource; Total number of production tasks Production tasks The total number of processes; Total number of resource types.

6. The adaptive intelligent scheduling optimization method for industrial production lines according to claim 1, characterized in that, Based on the real-time constraints reflected by the production state data Φ(t') at time t', update the production state constraints B(t') and the selected scheduling decision rules R(t'), and resolve the multi-objective scheduling optimization model. Then, within the current feasible scheduling solution set... neighborhood By finding the optimal solution within the set, a new non-dominated solution set is obtained. : Constraints : Among them, the constraints Corresponding to The production status at any given moment includes dynamically imposed constraints such as new equipment availability, urgent orders, and delivery date changes. for The scheduling decision rules in the process.

7. The adaptive intelligent scheduling optimization method for industrial production lines according to claim 1, characterized in that, The rule base update involves self-learning and updating the adaptive scheduling rule base, following the reinforcement learning model: middle For changes in production status The following scheduling decision rules are adopted. The utility value function, This is the learning rate.

8. The adaptive intelligent scheduling optimization method for industrial production lines according to claim 6, characterized in that, The simulated annealing algorithm employs an adaptive annealing strategy, including adaptive temperature decay and adaptive neighborhood radius adjustment, to dynamically adjust the annealing parameters; The adaptive neighborhood radius adjustment is based on the acceptance rate. Dynamically adjust the neighborhood radius generated by the new scheduling solution : ,in and These are the lower and upper limits of the acceptance rate, respectively. and Let be the neighborhood scaling factor, and When the acceptance rate If the acceptance rate is too low, it indicates that the neighborhood is too large, and the neighborhood radius needs to be appropriately reduced to improve search accuracy; while if the acceptance rate is too high, it indicates that the neighborhood is too small, and the neighborhood radius needs to be expanded to speed up the search.

9. The adaptive intelligent scheduling and optimization method for industrial production lines according to claim 1, characterized in that, Once the optimal production scheduling plan is actually issued to the workshop for execution, the workshop production status data is continuously monitored. Deviation from Plan Status Data The degree, definition and The Mahalanobis distance is: ,in, The covariance matrix of the state variables measures the correlation between the components; when Mahalanobis distance... When the preset distance threshold is exceeded, the system will promptly issue a warning of scheduling deviation risk and automatically diagnose the root cause.

Images