Universal intelligent scheduling method and system for intelligent beam field

Abstract

The invention discloses a universal intelligent scheduling method and system for an intelligent beam field, which can effectively extract and utilize characteristics in a precast beam field scheduling problem by combining the advantages of deep reinforcement learning and the characteristics of a multi-pointer graph network and fully utilizing graph structure information in the scheduling problem so as to realize efficient solution of the precast beam field scheduling problem. And corresponding scheduling problems can be adaptively processed according to different problem scales, so that the method has good generalization performance and can achieve a better effect in large-scale and complex precast beam field practical application.

Description

Technical Field

[0001] The present invention relates to the technical field of precast beam yard scheduling. More specifically, the present invention relates to a universal intelligent scheduling method and system for a smart beam yard. Background Art

[0002] In modern bridge construction, production efficiency and quality control at the beam yard (i.e., the production site for precast and prefabricated beams) are critical factors in ensuring the smooth progress of the project. Existing beam yard scheduling methods primarily rely on manual experience or simple planning software. This reliance on manual experience is susceptible to human factors, which can lead to inaccurate scheduling and affect production efficiency. Furthermore, the lack of an effective resource scheduling mechanism results in inefficient utilization of equipment and human resources, making comprehensive and real-time quality control during the production process difficult, impacting the quality of the final product. Existing scheduling systems often fail to effectively integrate with other systems, resulting in information silos and hindering overall coordination and management efficiency.

[0003] While existing scheduling methods based on deep reinforcement learning can handle scheduling problems in general workshops, beam yards, and other production locations, they are difficult to directly apply to the complex on-site scheduling requirements and constraints of precast beam yards. Uncertainty, in particular, increases the challenges of feature extraction and scheduling decisions. Summary of the Invention

[0004] An object of the present invention is to solve at least the above problems and to provide at least the advantages which will be described hereinafter.

[0005] Another object of the present invention is to provide a general intelligent scheduling method and system for a smart beam yard to solve the technical problem of insufficient overall scheduling capacity of a prefabricated beam yard in the existing technology.

[0006] In order to achieve these purposes and other advantages according to the present invention, a general intelligent scheduling method for a smart beam yard is provided, comprising the following steps:

[0007] S1. Mathematically model the production scheduling problem of the precast beam yard to obtain a mathematical model for the precast beam yard scheduling;

[0008] S2. Based on all relevant data on precast beam yard scheduling, randomly generate a precast beam yard scheduling data set that includes various uncertain factors;

[0009] S3. Construct a multi-pointer graph network model and use it to extract and encode features of the precast beam yard scheduling dataset;

[0010] S4. Convert the mathematical model of precast beam yard scheduling into a Markov decision process, design the state space, action space, reward function and parameter strategy, build a simulation scheduling environment, and form the precast beam process sub-strategy and pedestal sub-strategy;

[0011] S5. Constructing a precast beam decoder and a pedestal decoder based on a multi-layer perceptron. According to the encoding result of step S3, the decoder calculates the probability distribution in the precast beam process action space and the pedestal action space respectively.

[0012] S6. Adopting the actor-critic architecture, using the multi-proximal strategy optimization algorithm combined with the probability distribution of step S5 to learn the two sub-strategies of step S4, iteratively optimize, and obtain the scheduling decision model;

[0013] S7. According to the actual business scenario of precast beam yard scheduling, the scheduling action to be taken next is given based on the scheduling decision model, the precast beam yard scheduling problem is solved, and the scheduling Gantt chart and scheduling plan corresponding to the actual business scenario are generated.

[0014] Preferably, the specific steps of performing mathematical modeling in step S1 are as follows:

[0015] S11. Define the precast beam yard scheduling problem, including the duration and completion time of each process;

[0016] S12. In the precast beam yard scheduling problem, determine the objective function as minimizing the maximum completion time;

[0017] S13. Describe and define the constraints of the precast beam yard scheduling problem, including:

[0018] Each process of prefabricating a beam has a sequence;

[0019] The completion time of each precast beam cannot exceed the total completion time;

[0020] Only one process can be processed on the same pedestal at the same time;

[0021] The same process can only be processed by one pedestal at the same time;

[0022] There can be a cycle operation on each pedestal.

[0023] Preferably, the specific steps of step S2 are as follows:

[0024] S21. Determine the problem scale of the precast beam yard scheduling problem including various uncertainties, the number of tasks j, the number of pedestals m, and the time required for each task process;

[0025] S21. For each process of each task, generate a real number to represent the time required for each process;

[0026] S22. For all beams to be prefabricated of all tasks and all available pedestals, generate data matching all processes of all beams to be prefabricated and all available pedestals to obtain a prefabricated beam yard scheduling data set.

