A digital twin platform with simulated optimization kernel

Abstract

This invention relates to a digital twin platform with a simulation optimization kernel, comprising a physical entity and a digital twin, which can achieve virtual-physical synchronization and bidirectional data and instruction exchange. Its key feature is that it further includes a simulation operation module and a simulation optimization module for decision-makers to optimize decisions. During the operation of the digital twin platform, when a decision-maker needs to select an optimal solution from several candidate solutions, the digital twin controls the simulation operation module to perform simulations under different candidate solutions, obtaining output results with random noise. The simulation optimization module controls the transition between candidate solutions and the simulation accuracy under each candidate solution, ultimately deriving the optimal solution and transmitting it to the physical entity through instructions from the digital twin.

Description

Technical Field

[0001] This invention relates to a digital twin platform, and more particularly to a digital twin platform with a simulation-optimized kernel. Background Technology

[0002] Digital twins comprehensively utilize information technologies such as sensing, computing, and modeling. Through software definition, they describe, diagnose, predict, and make decisions about physical space, thereby achieving interactive mapping between physical and digital spaces and enabling intelligent decision-making. Most existing digital twin platforms for operations management achieve virtual-physical interaction between physical entities and their digital twins. They can perform simple decision optimizations in the digital space based on the digital twin and return the optimization results to the physical space for control of the physical entities.

[0003] However, in these practices, decision optimization based on digital twins in the digital space is either based on a series of predetermined rules or on simple, solvable mathematical models.

[0004] For example, CN 111857065 A discloses an intelligent production system and method based on edge computing and digital twins, including a physical system, edge digital twin nodes, a digital twin management and control system, a production simulation system, an order system, and an AI algorithm model library. The method includes: intelligent sensing devices acquiring physical production line information in real time and transmitting it to the edge digital twin nodes; the edge digital twin nodes constructing equipment models, predicting equipment failures and lifespan, and providing visualization; the digital twin management and control system generating simulation analysis tasks for production equipment scheduling, optimizing production scheduling strategies based on the simulation results of the production simulation system, and sending production scheduling instructions to the physical system. This reduces the computational burden on the terminal processor, decreases latency, and improves the information mapping efficiency and work efficiency of the entire intelligent production system, realizing equipment failure monitoring, prediction, and maintenance at the edge. The digital twin management and control system further optimizes the target operating status data of the production line based on the simulation results of the production simulation system and the production equipment scheduling optimization model in the AI ​​algorithm model library, generating production scheduling instructions for the physical system. However, the scheduling optimization model in this AI algorithm model library is used to analyze information from the physical production line in real time or to perform initial resource allocation for the manufacturing strategy of customized products. The optimization here is not based on the simulation model to make decisions. In addition, after receiving the simulation analysis results from the production simulation system, the digital twin management and control system optimizes the digital twin model and resource allocation scheme according to the simulation analysis results. This method is a well-known simple scheme comparison method, that is, running a simulation (a certain number of repetitions or time) and then comparing which scheme is the best. Because the output contains random noise, this method cannot guarantee that the optimal scheme will be selected in the end, and it does not achieve specific optimal scheme decision-making.

[0005] For example, CN114912369A discloses a mine intelligent management and optimization system based on digital twins, involving the field of mine management and control. This includes an intelligent mine model optimization system, a mine digital twin model optimization system, and a mine intelligent management and control platform optimization system. The intelligent mine model optimization system adopts a cyber-physical fusion approach to establish a virtual simulation model of the digital twin mine scene, equipment, and operational processes. It constructs and optimizes the intelligent mine model based on the interaction and data synchronization between the actual physical scene and the corresponding virtual simulation scene. Comparative document 5, through object twins, process twins, and performance twins of intelligent equipment, achieves performance evaluation of the mine digital twin, realizes the mapping of the entire lifecycle of physical equipment at each working face of the mine, and provides decision support for performance simulation and health prediction of the entire mine production process. This optimization scheme can only optimize specific mine equipment components and system integration descriptions, as well as optimize the real-time updating and visualization of dynamic data of mine virtual-physical interaction. Real-world production problems are often highly complex, making it impossible to cover all possibilities using established rules or to establish a solvable mathematical model without simplification. Simulation is essential to evaluate the effectiveness of candidate solutions and select the optimal solution that provides guidance for real-world operations. However, existing digital twin platform designs for operations management lack a simulation optimization kernel, namely a simulation execution module and a corresponding simulation optimization module. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a digital twin platform with a simulation optimization kernel, which can perform scheme optimization through simulation operation module and simulation optimization module, and provide the preferred scheme after the decision-making cycle ends.