[0027] Preferably, the specific steps of step S3 are as follows:

[0028] S31. Design a network structure consisting of a graph isomorphism network, a multi-layer perceptron, and task pointers to obtain the characteristics of the prefabricated beam process:

[0029]

[0030] in, is the embedding of node v in the l-th layer graph isomorphism network, is the multilayer perceptron at layer l, θ l is the parameter of the layer, ∈ (l) represents the parameters that need to be learned, t represents the time step, Indicates the total number of processes;

[0031] S32. Each node in the pedestal state information represents the main features of each pedestal, and the fully connected layer is used to encode the local state.

[0032] Preferably, the specific steps of step S4 are as follows:

[0033] S41. Define and design the state space. The state of the environment at time t is determined by the local state. and composition, Indicates the status of the precast beam process. Indicates the status of the pedestal;

[0034] S42. Define and design the action space. The action space at time t includes the prefabricated beam process action a. o and pedestal action a m ;

[0035] S43. Define the transfer space. At time t, the agent samples a process and selects a pedestal for the process. Then, the direction of the disjunctive arc is updated according to the current precast beam process action and the pedestal action. The newly generated disjunctive graph is used as the new local state. The pedestal status update is completed by the pedestal at time T t+1 (M k ) and prefabricated beam process a o Processing time p ijk Complete together and get the new state of the pedestal

[0036] S44. Design and define the reward function. The objective function is to minimize the maximum completion time of all precast beams. max , set the difference between the solution at time t+1 and time t to d t =C(s t+1 )-C(s t ), set the reward to -d t ;

[0037] S45. Set strategy, total strategy π θ (a o ,a m |s)Precast beam process sub-strategy and pedestal strategies Composition, where θ0 and θ m They represent the parameters of the prefabricated beam process sub-strategy and the pedestal sub-strategy, respectively. The total strategy π θ (a o ,a m |s) Select the precast beam process action and the pedestal action according to the probability distribution of the precast beam process action space and the pedestal action space respectively.

[0038] Preferably, the specific steps of step S5 are as follows:

[0039] S51. Construct a prefabricated beam decoder based on a multi-layer perceptron and use the decoder to calculate the prefabricated beam process action score.

[0040] S52. Convert the prefabricated beam process action scores into probabilities by performing Softmax normalization on the scores.

[0041] S53. Build a pedestal decoder based on a multi-layer perceptron and use the decoder to calculate the pedestal action score.

[0042]

[0043] S54. Convert the pedestal action scores into probabilities by performing Softmax normalization on the scores

[0044] Preferably, the specific steps of step S6 are as follows:

[0045] S61. Input parameters, including:

[0046] Number of training sessions E t , the number of updates of the multi-proximal strategy optimization algorithm E s , the number of samples B used in a single training, training actor network Behavior Actor Network And the parameters θ corresponding to the two networks h 、 Critics Network v φ And the corresponding parameter φ, the strategy loss coefficient c p , the value function loss coefficient c v , entropy loss coefficient c e , shear parameter∈;

[0047] S62, initialize relevant parameters, including parameters θ for training the actor network h and the critic network parameter φ;

[0048] S63, randomly sampling from a uniform distribution, using a multi-proximal strategy optimization algorithm to learn the precast beam process sub-strategy and the pedestal sub-strategy;

[0049] S64, use the state-value function to evaluate the value of the two sub-strategies learned in S63, by inputting the state and outputting the value;

[0050] S65, after multiple iterations of training, output the trained prefabricated beam process actor network parameters θ0 and the pedestal actor network parameters θ m , and obtain the scheduling decision model.

[0051] Preferably, the specific steps of step S7 are as follows:

[0052] S71. Convert the actual erection requirements of precast beams, the number of pedestals involved in the business scenario, and actual business constraints into mathematical symbols and corresponding formulas.

[0053] S72. Based on the scheduling decision model completed through training optimization, input data including the scheduling start and end dates, the total number of prefabricated beams to be scheduled, the total number of pedestals, and the beam type, substitute the data into the scheduling decision model for calculation, and automatically generate the scheduling Gantt chart and scheduling plan table for the corresponding period.

[0054] Preferably, for the scheduling decision model, the definition of the mathematical symbols includes:

[0055] Process set P = {p1, p2, ..., p O}, where O represents the total number of processes;

[0056] Process path set P ij , represents the process p i On which machines can it be processed?

[0057] Machine set M={m1,m2,...,m n}, where n represents the total number of machines;

[0058] Processing time t ij , indicating that on machine mj Upper processing step p i the time required;

[0059] The processing sequence S={s1,s2,...,s m}, indicating the processing sequence constraints between processes;

[0060] The number of pedestals k represents the number of available pedestals;

[0061] Base number K = {k1, k2, ..., k k};

[0062] The start time of the process on the pedestal Represents process p i on pedestal k j Start time on

[0063] End time of the process on the pedestal Represents process p i on pedestal k j End time on;

[0064] Precondition constraint d ij , represents the process p i The prerequisite p that must be met before completion j completion time;

[0065] The objective function is defined as follows:

[0066] The first objective function is to minimize the delivery cycle Minimize C max ,

[0067] Among them, C max represents the maximum completion time of all processes.