[0007] To address the aforementioned technical problems, the technical solution of this invention is as follows: a digital twin platform with a simulation optimization kernel, comprising a physical entity and a digital twin. The digital twin can integrate various data of the physical entity, move and evolve synchronously with the physical entity, and send instructions to the physical entity through a controller to control the physical entity, thereby achieving virtual-real synchronization and bidirectional data and instruction interaction between the physical entity and the digital twin. Its innovation lies in the inclusion of a simulation operation module and a simulation optimization module for decision-makers to optimize decisions. During the operation of the digital twin platform, when a decision-maker needs to select an optimal solution from several candidate solutions, the simulation operation module can control the digital twin to perform simulation operations under different candidate solutions, obtaining output results with random noise. The simulation optimization module controls the conversion between candidate solutions and the simulation accuracy under each candidate solution, ultimately deriving the optimal solution, which is then transmitted to the physical entity through instructions from the digital twin.

[0008] Specifically, decision-makers need to select a batch of candidate solutions and the performance metrics they will focus on during optimization. Let this performance index be denoted as a numerical value. ,remember The solution with the largest value is the optimal solution; the simulation module considers candidate solutions. The output of the next simulation run is For the expected random variable The simulation running module is in the candidate scheme Next Repeat the process several times to obtain the simulation output. Then calculate the sample mean. .

[0009] Furthermore, when the simulation running module simulates and runs the batch of candidate schemes, the number of simulations for each candidate scheme is determined by the simulation optimization algorithm of the simulation optimization module.

[0010] Preferred, recorded Candidate solutions In obtaining Each simulation output The sample variance after that, Describing the degrees of freedom as of - The cumulative distribution function of the distribution, and the specific steps of the simulation optimization algorithm are as follows:

[0011] S1: Select algorithm parameters Run for each candidate solution The next simulation, let ← ,

[0012] make ← ,

[0013] S2: Perform the following steps until the stop condition is triggered.

[0014] make ← - , ;

[0015] make ← ;

[0016] Add one more simulation run to candidate solution b, and let ← +1;

[0017] renew , , ;

[0018] S3: Return The final selected option.

[0019] More specifically, in practical applications, the stopping conditions for the simulation optimization process include two different triggering methods: when the decision time is given as... At that time, the simulation optimization run took a long time. The time-stopping condition is triggered; when the decision-making time is unlimited, the optimal solution found by the algorithm reaches the preset accuracy. The time-stop condition is triggered; specifically, the parameters can be freely set. Then, when the cumulative number of simulations reaches If the following criteria are met, the stop condition is triggered:

[0020] ,

[0021] in , Describing the degrees of freedom as of -distribution quantiles.

[0022] The advantages of this invention are as follows: The digital twin platform with a simulation optimization kernel in this invention achieves decision optimization in the digital space based on the simulation operation of the digital twin and the simulation operation module, relying on the simulation optimization module with a matching simulation optimization algorithm. During the operation of the digital twin platform, when the decision-maker needs to optimize, that is, when it is necessary to select an optimal solution from a series of candidate solutions, the simulation operation is triggered, and an output result with random noise is obtained. The simulation optimization algorithm controls the transformation between solutions and the accuracy of the simulation under each solution, and finally selects the optimal solution or one that is as close to the optimal as possible, and transmits the solution to the physical space through control commands.