[0068] The second objective function is to maximize the utilization of the platform.

[0069]

[0070] in, Indicates the total processing time of all processes on all platforms, Indicates the total available time of all pedestals;

[0071] Comprehensive objective function Minimize f(x) = w1·C max -w2·(1-Utilization);

[0072] The constraints considered in actual production planning include:

[0073] Timing constraints: the sum of the precast beam start time and processing time must be less than the completion time, the end time of the previous process must be less than the start time of the next process, and the completion time of each precast beam must be before the total completion time;

[0074] Due to resource constraints, each pedestal can only process one process at a time;

[0075] Sequential constraints: different processes should have a strict processing order;

[0076] Flow constraints: the sequential movement, parallel movement and smooth movement of the pedestal must comply with the set route.

[0077] On the other hand, the present invention also provides a general intelligent scheduling system for a smart beam yard, comprising an input layer, a strategy layer and an output layer, wherein the input layer comprises an information preprocessing module, the strategy layer comprises a mathematical modeling module, an algorithm solving module and a deduction module, and the output layer comprises a result visualization module, wherein:

[0078] The preprocessing module is used to input and read the input total production plan, change information, number of pedestals, process types supported by each pedestal, parameters and quantity of beams to be produced, process type, number of processes, and process time data for each beam production process, standardize the input data format to form information data, and transmit it to the mathematical modeling module;

[0079] The mathematical modeling module is used to read information data, perform mathematical modeling based on actual business scenarios, define mathematical symbols, build objective functions, establish actual constraints, and sequentially build a precast beam yard scheduling mathematical model, a multi-pointer graph network model, and a scheduling decision model;

[0080] The algorithm solving module is used to determine the next scheduling action to be taken based on the scheduling decision model according to the actual business scenario of precast beam yard scheduling, complete the solution of the precast beam yard scheduling problem, and transmit the calculation results to the deduction module;

[0081] The deduction module is used to place the scheduling sequences in the calculation results on the idle production resources in chronological order according to the actual production resources, constraints and calculation results returned by the algorithm solution module, simulate the processing of the scheduling sequences and analyze the resource consumption and time consumption of all scheduling sequences under all constraints, and perform strategy optimization. The optimized results are used as the final scheduling layout plan, and the final scheduling layout plan is sent to the result visualization module. At the same time, the change information after the implementation of the final scheduling layout plan is sent to the preprocessing module for update;

[0082] The result visualization module is used to read the final scheduling plan and generate the corresponding Gantt chart for visual display.

[0083] The present invention includes at least the following beneficial effects: the general intelligent scheduling method and system of the smart beam yard of the present invention are based on the actual production scheduling problem scenario description and problem modeling of different beam production lines, and through the general scheduling system architecture with the prefabricated beam yard scheduling method based on deep reinforcement learning and multi-pointer graph network as the core, it optimizes the matching of production resources and scheduling results, takes minimizing delivery cycle and maximizing platform utilization as optimization goals, and comprehensively considers potential constraints to coordinate the overall scheduling plan. The planning decision obtained by the scheduling system has high solution efficiency and can ensure the solution quality, effectively improve the effectiveness and executability of the scheduling series production plan, reduce the investment of labor costs, improve production efficiency, and shorten the processing cycle; and the universal system architecture has strong scalability, the solution process is explainable, the solution process and calculation logic are easy to be understood by scheduling staff, easy to implement, and can be extended to the beam production planning links of different processes and process flows, and has universality.

[0084] Other advantages, objectives and features of the present invention will be reflected in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 This is a diagram showing the overall architecture of the precast beam yard scheduling method and system of the present invention;

[0086] Figure 2 This is a diagram of the multi-pointer graph network architecture of the present invention;

[0087] Figure 3 This is a diagram of the multi-Actor-Critic architecture for the multi-action space scheduling problem of the present invention;

[0088] Figure 4 A Gantt chart showing the scheduling results of a precast beam yard according to an embodiment of the present invention;

[0089] Figure 5 This is a partial scheduling result plan corresponding to the scheduling Gantt chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0090] The present invention will be described in further detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0091] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials are commercially available unless otherwise specified; in the description of the present invention, the terms "horizontal", "longitudinal", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present invention.

[0092] like Figure 1-5 As shown, on the one hand, the present invention provides a general intelligent scheduling method for a smart beam yard, comprising the following steps:

[0093] S1. Mathematically model the production scheduling problem of the precast beam yard to obtain a mathematical model for the precast beam yard scheduling;

[0094] S2. Based on all relevant data on precast beam yard scheduling, randomly generate a precast beam yard scheduling data set that includes various uncertain factors;

[0095] S3. Construct a multi-pointer graph network model and use it to extract and encode features of the precast beam yard scheduling dataset;

[0096] S4. Convert the mathematical model of precast beam yard scheduling into a Markov decision process, design the state space, action space, reward function and parameter strategy, build a simulation scheduling environment, and form the precast beam process sub-strategy and pedestal sub-strategy;

[0097] S5. Constructing a precast beam decoder and a pedestal decoder based on a multi-layer perceptron. According to the encoding result of step S3, the decoder calculates the probability distribution in the precast beam process action space and the pedestal action space respectively.

[0098] S6: Using the actor-critic architecture, the multi-proximal strategy optimization algorithm is combined with the probability distribution of step S5 to learn the two sub-strategies of step S4, that is, the neural network model constructed by training. After evaluating the network and iterative optimization, the scheduling decision model is obtained;

[0099] S7. According to the actual business scenario of precast beam yard scheduling, the scheduling action to be taken next is given based on the scheduling decision model, the precast beam yard scheduling problem is solved, and the scheduling Gantt chart and scheduling plan corresponding to the actual business scenario are generated.

[0100] The universal intelligent scheduling method for smart beam yards of the present invention combines the advantages of deep reinforcement learning and the characteristics of multi-pointer graph networks, fully utilizes the graph structure information in the scheduling problem, can effectively extract and utilize the features of the prefabricated beam yard scheduling problem, and realize efficient solution to the prefabricated beam yard scheduling problem. In addition, it can adaptively handle the corresponding scheduling problems according to different problem scales, has good generalization performance, and can achieve better results in the actual application of large-scale and complex prefabricated beam yards.

[0101] In another technical solution, Figure 1 As shown, the specific steps of mathematical modeling in step S1 are as follows:

[0102] S11. Determine the definition of the precast beam yard scheduling problem, including the duration and completion time of each process;

[0103] S12. In the precast beam yard scheduling problem, the objective function is to minimize the maximum completion time minC max ;

[0104] S13. Describe and define the constraints of the precast beam yard scheduling problem, including:

[0105] Each process of prefabricating a beam has a sequence;

[0106] The completion time of each precast beam cannot exceed the total completion time;

[0107] Only one process can be processed on the same pedestal at the same time;

[0108] The same process can only be processed by one pedestal at the same time;

[0109] There can be a cycle operation on each pedestal.

[0110] During the mathematical modeling process, various related data are defined, such as the total production plan, the number of beam models to be produced, the number of beams to be produced corresponding to different models, the process types and number of processes for each beam production, the time consumption of each different process, the number of pedestals, the process types that each pedestal can support, planned resource change information, etc., so as to more comprehensively grasp the information data related to the actual construction scheduling of the beam yard, standardize the format, realize information integration, and improve the accuracy of subsequent scheduling in line with the actual business environment.

[0111] In another technical solution, Figure 1 As shown, the specific steps of step S2 are as follows:

[0112] S21. Determine the problem scale of the precast beam yard scheduling problem, including various uncertainties, as well as the number of tasks j, the number of pedestals m, and the time required for each task process; in order to better capture the complex characteristics of the data, improve the prediction accuracy of the model, and control computing resources and time.

[0113] S21. For each process of each task, generate a real number to represent the time required for each process.

[0114] S22. For all beams to be prefabricated in all tasks and all available pedestals, generate data matching all processes of all beams to be prefabricated and all available pedestals, and obtain a prefabricated beam yard scheduling data set to facilitate the subsequent solution of the prefabricated beam yard scheduling problem.

[0115] In another technical solution, Figure 1-2 As shown, the specific steps of step S3 are as follows:

[0116] S31. Design a network structure consisting of a graph isomorphism network, a multi-layer perceptron, and task pointers to obtain the characteristics of the prefabricated beam process:

[0117]

[0118] in, is the embedding of node v in the l-th layer graph isomorphism network, is the multilayer perceptron at layer l, θ l is the parameter of the layer, ∈ (l) represents the parameters that need to be learned, t represents the time step, Indicates the total number of processes;

[0119] S32. Each node in the pedestal state information represents the main features of each pedestal, and the fully connected layer is used to encode the local state.

[0120] Using node classification instead of traditional sequence modeling methods greatly reduces the computational complexity. Multi-task pointer feature fusion can improve the accuracy and robustness of the model and enhance its adaptability to complex application scenarios.