[0023] The digital twin platform with a simulation optimization kernel in this invention can be used in smart manufacturing scenarios, such as a production line. In this case, the physical entities are production equipment, conveyor belts, AGVs, etc., and the digital twin is their mapping in digital space. When a new order arrives, decision-makers may need to make a series of decisions within a given timeframe, including adjusting the production line structure and reordering the new order with existing orders. Decision-makers can then combine various decision factors to obtain... A comprehensive set of candidate solutions is provided. The simulation module allows for simulating production line operation under any candidate solution, calculating performance metrics of interest, such as net profit per unit time. The simulation optimization module seeks the best solution and returns a final solution at the end of the decision-making cycle.

[0024] The digital twin platform with a simulation optimization kernel in this invention can also be used in digital supply chain scenarios. In this case, the physical entity is a specific supply chain, and the digital twin is its mapping in the digital space. When a sudden event disrupts the original supply chain structure, decision-makers may need to make adjustments within a given timeframe, including changes to supply chain structure, inventory management, and work scheduling. Decision-makers can combine various decision factors to arrive at a solution. A comprehensive set of candidate solutions is provided. The simulation module allows for simulating the supply chain's operation under any candidate solution, calculating key performance indicators such as total supply chain cost. The simulation optimization module seeks the best solution and returns an optimal solution at the end of the decision-making cycle. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the working principle of the digital twin platform with simulation optimization kernel of the present invention.

[0026] Figure 2 This refers to the physical entity of the AGV system in this embodiment of the invention.

[0027] Figure 3 This is a digital twin of the AGV system in this embodiment of the invention. Detailed Implementation

[0028] like Figure 1 As shown, the digital twin platform of the present invention includes a physical entity and a digital twin. The digital twin can integrate various types of data of the physical entity, move and evolve synchronously with the physical entity, and send instructions to the physical entity through a controller to control the physical entity, thereby realizing virtual-real synchronization and two-way interaction of data and instructions between the physical entity and the digital twin. The specific interaction principles and methods have been implemented in existing digital twin platform practices in the industry, and will not be elaborated here.

[0029] The digital twin platform of the present invention also includes a simulation operation module and a simulation optimization module for decision optimization. The decision optimization in its digital space is achieved by the simulation operation module simulating the digital twin and relying on the simulation optimization algorithm in the simulation optimization module.

[0030] Specifically, during the operation of the digital twin platform, when decision-makers need to select an optimal solution from several candidate solutions, the simulation operation module can control the digital twin to perform simulation operation under different candidate solutions, obtain output results with random noise, and control the conversion between candidate solutions and the simulation accuracy under each candidate solution through the simulation optimization module, finally obtaining the optimal solution, and transmitting the optimal solution to the physical entity through the instructions of the digital twin.

[0031] When decision optimization is required and the simulation module is triggered, the decision-maker must first select a batch of candidate solutions and the performance indicators to be focused on during optimization. Without loss of generality, this performance index is denoted as a numerical value. ,remember The solution with the largest value is the optimal solution.

[0032] It is unknown and needs to be estimated through simulation. Simulation involves the digital twin deducing the operational logic of the physical entity in physical space under specific candidate scenarios. This is implemented in the simulation module. It involves various parameters in the operation process (such as the operating speed of the equipment, the time required for loading and unloading, etc.), which are fitted based on data accumulated during the actual operation of the physical entity. They usually involve randomness and are represented in the form of random distribution.

[0033] Due to uncertainties in the parameters, digital twins are used in candidate solutions. The output of the next simulation run is not Instead, For the expected random variable In order to get closer The simulation running module is in the candidate scheme The following needs to be done Repeat the process several times to obtain the simulation output. Then calculate the sample mean. = (

[0034] However, in order to ultimately obtain the truly optimal decision result efficiently, the key is how many simulations to run under which candidate scheme in this batch. This needs to be determined by the simulation optimization algorithm, which is implemented in the simulation optimization module.

[0035] Specifically: Record Candidate solutions In obtaining Each simulation output The sample variance after that, Describing the degrees of freedom as of - The cumulative distribution function of the distribution, and the specific steps of the simulation optimization algorithm are as follows:

[0036] S1: Select algorithm parameters Run for each candidate solution The next simulation, let ← ,

[0037] make ← ,

[0038] S2: Perform the following steps until the stop condition is triggered.