[0121] In another technical solution, Figure 1 As shown in Figure 3, the mathematical model of precast beam yard scheduling is transformed into a Markov decision process, which effectively handles the temporal decision problem and reduces the modeling complexity by simplifying the environmental model. The specific steps of step S4 are as follows:

[0122] S41. Define and design the state space. The state of the environment at time t is determined by the local state. and composition, Indicates the status of the precast beam process. Indicates the status of the pedestal;

[0123] S42. Define and design the action space. The action space at time t includes the prefabricated beam process action a. o and pedestal action a m ;

[0124] S43. Define the transfer space. At time t, the agent samples a process and selects a pedestal for the process. Then, the direction of the disjunctive arc is updated according to the current precast beam process action and the pedestal action. The newly generated disjunctive graph is used as the new local state. The pedestal status update is completed by the pedestal at time T t+1 (M k ) and prefabricated beam process a o Processing time p ijk Complete together and get the new state of the pedestal

[0125] S44. Design and define the reward function. The objective function is to minimize the maximum completion time of all precast beams. max , set the difference between the solution at time t+1 and time t to d t =C(s t+1 )-C(s t ), set the reward to -d t ;

[0126] S45. Set strategy, total strategy π θ (a o ,a m |s)Precast beam process sub-strategy and pedestal strategies Composition, where θ0 and θ m They represent the parameters of the prefabricated beam process sub-strategy and the pedestal sub-strategy, respectively. The total strategy π θ (a o ,a m |s) Select the precast beam process action and the pedestal action according to the probability distribution of the precast beam process action space and the pedestal action space respectively.

[0127] In another technical solution, Figure 1-2 As shown, the specific steps of step S5 are as follows:

[0128] S51. Construct a prefabricated beam decoder based on a multi-layer perceptron and use the decoder to calculate the prefabricated beam process action score.

[0129] S52. Convert the prefabricated beam process action scores into probabilities by performing Softmax normalization on the scores.

[0130] S53. Build a pedestal decoder based on a multi-layer perceptron and use the decoder to calculate the pedestal action score.

[0131] S54. Convert the pedestal action scores into probabilities by performing Softmax normalization on the scores

[0132] In another technical solution, Figure 1 、 3 As shown, a multi-actor-critic architecture is adopted and a multi-proximal policy optimization algorithm is used for reinforcement learning. The multi-actor-critic architecture reduces the amount of parameter updates in the policy network. The multi-proximal policy optimization algorithm ensures that the updated policy network is close to the network before the parameter update. The clipping parameters are used to control the policy differences, making the reinforcement learning process smoother. The specific steps of step S6 are as follows:

[0133] S61. Input parameters, including:

[0134] Number of training sessions E t , the number of updates of the multi-proximal strategy optimization algorithm E s , the number of samples B used in a single training, training actor network Behavior Actor Network And the parameters θ corresponding to the two networks h 、 Critics Network ν φ And the corresponding parameter φ, the strategy loss coefficient c p , the value function loss coefficient c v , entropy loss coefficient c e , shear parameter∈;

[0135] S62, initialize relevant parameters, including parameters θ for training the actor network h and the critic network parameter φ;

[0136] S63, randomly sampling from a uniform distribution, using a multi-proximal strategy optimization algorithm to learn the precast beam process sub-strategy and the pedestal sub-strategy;

[0137] S64, use the state-value function to evaluate the value of the two sub-strategies learned in S63, by inputting the state and outputting the value;

[0138] S65, after multiple iterations of training, output the trained prefabricated beam process actor network parameters θ0 and the pedestal actor network parameters θ m , and obtain the scheduling decision model.

[0139] In another technical solution, Figure 1-2 As shown, the specific steps of step S7 are as follows:

[0140] S71. Convert the actual erection requirements of precast beams, the number of pedestals involved in the business scenario, and actual business constraints into mathematical symbols and corresponding formulas.

[0141] S72. Based on the scheduling decision model completed through training optimization, input data including the scheduling start and deadline, the total number of prefabricated beams to be scheduled, the total number of pedestals, and the beam type, and substitute the data into the scheduling decision model for calculation to automatically generate the scheduling Gantt chart and scheduling schedule for the corresponding period, and obtain the scheduling plan results that minimize the delivery cycle and maximize the pedestal utilization rate.

[0142] In another technical solution, Figure 1 As shown, for the scheduling decision model, the definition of the mathematical symbols includes:

[0143] Process set P = {p1, p2, ..., p O}, where O represents the total number of processes;

[0144] Process path set P ij , represents the process p i On which machines can it be processed?

[0145] Machine set M={m1,m2,...,m n}, where n represents the total number of machines;

[0146] Processing time t ij , indicating that on machine m j Upper processing step p i the time required;

[0147] The processing sequence S={s1,s2,...,s m}, indicating the processing sequence constraints between processes;

[0148] The number of pedestals k represents the number of available pedestals;

[0149] Base number K = {k1, k2, ..., k k};

[0150] The start time of the process on the pedestal Represents process p i on pedestal k j Start time on

[0151] End time of the process on the pedestal Represents process p i on pedestal k j End time on;

[0152] Precondition constraint dij , represents the process p i The prerequisite p that must be met before completion j completion time;

[0153] The objective function is defined as follows:

[0154] The first objective function is to minimize the delivery cycle Minimize C max ,

[0155] Among them, C max represents the maximum completion time of all processes.

[0156] The second objective function is to maximize the utilization of the platform.

[0157]

[0158] in, Indicates the total processing time of all processes on all platforms, Indicates the total available time of all pedestals;

[0159] Comprehensive objective function Minimize f(x) = w1·C max -w2·(1-Utilization);

[0160] The constraints considered in actual production planning include:

[0161] Timing constraints: the sum of the precast beam start time and processing time must be less than the completion time, the end time of the previous process must be less than the start time of the next process, and the completion time of each precast beam must be before the total completion time;

[0162] Due to resource constraints, each pedestal can only process one process at a time;

[0163] Sequential constraints: different processes should have a strict processing order;

[0164] Flow constraints: the sequential movement, parallel movement and smooth movement of the pedestal must comply with the set route.

[0165] The present invention also provides a general intelligent scheduling system for smart beam yards, such as Figure 1 As shown, it includes input layer, strategy layer and output layer. The input layer includes information preprocessing module, strategy layer includes mathematical modeling module, algorithm solving module and deduction module, and output layer includes result visualization module.

[0166] The preprocessing module is used to input and read the input total production plan, change information, number of pedestals, process types supported by each pedestal, parameters and quantity of beams to be produced, process type, number of processes, and process time data for each beam production process, standardize the input data format to form information data, and transmit it to the mathematical modeling module;

[0167] The mathematical modeling module is used to read information data, perform mathematical modeling based on actual business scenarios, define mathematical symbols, build objective functions, establish actual constraints, and sequentially build a precast beam yard scheduling mathematical model, a multi-pointer graph network model, and a scheduling decision model;

[0168] The algorithm solving module is used to determine the next scheduling action to be taken based on the scheduling decision model according to the actual business scenario of precast beam yard scheduling, complete the solution of the precast beam yard scheduling problem, and transmit the calculation results to the deduction module;

[0169] The deduction module is used to place the scheduling sequences in the calculation results on the idle production resources in chronological order according to the actual production resources, constraints and calculation results returned by the algorithm solution module, simulate the processing of the scheduling sequences and analyze the resource consumption and time consumption of all scheduling sequences under all constraints, and perform strategy optimization. The optimized results are used as the final scheduling layout plan, and the final scheduling layout plan is sent to the result visualization module. At the same time, the change information after the implementation of the final scheduling layout plan is sent to the preprocessing module for update;

[0170] The result visualization module is used to read the final scheduling plan and generate the corresponding Gantt chart for visual display.

[0171] The universal intelligent scheduling system and method for the smart beam yard of the present invention was used for trial operation in the Yuxikou precast beam yard in Anhui Province. Six sets of pedestals were set up and 20 sets of beams were produced simultaneously. The beam process included steel bar binding, embedded parts installation, formwork installation, concrete pouring, steaming, demoulding and finishing. Each process of each beam took different time. The time taken for different processes of each beam on different pedestals was given in advance in minutes. The Gantt chart of the precast beam yard results calculated by the universal intelligent scheduling system for the smart beam yard developed according to the present invention is as follows: Figure 4 As shown, the scheduling result plan corresponding to the Gantt chart is as follows Figure 5 As shown, compared with the total production time of the production plan based on manual experience, after manual scheduling, the total production time of 20 beams is 160 hours (9600 minutes), while after calculation by the corresponding algorithm solution module of the present invention, the total production time of 20 beams is 6810 minutes, and the efficiency is improved by (9600-6810) / 9600=29.06%; and most of the pedestals can end their occupation ahead of time, and the production sequence of each group of beams is executed correctly, meeting the constraints of various settings.

[0172] In summary, the universal intelligent scheduling method and system of the smart beam yard of the present invention are used to describe and model the actual production scheduling problem scenarios of different beam production lines. Through the universal scheduling system architecture with the prefabricated beam yard scheduling method based on deep reinforcement learning and multi-pointer graph network as the core, the matching of production resources and scheduling results is optimized, and minimizing the delivery cycle and maximizing the base utilization rate are taken as optimization goals. The potential constraints are comprehensively considered and the overall scheduling plan is coordinated and arranged. The planning decision obtained by the scheduling system has high solution efficiency and can ensure the solution quality, effectively improve the effectiveness and executability of the scheduling series production plan, reduce the investment of labor costs, improve production efficiency, and shorten the processing cycle; and the universal system architecture has strong scalability, the solution process is explainable, the solution process and calculation logic are easy for scheduling staff to understand, easy to implement, and can be extended to the beam production planning links of different processes and process flows, and has universality.