[0039] make ← - , ;

[0040] make ← ;

[0041] Add one more simulation run to candidate solution b, and let ← +1;

[0042] renew , , ;

[0043] S3: Return The final selected solution;

[0044] in, These are artificially introduced symbols, intended solely for simplification. Indicates that The candidate solution with the optimal value. .

[0045] In practical applications, the stopping conditions for the simulation optimization process include two different triggering methods:

[0046] When the decision time is given as At that time, the simulation optimization run took a long time. The stop condition is triggered;

[0047] When there is no time limit for decision-making, the optimal solution found by the algorithm achieves the preset accuracy. The time-stop condition is triggered; specifically, the parameters can be freely set. Then, when the cumulative number of simulations reaches If the following criteria are met, the stop condition is triggered:

[0048] ,

[0049] in , Describing the degrees of freedom as of -distribution quantiles.

[0050] Example:

[0051] To better illustrate the operation process of the digital twin platform, this embodiment uses... Taking a grid-layout bidirectional single-channel AGV system for logistics handling as an example, the length and width of each grid in this bidirectional single-channel AGV system are... AGVs can move shelves from one node to another. The rated travel speed of an AGV is [missing information]. The speed is meters per second, but with some random deviation. When a transport task arrives, it has a starting point and a destination. If an AGV is idle, the task is assigned to the idle AGV closest to the starting point; if no AGV is idle, the transport task is cached in the task pool. When an AGV becomes idle, it executes the task whose starting point is closest to it among all waiting tasks. After task assignment is determined, the AGV uses a time window shortest path algorithm to determine the route to the starting point and the route from the starting point to the destination. Due to the random deviation of the AGV's travel speed and the possible random obstacles in the road network, path conflicts may occur during AGV operation. When a path conflict occurs, the relevant AGVs need to be replanned to resolve the conflict.

[0052] Suppose we know the intensity of transportation tasks over a future period at a certain point in time, but the start and end points of the tasks are unknown. The decision-maker wants to make an optimal decision regarding the number of AGVs to configure, such that the time required to complete all transportation tasks (denoted as ) is ... The shortest. As the number of AGVs increases, more transport tasks can be performed simultaneously. On the one hand, it reduces the number of paths, but on the other hand, it also leads to more path conflicts. The problem is to increase the size of the object. This is a complex decision-making problem that cannot be solved by establishing a mathematical model due to the presence of randomness, but it can be solved by using the simulation optimization kernel in this design.

[0053] For ease of explanation, assume the decision-maker is considering three AGV configurations: 12, 14, and 16 AGVs, resulting in three candidate schemes: Scheme 1 with 12 AGVs, Scheme 2 with 14 AGVs, and Scheme 3 with 16 AGVs. The goal is to select the optimal configuration. The minimum solution, that is, making The largest possible solution. For candidate solutions... Its The expected value is denoted as ,but It is unknown. By leveraging the simulation execution module within the simulation optimization kernel, candidate solutions can be identified. Multiple random simulations were performed to obtain the following results: For the expected random observation However, unless the simulation is repeated an infinite number of times, the sample mean of random observations will not be the same as... There is always random error involved, therefore decisions based on the sample mean cannot achieve [the desired outcome]. The accuracy. Now set the target accuracy as... ,Right now The simulation optimization module in the simulation optimization kernel is used to find the optimal solution. Algorithm parameters are set. Stop condition parameters The algorithm can be run, and the results are shown in the table below.

[0054]

[0055] First, run algorithm step 1, performing 10 simulations for each scheme, and observe the results. , , (See lines 1-10). Note that some observations in Scheme 3 are greater than those in Scheme 2, and some are less than those in Scheme 2; Scheme 1 also has observations greater than those in Scheme 2. At this point... The total number of simulations was Calculated , , , , , (See line 10). Entering the loop of step 2, we calculate the result at this point. The simulation took 53.5 seconds to reach this step (see line 10).

[0056] Next, an additional simulation run was performed for Scheme 3, and the following observations were made. , The number of simulations changes to 11, the total number of simulations becomes 31, and the update is performed. , The results for Scheme 1 and Scheme 2 are not updated (see line 11). The calculation at this point... The simulation took 55.3 seconds to reach this step (see line 11).

[0057] By repeating this process, we can see that the algorithm will append simulation runs to different schemes and update the results. Due to the settings... When the total number of simulations reaches At that time, it will determine whether the condition for triggering the stop is met. When the total number of simulations reaches When (see line 60), the calculation shows that the discrimination condition is not met, so the algorithm continues. When the total number of simulations reaches... When (see line 160), the judgment condition is met after calculation. Therefore, when the stopping condition is reached, the accuracy of the algorithm is... The algorithm will stop at this point. , , Therefore, option 2 will be chosen as the optimal solution. The algorithm guarantees that when the value is greater than or equal to... Under these circumstances, the solution derived in this way is indeed the optimal solution.

[0058] Decision-makers can also use another stopping condition, namely, the simulation optimization run time reaches a certain threshold. Stop at that time. For example, let... At this point, given the simulation output above, the simulation will stop calculating line 24 and use the result of line 23 to make a decision. Under normal circumstances, the decision made at this point has no guarantee of accuracy, unless otherwise specified. The time required for the judgment condition to be met is greater than the time required, as shown in the simulation output above. Second.

[0059] The foregoing has shown and described the basic principles and main features of the present invention, as well as its advantages. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A digital twin platform with a simulation optimization kernel, comprising a physical entity and a digital twin, wherein the digital twin is capable of integrating various data of the physical entity, moving and evolving synchronously with the physical entity, and sending instructions to the physical entity via a controller to control the physical entity, thereby achieving virtual-physical synchronization and bidirectional data and instruction interaction between the physical entity and the digital twin; characterized in that: It also includes a simulation execution module and a simulation optimization module for decision-makers to optimize their decisions. During the operation of the digital twin platform, when decision-makers need to select an optimal solution from several candidate solutions, they can control the digital twin to perform simulations under different candidate solutions through the simulation operation module to obtain output results with random noise. The simulation optimization module controls the conversion between candidate solutions and the simulation accuracy under each candidate solution, and finally obtains the optimal solution. The optimal solution is then transmitted to the physical entity through the instructions of the digital twin. Decision-makers need to select a batch of candidate solutions and the performance metrics they will focus on during optimization. Let this performance index be denoted as a numerical value. ,remember The solution with the largest value is the optimal solution; the simulation module considers candidate solutions. The output of the next simulation run is For the expected random variable The simulation running module is in the candidate scheme Next Repeat the process several times to obtain the simulation output. Then calculate the sample mean. ; When the simulation running module simulates and runs the batch of candidate schemes, the number of simulations for each candidate scheme is determined by the simulation optimization algorithm of the simulation optimization module; Recall that the candidate solution is obtained after a number of simulation runs and that the sample variance is given by the cumulative distribution function of a chi-squared distribution with degrees of freedom The specific steps of the simulation optimization algorithm are as follows: S1: select algorithm parameters run for each candidate sub-simulation, let ← , make ← , S2: Perform the following steps until the stop condition is triggered. make ← - , ; make ← ; Add one more simulation run to candidate solution b, and let ← +1; renew , , ; S3: Return The final selected option.

2. The digital twin platform with a simulation-optimized kernel according to claim 1, characterized in that: In practical applications, the stopping conditions for the simulation optimization process include two different triggering methods: When the decision time is given as At that time, the simulation optimization run took a long time. The stop condition is triggered; When there is no time limit for decision-making, the optimal solution found by the algorithm achieves the preset accuracy. The time-stop condition is triggered; specifically, the parameters can be freely set. Then, when the cumulative number of simulations reaches If the following criteria are met, the stop condition is triggered: ， in , Describing the degrees of freedom as of -distribution quantiles.

Images