[0173] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the description and implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A general intelligent scheduling method for a smart beam yard, characterized in that: The steps include: S1. Mathematically model the production scheduling problem of the precast beam yard to obtain a mathematical model for the precast beam yard scheduling; S2. Based on all relevant data on precast beam yard scheduling, randomly generate a precast beam yard scheduling data set that includes various uncertain factors; S3. Construct a multi-pointer graph network model and use it to extract and encode features of the precast beam yard scheduling dataset; S4. Convert the mathematical model of precast beam yard scheduling into a Markov decision process, design the state space, action space, reward function and parameter strategy, build a simulation scheduling environment, and form the precast beam process sub-strategy and pedestal sub-strategy; S5. Constructing a precast beam decoder and a pedestal decoder based on a multi-layer perceptron. According to the encoding result of step S3, the decoder calculates the probability distribution in the precast beam process action space and the pedestal action space respectively. S6. Adopting the actor-critic architecture, using the multi-proximal strategy optimization algorithm combined with the probability distribution of step S5 to learn the two sub-strategies of step S4, iteratively optimize, and obtain the scheduling decision model; S7. According to the actual business scenario of precast beam yard scheduling, the scheduling action to be taken next is given based on the scheduling decision model, the precast beam yard scheduling problem is solved, and the scheduling Gantt chart and scheduling plan corresponding to the actual business scenario are generated.

2. A general intelligent scheduling method for a smart beam yard as claimed in claim 1, characterized in that: The specific steps of mathematical modeling in step S1 are as follows: S11. Define the precast beam yard scheduling problem, including the duration and completion time of each process; S12. In the precast beam yard scheduling problem, determine the objective function as minimizing the maximum completion time; S13. Describe and define the constraints of the precast beam yard scheduling problem, including: Each process of prefabricating a beam has a sequence; The completion time of each precast beam cannot exceed the total completion time; Only one process can be processed on the same pedestal at the same time; The same process can only be processed by one pedestal at the same time; There can be a cycle operation on each pedestal.

3. A general intelligent scheduling method for a smart beam yard as claimed in claim 1, characterized in that: The specific steps of step S2 are as follows: S21. Determine the problem scale of the precast beam yard scheduling problem including various uncertainties, the number of tasks j, the number of pedestals m, and the time required for each task process; S21. For each process of each task, generate a real number to represent the time required for each process; S22. For all beams to be prefabricated of all tasks and all available pedestals, generate data matching all processes of all beams to be prefabricated and all available pedestals to obtain a prefabricated beam yard scheduling data set.

4. A general intelligent scheduling method for a smart beam yard as claimed in claim 1, characterized in that: The specific steps of step S3 are as follows: S31. Design a network structure consisting of a graph isomorphism network, a multi-layer perceptron, and task pointers to obtain the characteristics of the prefabricated beam process: in, is the embedding of node v in the l-th layer graph isomorphism network, is the multilayer perceptron at layer l, θ l is the parameter of the layer, ∈ (l) represents the parameters to be learned, t represents the time step, l∈{1,...,L}, Indicates the total number of processes; S32. Each node in the pedestal state information represents the main features of each pedestal, and the fully connected layer is used to encode the local state.

5. A general intelligent scheduling method for a smart beam yard as claimed in claim 1, characterized in that: The specific steps of step S4 are as follows: S41. Define and design the state space. The state of the environment at time t is determined by the local state. and composition, Indicates the status of the precast beam process. Indicates the status of the pedestal; S42. Define and design the action space. The action space at time t includes the prefabricated beam process action a. o and pedestal action a m ; S43. Define the transfer space. At time t, the agent samples a process and selects a pedestal for the process. Then, the direction of the disjunctive arc is updated according to the current precast beam process action and the pedestal action. The newly generated disjunctive graph is used as the new local state. The pedestal status update is completed by the pedestal at time T t+1 (M k ) and prefabricated beam process a o Processing time p ijk Complete together and get the new state of the pedestal S44. Design and define the reward function. The objective function is to minimize the maximum completion time of all precast beams. max , set the difference between the solution at time t+1 and time t to d t =C(s t+1 )-C(s t ), set the reward to -d t ; S45. Set strategy, total strategy π θ (a o ,a m |s)Precast beam process sub-strategy and pedestal strategies Composition, where θ0 and θ m They represent the parameters of the prefabricated beam process sub-strategy and the pedestal sub-strategy, respectively. The total strategy π θ (a o ,a m |s) Select the precast beam process action and the pedestal action according to the probability distribution of the precast beam process action space and the pedestal action space respectively.

6. A general intelligent scheduling method for a smart beam yard as claimed in claim 1, characterized in that: The specific steps of step S5 are as follows: S51. Construct a prefabricated beam decoder based on a multi-layer perceptron and use the decoder to calculate the prefabricated beam process action score. S52. Convert the prefabricated beam process action scores into probabilities by performing Softmax normalization on the scores. S53. Build a pedestal decoder based on a multi-layer perceptron and use the decoder to calculate the pedestal action score. S54. Convert the pedestal action scores into probabilities by performing Softmax normalization on the scores 7. A general intelligent scheduling method for a smart beam yard as claimed in claim 1, characterized in that: The specific steps of step S6 are as follows: S61. Input parameters, including: Number of training sessions E t , the number of updates of the multi-proximal strategy optimization algorithm E s , the number of samples B used in a single training, training actor network Behavior Actor Network And the parameters θ corresponding to the two networks h 、 Critics Network v φ And the corresponding parameter φ, the strategy loss coefficient c p , the value function loss coefficient c v , entropy loss coefficient c e , shear parameter∈; S62, initialize relevant parameters, including parameters θ for training the actor network h and the critic network parameter φ; S63, randomly sampling from a uniform distribution, using a multi-proximal strategy optimization algorithm to learn the precast beam process sub-strategy and the pedestal sub-strategy; S64, use the state-value function to evaluate the value of the two sub-strategies learned in S63, by inputting the state and outputting the value; S65, after multiple iterations of training, output the trained prefabricated beam process actor network parameters θ0 and the pedestal actor network parameters θ m , and obtain the scheduling decision model.

8. A general intelligent scheduling method for a smart beam yard as claimed in claim 1, characterized in that: The specific steps of step S7 are as follows: S71. Convert the actual erection requirements of precast beams, the number of pedestals involved in the business scenario, and actual business constraints into mathematical symbols and corresponding formulas. S72. Based on the scheduling decision model completed through training optimization, input data including the scheduling start and end dates, the total number of prefabricated beams to be scheduled, the total number of pedestals, and the beam type, substitute the data into the scheduling decision model for calculation, and automatically generate the scheduling Gantt chart and scheduling plan table for the corresponding period.

9. A general intelligent scheduling method for a smart beam yard as claimed in claim 8, characterized in that: For the scheduling decision model, the definitions of the mathematical symbols include: Process set P = {p1, p2, ..., p O }, where O represents the total number of processes; Process path set P ij , represents the process p i On which machines can it be processed? Machine set M={m1,m2,...,m n }, where n represents the total number of machines; Processing time t ij , indicating that on machine m j Upper processing step p i the time required; The processing sequence S={s1,s2,...,s m }, indicating the processing sequence constraints between processes; The number of pedestals k represents the number of available pedestals; Base number K = {k1, k2, ..., k k }; The start time of the process on the pedestal Represents process p i on pedestal k j Start time on End time of the process on the pedestal Represents process p i on pedestal k j End time on; Precondition constraint d ij , represents the process p i The prerequisite p that must be met before completion j completion time; The objective function is defined as follows: The first objective function is to minimize the delivery cycle Minimize C max , Among them, C max represents the maximum completion time of all processes. The second objective function is to maximize the utilization of the platform. in, Indicates the total processing time of all processes on all platforms, Indicates the total available time of all pedestals; Comprehensive objective function Minimize f(x) = w1·C max -w2·(1-Utilization); The constraints considered in actual production planning include: Timing constraints: the sum of the precast beam start time and processing time must be less than the completion time, the end time of the previous process must be less than the start time of the next process, and the completion time of each precast beam must be before the total completion time; Due to resource constraints, each pedestal can only process one process at a time; Sequential constraints: different processes should have a strict processing order; Flow constraints: the sequential movement, parallel movement and smooth movement of the pedestal must comply with the set route.

10. The universal intelligent scheduling system for a smart beam yard according to claim 1, characterized in that: It includes input layer, strategy layer and output layer. The input layer includes information preprocessing module, strategy layer includes mathematical modeling module, algorithm solving module and deduction module, and output layer includes result visualization module. The preprocessing module is used to input and read the input total production plan, change information, number of pedestals, process types supported by each pedestal, parameters and quantity of beams to be produced, process type, number of processes, and process time data for each beam production process, standardize the input data format to form information data, and transmit it to the mathematical modeling module; The mathematical modeling module is used to read information data, perform mathematical modeling based on actual business scenarios, define mathematical symbols, build objective functions, establish actual constraints, and sequentially build a precast beam yard scheduling mathematical model, a multi-pointer graph network model, and a scheduling decision model; The algorithm solving module is used to determine the next scheduling action to be taken based on the scheduling decision model according to the actual business scenario of precast beam yard scheduling, complete the solution of the precast beam yard scheduling problem, and transmit the calculation results to the deduction module; The deduction module is used to place the scheduling sequences in the calculation results on the idle production resources in chronological order according to the actual production resources, constraints and calculation results returned by the algorithm solution module, simulate the processing process of the scheduling sequences and the resource consumption and time consumption analysis of all scheduling sequences under all constraints, and perform strategy optimization. The optimized results are used as the final scheduling layout plan, and the final scheduling layout plan is sent to the result visualization module. At the same time, the change information after the implementation of the final scheduling layout plan is sent to the preprocessing module for updating; the result visualization module is used to read the final scheduling layout plan and generate the corresponding Gantt chart for visual display.

Images